net/ipv6/ip6_output.c

   1 /*
   2  *      IPv6 output functions
   3  *      Linux INET6 implementation
   4  *
   5  *      Authors:
   6  *      Pedro Roque             <roque@di.fc.ul.pt>
   7  *
   8  *      Based on linux/net/ipv4/ip_output.c
   9  *
  10  *      This program is free software; you can redistribute it and/or
  11  *      modify it under the terms of the GNU General Public License
  12  *      as published by the Free Software Foundation; either version
  13  *      2 of the License, or (at your option) any later version.
  14  *
  15  *      Changes:
  16  *      A.N.Kuznetsov   :       airthmetics in fragmentation.
  17  *                              extension headers are implemented.
  18  *                              route changes now work.
  19  *                              ip6_forward does not confuse sniffers.
  20  *                              etc.
  21  *
  22  *      H. von Brand    :       Added missing #include <linux/string.h>
  23  *      Imran Patel     :       frag id should be in NBO
  24  *      Kazunori MIYAZAWA @USAGI
  25  *                      :       add ip6_append_data and related functions
  26  *                              for datagram xmit
  27  */
  28
  29 #include <linux/errno.h>
  30 #include <linux/kernel.h>
  31 #include <linux/string.h>
  32 #include <linux/socket.h>
  33 #include <linux/net.h>
  34 #include <linux/netdevice.h>
  35 #include <linux/if_arp.h>
  36 #include <linux/in6.h>
  37 #include <linux/tcp.h>
  38 #include <linux/route.h>
  39 #include <linux/module.h>
  40 #include <linux/slab.h>
  41
  42 #include <linux/netfilter.h>
  43 #include <linux/netfilter_ipv6.h>
  44
  45 #include <net/sock.h>
  46 #include <net/snmp.h>
  47
  48 #include <net/ipv6.h>
  49 #include <net/ndisc.h>
  50 #include <net/protocol.h>
  51 #include <net/ip6_route.h>
  52 #include <net/addrconf.h>
  53 #include <net/rawv6.h>
  54 #include <net/icmp.h>
  55 #include <net/xfrm.h>
  56 #include <net/checksum.h>
  57 #include <linux/mroute6.h>
  58
  59 static int ip6_finish_output2(struct sk_buff *skb)
  60 {
  61         struct dst_entry *dst = skb_dst(skb);
  62         struct net_device *dev = dst->dev;
  63         struct neighbour *neigh;
  64         struct in6_addr *nexthop;
  65         int ret;
  66
  67         skb->protocol = htons(ETH_P_IPV6);
  68         skb->dev = dev;
  69
  70         if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
  71                 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
  72
  73                 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(skb->sk) &&
  74                     ((mroute6_socket(dev_net(dev), skb) &&
  75                      !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
  76                      ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
  77                                          &ipv6_hdr(skb)->saddr))) {
  78                         struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
  79
  80                         /* Do not check for IFF_ALLMULTI; multicast routing
  81                            is not supported in any case.
  82                          */
  83                         if (newskb)
  84                                 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
  85                                         newskb, NULL, newskb->dev,
  86                                         dev_loopback_xmit);
  87
  88                         if (ipv6_hdr(skb)->hop_limit == 0) {
  89                                 IP6_INC_STATS(dev_net(dev), idev,
  90                                               IPSTATS_MIB_OUTDISCARDS);
  91                                 kfree_skb(skb);
  92                                 return 0;
  93                         }
  94                 }
  95
  96                 IP6_UPD_PO_STATS(dev_net(dev), idev, IPSTATS_MIB_OUTMCAST,
  97                                 skb->len);
  98
  99                 if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb)->daddr) <=
 100                     IPV6_ADDR_SCOPE_NODELOCAL &&
 101                     !(dev->flags & IFF_LOOPBACK)) {
 102                         kfree_skb(skb);
 103                         return 0;
 104                 }
 105         }
 106
 107         rcu_read_lock_bh();
 108         nexthop = rt6_nexthop((struct rt6_info *)dst);
 109         neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
 110         if (unlikely(!neigh))
 111                 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
 112         if (!IS_ERR(neigh)) {
 113                 ret = dst_neigh_output(dst, neigh, skb);
 114                 rcu_read_unlock_bh();
 115                 return ret;
 116         }
 117         rcu_read_unlock_bh();
 118
 119         IP6_INC_STATS(dev_net(dst->dev),
 120                       ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
 121         kfree_skb(skb);
 122         return -EINVAL;
 123 }
 124
 125 static int ip6_finish_output(struct sk_buff *skb)
 126 {
 127         if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
 128             dst_allfrag(skb_dst(skb)) ||
 129             (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
 130                 return ip6_fragment(skb, ip6_finish_output2);
 131         else
 132                 return ip6_finish_output2(skb);
 133 }
 134
 135 int ip6_output(struct sk_buff *skb)
 136 {
 137         struct net_device *dev = skb_dst(skb)->dev;
 138         struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
 139         if (unlikely(idev->cnf.disable_ipv6)) {
 140                 IP6_INC_STATS(dev_net(dev), idev,
 141                               IPSTATS_MIB_OUTDISCARDS);
 142                 kfree_skb(skb);
 143                 return 0;
 144         }
 145
 146         return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, skb, NULL, dev,
 147                             ip6_finish_output,
 148                             !(IP6CB(skb)->flags & IP6SKB_REROUTED));
 149 }
 150
 151 /*
 152  *      xmit an sk_buff (used by TCP, SCTP and DCCP)
 153  */
 154
 155 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
 156              struct ipv6_txoptions *opt, int tclass)
 157 {
 158         struct net *net = sock_net(sk);
 159         struct ipv6_pinfo *np = inet6_sk(sk);
 160         struct in6_addr *first_hop = &fl6->daddr;
 161         struct dst_entry *dst = skb_dst(skb);
 162         struct ipv6hdr *hdr;
 163         u8  proto = fl6->flowi6_proto;
 164         int seg_len = skb->len;
 165         int hlimit = -1;
 166         u32 mtu;
 167
 168         if (opt) {
 169                 unsigned int head_room;
 170
 171                 /* First: exthdrs may take lots of space (~8K for now)
 172                    MAX_HEADER is not enough.
 173                  */
 174                 head_room = opt->opt_nflen + opt->opt_flen;
 175                 seg_len += head_room;
 176                 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
 177
 178                 if (skb_headroom(skb) < head_room) {
 179                         struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
 180                         if (skb2 == NULL) {
 181                                 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 182                                               IPSTATS_MIB_OUTDISCARDS);
 183                                 kfree_skb(skb);
 184                                 return -ENOBUFS;
 185                         }
 186                         consume_skb(skb);
 187                         skb = skb2;
 188                         skb_set_owner_w(skb, sk);
 189                 }
 190                 if (opt->opt_flen)
 191                         ipv6_push_frag_opts(skb, opt, &proto);
 192                 if (opt->opt_nflen)
 193                         ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
 194         }
 195
 196         skb_push(skb, sizeof(struct ipv6hdr));
 197         skb_reset_network_header(skb);
 198         hdr = ipv6_hdr(skb);
 199
 200         /*
 201          *      Fill in the IPv6 header
 202          */
 203         if (np)
 204                 hlimit = np->hop_limit;
 205         if (hlimit < 0)
 206                 hlimit = ip6_dst_hoplimit(dst);
 207
 208         ip6_flow_hdr(hdr, tclass, fl6->flowlabel);
 209
 210         hdr->payload_len = htons(seg_len);
 211         hdr->nexthdr = proto;
 212         hdr->hop_limit = hlimit;
 213
 214         hdr->saddr = fl6->saddr;
 215         hdr->daddr = *first_hop;
 216
 217         skb->protocol = htons(ETH_P_IPV6);
 218         skb->priority = sk->sk_priority;
 219         skb->mark = sk->sk_mark;
 220
 221         mtu = dst_mtu(dst);
 222         if ((skb->len <= mtu) || skb->local_df || skb_is_gso(skb)) {
 223                 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
 224                               IPSTATS_MIB_OUT, skb->len);
 225                 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, skb, NULL,
 226                                dst->dev, dst_output);
 227         }
 228
 229         skb->dev = dst->dev;
 230         ipv6_local_error(sk, EMSGSIZE, fl6, mtu);
 231         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
 232         kfree_skb(skb);
 233         return -EMSGSIZE;
 234 }
 235
 236 EXPORT_SYMBOL(ip6_xmit);
 237
 238 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
 239 {
 240         struct ip6_ra_chain *ra;
 241         struct sock *last = NULL;
 242
 243         read_lock(&ip6_ra_lock);
 244         for (ra = ip6_ra_chain; ra; ra = ra->next) {
 245                 struct sock *sk = ra->sk;
 246                 if (sk && ra->sel == sel &&
 247                     (!sk->sk_bound_dev_if ||
 248                      sk->sk_bound_dev_if == skb->dev->ifindex)) {
 249                         if (last) {
 250                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
 251                                 if (skb2)
 252                                         rawv6_rcv(last, skb2);
 253                         }
 254                         last = sk;
 255                 }
 256         }
 257
 258         if (last) {
 259                 rawv6_rcv(last, skb);
 260                 read_unlock(&ip6_ra_lock);
 261                 return 1;
 262         }
 263         read_unlock(&ip6_ra_lock);
 264         return 0;
 265 }
 266
 267 static int ip6_forward_proxy_check(struct sk_buff *skb)
 268 {
 269         struct ipv6hdr *hdr = ipv6_hdr(skb);
 270         u8 nexthdr = hdr->nexthdr;
 271         __be16 frag_off;
 272         int offset;
 273
 274         if (ipv6_ext_hdr(nexthdr)) {
 275                 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
 276                 if (offset < 0)
 277                         return 0;
 278         } else
 279                 offset = sizeof(struct ipv6hdr);
 280
 281         if (nexthdr == IPPROTO_ICMPV6) {
 282                 struct icmp6hdr *icmp6;
 283
 284                 if (!pskb_may_pull(skb, (skb_network_header(skb) +
 285                                          offset + 1 - skb->data)))
 286                         return 0;
 287
 288                 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
 289
 290                 switch (icmp6->icmp6_type) {
 291                 case NDISC_ROUTER_SOLICITATION:
 292                 case NDISC_ROUTER_ADVERTISEMENT:
 293                 case NDISC_NEIGHBOUR_SOLICITATION:
 294                 case NDISC_NEIGHBOUR_ADVERTISEMENT:
 295                 case NDISC_REDIRECT:
 296                         /* For reaction involving unicast neighbor discovery
 297                          * message destined to the proxied address, pass it to
 298                          * input function.
 299                          */
 300                         return 1;
 301                 default:
 302                         break;
 303                 }
 304         }
 305
 306         /*
 307          * The proxying router can't forward traffic sent to a link-local
 308          * address, so signal the sender and discard the packet. This
 309          * behavior is clarified by the MIPv6 specification.
 310          */
 311         if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
 312                 dst_link_failure(skb);
 313                 return -1;
 314         }
 315
 316         return 0;
 317 }
 318
 319 static inline int ip6_forward_finish(struct sk_buff *skb)
 320 {
 321         return dst_output(skb);
 322 }
 323
 324 static unsigned int ip6_dst_mtu_forward(const struct dst_entry *dst)
 325 {
 326         unsigned int mtu;
 327         struct inet6_dev *idev;
 328
 329         if (dst_metric_locked(dst, RTAX_MTU)) {
 330                 mtu = dst_metric_raw(dst, RTAX_MTU);
 331                 if (mtu)
 332                         return mtu;
 333         }
 334
 335         mtu = IPV6_MIN_MTU;
 336         rcu_read_lock();
 337         idev = __in6_dev_get(dst->dev);
 338         if (idev)
 339                 mtu = idev->cnf.mtu6;
 340         rcu_read_unlock();
 341
 342         return mtu;
 343 }
 344
 345 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
 346 {
 347         if (skb->len <= mtu)
 348                 return false;
 349
 350         /* ipv6 conntrack defrag sets max_frag_size + local_df */
 351         if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
 352                 return true;
 353
 354         if (skb->local_df)
 355                 return false;
 356
 357         if (skb_is_gso(skb) && skb_gso_network_seglen(skb) <= mtu)
 358                 return false;
 359
 360         return true;
 361 }
 362
 363 int ip6_forward(struct sk_buff *skb)
 364 {
 365         struct dst_entry *dst = skb_dst(skb);
 366         struct ipv6hdr *hdr = ipv6_hdr(skb);
 367         struct inet6_skb_parm *opt = IP6CB(skb);
 368         struct net *net = dev_net(dst->dev);
 369         u32 mtu;
 370
 371         if (net->ipv6.devconf_all->forwarding == 0)
 372                 goto error;
 373
 374         if (skb_warn_if_lro(skb))
 375                 goto drop;
 376
 377         if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
 378                 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
 379                                  IPSTATS_MIB_INDISCARDS);
 380                 goto drop;
 381         }
 382
 383         if (skb->pkt_type != PACKET_HOST)
 384                 goto drop;
 385
 386         skb_forward_csum(skb);
 387
 388         /*
 389          *      We DO NOT make any processing on
 390          *      RA packets, pushing them to user level AS IS
 391          *      without ane WARRANTY that application will be able
 392          *      to interpret them. The reason is that we
 393          *      cannot make anything clever here.
 394          *
 395          *      We are not end-node, so that if packet contains
 396          *      AH/ESP, we cannot make anything.
 397          *      Defragmentation also would be mistake, RA packets
 398          *      cannot be fragmented, because there is no warranty
 399          *      that different fragments will go along one path. --ANK
 400          */
 401         if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
 402                 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
 403                         return 0;
 404         }
 405
 406         /*
 407          *      check and decrement ttl
 408          */
 409         if (hdr->hop_limit <= 1) {
 410                 /* Force OUTPUT device used as source address */
 411                 skb->dev = dst->dev;
 412                 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
 413                 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
 414                                  IPSTATS_MIB_INHDRERRORS);
 415
 416                 kfree_skb(skb);
 417                 return -ETIMEDOUT;
 418         }
 419
 420         /* XXX: idev->cnf.proxy_ndp? */
 421         if (net->ipv6.devconf_all->proxy_ndp &&
 422             pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
 423                 int proxied = ip6_forward_proxy_check(skb);
 424                 if (proxied > 0)
 425                         return ip6_input(skb);
 426                 else if (proxied < 0) {
 427                         IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
 428                                          IPSTATS_MIB_INDISCARDS);
 429                         goto drop;
 430                 }
 431         }
 432
 433         if (!xfrm6_route_forward(skb)) {
 434                 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
 435                                  IPSTATS_MIB_INDISCARDS);
 436                 goto drop;
 437         }
 438         dst = skb_dst(skb);
 439
 440         /* IPv6 specs say nothing about it, but it is clear that we cannot
 441            send redirects to source routed frames.
 442            We don't send redirects to frames decapsulated from IPsec.
 443          */
 444         if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
 445                 struct in6_addr *target = NULL;
 446                 struct inet_peer *peer;
 447                 struct rt6_info *rt;
 448
 449                 /*
 450                  *      incoming and outgoing devices are the same
 451                  *      send a redirect.
 452                  */
 453
 454                 rt = (struct rt6_info *) dst;
 455                 if (rt->rt6i_flags & RTF_GATEWAY)
 456                         target = &rt->rt6i_gateway;
 457                 else
 458                         target = &hdr->daddr;
 459
 460                 peer = inet_getpeer_v6(net->ipv6.peers, &rt->rt6i_dst.addr, 1);
 461
 462                 /* Limit redirects both by destination (here)
 463                    and by source (inside ndisc_send_redirect)
 464                  */
 465                 if (inet_peer_xrlim_allow(peer, 1*HZ))
 466                         ndisc_send_redirect(skb, target);
 467                 if (peer)
 468                         inet_putpeer(peer);
 469         } else {
 470                 int addrtype = ipv6_addr_type(&hdr->saddr);
 471
 472                 /* This check is security critical. */
 473                 if (addrtype == IPV6_ADDR_ANY ||
 474                     addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
 475                         goto error;
 476                 if (addrtype & IPV6_ADDR_LINKLOCAL) {
 477                         icmpv6_send(skb, ICMPV6_DEST_UNREACH,
 478                                     ICMPV6_NOT_NEIGHBOUR, 0);
 479                         goto error;
 480                 }
 481         }
 482
 483         mtu = ip6_dst_mtu_forward(dst);
 484         if (mtu < IPV6_MIN_MTU)
 485                 mtu = IPV6_MIN_MTU;
 486
 487         if (ip6_pkt_too_big(skb, mtu)) {
 488                 /* Again, force OUTPUT device used as source address */
 489                 skb->dev = dst->dev;
 490                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
 491                 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
 492                                  IPSTATS_MIB_INTOOBIGERRORS);
 493                 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
 494                                  IPSTATS_MIB_FRAGFAILS);
 495                 kfree_skb(skb);
 496                 return -EMSGSIZE;
 497         }
 498
 499         if (skb_cow(skb, dst->dev->hard_header_len)) {
 500                 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
 501                                  IPSTATS_MIB_OUTDISCARDS);
 502                 goto drop;
 503         }
 504
 505         hdr = ipv6_hdr(skb);
 506
 507         /* Mangling hops number delayed to point after skb COW */
 508
 509         hdr->hop_limit--;
 510
 511         IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
 512         IP6_ADD_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
 513         return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD, skb, skb->dev, dst->dev,
 514                        ip6_forward_finish);
 515
 516 error:
 517         IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
 518 drop:
 519         kfree_skb(skb);
 520         return -EINVAL;
 521 }
 522
 523 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
 524 {
 525         to->pkt_type = from->pkt_type;
 526         to->priority = from->priority;
 527         to->protocol = from->protocol;
 528         skb_dst_drop(to);
 529         skb_dst_set(to, dst_clone(skb_dst(from)));
 530         to->dev = from->dev;
 531         to->mark = from->mark;
 532
 533 #ifdef CONFIG_NET_SCHED
 534         to->tc_index = from->tc_index;
 535 #endif
 536         nf_copy(to, from);
 537         skb_copy_secmark(to, from);
 538 }
 539
 540 static void ipv6_select_ident(struct frag_hdr *fhdr, struct rt6_info *rt)
 541 {
 542         static u32 ip6_idents_hashrnd __read_mostly;
 543         u32 hash, id;
 544
 545         net_get_random_once(&ip6_idents_hashrnd, sizeof(ip6_idents_hashrnd));
 546
 547         hash = __ipv6_addr_jhash(&rt->rt6i_dst.addr, ip6_idents_hashrnd);
 548         hash = __ipv6_addr_jhash(&rt->rt6i_src.addr, hash);
 549
 550         id = ip_idents_reserve(hash, 1);
 551         fhdr->identification = htonl(id);
 552 }
 553
 554 int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
 555 {
 556         struct sk_buff *frag;
 557         struct rt6_info *rt = (struct rt6_info*)skb_dst(skb);
 558         struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
 559         struct ipv6hdr *tmp_hdr;
 560         struct frag_hdr *fh;
 561         unsigned int mtu, hlen, left, len;
 562         int hroom, troom;
 563         __be32 frag_id = 0;
 564         int ptr, offset = 0, err=0;
 565         u8 *prevhdr, nexthdr = 0;
 566         struct net *net = dev_net(skb_dst(skb)->dev);
 567
 568         hlen = ip6_find_1stfragopt(skb, &prevhdr);
 569         nexthdr = *prevhdr;
 570
 571         mtu = ip6_skb_dst_mtu(skb);
 572
 573         /* We must not fragment if the socket is set to force MTU discovery
 574          * or if the skb it not generated by a local socket.
 575          */
 576         if (unlikely(!skb->local_df && skb->len > mtu) ||
 577                      (IP6CB(skb)->frag_max_size &&
 578                       IP6CB(skb)->frag_max_size > mtu)) {
 579                 if (skb->sk && dst_allfrag(skb_dst(skb)))
 580                         sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
 581
 582                 skb->dev = skb_dst(skb)->dev;
 583                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
 584                 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 585                               IPSTATS_MIB_FRAGFAILS);
 586                 kfree_skb(skb);
 587                 return -EMSGSIZE;
 588         }
 589
 590         if (np && np->frag_size < mtu) {
 591                 if (np->frag_size)
 592                         mtu = np->frag_size;
 593         }
 594         mtu -= hlen + sizeof(struct frag_hdr);
 595
 596         if (skb_has_frag_list(skb)) {
 597                 int first_len = skb_pagelen(skb);
 598                 struct sk_buff *frag2;
 599
 600                 if (first_len - hlen > mtu ||
 601                     ((first_len - hlen) & 7) ||
 602                     skb_cloned(skb))
 603                         goto slow_path;
 604
 605                 skb_walk_frags(skb, frag) {
 606                         /* Correct geometry. */
 607                         if (frag->len > mtu ||
 608                             ((frag->len & 7) && frag->next) ||
 609                             skb_headroom(frag) < hlen)
 610                                 goto slow_path_clean;
 611
 612                         /* Partially cloned skb? */
 613                         if (skb_shared(frag))
 614                                 goto slow_path_clean;
 615
 616                         BUG_ON(frag->sk);
 617                         if (skb->sk) {
 618                                 frag->sk = skb->sk;
 619                                 frag->destructor = sock_wfree;
 620                         }
 621                         skb->truesize -= frag->truesize;
 622                 }
 623
 624                 err = 0;
 625                 offset = 0;
 626                 frag = skb_shinfo(skb)->frag_list;
 627                 skb_frag_list_init(skb);
 628                 /* BUILD HEADER */
 629
 630                 *prevhdr = NEXTHDR_FRAGMENT;
 631                 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
 632                 if (!tmp_hdr) {
 633                         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 634                                       IPSTATS_MIB_FRAGFAILS);
 635                         return -ENOMEM;
 636                 }
 637
 638                 __skb_pull(skb, hlen);
 639                 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
 640                 __skb_push(skb, hlen);
 641                 skb_reset_network_header(skb);
 642                 memcpy(skb_network_header(skb), tmp_hdr, hlen);
 643
 644                 ipv6_select_ident(fh, rt);
 645                 fh->nexthdr = nexthdr;
 646                 fh->reserved = 0;
 647                 fh->frag_off = htons(IP6_MF);
 648                 frag_id = fh->identification;
 649
 650                 first_len = skb_pagelen(skb);
 651                 skb->data_len = first_len - skb_headlen(skb);
 652                 skb->len = first_len;
 653                 ipv6_hdr(skb)->payload_len = htons(first_len -
 654                                                    sizeof(struct ipv6hdr));
 655
 656                 dst_hold(&rt->dst);
 657
 658                 for (;;) {
 659                         /* Prepare header of the next frame,
 660                          * before previous one went down. */
 661                         if (frag) {
 662                                 frag->ip_summed = CHECKSUM_NONE;
 663                                 skb_reset_transport_header(frag);
 664                                 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
 665                                 __skb_push(frag, hlen);
 666                                 skb_reset_network_header(frag);
 667                                 memcpy(skb_network_header(frag), tmp_hdr,
 668                                        hlen);
 669                                 offset += skb->len - hlen - sizeof(struct frag_hdr);
 670                                 fh->nexthdr = nexthdr;
 671                                 fh->reserved = 0;
 672                                 fh->frag_off = htons(offset);
 673                                 if (frag->next != NULL)
 674                                         fh->frag_off |= htons(IP6_MF);
 675                                 fh->identification = frag_id;
 676                                 ipv6_hdr(frag)->payload_len =
 677                                                 htons(frag->len -
 678                                                       sizeof(struct ipv6hdr));
 679                                 ip6_copy_metadata(frag, skb);
 680                         }
 681
 682                         err = output(skb);
 683                         if(!err)
 684                                 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
 685                                               IPSTATS_MIB_FRAGCREATES);
 686
 687                         if (err || !frag)
 688                                 break;
 689
 690                         skb = frag;
 691                         frag = skb->next;
 692                         skb->next = NULL;
 693                 }
 694
 695                 kfree(tmp_hdr);
 696
 697                 if (err == 0) {
 698                         IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
 699                                       IPSTATS_MIB_FRAGOKS);
 700                         ip6_rt_put(rt);
 701                         return 0;
 702                 }
 703
 704                 while (frag) {
 705                         skb = frag->next;
 706                         kfree_skb(frag);
 707                         frag = skb;
 708                 }
 709
 710                 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
 711                               IPSTATS_MIB_FRAGFAILS);
 712                 ip6_rt_put(rt);
 713                 return err;
 714
 715 slow_path_clean:
 716                 skb_walk_frags(skb, frag2) {
 717                         if (frag2 == frag)
 718                                 break;
 719                         frag2->sk = NULL;
 720                         frag2->destructor = NULL;
 721                         skb->truesize += frag2->truesize;
 722                 }
 723         }
 724
 725 slow_path:
 726         if ((skb->ip_summed == CHECKSUM_PARTIAL) &&
 727             skb_checksum_help(skb))
 728                 goto fail;
 729
 730         left = skb->len - hlen;         /* Space per frame */
 731         ptr = hlen;                     /* Where to start from */
 732
 733         /*
 734          *      Fragment the datagram.
 735          */
 736
 737         *prevhdr = NEXTHDR_FRAGMENT;
 738         hroom = LL_RESERVED_SPACE(rt->dst.dev);
 739         troom = rt->dst.dev->needed_tailroom;
 740
 741         /*
 742          *      Keep copying data until we run out.
 743          */
 744         while(left > 0) {
 745                 len = left;
 746                 /* IF: it doesn't fit, use 'mtu' - the data space left */
 747                 if (len > mtu)
 748                         len = mtu;
 749                 /* IF: we are not sending up to and including the packet end
 750                    then align the next start on an eight byte boundary */
 751                 if (len < left) {
 752                         len &= ~7;
 753                 }
 754                 /*
 755                  *      Allocate buffer.
 756                  */
 757
 758                 if ((frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
 759                                       hroom + troom, GFP_ATOMIC)) == NULL) {
 760                         NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
 761                         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 762                                       IPSTATS_MIB_FRAGFAILS);
 763                         err = -ENOMEM;
 764                         goto fail;
 765                 }
 766
 767                 /*
 768                  *      Set up data on packet
 769                  */
 770
 771                 ip6_copy_metadata(frag, skb);
 772                 skb_reserve(frag, hroom);
 773                 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
 774                 skb_reset_network_header(frag);
 775                 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
 776                 frag->transport_header = (frag->network_header + hlen +
 777                                           sizeof(struct frag_hdr));
 778
 779                 /*
 780                  *      Charge the memory for the fragment to any owner
 781                  *      it might possess
 782                  */
 783                 if (skb->sk)
 784                         skb_set_owner_w(frag, skb->sk);
 785
 786                 /*
 787                  *      Copy the packet header into the new buffer.
 788                  */
 789                 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
 790
 791                 /*
 792                  *      Build fragment header.
 793                  */
 794                 fh->nexthdr = nexthdr;
 795                 fh->reserved = 0;
 796                 if (!frag_id) {
 797                         ipv6_select_ident(fh, rt);
 798                         frag_id = fh->identification;
 799                 } else
 800                         fh->identification = frag_id;
 801
 802                 /*
 803                  *      Copy a block of the IP datagram.
 804                  */
 805                 if (skb_copy_bits(skb, ptr, skb_transport_header(frag), len))
 806                         BUG();
 807                 left -= len;
 808
 809                 fh->frag_off = htons(offset);
 810                 if (left > 0)
 811                         fh->frag_off |= htons(IP6_MF);
 812                 ipv6_hdr(frag)->payload_len = htons(frag->len -
 813                                                     sizeof(struct ipv6hdr));
 814
 815                 ptr += len;
 816                 offset += len;
 817
 818                 /*
 819                  *      Put this fragment into the sending queue.
 820                  */
 821                 err = output(frag);
 822                 if (err)
 823                         goto fail;
 824
 825                 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 826                               IPSTATS_MIB_FRAGCREATES);
 827         }
 828         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 829                       IPSTATS_MIB_FRAGOKS);
 830         consume_skb(skb);
 831         return err;
 832
 833 fail:
 834         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 835                       IPSTATS_MIB_FRAGFAILS);
 836         kfree_skb(skb);
 837         return err;
 838 }
 839
 840 static inline int ip6_rt_check(const struct rt6key *rt_key,
 841                                const struct in6_addr *fl_addr,
 842                                const struct in6_addr *addr_cache)
 843 {
 844         return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
 845                 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache));
 846 }
 847
 848 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
 849                                           struct dst_entry *dst,
 850                                           const struct flowi6 *fl6)
 851 {
 852         struct ipv6_pinfo *np = inet6_sk(sk);
 853         struct rt6_info *rt;
 854
 855         if (!dst)
 856                 goto out;
 857
 858         if (dst->ops->family != AF_INET6) {
 859                 dst_release(dst);
 860                 return NULL;
 861         }
 862
 863         rt = (struct rt6_info *)dst;
 864         /* Yes, checking route validity in not connected
 865          * case is not very simple. Take into account,
 866          * that we do not support routing by source, TOS,
 867          * and MSG_DONTROUTE            --ANK (980726)
 868          *
 869          * 1. ip6_rt_check(): If route was host route,
 870          *    check that cached destination is current.
 871          *    If it is network route, we still may
 872          *    check its validity using saved pointer
 873          *    to the last used address: daddr_cache.
 874          *    We do not want to save whole address now,
 875          *    (because main consumer of this service
 876          *    is tcp, which has not this problem),
 877          *    so that the last trick works only on connected
 878          *    sockets.
 879          * 2. oif also should be the same.
 880          */
 881         if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
 882 #ifdef CONFIG_IPV6_SUBTREES
 883             ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
 884 #endif
 885             (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
 886                 dst_release(dst);
 887                 dst = NULL;
 888         }
 889
 890 out:
 891         return dst;
 892 }
 893
 894 static int ip6_dst_lookup_tail(struct sock *sk,
 895                                struct dst_entry **dst, struct flowi6 *fl6)
 896 {
 897         struct net *net = sock_net(sk);
 898 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
 899         struct neighbour *n;
 900         struct rt6_info *rt;
 901 #endif
 902         int err;
 903
 904         if (*dst == NULL)
 905                 *dst = ip6_route_output(net, sk, fl6);
 906
 907         if ((err = (*dst)->error))
 908                 goto out_err_release;
 909
 910         if (ipv6_addr_any(&fl6->saddr)) {
 911                 struct rt6_info *rt = (struct rt6_info *) *dst;
 912                 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
 913                                           sk ? inet6_sk(sk)->srcprefs : 0,
 914                                           &fl6->saddr);
 915                 if (err)
 916                         goto out_err_release;
 917         }
 918
 919 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
 920         /*
 921          * Here if the dst entry we've looked up
 922          * has a neighbour entry that is in the INCOMPLETE
 923          * state and the src address from the flow is
 924          * marked as OPTIMISTIC, we release the found
 925          * dst entry and replace it instead with the
 926          * dst entry of the nexthop router
 927          */
 928         rt = (struct rt6_info *) *dst;
 929         rcu_read_lock_bh();
 930         n = __ipv6_neigh_lookup_noref(rt->dst.dev, rt6_nexthop(rt));
 931         err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
 932         rcu_read_unlock_bh();
 933
 934         if (err) {
 935                 struct inet6_ifaddr *ifp;
 936                 struct flowi6 fl_gw6;
 937                 int redirect;
 938
 939                 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
 940                                       (*dst)->dev, 1);
 941
 942                 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
 943                 if (ifp)
 944                         in6_ifa_put(ifp);
 945
 946                 if (redirect) {
 947                         /*
 948                          * We need to get the dst entry for the
 949                          * default router instead
 950                          */
 951                         dst_release(*dst);
 952                         memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
 953                         memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
 954                         *dst = ip6_route_output(net, sk, &fl_gw6);
 955                         if ((err = (*dst)->error))
 956                                 goto out_err_release;
 957                 }
 958         }
 959 #endif
 960
 961         return 0;
 962
 963 out_err_release:
 964         if (err == -ENETUNREACH)
 965                 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
 966         dst_release(*dst);
 967         *dst = NULL;
 968         return err;
 969 }
 970
 971 /**
 972  *      ip6_dst_lookup - perform route lookup on flow
 973  *      @sk: socket which provides route info
 974  *      @dst: pointer to dst_entry * for result
 975  *      @fl6: flow to lookup
 976  *
 977  *      This function performs a route lookup on the given flow.
 978  *
 979  *      It returns zero on success, or a standard errno code on error.
 980  */
 981 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi6 *fl6)
 982 {
 983         *dst = NULL;
 984         return ip6_dst_lookup_tail(sk, dst, fl6);
 985 }
 986 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
 987
 988 /**
 989  *      ip6_dst_lookup_flow - perform route lookup on flow with ipsec
 990  *      @sk: socket which provides route info
 991  *      @fl6: flow to lookup
 992  *      @final_dst: final destination address for ipsec lookup
 993  *
 994  *      This function performs a route lookup on the given flow.
 995  *
 996  *      It returns a valid dst pointer on success, or a pointer encoded
 997  *      error code.
 998  */
 999 struct dst_entry *ip6_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1000                                       const struct in6_addr *final_dst)
1001 {
1002         struct dst_entry *dst = NULL;
1003         int err;
1004
1005         err = ip6_dst_lookup_tail(sk, &dst, fl6);
1006         if (err)
1007                 return ERR_PTR(err);
1008         if (final_dst)
1009                 fl6->daddr = *final_dst;
1010
1011         return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1012 }
1013 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1014
1015 /**
1016  *      ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1017  *      @sk: socket which provides the dst cache and route info
1018  *      @fl6: flow to lookup
1019  *      @final_dst: final destination address for ipsec lookup
1020  *
1021  *      This function performs a route lookup on the given flow with the
1022  *      possibility of using the cached route in the socket if it is valid.
1023  *      It will take the socket dst lock when operating on the dst cache.
1024  *      As a result, this function can only be used in process context.
1025  *
1026  *      It returns a valid dst pointer on success, or a pointer encoded
1027  *      error code.
1028  */
1029 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1030                                          const struct in6_addr *final_dst)
1031 {
1032         struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1033         int err;
1034
1035         dst = ip6_sk_dst_check(sk, dst, fl6);
1036
1037         err = ip6_dst_lookup_tail(sk, &dst, fl6);
1038         if (err)
1039                 return ERR_PTR(err);
1040         if (final_dst)
1041                 fl6->daddr = *final_dst;
1042
1043         return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1044 }
1045 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1046
1047 static inline int ip6_ufo_append_data(struct sock *sk,
1048                         int getfrag(void *from, char *to, int offset, int len,
1049                         int odd, struct sk_buff *skb),
1050                         void *from, int length, int hh_len, int fragheaderlen,
1051                         int transhdrlen, int mtu,unsigned int flags,
1052                         struct rt6_info *rt)
1053
1054 {
1055         struct sk_buff *skb;
1056         struct frag_hdr fhdr;
1057         int err;
1058
1059         /* There is support for UDP large send offload by network
1060          * device, so create one single skb packet containing complete
1061          * udp datagram
1062          */
1063         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
1064                 skb = sock_alloc_send_skb(sk,
1065                         hh_len + fragheaderlen + transhdrlen + 20,
1066                         (flags & MSG_DONTWAIT), &err);
1067                 if (skb == NULL)
1068                         return err;
1069
1070                 /* reserve space for Hardware header */
1071                 skb_reserve(skb, hh_len);
1072
1073                 /* create space for UDP/IP header */
1074                 skb_put(skb,fragheaderlen + transhdrlen);
1075
1076                 /* initialize network header pointer */
1077                 skb_reset_network_header(skb);
1078
1079                 /* initialize protocol header pointer */
1080                 skb->transport_header = skb->network_header + fragheaderlen;
1081
1082                 skb->protocol = htons(ETH_P_IPV6);
1083                 skb->csum = 0;
1084
1085                 __skb_queue_tail(&sk->sk_write_queue, skb);
1086         } else if (skb_is_gso(skb)) {
1087                 goto append;
1088         }
1089
1090         skb->ip_summed = CHECKSUM_PARTIAL;
1091         /* Specify the length of each IPv6 datagram fragment.
1092          * It has to be a multiple of 8.
1093          */
1094         skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1095                                      sizeof(struct frag_hdr)) & ~7;
1096         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1097         ipv6_select_ident(&fhdr, rt);
1098         skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1099
1100 append:
1101         return skb_append_datato_frags(sk, skb, getfrag, from,
1102                                        (length - transhdrlen));
1103 }
1104
1105 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1106                                                gfp_t gfp)
1107 {
1108         return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1109 }
1110
1111 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1112                                                 gfp_t gfp)
1113 {
1114         return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1115 }
1116
1117 static void ip6_append_data_mtu(unsigned int *mtu,
1118                                 int *maxfraglen,
1119                                 unsigned int fragheaderlen,
1120                                 struct sk_buff *skb,
1121                                 struct rt6_info *rt,
1122                                 unsigned int orig_mtu)
1123 {
1124         if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1125                 if (skb == NULL) {
1126                         /* first fragment, reserve header_len */
1127                         *mtu = orig_mtu - rt->dst.header_len;
1128
1129                 } else {
1130                         /*
1131                          * this fragment is not first, the headers
1132                          * space is regarded as data space.
1133                          */
1134                         *mtu = orig_mtu;
1135                 }
1136                 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1137                               + fragheaderlen - sizeof(struct frag_hdr);
1138         }
1139 }
1140
1141 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1142         int offset, int len, int odd, struct sk_buff *skb),
1143         void *from, int length, int transhdrlen,
1144         int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1145         struct rt6_info *rt, unsigned int flags, int dontfrag)
1146 {
1147         struct inet_sock *inet = inet_sk(sk);
1148         struct ipv6_pinfo *np = inet6_sk(sk);
1149         struct inet_cork *cork;
1150         struct sk_buff *skb, *skb_prev = NULL;
1151         unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu;
1152         int exthdrlen;
1153         int dst_exthdrlen;
1154         int hh_len;
1155         int copy;
1156         int err;
1157         int offset = 0;
1158         __u8 tx_flags = 0;
1159
1160         if (flags&MSG_PROBE)
1161                 return 0;
1162         cork = &inet->cork.base;
1163         if (skb_queue_empty(&sk->sk_write_queue)) {
1164                 /*
1165                  * setup for corking
1166                  */
1167                 if (opt) {
1168                         if (WARN_ON(np->cork.opt))
1169                                 return -EINVAL;
1170
1171                         np->cork.opt = kzalloc(opt->tot_len, sk->sk_allocation);
1172                         if (unlikely(np->cork.opt == NULL))
1173                                 return -ENOBUFS;
1174
1175                         np->cork.opt->tot_len = opt->tot_len;
1176                         np->cork.opt->opt_flen = opt->opt_flen;
1177                         np->cork.opt->opt_nflen = opt->opt_nflen;
1178
1179                         np->cork.opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1180                                                             sk->sk_allocation);
1181                         if (opt->dst0opt && !np->cork.opt->dst0opt)
1182                                 return -ENOBUFS;
1183
1184                         np->cork.opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1185                                                             sk->sk_allocation);
1186                         if (opt->dst1opt && !np->cork.opt->dst1opt)
1187                                 return -ENOBUFS;
1188
1189                         np->cork.opt->hopopt = ip6_opt_dup(opt->hopopt,
1190                                                            sk->sk_allocation);
1191                         if (opt->hopopt && !np->cork.opt->hopopt)
1192                                 return -ENOBUFS;
1193
1194                         np->cork.opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1195                                                             sk->sk_allocation);
1196                         if (opt->srcrt && !np->cork.opt->srcrt)
1197                                 return -ENOBUFS;
1198
1199                         /* need source address above miyazawa*/
1200                 }
1201                 dst_hold(&rt->dst);
1202                 cork->dst = &rt->dst;
1203                 inet->cork.fl.u.ip6 = *fl6;
1204                 np->cork.hop_limit = hlimit;
1205                 np->cork.tclass = tclass;
1206                 if (rt->dst.flags & DST_XFRM_TUNNEL)
1207                         mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1208                               rt->dst.dev->mtu : dst_mtu(&rt->dst);
1209                 else
1210                         mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1211                               rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1212                 if (np->frag_size < mtu) {
1213                         if (np->frag_size)
1214                                 mtu = np->frag_size;
1215                 }
1216                 cork->fragsize = mtu;
1217                 if (dst_allfrag(rt->dst.path))
1218                         cork->flags |= IPCORK_ALLFRAG;
1219                 cork->length = 0;
1220                 exthdrlen = (opt ? opt->opt_flen : 0);
1221                 length += exthdrlen;
1222                 transhdrlen += exthdrlen;
1223                 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1224         } else {
1225                 rt = (struct rt6_info *)cork->dst;
1226                 fl6 = &inet->cork.fl.u.ip6;
1227                 opt = np->cork.opt;
1228                 transhdrlen = 0;
1229                 exthdrlen = 0;
1230                 dst_exthdrlen = 0;
1231                 mtu = cork->fragsize;
1232         }
1233         orig_mtu = mtu;
1234
1235         hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1236
1237         fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1238                         (opt ? opt->opt_nflen : 0);
1239         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1240                      sizeof(struct frag_hdr);
1241
1242         if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1243                 unsigned int maxnonfragsize, headersize;
1244
1245                 headersize = sizeof(struct ipv6hdr) +
1246                              (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1247                              (dst_allfrag(&rt->dst) ?
1248                               sizeof(struct frag_hdr) : 0) +
1249                              rt->rt6i_nfheader_len;
1250
1251                 maxnonfragsize = (np->pmtudisc >= IPV6_PMTUDISC_DO) ?
1252                                  mtu : sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1253
1254                 /* dontfrag active */
1255                 if ((cork->length + length > mtu - headersize) && dontfrag &&
1256                     (sk->sk_protocol == IPPROTO_UDP ||
1257                      sk->sk_protocol == IPPROTO_RAW)) {
1258                         ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1259                                                    sizeof(struct ipv6hdr));
1260                         goto emsgsize;
1261                 }
1262
1263                 if (cork->length + length > maxnonfragsize - headersize) {
1264 emsgsize:
1265                         ipv6_local_error(sk, EMSGSIZE, fl6,
1266                                          mtu - headersize +
1267                                          sizeof(struct ipv6hdr));
1268                         return -EMSGSIZE;
1269                 }
1270         }
1271
1272         /* For UDP, check if TX timestamp is enabled */
1273         if (sk->sk_type == SOCK_DGRAM)
1274                 sock_tx_timestamp(sk, &tx_flags);
1275
1276         /*
1277          * Let's try using as much space as possible.
1278          * Use MTU if total length of the message fits into the MTU.
1279          * Otherwise, we need to reserve fragment header and
1280          * fragment alignment (= 8-15 octects, in total).
1281          *
1282          * Note that we may need to "move" the data from the tail of
1283          * of the buffer to the new fragment when we split
1284          * the message.
1285          *
1286          * FIXME: It may be fragmented into multiple chunks
1287          *        at once if non-fragmentable extension headers
1288          *        are too large.
1289          * --yoshfuji
1290          */
1291
1292         skb = skb_peek_tail(&sk->sk_write_queue);
1293         cork->length += length;
1294         if (((length > mtu) ||
1295              (skb && skb_is_gso(skb))) &&
1296             (sk->sk_protocol == IPPROTO_UDP) &&
1297             (rt->dst.dev->features & NETIF_F_UFO)) {
1298                 err = ip6_ufo_append_data(sk, getfrag, from, length,
1299                                           hh_len, fragheaderlen,
1300                                           transhdrlen, mtu, flags, rt);
1301                 if (err)
1302                         goto error;
1303                 return 0;
1304         }
1305
1306         if (!skb)
1307                 goto alloc_new_skb;
1308
1309         while (length > 0) {
1310                 /* Check if the remaining data fits into current packet. */
1311                 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1312                 if (copy < length)
1313                         copy = maxfraglen - skb->len;
1314
1315                 if (copy <= 0) {
1316                         char *data;
1317                         unsigned int datalen;
1318                         unsigned int fraglen;
1319                         unsigned int fraggap;
1320                         unsigned int alloclen;
1321 alloc_new_skb:
1322                         /* There's no room in the current skb */
1323                         if (skb)
1324                                 fraggap = skb->len - maxfraglen;
1325                         else
1326                                 fraggap = 0;
1327                         /* update mtu and maxfraglen if necessary */
1328                         if (skb == NULL || skb_prev == NULL)
1329                                 ip6_append_data_mtu(&mtu, &maxfraglen,
1330                                                     fragheaderlen, skb, rt,
1331                                                     orig_mtu);
1332
1333                         skb_prev = skb;
1334
1335                         /*
1336                          * If remaining data exceeds the mtu,
1337                          * we know we need more fragment(s).
1338                          */
1339                         datalen = length + fraggap;
1340
1341                         if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1342                                 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1343                         if ((flags & MSG_MORE) &&
1344                             !(rt->dst.dev->features&NETIF_F_SG))
1345                                 alloclen = mtu;
1346                         else
1347                                 alloclen = datalen + fragheaderlen;
1348
1349                         alloclen += dst_exthdrlen;
1350
1351                         if (datalen != length + fraggap) {
1352                                 /*
1353                                  * this is not the last fragment, the trailer
1354                                  * space is regarded as data space.
1355                                  */
1356                                 datalen += rt->dst.trailer_len;
1357                         }
1358
1359                         alloclen += rt->dst.trailer_len;
1360                         fraglen = datalen + fragheaderlen;
1361
1362                         /*
1363                          * We just reserve space for fragment header.
1364                          * Note: this may be overallocation if the message
1365                          * (without MSG_MORE) fits into the MTU.
1366                          */
1367                         alloclen += sizeof(struct frag_hdr);
1368
1369                         if (transhdrlen) {
1370                                 skb = sock_alloc_send_skb(sk,
1371                                                 alloclen + hh_len,
1372                                                 (flags & MSG_DONTWAIT), &err);
1373                         } else {
1374                                 skb = NULL;
1375                                 if (atomic_read(&sk->sk_wmem_alloc) <=
1376                                     2 * sk->sk_sndbuf)
1377                                         skb = sock_wmalloc(sk,
1378                                                            alloclen + hh_len, 1,
1379                                                            sk->sk_allocation);
1380                                 if (unlikely(skb == NULL))
1381                                         err = -ENOBUFS;
1382                                 else {
1383                                         /* Only the initial fragment
1384                                          * is time stamped.
1385                                          */
1386                                         tx_flags = 0;
1387                                 }
1388                         }
1389                         if (skb == NULL)
1390                                 goto error;
1391                         /*
1392                          *      Fill in the control structures
1393                          */
1394                         skb->protocol = htons(ETH_P_IPV6);
1395                         skb->ip_summed = CHECKSUM_NONE;
1396                         skb->csum = 0;
1397                         /* reserve for fragmentation and ipsec header */
1398                         skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1399                                     dst_exthdrlen);
1400
1401                         if (sk->sk_type == SOCK_DGRAM)
1402                                 skb_shinfo(skb)->tx_flags = tx_flags;
1403
1404                         /*
1405                          *      Find where to start putting bytes
1406                          */
1407                         data = skb_put(skb, fraglen);
1408                         skb_set_network_header(skb, exthdrlen);
1409                         data += fragheaderlen;
1410                         skb->transport_header = (skb->network_header +
1411                                                  fragheaderlen);
1412                         if (fraggap) {
1413                                 skb->csum = skb_copy_and_csum_bits(
1414                                         skb_prev, maxfraglen,
1415                                         data + transhdrlen, fraggap, 0);
1416                                 skb_prev->csum = csum_sub(skb_prev->csum,
1417                                                           skb->csum);
1418                                 data += fraggap;
1419                                 pskb_trim_unique(skb_prev, maxfraglen);
1420                         }
1421                         copy = datalen - transhdrlen - fraggap;
1422
1423                         if (copy < 0) {
1424                                 err = -EINVAL;
1425                                 kfree_skb(skb);
1426                                 goto error;
1427                         } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1428                                 err = -EFAULT;
1429                                 kfree_skb(skb);
1430                                 goto error;
1431                         }
1432
1433                         offset += copy;
1434                         length -= datalen - fraggap;
1435                         transhdrlen = 0;
1436                         exthdrlen = 0;
1437                         dst_exthdrlen = 0;
1438
1439                         /*
1440                          * Put the packet on the pending queue
1441                          */
1442                         __skb_queue_tail(&sk->sk_write_queue, skb);
1443                         continue;
1444                 }
1445
1446                 if (copy > length)
1447                         copy = length;
1448
1449                 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1450                         unsigned int off;
1451
1452                         off = skb->len;
1453                         if (getfrag(from, skb_put(skb, copy),
1454                                                 offset, copy, off, skb) < 0) {
1455                                 __skb_trim(skb, off);
1456                                 err = -EFAULT;
1457                                 goto error;
1458                         }
1459                 } else {
1460                         int i = skb_shinfo(skb)->nr_frags;
1461                         struct page_frag *pfrag = sk_page_frag(sk);
1462
1463                         err = -ENOMEM;
1464                         if (!sk_page_frag_refill(sk, pfrag))
1465                                 goto error;
1466
1467                         if (!skb_can_coalesce(skb, i, pfrag->page,
1468                                               pfrag->offset)) {
1469                                 err = -EMSGSIZE;
1470                                 if (i == MAX_SKB_FRAGS)
1471                                         goto error;
1472
1473                                 __skb_fill_page_desc(skb, i, pfrag->page,
1474                                                      pfrag->offset, 0);
1475                                 skb_shinfo(skb)->nr_frags = ++i;
1476                                 get_page(pfrag->page);
1477                         }
1478                         copy = min_t(int, copy, pfrag->size - pfrag->offset);
1479                         if (getfrag(from,
1480                                     page_address(pfrag->page) + pfrag->offset,
1481                                     offset, copy, skb->len, skb) < 0)
1482                                 goto error_efault;
1483
1484                         pfrag->offset += copy;
1485                         skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1486                         skb->len += copy;
1487                         skb->data_len += copy;
1488                         skb->truesize += copy;
1489                         atomic_add(copy, &sk->sk_wmem_alloc);
1490                 }
1491                 offset += copy;
1492                 length -= copy;
1493         }
1494
1495         return 0;
1496
1497 error_efault:
1498         err = -EFAULT;
1499 error:
1500         cork->length -= length;
1501         IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1502         return err;
1503 }
1504 EXPORT_SYMBOL_GPL(ip6_append_data);
1505
1506 static void ip6_cork_release(struct inet_sock *inet, struct ipv6_pinfo *np)
1507 {
1508         if (np->cork.opt) {
1509                 kfree(np->cork.opt->dst0opt);
1510                 kfree(np->cork.opt->dst1opt);
1511                 kfree(np->cork.opt->hopopt);
1512                 kfree(np->cork.opt->srcrt);
1513                 kfree(np->cork.opt);
1514                 np->cork.opt = NULL;
1515         }
1516
1517         if (inet->cork.base.dst) {
1518                 dst_release(inet->cork.base.dst);
1519                 inet->cork.base.dst = NULL;
1520                 inet->cork.base.flags &= ~IPCORK_ALLFRAG;
1521         }
1522         memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1523 }
1524
1525 int ip6_push_pending_frames(struct sock *sk)
1526 {
1527         struct sk_buff *skb, *tmp_skb;
1528         struct sk_buff **tail_skb;
1529         struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1530         struct inet_sock *inet = inet_sk(sk);
1531         struct ipv6_pinfo *np = inet6_sk(sk);
1532         struct net *net = sock_net(sk);
1533         struct ipv6hdr *hdr;
1534         struct ipv6_txoptions *opt = np->cork.opt;
1535         struct rt6_info *rt = (struct rt6_info *)inet->cork.base.dst;
1536         struct flowi6 *fl6 = &inet->cork.fl.u.ip6;
1537         unsigned char proto = fl6->flowi6_proto;
1538         int err = 0;
1539
1540         if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1541                 goto out;
1542         tail_skb = &(skb_shinfo(skb)->frag_list);
1543
1544         /* move skb->data to ip header from ext header */
1545         if (skb->data < skb_network_header(skb))
1546                 __skb_pull(skb, skb_network_offset(skb));
1547         while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1548                 __skb_pull(tmp_skb, skb_network_header_len(skb));
1549                 *tail_skb = tmp_skb;
1550                 tail_skb = &(tmp_skb->next);
1551                 skb->len += tmp_skb->len;
1552                 skb->data_len += tmp_skb->len;
1553                 skb->truesize += tmp_skb->truesize;
1554                 tmp_skb->destructor = NULL;
1555                 tmp_skb->sk = NULL;
1556         }
1557
1558         /* Allow local fragmentation. */
1559         if (np->pmtudisc < IPV6_PMTUDISC_DO)
1560                 skb->local_df = 1;
1561
1562         *final_dst = fl6->daddr;
1563         __skb_pull(skb, skb_network_header_len(skb));
1564         if (opt && opt->opt_flen)
1565                 ipv6_push_frag_opts(skb, opt, &proto);
1566         if (opt && opt->opt_nflen)
1567                 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1568
1569         skb_push(skb, sizeof(struct ipv6hdr));
1570         skb_reset_network_header(skb);
1571         hdr = ipv6_hdr(skb);
1572
1573         ip6_flow_hdr(hdr, np->cork.tclass, fl6->flowlabel);
1574         hdr->hop_limit = np->cork.hop_limit;
1575         hdr->nexthdr = proto;
1576         hdr->saddr = fl6->saddr;
1577         hdr->daddr = *final_dst;
1578
1579         skb->priority = sk->sk_priority;
1580         skb->mark = sk->sk_mark;
1581
1582         skb_dst_set(skb, dst_clone(&rt->dst));
1583         IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1584         if (proto == IPPROTO_ICMPV6) {
1585                 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1586
1587                 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1588                 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1589         }
1590
1591         err = ip6_local_out(skb);
1592         if (err) {
1593                 if (err > 0)
1594                         err = net_xmit_errno(err);
1595                 if (err)
1596                         goto error;
1597         }
1598
1599 out:
1600         ip6_cork_release(inet, np);
1601         return err;
1602 error:
1603         IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1604         goto out;
1605 }
1606 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1607
1608 void ip6_flush_pending_frames(struct sock *sk)
1609 {
1610         struct sk_buff *skb;
1611
1612         while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1613                 if (skb_dst(skb))
1614                         IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1615                                       IPSTATS_MIB_OUTDISCARDS);
1616                 kfree_skb(skb);
1617         }
1618
1619         ip6_cork_release(inet_sk(sk), inet6_sk(sk));
1620 }
1621 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);