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_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *));
  60
  61 int __ip6_local_out(struct sk_buff *skb)
  62 {
  63         int len;
  64
  65         len = skb->len - sizeof(struct ipv6hdr);
  66         if (len > IPV6_MAXPLEN)
  67                 len = 0;
  68         ipv6_hdr(skb)->payload_len = htons(len);
  69
  70         return nf_hook(PF_INET6, NF_INET_LOCAL_OUT, skb, NULL, skb_dst(skb)->dev,
  71                        dst_output);
  72 }
  73
  74 int ip6_local_out(struct sk_buff *skb)
  75 {
  76         int err;
  77
  78         err = __ip6_local_out(skb);
  79         if (likely(err == 1))
  80                 err = dst_output(skb);
  81
  82         return err;
  83 }
  84 EXPORT_SYMBOL_GPL(ip6_local_out);
  85
  86 static int ip6_output_finish(struct sk_buff *skb)
  87 {
  88         struct dst_entry *dst = skb_dst(skb);
  89
  90         if (dst->hh)
  91                 return neigh_hh_output(dst->hh, skb);
  92         else if (dst->neighbour)
  93                 return dst->neighbour->output(skb);
  94
  95         IP6_INC_STATS_BH(dev_net(dst->dev),
  96                          ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
  97         kfree_skb(skb);
  98         return -EINVAL;
  99
 100 }
 101
 102 /* dev_loopback_xmit for use with netfilter. */
 103 static int ip6_dev_loopback_xmit(struct sk_buff *newskb)
 104 {
 105         skb_reset_mac_header(newskb);
 106         __skb_pull(newskb, skb_network_offset(newskb));
 107         newskb->pkt_type = PACKET_LOOPBACK;
 108         newskb->ip_summed = CHECKSUM_UNNECESSARY;
 109         WARN_ON(!skb_dst(newskb));
 110
 111         netif_rx(newskb);
 112         return 0;
 113 }
 114
 115
 116 static int ip6_output2(struct sk_buff *skb)
 117 {
 118         struct dst_entry *dst = skb_dst(skb);
 119         struct net_device *dev = dst->dev;
 120
 121         skb->protocol = htons(ETH_P_IPV6);
 122         skb->dev = dev;
 123
 124         if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
 125                 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
 126
 127                 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(skb->sk) &&
 128                     ((mroute6_socket(dev_net(dev)) &&
 129                      !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
 130                      ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
 131                                          &ipv6_hdr(skb)->saddr))) {
 132                         struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
 133
 134                         /* Do not check for IFF_ALLMULTI; multicast routing
 135                            is not supported in any case.
 136                          */
 137                         if (newskb)
 138                                 NF_HOOK(PF_INET6, NF_INET_POST_ROUTING, newskb,
 139                                         NULL, newskb->dev,
 140                                         ip6_dev_loopback_xmit);
 141
 142                         if (ipv6_hdr(skb)->hop_limit == 0) {
 143                                 IP6_INC_STATS(dev_net(dev), idev,
 144                                               IPSTATS_MIB_OUTDISCARDS);
 145                                 kfree_skb(skb);
 146                                 return 0;
 147                         }
 148                 }
 149
 150                 IP6_UPD_PO_STATS(dev_net(dev), idev, IPSTATS_MIB_OUTMCAST,
 151                                 skb->len);
 152         }
 153
 154         return NF_HOOK(PF_INET6, NF_INET_POST_ROUTING, skb, NULL, skb->dev,
 155                        ip6_output_finish);
 156 }
 157
 158 static inline int ip6_skb_dst_mtu(struct sk_buff *skb)
 159 {
 160         struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
 161
 162         return (np && np->pmtudisc == IPV6_PMTUDISC_PROBE) ?
 163                skb_dst(skb)->dev->mtu : dst_mtu(skb_dst(skb));
 164 }
 165
 166 int ip6_output(struct sk_buff *skb)
 167 {
 168         struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
 169         if (unlikely(idev->cnf.disable_ipv6)) {
 170                 IP6_INC_STATS(dev_net(skb_dst(skb)->dev), idev,
 171                               IPSTATS_MIB_OUTDISCARDS);
 172                 kfree_skb(skb);
 173                 return 0;
 174         }
 175
 176         if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
 177                                 dst_allfrag(skb_dst(skb)))
 178                 return ip6_fragment(skb, ip6_output2);
 179         else
 180                 return ip6_output2(skb);
 181 }
 182
 183 /*
 184  *      xmit an sk_buff (used by TCP, SCTP and DCCP)
 185  */
 186
 187 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
 188              struct ipv6_txoptions *opt)
 189 {
 190         struct net *net = sock_net(sk);
 191         struct ipv6_pinfo *np = inet6_sk(sk);
 192         struct in6_addr *first_hop = &fl->fl6_dst;
 193         struct dst_entry *dst = skb_dst(skb);
 194         struct ipv6hdr *hdr;
 195         u8  proto = fl->proto;
 196         int seg_len = skb->len;
 197         int hlimit = -1;
 198         int tclass = 0;
 199         u32 mtu;
 200
 201         if (opt) {
 202                 unsigned int head_room;
 203
 204                 /* First: exthdrs may take lots of space (~8K for now)
 205                    MAX_HEADER is not enough.
 206                  */
 207                 head_room = opt->opt_nflen + opt->opt_flen;
 208                 seg_len += head_room;
 209                 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
 210
 211                 if (skb_headroom(skb) < head_room) {
 212                         struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
 213                         if (skb2 == NULL) {
 214                                 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 215                                               IPSTATS_MIB_OUTDISCARDS);
 216                                 kfree_skb(skb);
 217                                 return -ENOBUFS;
 218                         }
 219                         kfree_skb(skb);
 220                         skb = skb2;
 221                         if (sk)
 222                                 skb_set_owner_w(skb, sk);
 223                 }
 224                 if (opt->opt_flen)
 225                         ipv6_push_frag_opts(skb, opt, &proto);
 226                 if (opt->opt_nflen)
 227                         ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
 228         }
 229
 230         skb_push(skb, sizeof(struct ipv6hdr));
 231         skb_reset_network_header(skb);
 232         hdr = ipv6_hdr(skb);
 233
 234         /*
 235          *      Fill in the IPv6 header
 236          */
 237         if (np) {
 238                 tclass = np->tclass;
 239                 hlimit = np->hop_limit;
 240         }
 241         if (hlimit < 0)
 242                 hlimit = ip6_dst_hoplimit(dst);
 243
 244         *(__be32 *)hdr = htonl(0x60000000 | (tclass << 20)) | fl->fl6_flowlabel;
 245
 246         hdr->payload_len = htons(seg_len);
 247         hdr->nexthdr = proto;
 248         hdr->hop_limit = hlimit;
 249
 250         ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
 251         ipv6_addr_copy(&hdr->daddr, first_hop);
 252
 253         skb->priority = sk->sk_priority;
 254         skb->mark = sk->sk_mark;
 255
 256         mtu = dst_mtu(dst);
 257         if ((skb->len <= mtu) || skb->local_df || skb_is_gso(skb)) {
 258                 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
 259                               IPSTATS_MIB_OUT, skb->len);
 260                 return NF_HOOK(PF_INET6, NF_INET_LOCAL_OUT, skb, NULL, dst->dev,
 261                                 dst_output);
 262         }
 263
 264         if (net_ratelimit())
 265                 printk(KERN_DEBUG "IPv6: sending pkt_too_big to self\n");
 266         skb->dev = dst->dev;
 267         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
 268         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
 269         kfree_skb(skb);
 270         return -EMSGSIZE;
 271 }
 272
 273 EXPORT_SYMBOL(ip6_xmit);
 274
 275 /*
 276  *      To avoid extra problems ND packets are send through this
 277  *      routine. It's code duplication but I really want to avoid
 278  *      extra checks since ipv6_build_header is used by TCP (which
 279  *      is for us performance critical)
 280  */
 281
 282 int ip6_nd_hdr(struct sock *sk, struct sk_buff *skb, struct net_device *dev,
 283                const struct in6_addr *saddr, const struct in6_addr *daddr,
 284                int proto, int len)
 285 {
 286         struct ipv6_pinfo *np = inet6_sk(sk);
 287         struct ipv6hdr *hdr;
 288         int totlen;
 289
 290         skb->protocol = htons(ETH_P_IPV6);
 291         skb->dev = dev;
 292
 293         totlen = len + sizeof(struct ipv6hdr);
 294
 295         skb_reset_network_header(skb);
 296         skb_put(skb, sizeof(struct ipv6hdr));
 297         hdr = ipv6_hdr(skb);
 298
 299         *(__be32*)hdr = htonl(0x60000000);
 300
 301         hdr->payload_len = htons(len);
 302         hdr->nexthdr = proto;
 303         hdr->hop_limit = np->hop_limit;
 304
 305         ipv6_addr_copy(&hdr->saddr, saddr);
 306         ipv6_addr_copy(&hdr->daddr, daddr);
 307
 308         return 0;
 309 }
 310
 311 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
 312 {
 313         struct ip6_ra_chain *ra;
 314         struct sock *last = NULL;
 315
 316         read_lock(&ip6_ra_lock);
 317         for (ra = ip6_ra_chain; ra; ra = ra->next) {
 318                 struct sock *sk = ra->sk;
 319                 if (sk && ra->sel == sel &&
 320                     (!sk->sk_bound_dev_if ||
 321                      sk->sk_bound_dev_if == skb->dev->ifindex)) {
 322                         if (last) {
 323                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
 324                                 if (skb2)
 325                                         rawv6_rcv(last, skb2);
 326                         }
 327                         last = sk;
 328                 }
 329         }
 330
 331         if (last) {
 332                 rawv6_rcv(last, skb);
 333                 read_unlock(&ip6_ra_lock);
 334                 return 1;
 335         }
 336         read_unlock(&ip6_ra_lock);
 337         return 0;
 338 }
 339
 340 static int ip6_forward_proxy_check(struct sk_buff *skb)
 341 {
 342         struct ipv6hdr *hdr = ipv6_hdr(skb);
 343         u8 nexthdr = hdr->nexthdr;
 344         int offset;
 345
 346         if (ipv6_ext_hdr(nexthdr)) {
 347                 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr);
 348                 if (offset < 0)
 349                         return 0;
 350         } else
 351                 offset = sizeof(struct ipv6hdr);
 352
 353         if (nexthdr == IPPROTO_ICMPV6) {
 354                 struct icmp6hdr *icmp6;
 355
 356                 if (!pskb_may_pull(skb, (skb_network_header(skb) +
 357                                          offset + 1 - skb->data)))
 358                         return 0;
 359
 360                 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
 361
 362                 switch (icmp6->icmp6_type) {
 363                 case NDISC_ROUTER_SOLICITATION:
 364                 case NDISC_ROUTER_ADVERTISEMENT:
 365                 case NDISC_NEIGHBOUR_SOLICITATION:
 366                 case NDISC_NEIGHBOUR_ADVERTISEMENT:
 367                 case NDISC_REDIRECT:
 368                         /* For reaction involving unicast neighbor discovery
 369                          * message destined to the proxied address, pass it to
 370                          * input function.
 371                          */
 372                         return 1;
 373                 default:
 374                         break;
 375                 }
 376         }
 377
 378         /*
 379          * The proxying router can't forward traffic sent to a link-local
 380          * address, so signal the sender and discard the packet. This
 381          * behavior is clarified by the MIPv6 specification.
 382          */
 383         if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
 384                 dst_link_failure(skb);
 385                 return -1;
 386         }
 387
 388         return 0;
 389 }
 390
 391 static inline int ip6_forward_finish(struct sk_buff *skb)
 392 {
 393         return dst_output(skb);
 394 }
 395
 396 int ip6_forward(struct sk_buff *skb)
 397 {
 398         struct dst_entry *dst = skb_dst(skb);
 399         struct ipv6hdr *hdr = ipv6_hdr(skb);
 400         struct inet6_skb_parm *opt = IP6CB(skb);
 401         struct net *net = dev_net(dst->dev);
 402         u32 mtu;
 403
 404         if (net->ipv6.devconf_all->forwarding == 0)
 405                 goto error;
 406
 407         if (skb_warn_if_lro(skb))
 408                 goto drop;
 409
 410         if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
 411                 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
 412                 goto drop;
 413         }
 414
 415         skb_forward_csum(skb);
 416
 417         /*
 418          *      We DO NOT make any processing on
 419          *      RA packets, pushing them to user level AS IS
 420          *      without ane WARRANTY that application will be able
 421          *      to interpret them. The reason is that we
 422          *      cannot make anything clever here.
 423          *
 424          *      We are not end-node, so that if packet contains
 425          *      AH/ESP, we cannot make anything.
 426          *      Defragmentation also would be mistake, RA packets
 427          *      cannot be fragmented, because there is no warranty
 428          *      that different fragments will go along one path. --ANK
 429          */
 430         if (opt->ra) {
 431                 u8 *ptr = skb_network_header(skb) + opt->ra;
 432                 if (ip6_call_ra_chain(skb, (ptr[2]<<8) + ptr[3]))
 433                         return 0;
 434         }
 435
 436         /*
 437          *      check and decrement ttl
 438          */
 439         if (hdr->hop_limit <= 1) {
 440                 /* Force OUTPUT device used as source address */
 441                 skb->dev = dst->dev;
 442                 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
 443                 IP6_INC_STATS_BH(net,
 444                                  ip6_dst_idev(dst), IPSTATS_MIB_INHDRERRORS);
 445
 446                 kfree_skb(skb);
 447                 return -ETIMEDOUT;
 448         }
 449
 450         /* XXX: idev->cnf.proxy_ndp? */
 451         if (net->ipv6.devconf_all->proxy_ndp &&
 452             pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
 453                 int proxied = ip6_forward_proxy_check(skb);
 454                 if (proxied > 0)
 455                         return ip6_input(skb);
 456                 else if (proxied < 0) {
 457                         IP6_INC_STATS(net, ip6_dst_idev(dst),
 458                                       IPSTATS_MIB_INDISCARDS);
 459                         goto drop;
 460                 }
 461         }
 462
 463         if (!xfrm6_route_forward(skb)) {
 464                 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
 465                 goto drop;
 466         }
 467         dst = skb_dst(skb);
 468
 469         /* IPv6 specs say nothing about it, but it is clear that we cannot
 470            send redirects to source routed frames.
 471            We don't send redirects to frames decapsulated from IPsec.
 472          */
 473         if (skb->dev == dst->dev && dst->neighbour && opt->srcrt == 0 &&
 474             !skb_sec_path(skb)) {
 475                 struct in6_addr *target = NULL;
 476                 struct rt6_info *rt;
 477                 struct neighbour *n = dst->neighbour;
 478
 479                 /*
 480                  *      incoming and outgoing devices are the same
 481                  *      send a redirect.
 482                  */
 483
 484                 rt = (struct rt6_info *) dst;
 485                 if ((rt->rt6i_flags & RTF_GATEWAY))
 486                         target = (struct in6_addr*)&n->primary_key;
 487                 else
 488                         target = &hdr->daddr;
 489
 490                 /* Limit redirects both by destination (here)
 491                    and by source (inside ndisc_send_redirect)
 492                  */
 493                 if (xrlim_allow(dst, 1*HZ))
 494                         ndisc_send_redirect(skb, n, target);
 495         } else {
 496                 int addrtype = ipv6_addr_type(&hdr->saddr);
 497
 498                 /* This check is security critical. */
 499                 if (addrtype == IPV6_ADDR_ANY ||
 500                     addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
 501                         goto error;
 502                 if (addrtype & IPV6_ADDR_LINKLOCAL) {
 503                         icmpv6_send(skb, ICMPV6_DEST_UNREACH,
 504                                     ICMPV6_NOT_NEIGHBOUR, 0);
 505                         goto error;
 506                 }
 507         }
 508
 509         mtu = dst_mtu(dst);
 510         if (mtu < IPV6_MIN_MTU)
 511                 mtu = IPV6_MIN_MTU;
 512
 513         if (skb->len > mtu) {
 514                 /* Again, force OUTPUT device used as source address */
 515                 skb->dev = dst->dev;
 516                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
 517                 IP6_INC_STATS_BH(net,
 518                                  ip6_dst_idev(dst), IPSTATS_MIB_INTOOBIGERRORS);
 519                 IP6_INC_STATS_BH(net,
 520                                  ip6_dst_idev(dst), IPSTATS_MIB_FRAGFAILS);
 521                 kfree_skb(skb);
 522                 return -EMSGSIZE;
 523         }
 524
 525         if (skb_cow(skb, dst->dev->hard_header_len)) {
 526                 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTDISCARDS);
 527                 goto drop;
 528         }
 529
 530         hdr = ipv6_hdr(skb);
 531
 532         /* Mangling hops number delayed to point after skb COW */
 533
 534         hdr->hop_limit--;
 535
 536         IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
 537         return NF_HOOK(PF_INET6, NF_INET_FORWARD, skb, skb->dev, dst->dev,
 538                        ip6_forward_finish);
 539
 540 error:
 541         IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
 542 drop:
 543         kfree_skb(skb);
 544         return -EINVAL;
 545 }
 546
 547 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
 548 {
 549         to->pkt_type = from->pkt_type;
 550         to->priority = from->priority;
 551         to->protocol = from->protocol;
 552         skb_dst_drop(to);
 553         skb_dst_set(to, dst_clone(skb_dst(from)));
 554         to->dev = from->dev;
 555         to->mark = from->mark;
 556
 557 #ifdef CONFIG_NET_SCHED
 558         to->tc_index = from->tc_index;
 559 #endif
 560         nf_copy(to, from);
 561 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
 562     defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
 563         to->nf_trace = from->nf_trace;
 564 #endif
 565         skb_copy_secmark(to, from);
 566 }
 567
 568 int ip6_find_1stfragopt(struct sk_buff *skb, u8 **nexthdr)
 569 {
 570         u16 offset = sizeof(struct ipv6hdr);
 571         struct ipv6_opt_hdr *exthdr =
 572                                 (struct ipv6_opt_hdr *)(ipv6_hdr(skb) + 1);
 573         unsigned int packet_len = skb->tail - skb->network_header;
 574         int found_rhdr = 0;
 575         *nexthdr = &ipv6_hdr(skb)->nexthdr;
 576
 577         while (offset + 1 <= packet_len) {
 578
 579                 switch (**nexthdr) {
 580
 581                 case NEXTHDR_HOP:
 582                         break;
 583                 case NEXTHDR_ROUTING:
 584                         found_rhdr = 1;
 585                         break;
 586                 case NEXTHDR_DEST:
 587 #if defined(CONFIG_IPV6_MIP6) || defined(CONFIG_IPV6_MIP6_MODULE)
 588                         if (ipv6_find_tlv(skb, offset, IPV6_TLV_HAO) >= 0)
 589                                 break;
 590 #endif
 591                         if (found_rhdr)
 592                                 return offset;
 593                         break;
 594                 default :
 595                         return offset;
 596                 }
 597
 598                 offset += ipv6_optlen(exthdr);
 599                 *nexthdr = &exthdr->nexthdr;
 600                 exthdr = (struct ipv6_opt_hdr *)(skb_network_header(skb) +
 601                                                  offset);
 602         }
 603
 604         return offset;
 605 }
 606
 607 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
 608 {
 609         struct sk_buff *frag;
 610         struct rt6_info *rt = (struct rt6_info*)skb_dst(skb);
 611         struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
 612         struct ipv6hdr *tmp_hdr;
 613         struct frag_hdr *fh;
 614         unsigned int mtu, hlen, left, len;
 615         __be32 frag_id = 0;
 616         int ptr, offset = 0, err=0;
 617         u8 *prevhdr, nexthdr = 0;
 618         struct net *net = dev_net(skb_dst(skb)->dev);
 619
 620         hlen = ip6_find_1stfragopt(skb, &prevhdr);
 621         nexthdr = *prevhdr;
 622
 623         mtu = ip6_skb_dst_mtu(skb);
 624
 625         /* We must not fragment if the socket is set to force MTU discovery
 626          * or if the skb it not generated by a local socket.
 627          */
 628         if (!skb->local_df) {
 629                 skb->dev = skb_dst(skb)->dev;
 630                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
 631                 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 632                               IPSTATS_MIB_FRAGFAILS);
 633                 kfree_skb(skb);
 634                 return -EMSGSIZE;
 635         }
 636
 637         if (np && np->frag_size < mtu) {
 638                 if (np->frag_size)
 639                         mtu = np->frag_size;
 640         }
 641         mtu -= hlen + sizeof(struct frag_hdr);
 642
 643         if (skb_has_frags(skb)) {
 644                 int first_len = skb_pagelen(skb);
 645                 int truesizes = 0;
 646
 647                 if (first_len - hlen > mtu ||
 648                     ((first_len - hlen) & 7) ||
 649                     skb_cloned(skb))
 650                         goto slow_path;
 651
 652                 skb_walk_frags(skb, frag) {
 653                         /* Correct geometry. */
 654                         if (frag->len > mtu ||
 655                             ((frag->len & 7) && frag->next) ||
 656                             skb_headroom(frag) < hlen)
 657                             goto slow_path;
 658
 659                         /* Partially cloned skb? */
 660                         if (skb_shared(frag))
 661                                 goto slow_path;
 662
 663                         BUG_ON(frag->sk);
 664                         if (skb->sk) {
 665                                 frag->sk = skb->sk;
 666                                 frag->destructor = sock_wfree;
 667                                 truesizes += frag->truesize;
 668                         }
 669                 }
 670
 671                 err = 0;
 672                 offset = 0;
 673                 frag = skb_shinfo(skb)->frag_list;
 674                 skb_frag_list_init(skb);
 675                 /* BUILD HEADER */
 676
 677                 *prevhdr = NEXTHDR_FRAGMENT;
 678                 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
 679                 if (!tmp_hdr) {
 680                         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 681                                       IPSTATS_MIB_FRAGFAILS);
 682                         return -ENOMEM;
 683                 }
 684
 685                 __skb_pull(skb, hlen);
 686                 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
 687                 __skb_push(skb, hlen);
 688                 skb_reset_network_header(skb);
 689                 memcpy(skb_network_header(skb), tmp_hdr, hlen);
 690
 691                 ipv6_select_ident(fh);
 692                 fh->nexthdr = nexthdr;
 693                 fh->reserved = 0;
 694                 fh->frag_off = htons(IP6_MF);
 695                 frag_id = fh->identification;
 696
 697                 first_len = skb_pagelen(skb);
 698                 skb->data_len = first_len - skb_headlen(skb);
 699                 skb->truesize -= truesizes;
 700                 skb->len = first_len;
 701                 ipv6_hdr(skb)->payload_len = htons(first_len -
 702                                                    sizeof(struct ipv6hdr));
 703
 704                 dst_hold(&rt->u.dst);
 705
 706                 for (;;) {
 707                         /* Prepare header of the next frame,
 708                          * before previous one went down. */
 709                         if (frag) {
 710                                 frag->ip_summed = CHECKSUM_NONE;
 711                                 skb_reset_transport_header(frag);
 712                                 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
 713                                 __skb_push(frag, hlen);
 714                                 skb_reset_network_header(frag);
 715                                 memcpy(skb_network_header(frag), tmp_hdr,
 716                                        hlen);
 717                                 offset += skb->len - hlen - sizeof(struct frag_hdr);
 718                                 fh->nexthdr = nexthdr;
 719                                 fh->reserved = 0;
 720                                 fh->frag_off = htons(offset);
 721                                 if (frag->next != NULL)
 722                                         fh->frag_off |= htons(IP6_MF);
 723                                 fh->identification = frag_id;
 724                                 ipv6_hdr(frag)->payload_len =
 725                                                 htons(frag->len -
 726                                                       sizeof(struct ipv6hdr));
 727                                 ip6_copy_metadata(frag, skb);
 728                         }
 729
 730                         err = output(skb);
 731                         if(!err)
 732                                 IP6_INC_STATS(net, ip6_dst_idev(&rt->u.dst),
 733                                               IPSTATS_MIB_FRAGCREATES);
 734
 735                         if (err || !frag)
 736                                 break;
 737
 738                         skb = frag;
 739                         frag = skb->next;
 740                         skb->next = NULL;
 741                 }
 742
 743                 kfree(tmp_hdr);
 744
 745                 if (err == 0) {
 746                         IP6_INC_STATS(net, ip6_dst_idev(&rt->u.dst),
 747                                       IPSTATS_MIB_FRAGOKS);
 748                         dst_release(&rt->u.dst);
 749                         return 0;
 750                 }
 751
 752                 while (frag) {
 753                         skb = frag->next;
 754                         kfree_skb(frag);
 755                         frag = skb;
 756                 }
 757
 758                 IP6_INC_STATS(net, ip6_dst_idev(&rt->u.dst),
 759                               IPSTATS_MIB_FRAGFAILS);
 760                 dst_release(&rt->u.dst);
 761                 return err;
 762         }
 763
 764 slow_path:
 765         left = skb->len - hlen;         /* Space per frame */
 766         ptr = hlen;                     /* Where to start from */
 767
 768         /*
 769          *      Fragment the datagram.
 770          */
 771
 772         *prevhdr = NEXTHDR_FRAGMENT;
 773
 774         /*
 775          *      Keep copying data until we run out.
 776          */
 777         while(left > 0) {
 778                 len = left;
 779                 /* IF: it doesn't fit, use 'mtu' - the data space left */
 780                 if (len > mtu)
 781                         len = mtu;
 782                 /* IF: we are not sending upto and including the packet end
 783                    then align the next start on an eight byte boundary */
 784                 if (len < left) {
 785                         len &= ~7;
 786                 }
 787                 /*
 788                  *      Allocate buffer.
 789                  */
 790
 791                 if ((frag = alloc_skb(len+hlen+sizeof(struct frag_hdr)+LL_ALLOCATED_SPACE(rt->u.dst.dev), GFP_ATOMIC)) == NULL) {
 792                         NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
 793                         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 794                                       IPSTATS_MIB_FRAGFAILS);
 795                         err = -ENOMEM;
 796                         goto fail;
 797                 }
 798
 799                 /*
 800                  *      Set up data on packet
 801                  */
 802
 803                 ip6_copy_metadata(frag, skb);
 804                 skb_reserve(frag, LL_RESERVED_SPACE(rt->u.dst.dev));
 805                 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
 806                 skb_reset_network_header(frag);
 807                 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
 808                 frag->transport_header = (frag->network_header + hlen +
 809                                           sizeof(struct frag_hdr));
 810
 811                 /*
 812                  *      Charge the memory for the fragment to any owner
 813                  *      it might possess
 814                  */
 815                 if (skb->sk)
 816                         skb_set_owner_w(frag, skb->sk);
 817
 818                 /*
 819                  *      Copy the packet header into the new buffer.
 820                  */
 821                 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
 822
 823                 /*
 824                  *      Build fragment header.
 825                  */
 826                 fh->nexthdr = nexthdr;
 827                 fh->reserved = 0;
 828                 if (!frag_id) {
 829                         ipv6_select_ident(fh);
 830                         frag_id = fh->identification;
 831                 } else
 832                         fh->identification = frag_id;
 833
 834                 /*
 835                  *      Copy a block of the IP datagram.
 836                  */
 837                 if (skb_copy_bits(skb, ptr, skb_transport_header(frag), len))
 838                         BUG();
 839                 left -= len;
 840
 841                 fh->frag_off = htons(offset);
 842                 if (left > 0)
 843                         fh->frag_off |= htons(IP6_MF);
 844                 ipv6_hdr(frag)->payload_len = htons(frag->len -
 845                                                     sizeof(struct ipv6hdr));
 846
 847                 ptr += len;
 848                 offset += len;
 849
 850                 /*
 851                  *      Put this fragment into the sending queue.
 852                  */
 853                 err = output(frag);
 854                 if (err)
 855                         goto fail;
 856
 857                 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 858                               IPSTATS_MIB_FRAGCREATES);
 859         }
 860         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 861                       IPSTATS_MIB_FRAGOKS);
 862         kfree_skb(skb);
 863         return err;
 864
 865 fail:
 866         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 867                       IPSTATS_MIB_FRAGFAILS);
 868         kfree_skb(skb);
 869         return err;
 870 }
 871
 872 static inline int ip6_rt_check(struct rt6key *rt_key,
 873                                struct in6_addr *fl_addr,
 874                                struct in6_addr *addr_cache)
 875 {
 876         return ((rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
 877                 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache)));
 878 }
 879
 880 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
 881                                           struct dst_entry *dst,
 882                                           struct flowi *fl)
 883 {
 884         struct ipv6_pinfo *np = inet6_sk(sk);
 885         struct rt6_info *rt = (struct rt6_info *)dst;
 886
 887         if (!dst)
 888                 goto out;
 889
 890         /* Yes, checking route validity in not connected
 891          * case is not very simple. Take into account,
 892          * that we do not support routing by source, TOS,
 893          * and MSG_DONTROUTE            --ANK (980726)
 894          *
 895          * 1. ip6_rt_check(): If route was host route,
 896          *    check that cached destination is current.
 897          *    If it is network route, we still may
 898          *    check its validity using saved pointer
 899          *    to the last used address: daddr_cache.
 900          *    We do not want to save whole address now,
 901          *    (because main consumer of this service
 902          *    is tcp, which has not this problem),
 903          *    so that the last trick works only on connected
 904          *    sockets.
 905          * 2. oif also should be the same.
 906          */
 907         if (ip6_rt_check(&rt->rt6i_dst, &fl->fl6_dst, np->daddr_cache) ||
 908 #ifdef CONFIG_IPV6_SUBTREES
 909             ip6_rt_check(&rt->rt6i_src, &fl->fl6_src, np->saddr_cache) ||
 910 #endif
 911             (fl->oif && fl->oif != dst->dev->ifindex)) {
 912                 dst_release(dst);
 913                 dst = NULL;
 914         }
 915
 916 out:
 917         return dst;
 918 }
 919
 920 static int ip6_dst_lookup_tail(struct sock *sk,
 921                                struct dst_entry **dst, struct flowi *fl)
 922 {
 923         int err;
 924         struct net *net = sock_net(sk);
 925
 926         if (*dst == NULL)
 927                 *dst = ip6_route_output(net, sk, fl);
 928
 929         if ((err = (*dst)->error))
 930                 goto out_err_release;
 931
 932         if (ipv6_addr_any(&fl->fl6_src)) {
 933                 err = ipv6_dev_get_saddr(net, ip6_dst_idev(*dst)->dev,
 934                                          &fl->fl6_dst,
 935                                          sk ? inet6_sk(sk)->srcprefs : 0,
 936                                          &fl->fl6_src);
 937                 if (err)
 938                         goto out_err_release;
 939         }
 940
 941 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
 942         /*
 943          * Here if the dst entry we've looked up
 944          * has a neighbour entry that is in the INCOMPLETE
 945          * state and the src address from the flow is
 946          * marked as OPTIMISTIC, we release the found
 947          * dst entry and replace it instead with the
 948          * dst entry of the nexthop router
 949          */
 950         if ((*dst)->neighbour && !((*dst)->neighbour->nud_state & NUD_VALID)) {
 951                 struct inet6_ifaddr *ifp;
 952                 struct flowi fl_gw;
 953                 int redirect;
 954
 955                 ifp = ipv6_get_ifaddr(net, &fl->fl6_src,
 956                                       (*dst)->dev, 1);
 957
 958                 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
 959                 if (ifp)
 960                         in6_ifa_put(ifp);
 961
 962                 if (redirect) {
 963                         /*
 964                          * We need to get the dst entry for the
 965                          * default router instead
 966                          */
 967                         dst_release(*dst);
 968                         memcpy(&fl_gw, fl, sizeof(struct flowi));
 969                         memset(&fl_gw.fl6_dst, 0, sizeof(struct in6_addr));
 970                         *dst = ip6_route_output(net, sk, &fl_gw);
 971                         if ((err = (*dst)->error))
 972                                 goto out_err_release;
 973                 }
 974         }
 975 #endif
 976
 977         return 0;
 978
 979 out_err_release:
 980         if (err == -ENETUNREACH)
 981                 IP6_INC_STATS_BH(net, NULL, IPSTATS_MIB_OUTNOROUTES);
 982         dst_release(*dst);
 983         *dst = NULL;
 984         return err;
 985 }
 986
 987 /**
 988  *      ip6_dst_lookup - perform route lookup on flow
 989  *      @sk: socket which provides route info
 990  *      @dst: pointer to dst_entry * for result
 991  *      @fl: flow to lookup
 992  *
 993  *      This function performs a route lookup on the given flow.
 994  *
 995  *      It returns zero on success, or a standard errno code on error.
 996  */
 997 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
 998 {
 999         *dst = NULL;
1000         return ip6_dst_lookup_tail(sk, dst, fl);
1001 }
1002 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1003
1004 /**
1005  *      ip6_sk_dst_lookup - perform socket cached route lookup on flow
1006  *      @sk: socket which provides the dst cache and route info
1007  *      @dst: pointer to dst_entry * for result
1008  *      @fl: flow to lookup
1009  *
1010  *      This function performs a route lookup on the given flow with the
1011  *      possibility of using the cached route in the socket if it is valid.
1012  *      It will take the socket dst lock when operating on the dst cache.
1013  *      As a result, this function can only be used in process context.
1014  *
1015  *      It returns zero on success, or a standard errno code on error.
1016  */
1017 int ip6_sk_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
1018 {
1019         *dst = NULL;
1020         if (sk) {
1021                 *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1022                 *dst = ip6_sk_dst_check(sk, *dst, fl);
1023         }
1024
1025         return ip6_dst_lookup_tail(sk, dst, fl);
1026 }
1027 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup);
1028
1029 static inline int ip6_ufo_append_data(struct sock *sk,
1030                         int getfrag(void *from, char *to, int offset, int len,
1031                         int odd, struct sk_buff *skb),
1032                         void *from, int length, int hh_len, int fragheaderlen,
1033                         int transhdrlen, int mtu,unsigned int flags)
1034
1035 {
1036         struct sk_buff *skb;
1037         int err;
1038
1039         /* There is support for UDP large send offload by network
1040          * device, so create one single skb packet containing complete
1041          * udp datagram
1042          */
1043         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
1044                 skb = sock_alloc_send_skb(sk,
1045                         hh_len + fragheaderlen + transhdrlen + 20,
1046                         (flags & MSG_DONTWAIT), &err);
1047                 if (skb == NULL)
1048                         return -ENOMEM;
1049
1050                 /* reserve space for Hardware header */
1051                 skb_reserve(skb, hh_len);
1052
1053                 /* create space for UDP/IP header */
1054                 skb_put(skb,fragheaderlen + transhdrlen);
1055
1056                 /* initialize network header pointer */
1057                 skb_reset_network_header(skb);
1058
1059                 /* initialize protocol header pointer */
1060                 skb->transport_header = skb->network_header + fragheaderlen;
1061
1062                 skb->ip_summed = CHECKSUM_PARTIAL;
1063                 skb->csum = 0;
1064                 sk->sk_sndmsg_off = 0;
1065         }
1066
1067         err = skb_append_datato_frags(sk,skb, getfrag, from,
1068                                       (length - transhdrlen));
1069         if (!err) {
1070                 struct frag_hdr fhdr;
1071
1072                 /* Specify the length of each IPv6 datagram fragment.
1073                  * It has to be a multiple of 8.
1074                  */
1075                 skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1076                                              sizeof(struct frag_hdr)) & ~7;
1077                 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1078                 ipv6_select_ident(&fhdr);
1079                 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1080                 __skb_queue_tail(&sk->sk_write_queue, skb);
1081
1082                 return 0;
1083         }
1084         /* There is not enough support do UPD LSO,
1085          * so follow normal path
1086          */
1087         kfree_skb(skb);
1088
1089         return err;
1090 }
1091
1092 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1093                                                gfp_t gfp)
1094 {
1095         return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1096 }
1097
1098 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1099                                                 gfp_t gfp)
1100 {
1101         return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1102 }
1103
1104 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1105         int offset, int len, int odd, struct sk_buff *skb),
1106         void *from, int length, int transhdrlen,
1107         int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi *fl,
1108         struct rt6_info *rt, unsigned int flags)
1109 {
1110         struct inet_sock *inet = inet_sk(sk);
1111         struct ipv6_pinfo *np = inet6_sk(sk);
1112         struct sk_buff *skb;
1113         unsigned int maxfraglen, fragheaderlen;
1114         int exthdrlen;
1115         int hh_len;
1116         int mtu;
1117         int copy;
1118         int err;
1119         int offset = 0;
1120         int csummode = CHECKSUM_NONE;
1121
1122         if (flags&MSG_PROBE)
1123                 return 0;
1124         if (skb_queue_empty(&sk->sk_write_queue)) {
1125                 /*
1126                  * setup for corking
1127                  */
1128                 if (opt) {
1129                         if (WARN_ON(np->cork.opt))
1130                                 return -EINVAL;
1131
1132                         np->cork.opt = kmalloc(opt->tot_len, sk->sk_allocation);
1133                         if (unlikely(np->cork.opt == NULL))
1134                                 return -ENOBUFS;
1135
1136                         np->cork.opt->tot_len = opt->tot_len;
1137                         np->cork.opt->opt_flen = opt->opt_flen;
1138                         np->cork.opt->opt_nflen = opt->opt_nflen;
1139
1140                         np->cork.opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1141                                                             sk->sk_allocation);
1142                         if (opt->dst0opt && !np->cork.opt->dst0opt)
1143                                 return -ENOBUFS;
1144
1145                         np->cork.opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1146                                                             sk->sk_allocation);
1147                         if (opt->dst1opt && !np->cork.opt->dst1opt)
1148                                 return -ENOBUFS;
1149
1150                         np->cork.opt->hopopt = ip6_opt_dup(opt->hopopt,
1151                                                            sk->sk_allocation);
1152                         if (opt->hopopt && !np->cork.opt->hopopt)
1153                                 return -ENOBUFS;
1154
1155                         np->cork.opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1156                                                             sk->sk_allocation);
1157                         if (opt->srcrt && !np->cork.opt->srcrt)
1158                                 return -ENOBUFS;
1159
1160                         /* need source address above miyazawa*/
1161                 }
1162                 dst_hold(&rt->u.dst);
1163                 inet->cork.dst = &rt->u.dst;
1164                 inet->cork.fl = *fl;
1165                 np->cork.hop_limit = hlimit;
1166                 np->cork.tclass = tclass;
1167                 mtu = np->pmtudisc == IPV6_PMTUDISC_PROBE ?
1168                       rt->u.dst.dev->mtu : dst_mtu(rt->u.dst.path);
1169                 if (np->frag_size < mtu) {
1170                         if (np->frag_size)
1171                                 mtu = np->frag_size;
1172                 }
1173                 inet->cork.fragsize = mtu;
1174                 if (dst_allfrag(rt->u.dst.path))
1175                         inet->cork.flags |= IPCORK_ALLFRAG;
1176                 inet->cork.length = 0;
1177                 sk->sk_sndmsg_page = NULL;
1178                 sk->sk_sndmsg_off = 0;
1179                 exthdrlen = rt->u.dst.header_len + (opt ? opt->opt_flen : 0) -
1180                             rt->rt6i_nfheader_len;
1181                 length += exthdrlen;
1182                 transhdrlen += exthdrlen;
1183         } else {
1184                 rt = (struct rt6_info *)inet->cork.dst;
1185                 fl = &inet->cork.fl;
1186                 opt = np->cork.opt;
1187                 transhdrlen = 0;
1188                 exthdrlen = 0;
1189                 mtu = inet->cork.fragsize;
1190         }
1191
1192         hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1193
1194         fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1195                         (opt ? opt->opt_nflen : 0);
1196         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
1197
1198         if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1199                 if (inet->cork.length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
1200                         ipv6_local_error(sk, EMSGSIZE, fl, mtu-exthdrlen);
1201                         return -EMSGSIZE;
1202                 }
1203         }
1204
1205         /*
1206          * Let's try using as much space as possible.
1207          * Use MTU if total length of the message fits into the MTU.
1208          * Otherwise, we need to reserve fragment header and
1209          * fragment alignment (= 8-15 octects, in total).
1210          *
1211          * Note that we may need to "move" the data from the tail of
1212          * of the buffer to the new fragment when we split
1213          * the message.
1214          *
1215          * FIXME: It may be fragmented into multiple chunks
1216          *        at once if non-fragmentable extension headers
1217          *        are too large.
1218          * --yoshfuji
1219          */
1220
1221         inet->cork.length += length;
1222         if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
1223             (rt->u.dst.dev->features & NETIF_F_UFO)) {
1224
1225                 err = ip6_ufo_append_data(sk, getfrag, from, length, hh_len,
1226                                           fragheaderlen, transhdrlen, mtu,
1227                                           flags);
1228                 if (err)
1229                         goto error;
1230                 return 0;
1231         }
1232
1233         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1234                 goto alloc_new_skb;
1235
1236         while (length > 0) {
1237                 /* Check if the remaining data fits into current packet. */
1238                 copy = (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1239                 if (copy < length)
1240                         copy = maxfraglen - skb->len;
1241
1242                 if (copy <= 0) {
1243                         char *data;
1244                         unsigned int datalen;
1245                         unsigned int fraglen;
1246                         unsigned int fraggap;
1247                         unsigned int alloclen;
1248                         struct sk_buff *skb_prev;
1249 alloc_new_skb:
1250                         skb_prev = skb;
1251
1252                         /* There's no room in the current skb */
1253                         if (skb_prev)
1254                                 fraggap = skb_prev->len - maxfraglen;
1255                         else
1256                                 fraggap = 0;
1257
1258                         /*
1259                          * If remaining data exceeds the mtu,
1260                          * we know we need more fragment(s).
1261                          */
1262                         datalen = length + fraggap;
1263                         if (datalen > (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1264                                 datalen = maxfraglen - fragheaderlen;
1265
1266                         fraglen = datalen + fragheaderlen;
1267                         if ((flags & MSG_MORE) &&
1268                             !(rt->u.dst.dev->features&NETIF_F_SG))
1269                                 alloclen = mtu;
1270                         else
1271                                 alloclen = datalen + fragheaderlen;
1272
1273                         /*
1274                          * The last fragment gets additional space at tail.
1275                          * Note: we overallocate on fragments with MSG_MODE
1276                          * because we have no idea if we're the last one.
1277                          */
1278                         if (datalen == length + fraggap)
1279                                 alloclen += rt->u.dst.trailer_len;
1280
1281                         /*
1282                          * We just reserve space for fragment header.
1283                          * Note: this may be overallocation if the message
1284                          * (without MSG_MORE) fits into the MTU.
1285                          */
1286                         alloclen += sizeof(struct frag_hdr);
1287
1288                         if (transhdrlen) {
1289                                 skb = sock_alloc_send_skb(sk,
1290                                                 alloclen + hh_len,
1291                                                 (flags & MSG_DONTWAIT), &err);
1292                         } else {
1293                                 skb = NULL;
1294                                 if (atomic_read(&sk->sk_wmem_alloc) <=
1295                                     2 * sk->sk_sndbuf)
1296                                         skb = sock_wmalloc(sk,
1297                                                            alloclen + hh_len, 1,
1298                                                            sk->sk_allocation);
1299                                 if (unlikely(skb == NULL))
1300                                         err = -ENOBUFS;
1301                         }
1302                         if (skb == NULL)
1303                                 goto error;
1304                         /*
1305                          *      Fill in the control structures
1306                          */
1307                         skb->ip_summed = csummode;
1308                         skb->csum = 0;
1309                         /* reserve for fragmentation */
1310                         skb_reserve(skb, hh_len+sizeof(struct frag_hdr));
1311
1312                         /*
1313                          *      Find where to start putting bytes
1314                          */
1315                         data = skb_put(skb, fraglen);
1316                         skb_set_network_header(skb, exthdrlen);
1317                         data += fragheaderlen;
1318                         skb->transport_header = (skb->network_header +
1319                                                  fragheaderlen);
1320                         if (fraggap) {
1321                                 skb->csum = skb_copy_and_csum_bits(
1322                                         skb_prev, maxfraglen,
1323                                         data + transhdrlen, fraggap, 0);
1324                                 skb_prev->csum = csum_sub(skb_prev->csum,
1325                                                           skb->csum);
1326                                 data += fraggap;
1327                                 pskb_trim_unique(skb_prev, maxfraglen);
1328                         }
1329                         copy = datalen - transhdrlen - fraggap;
1330                         if (copy < 0) {
1331                                 err = -EINVAL;
1332                                 kfree_skb(skb);
1333                                 goto error;
1334                         } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1335                                 err = -EFAULT;
1336                                 kfree_skb(skb);
1337                                 goto error;
1338                         }
1339
1340                         offset += copy;
1341                         length -= datalen - fraggap;
1342                         transhdrlen = 0;
1343                         exthdrlen = 0;
1344                         csummode = CHECKSUM_NONE;
1345
1346                         /*
1347                          * Put the packet on the pending queue
1348                          */
1349                         __skb_queue_tail(&sk->sk_write_queue, skb);
1350                         continue;
1351                 }
1352
1353                 if (copy > length)
1354                         copy = length;
1355
1356                 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
1357                         unsigned int off;
1358
1359                         off = skb->len;
1360                         if (getfrag(from, skb_put(skb, copy),
1361                                                 offset, copy, off, skb) < 0) {
1362                                 __skb_trim(skb, off);
1363                                 err = -EFAULT;
1364                                 goto error;
1365                         }
1366                 } else {
1367                         int i = skb_shinfo(skb)->nr_frags;
1368                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1369                         struct page *page = sk->sk_sndmsg_page;
1370                         int off = sk->sk_sndmsg_off;
1371                         unsigned int left;
1372
1373                         if (page && (left = PAGE_SIZE - off) > 0) {
1374                                 if (copy >= left)
1375                                         copy = left;
1376                                 if (page != frag->page) {
1377                                         if (i == MAX_SKB_FRAGS) {
1378                                                 err = -EMSGSIZE;
1379                                                 goto error;
1380                                         }
1381                                         get_page(page);
1382                                         skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1383                                         frag = &skb_shinfo(skb)->frags[i];
1384                                 }
1385                         } else if(i < MAX_SKB_FRAGS) {
1386                                 if (copy > PAGE_SIZE)
1387                                         copy = PAGE_SIZE;
1388                                 page = alloc_pages(sk->sk_allocation, 0);
1389                                 if (page == NULL) {
1390                                         err = -ENOMEM;
1391                                         goto error;
1392                                 }
1393                                 sk->sk_sndmsg_page = page;
1394                                 sk->sk_sndmsg_off = 0;
1395
1396                                 skb_fill_page_desc(skb, i, page, 0, 0);
1397                                 frag = &skb_shinfo(skb)->frags[i];
1398                         } else {
1399                                 err = -EMSGSIZE;
1400                                 goto error;
1401                         }
1402                         if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1403                                 err = -EFAULT;
1404                                 goto error;
1405                         }
1406                         sk->sk_sndmsg_off += copy;
1407                         frag->size += copy;
1408                         skb->len += copy;
1409                         skb->data_len += copy;
1410                         skb->truesize += copy;
1411                         atomic_add(copy, &sk->sk_wmem_alloc);
1412                 }
1413                 offset += copy;
1414                 length -= copy;
1415         }
1416         return 0;
1417 error:
1418         inet->cork.length -= length;
1419         IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1420         return err;
1421 }
1422
1423 static void ip6_cork_release(struct inet_sock *inet, struct ipv6_pinfo *np)
1424 {
1425         if (np->cork.opt) {
1426                 kfree(np->cork.opt->dst0opt);
1427                 kfree(np->cork.opt->dst1opt);
1428                 kfree(np->cork.opt->hopopt);
1429                 kfree(np->cork.opt->srcrt);
1430                 kfree(np->cork.opt);
1431                 np->cork.opt = NULL;
1432         }
1433
1434         if (inet->cork.dst) {
1435                 dst_release(inet->cork.dst);
1436                 inet->cork.dst = NULL;
1437                 inet->cork.flags &= ~IPCORK_ALLFRAG;
1438         }
1439         memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1440 }
1441
1442 int ip6_push_pending_frames(struct sock *sk)
1443 {
1444         struct sk_buff *skb, *tmp_skb;
1445         struct sk_buff **tail_skb;
1446         struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1447         struct inet_sock *inet = inet_sk(sk);
1448         struct ipv6_pinfo *np = inet6_sk(sk);
1449         struct net *net = sock_net(sk);
1450         struct ipv6hdr *hdr;
1451         struct ipv6_txoptions *opt = np->cork.opt;
1452         struct rt6_info *rt = (struct rt6_info *)inet->cork.dst;
1453         struct flowi *fl = &inet->cork.fl;
1454         unsigned char proto = fl->proto;
1455         int err = 0;
1456
1457         if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1458                 goto out;
1459         tail_skb = &(skb_shinfo(skb)->frag_list);
1460
1461         /* move skb->data to ip header from ext header */
1462         if (skb->data < skb_network_header(skb))
1463                 __skb_pull(skb, skb_network_offset(skb));
1464         while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1465                 __skb_pull(tmp_skb, skb_network_header_len(skb));
1466                 *tail_skb = tmp_skb;
1467                 tail_skb = &(tmp_skb->next);
1468                 skb->len += tmp_skb->len;
1469                 skb->data_len += tmp_skb->len;
1470                 skb->truesize += tmp_skb->truesize;
1471                 tmp_skb->destructor = NULL;
1472                 tmp_skb->sk = NULL;
1473         }
1474
1475         /* Allow local fragmentation. */
1476         if (np->pmtudisc < IPV6_PMTUDISC_DO)
1477                 skb->local_df = 1;
1478
1479         ipv6_addr_copy(final_dst, &fl->fl6_dst);
1480         __skb_pull(skb, skb_network_header_len(skb));
1481         if (opt && opt->opt_flen)
1482                 ipv6_push_frag_opts(skb, opt, &proto);
1483         if (opt && opt->opt_nflen)
1484                 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1485
1486         skb_push(skb, sizeof(struct ipv6hdr));
1487         skb_reset_network_header(skb);
1488         hdr = ipv6_hdr(skb);
1489
1490         *(__be32*)hdr = fl->fl6_flowlabel |
1491                      htonl(0x60000000 | ((int)np->cork.tclass << 20));
1492
1493         hdr->hop_limit = np->cork.hop_limit;
1494         hdr->nexthdr = proto;
1495         ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
1496         ipv6_addr_copy(&hdr->daddr, final_dst);
1497
1498         skb->priority = sk->sk_priority;
1499         skb->mark = sk->sk_mark;
1500
1501         skb_dst_set(skb, dst_clone(&rt->u.dst));
1502         IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1503         if (proto == IPPROTO_ICMPV6) {
1504                 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1505
1506                 ICMP6MSGOUT_INC_STATS_BH(net, idev, icmp6_hdr(skb)->icmp6_type);
1507                 ICMP6_INC_STATS_BH(net, idev, ICMP6_MIB_OUTMSGS);
1508         }
1509
1510         err = ip6_local_out(skb);
1511         if (err) {
1512                 if (err > 0)
1513                         err = net_xmit_errno(err);
1514                 if (err)
1515                         goto error;
1516         }
1517
1518 out:
1519         ip6_cork_release(inet, np);
1520         return err;
1521 error:
1522         IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1523         goto out;
1524 }
1525
1526 void ip6_flush_pending_frames(struct sock *sk)
1527 {
1528         struct sk_buff *skb;
1529
1530         while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1531                 if (skb_dst(skb))
1532                         IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1533                                       IPSTATS_MIB_OUTDISCARDS);
1534                 kfree_skb(skb);
1535         }
1536
1537         ip6_cork_release(inet_sk(sk), inet6_sk(sk));
1538 }