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  *      $Id: ip6_output.c,v 1.34 2002/02/01 22:01:04 davem Exp $
   9  *
  10  *      Based on linux/net/ipv4/ip_output.c
  11  *
  12  *      This program is free software; you can redistribute it and/or
  13  *      modify it under the terms of the GNU General Public License
  14  *      as published by the Free Software Foundation; either version
  15  *      2 of the License, or (at your option) any later version.
  16  *
  17  *      Changes:
  18  *      A.N.Kuznetsov   :       airthmetics in fragmentation.
  19  *                              extension headers are implemented.
  20  *                              route changes now work.
  21  *                              ip6_forward does not confuse sniffers.
  22  *                              etc.
  23  *
  24  *      H. von Brand    :       Added missing #include <linux/string.h>
  25  *      Imran Patel     :       frag id should be in NBO
  26  *      Kazunori MIYAZAWA @USAGI
  27  *                      :       add ip6_append_data and related functions
  28  *                              for datagram xmit
  29  */
  30
  31 #include <linux/errno.h>
  32 #include <linux/types.h>
  33 #include <linux/string.h>
  34 #include <linux/socket.h>
  35 #include <linux/net.h>
  36 #include <linux/netdevice.h>
  37 #include <linux/if_arp.h>
  38 #include <linux/in6.h>
  39 #include <linux/tcp.h>
  40 #include <linux/route.h>
  41 #include <linux/module.h>
  42
  43 #include <linux/netfilter.h>
  44 #include <linux/netfilter_ipv6.h>
  45
  46 #include <net/sock.h>
  47 #include <net/snmp.h>
  48
  49 #include <net/ipv6.h>
  50 #include <net/ndisc.h>
  51 #include <net/protocol.h>
  52 #include <net/ip6_route.h>
  53 #include <net/addrconf.h>
  54 #include <net/rawv6.h>
  55 #include <net/icmp.h>
  56 #include <net/xfrm.h>
  57 #include <net/checksum.h>
  58
  59 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *));
  60
  61 static __inline__ void ipv6_select_ident(struct sk_buff *skb, struct frag_hdr *fhdr)
  62 {
  63         static u32 ipv6_fragmentation_id = 1;
  64         static DEFINE_SPINLOCK(ip6_id_lock);
  65
  66         spin_lock_bh(&ip6_id_lock);
  67         fhdr->identification = htonl(ipv6_fragmentation_id);
  68         if (++ipv6_fragmentation_id == 0)
  69                 ipv6_fragmentation_id = 1;
  70         spin_unlock_bh(&ip6_id_lock);
  71 }
  72
  73 static inline int ip6_output_finish(struct sk_buff *skb)
  74 {
  75
  76         struct dst_entry *dst = skb->dst;
  77         struct hh_cache *hh = dst->hh;
  78
  79         if (hh) {
  80                 int hh_alen;
  81
  82                 read_lock_bh(&hh->hh_lock);
  83                 hh_alen = HH_DATA_ALIGN(hh->hh_len);
  84                 memcpy(skb->data - hh_alen, hh->hh_data, hh_alen);
  85                 read_unlock_bh(&hh->hh_lock);
  86                 skb_push(skb, hh->hh_len);
  87                 return hh->hh_output(skb);
  88         } else if (dst->neighbour)
  89                 return dst->neighbour->output(skb);
  90
  91         IP6_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
  92         kfree_skb(skb);
  93         return -EINVAL;
  94
  95 }
  96
  97 /* dev_loopback_xmit for use with netfilter. */
  98 static int ip6_dev_loopback_xmit(struct sk_buff *newskb)
  99 {
 100         newskb->mac.raw = newskb->data;
 101         __skb_pull(newskb, newskb->nh.raw - newskb->data);
 102         newskb->pkt_type = PACKET_LOOPBACK;
 103         newskb->ip_summed = CHECKSUM_UNNECESSARY;
 104         BUG_TRAP(newskb->dst);
 105
 106         netif_rx(newskb);
 107         return 0;
 108 }
 109
 110
 111 static int ip6_output2(struct sk_buff *skb)
 112 {
 113         struct dst_entry *dst = skb->dst;
 114         struct net_device *dev = dst->dev;
 115
 116         skb->protocol = htons(ETH_P_IPV6);
 117         skb->dev = dev;
 118
 119         if (ipv6_addr_is_multicast(&skb->nh.ipv6h->daddr)) {
 120                 struct ipv6_pinfo* np = skb->sk ? inet6_sk(skb->sk) : NULL;
 121
 122                 if (!(dev->flags & IFF_LOOPBACK) && (!np || np->mc_loop) &&
 123                     ipv6_chk_mcast_addr(dev, &skb->nh.ipv6h->daddr,
 124                                 &skb->nh.ipv6h->saddr)) {
 125                         struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
 126
 127                         /* Do not check for IFF_ALLMULTI; multicast routing
 128                            is not supported in any case.
 129                          */
 130                         if (newskb)
 131                                 NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, newskb, NULL,
 132                                         newskb->dev,
 133                                         ip6_dev_loopback_xmit);
 134
 135                         if (skb->nh.ipv6h->hop_limit == 0) {
 136                                 IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
 137                                 kfree_skb(skb);
 138                                 return 0;
 139                         }
 140                 }
 141
 142                 IP6_INC_STATS(IPSTATS_MIB_OUTMCASTPKTS);
 143         }
 144
 145         return NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, skb,NULL, skb->dev,ip6_output_finish);
 146 }
 147
 148 int ip6_output(struct sk_buff *skb)
 149 {
 150         if ((skb->len > dst_mtu(skb->dst) && !skb_is_gso(skb)) ||
 151                                 dst_allfrag(skb->dst))
 152                 return ip6_fragment(skb, ip6_output2);
 153         else
 154                 return ip6_output2(skb);
 155 }
 156
 157 /*
 158  *      xmit an sk_buff (used by TCP)
 159  */
 160
 161 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
 162              struct ipv6_txoptions *opt, int ipfragok)
 163 {
 164         struct ipv6_pinfo *np = inet6_sk(sk);
 165         struct in6_addr *first_hop = &fl->fl6_dst;
 166         struct dst_entry *dst = skb->dst;
 167         struct ipv6hdr *hdr;
 168         u8  proto = fl->proto;
 169         int seg_len = skb->len;
 170         int hlimit, tclass;
 171         u32 mtu;
 172
 173         if (opt) {
 174                 int head_room;
 175
 176                 /* First: exthdrs may take lots of space (~8K for now)
 177                    MAX_HEADER is not enough.
 178                  */
 179                 head_room = opt->opt_nflen + opt->opt_flen;
 180                 seg_len += head_room;
 181                 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
 182
 183                 if (skb_headroom(skb) < head_room) {
 184                         struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
 185                         kfree_skb(skb);
 186                         skb = skb2;
 187                         if (skb == NULL) {
 188                                 IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
 189                                 return -ENOBUFS;
 190                         }
 191                         if (sk)
 192                                 skb_set_owner_w(skb, sk);
 193                 }
 194                 if (opt->opt_flen)
 195                         ipv6_push_frag_opts(skb, opt, &proto);
 196                 if (opt->opt_nflen)
 197                         ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
 198         }
 199
 200         hdr = skb->nh.ipv6h = (struct ipv6hdr*)skb_push(skb, sizeof(struct ipv6hdr));
 201
 202         /*
 203          *      Fill in the IPv6 header
 204          */
 205
 206         hlimit = -1;
 207         if (np)
 208                 hlimit = np->hop_limit;
 209         if (hlimit < 0)
 210                 hlimit = dst_metric(dst, RTAX_HOPLIMIT);
 211         if (hlimit < 0)
 212                 hlimit = ipv6_get_hoplimit(dst->dev);
 213
 214         tclass = -1;
 215         if (np)
 216                 tclass = np->tclass;
 217         if (tclass < 0)
 218                 tclass = 0;
 219
 220         *(u32 *)hdr = htonl(0x60000000 | (tclass << 20)) | fl->fl6_flowlabel;
 221
 222         hdr->payload_len = htons(seg_len);
 223         hdr->nexthdr = proto;
 224         hdr->hop_limit = hlimit;
 225
 226         ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
 227         ipv6_addr_copy(&hdr->daddr, first_hop);
 228
 229         skb->priority = sk->sk_priority;
 230
 231         mtu = dst_mtu(dst);
 232         if ((skb->len <= mtu) || ipfragok || skb_is_gso(skb)) {
 233                 IP6_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
 234                 return NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, dst->dev,
 235                                 dst_output);
 236         }
 237
 238         if (net_ratelimit())
 239                 printk(KERN_DEBUG "IPv6: sending pkt_too_big to self\n");
 240         skb->dev = dst->dev;
 241         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev);
 242         IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
 243         kfree_skb(skb);
 244         return -EMSGSIZE;
 245 }
 246
 247 /*
 248  *      To avoid extra problems ND packets are send through this
 249  *      routine. It's code duplication but I really want to avoid
 250  *      extra checks since ipv6_build_header is used by TCP (which
 251  *      is for us performance critical)
 252  */
 253
 254 int ip6_nd_hdr(struct sock *sk, struct sk_buff *skb, struct net_device *dev,
 255                struct in6_addr *saddr, struct in6_addr *daddr,
 256                int proto, int len)
 257 {
 258         struct ipv6_pinfo *np = inet6_sk(sk);
 259         struct ipv6hdr *hdr;
 260         int totlen;
 261
 262         skb->protocol = htons(ETH_P_IPV6);
 263         skb->dev = dev;
 264
 265         totlen = len + sizeof(struct ipv6hdr);
 266
 267         hdr = (struct ipv6hdr *) skb_put(skb, sizeof(struct ipv6hdr));
 268         skb->nh.ipv6h = hdr;
 269
 270         *(u32*)hdr = htonl(0x60000000);
 271
 272         hdr->payload_len = htons(len);
 273         hdr->nexthdr = proto;
 274         hdr->hop_limit = np->hop_limit;
 275
 276         ipv6_addr_copy(&hdr->saddr, saddr);
 277         ipv6_addr_copy(&hdr->daddr, daddr);
 278
 279         return 0;
 280 }
 281
 282 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
 283 {
 284         struct ip6_ra_chain *ra;
 285         struct sock *last = NULL;
 286
 287         read_lock(&ip6_ra_lock);
 288         for (ra = ip6_ra_chain; ra; ra = ra->next) {
 289                 struct sock *sk = ra->sk;
 290                 if (sk && ra->sel == sel &&
 291                     (!sk->sk_bound_dev_if ||
 292                      sk->sk_bound_dev_if == skb->dev->ifindex)) {
 293                         if (last) {
 294                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
 295                                 if (skb2)
 296                                         rawv6_rcv(last, skb2);
 297                         }
 298                         last = sk;
 299                 }
 300         }
 301
 302         if (last) {
 303                 rawv6_rcv(last, skb);
 304                 read_unlock(&ip6_ra_lock);
 305                 return 1;
 306         }
 307         read_unlock(&ip6_ra_lock);
 308         return 0;
 309 }
 310
 311 static int ip6_forward_proxy_check(struct sk_buff *skb)
 312 {
 313         struct ipv6hdr *hdr = skb->nh.ipv6h;
 314         u8 nexthdr = hdr->nexthdr;
 315         int offset;
 316
 317         if (ipv6_ext_hdr(nexthdr)) {
 318                 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr);
 319                 if (offset < 0)
 320                         return 0;
 321         } else
 322                 offset = sizeof(struct ipv6hdr);
 323
 324         if (nexthdr == IPPROTO_ICMPV6) {
 325                 struct icmp6hdr *icmp6;
 326
 327                 if (!pskb_may_pull(skb, skb->nh.raw + offset + 1 - skb->data))
 328                         return 0;
 329
 330                 icmp6 = (struct icmp6hdr *)(skb->nh.raw + offset);
 331
 332                 switch (icmp6->icmp6_type) {
 333                 case NDISC_ROUTER_SOLICITATION:
 334                 case NDISC_ROUTER_ADVERTISEMENT:
 335                 case NDISC_NEIGHBOUR_SOLICITATION:
 336                 case NDISC_NEIGHBOUR_ADVERTISEMENT:
 337                 case NDISC_REDIRECT:
 338                         /* For reaction involving unicast neighbor discovery
 339                          * message destined to the proxied address, pass it to
 340                          * input function.
 341                          */
 342                         return 1;
 343                 default:
 344                         break;
 345                 }
 346         }
 347
 348         /*
 349          * The proxying router can't forward traffic sent to a link-local
 350          * address, so signal the sender and discard the packet. This
 351          * behavior is clarified by the MIPv6 specification.
 352          */
 353         if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
 354                 dst_link_failure(skb);
 355                 return -1;
 356         }
 357
 358         return 0;
 359 }
 360
 361 static inline int ip6_forward_finish(struct sk_buff *skb)
 362 {
 363         return dst_output(skb);
 364 }
 365
 366 int ip6_forward(struct sk_buff *skb)
 367 {
 368         struct dst_entry *dst = skb->dst;
 369         struct ipv6hdr *hdr = skb->nh.ipv6h;
 370         struct inet6_skb_parm *opt = IP6CB(skb);
 371
 372         if (ipv6_devconf.forwarding == 0)
 373                 goto error;
 374
 375         if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
 376                 IP6_INC_STATS(IPSTATS_MIB_INDISCARDS);
 377                 goto drop;
 378         }
 379
 380         skb->ip_summed = CHECKSUM_NONE;
 381
 382         /*
 383          *      We DO NOT make any processing on
 384          *      RA packets, pushing them to user level AS IS
 385          *      without ane WARRANTY that application will be able
 386          *      to interpret them. The reason is that we
 387          *      cannot make anything clever here.
 388          *
 389          *      We are not end-node, so that if packet contains
 390          *      AH/ESP, we cannot make anything.
 391          *      Defragmentation also would be mistake, RA packets
 392          *      cannot be fragmented, because there is no warranty
 393          *      that different fragments will go along one path. --ANK
 394          */
 395         if (opt->ra) {
 396                 u8 *ptr = skb->nh.raw + opt->ra;
 397                 if (ip6_call_ra_chain(skb, (ptr[2]<<8) + ptr[3]))
 398                         return 0;
 399         }
 400
 401         /*
 402          *      check and decrement ttl
 403          */
 404         if (hdr->hop_limit <= 1) {
 405                 /* Force OUTPUT device used as source address */
 406                 skb->dev = dst->dev;
 407                 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT,
 408                             0, skb->dev);
 409                 IP6_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS);
 410
 411                 kfree_skb(skb);
 412                 return -ETIMEDOUT;
 413         }
 414
 415         if (pneigh_lookup(&nd_tbl, &hdr->daddr, skb->dev, 0)) {
 416                 int proxied = ip6_forward_proxy_check(skb);
 417                 if (proxied > 0)
 418                         return ip6_input(skb);
 419                 else if (proxied < 0) {
 420                         IP6_INC_STATS(IPSTATS_MIB_INDISCARDS);
 421                         goto drop;
 422                 }
 423         }
 424
 425         if (!xfrm6_route_forward(skb)) {
 426                 IP6_INC_STATS(IPSTATS_MIB_INDISCARDS);
 427                 goto drop;
 428         }
 429         dst = skb->dst;
 430
 431         /* IPv6 specs say nothing about it, but it is clear that we cannot
 432            send redirects to source routed frames.
 433          */
 434         if (skb->dev == dst->dev && dst->neighbour && opt->srcrt == 0) {
 435                 struct in6_addr *target = NULL;
 436                 struct rt6_info *rt;
 437                 struct neighbour *n = dst->neighbour;
 438
 439                 /*
 440                  *      incoming and outgoing devices are the same
 441                  *      send a redirect.
 442                  */
 443
 444                 rt = (struct rt6_info *) dst;
 445                 if ((rt->rt6i_flags & RTF_GATEWAY))
 446                         target = (struct in6_addr*)&n->primary_key;
 447                 else
 448                         target = &hdr->daddr;
 449
 450                 /* Limit redirects both by destination (here)
 451                    and by source (inside ndisc_send_redirect)
 452                  */
 453                 if (xrlim_allow(dst, 1*HZ))
 454                         ndisc_send_redirect(skb, n, target);
 455         } else if (ipv6_addr_type(&hdr->saddr)&(IPV6_ADDR_MULTICAST|IPV6_ADDR_LOOPBACK
 456                                                 |IPV6_ADDR_LINKLOCAL)) {
 457                 /* This check is security critical. */
 458                 goto error;
 459         }
 460
 461         if (skb->len > dst_mtu(dst)) {
 462                 /* Again, force OUTPUT device used as source address */
 463                 skb->dev = dst->dev;
 464                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, dst_mtu(dst), skb->dev);
 465                 IP6_INC_STATS_BH(IPSTATS_MIB_INTOOBIGERRORS);
 466                 IP6_INC_STATS_BH(IPSTATS_MIB_FRAGFAILS);
 467                 kfree_skb(skb);
 468                 return -EMSGSIZE;
 469         }
 470
 471         if (skb_cow(skb, dst->dev->hard_header_len)) {
 472                 IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
 473                 goto drop;
 474         }
 475
 476         hdr = skb->nh.ipv6h;
 477
 478         /* Mangling hops number delayed to point after skb COW */
 479
 480         hdr->hop_limit--;
 481
 482         IP6_INC_STATS_BH(IPSTATS_MIB_OUTFORWDATAGRAMS);
 483         return NF_HOOK(PF_INET6,NF_IP6_FORWARD, skb, skb->dev, dst->dev, ip6_forward_finish);
 484
 485 error:
 486         IP6_INC_STATS_BH(IPSTATS_MIB_INADDRERRORS);
 487 drop:
 488         kfree_skb(skb);
 489         return -EINVAL;
 490 }
 491
 492 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
 493 {
 494         to->pkt_type = from->pkt_type;
 495         to->priority = from->priority;
 496         to->protocol = from->protocol;
 497         dst_release(to->dst);
 498         to->dst = dst_clone(from->dst);
 499         to->dev = from->dev;
 500
 501 #ifdef CONFIG_NET_SCHED
 502         to->tc_index = from->tc_index;
 503 #endif
 504 #ifdef CONFIG_NETFILTER
 505         to->nfmark = from->nfmark;
 506         /* Connection association is same as pre-frag packet */
 507         nf_conntrack_put(to->nfct);
 508         to->nfct = from->nfct;
 509         nf_conntrack_get(to->nfct);
 510         to->nfctinfo = from->nfctinfo;
 511 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
 512         nf_conntrack_put_reasm(to->nfct_reasm);
 513         to->nfct_reasm = from->nfct_reasm;
 514         nf_conntrack_get_reasm(to->nfct_reasm);
 515 #endif
 516 #ifdef CONFIG_BRIDGE_NETFILTER
 517         nf_bridge_put(to->nf_bridge);
 518         to->nf_bridge = from->nf_bridge;
 519         nf_bridge_get(to->nf_bridge);
 520 #endif
 521 #endif
 522         skb_copy_secmark(to, from);
 523 }
 524
 525 int ip6_find_1stfragopt(struct sk_buff *skb, u8 **nexthdr)
 526 {
 527         u16 offset = sizeof(struct ipv6hdr);
 528         struct ipv6_opt_hdr *exthdr = (struct ipv6_opt_hdr*)(skb->nh.ipv6h + 1);
 529         unsigned int packet_len = skb->tail - skb->nh.raw;
 530         int found_rhdr = 0;
 531         *nexthdr = &skb->nh.ipv6h->nexthdr;
 532
 533         while (offset + 1 <= packet_len) {
 534
 535                 switch (**nexthdr) {
 536
 537                 case NEXTHDR_HOP:
 538                         break;
 539                 case NEXTHDR_ROUTING:
 540                         found_rhdr = 1;
 541                         break;
 542                 case NEXTHDR_DEST:
 543 #ifdef CONFIG_IPV6_MIP6
 544                         if (ipv6_find_tlv(skb, offset, IPV6_TLV_HAO) >= 0)
 545                                 break;
 546 #endif
 547                         if (found_rhdr)
 548                                 return offset;
 549                         break;
 550                 default :
 551                         return offset;
 552                 }
 553
 554                 offset += ipv6_optlen(exthdr);
 555                 *nexthdr = &exthdr->nexthdr;
 556                 exthdr = (struct ipv6_opt_hdr*)(skb->nh.raw + offset);
 557         }
 558
 559         return offset;
 560 }
 561 EXPORT_SYMBOL_GPL(ip6_find_1stfragopt);
 562
 563 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
 564 {
 565         struct net_device *dev;
 566         struct sk_buff *frag;
 567         struct rt6_info *rt = (struct rt6_info*)skb->dst;
 568         struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
 569         struct ipv6hdr *tmp_hdr;
 570         struct frag_hdr *fh;
 571         unsigned int mtu, hlen, left, len;
 572         u32 frag_id = 0;
 573         int ptr, offset = 0, err=0;
 574         u8 *prevhdr, nexthdr = 0;
 575
 576         dev = rt->u.dst.dev;
 577         hlen = ip6_find_1stfragopt(skb, &prevhdr);
 578         nexthdr = *prevhdr;
 579
 580         mtu = dst_mtu(&rt->u.dst);
 581         if (np && np->frag_size < mtu) {
 582                 if (np->frag_size)
 583                         mtu = np->frag_size;
 584         }
 585         mtu -= hlen + sizeof(struct frag_hdr);
 586
 587         if (skb_shinfo(skb)->frag_list) {
 588                 int first_len = skb_pagelen(skb);
 589
 590                 if (first_len - hlen > mtu ||
 591                     ((first_len - hlen) & 7) ||
 592                     skb_cloned(skb))
 593                         goto slow_path;
 594
 595                 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
 596                         /* Correct geometry. */
 597                         if (frag->len > mtu ||
 598                             ((frag->len & 7) && frag->next) ||
 599                             skb_headroom(frag) < hlen)
 600                             goto slow_path;
 601
 602                         /* Partially cloned skb? */
 603                         if (skb_shared(frag))
 604                                 goto slow_path;
 605
 606                         BUG_ON(frag->sk);
 607                         if (skb->sk) {
 608                                 sock_hold(skb->sk);
 609                                 frag->sk = skb->sk;
 610                                 frag->destructor = sock_wfree;
 611                                 skb->truesize -= frag->truesize;
 612                         }
 613                 }
 614
 615                 err = 0;
 616                 offset = 0;
 617                 frag = skb_shinfo(skb)->frag_list;
 618                 skb_shinfo(skb)->frag_list = NULL;
 619                 /* BUILD HEADER */
 620
 621                 tmp_hdr = kmalloc(hlen, GFP_ATOMIC);
 622                 if (!tmp_hdr) {
 623                         IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
 624                         return -ENOMEM;
 625                 }
 626
 627                 *prevhdr = NEXTHDR_FRAGMENT;
 628                 memcpy(tmp_hdr, skb->nh.raw, hlen);
 629                 __skb_pull(skb, hlen);
 630                 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
 631                 skb->nh.raw = __skb_push(skb, hlen);
 632                 memcpy(skb->nh.raw, tmp_hdr, hlen);
 633
 634                 ipv6_select_ident(skb, fh);
 635                 fh->nexthdr = nexthdr;
 636                 fh->reserved = 0;
 637                 fh->frag_off = htons(IP6_MF);
 638                 frag_id = fh->identification;
 639
 640                 first_len = skb_pagelen(skb);
 641                 skb->data_len = first_len - skb_headlen(skb);
 642                 skb->len = first_len;
 643                 skb->nh.ipv6h->payload_len = htons(first_len - sizeof(struct ipv6hdr));
 644
 645
 646                 for (;;) {
 647                         /* Prepare header of the next frame,
 648                          * before previous one went down. */
 649                         if (frag) {
 650                                 frag->ip_summed = CHECKSUM_NONE;
 651                                 frag->h.raw = frag->data;
 652                                 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
 653                                 frag->nh.raw = __skb_push(frag, hlen);
 654                                 memcpy(frag->nh.raw, tmp_hdr, hlen);
 655                                 offset += skb->len - hlen - sizeof(struct frag_hdr);
 656                                 fh->nexthdr = nexthdr;
 657                                 fh->reserved = 0;
 658                                 fh->frag_off = htons(offset);
 659                                 if (frag->next != NULL)
 660                                         fh->frag_off |= htons(IP6_MF);
 661                                 fh->identification = frag_id;
 662                                 frag->nh.ipv6h->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
 663                                 ip6_copy_metadata(frag, skb);
 664                         }
 665
 666                         err = output(skb);
 667                         if(!err)
 668                                 IP6_INC_STATS(IPSTATS_MIB_FRAGCREATES);
 669
 670                         if (err || !frag)
 671                                 break;
 672
 673                         skb = frag;
 674                         frag = skb->next;
 675                         skb->next = NULL;
 676                 }
 677
 678                 kfree(tmp_hdr);
 679
 680                 if (err == 0) {
 681                         IP6_INC_STATS(IPSTATS_MIB_FRAGOKS);
 682                         return 0;
 683                 }
 684
 685                 while (frag) {
 686                         skb = frag->next;
 687                         kfree_skb(frag);
 688                         frag = skb;
 689                 }
 690
 691                 IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
 692                 return err;
 693         }
 694
 695 slow_path:
 696         left = skb->len - hlen;         /* Space per frame */
 697         ptr = hlen;                     /* Where to start from */
 698
 699         /*
 700          *      Fragment the datagram.
 701          */
 702
 703         *prevhdr = NEXTHDR_FRAGMENT;
 704
 705         /*
 706          *      Keep copying data until we run out.
 707          */
 708         while(left > 0) {
 709                 len = left;
 710                 /* IF: it doesn't fit, use 'mtu' - the data space left */
 711                 if (len > mtu)
 712                         len = mtu;
 713                 /* IF: we are not sending upto and including the packet end
 714                    then align the next start on an eight byte boundary */
 715                 if (len < left) {
 716                         len &= ~7;
 717                 }
 718                 /*
 719                  *      Allocate buffer.
 720                  */
 721
 722                 if ((frag = alloc_skb(len+hlen+sizeof(struct frag_hdr)+LL_RESERVED_SPACE(rt->u.dst.dev), GFP_ATOMIC)) == NULL) {
 723                         NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
 724                         IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
 725                         err = -ENOMEM;
 726                         goto fail;
 727                 }
 728
 729                 /*
 730                  *      Set up data on packet
 731                  */
 732
 733                 ip6_copy_metadata(frag, skb);
 734                 skb_reserve(frag, LL_RESERVED_SPACE(rt->u.dst.dev));
 735                 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
 736                 frag->nh.raw = frag->data;
 737                 fh = (struct frag_hdr*)(frag->data + hlen);
 738                 frag->h.raw = frag->data + hlen + sizeof(struct frag_hdr);
 739
 740                 /*
 741                  *      Charge the memory for the fragment to any owner
 742                  *      it might possess
 743                  */
 744                 if (skb->sk)
 745                         skb_set_owner_w(frag, skb->sk);
 746
 747                 /*
 748                  *      Copy the packet header into the new buffer.
 749                  */
 750                 memcpy(frag->nh.raw, skb->data, hlen);
 751
 752                 /*
 753                  *      Build fragment header.
 754                  */
 755                 fh->nexthdr = nexthdr;
 756                 fh->reserved = 0;
 757                 if (!frag_id) {
 758                         ipv6_select_ident(skb, fh);
 759                         frag_id = fh->identification;
 760                 } else
 761                         fh->identification = frag_id;
 762
 763                 /*
 764                  *      Copy a block of the IP datagram.
 765                  */
 766                 if (skb_copy_bits(skb, ptr, frag->h.raw, len))
 767                         BUG();
 768                 left -= len;
 769
 770                 fh->frag_off = htons(offset);
 771                 if (left > 0)
 772                         fh->frag_off |= htons(IP6_MF);
 773                 frag->nh.ipv6h->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
 774
 775                 ptr += len;
 776                 offset += len;
 777
 778                 /*
 779                  *      Put this fragment into the sending queue.
 780                  */
 781                 err = output(frag);
 782                 if (err)
 783                         goto fail;
 784
 785                 IP6_INC_STATS(IPSTATS_MIB_FRAGCREATES);
 786         }
 787         kfree_skb(skb);
 788         IP6_INC_STATS(IPSTATS_MIB_FRAGOKS);
 789         return err;
 790
 791 fail:
 792         kfree_skb(skb);
 793         IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
 794         return err;
 795 }
 796
 797 static inline int ip6_rt_check(struct rt6key *rt_key,
 798                                struct in6_addr *fl_addr,
 799                                struct in6_addr *addr_cache)
 800 {
 801         return ((rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
 802                 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache)));
 803 }
 804
 805 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
 806                                           struct dst_entry *dst,
 807                                           struct flowi *fl)
 808 {
 809         struct ipv6_pinfo *np = inet6_sk(sk);
 810         struct rt6_info *rt = (struct rt6_info *)dst;
 811
 812         if (!dst)
 813                 goto out;
 814
 815         /* Yes, checking route validity in not connected
 816          * case is not very simple. Take into account,
 817          * that we do not support routing by source, TOS,
 818          * and MSG_DONTROUTE            --ANK (980726)
 819          *
 820          * 1. ip6_rt_check(): If route was host route,
 821          *    check that cached destination is current.
 822          *    If it is network route, we still may
 823          *    check its validity using saved pointer
 824          *    to the last used address: daddr_cache.
 825          *    We do not want to save whole address now,
 826          *    (because main consumer of this service
 827          *    is tcp, which has not this problem),
 828          *    so that the last trick works only on connected
 829          *    sockets.
 830          * 2. oif also should be the same.
 831          */
 832         if (ip6_rt_check(&rt->rt6i_dst, &fl->fl6_dst, np->daddr_cache) ||
 833 #ifdef CONFIG_IPV6_SUBTREES
 834             ip6_rt_check(&rt->rt6i_src, &fl->fl6_src, np->saddr_cache) ||
 835 #endif
 836             (fl->oif && fl->oif != dst->dev->ifindex)) {
 837                 dst_release(dst);
 838                 dst = NULL;
 839         }
 840
 841 out:
 842         return dst;
 843 }
 844
 845 static int ip6_dst_lookup_tail(struct sock *sk,
 846                                struct dst_entry **dst, struct flowi *fl)
 847 {
 848         int err;
 849
 850         if (*dst == NULL)
 851                 *dst = ip6_route_output(sk, fl);
 852
 853         if ((err = (*dst)->error))
 854                 goto out_err_release;
 855
 856         if (ipv6_addr_any(&fl->fl6_src)) {
 857                 err = ipv6_get_saddr(*dst, &fl->fl6_dst, &fl->fl6_src);
 858                 if (err)
 859                         goto out_err_release;
 860         }
 861
 862         return 0;
 863
 864 out_err_release:
 865         dst_release(*dst);
 866         *dst = NULL;
 867         return err;
 868 }
 869
 870 /**
 871  *      ip6_dst_lookup - perform route lookup on flow
 872  *      @sk: socket which provides route info
 873  *      @dst: pointer to dst_entry * for result
 874  *      @fl: flow to lookup
 875  *
 876  *      This function performs a route lookup on the given flow.
 877  *
 878  *      It returns zero on success, or a standard errno code on error.
 879  */
 880 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
 881 {
 882         *dst = NULL;
 883         return ip6_dst_lookup_tail(sk, dst, fl);
 884 }
 885 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
 886
 887 /**
 888  *      ip6_sk_dst_lookup - perform socket cached route lookup on flow
 889  *      @sk: socket which provides the dst cache and route info
 890  *      @dst: pointer to dst_entry * for result
 891  *      @fl: flow to lookup
 892  *
 893  *      This function performs a route lookup on the given flow with the
 894  *      possibility of using the cached route in the socket if it is valid.
 895  *      It will take the socket dst lock when operating on the dst cache.
 896  *      As a result, this function can only be used in process context.
 897  *
 898  *      It returns zero on success, or a standard errno code on error.
 899  */
 900 int ip6_sk_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
 901 {
 902         *dst = NULL;
 903         if (sk) {
 904                 *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
 905                 *dst = ip6_sk_dst_check(sk, *dst, fl);
 906         }
 907
 908         return ip6_dst_lookup_tail(sk, dst, fl);
 909 }
 910 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup);
 911
 912 static inline int ip6_ufo_append_data(struct sock *sk,
 913                         int getfrag(void *from, char *to, int offset, int len,
 914                         int odd, struct sk_buff *skb),
 915                         void *from, int length, int hh_len, int fragheaderlen,
 916                         int transhdrlen, int mtu,unsigned int flags)
 917
 918 {
 919         struct sk_buff *skb;
 920         int err;
 921
 922         /* There is support for UDP large send offload by network
 923          * device, so create one single skb packet containing complete
 924          * udp datagram
 925          */
 926         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
 927                 skb = sock_alloc_send_skb(sk,
 928                         hh_len + fragheaderlen + transhdrlen + 20,
 929                         (flags & MSG_DONTWAIT), &err);
 930                 if (skb == NULL)
 931                         return -ENOMEM;
 932
 933                 /* reserve space for Hardware header */
 934                 skb_reserve(skb, hh_len);
 935
 936                 /* create space for UDP/IP header */
 937                 skb_put(skb,fragheaderlen + transhdrlen);
 938
 939                 /* initialize network header pointer */
 940                 skb->nh.raw = skb->data;
 941
 942                 /* initialize protocol header pointer */
 943                 skb->h.raw = skb->data + fragheaderlen;
 944
 945                 skb->ip_summed = CHECKSUM_PARTIAL;
 946                 skb->csum = 0;
 947                 sk->sk_sndmsg_off = 0;
 948         }
 949
 950         err = skb_append_datato_frags(sk,skb, getfrag, from,
 951                                       (length - transhdrlen));
 952         if (!err) {
 953                 struct frag_hdr fhdr;
 954
 955                 /* specify the length of each IP datagram fragment*/
 956                 skb_shinfo(skb)->gso_size = mtu - fragheaderlen -
 957                                             sizeof(struct frag_hdr);
 958                 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
 959                 ipv6_select_ident(skb, &fhdr);
 960                 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
 961                 __skb_queue_tail(&sk->sk_write_queue, skb);
 962
 963                 return 0;
 964         }
 965         /* There is not enough support do UPD LSO,
 966          * so follow normal path
 967          */
 968         kfree_skb(skb);
 969
 970         return err;
 971 }
 972
 973 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
 974         int offset, int len, int odd, struct sk_buff *skb),
 975         void *from, int length, int transhdrlen,
 976         int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi *fl,
 977         struct rt6_info *rt, unsigned int flags)
 978 {
 979         struct inet_sock *inet = inet_sk(sk);
 980         struct ipv6_pinfo *np = inet6_sk(sk);
 981         struct sk_buff *skb;
 982         unsigned int maxfraglen, fragheaderlen;
 983         int exthdrlen;
 984         int hh_len;
 985         int mtu;
 986         int copy;
 987         int err;
 988         int offset = 0;
 989         int csummode = CHECKSUM_NONE;
 990
 991         if (flags&MSG_PROBE)
 992                 return 0;
 993         if (skb_queue_empty(&sk->sk_write_queue)) {
 994                 /*
 995                  * setup for corking
 996                  */
 997                 if (opt) {
 998                         if (np->cork.opt == NULL) {
 999                                 np->cork.opt = kmalloc(opt->tot_len,
1000                                                        sk->sk_allocation);
1001                                 if (unlikely(np->cork.opt == NULL))
1002                                         return -ENOBUFS;
1003                         } else if (np->cork.opt->tot_len < opt->tot_len) {
1004                                 printk(KERN_DEBUG "ip6_append_data: invalid option length\n");
1005                                 return -EINVAL;
1006                         }
1007                         memcpy(np->cork.opt, opt, opt->tot_len);
1008                         inet->cork.flags |= IPCORK_OPT;
1009                         /* need source address above miyazawa*/
1010                 }
1011                 dst_hold(&rt->u.dst);
1012                 np->cork.rt = rt;
1013                 inet->cork.fl = *fl;
1014                 np->cork.hop_limit = hlimit;
1015                 np->cork.tclass = tclass;
1016                 mtu = dst_mtu(rt->u.dst.path);
1017                 if (np->frag_size < mtu) {
1018                         if (np->frag_size)
1019                                 mtu = np->frag_size;
1020                 }
1021                 inet->cork.fragsize = mtu;
1022                 if (dst_allfrag(rt->u.dst.path))
1023                         inet->cork.flags |= IPCORK_ALLFRAG;
1024                 inet->cork.length = 0;
1025                 sk->sk_sndmsg_page = NULL;
1026                 sk->sk_sndmsg_off = 0;
1027                 exthdrlen = rt->u.dst.header_len + (opt ? opt->opt_flen : 0);
1028                 length += exthdrlen;
1029                 transhdrlen += exthdrlen;
1030         } else {
1031                 rt = np->cork.rt;
1032                 fl = &inet->cork.fl;
1033                 if (inet->cork.flags & IPCORK_OPT)
1034                         opt = np->cork.opt;
1035                 transhdrlen = 0;
1036                 exthdrlen = 0;
1037                 mtu = inet->cork.fragsize;
1038         }
1039
1040         hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1041
1042         fragheaderlen = sizeof(struct ipv6hdr) + rt->u.dst.nfheader_len + (opt ? opt->opt_nflen : 0);
1043         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
1044
1045         if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1046                 if (inet->cork.length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
1047                         ipv6_local_error(sk, EMSGSIZE, fl, mtu-exthdrlen);
1048                         return -EMSGSIZE;
1049                 }
1050         }
1051
1052         /*
1053          * Let's try using as much space as possible.
1054          * Use MTU if total length of the message fits into the MTU.
1055          * Otherwise, we need to reserve fragment header and
1056          * fragment alignment (= 8-15 octects, in total).
1057          *
1058          * Note that we may need to "move" the data from the tail of
1059          * of the buffer to the new fragment when we split
1060          * the message.
1061          *
1062          * FIXME: It may be fragmented into multiple chunks
1063          *        at once if non-fragmentable extension headers
1064          *        are too large.
1065          * --yoshfuji
1066          */
1067
1068         inet->cork.length += length;
1069         if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
1070             (rt->u.dst.dev->features & NETIF_F_UFO)) {
1071
1072                 err = ip6_ufo_append_data(sk, getfrag, from, length, hh_len,
1073                                           fragheaderlen, transhdrlen, mtu,
1074                                           flags);
1075                 if (err)
1076                         goto error;
1077                 return 0;
1078         }
1079
1080         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1081                 goto alloc_new_skb;
1082
1083         while (length > 0) {
1084                 /* Check if the remaining data fits into current packet. */
1085                 copy = (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1086                 if (copy < length)
1087                         copy = maxfraglen - skb->len;
1088
1089                 if (copy <= 0) {
1090                         char *data;
1091                         unsigned int datalen;
1092                         unsigned int fraglen;
1093                         unsigned int fraggap;
1094                         unsigned int alloclen;
1095                         struct sk_buff *skb_prev;
1096 alloc_new_skb:
1097                         skb_prev = skb;
1098
1099                         /* There's no room in the current skb */
1100                         if (skb_prev)
1101                                 fraggap = skb_prev->len - maxfraglen;
1102                         else
1103                                 fraggap = 0;
1104
1105                         /*
1106                          * If remaining data exceeds the mtu,
1107                          * we know we need more fragment(s).
1108                          */
1109                         datalen = length + fraggap;
1110                         if (datalen > (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1111                                 datalen = maxfraglen - fragheaderlen;
1112
1113                         fraglen = datalen + fragheaderlen;
1114                         if ((flags & MSG_MORE) &&
1115                             !(rt->u.dst.dev->features&NETIF_F_SG))
1116                                 alloclen = mtu;
1117                         else
1118                                 alloclen = datalen + fragheaderlen;
1119
1120                         /*
1121                          * The last fragment gets additional space at tail.
1122                          * Note: we overallocate on fragments with MSG_MODE
1123                          * because we have no idea if we're the last one.
1124                          */
1125                         if (datalen == length + fraggap)
1126                                 alloclen += rt->u.dst.trailer_len;
1127
1128                         /*
1129                          * We just reserve space for fragment header.
1130                          * Note: this may be overallocation if the message
1131                          * (without MSG_MORE) fits into the MTU.
1132                          */
1133                         alloclen += sizeof(struct frag_hdr);
1134
1135                         if (transhdrlen) {
1136                                 skb = sock_alloc_send_skb(sk,
1137                                                 alloclen + hh_len,
1138                                                 (flags & MSG_DONTWAIT), &err);
1139                         } else {
1140                                 skb = NULL;
1141                                 if (atomic_read(&sk->sk_wmem_alloc) <=
1142                                     2 * sk->sk_sndbuf)
1143                                         skb = sock_wmalloc(sk,
1144                                                            alloclen + hh_len, 1,
1145                                                            sk->sk_allocation);
1146                                 if (unlikely(skb == NULL))
1147                                         err = -ENOBUFS;
1148                         }
1149                         if (skb == NULL)
1150                                 goto error;
1151                         /*
1152                          *      Fill in the control structures
1153                          */
1154                         skb->ip_summed = csummode;
1155                         skb->csum = 0;
1156                         /* reserve for fragmentation */
1157                         skb_reserve(skb, hh_len+sizeof(struct frag_hdr));
1158
1159                         /*
1160                          *      Find where to start putting bytes
1161                          */
1162                         data = skb_put(skb, fraglen);
1163                         skb->nh.raw = data + exthdrlen;
1164                         data += fragheaderlen;
1165                         skb->h.raw = data + exthdrlen;
1166
1167                         if (fraggap) {
1168                                 skb->csum = skb_copy_and_csum_bits(
1169                                         skb_prev, maxfraglen,
1170                                         data + transhdrlen, fraggap, 0);
1171                                 skb_prev->csum = csum_sub(skb_prev->csum,
1172                                                           skb->csum);
1173                                 data += fraggap;
1174                                 pskb_trim_unique(skb_prev, maxfraglen);
1175                         }
1176                         copy = datalen - transhdrlen - fraggap;
1177                         if (copy < 0) {
1178                                 err = -EINVAL;
1179                                 kfree_skb(skb);
1180                                 goto error;
1181                         } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1182                                 err = -EFAULT;
1183                                 kfree_skb(skb);
1184                                 goto error;
1185                         }
1186
1187                         offset += copy;
1188                         length -= datalen - fraggap;
1189                         transhdrlen = 0;
1190                         exthdrlen = 0;
1191                         csummode = CHECKSUM_NONE;
1192
1193                         /*
1194                          * Put the packet on the pending queue
1195                          */
1196                         __skb_queue_tail(&sk->sk_write_queue, skb);
1197                         continue;
1198                 }
1199
1200                 if (copy > length)
1201                         copy = length;
1202
1203                 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
1204                         unsigned int off;
1205
1206                         off = skb->len;
1207                         if (getfrag(from, skb_put(skb, copy),
1208                                                 offset, copy, off, skb) < 0) {
1209                                 __skb_trim(skb, off);
1210                                 err = -EFAULT;
1211                                 goto error;
1212                         }
1213                 } else {
1214                         int i = skb_shinfo(skb)->nr_frags;
1215                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1216                         struct page *page = sk->sk_sndmsg_page;
1217                         int off = sk->sk_sndmsg_off;
1218                         unsigned int left;
1219
1220                         if (page && (left = PAGE_SIZE - off) > 0) {
1221                                 if (copy >= left)
1222                                         copy = left;
1223                                 if (page != frag->page) {
1224                                         if (i == MAX_SKB_FRAGS) {
1225                                                 err = -EMSGSIZE;
1226                                                 goto error;
1227                                         }
1228                                         get_page(page);
1229                                         skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1230                                         frag = &skb_shinfo(skb)->frags[i];
1231                                 }
1232                         } else if(i < MAX_SKB_FRAGS) {
1233                                 if (copy > PAGE_SIZE)
1234                                         copy = PAGE_SIZE;
1235                                 page = alloc_pages(sk->sk_allocation, 0);
1236                                 if (page == NULL) {
1237                                         err = -ENOMEM;
1238                                         goto error;
1239                                 }
1240                                 sk->sk_sndmsg_page = page;
1241                                 sk->sk_sndmsg_off = 0;
1242
1243                                 skb_fill_page_desc(skb, i, page, 0, 0);
1244                                 frag = &skb_shinfo(skb)->frags[i];
1245                                 skb->truesize += PAGE_SIZE;
1246                                 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1247                         } else {
1248                                 err = -EMSGSIZE;
1249                                 goto error;
1250                         }
1251                         if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1252                                 err = -EFAULT;
1253                                 goto error;
1254                         }
1255                         sk->sk_sndmsg_off += copy;
1256                         frag->size += copy;
1257                         skb->len += copy;
1258                         skb->data_len += copy;
1259                 }
1260                 offset += copy;
1261                 length -= copy;
1262         }
1263         return 0;
1264 error:
1265         inet->cork.length -= length;
1266         IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1267         return err;
1268 }
1269
1270 int ip6_push_pending_frames(struct sock *sk)
1271 {
1272         struct sk_buff *skb, *tmp_skb;
1273         struct sk_buff **tail_skb;
1274         struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1275         struct inet_sock *inet = inet_sk(sk);
1276         struct ipv6_pinfo *np = inet6_sk(sk);
1277         struct ipv6hdr *hdr;
1278         struct ipv6_txoptions *opt = np->cork.opt;
1279         struct rt6_info *rt = np->cork.rt;
1280         struct flowi *fl = &inet->cork.fl;
1281         unsigned char proto = fl->proto;
1282         int err = 0;
1283
1284         if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1285                 goto out;
1286         tail_skb = &(skb_shinfo(skb)->frag_list);
1287
1288         /* move skb->data to ip header from ext header */
1289         if (skb->data < skb->nh.raw)
1290                 __skb_pull(skb, skb->nh.raw - skb->data);
1291         while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1292                 __skb_pull(tmp_skb, skb->h.raw - skb->nh.raw);
1293                 *tail_skb = tmp_skb;
1294                 tail_skb = &(tmp_skb->next);
1295                 skb->len += tmp_skb->len;
1296                 skb->data_len += tmp_skb->len;
1297                 skb->truesize += tmp_skb->truesize;
1298                 __sock_put(tmp_skb->sk);
1299                 tmp_skb->destructor = NULL;
1300                 tmp_skb->sk = NULL;
1301         }
1302
1303         ipv6_addr_copy(final_dst, &fl->fl6_dst);
1304         __skb_pull(skb, skb->h.raw - skb->nh.raw);
1305         if (opt && opt->opt_flen)
1306                 ipv6_push_frag_opts(skb, opt, &proto);
1307         if (opt && opt->opt_nflen)
1308                 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1309
1310         skb->nh.ipv6h = hdr = (struct ipv6hdr*) skb_push(skb, sizeof(struct ipv6hdr));
1311
1312         *(u32*)hdr = fl->fl6_flowlabel |
1313                      htonl(0x60000000 | ((int)np->cork.tclass << 20));
1314
1315         if (skb->len <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN)
1316                 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
1317         else
1318                 hdr->payload_len = 0;
1319         hdr->hop_limit = np->cork.hop_limit;
1320         hdr->nexthdr = proto;
1321         ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
1322         ipv6_addr_copy(&hdr->daddr, final_dst);
1323
1324         skb->priority = sk->sk_priority;
1325
1326         skb->dst = dst_clone(&rt->u.dst);
1327         IP6_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
1328         err = NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, skb->dst->dev, dst_output);
1329         if (err) {
1330                 if (err > 0)
1331                         err = np->recverr ? net_xmit_errno(err) : 0;
1332                 if (err)
1333                         goto error;
1334         }
1335
1336 out:
1337         inet->cork.flags &= ~IPCORK_OPT;
1338         kfree(np->cork.opt);
1339         np->cork.opt = NULL;
1340         if (np->cork.rt) {
1341                 dst_release(&np->cork.rt->u.dst);
1342                 np->cork.rt = NULL;
1343                 inet->cork.flags &= ~IPCORK_ALLFRAG;
1344         }
1345         memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1346         return err;
1347 error:
1348         goto out;
1349 }
1350
1351 void ip6_flush_pending_frames(struct sock *sk)
1352 {
1353         struct inet_sock *inet = inet_sk(sk);
1354         struct ipv6_pinfo *np = inet6_sk(sk);
1355         struct sk_buff *skb;
1356
1357         while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1358                 IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1359                 kfree_skb(skb);
1360         }
1361
1362         inet->cork.flags &= ~IPCORK_OPT;
1363
1364         kfree(np->cork.opt);
1365         np->cork.opt = NULL;
1366         if (np->cork.rt) {
1367                 dst_release(&np->cork.rt->u.dst);
1368                 np->cork.rt = NULL;
1369                 inet->cork.flags &= ~IPCORK_ALLFRAG;
1370         }
1371         memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1372 }