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         struct dst_entry *dst = skb->dst;
  76
  77         if (dst->hh)
  78                 return neigh_hh_output(dst->hh, skb);
  79         else if (dst->neighbour)
  80                 return dst->neighbour->output(skb);
  81
  82         IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
  83         kfree_skb(skb);
  84         return -EINVAL;
  85
  86 }
  87
  88 /* dev_loopback_xmit for use with netfilter. */
  89 static int ip6_dev_loopback_xmit(struct sk_buff *newskb)
  90 {
  91         skb_reset_mac_header(newskb);
  92         __skb_pull(newskb, skb_network_offset(newskb));
  93         newskb->pkt_type = PACKET_LOOPBACK;
  94         newskb->ip_summed = CHECKSUM_UNNECESSARY;
  95         BUG_TRAP(newskb->dst);
  96
  97         netif_rx(newskb);
  98         return 0;
  99 }
 100
 101
 102 static int ip6_output2(struct sk_buff *skb)
 103 {
 104         struct dst_entry *dst = skb->dst;
 105         struct net_device *dev = dst->dev;
 106
 107         skb->protocol = htons(ETH_P_IPV6);
 108         skb->dev = dev;
 109
 110         if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
 111                 struct ipv6_pinfo* np = skb->sk ? inet6_sk(skb->sk) : NULL;
 112                 struct inet6_dev *idev = ip6_dst_idev(skb->dst);
 113
 114                 if (!(dev->flags & IFF_LOOPBACK) && (!np || np->mc_loop) &&
 115                     ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
 116                                         &ipv6_hdr(skb)->saddr)) {
 117                         struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
 118
 119                         /* Do not check for IFF_ALLMULTI; multicast routing
 120                            is not supported in any case.
 121                          */
 122                         if (newskb)
 123                                 NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, newskb, NULL,
 124                                         newskb->dev,
 125                                         ip6_dev_loopback_xmit);
 126
 127                         if (ipv6_hdr(skb)->hop_limit == 0) {
 128                                 IP6_INC_STATS(idev, IPSTATS_MIB_OUTDISCARDS);
 129                                 kfree_skb(skb);
 130                                 return 0;
 131                         }
 132                 }
 133
 134                 IP6_INC_STATS(idev, IPSTATS_MIB_OUTMCASTPKTS);
 135         }
 136
 137         return NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, skb,NULL, skb->dev,ip6_output_finish);
 138 }
 139
 140 static inline int ip6_skb_dst_mtu(struct sk_buff *skb)
 141 {
 142         struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
 143
 144         return (np && np->pmtudisc == IPV6_PMTUDISC_PROBE) ?
 145                skb->dst->dev->mtu : dst_mtu(skb->dst);
 146 }
 147
 148 int ip6_output(struct sk_buff *skb)
 149 {
 150         if ((skb->len > ip6_skb_dst_mtu(skb) && !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                         if (skb2 == NULL) {
 186                                 IP6_INC_STATS(ip6_dst_idev(skb->dst),
 187                                               IPSTATS_MIB_OUTDISCARDS);
 188                                 kfree_skb(skb);
 189                                 return -ENOBUFS;
 190                         }
 191                         kfree_skb(skb);
 192                         skb = skb2;
 193                         if (sk)
 194                                 skb_set_owner_w(skb, sk);
 195                 }
 196                 if (opt->opt_flen)
 197                         ipv6_push_frag_opts(skb, opt, &proto);
 198                 if (opt->opt_nflen)
 199                         ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
 200         }
 201
 202         skb_push(skb, sizeof(struct ipv6hdr));
 203         skb_reset_network_header(skb);
 204         hdr = ipv6_hdr(skb);
 205
 206         /*
 207          *      Fill in the IPv6 header
 208          */
 209
 210         hlimit = -1;
 211         if (np)
 212                 hlimit = np->hop_limit;
 213         if (hlimit < 0)
 214                 hlimit = dst_metric(dst, RTAX_HOPLIMIT);
 215         if (hlimit < 0)
 216                 hlimit = ipv6_get_hoplimit(dst->dev);
 217
 218         tclass = -1;
 219         if (np)
 220                 tclass = np->tclass;
 221         if (tclass < 0)
 222                 tclass = 0;
 223
 224         *(__be32 *)hdr = htonl(0x60000000 | (tclass << 20)) | fl->fl6_flowlabel;
 225
 226         hdr->payload_len = htons(seg_len);
 227         hdr->nexthdr = proto;
 228         hdr->hop_limit = hlimit;
 229
 230         ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
 231         ipv6_addr_copy(&hdr->daddr, first_hop);
 232
 233         skb->priority = sk->sk_priority;
 234
 235         mtu = dst_mtu(dst);
 236         if ((skb->len <= mtu) || ipfragok || skb_is_gso(skb)) {
 237                 IP6_INC_STATS(ip6_dst_idev(skb->dst),
 238                               IPSTATS_MIB_OUTREQUESTS);
 239                 return NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, dst->dev,
 240                                 dst_output);
 241         }
 242
 243         if (net_ratelimit())
 244                 printk(KERN_DEBUG "IPv6: sending pkt_too_big to self\n");
 245         skb->dev = dst->dev;
 246         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev);
 247         IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGFAILS);
 248         kfree_skb(skb);
 249         return -EMSGSIZE;
 250 }
 251
 252 EXPORT_SYMBOL(ip6_xmit);
 253
 254 /*
 255  *      To avoid extra problems ND packets are send through this
 256  *      routine. It's code duplication but I really want to avoid
 257  *      extra checks since ipv6_build_header is used by TCP (which
 258  *      is for us performance critical)
 259  */
 260
 261 int ip6_nd_hdr(struct sock *sk, struct sk_buff *skb, struct net_device *dev,
 262                struct in6_addr *saddr, struct in6_addr *daddr,
 263                int proto, int len)
 264 {
 265         struct ipv6_pinfo *np = inet6_sk(sk);
 266         struct ipv6hdr *hdr;
 267         int totlen;
 268
 269         skb->protocol = htons(ETH_P_IPV6);
 270         skb->dev = dev;
 271
 272         totlen = len + sizeof(struct ipv6hdr);
 273
 274         skb_reset_network_header(skb);
 275         skb_put(skb, sizeof(struct ipv6hdr));
 276         hdr = ipv6_hdr(skb);
 277
 278         *(__be32*)hdr = htonl(0x60000000);
 279
 280         hdr->payload_len = htons(len);
 281         hdr->nexthdr = proto;
 282         hdr->hop_limit = np->hop_limit;
 283
 284         ipv6_addr_copy(&hdr->saddr, saddr);
 285         ipv6_addr_copy(&hdr->daddr, daddr);
 286
 287         return 0;
 288 }
 289
 290 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
 291 {
 292         struct ip6_ra_chain *ra;
 293         struct sock *last = NULL;
 294
 295         read_lock(&ip6_ra_lock);
 296         for (ra = ip6_ra_chain; ra; ra = ra->next) {
 297                 struct sock *sk = ra->sk;
 298                 if (sk && ra->sel == sel &&
 299                     (!sk->sk_bound_dev_if ||
 300                      sk->sk_bound_dev_if == skb->dev->ifindex)) {
 301                         if (last) {
 302                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
 303                                 if (skb2)
 304                                         rawv6_rcv(last, skb2);
 305                         }
 306                         last = sk;
 307                 }
 308         }
 309
 310         if (last) {
 311                 rawv6_rcv(last, skb);
 312                 read_unlock(&ip6_ra_lock);
 313                 return 1;
 314         }
 315         read_unlock(&ip6_ra_lock);
 316         return 0;
 317 }
 318
 319 static int ip6_forward_proxy_check(struct sk_buff *skb)
 320 {
 321         struct ipv6hdr *hdr = ipv6_hdr(skb);
 322         u8 nexthdr = hdr->nexthdr;
 323         int offset;
 324
 325         if (ipv6_ext_hdr(nexthdr)) {
 326                 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr);
 327                 if (offset < 0)
 328                         return 0;
 329         } else
 330                 offset = sizeof(struct ipv6hdr);
 331
 332         if (nexthdr == IPPROTO_ICMPV6) {
 333                 struct icmp6hdr *icmp6;
 334
 335                 if (!pskb_may_pull(skb, (skb_network_header(skb) +
 336                                          offset + 1 - skb->data)))
 337                         return 0;
 338
 339                 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
 340
 341                 switch (icmp6->icmp6_type) {
 342                 case NDISC_ROUTER_SOLICITATION:
 343                 case NDISC_ROUTER_ADVERTISEMENT:
 344                 case NDISC_NEIGHBOUR_SOLICITATION:
 345                 case NDISC_NEIGHBOUR_ADVERTISEMENT:
 346                 case NDISC_REDIRECT:
 347                         /* For reaction involving unicast neighbor discovery
 348                          * message destined to the proxied address, pass it to
 349                          * input function.
 350                          */
 351                         return 1;
 352                 default:
 353                         break;
 354                 }
 355         }
 356
 357         /*
 358          * The proxying router can't forward traffic sent to a link-local
 359          * address, so signal the sender and discard the packet. This
 360          * behavior is clarified by the MIPv6 specification.
 361          */
 362         if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
 363                 dst_link_failure(skb);
 364                 return -1;
 365         }
 366
 367         return 0;
 368 }
 369
 370 static inline int ip6_forward_finish(struct sk_buff *skb)
 371 {
 372         return dst_output(skb);
 373 }
 374
 375 int ip6_forward(struct sk_buff *skb)
 376 {
 377         struct dst_entry *dst = skb->dst;
 378         struct ipv6hdr *hdr = ipv6_hdr(skb);
 379         struct inet6_skb_parm *opt = IP6CB(skb);
 380
 381         if (ipv6_devconf.forwarding == 0)
 382                 goto error;
 383
 384         if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
 385                 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
 386                 goto drop;
 387         }
 388
 389         skb_forward_csum(skb);
 390
 391         /*
 392          *      We DO NOT make any processing on
 393          *      RA packets, pushing them to user level AS IS
 394          *      without ane WARRANTY that application will be able
 395          *      to interpret them. The reason is that we
 396          *      cannot make anything clever here.
 397          *
 398          *      We are not end-node, so that if packet contains
 399          *      AH/ESP, we cannot make anything.
 400          *      Defragmentation also would be mistake, RA packets
 401          *      cannot be fragmented, because there is no warranty
 402          *      that different fragments will go along one path. --ANK
 403          */
 404         if (opt->ra) {
 405                 u8 *ptr = skb_network_header(skb) + opt->ra;
 406                 if (ip6_call_ra_chain(skb, (ptr[2]<<8) + ptr[3]))
 407                         return 0;
 408         }
 409
 410         /*
 411          *      check and decrement ttl
 412          */
 413         if (hdr->hop_limit <= 1) {
 414                 /* Force OUTPUT device used as source address */
 415                 skb->dev = dst->dev;
 416                 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT,
 417                             0, skb->dev);
 418                 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INHDRERRORS);
 419
 420                 kfree_skb(skb);
 421                 return -ETIMEDOUT;
 422         }
 423
 424         /* XXX: idev->cnf.proxy_ndp? */
 425         if (ipv6_devconf.proxy_ndp &&
 426             pneigh_lookup(&nd_tbl, &hdr->daddr, skb->dev, 0)) {
 427                 int proxied = ip6_forward_proxy_check(skb);
 428                 if (proxied > 0)
 429                         return ip6_input(skb);
 430                 else if (proxied < 0) {
 431                         IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
 432                         goto drop;
 433                 }
 434         }
 435
 436         if (!xfrm6_route_forward(skb)) {
 437                 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
 438                 goto drop;
 439         }
 440         dst = skb->dst;
 441
 442         /* IPv6 specs say nothing about it, but it is clear that we cannot
 443            send redirects to source routed frames.
 444          */
 445         if (skb->dev == dst->dev && dst->neighbour && opt->srcrt == 0) {
 446                 struct in6_addr *target = NULL;
 447                 struct rt6_info *rt;
 448                 struct neighbour *n = dst->neighbour;
 449
 450                 /*
 451                  *      incoming and outgoing devices are the same
 452                  *      send a redirect.
 453                  */
 454
 455                 rt = (struct rt6_info *) dst;
 456                 if ((rt->rt6i_flags & RTF_GATEWAY))
 457                         target = (struct in6_addr*)&n->primary_key;
 458                 else
 459                         target = &hdr->daddr;
 460
 461                 /* Limit redirects both by destination (here)
 462                    and by source (inside ndisc_send_redirect)
 463                  */
 464                 if (xrlim_allow(dst, 1*HZ))
 465                         ndisc_send_redirect(skb, n, target);
 466         } else {
 467                 int addrtype = ipv6_addr_type(&hdr->saddr);
 468
 469                 /* This check is security critical. */
 470                 if (addrtype & (IPV6_ADDR_MULTICAST|IPV6_ADDR_LOOPBACK))
 471                         goto error;
 472                 if (addrtype & IPV6_ADDR_LINKLOCAL) {
 473                         icmpv6_send(skb, ICMPV6_DEST_UNREACH,
 474                                 ICMPV6_NOT_NEIGHBOUR, 0, skb->dev);
 475                         goto error;
 476                 }
 477         }
 478
 479         if (skb->len > dst_mtu(dst)) {
 480                 /* Again, force OUTPUT device used as source address */
 481                 skb->dev = dst->dev;
 482                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, dst_mtu(dst), skb->dev);
 483                 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INTOOBIGERRORS);
 484                 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_FRAGFAILS);
 485                 kfree_skb(skb);
 486                 return -EMSGSIZE;
 487         }
 488
 489         if (skb_cow(skb, dst->dev->hard_header_len)) {
 490                 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_OUTDISCARDS);
 491                 goto drop;
 492         }
 493
 494         hdr = ipv6_hdr(skb);
 495
 496         /* Mangling hops number delayed to point after skb COW */
 497
 498         hdr->hop_limit--;
 499
 500         IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
 501         return NF_HOOK(PF_INET6,NF_IP6_FORWARD, skb, skb->dev, dst->dev, ip6_forward_finish);
 502
 503 error:
 504         IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
 505 drop:
 506         kfree_skb(skb);
 507         return -EINVAL;
 508 }
 509
 510 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
 511 {
 512         to->pkt_type = from->pkt_type;
 513         to->priority = from->priority;
 514         to->protocol = from->protocol;
 515         dst_release(to->dst);
 516         to->dst = dst_clone(from->dst);
 517         to->dev = from->dev;
 518         to->mark = from->mark;
 519
 520 #ifdef CONFIG_NET_SCHED
 521         to->tc_index = from->tc_index;
 522 #endif
 523         nf_copy(to, from);
 524 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
 525     defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
 526         to->nf_trace = from->nf_trace;
 527 #endif
 528         skb_copy_secmark(to, from);
 529 }
 530
 531 int ip6_find_1stfragopt(struct sk_buff *skb, u8 **nexthdr)
 532 {
 533         u16 offset = sizeof(struct ipv6hdr);
 534         struct ipv6_opt_hdr *exthdr =
 535                                 (struct ipv6_opt_hdr *)(ipv6_hdr(skb) + 1);
 536         unsigned int packet_len = skb->tail - skb->network_header;
 537         int found_rhdr = 0;
 538         *nexthdr = &ipv6_hdr(skb)->nexthdr;
 539
 540         while (offset + 1 <= packet_len) {
 541
 542                 switch (**nexthdr) {
 543
 544                 case NEXTHDR_HOP:
 545                         break;
 546                 case NEXTHDR_ROUTING:
 547                         found_rhdr = 1;
 548                         break;
 549                 case NEXTHDR_DEST:
 550 #if defined(CONFIG_IPV6_MIP6) || defined(CONFIG_IPV6_MIP6_MODULE)
 551                         if (ipv6_find_tlv(skb, offset, IPV6_TLV_HAO) >= 0)
 552                                 break;
 553 #endif
 554                         if (found_rhdr)
 555                                 return offset;
 556                         break;
 557                 default :
 558                         return offset;
 559                 }
 560
 561                 offset += ipv6_optlen(exthdr);
 562                 *nexthdr = &exthdr->nexthdr;
 563                 exthdr = (struct ipv6_opt_hdr *)(skb_network_header(skb) +
 564                                                  offset);
 565         }
 566
 567         return offset;
 568 }
 569 EXPORT_SYMBOL_GPL(ip6_find_1stfragopt);
 570
 571 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
 572 {
 573         struct net_device *dev;
 574         struct sk_buff *frag;
 575         struct rt6_info *rt = (struct rt6_info*)skb->dst;
 576         struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
 577         struct ipv6hdr *tmp_hdr;
 578         struct frag_hdr *fh;
 579         unsigned int mtu, hlen, left, len;
 580         __be32 frag_id = 0;
 581         int ptr, offset = 0, err=0;
 582         u8 *prevhdr, nexthdr = 0;
 583
 584         dev = rt->u.dst.dev;
 585         hlen = ip6_find_1stfragopt(skb, &prevhdr);
 586         nexthdr = *prevhdr;
 587
 588         mtu = ip6_skb_dst_mtu(skb);
 589
 590         /* We must not fragment if the socket is set to force MTU discovery
 591          * or if the skb it not generated by a local socket.  (This last
 592          * check should be redundant, but it's free.)
 593          */
 594         if (!np || np->pmtudisc >= IPV6_PMTUDISC_DO) {
 595                 skb->dev = skb->dst->dev;
 596                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev);
 597                 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGFAILS);
 598                 kfree_skb(skb);
 599                 return -EMSGSIZE;
 600         }
 601
 602         if (np && np->frag_size < mtu) {
 603                 if (np->frag_size)
 604                         mtu = np->frag_size;
 605         }
 606         mtu -= hlen + sizeof(struct frag_hdr);
 607
 608         if (skb_shinfo(skb)->frag_list) {
 609                 int first_len = skb_pagelen(skb);
 610
 611                 if (first_len - hlen > mtu ||
 612                     ((first_len - hlen) & 7) ||
 613                     skb_cloned(skb))
 614                         goto slow_path;
 615
 616                 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
 617                         /* Correct geometry. */
 618                         if (frag->len > mtu ||
 619                             ((frag->len & 7) && frag->next) ||
 620                             skb_headroom(frag) < hlen)
 621                             goto slow_path;
 622
 623                         /* Partially cloned skb? */
 624                         if (skb_shared(frag))
 625                                 goto slow_path;
 626
 627                         BUG_ON(frag->sk);
 628                         if (skb->sk) {
 629                                 sock_hold(skb->sk);
 630                                 frag->sk = skb->sk;
 631                                 frag->destructor = sock_wfree;
 632                                 skb->truesize -= frag->truesize;
 633                         }
 634                 }
 635
 636                 err = 0;
 637                 offset = 0;
 638                 frag = skb_shinfo(skb)->frag_list;
 639                 skb_shinfo(skb)->frag_list = NULL;
 640                 /* BUILD HEADER */
 641
 642                 *prevhdr = NEXTHDR_FRAGMENT;
 643                 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
 644                 if (!tmp_hdr) {
 645                         IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGFAILS);
 646                         return -ENOMEM;
 647                 }
 648
 649                 __skb_pull(skb, hlen);
 650                 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
 651                 __skb_push(skb, hlen);
 652                 skb_reset_network_header(skb);
 653                 memcpy(skb_network_header(skb), tmp_hdr, hlen);
 654
 655                 ipv6_select_ident(skb, fh);
 656                 fh->nexthdr = nexthdr;
 657                 fh->reserved = 0;
 658                 fh->frag_off = htons(IP6_MF);
 659                 frag_id = fh->identification;
 660
 661                 first_len = skb_pagelen(skb);
 662                 skb->data_len = first_len - skb_headlen(skb);
 663                 skb->len = first_len;
 664                 ipv6_hdr(skb)->payload_len = htons(first_len -
 665                                                    sizeof(struct ipv6hdr));
 666
 667                 dst_hold(&rt->u.dst);
 668
 669                 for (;;) {
 670                         /* Prepare header of the next frame,
 671                          * before previous one went down. */
 672                         if (frag) {
 673                                 frag->ip_summed = CHECKSUM_NONE;
 674                                 skb_reset_transport_header(frag);
 675                                 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
 676                                 __skb_push(frag, hlen);
 677                                 skb_reset_network_header(frag);
 678                                 memcpy(skb_network_header(frag), tmp_hdr,
 679                                        hlen);
 680                                 offset += skb->len - hlen - sizeof(struct frag_hdr);
 681                                 fh->nexthdr = nexthdr;
 682                                 fh->reserved = 0;
 683                                 fh->frag_off = htons(offset);
 684                                 if (frag->next != NULL)
 685                                         fh->frag_off |= htons(IP6_MF);
 686                                 fh->identification = frag_id;
 687                                 ipv6_hdr(frag)->payload_len =
 688                                                 htons(frag->len -
 689                                                       sizeof(struct ipv6hdr));
 690                                 ip6_copy_metadata(frag, skb);
 691                         }
 692
 693                         err = output(skb);
 694                         if(!err)
 695                                 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGCREATES);
 696
 697                         if (err || !frag)
 698                                 break;
 699
 700                         skb = frag;
 701                         frag = skb->next;
 702                         skb->next = NULL;
 703                 }
 704
 705                 kfree(tmp_hdr);
 706
 707                 if (err == 0) {
 708                         IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGOKS);
 709                         dst_release(&rt->u.dst);
 710                         return 0;
 711                 }
 712
 713                 while (frag) {
 714                         skb = frag->next;
 715                         kfree_skb(frag);
 716                         frag = skb;
 717                 }
 718
 719                 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGFAILS);
 720                 dst_release(&rt->u.dst);
 721                 return err;
 722         }
 723
 724 slow_path:
 725         left = skb->len - hlen;         /* Space per frame */
 726         ptr = hlen;                     /* Where to start from */
 727
 728         /*
 729          *      Fragment the datagram.
 730          */
 731
 732         *prevhdr = NEXTHDR_FRAGMENT;
 733
 734         /*
 735          *      Keep copying data until we run out.
 736          */
 737         while(left > 0) {
 738                 len = left;
 739                 /* IF: it doesn't fit, use 'mtu' - the data space left */
 740                 if (len > mtu)
 741                         len = mtu;
 742                 /* IF: we are not sending upto and including the packet end
 743                    then align the next start on an eight byte boundary */
 744                 if (len < left) {
 745                         len &= ~7;
 746                 }
 747                 /*
 748                  *      Allocate buffer.
 749                  */
 750
 751                 if ((frag = alloc_skb(len+hlen+sizeof(struct frag_hdr)+LL_RESERVED_SPACE(rt->u.dst.dev), GFP_ATOMIC)) == NULL) {
 752                         NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
 753                         IP6_INC_STATS(ip6_dst_idev(skb->dst),
 754                                       IPSTATS_MIB_FRAGFAILS);
 755                         err = -ENOMEM;
 756                         goto fail;
 757                 }
 758
 759                 /*
 760                  *      Set up data on packet
 761                  */
 762
 763                 ip6_copy_metadata(frag, skb);
 764                 skb_reserve(frag, LL_RESERVED_SPACE(rt->u.dst.dev));
 765                 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
 766                 skb_reset_network_header(frag);
 767                 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
 768                 frag->transport_header = (frag->network_header + hlen +
 769                                           sizeof(struct frag_hdr));
 770
 771                 /*
 772                  *      Charge the memory for the fragment to any owner
 773                  *      it might possess
 774                  */
 775                 if (skb->sk)
 776                         skb_set_owner_w(frag, skb->sk);
 777
 778                 /*
 779                  *      Copy the packet header into the new buffer.
 780                  */
 781                 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
 782
 783                 /*
 784                  *      Build fragment header.
 785                  */
 786                 fh->nexthdr = nexthdr;
 787                 fh->reserved = 0;
 788                 if (!frag_id) {
 789                         ipv6_select_ident(skb, fh);
 790                         frag_id = fh->identification;
 791                 } else
 792                         fh->identification = frag_id;
 793
 794                 /*
 795                  *      Copy a block of the IP datagram.
 796                  */
 797                 if (skb_copy_bits(skb, ptr, skb_transport_header(frag), len))
 798                         BUG();
 799                 left -= len;
 800
 801                 fh->frag_off = htons(offset);
 802                 if (left > 0)
 803                         fh->frag_off |= htons(IP6_MF);
 804                 ipv6_hdr(frag)->payload_len = htons(frag->len -
 805                                                     sizeof(struct ipv6hdr));
 806
 807                 ptr += len;
 808                 offset += len;
 809
 810                 /*
 811                  *      Put this fragment into the sending queue.
 812                  */
 813                 err = output(frag);
 814                 if (err)
 815                         goto fail;
 816
 817                 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGCREATES);
 818         }
 819         IP6_INC_STATS(ip6_dst_idev(skb->dst),
 820                       IPSTATS_MIB_FRAGOKS);
 821         kfree_skb(skb);
 822         return err;
 823
 824 fail:
 825         IP6_INC_STATS(ip6_dst_idev(skb->dst),
 826                       IPSTATS_MIB_FRAGFAILS);
 827         kfree_skb(skb);
 828         return err;
 829 }
 830
 831 static inline int ip6_rt_check(struct rt6key *rt_key,
 832                                struct in6_addr *fl_addr,
 833                                struct in6_addr *addr_cache)
 834 {
 835         return ((rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
 836                 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache)));
 837 }
 838
 839 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
 840                                           struct dst_entry *dst,
 841                                           struct flowi *fl)
 842 {
 843         struct ipv6_pinfo *np = inet6_sk(sk);
 844         struct rt6_info *rt = (struct rt6_info *)dst;
 845
 846         if (!dst)
 847                 goto out;
 848
 849         /* Yes, checking route validity in not connected
 850          * case is not very simple. Take into account,
 851          * that we do not support routing by source, TOS,
 852          * and MSG_DONTROUTE            --ANK (980726)
 853          *
 854          * 1. ip6_rt_check(): If route was host route,
 855          *    check that cached destination is current.
 856          *    If it is network route, we still may
 857          *    check its validity using saved pointer
 858          *    to the last used address: daddr_cache.
 859          *    We do not want to save whole address now,
 860          *    (because main consumer of this service
 861          *    is tcp, which has not this problem),
 862          *    so that the last trick works only on connected
 863          *    sockets.
 864          * 2. oif also should be the same.
 865          */
 866         if (ip6_rt_check(&rt->rt6i_dst, &fl->fl6_dst, np->daddr_cache) ||
 867 #ifdef CONFIG_IPV6_SUBTREES
 868             ip6_rt_check(&rt->rt6i_src, &fl->fl6_src, np->saddr_cache) ||
 869 #endif
 870             (fl->oif && fl->oif != dst->dev->ifindex)) {
 871                 dst_release(dst);
 872                 dst = NULL;
 873         }
 874
 875 out:
 876         return dst;
 877 }
 878
 879 static int ip6_dst_lookup_tail(struct sock *sk,
 880                                struct dst_entry **dst, struct flowi *fl)
 881 {
 882         int err;
 883
 884         if (*dst == NULL)
 885                 *dst = ip6_route_output(sk, fl);
 886
 887         if ((err = (*dst)->error))
 888                 goto out_err_release;
 889
 890         if (ipv6_addr_any(&fl->fl6_src)) {
 891                 err = ipv6_get_saddr(*dst, &fl->fl6_dst, &fl->fl6_src);
 892                 if (err)
 893                         goto out_err_release;
 894         }
 895
 896 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
 897                 /*
 898                  * Here if the dst entry we've looked up
 899                  * has a neighbour entry that is in the INCOMPLETE
 900                  * state and the src address from the flow is
 901                  * marked as OPTIMISTIC, we release the found
 902                  * dst entry and replace it instead with the
 903                  * dst entry of the nexthop router
 904                  */
 905                 if (!((*dst)->neighbour->nud_state & NUD_VALID)) {
 906                         struct inet6_ifaddr *ifp;
 907                         struct flowi fl_gw;
 908                         int redirect;
 909
 910                         ifp = ipv6_get_ifaddr(&fl->fl6_src, (*dst)->dev, 1);
 911
 912                         redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
 913                         if (ifp)
 914                                 in6_ifa_put(ifp);
 915
 916                         if (redirect) {
 917                                 /*
 918                                  * We need to get the dst entry for the
 919                                  * default router instead
 920                                  */
 921                                 dst_release(*dst);
 922                                 memcpy(&fl_gw, fl, sizeof(struct flowi));
 923                                 memset(&fl_gw.fl6_dst, 0, sizeof(struct in6_addr));
 924                                 *dst = ip6_route_output(sk, &fl_gw);
 925                                 if ((err = (*dst)->error))
 926                                         goto out_err_release;
 927                         }
 928                 }
 929 #endif
 930
 931         return 0;
 932
 933 out_err_release:
 934         dst_release(*dst);
 935         *dst = NULL;
 936         return err;
 937 }
 938
 939 /**
 940  *      ip6_dst_lookup - perform route lookup on flow
 941  *      @sk: socket which provides route info
 942  *      @dst: pointer to dst_entry * for result
 943  *      @fl: flow to lookup
 944  *
 945  *      This function performs a route lookup on the given flow.
 946  *
 947  *      It returns zero on success, or a standard errno code on error.
 948  */
 949 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
 950 {
 951         *dst = NULL;
 952         return ip6_dst_lookup_tail(sk, dst, fl);
 953 }
 954 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
 955
 956 /**
 957  *      ip6_sk_dst_lookup - perform socket cached route lookup on flow
 958  *      @sk: socket which provides the dst cache and route info
 959  *      @dst: pointer to dst_entry * for result
 960  *      @fl: flow to lookup
 961  *
 962  *      This function performs a route lookup on the given flow with the
 963  *      possibility of using the cached route in the socket if it is valid.
 964  *      It will take the socket dst lock when operating on the dst cache.
 965  *      As a result, this function can only be used in process context.
 966  *
 967  *      It returns zero on success, or a standard errno code on error.
 968  */
 969 int ip6_sk_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
 970 {
 971         *dst = NULL;
 972         if (sk) {
 973                 *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
 974                 *dst = ip6_sk_dst_check(sk, *dst, fl);
 975         }
 976
 977         return ip6_dst_lookup_tail(sk, dst, fl);
 978 }
 979 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup);
 980
 981 static inline int ip6_ufo_append_data(struct sock *sk,
 982                         int getfrag(void *from, char *to, int offset, int len,
 983                         int odd, struct sk_buff *skb),
 984                         void *from, int length, int hh_len, int fragheaderlen,
 985                         int transhdrlen, int mtu,unsigned int flags)
 986
 987 {
 988         struct sk_buff *skb;
 989         int err;
 990
 991         /* There is support for UDP large send offload by network
 992          * device, so create one single skb packet containing complete
 993          * udp datagram
 994          */
 995         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
 996                 skb = sock_alloc_send_skb(sk,
 997                         hh_len + fragheaderlen + transhdrlen + 20,
 998                         (flags & MSG_DONTWAIT), &err);
 999                 if (skb == NULL)
1000                         return -ENOMEM;
1001
1002                 /* reserve space for Hardware header */
1003                 skb_reserve(skb, hh_len);
1004
1005                 /* create space for UDP/IP header */
1006                 skb_put(skb,fragheaderlen + transhdrlen);
1007
1008                 /* initialize network header pointer */
1009                 skb_reset_network_header(skb);
1010
1011                 /* initialize protocol header pointer */
1012                 skb->transport_header = skb->network_header + fragheaderlen;
1013
1014                 skb->ip_summed = CHECKSUM_PARTIAL;
1015                 skb->csum = 0;
1016                 sk->sk_sndmsg_off = 0;
1017         }
1018
1019         err = skb_append_datato_frags(sk,skb, getfrag, from,
1020                                       (length - transhdrlen));
1021         if (!err) {
1022                 struct frag_hdr fhdr;
1023
1024                 /* specify the length of each IP datagram fragment*/
1025                 skb_shinfo(skb)->gso_size = mtu - fragheaderlen -
1026                                             sizeof(struct frag_hdr);
1027                 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1028                 ipv6_select_ident(skb, &fhdr);
1029                 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1030                 __skb_queue_tail(&sk->sk_write_queue, skb);
1031
1032                 return 0;
1033         }
1034         /* There is not enough support do UPD LSO,
1035          * so follow normal path
1036          */
1037         kfree_skb(skb);
1038
1039         return err;
1040 }
1041
1042 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1043         int offset, int len, int odd, struct sk_buff *skb),
1044         void *from, int length, int transhdrlen,
1045         int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi *fl,
1046         struct rt6_info *rt, unsigned int flags)
1047 {
1048         struct inet_sock *inet = inet_sk(sk);
1049         struct ipv6_pinfo *np = inet6_sk(sk);
1050         struct sk_buff *skb;
1051         unsigned int maxfraglen, fragheaderlen;
1052         int exthdrlen;
1053         int hh_len;
1054         int mtu;
1055         int copy;
1056         int err;
1057         int offset = 0;
1058         int csummode = CHECKSUM_NONE;
1059
1060         if (flags&MSG_PROBE)
1061                 return 0;
1062         if (skb_queue_empty(&sk->sk_write_queue)) {
1063                 /*
1064                  * setup for corking
1065                  */
1066                 if (opt) {
1067                         if (np->cork.opt == NULL) {
1068                                 np->cork.opt = kmalloc(opt->tot_len,
1069                                                        sk->sk_allocation);
1070                                 if (unlikely(np->cork.opt == NULL))
1071                                         return -ENOBUFS;
1072                         } else if (np->cork.opt->tot_len < opt->tot_len) {
1073                                 printk(KERN_DEBUG "ip6_append_data: invalid option length\n");
1074                                 return -EINVAL;
1075                         }
1076                         memcpy(np->cork.opt, opt, opt->tot_len);
1077                         inet->cork.flags |= IPCORK_OPT;
1078                         /* need source address above miyazawa*/
1079                 }
1080                 dst_hold(&rt->u.dst);
1081                 np->cork.rt = rt;
1082                 inet->cork.fl = *fl;
1083                 np->cork.hop_limit = hlimit;
1084                 np->cork.tclass = tclass;
1085                 mtu = np->pmtudisc == IPV6_PMTUDISC_PROBE ?
1086                       rt->u.dst.dev->mtu : dst_mtu(rt->u.dst.path);
1087                 if (np->frag_size < mtu) {
1088                         if (np->frag_size)
1089                                 mtu = np->frag_size;
1090                 }
1091                 inet->cork.fragsize = mtu;
1092                 if (dst_allfrag(rt->u.dst.path))
1093                         inet->cork.flags |= IPCORK_ALLFRAG;
1094                 inet->cork.length = 0;
1095                 sk->sk_sndmsg_page = NULL;
1096                 sk->sk_sndmsg_off = 0;
1097                 exthdrlen = rt->u.dst.header_len + (opt ? opt->opt_flen : 0);
1098                 length += exthdrlen;
1099                 transhdrlen += exthdrlen;
1100         } else {
1101                 rt = np->cork.rt;
1102                 fl = &inet->cork.fl;
1103                 if (inet->cork.flags & IPCORK_OPT)
1104                         opt = np->cork.opt;
1105                 transhdrlen = 0;
1106                 exthdrlen = 0;
1107                 mtu = inet->cork.fragsize;
1108         }
1109
1110         hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1111
1112         fragheaderlen = sizeof(struct ipv6hdr) + rt->u.dst.nfheader_len + (opt ? opt->opt_nflen : 0);
1113         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
1114
1115         if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1116                 if (inet->cork.length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
1117                         ipv6_local_error(sk, EMSGSIZE, fl, mtu-exthdrlen);
1118                         return -EMSGSIZE;
1119                 }
1120         }
1121
1122         /*
1123          * Let's try using as much space as possible.
1124          * Use MTU if total length of the message fits into the MTU.
1125          * Otherwise, we need to reserve fragment header and
1126          * fragment alignment (= 8-15 octects, in total).
1127          *
1128          * Note that we may need to "move" the data from the tail of
1129          * of the buffer to the new fragment when we split
1130          * the message.
1131          *
1132          * FIXME: It may be fragmented into multiple chunks
1133          *        at once if non-fragmentable extension headers
1134          *        are too large.
1135          * --yoshfuji
1136          */
1137
1138         inet->cork.length += length;
1139         if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
1140             (rt->u.dst.dev->features & NETIF_F_UFO)) {
1141
1142                 err = ip6_ufo_append_data(sk, getfrag, from, length, hh_len,
1143                                           fragheaderlen, transhdrlen, mtu,
1144                                           flags);
1145                 if (err)
1146                         goto error;
1147                 return 0;
1148         }
1149
1150         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1151                 goto alloc_new_skb;
1152
1153         while (length > 0) {
1154                 /* Check if the remaining data fits into current packet. */
1155                 copy = (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1156                 if (copy < length)
1157                         copy = maxfraglen - skb->len;
1158
1159                 if (copy <= 0) {
1160                         char *data;
1161                         unsigned int datalen;
1162                         unsigned int fraglen;
1163                         unsigned int fraggap;
1164                         unsigned int alloclen;
1165                         struct sk_buff *skb_prev;
1166 alloc_new_skb:
1167                         skb_prev = skb;
1168
1169                         /* There's no room in the current skb */
1170                         if (skb_prev)
1171                                 fraggap = skb_prev->len - maxfraglen;
1172                         else
1173                                 fraggap = 0;
1174
1175                         /*
1176                          * If remaining data exceeds the mtu,
1177                          * we know we need more fragment(s).
1178                          */
1179                         datalen = length + fraggap;
1180                         if (datalen > (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1181                                 datalen = maxfraglen - fragheaderlen;
1182
1183                         fraglen = datalen + fragheaderlen;
1184                         if ((flags & MSG_MORE) &&
1185                             !(rt->u.dst.dev->features&NETIF_F_SG))
1186                                 alloclen = mtu;
1187                         else
1188                                 alloclen = datalen + fragheaderlen;
1189
1190                         /*
1191                          * The last fragment gets additional space at tail.
1192                          * Note: we overallocate on fragments with MSG_MODE
1193                          * because we have no idea if we're the last one.
1194                          */
1195                         if (datalen == length + fraggap)
1196                                 alloclen += rt->u.dst.trailer_len;
1197
1198                         /*
1199                          * We just reserve space for fragment header.
1200                          * Note: this may be overallocation if the message
1201                          * (without MSG_MORE) fits into the MTU.
1202                          */
1203                         alloclen += sizeof(struct frag_hdr);
1204
1205                         if (transhdrlen) {
1206                                 skb = sock_alloc_send_skb(sk,
1207                                                 alloclen + hh_len,
1208                                                 (flags & MSG_DONTWAIT), &err);
1209                         } else {
1210                                 skb = NULL;
1211                                 if (atomic_read(&sk->sk_wmem_alloc) <=
1212                                     2 * sk->sk_sndbuf)
1213                                         skb = sock_wmalloc(sk,
1214                                                            alloclen + hh_len, 1,
1215                                                            sk->sk_allocation);
1216                                 if (unlikely(skb == NULL))
1217                                         err = -ENOBUFS;
1218                         }
1219                         if (skb == NULL)
1220                                 goto error;
1221                         /*
1222                          *      Fill in the control structures
1223                          */
1224                         skb->ip_summed = csummode;
1225                         skb->csum = 0;
1226                         /* reserve for fragmentation */
1227                         skb_reserve(skb, hh_len+sizeof(struct frag_hdr));
1228
1229                         /*
1230                          *      Find where to start putting bytes
1231                          */
1232                         data = skb_put(skb, fraglen);
1233                         skb_set_network_header(skb, exthdrlen);
1234                         data += fragheaderlen;
1235                         skb->transport_header = (skb->network_header +
1236                                                  fragheaderlen);
1237                         if (fraggap) {
1238                                 skb->csum = skb_copy_and_csum_bits(
1239                                         skb_prev, maxfraglen,
1240                                         data + transhdrlen, fraggap, 0);
1241                                 skb_prev->csum = csum_sub(skb_prev->csum,
1242                                                           skb->csum);
1243                                 data += fraggap;
1244                                 pskb_trim_unique(skb_prev, maxfraglen);
1245                         }
1246                         copy = datalen - transhdrlen - fraggap;
1247                         if (copy < 0) {
1248                                 err = -EINVAL;
1249                                 kfree_skb(skb);
1250                                 goto error;
1251                         } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1252                                 err = -EFAULT;
1253                                 kfree_skb(skb);
1254                                 goto error;
1255                         }
1256
1257                         offset += copy;
1258                         length -= datalen - fraggap;
1259                         transhdrlen = 0;
1260                         exthdrlen = 0;
1261                         csummode = CHECKSUM_NONE;
1262
1263                         /*
1264                          * Put the packet on the pending queue
1265                          */
1266                         __skb_queue_tail(&sk->sk_write_queue, skb);
1267                         continue;
1268                 }
1269
1270                 if (copy > length)
1271                         copy = length;
1272
1273                 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
1274                         unsigned int off;
1275
1276                         off = skb->len;
1277                         if (getfrag(from, skb_put(skb, copy),
1278                                                 offset, copy, off, skb) < 0) {
1279                                 __skb_trim(skb, off);
1280                                 err = -EFAULT;
1281                                 goto error;
1282                         }
1283                 } else {
1284                         int i = skb_shinfo(skb)->nr_frags;
1285                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1286                         struct page *page = sk->sk_sndmsg_page;
1287                         int off = sk->sk_sndmsg_off;
1288                         unsigned int left;
1289
1290                         if (page && (left = PAGE_SIZE - off) > 0) {
1291                                 if (copy >= left)
1292                                         copy = left;
1293                                 if (page != frag->page) {
1294                                         if (i == MAX_SKB_FRAGS) {
1295                                                 err = -EMSGSIZE;
1296                                                 goto error;
1297                                         }
1298                                         get_page(page);
1299                                         skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1300                                         frag = &skb_shinfo(skb)->frags[i];
1301                                 }
1302                         } else if(i < MAX_SKB_FRAGS) {
1303                                 if (copy > PAGE_SIZE)
1304                                         copy = PAGE_SIZE;
1305                                 page = alloc_pages(sk->sk_allocation, 0);
1306                                 if (page == NULL) {
1307                                         err = -ENOMEM;
1308                                         goto error;
1309                                 }
1310                                 sk->sk_sndmsg_page = page;
1311                                 sk->sk_sndmsg_off = 0;
1312
1313                                 skb_fill_page_desc(skb, i, page, 0, 0);
1314                                 frag = &skb_shinfo(skb)->frags[i];
1315                                 skb->truesize += PAGE_SIZE;
1316                                 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1317                         } else {
1318                                 err = -EMSGSIZE;
1319                                 goto error;
1320                         }
1321                         if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1322                                 err = -EFAULT;
1323                                 goto error;
1324                         }
1325                         sk->sk_sndmsg_off += copy;
1326                         frag->size += copy;
1327                         skb->len += copy;
1328                         skb->data_len += copy;
1329                 }
1330                 offset += copy;
1331                 length -= copy;
1332         }
1333         return 0;
1334 error:
1335         inet->cork.length -= length;
1336         IP6_INC_STATS(rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1337         return err;
1338 }
1339
1340 int ip6_push_pending_frames(struct sock *sk)
1341 {
1342         struct sk_buff *skb, *tmp_skb;
1343         struct sk_buff **tail_skb;
1344         struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1345         struct inet_sock *inet = inet_sk(sk);
1346         struct ipv6_pinfo *np = inet6_sk(sk);
1347         struct ipv6hdr *hdr;
1348         struct ipv6_txoptions *opt = np->cork.opt;
1349         struct rt6_info *rt = np->cork.rt;
1350         struct flowi *fl = &inet->cork.fl;
1351         unsigned char proto = fl->proto;
1352         int err = 0;
1353
1354         if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1355                 goto out;
1356         tail_skb = &(skb_shinfo(skb)->frag_list);
1357
1358         /* move skb->data to ip header from ext header */
1359         if (skb->data < skb_network_header(skb))
1360                 __skb_pull(skb, skb_network_offset(skb));
1361         while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1362                 __skb_pull(tmp_skb, skb_network_header_len(skb));
1363                 *tail_skb = tmp_skb;
1364                 tail_skb = &(tmp_skb->next);
1365                 skb->len += tmp_skb->len;
1366                 skb->data_len += tmp_skb->len;
1367                 skb->truesize += tmp_skb->truesize;
1368                 __sock_put(tmp_skb->sk);
1369                 tmp_skb->destructor = NULL;
1370                 tmp_skb->sk = NULL;
1371         }
1372
1373         ipv6_addr_copy(final_dst, &fl->fl6_dst);
1374         __skb_pull(skb, skb_network_header_len(skb));
1375         if (opt && opt->opt_flen)
1376                 ipv6_push_frag_opts(skb, opt, &proto);
1377         if (opt && opt->opt_nflen)
1378                 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1379
1380         skb_push(skb, sizeof(struct ipv6hdr));
1381         skb_reset_network_header(skb);
1382         hdr = ipv6_hdr(skb);
1383
1384         *(__be32*)hdr = fl->fl6_flowlabel |
1385                      htonl(0x60000000 | ((int)np->cork.tclass << 20));
1386
1387         if (skb->len <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN)
1388                 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
1389         else
1390                 hdr->payload_len = 0;
1391         hdr->hop_limit = np->cork.hop_limit;
1392         hdr->nexthdr = proto;
1393         ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
1394         ipv6_addr_copy(&hdr->daddr, final_dst);
1395
1396         skb->priority = sk->sk_priority;
1397
1398         skb->dst = dst_clone(&rt->u.dst);
1399         IP6_INC_STATS(rt->rt6i_idev, IPSTATS_MIB_OUTREQUESTS);
1400         err = NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, skb->dst->dev, dst_output);
1401         if (err) {
1402                 if (err > 0)
1403                         err = np->recverr ? net_xmit_errno(err) : 0;
1404                 if (err)
1405                         goto error;
1406         }
1407
1408 out:
1409         inet->cork.flags &= ~IPCORK_OPT;
1410         kfree(np->cork.opt);
1411         np->cork.opt = NULL;
1412         if (np->cork.rt) {
1413                 dst_release(&np->cork.rt->u.dst);
1414                 np->cork.rt = NULL;
1415                 inet->cork.flags &= ~IPCORK_ALLFRAG;
1416         }
1417         memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1418         return err;
1419 error:
1420         goto out;
1421 }
1422
1423 void ip6_flush_pending_frames(struct sock *sk)
1424 {
1425         struct inet_sock *inet = inet_sk(sk);
1426         struct ipv6_pinfo *np = inet6_sk(sk);
1427         struct sk_buff *skb;
1428
1429         while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1430                 if (skb->dst)
1431                         IP6_INC_STATS(ip6_dst_idev(skb->dst),
1432                                       IPSTATS_MIB_OUTDISCARDS);
1433                 kfree_skb(skb);
1434         }
1435
1436         inet->cork.flags &= ~IPCORK_OPT;
1437
1438         kfree(np->cork.opt);
1439         np->cork.opt = NULL;
1440         if (np->cork.rt) {
1441                 dst_release(&np->cork.rt->u.dst);
1442                 np->cork.rt = NULL;
1443                 inet->cork.flags &= ~IPCORK_ALLFRAG;
1444         }
1445         memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1446 }