git.oblomov.eu Git - linux-2.6/blob - 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/kernel.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 int __ip6_local_out(struct sk_buff *skb)
  74 {
  75         int len;
  76
  77         len = skb->len - sizeof(struct ipv6hdr);
  78         if (len > IPV6_MAXPLEN)
  79                 len = 0;
  80         ipv6_hdr(skb)->payload_len = htons(len);
  81
  82         return nf_hook(PF_INET6, NF_INET_LOCAL_OUT, skb, NULL, skb->dst->dev,
  83                        dst_output);
  84 }
  85
  86 int ip6_local_out(struct sk_buff *skb)
  87 {
  88         int err;
  89
  90         err = __ip6_local_out(skb);
  91         if (likely(err == 1))
  92                 err = dst_output(skb);
  93
  94         return err;
  95 }
  96 EXPORT_SYMBOL_GPL(ip6_local_out);
  97
  98 static int ip6_output_finish(struct sk_buff *skb)
  99 {
 100         struct dst_entry *dst = skb->dst;
 101
 102         if (dst->hh)
 103                 return neigh_hh_output(dst->hh, skb);
 104         else if (dst->neighbour)
 105                 return dst->neighbour->output(skb);
 106
 107         IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
 108         kfree_skb(skb);
 109         return -EINVAL;
 110
 111 }
 112
 113 /* dev_loopback_xmit for use with netfilter. */
 114 static int ip6_dev_loopback_xmit(struct sk_buff *newskb)
 115 {
 116         skb_reset_mac_header(newskb);
 117         __skb_pull(newskb, skb_network_offset(newskb));
 118         newskb->pkt_type = PACKET_LOOPBACK;
 119         newskb->ip_summed = CHECKSUM_UNNECESSARY;
 120         BUG_TRAP(newskb->dst);
 121
 122         netif_rx(newskb);
 123         return 0;
 124 }
 125
 126
 127 static int ip6_output2(struct sk_buff *skb)
 128 {
 129         struct dst_entry *dst = skb->dst;
 130         struct net_device *dev = dst->dev;
 131
 132         skb->protocol = htons(ETH_P_IPV6);
 133         skb->dev = dev;
 134
 135         if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
 136                 struct ipv6_pinfo* np = skb->sk ? inet6_sk(skb->sk) : NULL;
 137                 struct inet6_dev *idev = ip6_dst_idev(skb->dst);
 138
 139                 if (!(dev->flags & IFF_LOOPBACK) && (!np || np->mc_loop) &&
 140                     ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
 141                                         &ipv6_hdr(skb)->saddr)) {
 142                         struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
 143
 144                         /* Do not check for IFF_ALLMULTI; multicast routing
 145                            is not supported in any case.
 146                          */
 147                         if (newskb)
 148                                 NF_HOOK(PF_INET6, NF_INET_POST_ROUTING, newskb,
 149                                         NULL, newskb->dev,
 150                                         ip6_dev_loopback_xmit);
 151
 152                         if (ipv6_hdr(skb)->hop_limit == 0) {
 153                                 IP6_INC_STATS(idev, IPSTATS_MIB_OUTDISCARDS);
 154                                 kfree_skb(skb);
 155                                 return 0;
 156                         }
 157                 }
 158
 159                 IP6_INC_STATS(idev, IPSTATS_MIB_OUTMCASTPKTS);
 160         }
 161
 162         return NF_HOOK(PF_INET6, NF_INET_POST_ROUTING, skb, NULL, skb->dev,
 163                        ip6_output_finish);
 164 }
 165
 166 static inline int ip6_skb_dst_mtu(struct sk_buff *skb)
 167 {
 168         struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
 169
 170         return (np && np->pmtudisc == IPV6_PMTUDISC_PROBE) ?
 171                skb->dst->dev->mtu : dst_mtu(skb->dst);
 172 }
 173
 174 int ip6_output(struct sk_buff *skb)
 175 {
 176         if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
 177                                 dst_allfrag(skb->dst))
 178                 return ip6_fragment(skb, ip6_output2);
 179         else
 180                 return ip6_output2(skb);
 181 }
 182
 183 /*
 184  *      xmit an sk_buff (used by TCP)
 185  */
 186
 187 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
 188              struct ipv6_txoptions *opt, int ipfragok)
 189 {
 190         struct ipv6_pinfo *np = inet6_sk(sk);
 191         struct in6_addr *first_hop = &fl->fl6_dst;
 192         struct dst_entry *dst = skb->dst;
 193         struct ipv6hdr *hdr;
 194         u8  proto = fl->proto;
 195         int seg_len = skb->len;
 196         int hlimit, tclass;
 197         u32 mtu;
 198
 199         if (opt) {
 200                 unsigned int head_room;
 201
 202                 /* First: exthdrs may take lots of space (~8K for now)
 203                    MAX_HEADER is not enough.
 204                  */
 205                 head_room = opt->opt_nflen + opt->opt_flen;
 206                 seg_len += head_room;
 207                 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
 208
 209                 if (skb_headroom(skb) < head_room) {
 210                         struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
 211                         if (skb2 == NULL) {
 212                                 IP6_INC_STATS(ip6_dst_idev(skb->dst),
 213                                               IPSTATS_MIB_OUTDISCARDS);
 214                                 kfree_skb(skb);
 215                                 return -ENOBUFS;
 216                         }
 217                         kfree_skb(skb);
 218                         skb = skb2;
 219                         if (sk)
 220                                 skb_set_owner_w(skb, sk);
 221                 }
 222                 if (opt->opt_flen)
 223                         ipv6_push_frag_opts(skb, opt, &proto);
 224                 if (opt->opt_nflen)
 225                         ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
 226         }
 227
 228         skb_push(skb, sizeof(struct ipv6hdr));
 229         skb_reset_network_header(skb);
 230         hdr = ipv6_hdr(skb);
 231
 232         /*
 233          *      Fill in the IPv6 header
 234          */
 235
 236         hlimit = -1;
 237         if (np)
 238                 hlimit = np->hop_limit;
 239         if (hlimit < 0)
 240                 hlimit = dst_metric(dst, RTAX_HOPLIMIT);
 241         if (hlimit < 0)
 242                 hlimit = ipv6_get_hoplimit(dst->dev);
 243
 244         tclass = -1;
 245         if (np)
 246                 tclass = np->tclass;
 247         if (tclass < 0)
 248                 tclass = 0;
 249
 250         *(__be32 *)hdr = htonl(0x60000000 | (tclass << 20)) | fl->fl6_flowlabel;
 251
 252         hdr->payload_len = htons(seg_len);
 253         hdr->nexthdr = proto;
 254         hdr->hop_limit = hlimit;
 255
 256         ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
 257         ipv6_addr_copy(&hdr->daddr, first_hop);
 258
 259         skb->priority = sk->sk_priority;
 260         skb->mark = sk->sk_mark;
 261
 262         mtu = dst_mtu(dst);
 263         if ((skb->len <= mtu) || ipfragok || skb_is_gso(skb)) {
 264                 IP6_INC_STATS(ip6_dst_idev(skb->dst),
 265                               IPSTATS_MIB_OUTREQUESTS);
 266                 return NF_HOOK(PF_INET6, NF_INET_LOCAL_OUT, skb, NULL, dst->dev,
 267                                 dst_output);
 268         }
 269
 270         if (net_ratelimit())
 271                 printk(KERN_DEBUG "IPv6: sending pkt_too_big to self\n");
 272         skb->dev = dst->dev;
 273         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev);
 274         IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGFAILS);
 275         kfree_skb(skb);
 276         return -EMSGSIZE;
 277 }
 278
 279 EXPORT_SYMBOL(ip6_xmit);
 280
 281 /*
 282  *      To avoid extra problems ND packets are send through this
 283  *      routine. It's code duplication but I really want to avoid
 284  *      extra checks since ipv6_build_header is used by TCP (which
 285  *      is for us performance critical)
 286  */
 287
 288 int ip6_nd_hdr(struct sock *sk, struct sk_buff *skb, struct net_device *dev,
 289                struct in6_addr *saddr, struct in6_addr *daddr,
 290                int proto, int len)
 291 {
 292         struct ipv6_pinfo *np = inet6_sk(sk);
 293         struct ipv6hdr *hdr;
 294         int totlen;
 295
 296         skb->protocol = htons(ETH_P_IPV6);
 297         skb->dev = dev;
 298
 299         totlen = len + sizeof(struct ipv6hdr);
 300
 301         skb_reset_network_header(skb);
 302         skb_put(skb, sizeof(struct ipv6hdr));
 303         hdr = ipv6_hdr(skb);
 304
 305         *(__be32*)hdr = htonl(0x60000000);
 306
 307         hdr->payload_len = htons(len);
 308         hdr->nexthdr = proto;
 309         hdr->hop_limit = np->hop_limit;
 310
 311         ipv6_addr_copy(&hdr->saddr, saddr);
 312         ipv6_addr_copy(&hdr->daddr, daddr);
 313
 314         return 0;
 315 }
 316
 317 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
 318 {
 319         struct ip6_ra_chain *ra;
 320         struct sock *last = NULL;
 321
 322         read_lock(&ip6_ra_lock);
 323         for (ra = ip6_ra_chain; ra; ra = ra->next) {
 324                 struct sock *sk = ra->sk;
 325                 if (sk && ra->sel == sel &&
 326                     (!sk->sk_bound_dev_if ||
 327                      sk->sk_bound_dev_if == skb->dev->ifindex)) {
 328                         if (last) {
 329                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
 330                                 if (skb2)
 331                                         rawv6_rcv(last, skb2);
 332                         }
 333                         last = sk;
 334                 }
 335         }
 336
 337         if (last) {
 338                 rawv6_rcv(last, skb);
 339                 read_unlock(&ip6_ra_lock);
 340                 return 1;
 341         }
 342         read_unlock(&ip6_ra_lock);
 343         return 0;
 344 }
 345
 346 static int ip6_forward_proxy_check(struct sk_buff *skb)
 347 {
 348         struct ipv6hdr *hdr = ipv6_hdr(skb);
 349         u8 nexthdr = hdr->nexthdr;
 350         int offset;
 351
 352         if (ipv6_ext_hdr(nexthdr)) {
 353                 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr);
 354                 if (offset < 0)
 355                         return 0;
 356         } else
 357                 offset = sizeof(struct ipv6hdr);
 358
 359         if (nexthdr == IPPROTO_ICMPV6) {
 360                 struct icmp6hdr *icmp6;
 361
 362                 if (!pskb_may_pull(skb, (skb_network_header(skb) +
 363                                          offset + 1 - skb->data)))
 364                         return 0;
 365
 366                 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
 367
 368                 switch (icmp6->icmp6_type) {
 369                 case NDISC_ROUTER_SOLICITATION:
 370                 case NDISC_ROUTER_ADVERTISEMENT:
 371                 case NDISC_NEIGHBOUR_SOLICITATION:
 372                 case NDISC_NEIGHBOUR_ADVERTISEMENT:
 373                 case NDISC_REDIRECT:
 374                         /* For reaction involving unicast neighbor discovery
 375                          * message destined to the proxied address, pass it to
 376                          * input function.
 377                          */
 378                         return 1;
 379                 default:
 380                         break;
 381                 }
 382         }
 383
 384         /*
 385          * The proxying router can't forward traffic sent to a link-local
 386          * address, so signal the sender and discard the packet. This
 387          * behavior is clarified by the MIPv6 specification.
 388          */
 389         if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
 390                 dst_link_failure(skb);
 391                 return -1;
 392         }
 393
 394         return 0;
 395 }
 396
 397 static inline int ip6_forward_finish(struct sk_buff *skb)
 398 {
 399         return dst_output(skb);
 400 }
 401
 402 int ip6_forward(struct sk_buff *skb)
 403 {
 404         struct dst_entry *dst = skb->dst;
 405         struct ipv6hdr *hdr = ipv6_hdr(skb);
 406         struct inet6_skb_parm *opt = IP6CB(skb);
 407         struct net *net = dst->dev->nd_net;
 408
 409         if (ipv6_devconf.forwarding == 0)
 410                 goto error;
 411
 412         if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
 413                 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
 414                 goto drop;
 415         }
 416
 417         skb_forward_csum(skb);
 418
 419         /*
 420          *      We DO NOT make any processing on
 421          *      RA packets, pushing them to user level AS IS
 422          *      without ane WARRANTY that application will be able
 423          *      to interpret them. The reason is that we
 424          *      cannot make anything clever here.
 425          *
 426          *      We are not end-node, so that if packet contains
 427          *      AH/ESP, we cannot make anything.
 428          *      Defragmentation also would be mistake, RA packets
 429          *      cannot be fragmented, because there is no warranty
 430          *      that different fragments will go along one path. --ANK
 431          */
 432         if (opt->ra) {
 433                 u8 *ptr = skb_network_header(skb) + opt->ra;
 434                 if (ip6_call_ra_chain(skb, (ptr[2]<<8) + ptr[3]))
 435                         return 0;
 436         }
 437
 438         /*
 439          *      check and decrement ttl
 440          */
 441         if (hdr->hop_limit <= 1) {
 442                 /* Force OUTPUT device used as source address */
 443                 skb->dev = dst->dev;
 444                 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT,
 445                             0, skb->dev);
 446                 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INHDRERRORS);
 447
 448                 kfree_skb(skb);
 449                 return -ETIMEDOUT;
 450         }
 451
 452         /* XXX: idev->cnf.proxy_ndp? */
 453         if (ipv6_devconf.proxy_ndp &&
 454             pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
 455                 int proxied = ip6_forward_proxy_check(skb);
 456                 if (proxied > 0)
 457                         return ip6_input(skb);
 458                 else if (proxied < 0) {
 459                         IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
 460                         goto drop;
 461                 }
 462         }
 463
 464         if (!xfrm6_route_forward(skb)) {
 465                 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
 466                 goto drop;
 467         }
 468         dst = skb->dst;
 469
 470         /* IPv6 specs say nothing about it, but it is clear that we cannot
 471            send redirects to source routed frames.
 472            We don't send redirects to frames decapsulated from IPsec.
 473          */
 474         if (skb->dev == dst->dev && dst->neighbour && opt->srcrt == 0 &&
 475             !skb->sp) {
 476                 struct in6_addr *target = NULL;
 477                 struct rt6_info *rt;
 478                 struct neighbour *n = dst->neighbour;
 479
 480                 /*
 481                  *      incoming and outgoing devices are the same
 482                  *      send a redirect.
 483                  */
 484
 485                 rt = (struct rt6_info *) dst;
 486                 if ((rt->rt6i_flags & RTF_GATEWAY))
 487                         target = (struct in6_addr*)&n->primary_key;
 488                 else
 489                         target = &hdr->daddr;
 490
 491                 /* Limit redirects both by destination (here)
 492                    and by source (inside ndisc_send_redirect)
 493                  */
 494                 if (xrlim_allow(dst, 1*HZ))
 495                         ndisc_send_redirect(skb, n, target);
 496         } else {
 497                 int addrtype = ipv6_addr_type(&hdr->saddr);
 498
 499                 /* This check is security critical. */
 500                 if (addrtype & (IPV6_ADDR_MULTICAST|IPV6_ADDR_LOOPBACK))
 501                         goto error;
 502                 if (addrtype & IPV6_ADDR_LINKLOCAL) {
 503                         icmpv6_send(skb, ICMPV6_DEST_UNREACH,
 504                                 ICMPV6_NOT_NEIGHBOUR, 0, skb->dev);
 505                         goto error;
 506                 }
 507         }
 508
 509         if (skb->len > dst_mtu(dst)) {
 510                 /* Again, force OUTPUT device used as source address */
 511                 skb->dev = dst->dev;
 512                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, dst_mtu(dst), skb->dev);
 513                 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INTOOBIGERRORS);
 514                 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_FRAGFAILS);
 515                 kfree_skb(skb);
 516                 return -EMSGSIZE;
 517         }
 518
 519         if (skb_cow(skb, dst->dev->hard_header_len)) {
 520                 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_OUTDISCARDS);
 521                 goto drop;
 522         }
 523
 524         hdr = ipv6_hdr(skb);
 525
 526         /* Mangling hops number delayed to point after skb COW */
 527
 528         hdr->hop_limit--;
 529
 530         IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
 531         return NF_HOOK(PF_INET6, NF_INET_FORWARD, skb, skb->dev, dst->dev,
 532                        ip6_forward_finish);
 533
 534 error:
 535         IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
 536 drop:
 537         kfree_skb(skb);
 538         return -EINVAL;
 539 }
 540
 541 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
 542 {
 543         to->pkt_type = from->pkt_type;
 544         to->priority = from->priority;
 545         to->protocol = from->protocol;
 546         dst_release(to->dst);
 547         to->dst = dst_clone(from->dst);
 548         to->dev = from->dev;
 549         to->mark = from->mark;
 550
 551 #ifdef CONFIG_NET_SCHED
 552         to->tc_index = from->tc_index;
 553 #endif
 554         nf_copy(to, from);
 555 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
 556     defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
 557         to->nf_trace = from->nf_trace;
 558 #endif
 559         skb_copy_secmark(to, from);
 560 }
 561
 562 int ip6_find_1stfragopt(struct sk_buff *skb, u8 **nexthdr)
 563 {
 564         u16 offset = sizeof(struct ipv6hdr);
 565         struct ipv6_opt_hdr *exthdr =
 566                                 (struct ipv6_opt_hdr *)(ipv6_hdr(skb) + 1);
 567         unsigned int packet_len = skb->tail - skb->network_header;
 568         int found_rhdr = 0;
 569         *nexthdr = &ipv6_hdr(skb)->nexthdr;
 570
 571         while (offset + 1 <= packet_len) {
 572
 573                 switch (**nexthdr) {
 574
 575                 case NEXTHDR_HOP:
 576                         break;
 577                 case NEXTHDR_ROUTING:
 578                         found_rhdr = 1;
 579                         break;
 580                 case NEXTHDR_DEST:
 581 #if defined(CONFIG_IPV6_MIP6) || defined(CONFIG_IPV6_MIP6_MODULE)
 582                         if (ipv6_find_tlv(skb, offset, IPV6_TLV_HAO) >= 0)
 583                                 break;
 584 #endif
 585                         if (found_rhdr)
 586                                 return offset;
 587                         break;
 588                 default :
 589                         return offset;
 590                 }
 591
 592                 offset += ipv6_optlen(exthdr);
 593                 *nexthdr = &exthdr->nexthdr;
 594                 exthdr = (struct ipv6_opt_hdr *)(skb_network_header(skb) +
 595                                                  offset);
 596         }
 597
 598         return offset;
 599 }
 600
 601 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
 602 {
 603         struct net_device *dev;
 604         struct sk_buff *frag;
 605         struct rt6_info *rt = (struct rt6_info*)skb->dst;
 606         struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
 607         struct ipv6hdr *tmp_hdr;
 608         struct frag_hdr *fh;
 609         unsigned int mtu, hlen, left, len;
 610         __be32 frag_id = 0;
 611         int ptr, offset = 0, err=0;
 612         u8 *prevhdr, nexthdr = 0;
 613
 614         dev = rt->u.dst.dev;
 615         hlen = ip6_find_1stfragopt(skb, &prevhdr);
 616         nexthdr = *prevhdr;
 617
 618         mtu = ip6_skb_dst_mtu(skb);
 619
 620         /* We must not fragment if the socket is set to force MTU discovery
 621          * or if the skb it not generated by a local socket.  (This last
 622          * check should be redundant, but it's free.)
 623          */
 624         if (!skb->local_df) {
 625                 skb->dev = skb->dst->dev;
 626                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev);
 627                 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGFAILS);
 628                 kfree_skb(skb);
 629                 return -EMSGSIZE;
 630         }
 631
 632         if (np && np->frag_size < mtu) {
 633                 if (np->frag_size)
 634                         mtu = np->frag_size;
 635         }
 636         mtu -= hlen + sizeof(struct frag_hdr);
 637
 638         if (skb_shinfo(skb)->frag_list) {
 639                 int first_len = skb_pagelen(skb);
 640                 int truesizes = 0;
 641
 642                 if (first_len - hlen > mtu ||
 643                     ((first_len - hlen) & 7) ||
 644                     skb_cloned(skb))
 645                         goto slow_path;
 646
 647                 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
 648                         /* Correct geometry. */
 649                         if (frag->len > mtu ||
 650                             ((frag->len & 7) && frag->next) ||
 651                             skb_headroom(frag) < hlen)
 652                             goto slow_path;
 653
 654                         /* Partially cloned skb? */
 655                         if (skb_shared(frag))
 656                                 goto slow_path;
 657
 658                         BUG_ON(frag->sk);
 659                         if (skb->sk) {
 660                                 sock_hold(skb->sk);
 661                                 frag->sk = skb->sk;
 662                                 frag->destructor = sock_wfree;
 663                                 truesizes += frag->truesize;
 664                         }
 665                 }
 666
 667                 err = 0;
 668                 offset = 0;
 669                 frag = skb_shinfo(skb)->frag_list;
 670                 skb_shinfo(skb)->frag_list = NULL;
 671                 /* BUILD HEADER */
 672
 673                 *prevhdr = NEXTHDR_FRAGMENT;
 674                 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
 675                 if (!tmp_hdr) {
 676                         IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGFAILS);
 677                         return -ENOMEM;
 678                 }
 679
 680                 __skb_pull(skb, hlen);
 681                 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
 682                 __skb_push(skb, hlen);
 683                 skb_reset_network_header(skb);
 684                 memcpy(skb_network_header(skb), tmp_hdr, hlen);
 685
 686                 ipv6_select_ident(skb, fh);
 687                 fh->nexthdr = nexthdr;
 688                 fh->reserved = 0;
 689                 fh->frag_off = htons(IP6_MF);
 690                 frag_id = fh->identification;
 691
 692                 first_len = skb_pagelen(skb);
 693                 skb->data_len = first_len - skb_headlen(skb);
 694                 skb->truesize -= truesizes;
 695                 skb->len = first_len;
 696                 ipv6_hdr(skb)->payload_len = htons(first_len -
 697                                                    sizeof(struct ipv6hdr));
 698
 699                 dst_hold(&rt->u.dst);
 700
 701                 for (;;) {
 702                         /* Prepare header of the next frame,
 703                          * before previous one went down. */
 704                         if (frag) {
 705                                 frag->ip_summed = CHECKSUM_NONE;
 706                                 skb_reset_transport_header(frag);
 707                                 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
 708                                 __skb_push(frag, hlen);
 709                                 skb_reset_network_header(frag);
 710                                 memcpy(skb_network_header(frag), tmp_hdr,
 711                                        hlen);
 712                                 offset += skb->len - hlen - sizeof(struct frag_hdr);
 713                                 fh->nexthdr = nexthdr;
 714                                 fh->reserved = 0;
 715                                 fh->frag_off = htons(offset);
 716                                 if (frag->next != NULL)
 717                                         fh->frag_off |= htons(IP6_MF);
 718                                 fh->identification = frag_id;
 719                                 ipv6_hdr(frag)->payload_len =
 720                                                 htons(frag->len -
 721                                                       sizeof(struct ipv6hdr));
 722                                 ip6_copy_metadata(frag, skb);
 723                         }
 724
 725                         err = output(skb);
 726                         if(!err)
 727                                 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGCREATES);
 728
 729                         if (err || !frag)
 730                                 break;
 731
 732                         skb = frag;
 733                         frag = skb->next;
 734                         skb->next = NULL;
 735                 }
 736
 737                 kfree(tmp_hdr);
 738
 739                 if (err == 0) {
 740                         IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGOKS);
 741                         dst_release(&rt->u.dst);
 742                         return 0;
 743                 }
 744
 745                 while (frag) {
 746                         skb = frag->next;
 747                         kfree_skb(frag);
 748                         frag = skb;
 749                 }
 750
 751                 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGFAILS);
 752                 dst_release(&rt->u.dst);
 753                 return err;
 754         }
 755
 756 slow_path:
 757         left = skb->len - hlen;         /* Space per frame */
 758         ptr = hlen;                     /* Where to start from */
 759
 760         /*
 761          *      Fragment the datagram.
 762          */
 763
 764         *prevhdr = NEXTHDR_FRAGMENT;
 765
 766         /*
 767          *      Keep copying data until we run out.
 768          */
 769         while(left > 0) {
 770                 len = left;
 771                 /* IF: it doesn't fit, use 'mtu' - the data space left */
 772                 if (len > mtu)
 773                         len = mtu;
 774                 /* IF: we are not sending upto and including the packet end
 775                    then align the next start on an eight byte boundary */
 776                 if (len < left) {
 777                         len &= ~7;
 778                 }
 779                 /*
 780                  *      Allocate buffer.
 781                  */
 782
 783                 if ((frag = alloc_skb(len+hlen+sizeof(struct frag_hdr)+LL_RESERVED_SPACE(rt->u.dst.dev), GFP_ATOMIC)) == NULL) {
 784                         NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
 785                         IP6_INC_STATS(ip6_dst_idev(skb->dst),
 786                                       IPSTATS_MIB_FRAGFAILS);
 787                         err = -ENOMEM;
 788                         goto fail;
 789                 }
 790
 791                 /*
 792                  *      Set up data on packet
 793                  */
 794
 795                 ip6_copy_metadata(frag, skb);
 796                 skb_reserve(frag, LL_RESERVED_SPACE(rt->u.dst.dev));
 797                 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
 798                 skb_reset_network_header(frag);
 799                 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
 800                 frag->transport_header = (frag->network_header + hlen +
 801                                           sizeof(struct frag_hdr));
 802
 803                 /*
 804                  *      Charge the memory for the fragment to any owner
 805                  *      it might possess
 806                  */
 807                 if (skb->sk)
 808                         skb_set_owner_w(frag, skb->sk);
 809
 810                 /*
 811                  *      Copy the packet header into the new buffer.
 812                  */
 813                 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
 814
 815                 /*
 816                  *      Build fragment header.
 817                  */
 818                 fh->nexthdr = nexthdr;
 819                 fh->reserved = 0;
 820                 if (!frag_id) {
 821                         ipv6_select_ident(skb, fh);
 822                         frag_id = fh->identification;
 823                 } else
 824                         fh->identification = frag_id;
 825
 826                 /*
 827                  *      Copy a block of the IP datagram.
 828                  */
 829                 if (skb_copy_bits(skb, ptr, skb_transport_header(frag), len))
 830                         BUG();
 831                 left -= len;
 832
 833                 fh->frag_off = htons(offset);
 834                 if (left > 0)
 835                         fh->frag_off |= htons(IP6_MF);
 836                 ipv6_hdr(frag)->payload_len = htons(frag->len -
 837                                                     sizeof(struct ipv6hdr));
 838
 839                 ptr += len;
 840                 offset += len;
 841
 842                 /*
 843                  *      Put this fragment into the sending queue.
 844                  */
 845                 err = output(frag);
 846                 if (err)
 847                         goto fail;
 848
 849                 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGCREATES);
 850         }
 851         IP6_INC_STATS(ip6_dst_idev(skb->dst),
 852                       IPSTATS_MIB_FRAGOKS);
 853         kfree_skb(skb);
 854         return err;
 855
 856 fail:
 857         IP6_INC_STATS(ip6_dst_idev(skb->dst),
 858                       IPSTATS_MIB_FRAGFAILS);
 859         kfree_skb(skb);
 860         return err;
 861 }
 862
 863 static inline int ip6_rt_check(struct rt6key *rt_key,
 864                                struct in6_addr *fl_addr,
 865                                struct in6_addr *addr_cache)
 866 {
 867         return ((rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
 868                 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache)));
 869 }
 870
 871 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
 872                                           struct dst_entry *dst,
 873                                           struct flowi *fl)
 874 {
 875         struct ipv6_pinfo *np = inet6_sk(sk);
 876         struct rt6_info *rt = (struct rt6_info *)dst;
 877
 878         if (!dst)
 879                 goto out;
 880
 881         /* Yes, checking route validity in not connected
 882          * case is not very simple. Take into account,
 883          * that we do not support routing by source, TOS,
 884          * and MSG_DONTROUTE            --ANK (980726)
 885          *
 886          * 1. ip6_rt_check(): If route was host route,
 887          *    check that cached destination is current.
 888          *    If it is network route, we still may
 889          *    check its validity using saved pointer
 890          *    to the last used address: daddr_cache.
 891          *    We do not want to save whole address now,
 892          *    (because main consumer of this service
 893          *    is tcp, which has not this problem),
 894          *    so that the last trick works only on connected
 895          *    sockets.
 896          * 2. oif also should be the same.
 897          */
 898         if (ip6_rt_check(&rt->rt6i_dst, &fl->fl6_dst, np->daddr_cache) ||
 899 #ifdef CONFIG_IPV6_SUBTREES
 900             ip6_rt_check(&rt->rt6i_src, &fl->fl6_src, np->saddr_cache) ||
 901 #endif
 902             (fl->oif && fl->oif != dst->dev->ifindex)) {
 903                 dst_release(dst);
 904                 dst = NULL;
 905         }
 906
 907 out:
 908         return dst;
 909 }
 910
 911 static int ip6_dst_lookup_tail(struct sock *sk,
 912                                struct dst_entry **dst, struct flowi *fl)
 913 {
 914         int err;
 915         struct net *net = sk->sk_net;
 916
 917         if (*dst == NULL)
 918                 *dst = ip6_route_output(net, sk, fl);
 919
 920         if ((err = (*dst)->error))
 921                 goto out_err_release;
 922
 923         if (ipv6_addr_any(&fl->fl6_src)) {
 924                 err = ipv6_dev_get_saddr(ip6_dst_idev(*dst)->dev,
 925                                          &fl->fl6_dst, &fl->fl6_src);
 926                 if (err)
 927                         goto out_err_release;
 928         }
 929
 930 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
 931                 /*
 932                  * Here if the dst entry we've looked up
 933                  * has a neighbour entry that is in the INCOMPLETE
 934                  * state and the src address from the flow is
 935                  * marked as OPTIMISTIC, we release the found
 936                  * dst entry and replace it instead with the
 937                  * dst entry of the nexthop router
 938                  */
 939                 if (!((*dst)->neighbour->nud_state & NUD_VALID)) {
 940                         struct inet6_ifaddr *ifp;
 941                         struct flowi fl_gw;
 942                         int redirect;
 943
 944                         ifp = ipv6_get_ifaddr(net, &fl->fl6_src,
 945                                               (*dst)->dev, 1);
 946
 947                         redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
 948                         if (ifp)
 949                                 in6_ifa_put(ifp);
 950
 951                         if (redirect) {
 952                                 /*
 953                                  * We need to get the dst entry for the
 954                                  * default router instead
 955                                  */
 956                                 dst_release(*dst);
 957                                 memcpy(&fl_gw, fl, sizeof(struct flowi));
 958                                 memset(&fl_gw.fl6_dst, 0, sizeof(struct in6_addr));
 959                                 *dst = ip6_route_output(net, sk, &fl_gw);
 960                                 if ((err = (*dst)->error))
 961                                         goto out_err_release;
 962                         }
 963                 }
 964 #endif
 965
 966         return 0;
 967
 968 out_err_release:
 969         if (err == -ENETUNREACH)
 970                 IP6_INC_STATS_BH(NULL, IPSTATS_MIB_OUTNOROUTES);
 971         dst_release(*dst);
 972         *dst = NULL;
 973         return err;
 974 }
 975
 976 /**
 977  *      ip6_dst_lookup - perform route lookup on flow
 978  *      @sk: socket which provides route info
 979  *      @dst: pointer to dst_entry * for result
 980  *      @fl: flow to lookup
 981  *
 982  *      This function performs a route lookup on the given flow.
 983  *
 984  *      It returns zero on success, or a standard errno code on error.
 985  */
 986 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
 987 {
 988         *dst = NULL;
 989         return ip6_dst_lookup_tail(sk, dst, fl);
 990 }
 991 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
 992
 993 /**
 994  *      ip6_sk_dst_lookup - perform socket cached route lookup on flow
 995  *      @sk: socket which provides the dst cache and route info
 996  *      @dst: pointer to dst_entry * for result
 997  *      @fl: flow to lookup
 998  *
 999  *      This function performs a route lookup on the given flow with the
1000  *      possibility of using the cached route in the socket if it is valid.
1001  *      It will take the socket dst lock when operating on the dst cache.
1002  *      As a result, this function can only be used in process context.
1003  *
1004  *      It returns zero on success, or a standard errno code on error.
1005  */
1006 int ip6_sk_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
1007 {
1008         *dst = NULL;
1009         if (sk) {
1010                 *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1011                 *dst = ip6_sk_dst_check(sk, *dst, fl);
1012         }
1013
1014         return ip6_dst_lookup_tail(sk, dst, fl);
1015 }
1016 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup);
1017
1018 static inline int ip6_ufo_append_data(struct sock *sk,
1019                         int getfrag(void *from, char *to, int offset, int len,
1020                         int odd, struct sk_buff *skb),
1021                         void *from, int length, int hh_len, int fragheaderlen,
1022                         int transhdrlen, int mtu,unsigned int flags)
1023
1024 {
1025         struct sk_buff *skb;
1026         int err;
1027
1028         /* There is support for UDP large send offload by network
1029          * device, so create one single skb packet containing complete
1030          * udp datagram
1031          */
1032         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
1033                 skb = sock_alloc_send_skb(sk,
1034                         hh_len + fragheaderlen + transhdrlen + 20,
1035                         (flags & MSG_DONTWAIT), &err);
1036                 if (skb == NULL)
1037                         return -ENOMEM;
1038
1039                 /* reserve space for Hardware header */
1040                 skb_reserve(skb, hh_len);
1041
1042                 /* create space for UDP/IP header */
1043                 skb_put(skb,fragheaderlen + transhdrlen);
1044
1045                 /* initialize network header pointer */
1046                 skb_reset_network_header(skb);
1047
1048                 /* initialize protocol header pointer */
1049                 skb->transport_header = skb->network_header + fragheaderlen;
1050
1051                 skb->ip_summed = CHECKSUM_PARTIAL;
1052                 skb->csum = 0;
1053                 sk->sk_sndmsg_off = 0;
1054         }
1055
1056         err = skb_append_datato_frags(sk,skb, getfrag, from,
1057                                       (length - transhdrlen));
1058         if (!err) {
1059                 struct frag_hdr fhdr;
1060
1061                 /* specify the length of each IP datagram fragment*/
1062                 skb_shinfo(skb)->gso_size = mtu - fragheaderlen -
1063                                             sizeof(struct frag_hdr);
1064                 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1065                 ipv6_select_ident(skb, &fhdr);
1066                 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1067                 __skb_queue_tail(&sk->sk_write_queue, skb);
1068
1069                 return 0;
1070         }
1071         /* There is not enough support do UPD LSO,
1072          * so follow normal path
1073          */
1074         kfree_skb(skb);
1075
1076         return err;
1077 }
1078
1079 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1080         int offset, int len, int odd, struct sk_buff *skb),
1081         void *from, int length, int transhdrlen,
1082         int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi *fl,
1083         struct rt6_info *rt, unsigned int flags)
1084 {
1085         struct inet_sock *inet = inet_sk(sk);
1086         struct ipv6_pinfo *np = inet6_sk(sk);
1087         struct sk_buff *skb;
1088         unsigned int maxfraglen, fragheaderlen;
1089         int exthdrlen;
1090         int hh_len;
1091         int mtu;
1092         int copy;
1093         int err;
1094         int offset = 0;
1095         int csummode = CHECKSUM_NONE;
1096
1097         if (flags&MSG_PROBE)
1098                 return 0;
1099         if (skb_queue_empty(&sk->sk_write_queue)) {
1100                 /*
1101                  * setup for corking
1102                  */
1103                 if (opt) {
1104                         if (np->cork.opt == NULL) {
1105                                 np->cork.opt = kmalloc(opt->tot_len,
1106                                                        sk->sk_allocation);
1107                                 if (unlikely(np->cork.opt == NULL))
1108                                         return -ENOBUFS;
1109                         } else if (np->cork.opt->tot_len < opt->tot_len) {
1110                                 printk(KERN_DEBUG "ip6_append_data: invalid option length\n");
1111                                 return -EINVAL;
1112                         }
1113                         memcpy(np->cork.opt, opt, opt->tot_len);
1114                         inet->cork.flags |= IPCORK_OPT;
1115                         /* need source address above miyazawa*/
1116                 }
1117                 dst_hold(&rt->u.dst);
1118                 np->cork.rt = rt;
1119                 inet->cork.fl = *fl;
1120                 np->cork.hop_limit = hlimit;
1121                 np->cork.tclass = tclass;
1122                 mtu = np->pmtudisc == IPV6_PMTUDISC_PROBE ?
1123                       rt->u.dst.dev->mtu : dst_mtu(rt->u.dst.path);
1124                 if (np->frag_size < mtu) {
1125                         if (np->frag_size)
1126                                 mtu = np->frag_size;
1127                 }
1128                 inet->cork.fragsize = mtu;
1129                 if (dst_allfrag(rt->u.dst.path))
1130                         inet->cork.flags |= IPCORK_ALLFRAG;
1131                 inet->cork.length = 0;
1132                 sk->sk_sndmsg_page = NULL;
1133                 sk->sk_sndmsg_off = 0;
1134                 exthdrlen = rt->u.dst.header_len + (opt ? opt->opt_flen : 0) -
1135                             rt->rt6i_nfheader_len;
1136                 length += exthdrlen;
1137                 transhdrlen += exthdrlen;
1138         } else {
1139                 rt = np->cork.rt;
1140                 fl = &inet->cork.fl;
1141                 if (inet->cork.flags & IPCORK_OPT)
1142                         opt = np->cork.opt;
1143                 transhdrlen = 0;
1144                 exthdrlen = 0;
1145                 mtu = inet->cork.fragsize;
1146         }
1147
1148         hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1149
1150         fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1151                         (opt ? opt->opt_nflen : 0);
1152         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
1153
1154         if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1155                 if (inet->cork.length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
1156                         ipv6_local_error(sk, EMSGSIZE, fl, mtu-exthdrlen);
1157                         return -EMSGSIZE;
1158                 }
1159         }
1160
1161         /*
1162          * Let's try using as much space as possible.
1163          * Use MTU if total length of the message fits into the MTU.
1164          * Otherwise, we need to reserve fragment header and
1165          * fragment alignment (= 8-15 octects, in total).
1166          *
1167          * Note that we may need to "move" the data from the tail of
1168          * of the buffer to the new fragment when we split
1169          * the message.
1170          *
1171          * FIXME: It may be fragmented into multiple chunks
1172          *        at once if non-fragmentable extension headers
1173          *        are too large.
1174          * --yoshfuji
1175          */
1176
1177         inet->cork.length += length;
1178         if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
1179             (rt->u.dst.dev->features & NETIF_F_UFO)) {
1180
1181                 err = ip6_ufo_append_data(sk, getfrag, from, length, hh_len,
1182                                           fragheaderlen, transhdrlen, mtu,
1183                                           flags);
1184                 if (err)
1185                         goto error;
1186                 return 0;
1187         }
1188
1189         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1190                 goto alloc_new_skb;
1191
1192         while (length > 0) {
1193                 /* Check if the remaining data fits into current packet. */
1194                 copy = (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1195                 if (copy < length)
1196                         copy = maxfraglen - skb->len;
1197
1198                 if (copy <= 0) {
1199                         char *data;
1200                         unsigned int datalen;
1201                         unsigned int fraglen;
1202                         unsigned int fraggap;
1203                         unsigned int alloclen;
1204                         struct sk_buff *skb_prev;
1205 alloc_new_skb:
1206                         skb_prev = skb;
1207
1208                         /* There's no room in the current skb */
1209                         if (skb_prev)
1210                                 fraggap = skb_prev->len - maxfraglen;
1211                         else
1212                                 fraggap = 0;
1213
1214                         /*
1215                          * If remaining data exceeds the mtu,
1216                          * we know we need more fragment(s).
1217                          */
1218                         datalen = length + fraggap;
1219                         if (datalen > (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1220                                 datalen = maxfraglen - fragheaderlen;
1221
1222                         fraglen = datalen + fragheaderlen;
1223                         if ((flags & MSG_MORE) &&
1224                             !(rt->u.dst.dev->features&NETIF_F_SG))
1225                                 alloclen = mtu;
1226                         else
1227                                 alloclen = datalen + fragheaderlen;
1228
1229                         /*
1230                          * The last fragment gets additional space at tail.
1231                          * Note: we overallocate on fragments with MSG_MODE
1232                          * because we have no idea if we're the last one.
1233                          */
1234                         if (datalen == length + fraggap)
1235                                 alloclen += rt->u.dst.trailer_len;
1236
1237                         /*
1238                          * We just reserve space for fragment header.
1239                          * Note: this may be overallocation if the message
1240                          * (without MSG_MORE) fits into the MTU.
1241                          */
1242                         alloclen += sizeof(struct frag_hdr);
1243
1244                         if (transhdrlen) {
1245                                 skb = sock_alloc_send_skb(sk,
1246                                                 alloclen + hh_len,
1247                                                 (flags & MSG_DONTWAIT), &err);
1248                         } else {
1249                                 skb = NULL;
1250                                 if (atomic_read(&sk->sk_wmem_alloc) <=
1251                                     2 * sk->sk_sndbuf)
1252                                         skb = sock_wmalloc(sk,
1253                                                            alloclen + hh_len, 1,
1254                                                            sk->sk_allocation);
1255                                 if (unlikely(skb == NULL))
1256                                         err = -ENOBUFS;
1257                         }
1258                         if (skb == NULL)
1259                                 goto error;
1260                         /*
1261                          *      Fill in the control structures
1262                          */
1263                         skb->ip_summed = csummode;
1264                         skb->csum = 0;
1265                         /* reserve for fragmentation */
1266                         skb_reserve(skb, hh_len+sizeof(struct frag_hdr));
1267
1268                         /*
1269                          *      Find where to start putting bytes
1270                          */
1271                         data = skb_put(skb, fraglen);
1272                         skb_set_network_header(skb, exthdrlen);
1273                         data += fragheaderlen;
1274                         skb->transport_header = (skb->network_header +
1275                                                  fragheaderlen);
1276                         if (fraggap) {
1277                                 skb->csum = skb_copy_and_csum_bits(
1278                                         skb_prev, maxfraglen,
1279                                         data + transhdrlen, fraggap, 0);
1280                                 skb_prev->csum = csum_sub(skb_prev->csum,
1281                                                           skb->csum);
1282                                 data += fraggap;
1283                                 pskb_trim_unique(skb_prev, maxfraglen);
1284                         }
1285                         copy = datalen - transhdrlen - fraggap;
1286                         if (copy < 0) {
1287                                 err = -EINVAL;
1288                                 kfree_skb(skb);
1289                                 goto error;
1290                         } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1291                                 err = -EFAULT;
1292                                 kfree_skb(skb);
1293                                 goto error;
1294                         }
1295
1296                         offset += copy;
1297                         length -= datalen - fraggap;
1298                         transhdrlen = 0;
1299                         exthdrlen = 0;
1300                         csummode = CHECKSUM_NONE;
1301
1302                         /*
1303                          * Put the packet on the pending queue
1304                          */
1305                         __skb_queue_tail(&sk->sk_write_queue, skb);
1306                         continue;
1307                 }
1308
1309                 if (copy > length)
1310                         copy = length;
1311
1312                 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
1313                         unsigned int off;
1314
1315                         off = skb->len;
1316                         if (getfrag(from, skb_put(skb, copy),
1317                                                 offset, copy, off, skb) < 0) {
1318                                 __skb_trim(skb, off);
1319                                 err = -EFAULT;
1320                                 goto error;
1321                         }
1322                 } else {
1323                         int i = skb_shinfo(skb)->nr_frags;
1324                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1325                         struct page *page = sk->sk_sndmsg_page;
1326                         int off = sk->sk_sndmsg_off;
1327                         unsigned int left;
1328
1329                         if (page && (left = PAGE_SIZE - off) > 0) {
1330                                 if (copy >= left)
1331                                         copy = left;
1332                                 if (page != frag->page) {
1333                                         if (i == MAX_SKB_FRAGS) {
1334                                                 err = -EMSGSIZE;
1335                                                 goto error;
1336                                         }
1337                                         get_page(page);
1338                                         skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1339                                         frag = &skb_shinfo(skb)->frags[i];
1340                                 }
1341                         } else if(i < MAX_SKB_FRAGS) {
1342                                 if (copy > PAGE_SIZE)
1343                                         copy = PAGE_SIZE;
1344                                 page = alloc_pages(sk->sk_allocation, 0);
1345                                 if (page == NULL) {
1346                                         err = -ENOMEM;
1347                                         goto error;
1348                                 }
1349                                 sk->sk_sndmsg_page = page;
1350                                 sk->sk_sndmsg_off = 0;
1351
1352                                 skb_fill_page_desc(skb, i, page, 0, 0);
1353                                 frag = &skb_shinfo(skb)->frags[i];
1354                         } else {
1355                                 err = -EMSGSIZE;
1356                                 goto error;
1357                         }
1358                         if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1359                                 err = -EFAULT;
1360                                 goto error;
1361                         }
1362                         sk->sk_sndmsg_off += copy;
1363                         frag->size += copy;
1364                         skb->len += copy;
1365                         skb->data_len += copy;
1366                         skb->truesize += copy;
1367                         atomic_add(copy, &sk->sk_wmem_alloc);
1368                 }
1369                 offset += copy;
1370                 length -= copy;
1371         }
1372         return 0;
1373 error:
1374         inet->cork.length -= length;
1375         IP6_INC_STATS(rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1376         return err;
1377 }
1378
1379 static void ip6_cork_release(struct inet_sock *inet, struct ipv6_pinfo *np)
1380 {
1381         inet->cork.flags &= ~IPCORK_OPT;
1382         kfree(np->cork.opt);
1383         np->cork.opt = NULL;
1384         if (np->cork.rt) {
1385                 dst_release(&np->cork.rt->u.dst);
1386                 np->cork.rt = NULL;
1387                 inet->cork.flags &= ~IPCORK_ALLFRAG;
1388         }
1389         memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1390 }
1391
1392 int ip6_push_pending_frames(struct sock *sk)
1393 {
1394         struct sk_buff *skb, *tmp_skb;
1395         struct sk_buff **tail_skb;
1396         struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1397         struct inet_sock *inet = inet_sk(sk);
1398         struct ipv6_pinfo *np = inet6_sk(sk);
1399         struct ipv6hdr *hdr;
1400         struct ipv6_txoptions *opt = np->cork.opt;
1401         struct rt6_info *rt = np->cork.rt;
1402         struct flowi *fl = &inet->cork.fl;
1403         unsigned char proto = fl->proto;
1404         int err = 0;
1405
1406         if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1407                 goto out;
1408         tail_skb = &(skb_shinfo(skb)->frag_list);
1409
1410         /* move skb->data to ip header from ext header */
1411         if (skb->data < skb_network_header(skb))
1412                 __skb_pull(skb, skb_network_offset(skb));
1413         while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1414                 __skb_pull(tmp_skb, skb_network_header_len(skb));
1415                 *tail_skb = tmp_skb;
1416                 tail_skb = &(tmp_skb->next);
1417                 skb->len += tmp_skb->len;
1418                 skb->data_len += tmp_skb->len;
1419                 skb->truesize += tmp_skb->truesize;
1420                 __sock_put(tmp_skb->sk);
1421                 tmp_skb->destructor = NULL;
1422                 tmp_skb->sk = NULL;
1423         }
1424
1425         /* Allow local fragmentation. */
1426         if (np->pmtudisc < IPV6_PMTUDISC_DO)
1427                 skb->local_df = 1;
1428
1429         ipv6_addr_copy(final_dst, &fl->fl6_dst);
1430         __skb_pull(skb, skb_network_header_len(skb));
1431         if (opt && opt->opt_flen)
1432                 ipv6_push_frag_opts(skb, opt, &proto);
1433         if (opt && opt->opt_nflen)
1434                 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1435
1436         skb_push(skb, sizeof(struct ipv6hdr));
1437         skb_reset_network_header(skb);
1438         hdr = ipv6_hdr(skb);
1439
1440         *(__be32*)hdr = fl->fl6_flowlabel |
1441                      htonl(0x60000000 | ((int)np->cork.tclass << 20));
1442
1443         hdr->hop_limit = np->cork.hop_limit;
1444         hdr->nexthdr = proto;
1445         ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
1446         ipv6_addr_copy(&hdr->daddr, final_dst);
1447
1448         skb->priority = sk->sk_priority;
1449         skb->mark = sk->sk_mark;
1450
1451         skb->dst = dst_clone(&rt->u.dst);
1452         IP6_INC_STATS(rt->rt6i_idev, IPSTATS_MIB_OUTREQUESTS);
1453         if (proto == IPPROTO_ICMPV6) {
1454                 struct inet6_dev *idev = ip6_dst_idev(skb->dst);
1455
1456                 ICMP6MSGOUT_INC_STATS_BH(idev, icmp6_hdr(skb)->icmp6_type);
1457                 ICMP6_INC_STATS_BH(idev, ICMP6_MIB_OUTMSGS);
1458         }
1459
1460         err = ip6_local_out(skb);
1461         if (err) {
1462                 if (err > 0)
1463                         err = np->recverr ? net_xmit_errno(err) : 0;
1464                 if (err)
1465                         goto error;
1466         }
1467
1468 out:
1469         ip6_cork_release(inet, np);
1470         return err;
1471 error:
1472         goto out;
1473 }
1474
1475 void ip6_flush_pending_frames(struct sock *sk)
1476 {
1477         struct sk_buff *skb;
1478
1479         while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1480                 if (skb->dst)
1481                         IP6_INC_STATS(ip6_dst_idev(skb->dst),
1482                                       IPSTATS_MIB_OUTDISCARDS);
1483                 kfree_skb(skb);
1484         }
1485
1486         ip6_cork_release(inet_sk(sk), inet6_sk(sk));
1487 }