SAFE public projects git trees. - safe/jmp/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/types.h>
  33 #include <linux/string.h>
  34 #include <linux/socket.h>
  35 #include <linux/net.h>
  36 #include <linux/netdevice.h>
  37 #include <linux/if_arp.h>
  38 #include <linux/in6.h>
  39 #include <linux/tcp.h>
  40 #include <linux/route.h>
  41 #include <linux/module.h>
  42
  43 #include <linux/netfilter.h>
  44 #include <linux/netfilter_ipv6.h>
  45
  46 #include <net/sock.h>
  47 #include <net/snmp.h>
  48
  49 #include <net/ipv6.h>
  50 #include <net/ndisc.h>
  51 #include <net/protocol.h>
  52 #include <net/ip6_route.h>
  53 #include <net/addrconf.h>
  54 #include <net/rawv6.h>
  55 #include <net/icmp.h>
  56 #include <net/xfrm.h>
  57 #include <net/checksum.h>
  58
  59 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *));
  60
  61 static __inline__ void ipv6_select_ident(struct sk_buff *skb, struct frag_hdr *fhdr)
  62 {
  63         static u32 ipv6_fragmentation_id = 1;
  64         static DEFINE_SPINLOCK(ip6_id_lock);
  65
  66         spin_lock_bh(&ip6_id_lock);
  67         fhdr->identification = htonl(ipv6_fragmentation_id);
  68         if (++ipv6_fragmentation_id == 0)
  69                 ipv6_fragmentation_id = 1;
  70         spin_unlock_bh(&ip6_id_lock);
  71 }
  72
  73 static inline int ip6_output_finish(struct sk_buff *skb)
  74 {
  75
  76         struct dst_entry *dst = skb->dst;
  77         struct hh_cache *hh = dst->hh;
  78
  79         if (hh) {
  80                 int hh_alen;
  81
  82                 read_lock_bh(&hh->hh_lock);
  83                 hh_alen = HH_DATA_ALIGN(hh->hh_len);
  84                 memcpy(skb->data - hh_alen, hh->hh_data, hh_alen);
  85                 read_unlock_bh(&hh->hh_lock);
  86                 skb_push(skb, hh->hh_len);
  87                 return hh->hh_output(skb);
  88         } else if (dst->neighbour)
  89                 return dst->neighbour->output(skb);
  90
  91         IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
  92         kfree_skb(skb);
  93         return -EINVAL;
  94
  95 }
  96
  97 /* dev_loopback_xmit for use with netfilter. */
  98 static int ip6_dev_loopback_xmit(struct sk_buff *newskb)
  99 {
 100         newskb->mac.raw = newskb->data;
 101         __skb_pull(newskb, newskb->nh.raw - newskb->data);
 102         newskb->pkt_type = PACKET_LOOPBACK;
 103         newskb->ip_summed = CHECKSUM_UNNECESSARY;
 104         BUG_TRAP(newskb->dst);
 105
 106         netif_rx(newskb);
 107         return 0;
 108 }
 109
 110
 111 static int ip6_output2(struct sk_buff *skb)
 112 {
 113         struct dst_entry *dst = skb->dst;
 114         struct net_device *dev = dst->dev;
 115
 116         skb->protocol = htons(ETH_P_IPV6);
 117         skb->dev = dev;
 118
 119         if (ipv6_addr_is_multicast(&skb->nh.ipv6h->daddr)) {
 120                 struct ipv6_pinfo* np = skb->sk ? inet6_sk(skb->sk) : NULL;
 121                 struct inet6_dev *idev = ip6_dst_idev(skb->dst);
 122
 123                 if (!(dev->flags & IFF_LOOPBACK) && (!np || np->mc_loop) &&
 124                     ipv6_chk_mcast_addr(dev, &skb->nh.ipv6h->daddr,
 125                                 &skb->nh.ipv6h->saddr)) {
 126                         struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
 127
 128                         /* Do not check for IFF_ALLMULTI; multicast routing
 129                            is not supported in any case.
 130                          */
 131                         if (newskb)
 132                                 NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, newskb, NULL,
 133                                         newskb->dev,
 134                                         ip6_dev_loopback_xmit);
 135
 136                         if (skb->nh.ipv6h->hop_limit == 0) {
 137                                 IP6_INC_STATS(idev, IPSTATS_MIB_OUTDISCARDS);
 138                                 kfree_skb(skb);
 139                                 return 0;
 140                         }
 141                 }
 142
 143                 IP6_INC_STATS(idev, IPSTATS_MIB_OUTMCASTPKTS);
 144         }
 145
 146         return NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, skb,NULL, skb->dev,ip6_output_finish);
 147 }
 148
 149 int ip6_output(struct sk_buff *skb)
 150 {
 151         if ((skb->len > dst_mtu(skb->dst) && !skb_is_gso(skb)) ||
 152                                 dst_allfrag(skb->dst))
 153                 return ip6_fragment(skb, ip6_output2);
 154         else
 155                 return ip6_output2(skb);
 156 }
 157
 158 /*
 159  *      xmit an sk_buff (used by TCP)
 160  */
 161
 162 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
 163              struct ipv6_txoptions *opt, int ipfragok)
 164 {
 165         struct ipv6_pinfo *np = inet6_sk(sk);
 166         struct in6_addr *first_hop = &fl->fl6_dst;
 167         struct dst_entry *dst = skb->dst;
 168         struct ipv6hdr *hdr;
 169         u8  proto = fl->proto;
 170         int seg_len = skb->len;
 171         int hlimit, tclass;
 172         u32 mtu;
 173
 174         if (opt) {
 175                 int head_room;
 176
 177                 /* First: exthdrs may take lots of space (~8K for now)
 178                    MAX_HEADER is not enough.
 179                  */
 180                 head_room = opt->opt_nflen + opt->opt_flen;
 181                 seg_len += head_room;
 182                 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
 183
 184                 if (skb_headroom(skb) < head_room) {
 185                         struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
 186                         if (skb2 == NULL) {
 187                                 IP6_INC_STATS(ip6_dst_idev(skb->dst),
 188                                               IPSTATS_MIB_OUTDISCARDS);
 189                                 kfree_skb(skb);
 190                                 return -ENOBUFS;
 191                         }
 192                         kfree_skb(skb);
 193                         skb = skb2;
 194                         if (sk)
 195                                 skb_set_owner_w(skb, sk);
 196                 }
 197                 if (opt->opt_flen)
 198                         ipv6_push_frag_opts(skb, opt, &proto);
 199                 if (opt->opt_nflen)
 200                         ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
 201         }
 202
 203         hdr = skb->nh.ipv6h = (struct ipv6hdr*)skb_push(skb, sizeof(struct ipv6hdr));
 204
 205         /*
 206          *      Fill in the IPv6 header
 207          */
 208
 209         hlimit = -1;
 210         if (np)
 211                 hlimit = np->hop_limit;
 212         if (hlimit < 0)
 213                 hlimit = dst_metric(dst, RTAX_HOPLIMIT);
 214         if (hlimit < 0)
 215                 hlimit = ipv6_get_hoplimit(dst->dev);
 216
 217         tclass = -1;
 218         if (np)
 219                 tclass = np->tclass;
 220         if (tclass < 0)
 221                 tclass = 0;
 222
 223         *(__be32 *)hdr = htonl(0x60000000 | (tclass << 20)) | fl->fl6_flowlabel;
 224
 225         hdr->payload_len = htons(seg_len);
 226         hdr->nexthdr = proto;
 227         hdr->hop_limit = hlimit;
 228
 229         ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
 230         ipv6_addr_copy(&hdr->daddr, first_hop);
 231
 232         skb->priority = sk->sk_priority;
 233
 234         mtu = dst_mtu(dst);
 235         if ((skb->len <= mtu) || ipfragok || skb_is_gso(skb)) {
 236                 IP6_INC_STATS(ip6_dst_idev(skb->dst),
 237                               IPSTATS_MIB_OUTREQUESTS);
 238                 return NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, dst->dev,
 239                                 dst_output);
 240         }
 241
 242         if (net_ratelimit())
 243                 printk(KERN_DEBUG "IPv6: sending pkt_too_big to self\n");
 244         skb->dev = dst->dev;
 245         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev);
 246         IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGFAILS);
 247         kfree_skb(skb);
 248         return -EMSGSIZE;
 249 }
 250
 251 /*
 252  *      To avoid extra problems ND packets are send through this
 253  *      routine. It's code duplication but I really want to avoid
 254  *      extra checks since ipv6_build_header is used by TCP (which
 255  *      is for us performance critical)
 256  */
 257
 258 int ip6_nd_hdr(struct sock *sk, struct sk_buff *skb, struct net_device *dev,
 259                struct in6_addr *saddr, struct in6_addr *daddr,
 260                int proto, int len)
 261 {
 262         struct ipv6_pinfo *np = inet6_sk(sk);
 263         struct ipv6hdr *hdr;
 264         int totlen;
 265
 266         skb->protocol = htons(ETH_P_IPV6);
 267         skb->dev = dev;
 268
 269         totlen = len + sizeof(struct ipv6hdr);
 270
 271         hdr = (struct ipv6hdr *) skb_put(skb, sizeof(struct ipv6hdr));
 272         skb->nh.ipv6h = hdr;
 273
 274         *(__be32*)hdr = htonl(0x60000000);
 275
 276         hdr->payload_len = htons(len);
 277         hdr->nexthdr = proto;
 278         hdr->hop_limit = np->hop_limit;
 279
 280         ipv6_addr_copy(&hdr->saddr, saddr);
 281         ipv6_addr_copy(&hdr->daddr, daddr);
 282
 283         return 0;
 284 }
 285
 286 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
 287 {
 288         struct ip6_ra_chain *ra;
 289         struct sock *last = NULL;
 290
 291         read_lock(&ip6_ra_lock);
 292         for (ra = ip6_ra_chain; ra; ra = ra->next) {
 293                 struct sock *sk = ra->sk;
 294                 if (sk && ra->sel == sel &&
 295                     (!sk->sk_bound_dev_if ||
 296                      sk->sk_bound_dev_if == skb->dev->ifindex)) {
 297                         if (last) {
 298                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
 299                                 if (skb2)
 300                                         rawv6_rcv(last, skb2);
 301                         }
 302                         last = sk;
 303                 }
 304         }
 305
 306         if (last) {
 307                 rawv6_rcv(last, skb);
 308                 read_unlock(&ip6_ra_lock);
 309                 return 1;
 310         }
 311         read_unlock(&ip6_ra_lock);
 312         return 0;
 313 }
 314
 315 static int ip6_forward_proxy_check(struct sk_buff *skb)
 316 {
 317         struct ipv6hdr *hdr = skb->nh.ipv6h;
 318         u8 nexthdr = hdr->nexthdr;
 319         int offset;
 320
 321         if (ipv6_ext_hdr(nexthdr)) {
 322                 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr);
 323                 if (offset < 0)
 324                         return 0;
 325         } else
 326                 offset = sizeof(struct ipv6hdr);
 327
 328         if (nexthdr == IPPROTO_ICMPV6) {
 329                 struct icmp6hdr *icmp6;
 330
 331                 if (!pskb_may_pull(skb, skb->nh.raw + offset + 1 - skb->data))
 332                         return 0;
 333
 334                 icmp6 = (struct icmp6hdr *)(skb->nh.raw + offset);
 335
 336                 switch (icmp6->icmp6_type) {
 337                 case NDISC_ROUTER_SOLICITATION:
 338                 case NDISC_ROUTER_ADVERTISEMENT:
 339                 case NDISC_NEIGHBOUR_SOLICITATION:
 340                 case NDISC_NEIGHBOUR_ADVERTISEMENT:
 341                 case NDISC_REDIRECT:
 342                         /* For reaction involving unicast neighbor discovery
 343                          * message destined to the proxied address, pass it to
 344                          * input function.
 345                          */
 346                         return 1;
 347                 default:
 348                         break;
 349                 }
 350         }
 351
 352         /*
 353          * The proxying router can't forward traffic sent to a link-local
 354          * address, so signal the sender and discard the packet. This
 355          * behavior is clarified by the MIPv6 specification.
 356          */
 357         if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
 358                 dst_link_failure(skb);
 359                 return -1;
 360         }
 361
 362         return 0;
 363 }
 364
 365 static inline int ip6_forward_finish(struct sk_buff *skb)
 366 {
 367         return dst_output(skb);
 368 }
 369
 370 int ip6_forward(struct sk_buff *skb)
 371 {
 372         struct dst_entry *dst = skb->dst;
 373         struct ipv6hdr *hdr = skb->nh.ipv6h;
 374         struct inet6_skb_parm *opt = IP6CB(skb);
 375
 376         if (ipv6_devconf.forwarding == 0)
 377                 goto error;
 378
 379         if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
 380                 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
 381                 goto drop;
 382         }
 383
 384         skb->ip_summed = CHECKSUM_NONE;
 385
 386         /*
 387          *      We DO NOT make any processing on
 388          *      RA packets, pushing them to user level AS IS
 389          *      without ane WARRANTY that application will be able
 390          *      to interpret them. The reason is that we
 391          *      cannot make anything clever here.
 392          *
 393          *      We are not end-node, so that if packet contains
 394          *      AH/ESP, we cannot make anything.
 395          *      Defragmentation also would be mistake, RA packets
 396          *      cannot be fragmented, because there is no warranty
 397          *      that different fragments will go along one path. --ANK
 398          */
 399         if (opt->ra) {
 400                 u8 *ptr = skb->nh.raw + opt->ra;
 401                 if (ip6_call_ra_chain(skb, (ptr[2]<<8) + ptr[3]))
 402                         return 0;
 403         }
 404
 405         /*
 406          *      check and decrement ttl
 407          */
 408         if (hdr->hop_limit <= 1) {
 409                 /* Force OUTPUT device used as source address */
 410                 skb->dev = dst->dev;
 411                 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT,
 412                             0, skb->dev);
 413                 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INHDRERRORS);
 414
 415                 kfree_skb(skb);
 416                 return -ETIMEDOUT;
 417         }
 418
 419         /* XXX: idev->cnf.proxy_ndp? */
 420         if (ipv6_devconf.proxy_ndp &&
 421             pneigh_lookup(&nd_tbl, &hdr->daddr, skb->dev, 0)) {
 422                 int proxied = ip6_forward_proxy_check(skb);
 423                 if (proxied > 0)
 424                         return ip6_input(skb);
 425                 else if (proxied < 0) {
 426                         IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
 427                         goto drop;
 428                 }
 429         }
 430
 431         if (!xfrm6_route_forward(skb)) {
 432                 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
 433                 goto drop;
 434         }
 435         dst = skb->dst;
 436
 437         /* IPv6 specs say nothing about it, but it is clear that we cannot
 438            send redirects to source routed frames.
 439          */
 440         if (skb->dev == dst->dev && dst->neighbour && opt->srcrt == 0) {
 441                 struct in6_addr *target = NULL;
 442                 struct rt6_info *rt;
 443                 struct neighbour *n = dst->neighbour;
 444
 445                 /*
 446                  *      incoming and outgoing devices are the same
 447                  *      send a redirect.
 448                  */
 449
 450                 rt = (struct rt6_info *) dst;
 451                 if ((rt->rt6i_flags & RTF_GATEWAY))
 452                         target = (struct in6_addr*)&n->primary_key;
 453                 else
 454                         target = &hdr->daddr;
 455
 456                 /* Limit redirects both by destination (here)
 457                    and by source (inside ndisc_send_redirect)
 458                  */
 459                 if (xrlim_allow(dst, 1*HZ))
 460                         ndisc_send_redirect(skb, n, target);
 461         } else if (ipv6_addr_type(&hdr->saddr)&(IPV6_ADDR_MULTICAST|IPV6_ADDR_LOOPBACK
 462                                                 |IPV6_ADDR_LINKLOCAL)) {
 463                 /* This check is security critical. */
 464                 goto error;
 465         }
 466
 467         if (skb->len > dst_mtu(dst)) {
 468                 /* Again, force OUTPUT device used as source address */
 469                 skb->dev = dst->dev;
 470                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, dst_mtu(dst), skb->dev);
 471                 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INTOOBIGERRORS);
 472                 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_FRAGFAILS);
 473                 kfree_skb(skb);
 474                 return -EMSGSIZE;
 475         }
 476
 477         if (skb_cow(skb, dst->dev->hard_header_len)) {
 478                 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_OUTDISCARDS);
 479                 goto drop;
 480         }
 481
 482         hdr = skb->nh.ipv6h;
 483
 484         /* Mangling hops number delayed to point after skb COW */
 485
 486         hdr->hop_limit--;
 487
 488         IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
 489         return NF_HOOK(PF_INET6,NF_IP6_FORWARD, skb, skb->dev, dst->dev, ip6_forward_finish);
 490
 491 error:
 492         IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
 493 drop:
 494         kfree_skb(skb);
 495         return -EINVAL;
 496 }
 497
 498 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
 499 {
 500         to->pkt_type = from->pkt_type;
 501         to->priority = from->priority;
 502         to->protocol = from->protocol;
 503         dst_release(to->dst);
 504         to->dst = dst_clone(from->dst);
 505         to->dev = from->dev;
 506         to->mark = from->mark;
 507
 508 #ifdef CONFIG_NET_SCHED
 509         to->tc_index = from->tc_index;
 510 #endif
 511 #ifdef CONFIG_NETFILTER
 512         /* Connection association is same as pre-frag packet */
 513         nf_conntrack_put(to->nfct);
 514         to->nfct = from->nfct;
 515         nf_conntrack_get(to->nfct);
 516         to->nfctinfo = from->nfctinfo;
 517 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
 518         nf_conntrack_put_reasm(to->nfct_reasm);
 519         to->nfct_reasm = from->nfct_reasm;
 520         nf_conntrack_get_reasm(to->nfct_reasm);
 521 #endif
 522 #ifdef CONFIG_BRIDGE_NETFILTER
 523         nf_bridge_put(to->nf_bridge);
 524         to->nf_bridge = from->nf_bridge;
 525         nf_bridge_get(to->nf_bridge);
 526 #endif
 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 = (struct ipv6_opt_hdr*)(skb->nh.ipv6h + 1);
 535         unsigned int packet_len = skb->tail - skb->nh.raw;
 536         int found_rhdr = 0;
 537         *nexthdr = &skb->nh.ipv6h->nexthdr;
 538
 539         while (offset + 1 <= packet_len) {
 540
 541                 switch (**nexthdr) {
 542
 543                 case NEXTHDR_HOP:
 544                         break;
 545                 case NEXTHDR_ROUTING:
 546                         found_rhdr = 1;
 547                         break;
 548                 case NEXTHDR_DEST:
 549 #ifdef CONFIG_IPV6_MIP6
 550                         if (ipv6_find_tlv(skb, offset, IPV6_TLV_HAO) >= 0)
 551                                 break;
 552 #endif
 553                         if (found_rhdr)
 554                                 return offset;
 555                         break;
 556                 default :
 557                         return offset;
 558                 }
 559
 560                 offset += ipv6_optlen(exthdr);
 561                 *nexthdr = &exthdr->nexthdr;
 562                 exthdr = (struct ipv6_opt_hdr*)(skb->nh.raw + offset);
 563         }
 564
 565         return offset;
 566 }
 567 EXPORT_SYMBOL_GPL(ip6_find_1stfragopt);
 568
 569 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
 570 {
 571         struct net_device *dev;
 572         struct sk_buff *frag;
 573         struct rt6_info *rt = (struct rt6_info*)skb->dst;
 574         struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
 575         struct ipv6hdr *tmp_hdr;
 576         struct frag_hdr *fh;
 577         unsigned int mtu, hlen, left, len;
 578         __be32 frag_id = 0;
 579         int ptr, offset = 0, err=0;
 580         u8 *prevhdr, nexthdr = 0;
 581
 582         dev = rt->u.dst.dev;
 583         hlen = ip6_find_1stfragopt(skb, &prevhdr);
 584         nexthdr = *prevhdr;
 585
 586         mtu = dst_mtu(&rt->u.dst);
 587         if (np && np->frag_size < mtu) {
 588                 if (np->frag_size)
 589                         mtu = np->frag_size;
 590         }
 591         mtu -= hlen + sizeof(struct frag_hdr);
 592
 593         if (skb_shinfo(skb)->frag_list) {
 594                 int first_len = skb_pagelen(skb);
 595
 596                 if (first_len - hlen > mtu ||
 597                     ((first_len - hlen) & 7) ||
 598                     skb_cloned(skb))
 599                         goto slow_path;
 600
 601                 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
 602                         /* Correct geometry. */
 603                         if (frag->len > mtu ||
 604                             ((frag->len & 7) && frag->next) ||
 605                             skb_headroom(frag) < hlen)
 606                             goto slow_path;
 607
 608                         /* Partially cloned skb? */
 609                         if (skb_shared(frag))
 610                                 goto slow_path;
 611
 612                         BUG_ON(frag->sk);
 613                         if (skb->sk) {
 614                                 sock_hold(skb->sk);
 615                                 frag->sk = skb->sk;
 616                                 frag->destructor = sock_wfree;
 617                                 skb->truesize -= frag->truesize;
 618                         }
 619                 }
 620
 621                 err = 0;
 622                 offset = 0;
 623                 frag = skb_shinfo(skb)->frag_list;
 624                 skb_shinfo(skb)->frag_list = NULL;
 625                 /* BUILD HEADER */
 626
 627                 tmp_hdr = kmalloc(hlen, GFP_ATOMIC);
 628                 if (!tmp_hdr) {
 629                         IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGFAILS);
 630                         return -ENOMEM;
 631                 }
 632
 633                 *prevhdr = NEXTHDR_FRAGMENT;
 634                 memcpy(tmp_hdr, skb->nh.raw, hlen);
 635                 __skb_pull(skb, hlen);
 636                 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
 637                 skb->nh.raw = __skb_push(skb, hlen);
 638                 memcpy(skb->nh.raw, tmp_hdr, hlen);
 639
 640                 ipv6_select_ident(skb, fh);
 641                 fh->nexthdr = nexthdr;
 642                 fh->reserved = 0;
 643                 fh->frag_off = htons(IP6_MF);
 644                 frag_id = fh->identification;
 645
 646                 first_len = skb_pagelen(skb);
 647                 skb->data_len = first_len - skb_headlen(skb);
 648                 skb->len = first_len;
 649                 skb->nh.ipv6h->payload_len = htons(first_len - sizeof(struct ipv6hdr));
 650
 651                 dst_hold(&rt->u.dst);
 652
 653                 for (;;) {
 654                         /* Prepare header of the next frame,
 655                          * before previous one went down. */
 656                         if (frag) {
 657                                 frag->ip_summed = CHECKSUM_NONE;
 658                                 frag->h.raw = frag->data;
 659                                 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
 660                                 frag->nh.raw = __skb_push(frag, hlen);
 661                                 memcpy(frag->nh.raw, tmp_hdr, hlen);
 662                                 offset += skb->len - hlen - sizeof(struct frag_hdr);
 663                                 fh->nexthdr = nexthdr;
 664                                 fh->reserved = 0;
 665                                 fh->frag_off = htons(offset);
 666                                 if (frag->next != NULL)
 667                                         fh->frag_off |= htons(IP6_MF);
 668                                 fh->identification = frag_id;
 669                                 frag->nh.ipv6h->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
 670                                 ip6_copy_metadata(frag, skb);
 671                         }
 672
 673                         err = output(skb);
 674                         if(!err)
 675                                 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGCREATES);
 676
 677                         if (err || !frag)
 678                                 break;
 679
 680                         skb = frag;
 681                         frag = skb->next;
 682                         skb->next = NULL;
 683                 }
 684
 685                 kfree(tmp_hdr);
 686
 687                 if (err == 0) {
 688                         IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGOKS);
 689                         dst_release(&rt->u.dst);
 690                         return 0;
 691                 }
 692
 693                 while (frag) {
 694                         skb = frag->next;
 695                         kfree_skb(frag);
 696                         frag = skb;
 697                 }
 698
 699                 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGFAILS);
 700                 dst_release(&rt->u.dst);
 701                 return err;
 702         }
 703
 704 slow_path:
 705         left = skb->len - hlen;         /* Space per frame */
 706         ptr = hlen;                     /* Where to start from */
 707
 708         /*
 709          *      Fragment the datagram.
 710          */
 711
 712         *prevhdr = NEXTHDR_FRAGMENT;
 713
 714         /*
 715          *      Keep copying data until we run out.
 716          */
 717         while(left > 0) {
 718                 len = left;
 719                 /* IF: it doesn't fit, use 'mtu' - the data space left */
 720                 if (len > mtu)
 721                         len = mtu;
 722                 /* IF: we are not sending upto and including the packet end
 723                    then align the next start on an eight byte boundary */
 724                 if (len < left) {
 725                         len &= ~7;
 726                 }
 727                 /*
 728                  *      Allocate buffer.
 729                  */
 730
 731                 if ((frag = alloc_skb(len+hlen+sizeof(struct frag_hdr)+LL_RESERVED_SPACE(rt->u.dst.dev), GFP_ATOMIC)) == NULL) {
 732                         NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
 733                         IP6_INC_STATS(ip6_dst_idev(skb->dst),
 734                                       IPSTATS_MIB_FRAGFAILS);
 735                         err = -ENOMEM;
 736                         goto fail;
 737                 }
 738
 739                 /*
 740                  *      Set up data on packet
 741                  */
 742
 743                 ip6_copy_metadata(frag, skb);
 744                 skb_reserve(frag, LL_RESERVED_SPACE(rt->u.dst.dev));
 745                 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
 746                 frag->nh.raw = frag->data;
 747                 fh = (struct frag_hdr*)(frag->data + hlen);
 748                 frag->h.raw = frag->data + hlen + sizeof(struct frag_hdr);
 749
 750                 /*
 751                  *      Charge the memory for the fragment to any owner
 752                  *      it might possess
 753                  */
 754                 if (skb->sk)
 755                         skb_set_owner_w(frag, skb->sk);
 756
 757                 /*
 758                  *      Copy the packet header into the new buffer.
 759                  */
 760                 memcpy(frag->nh.raw, skb->data, hlen);
 761
 762                 /*
 763                  *      Build fragment header.
 764                  */
 765                 fh->nexthdr = nexthdr;
 766                 fh->reserved = 0;
 767                 if (!frag_id) {
 768                         ipv6_select_ident(skb, fh);
 769                         frag_id = fh->identification;
 770                 } else
 771                         fh->identification = frag_id;
 772
 773                 /*
 774                  *      Copy a block of the IP datagram.
 775                  */
 776                 if (skb_copy_bits(skb, ptr, frag->h.raw, len))
 777                         BUG();
 778                 left -= len;
 779
 780                 fh->frag_off = htons(offset);
 781                 if (left > 0)
 782                         fh->frag_off |= htons(IP6_MF);
 783                 frag->nh.ipv6h->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
 784
 785                 ptr += len;
 786                 offset += len;
 787
 788                 /*
 789                  *      Put this fragment into the sending queue.
 790                  */
 791                 err = output(frag);
 792                 if (err)
 793                         goto fail;
 794
 795                 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGCREATES);
 796         }
 797         IP6_INC_STATS(ip6_dst_idev(skb->dst),
 798                       IPSTATS_MIB_FRAGOKS);
 799         kfree_skb(skb);
 800         return err;
 801
 802 fail:
 803         IP6_INC_STATS(ip6_dst_idev(skb->dst),
 804                       IPSTATS_MIB_FRAGFAILS);
 805         kfree_skb(skb);
 806         return err;
 807 }
 808
 809 static inline int ip6_rt_check(struct rt6key *rt_key,
 810                                struct in6_addr *fl_addr,
 811                                struct in6_addr *addr_cache)
 812 {
 813         return ((rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
 814                 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache)));
 815 }
 816
 817 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
 818                                           struct dst_entry *dst,
 819                                           struct flowi *fl)
 820 {
 821         struct ipv6_pinfo *np = inet6_sk(sk);
 822         struct rt6_info *rt = (struct rt6_info *)dst;
 823
 824         if (!dst)
 825                 goto out;
 826
 827         /* Yes, checking route validity in not connected
 828          * case is not very simple. Take into account,
 829          * that we do not support routing by source, TOS,
 830          * and MSG_DONTROUTE            --ANK (980726)
 831          *
 832          * 1. ip6_rt_check(): If route was host route,
 833          *    check that cached destination is current.
 834          *    If it is network route, we still may
 835          *    check its validity using saved pointer
 836          *    to the last used address: daddr_cache.
 837          *    We do not want to save whole address now,
 838          *    (because main consumer of this service
 839          *    is tcp, which has not this problem),
 840          *    so that the last trick works only on connected
 841          *    sockets.
 842          * 2. oif also should be the same.
 843          */
 844         if (ip6_rt_check(&rt->rt6i_dst, &fl->fl6_dst, np->daddr_cache) ||
 845 #ifdef CONFIG_IPV6_SUBTREES
 846             ip6_rt_check(&rt->rt6i_src, &fl->fl6_src, np->saddr_cache) ||
 847 #endif
 848             (fl->oif && fl->oif != dst->dev->ifindex)) {
 849                 dst_release(dst);
 850                 dst = NULL;
 851         }
 852
 853 out:
 854         return dst;
 855 }
 856
 857 static int ip6_dst_lookup_tail(struct sock *sk,
 858                                struct dst_entry **dst, struct flowi *fl)
 859 {
 860         int err;
 861
 862         if (*dst == NULL)
 863                 *dst = ip6_route_output(sk, fl);
 864
 865         if ((err = (*dst)->error))
 866                 goto out_err_release;
 867
 868         if (ipv6_addr_any(&fl->fl6_src)) {
 869                 err = ipv6_get_saddr(*dst, &fl->fl6_dst, &fl->fl6_src);
 870                 if (err)
 871                         goto out_err_release;
 872         }
 873
 874         return 0;
 875
 876 out_err_release:
 877         dst_release(*dst);
 878         *dst = NULL;
 879         return err;
 880 }
 881
 882 /**
 883  *      ip6_dst_lookup - perform route lookup on flow
 884  *      @sk: socket which provides route info
 885  *      @dst: pointer to dst_entry * for result
 886  *      @fl: flow to lookup
 887  *
 888  *      This function performs a route lookup on the given flow.
 889  *
 890  *      It returns zero on success, or a standard errno code on error.
 891  */
 892 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
 893 {
 894         *dst = NULL;
 895         return ip6_dst_lookup_tail(sk, dst, fl);
 896 }
 897 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
 898
 899 /**
 900  *      ip6_sk_dst_lookup - perform socket cached route lookup on flow
 901  *      @sk: socket which provides the dst cache and route info
 902  *      @dst: pointer to dst_entry * for result
 903  *      @fl: flow to lookup
 904  *
 905  *      This function performs a route lookup on the given flow with the
 906  *      possibility of using the cached route in the socket if it is valid.
 907  *      It will take the socket dst lock when operating on the dst cache.
 908  *      As a result, this function can only be used in process context.
 909  *
 910  *      It returns zero on success, or a standard errno code on error.
 911  */
 912 int ip6_sk_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
 913 {
 914         *dst = NULL;
 915         if (sk) {
 916                 *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
 917                 *dst = ip6_sk_dst_check(sk, *dst, fl);
 918         }
 919
 920         return ip6_dst_lookup_tail(sk, dst, fl);
 921 }
 922 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup);
 923
 924 static inline int ip6_ufo_append_data(struct sock *sk,
 925                         int getfrag(void *from, char *to, int offset, int len,
 926                         int odd, struct sk_buff *skb),
 927                         void *from, int length, int hh_len, int fragheaderlen,
 928                         int transhdrlen, int mtu,unsigned int flags)
 929
 930 {
 931         struct sk_buff *skb;
 932         int err;
 933
 934         /* There is support for UDP large send offload by network
 935          * device, so create one single skb packet containing complete
 936          * udp datagram
 937          */
 938         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
 939                 skb = sock_alloc_send_skb(sk,
 940                         hh_len + fragheaderlen + transhdrlen + 20,
 941                         (flags & MSG_DONTWAIT), &err);
 942                 if (skb == NULL)
 943                         return -ENOMEM;
 944
 945                 /* reserve space for Hardware header */
 946                 skb_reserve(skb, hh_len);
 947
 948                 /* create space for UDP/IP header */
 949                 skb_put(skb,fragheaderlen + transhdrlen);
 950
 951                 /* initialize network header pointer */
 952                 skb->nh.raw = skb->data;
 953
 954                 /* initialize protocol header pointer */
 955                 skb->h.raw = skb->data + fragheaderlen;
 956
 957                 skb->ip_summed = CHECKSUM_PARTIAL;
 958                 skb->csum = 0;
 959                 sk->sk_sndmsg_off = 0;
 960         }
 961
 962         err = skb_append_datato_frags(sk,skb, getfrag, from,
 963                                       (length - transhdrlen));
 964         if (!err) {
 965                 struct frag_hdr fhdr;
 966
 967                 /* specify the length of each IP datagram fragment*/
 968                 skb_shinfo(skb)->gso_size = mtu - fragheaderlen -
 969                                             sizeof(struct frag_hdr);
 970                 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
 971                 ipv6_select_ident(skb, &fhdr);
 972                 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
 973                 __skb_queue_tail(&sk->sk_write_queue, skb);
 974
 975                 return 0;
 976         }
 977         /* There is not enough support do UPD LSO,
 978          * so follow normal path
 979          */
 980         kfree_skb(skb);
 981
 982         return err;
 983 }
 984
 985 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
 986         int offset, int len, int odd, struct sk_buff *skb),
 987         void *from, int length, int transhdrlen,
 988         int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi *fl,
 989         struct rt6_info *rt, unsigned int flags)
 990 {
 991         struct inet_sock *inet = inet_sk(sk);
 992         struct ipv6_pinfo *np = inet6_sk(sk);
 993         struct sk_buff *skb;
 994         unsigned int maxfraglen, fragheaderlen;
 995         int exthdrlen;
 996         int hh_len;
 997         int mtu;
 998         int copy;
 999         int err;
1000         int offset = 0;
1001         int csummode = CHECKSUM_NONE;
1002
1003         if (flags&MSG_PROBE)
1004                 return 0;
1005         if (skb_queue_empty(&sk->sk_write_queue)) {
1006                 /*
1007                  * setup for corking
1008                  */
1009                 if (opt) {
1010                         if (np->cork.opt == NULL) {
1011                                 np->cork.opt = kmalloc(opt->tot_len,
1012                                                        sk->sk_allocation);
1013                                 if (unlikely(np->cork.opt == NULL))
1014                                         return -ENOBUFS;
1015                         } else if (np->cork.opt->tot_len < opt->tot_len) {
1016                                 printk(KERN_DEBUG "ip6_append_data: invalid option length\n");
1017                                 return -EINVAL;
1018                         }
1019                         memcpy(np->cork.opt, opt, opt->tot_len);
1020                         inet->cork.flags |= IPCORK_OPT;
1021                         /* need source address above miyazawa*/
1022                 }
1023                 dst_hold(&rt->u.dst);
1024                 np->cork.rt = rt;
1025                 inet->cork.fl = *fl;
1026                 np->cork.hop_limit = hlimit;
1027                 np->cork.tclass = tclass;
1028                 mtu = dst_mtu(rt->u.dst.path);
1029                 if (np->frag_size < mtu) {
1030                         if (np->frag_size)
1031                                 mtu = np->frag_size;
1032                 }
1033                 inet->cork.fragsize = mtu;
1034                 if (dst_allfrag(rt->u.dst.path))
1035                         inet->cork.flags |= IPCORK_ALLFRAG;
1036                 inet->cork.length = 0;
1037                 sk->sk_sndmsg_page = NULL;
1038                 sk->sk_sndmsg_off = 0;
1039                 exthdrlen = rt->u.dst.header_len + (opt ? opt->opt_flen : 0);
1040                 length += exthdrlen;
1041                 transhdrlen += exthdrlen;
1042         } else {
1043                 rt = np->cork.rt;
1044                 fl = &inet->cork.fl;
1045                 if (inet->cork.flags & IPCORK_OPT)
1046                         opt = np->cork.opt;
1047                 transhdrlen = 0;
1048                 exthdrlen = 0;
1049                 mtu = inet->cork.fragsize;
1050         }
1051
1052         hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1053
1054         fragheaderlen = sizeof(struct ipv6hdr) + rt->u.dst.nfheader_len + (opt ? opt->opt_nflen : 0);
1055         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
1056
1057         if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1058                 if (inet->cork.length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
1059                         ipv6_local_error(sk, EMSGSIZE, fl, mtu-exthdrlen);
1060                         return -EMSGSIZE;
1061                 }
1062         }
1063
1064         /*
1065          * Let's try using as much space as possible.
1066          * Use MTU if total length of the message fits into the MTU.
1067          * Otherwise, we need to reserve fragment header and
1068          * fragment alignment (= 8-15 octects, in total).
1069          *
1070          * Note that we may need to "move" the data from the tail of
1071          * of the buffer to the new fragment when we split
1072          * the message.
1073          *
1074          * FIXME: It may be fragmented into multiple chunks
1075          *        at once if non-fragmentable extension headers
1076          *        are too large.
1077          * --yoshfuji
1078          */
1079
1080         inet->cork.length += length;
1081         if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
1082             (rt->u.dst.dev->features & NETIF_F_UFO)) {
1083
1084                 err = ip6_ufo_append_data(sk, getfrag, from, length, hh_len,
1085                                           fragheaderlen, transhdrlen, mtu,
1086                                           flags);
1087                 if (err)
1088                         goto error;
1089                 return 0;
1090         }
1091
1092         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1093                 goto alloc_new_skb;
1094
1095         while (length > 0) {
1096                 /* Check if the remaining data fits into current packet. */
1097                 copy = (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1098                 if (copy < length)
1099                         copy = maxfraglen - skb->len;
1100
1101                 if (copy <= 0) {
1102                         char *data;
1103                         unsigned int datalen;
1104                         unsigned int fraglen;
1105                         unsigned int fraggap;
1106                         unsigned int alloclen;
1107                         struct sk_buff *skb_prev;
1108 alloc_new_skb:
1109                         skb_prev = skb;
1110
1111                         /* There's no room in the current skb */
1112                         if (skb_prev)
1113                                 fraggap = skb_prev->len - maxfraglen;
1114                         else
1115                                 fraggap = 0;
1116
1117                         /*
1118                          * If remaining data exceeds the mtu,
1119                          * we know we need more fragment(s).
1120                          */
1121                         datalen = length + fraggap;
1122                         if (datalen > (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1123                                 datalen = maxfraglen - fragheaderlen;
1124
1125                         fraglen = datalen + fragheaderlen;
1126                         if ((flags & MSG_MORE) &&
1127                             !(rt->u.dst.dev->features&NETIF_F_SG))
1128                                 alloclen = mtu;
1129                         else
1130                                 alloclen = datalen + fragheaderlen;
1131
1132                         /*
1133                          * The last fragment gets additional space at tail.
1134                          * Note: we overallocate on fragments with MSG_MODE
1135                          * because we have no idea if we're the last one.
1136                          */
1137                         if (datalen == length + fraggap)
1138                                 alloclen += rt->u.dst.trailer_len;
1139
1140                         /*
1141                          * We just reserve space for fragment header.
1142                          * Note: this may be overallocation if the message
1143                          * (without MSG_MORE) fits into the MTU.
1144                          */
1145                         alloclen += sizeof(struct frag_hdr);
1146
1147                         if (transhdrlen) {
1148                                 skb = sock_alloc_send_skb(sk,
1149                                                 alloclen + hh_len,
1150                                                 (flags & MSG_DONTWAIT), &err);
1151                         } else {
1152                                 skb = NULL;
1153                                 if (atomic_read(&sk->sk_wmem_alloc) <=
1154                                     2 * sk->sk_sndbuf)
1155                                         skb = sock_wmalloc(sk,
1156                                                            alloclen + hh_len, 1,
1157                                                            sk->sk_allocation);
1158                                 if (unlikely(skb == NULL))
1159                                         err = -ENOBUFS;
1160                         }
1161                         if (skb == NULL)
1162                                 goto error;
1163                         /*
1164                          *      Fill in the control structures
1165                          */
1166                         skb->ip_summed = csummode;
1167                         skb->csum = 0;
1168                         /* reserve for fragmentation */
1169                         skb_reserve(skb, hh_len+sizeof(struct frag_hdr));
1170
1171                         /*
1172                          *      Find where to start putting bytes
1173                          */
1174                         data = skb_put(skb, fraglen);
1175                         skb->nh.raw = data + exthdrlen;
1176                         data += fragheaderlen;
1177                         skb->h.raw = data + exthdrlen;
1178
1179                         if (fraggap) {
1180                                 skb->csum = skb_copy_and_csum_bits(
1181                                         skb_prev, maxfraglen,
1182                                         data + transhdrlen, fraggap, 0);
1183                                 skb_prev->csum = csum_sub(skb_prev->csum,
1184                                                           skb->csum);
1185                                 data += fraggap;
1186                                 pskb_trim_unique(skb_prev, maxfraglen);
1187                         }
1188                         copy = datalen - transhdrlen - fraggap;
1189                         if (copy < 0) {
1190                                 err = -EINVAL;
1191                                 kfree_skb(skb);
1192                                 goto error;
1193                         } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1194                                 err = -EFAULT;
1195                                 kfree_skb(skb);
1196                                 goto error;
1197                         }
1198
1199                         offset += copy;
1200                         length -= datalen - fraggap;
1201                         transhdrlen = 0;
1202                         exthdrlen = 0;
1203                         csummode = CHECKSUM_NONE;
1204
1205                         /*
1206                          * Put the packet on the pending queue
1207                          */
1208                         __skb_queue_tail(&sk->sk_write_queue, skb);
1209                         continue;
1210                 }
1211
1212                 if (copy > length)
1213                         copy = length;
1214
1215                 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
1216                         unsigned int off;
1217
1218                         off = skb->len;
1219                         if (getfrag(from, skb_put(skb, copy),
1220                                                 offset, copy, off, skb) < 0) {
1221                                 __skb_trim(skb, off);
1222                                 err = -EFAULT;
1223                                 goto error;
1224                         }
1225                 } else {
1226                         int i = skb_shinfo(skb)->nr_frags;
1227                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1228                         struct page *page = sk->sk_sndmsg_page;
1229                         int off = sk->sk_sndmsg_off;
1230                         unsigned int left;
1231
1232                         if (page && (left = PAGE_SIZE - off) > 0) {
1233                                 if (copy >= left)
1234                                         copy = left;
1235                                 if (page != frag->page) {
1236                                         if (i == MAX_SKB_FRAGS) {
1237                                                 err = -EMSGSIZE;
1238                                                 goto error;
1239                                         }
1240                                         get_page(page);
1241                                         skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1242                                         frag = &skb_shinfo(skb)->frags[i];
1243                                 }
1244                         } else if(i < MAX_SKB_FRAGS) {
1245                                 if (copy > PAGE_SIZE)
1246                                         copy = PAGE_SIZE;
1247                                 page = alloc_pages(sk->sk_allocation, 0);
1248                                 if (page == NULL) {
1249                                         err = -ENOMEM;
1250                                         goto error;
1251                                 }
1252                                 sk->sk_sndmsg_page = page;
1253                                 sk->sk_sndmsg_off = 0;
1254
1255                                 skb_fill_page_desc(skb, i, page, 0, 0);
1256                                 frag = &skb_shinfo(skb)->frags[i];
1257                                 skb->truesize += PAGE_SIZE;
1258                                 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1259                         } else {
1260                                 err = -EMSGSIZE;
1261                                 goto error;
1262                         }
1263                         if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1264                                 err = -EFAULT;
1265                                 goto error;
1266                         }
1267                         sk->sk_sndmsg_off += copy;
1268                         frag->size += copy;
1269                         skb->len += copy;
1270                         skb->data_len += copy;
1271                 }
1272                 offset += copy;
1273                 length -= copy;
1274         }
1275         return 0;
1276 error:
1277         inet->cork.length -= length;
1278         IP6_INC_STATS(rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1279         return err;
1280 }
1281
1282 int ip6_push_pending_frames(struct sock *sk)
1283 {
1284         struct sk_buff *skb, *tmp_skb;
1285         struct sk_buff **tail_skb;
1286         struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1287         struct inet_sock *inet = inet_sk(sk);
1288         struct ipv6_pinfo *np = inet6_sk(sk);
1289         struct ipv6hdr *hdr;
1290         struct ipv6_txoptions *opt = np->cork.opt;
1291         struct rt6_info *rt = np->cork.rt;
1292         struct flowi *fl = &inet->cork.fl;
1293         unsigned char proto = fl->proto;
1294         int err = 0;
1295
1296         if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1297                 goto out;
1298         tail_skb = &(skb_shinfo(skb)->frag_list);
1299
1300         /* move skb->data to ip header from ext header */
1301         if (skb->data < skb->nh.raw)
1302                 __skb_pull(skb, skb->nh.raw - skb->data);
1303         while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1304                 __skb_pull(tmp_skb, skb->h.raw - skb->nh.raw);
1305                 *tail_skb = tmp_skb;
1306                 tail_skb = &(tmp_skb->next);
1307                 skb->len += tmp_skb->len;
1308                 skb->data_len += tmp_skb->len;
1309                 skb->truesize += tmp_skb->truesize;
1310                 __sock_put(tmp_skb->sk);
1311                 tmp_skb->destructor = NULL;
1312                 tmp_skb->sk = NULL;
1313         }
1314
1315         ipv6_addr_copy(final_dst, &fl->fl6_dst);
1316         __skb_pull(skb, skb->h.raw - skb->nh.raw);
1317         if (opt && opt->opt_flen)
1318                 ipv6_push_frag_opts(skb, opt, &proto);
1319         if (opt && opt->opt_nflen)
1320                 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1321
1322         skb->nh.ipv6h = hdr = (struct ipv6hdr*) skb_push(skb, sizeof(struct ipv6hdr));
1323
1324         *(__be32*)hdr = fl->fl6_flowlabel |
1325                      htonl(0x60000000 | ((int)np->cork.tclass << 20));
1326
1327         if (skb->len <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN)
1328                 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
1329         else
1330                 hdr->payload_len = 0;
1331         hdr->hop_limit = np->cork.hop_limit;
1332         hdr->nexthdr = proto;
1333         ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
1334         ipv6_addr_copy(&hdr->daddr, final_dst);
1335
1336         skb->priority = sk->sk_priority;
1337
1338         skb->dst = dst_clone(&rt->u.dst);
1339         IP6_INC_STATS(rt->rt6i_idev, IPSTATS_MIB_OUTREQUESTS);
1340         err = NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, skb->dst->dev, dst_output);
1341         if (err) {
1342                 if (err > 0)
1343                         err = np->recverr ? net_xmit_errno(err) : 0;
1344                 if (err)
1345                         goto error;
1346         }
1347
1348 out:
1349         inet->cork.flags &= ~IPCORK_OPT;
1350         kfree(np->cork.opt);
1351         np->cork.opt = NULL;
1352         if (np->cork.rt) {
1353                 dst_release(&np->cork.rt->u.dst);
1354                 np->cork.rt = NULL;
1355                 inet->cork.flags &= ~IPCORK_ALLFRAG;
1356         }
1357         memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1358         return err;
1359 error:
1360         goto out;
1361 }
1362
1363 void ip6_flush_pending_frames(struct sock *sk)
1364 {
1365         struct inet_sock *inet = inet_sk(sk);
1366         struct ipv6_pinfo *np = inet6_sk(sk);
1367         struct sk_buff *skb;
1368
1369         while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1370                 IP6_INC_STATS(ip6_dst_idev(skb->dst),
1371                               IPSTATS_MIB_OUTDISCARDS);
1372                 kfree_skb(skb);
1373         }
1374
1375         inet->cork.flags &= ~IPCORK_OPT;
1376
1377         kfree(np->cork.opt);
1378         np->cork.opt = NULL;
1379         if (np->cork.rt) {
1380                 dst_release(&np->cork.rt->u.dst);
1381                 np->cork.rt = NULL;
1382                 inet->cork.flags &= ~IPCORK_ALLFRAG;
1383         }
1384         memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1385 }