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         struct dst_entry *dst = skb->dst;
  76
  77         if (dst->hh)
  78                 return neigh_hh_output(dst->hh, skb);
  79         else if (dst->neighbour)
  80                 return dst->neighbour->output(skb);
  81
  82         IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
  83         kfree_skb(skb);
  84         return -EINVAL;
  85
  86 }
  87
  88 /* dev_loopback_xmit for use with netfilter. */
  89 static int ip6_dev_loopback_xmit(struct sk_buff *newskb)
  90 {
  91         skb_reset_mac_header(newskb);
  92         __skb_pull(newskb, skb_network_offset(newskb));
  93         newskb->pkt_type = PACKET_LOOPBACK;
  94         newskb->ip_summed = CHECKSUM_UNNECESSARY;
  95         BUG_TRAP(newskb->dst);
  96
  97         netif_rx(newskb);
  98         return 0;
  99 }
 100
 101
 102 static int ip6_output2(struct sk_buff *skb)
 103 {
 104         struct dst_entry *dst = skb->dst;
 105         struct net_device *dev = dst->dev;
 106
 107         skb->protocol = htons(ETH_P_IPV6);
 108         skb->dev = dev;
 109
 110         if (ipv6_addr_is_multicast(&skb->nh.ipv6h->daddr)) {
 111                 struct ipv6_pinfo* np = skb->sk ? inet6_sk(skb->sk) : NULL;
 112                 struct inet6_dev *idev = ip6_dst_idev(skb->dst);
 113
 114                 if (!(dev->flags & IFF_LOOPBACK) && (!np || np->mc_loop) &&
 115                     ipv6_chk_mcast_addr(dev, &skb->nh.ipv6h->daddr,
 116                                 &skb->nh.ipv6h->saddr)) {
 117                         struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
 118
 119                         /* Do not check for IFF_ALLMULTI; multicast routing
 120                            is not supported in any case.
 121                          */
 122                         if (newskb)
 123                                 NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, newskb, NULL,
 124                                         newskb->dev,
 125                                         ip6_dev_loopback_xmit);
 126
 127                         if (skb->nh.ipv6h->hop_limit == 0) {
 128                                 IP6_INC_STATS(idev, IPSTATS_MIB_OUTDISCARDS);
 129                                 kfree_skb(skb);
 130                                 return 0;
 131                         }
 132                 }
 133
 134                 IP6_INC_STATS(idev, IPSTATS_MIB_OUTMCASTPKTS);
 135         }
 136
 137         return NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, skb,NULL, skb->dev,ip6_output_finish);
 138 }
 139
 140 int ip6_output(struct sk_buff *skb)
 141 {
 142         if ((skb->len > dst_mtu(skb->dst) && !skb_is_gso(skb)) ||
 143                                 dst_allfrag(skb->dst))
 144                 return ip6_fragment(skb, ip6_output2);
 145         else
 146                 return ip6_output2(skb);
 147 }
 148
 149 /*
 150  *      xmit an sk_buff (used by TCP)
 151  */
 152
 153 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
 154              struct ipv6_txoptions *opt, int ipfragok)
 155 {
 156         struct ipv6_pinfo *np = inet6_sk(sk);
 157         struct in6_addr *first_hop = &fl->fl6_dst;
 158         struct dst_entry *dst = skb->dst;
 159         struct ipv6hdr *hdr;
 160         u8  proto = fl->proto;
 161         int seg_len = skb->len;
 162         int hlimit, tclass;
 163         u32 mtu;
 164
 165         if (opt) {
 166                 int head_room;
 167
 168                 /* First: exthdrs may take lots of space (~8K for now)
 169                    MAX_HEADER is not enough.
 170                  */
 171                 head_room = opt->opt_nflen + opt->opt_flen;
 172                 seg_len += head_room;
 173                 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
 174
 175                 if (skb_headroom(skb) < head_room) {
 176                         struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
 177                         if (skb2 == NULL) {
 178                                 IP6_INC_STATS(ip6_dst_idev(skb->dst),
 179                                               IPSTATS_MIB_OUTDISCARDS);
 180                                 kfree_skb(skb);
 181                                 return -ENOBUFS;
 182                         }
 183                         kfree_skb(skb);
 184                         skb = skb2;
 185                         if (sk)
 186                                 skb_set_owner_w(skb, sk);
 187                 }
 188                 if (opt->opt_flen)
 189                         ipv6_push_frag_opts(skb, opt, &proto);
 190                 if (opt->opt_nflen)
 191                         ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
 192         }
 193
 194         skb_push(skb, sizeof(struct ipv6hdr));
 195         skb_reset_network_header(skb);
 196         hdr = skb->nh.ipv6h;
 197
 198         /*
 199          *      Fill in the IPv6 header
 200          */
 201
 202         hlimit = -1;
 203         if (np)
 204                 hlimit = np->hop_limit;
 205         if (hlimit < 0)
 206                 hlimit = dst_metric(dst, RTAX_HOPLIMIT);
 207         if (hlimit < 0)
 208                 hlimit = ipv6_get_hoplimit(dst->dev);
 209
 210         tclass = -1;
 211         if (np)
 212                 tclass = np->tclass;
 213         if (tclass < 0)
 214                 tclass = 0;
 215
 216         *(__be32 *)hdr = htonl(0x60000000 | (tclass << 20)) | fl->fl6_flowlabel;
 217
 218         hdr->payload_len = htons(seg_len);
 219         hdr->nexthdr = proto;
 220         hdr->hop_limit = hlimit;
 221
 222         ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
 223         ipv6_addr_copy(&hdr->daddr, first_hop);
 224
 225         skb->priority = sk->sk_priority;
 226
 227         mtu = dst_mtu(dst);
 228         if ((skb->len <= mtu) || ipfragok || skb_is_gso(skb)) {
 229                 IP6_INC_STATS(ip6_dst_idev(skb->dst),
 230                               IPSTATS_MIB_OUTREQUESTS);
 231                 return NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, dst->dev,
 232                                 dst_output);
 233         }
 234
 235         if (net_ratelimit())
 236                 printk(KERN_DEBUG "IPv6: sending pkt_too_big to self\n");
 237         skb->dev = dst->dev;
 238         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev);
 239         IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGFAILS);
 240         kfree_skb(skb);
 241         return -EMSGSIZE;
 242 }
 243
 244 EXPORT_SYMBOL(ip6_xmit);
 245
 246 /*
 247  *      To avoid extra problems ND packets are send through this
 248  *      routine. It's code duplication but I really want to avoid
 249  *      extra checks since ipv6_build_header is used by TCP (which
 250  *      is for us performance critical)
 251  */
 252
 253 int ip6_nd_hdr(struct sock *sk, struct sk_buff *skb, struct net_device *dev,
 254                struct in6_addr *saddr, struct in6_addr *daddr,
 255                int proto, int len)
 256 {
 257         struct ipv6_pinfo *np = inet6_sk(sk);
 258         struct ipv6hdr *hdr;
 259         int totlen;
 260
 261         skb->protocol = htons(ETH_P_IPV6);
 262         skb->dev = dev;
 263
 264         totlen = len + sizeof(struct ipv6hdr);
 265
 266         hdr = (struct ipv6hdr *) skb_put(skb, sizeof(struct ipv6hdr));
 267         skb->nh.ipv6h = hdr;
 268
 269         *(__be32*)hdr = htonl(0x60000000);
 270
 271         hdr->payload_len = htons(len);
 272         hdr->nexthdr = proto;
 273         hdr->hop_limit = np->hop_limit;
 274
 275         ipv6_addr_copy(&hdr->saddr, saddr);
 276         ipv6_addr_copy(&hdr->daddr, daddr);
 277
 278         return 0;
 279 }
 280
 281 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
 282 {
 283         struct ip6_ra_chain *ra;
 284         struct sock *last = NULL;
 285
 286         read_lock(&ip6_ra_lock);
 287         for (ra = ip6_ra_chain; ra; ra = ra->next) {
 288                 struct sock *sk = ra->sk;
 289                 if (sk && ra->sel == sel &&
 290                     (!sk->sk_bound_dev_if ||
 291                      sk->sk_bound_dev_if == skb->dev->ifindex)) {
 292                         if (last) {
 293                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
 294                                 if (skb2)
 295                                         rawv6_rcv(last, skb2);
 296                         }
 297                         last = sk;
 298                 }
 299         }
 300
 301         if (last) {
 302                 rawv6_rcv(last, skb);
 303                 read_unlock(&ip6_ra_lock);
 304                 return 1;
 305         }
 306         read_unlock(&ip6_ra_lock);
 307         return 0;
 308 }
 309
 310 static int ip6_forward_proxy_check(struct sk_buff *skb)
 311 {
 312         struct ipv6hdr *hdr = skb->nh.ipv6h;
 313         u8 nexthdr = hdr->nexthdr;
 314         int offset;
 315
 316         if (ipv6_ext_hdr(nexthdr)) {
 317                 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr);
 318                 if (offset < 0)
 319                         return 0;
 320         } else
 321                 offset = sizeof(struct ipv6hdr);
 322
 323         if (nexthdr == IPPROTO_ICMPV6) {
 324                 struct icmp6hdr *icmp6;
 325
 326                 if (!pskb_may_pull(skb, skb->nh.raw + offset + 1 - skb->data))
 327                         return 0;
 328
 329                 icmp6 = (struct icmp6hdr *)(skb->nh.raw + offset);
 330
 331                 switch (icmp6->icmp6_type) {
 332                 case NDISC_ROUTER_SOLICITATION:
 333                 case NDISC_ROUTER_ADVERTISEMENT:
 334                 case NDISC_NEIGHBOUR_SOLICITATION:
 335                 case NDISC_NEIGHBOUR_ADVERTISEMENT:
 336                 case NDISC_REDIRECT:
 337                         /* For reaction involving unicast neighbor discovery
 338                          * message destined to the proxied address, pass it to
 339                          * input function.
 340                          */
 341                         return 1;
 342                 default:
 343                         break;
 344                 }
 345         }
 346
 347         /*
 348          * The proxying router can't forward traffic sent to a link-local
 349          * address, so signal the sender and discard the packet. This
 350          * behavior is clarified by the MIPv6 specification.
 351          */
 352         if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
 353                 dst_link_failure(skb);
 354                 return -1;
 355         }
 356
 357         return 0;
 358 }
 359
 360 static inline int ip6_forward_finish(struct sk_buff *skb)
 361 {
 362         return dst_output(skb);
 363 }
 364
 365 int ip6_forward(struct sk_buff *skb)
 366 {
 367         struct dst_entry *dst = skb->dst;
 368         struct ipv6hdr *hdr = skb->nh.ipv6h;
 369         struct inet6_skb_parm *opt = IP6CB(skb);
 370
 371         if (ipv6_devconf.forwarding == 0)
 372                 goto error;
 373
 374         if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
 375                 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
 376                 goto drop;
 377         }
 378
 379         skb->ip_summed = CHECKSUM_NONE;
 380
 381         /*
 382          *      We DO NOT make any processing on
 383          *      RA packets, pushing them to user level AS IS
 384          *      without ane WARRANTY that application will be able
 385          *      to interpret them. The reason is that we
 386          *      cannot make anything clever here.
 387          *
 388          *      We are not end-node, so that if packet contains
 389          *      AH/ESP, we cannot make anything.
 390          *      Defragmentation also would be mistake, RA packets
 391          *      cannot be fragmented, because there is no warranty
 392          *      that different fragments will go along one path. --ANK
 393          */
 394         if (opt->ra) {
 395                 u8 *ptr = skb->nh.raw + opt->ra;
 396                 if (ip6_call_ra_chain(skb, (ptr[2]<<8) + ptr[3]))
 397                         return 0;
 398         }
 399
 400         /*
 401          *      check and decrement ttl
 402          */
 403         if (hdr->hop_limit <= 1) {
 404                 /* Force OUTPUT device used as source address */
 405                 skb->dev = dst->dev;
 406                 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT,
 407                             0, skb->dev);
 408                 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INHDRERRORS);
 409
 410                 kfree_skb(skb);
 411                 return -ETIMEDOUT;
 412         }
 413
 414         /* XXX: idev->cnf.proxy_ndp? */
 415         if (ipv6_devconf.proxy_ndp &&
 416             pneigh_lookup(&nd_tbl, &hdr->daddr, skb->dev, 0)) {
 417                 int proxied = ip6_forward_proxy_check(skb);
 418                 if (proxied > 0)
 419                         return ip6_input(skb);
 420                 else if (proxied < 0) {
 421                         IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
 422                         goto drop;
 423                 }
 424         }
 425
 426         if (!xfrm6_route_forward(skb)) {
 427                 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
 428                 goto drop;
 429         }
 430         dst = skb->dst;
 431
 432         /* IPv6 specs say nothing about it, but it is clear that we cannot
 433            send redirects to source routed frames.
 434          */
 435         if (skb->dev == dst->dev && dst->neighbour && opt->srcrt == 0) {
 436                 struct in6_addr *target = NULL;
 437                 struct rt6_info *rt;
 438                 struct neighbour *n = dst->neighbour;
 439
 440                 /*
 441                  *      incoming and outgoing devices are the same
 442                  *      send a redirect.
 443                  */
 444
 445                 rt = (struct rt6_info *) dst;
 446                 if ((rt->rt6i_flags & RTF_GATEWAY))
 447                         target = (struct in6_addr*)&n->primary_key;
 448                 else
 449                         target = &hdr->daddr;
 450
 451                 /* Limit redirects both by destination (here)
 452                    and by source (inside ndisc_send_redirect)
 453                  */
 454                 if (xrlim_allow(dst, 1*HZ))
 455                         ndisc_send_redirect(skb, n, target);
 456         } else if (ipv6_addr_type(&hdr->saddr)&(IPV6_ADDR_MULTICAST|IPV6_ADDR_LOOPBACK
 457                                                 |IPV6_ADDR_LINKLOCAL)) {
 458                 /* This check is security critical. */
 459                 goto error;
 460         }
 461
 462         if (skb->len > dst_mtu(dst)) {
 463                 /* Again, force OUTPUT device used as source address */
 464                 skb->dev = dst->dev;
 465                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, dst_mtu(dst), skb->dev);
 466                 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INTOOBIGERRORS);
 467                 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_FRAGFAILS);
 468                 kfree_skb(skb);
 469                 return -EMSGSIZE;
 470         }
 471
 472         if (skb_cow(skb, dst->dev->hard_header_len)) {
 473                 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_OUTDISCARDS);
 474                 goto drop;
 475         }
 476
 477         hdr = skb->nh.ipv6h;
 478
 479         /* Mangling hops number delayed to point after skb COW */
 480
 481         hdr->hop_limit--;
 482
 483         IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
 484         return NF_HOOK(PF_INET6,NF_IP6_FORWARD, skb, skb->dev, dst->dev, ip6_forward_finish);
 485
 486 error:
 487         IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
 488 drop:
 489         kfree_skb(skb);
 490         return -EINVAL;
 491 }
 492
 493 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
 494 {
 495         to->pkt_type = from->pkt_type;
 496         to->priority = from->priority;
 497         to->protocol = from->protocol;
 498         dst_release(to->dst);
 499         to->dst = dst_clone(from->dst);
 500         to->dev = from->dev;
 501         to->mark = from->mark;
 502
 503 #ifdef CONFIG_NET_SCHED
 504         to->tc_index = from->tc_index;
 505 #endif
 506 #ifdef CONFIG_NETFILTER
 507         /* Connection association is same as pre-frag packet */
 508         nf_conntrack_put(to->nfct);
 509         to->nfct = from->nfct;
 510         nf_conntrack_get(to->nfct);
 511         to->nfctinfo = from->nfctinfo;
 512 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
 513         nf_conntrack_put_reasm(to->nfct_reasm);
 514         to->nfct_reasm = from->nfct_reasm;
 515         nf_conntrack_get_reasm(to->nfct_reasm);
 516 #endif
 517 #ifdef CONFIG_BRIDGE_NETFILTER
 518         nf_bridge_put(to->nf_bridge);
 519         to->nf_bridge = from->nf_bridge;
 520         nf_bridge_get(to->nf_bridge);
 521 #endif
 522 #endif
 523         skb_copy_secmark(to, from);
 524 }
 525
 526 int ip6_find_1stfragopt(struct sk_buff *skb, u8 **nexthdr)
 527 {
 528         u16 offset = sizeof(struct ipv6hdr);
 529         struct ipv6_opt_hdr *exthdr = (struct ipv6_opt_hdr*)(skb->nh.ipv6h + 1);
 530         unsigned int packet_len = skb->tail - skb->nh.raw;
 531         int found_rhdr = 0;
 532         *nexthdr = &skb->nh.ipv6h->nexthdr;
 533
 534         while (offset + 1 <= packet_len) {
 535
 536                 switch (**nexthdr) {
 537
 538                 case NEXTHDR_HOP:
 539                         break;
 540                 case NEXTHDR_ROUTING:
 541                         found_rhdr = 1;
 542                         break;
 543                 case NEXTHDR_DEST:
 544 #ifdef CONFIG_IPV6_MIP6
 545                         if (ipv6_find_tlv(skb, offset, IPV6_TLV_HAO) >= 0)
 546                                 break;
 547 #endif
 548                         if (found_rhdr)
 549                                 return offset;
 550                         break;
 551                 default :
 552                         return offset;
 553                 }
 554
 555                 offset += ipv6_optlen(exthdr);
 556                 *nexthdr = &exthdr->nexthdr;
 557                 exthdr = (struct ipv6_opt_hdr*)(skb->nh.raw + offset);
 558         }
 559
 560         return offset;
 561 }
 562 EXPORT_SYMBOL_GPL(ip6_find_1stfragopt);
 563
 564 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
 565 {
 566         struct net_device *dev;
 567         struct sk_buff *frag;
 568         struct rt6_info *rt = (struct rt6_info*)skb->dst;
 569         struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
 570         struct ipv6hdr *tmp_hdr;
 571         struct frag_hdr *fh;
 572         unsigned int mtu, hlen, left, len;
 573         __be32 frag_id = 0;
 574         int ptr, offset = 0, err=0;
 575         u8 *prevhdr, nexthdr = 0;
 576
 577         dev = rt->u.dst.dev;
 578         hlen = ip6_find_1stfragopt(skb, &prevhdr);
 579         nexthdr = *prevhdr;
 580
 581         mtu = dst_mtu(&rt->u.dst);
 582         if (np && np->frag_size < mtu) {
 583                 if (np->frag_size)
 584                         mtu = np->frag_size;
 585         }
 586         mtu -= hlen + sizeof(struct frag_hdr);
 587
 588         if (skb_shinfo(skb)->frag_list) {
 589                 int first_len = skb_pagelen(skb);
 590
 591                 if (first_len - hlen > mtu ||
 592                     ((first_len - hlen) & 7) ||
 593                     skb_cloned(skb))
 594                         goto slow_path;
 595
 596                 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
 597                         /* Correct geometry. */
 598                         if (frag->len > mtu ||
 599                             ((frag->len & 7) && frag->next) ||
 600                             skb_headroom(frag) < hlen)
 601                             goto slow_path;
 602
 603                         /* Partially cloned skb? */
 604                         if (skb_shared(frag))
 605                                 goto slow_path;
 606
 607                         BUG_ON(frag->sk);
 608                         if (skb->sk) {
 609                                 sock_hold(skb->sk);
 610                                 frag->sk = skb->sk;
 611                                 frag->destructor = sock_wfree;
 612                                 skb->truesize -= frag->truesize;
 613                         }
 614                 }
 615
 616                 err = 0;
 617                 offset = 0;
 618                 frag = skb_shinfo(skb)->frag_list;
 619                 skb_shinfo(skb)->frag_list = NULL;
 620                 /* BUILD HEADER */
 621
 622                 *prevhdr = NEXTHDR_FRAGMENT;
 623                 tmp_hdr = kmemdup(skb->nh.raw, hlen, GFP_ATOMIC);
 624                 if (!tmp_hdr) {
 625                         IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGFAILS);
 626                         return -ENOMEM;
 627                 }
 628
 629                 __skb_pull(skb, hlen);
 630                 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
 631                 __skb_push(skb, hlen);
 632                 skb_reset_network_header(skb);
 633                 memcpy(skb->nh.raw, tmp_hdr, hlen);
 634
 635                 ipv6_select_ident(skb, fh);
 636                 fh->nexthdr = nexthdr;
 637                 fh->reserved = 0;
 638                 fh->frag_off = htons(IP6_MF);
 639                 frag_id = fh->identification;
 640
 641                 first_len = skb_pagelen(skb);
 642                 skb->data_len = first_len - skb_headlen(skb);
 643                 skb->len = first_len;
 644                 skb->nh.ipv6h->payload_len = htons(first_len - sizeof(struct ipv6hdr));
 645
 646                 dst_hold(&rt->u.dst);
 647
 648                 for (;;) {
 649                         /* Prepare header of the next frame,
 650                          * before previous one went down. */
 651                         if (frag) {
 652                                 frag->ip_summed = CHECKSUM_NONE;
 653                                 frag->h.raw = frag->data;
 654                                 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
 655                                 __skb_push(frag, hlen);
 656                                 skb_reset_network_header(frag);
 657                                 memcpy(frag->nh.raw, tmp_hdr, hlen);
 658                                 offset += skb->len - hlen - sizeof(struct frag_hdr);
 659                                 fh->nexthdr = nexthdr;
 660                                 fh->reserved = 0;
 661                                 fh->frag_off = htons(offset);
 662                                 if (frag->next != NULL)
 663                                         fh->frag_off |= htons(IP6_MF);
 664                                 fh->identification = frag_id;
 665                                 frag->nh.ipv6h->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
 666                                 ip6_copy_metadata(frag, skb);
 667                         }
 668
 669                         err = output(skb);
 670                         if(!err)
 671                                 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGCREATES);
 672
 673                         if (err || !frag)
 674                                 break;
 675
 676                         skb = frag;
 677                         frag = skb->next;
 678                         skb->next = NULL;
 679                 }
 680
 681                 kfree(tmp_hdr);
 682
 683                 if (err == 0) {
 684                         IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGOKS);
 685                         dst_release(&rt->u.dst);
 686                         return 0;
 687                 }
 688
 689                 while (frag) {
 690                         skb = frag->next;
 691                         kfree_skb(frag);
 692                         frag = skb;
 693                 }
 694
 695                 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGFAILS);
 696                 dst_release(&rt->u.dst);
 697                 return err;
 698         }
 699
 700 slow_path:
 701         left = skb->len - hlen;         /* Space per frame */
 702         ptr = hlen;                     /* Where to start from */
 703
 704         /*
 705          *      Fragment the datagram.
 706          */
 707
 708         *prevhdr = NEXTHDR_FRAGMENT;
 709
 710         /*
 711          *      Keep copying data until we run out.
 712          */
 713         while(left > 0) {
 714                 len = left;
 715                 /* IF: it doesn't fit, use 'mtu' - the data space left */
 716                 if (len > mtu)
 717                         len = mtu;
 718                 /* IF: we are not sending upto and including the packet end
 719                    then align the next start on an eight byte boundary */
 720                 if (len < left) {
 721                         len &= ~7;
 722                 }
 723                 /*
 724                  *      Allocate buffer.
 725                  */
 726
 727                 if ((frag = alloc_skb(len+hlen+sizeof(struct frag_hdr)+LL_RESERVED_SPACE(rt->u.dst.dev), GFP_ATOMIC)) == NULL) {
 728                         NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
 729                         IP6_INC_STATS(ip6_dst_idev(skb->dst),
 730                                       IPSTATS_MIB_FRAGFAILS);
 731                         err = -ENOMEM;
 732                         goto fail;
 733                 }
 734
 735                 /*
 736                  *      Set up data on packet
 737                  */
 738
 739                 ip6_copy_metadata(frag, skb);
 740                 skb_reserve(frag, LL_RESERVED_SPACE(rt->u.dst.dev));
 741                 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
 742                 skb_reset_network_header(frag);
 743                 fh = (struct frag_hdr*)(frag->data + hlen);
 744                 frag->h.raw = frag->data + hlen + sizeof(struct frag_hdr);
 745
 746                 /*
 747                  *      Charge the memory for the fragment to any owner
 748                  *      it might possess
 749                  */
 750                 if (skb->sk)
 751                         skb_set_owner_w(frag, skb->sk);
 752
 753                 /*
 754                  *      Copy the packet header into the new buffer.
 755                  */
 756                 memcpy(frag->nh.raw, skb->data, hlen);
 757
 758                 /*
 759                  *      Build fragment header.
 760                  */
 761                 fh->nexthdr = nexthdr;
 762                 fh->reserved = 0;
 763                 if (!frag_id) {
 764                         ipv6_select_ident(skb, fh);
 765                         frag_id = fh->identification;
 766                 } else
 767                         fh->identification = frag_id;
 768
 769                 /*
 770                  *      Copy a block of the IP datagram.
 771                  */
 772                 if (skb_copy_bits(skb, ptr, frag->h.raw, len))
 773                         BUG();
 774                 left -= len;
 775
 776                 fh->frag_off = htons(offset);
 777                 if (left > 0)
 778                         fh->frag_off |= htons(IP6_MF);
 779                 frag->nh.ipv6h->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
 780
 781                 ptr += len;
 782                 offset += len;
 783
 784                 /*
 785                  *      Put this fragment into the sending queue.
 786                  */
 787                 err = output(frag);
 788                 if (err)
 789                         goto fail;
 790
 791                 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGCREATES);
 792         }
 793         IP6_INC_STATS(ip6_dst_idev(skb->dst),
 794                       IPSTATS_MIB_FRAGOKS);
 795         kfree_skb(skb);
 796         return err;
 797
 798 fail:
 799         IP6_INC_STATS(ip6_dst_idev(skb->dst),
 800                       IPSTATS_MIB_FRAGFAILS);
 801         kfree_skb(skb);
 802         return err;
 803 }
 804
 805 static inline int ip6_rt_check(struct rt6key *rt_key,
 806                                struct in6_addr *fl_addr,
 807                                struct in6_addr *addr_cache)
 808 {
 809         return ((rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
 810                 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache)));
 811 }
 812
 813 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
 814                                           struct dst_entry *dst,
 815                                           struct flowi *fl)
 816 {
 817         struct ipv6_pinfo *np = inet6_sk(sk);
 818         struct rt6_info *rt = (struct rt6_info *)dst;
 819
 820         if (!dst)
 821                 goto out;
 822
 823         /* Yes, checking route validity in not connected
 824          * case is not very simple. Take into account,
 825          * that we do not support routing by source, TOS,
 826          * and MSG_DONTROUTE            --ANK (980726)
 827          *
 828          * 1. ip6_rt_check(): If route was host route,
 829          *    check that cached destination is current.
 830          *    If it is network route, we still may
 831          *    check its validity using saved pointer
 832          *    to the last used address: daddr_cache.
 833          *    We do not want to save whole address now,
 834          *    (because main consumer of this service
 835          *    is tcp, which has not this problem),
 836          *    so that the last trick works only on connected
 837          *    sockets.
 838          * 2. oif also should be the same.
 839          */
 840         if (ip6_rt_check(&rt->rt6i_dst, &fl->fl6_dst, np->daddr_cache) ||
 841 #ifdef CONFIG_IPV6_SUBTREES
 842             ip6_rt_check(&rt->rt6i_src, &fl->fl6_src, np->saddr_cache) ||
 843 #endif
 844             (fl->oif && fl->oif != dst->dev->ifindex)) {
 845                 dst_release(dst);
 846                 dst = NULL;
 847         }
 848
 849 out:
 850         return dst;
 851 }
 852
 853 static int ip6_dst_lookup_tail(struct sock *sk,
 854                                struct dst_entry **dst, struct flowi *fl)
 855 {
 856         int err;
 857
 858         if (*dst == NULL)
 859                 *dst = ip6_route_output(sk, fl);
 860
 861         if ((err = (*dst)->error))
 862                 goto out_err_release;
 863
 864         if (ipv6_addr_any(&fl->fl6_src)) {
 865                 err = ipv6_get_saddr(*dst, &fl->fl6_dst, &fl->fl6_src);
 866                 if (err)
 867                         goto out_err_release;
 868         }
 869
 870 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
 871                 /*
 872                  * Here if the dst entry we've looked up
 873                  * has a neighbour entry that is in the INCOMPLETE
 874                  * state and the src address from the flow is
 875                  * marked as OPTIMISTIC, we release the found
 876                  * dst entry and replace it instead with the
 877                  * dst entry of the nexthop router
 878                  */
 879                 if (!((*dst)->neighbour->nud_state & NUD_VALID)) {
 880                         struct inet6_ifaddr *ifp;
 881                         struct flowi fl_gw;
 882                         int redirect;
 883
 884                         ifp = ipv6_get_ifaddr(&fl->fl6_src, (*dst)->dev, 1);
 885
 886                         redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
 887                         if (ifp)
 888                                 in6_ifa_put(ifp);
 889
 890                         if (redirect) {
 891                                 /*
 892                                  * We need to get the dst entry for the
 893                                  * default router instead
 894                                  */
 895                                 dst_release(*dst);
 896                                 memcpy(&fl_gw, fl, sizeof(struct flowi));
 897                                 memset(&fl_gw.fl6_dst, 0, sizeof(struct in6_addr));
 898                                 *dst = ip6_route_output(sk, &fl_gw);
 899                                 if ((err = (*dst)->error))
 900                                         goto out_err_release;
 901                         }
 902                 }
 903 #endif
 904
 905         return 0;
 906
 907 out_err_release:
 908         dst_release(*dst);
 909         *dst = NULL;
 910         return err;
 911 }
 912
 913 /**
 914  *      ip6_dst_lookup - perform route lookup on flow
 915  *      @sk: socket which provides route info
 916  *      @dst: pointer to dst_entry * for result
 917  *      @fl: flow to lookup
 918  *
 919  *      This function performs a route lookup on the given flow.
 920  *
 921  *      It returns zero on success, or a standard errno code on error.
 922  */
 923 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
 924 {
 925         *dst = NULL;
 926         return ip6_dst_lookup_tail(sk, dst, fl);
 927 }
 928 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
 929
 930 /**
 931  *      ip6_sk_dst_lookup - perform socket cached route lookup on flow
 932  *      @sk: socket which provides the dst cache and route info
 933  *      @dst: pointer to dst_entry * for result
 934  *      @fl: flow to lookup
 935  *
 936  *      This function performs a route lookup on the given flow with the
 937  *      possibility of using the cached route in the socket if it is valid.
 938  *      It will take the socket dst lock when operating on the dst cache.
 939  *      As a result, this function can only be used in process context.
 940  *
 941  *      It returns zero on success, or a standard errno code on error.
 942  */
 943 int ip6_sk_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
 944 {
 945         *dst = NULL;
 946         if (sk) {
 947                 *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
 948                 *dst = ip6_sk_dst_check(sk, *dst, fl);
 949         }
 950
 951         return ip6_dst_lookup_tail(sk, dst, fl);
 952 }
 953 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup);
 954
 955 static inline int ip6_ufo_append_data(struct sock *sk,
 956                         int getfrag(void *from, char *to, int offset, int len,
 957                         int odd, struct sk_buff *skb),
 958                         void *from, int length, int hh_len, int fragheaderlen,
 959                         int transhdrlen, int mtu,unsigned int flags)
 960
 961 {
 962         struct sk_buff *skb;
 963         int err;
 964
 965         /* There is support for UDP large send offload by network
 966          * device, so create one single skb packet containing complete
 967          * udp datagram
 968          */
 969         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
 970                 skb = sock_alloc_send_skb(sk,
 971                         hh_len + fragheaderlen + transhdrlen + 20,
 972                         (flags & MSG_DONTWAIT), &err);
 973                 if (skb == NULL)
 974                         return -ENOMEM;
 975
 976                 /* reserve space for Hardware header */
 977                 skb_reserve(skb, hh_len);
 978
 979                 /* create space for UDP/IP header */
 980                 skb_put(skb,fragheaderlen + transhdrlen);
 981
 982                 /* initialize network header pointer */
 983                 skb_reset_network_header(skb);
 984
 985                 /* initialize protocol header pointer */
 986                 skb->h.raw = skb->data + fragheaderlen;
 987
 988                 skb->ip_summed = CHECKSUM_PARTIAL;
 989                 skb->csum = 0;
 990                 sk->sk_sndmsg_off = 0;
 991         }
 992
 993         err = skb_append_datato_frags(sk,skb, getfrag, from,
 994                                       (length - transhdrlen));
 995         if (!err) {
 996                 struct frag_hdr fhdr;
 997
 998                 /* specify the length of each IP datagram fragment*/
 999                 skb_shinfo(skb)->gso_size = mtu - fragheaderlen -
1000                                             sizeof(struct frag_hdr);
1001                 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1002                 ipv6_select_ident(skb, &fhdr);
1003                 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1004                 __skb_queue_tail(&sk->sk_write_queue, skb);
1005
1006                 return 0;
1007         }
1008         /* There is not enough support do UPD LSO,
1009          * so follow normal path
1010          */
1011         kfree_skb(skb);
1012
1013         return err;
1014 }
1015
1016 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1017         int offset, int len, int odd, struct sk_buff *skb),
1018         void *from, int length, int transhdrlen,
1019         int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi *fl,
1020         struct rt6_info *rt, unsigned int flags)
1021 {
1022         struct inet_sock *inet = inet_sk(sk);
1023         struct ipv6_pinfo *np = inet6_sk(sk);
1024         struct sk_buff *skb;
1025         unsigned int maxfraglen, fragheaderlen;
1026         int exthdrlen;
1027         int hh_len;
1028         int mtu;
1029         int copy;
1030         int err;
1031         int offset = 0;
1032         int csummode = CHECKSUM_NONE;
1033
1034         if (flags&MSG_PROBE)
1035                 return 0;
1036         if (skb_queue_empty(&sk->sk_write_queue)) {
1037                 /*
1038                  * setup for corking
1039                  */
1040                 if (opt) {
1041                         if (np->cork.opt == NULL) {
1042                                 np->cork.opt = kmalloc(opt->tot_len,
1043                                                        sk->sk_allocation);
1044                                 if (unlikely(np->cork.opt == NULL))
1045                                         return -ENOBUFS;
1046                         } else if (np->cork.opt->tot_len < opt->tot_len) {
1047                                 printk(KERN_DEBUG "ip6_append_data: invalid option length\n");
1048                                 return -EINVAL;
1049                         }
1050                         memcpy(np->cork.opt, opt, opt->tot_len);
1051                         inet->cork.flags |= IPCORK_OPT;
1052                         /* need source address above miyazawa*/
1053                 }
1054                 dst_hold(&rt->u.dst);
1055                 np->cork.rt = rt;
1056                 inet->cork.fl = *fl;
1057                 np->cork.hop_limit = hlimit;
1058                 np->cork.tclass = tclass;
1059                 mtu = dst_mtu(rt->u.dst.path);
1060                 if (np->frag_size < mtu) {
1061                         if (np->frag_size)
1062                                 mtu = np->frag_size;
1063                 }
1064                 inet->cork.fragsize = mtu;
1065                 if (dst_allfrag(rt->u.dst.path))
1066                         inet->cork.flags |= IPCORK_ALLFRAG;
1067                 inet->cork.length = 0;
1068                 sk->sk_sndmsg_page = NULL;
1069                 sk->sk_sndmsg_off = 0;
1070                 exthdrlen = rt->u.dst.header_len + (opt ? opt->opt_flen : 0);
1071                 length += exthdrlen;
1072                 transhdrlen += exthdrlen;
1073         } else {
1074                 rt = np->cork.rt;
1075                 fl = &inet->cork.fl;
1076                 if (inet->cork.flags & IPCORK_OPT)
1077                         opt = np->cork.opt;
1078                 transhdrlen = 0;
1079                 exthdrlen = 0;
1080                 mtu = inet->cork.fragsize;
1081         }
1082
1083         hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1084
1085         fragheaderlen = sizeof(struct ipv6hdr) + rt->u.dst.nfheader_len + (opt ? opt->opt_nflen : 0);
1086         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
1087
1088         if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1089                 if (inet->cork.length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
1090                         ipv6_local_error(sk, EMSGSIZE, fl, mtu-exthdrlen);
1091                         return -EMSGSIZE;
1092                 }
1093         }
1094
1095         /*
1096          * Let's try using as much space as possible.
1097          * Use MTU if total length of the message fits into the MTU.
1098          * Otherwise, we need to reserve fragment header and
1099          * fragment alignment (= 8-15 octects, in total).
1100          *
1101          * Note that we may need to "move" the data from the tail of
1102          * of the buffer to the new fragment when we split
1103          * the message.
1104          *
1105          * FIXME: It may be fragmented into multiple chunks
1106          *        at once if non-fragmentable extension headers
1107          *        are too large.
1108          * --yoshfuji
1109          */
1110
1111         inet->cork.length += length;
1112         if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
1113             (rt->u.dst.dev->features & NETIF_F_UFO)) {
1114
1115                 err = ip6_ufo_append_data(sk, getfrag, from, length, hh_len,
1116                                           fragheaderlen, transhdrlen, mtu,
1117                                           flags);
1118                 if (err)
1119                         goto error;
1120                 return 0;
1121         }
1122
1123         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1124                 goto alloc_new_skb;
1125
1126         while (length > 0) {
1127                 /* Check if the remaining data fits into current packet. */
1128                 copy = (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1129                 if (copy < length)
1130                         copy = maxfraglen - skb->len;
1131
1132                 if (copy <= 0) {
1133                         char *data;
1134                         unsigned int datalen;
1135                         unsigned int fraglen;
1136                         unsigned int fraggap;
1137                         unsigned int alloclen;
1138                         struct sk_buff *skb_prev;
1139 alloc_new_skb:
1140                         skb_prev = skb;
1141
1142                         /* There's no room in the current skb */
1143                         if (skb_prev)
1144                                 fraggap = skb_prev->len - maxfraglen;
1145                         else
1146                                 fraggap = 0;
1147
1148                         /*
1149                          * If remaining data exceeds the mtu,
1150                          * we know we need more fragment(s).
1151                          */
1152                         datalen = length + fraggap;
1153                         if (datalen > (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1154                                 datalen = maxfraglen - fragheaderlen;
1155
1156                         fraglen = datalen + fragheaderlen;
1157                         if ((flags & MSG_MORE) &&
1158                             !(rt->u.dst.dev->features&NETIF_F_SG))
1159                                 alloclen = mtu;
1160                         else
1161                                 alloclen = datalen + fragheaderlen;
1162
1163                         /*
1164                          * The last fragment gets additional space at tail.
1165                          * Note: we overallocate on fragments with MSG_MODE
1166                          * because we have no idea if we're the last one.
1167                          */
1168                         if (datalen == length + fraggap)
1169                                 alloclen += rt->u.dst.trailer_len;
1170
1171                         /*
1172                          * We just reserve space for fragment header.
1173                          * Note: this may be overallocation if the message
1174                          * (without MSG_MORE) fits into the MTU.
1175                          */
1176                         alloclen += sizeof(struct frag_hdr);
1177
1178                         if (transhdrlen) {
1179                                 skb = sock_alloc_send_skb(sk,
1180                                                 alloclen + hh_len,
1181                                                 (flags & MSG_DONTWAIT), &err);
1182                         } else {
1183                                 skb = NULL;
1184                                 if (atomic_read(&sk->sk_wmem_alloc) <=
1185                                     2 * sk->sk_sndbuf)
1186                                         skb = sock_wmalloc(sk,
1187                                                            alloclen + hh_len, 1,
1188                                                            sk->sk_allocation);
1189                                 if (unlikely(skb == NULL))
1190                                         err = -ENOBUFS;
1191                         }
1192                         if (skb == NULL)
1193                                 goto error;
1194                         /*
1195                          *      Fill in the control structures
1196                          */
1197                         skb->ip_summed = csummode;
1198                         skb->csum = 0;
1199                         /* reserve for fragmentation */
1200                         skb_reserve(skb, hh_len+sizeof(struct frag_hdr));
1201
1202                         /*
1203                          *      Find where to start putting bytes
1204                          */
1205                         data = skb_put(skb, fraglen);
1206                         skb->nh.raw = data + exthdrlen;
1207                         data += fragheaderlen;
1208                         skb->h.raw = data + exthdrlen;
1209
1210                         if (fraggap) {
1211                                 skb->csum = skb_copy_and_csum_bits(
1212                                         skb_prev, maxfraglen,
1213                                         data + transhdrlen, fraggap, 0);
1214                                 skb_prev->csum = csum_sub(skb_prev->csum,
1215                                                           skb->csum);
1216                                 data += fraggap;
1217                                 pskb_trim_unique(skb_prev, maxfraglen);
1218                         }
1219                         copy = datalen - transhdrlen - fraggap;
1220                         if (copy < 0) {
1221                                 err = -EINVAL;
1222                                 kfree_skb(skb);
1223                                 goto error;
1224                         } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1225                                 err = -EFAULT;
1226                                 kfree_skb(skb);
1227                                 goto error;
1228                         }
1229
1230                         offset += copy;
1231                         length -= datalen - fraggap;
1232                         transhdrlen = 0;
1233                         exthdrlen = 0;
1234                         csummode = CHECKSUM_NONE;
1235
1236                         /*
1237                          * Put the packet on the pending queue
1238                          */
1239                         __skb_queue_tail(&sk->sk_write_queue, skb);
1240                         continue;
1241                 }
1242
1243                 if (copy > length)
1244                         copy = length;
1245
1246                 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
1247                         unsigned int off;
1248
1249                         off = skb->len;
1250                         if (getfrag(from, skb_put(skb, copy),
1251                                                 offset, copy, off, skb) < 0) {
1252                                 __skb_trim(skb, off);
1253                                 err = -EFAULT;
1254                                 goto error;
1255                         }
1256                 } else {
1257                         int i = skb_shinfo(skb)->nr_frags;
1258                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1259                         struct page *page = sk->sk_sndmsg_page;
1260                         int off = sk->sk_sndmsg_off;
1261                         unsigned int left;
1262
1263                         if (page && (left = PAGE_SIZE - off) > 0) {
1264                                 if (copy >= left)
1265                                         copy = left;
1266                                 if (page != frag->page) {
1267                                         if (i == MAX_SKB_FRAGS) {
1268                                                 err = -EMSGSIZE;
1269                                                 goto error;
1270                                         }
1271                                         get_page(page);
1272                                         skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1273                                         frag = &skb_shinfo(skb)->frags[i];
1274                                 }
1275                         } else if(i < MAX_SKB_FRAGS) {
1276                                 if (copy > PAGE_SIZE)
1277                                         copy = PAGE_SIZE;
1278                                 page = alloc_pages(sk->sk_allocation, 0);
1279                                 if (page == NULL) {
1280                                         err = -ENOMEM;
1281                                         goto error;
1282                                 }
1283                                 sk->sk_sndmsg_page = page;
1284                                 sk->sk_sndmsg_off = 0;
1285
1286                                 skb_fill_page_desc(skb, i, page, 0, 0);
1287                                 frag = &skb_shinfo(skb)->frags[i];
1288                                 skb->truesize += PAGE_SIZE;
1289                                 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1290                         } else {
1291                                 err = -EMSGSIZE;
1292                                 goto error;
1293                         }
1294                         if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1295                                 err = -EFAULT;
1296                                 goto error;
1297                         }
1298                         sk->sk_sndmsg_off += copy;
1299                         frag->size += copy;
1300                         skb->len += copy;
1301                         skb->data_len += copy;
1302                 }
1303                 offset += copy;
1304                 length -= copy;
1305         }
1306         return 0;
1307 error:
1308         inet->cork.length -= length;
1309         IP6_INC_STATS(rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1310         return err;
1311 }
1312
1313 int ip6_push_pending_frames(struct sock *sk)
1314 {
1315         struct sk_buff *skb, *tmp_skb;
1316         struct sk_buff **tail_skb;
1317         struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1318         struct inet_sock *inet = inet_sk(sk);
1319         struct ipv6_pinfo *np = inet6_sk(sk);
1320         struct ipv6hdr *hdr;
1321         struct ipv6_txoptions *opt = np->cork.opt;
1322         struct rt6_info *rt = np->cork.rt;
1323         struct flowi *fl = &inet->cork.fl;
1324         unsigned char proto = fl->proto;
1325         int err = 0;
1326
1327         if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1328                 goto out;
1329         tail_skb = &(skb_shinfo(skb)->frag_list);
1330
1331         /* move skb->data to ip header from ext header */
1332         if (skb->data < skb->nh.raw)
1333                 __skb_pull(skb, skb_network_offset(skb));
1334         while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1335                 __skb_pull(tmp_skb, skb->h.raw - skb->nh.raw);
1336                 *tail_skb = tmp_skb;
1337                 tail_skb = &(tmp_skb->next);
1338                 skb->len += tmp_skb->len;
1339                 skb->data_len += tmp_skb->len;
1340                 skb->truesize += tmp_skb->truesize;
1341                 __sock_put(tmp_skb->sk);
1342                 tmp_skb->destructor = NULL;
1343                 tmp_skb->sk = NULL;
1344         }
1345
1346         ipv6_addr_copy(final_dst, &fl->fl6_dst);
1347         __skb_pull(skb, skb->h.raw - skb->nh.raw);
1348         if (opt && opt->opt_flen)
1349                 ipv6_push_frag_opts(skb, opt, &proto);
1350         if (opt && opt->opt_nflen)
1351                 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1352
1353         skb_push(skb, sizeof(struct ipv6hdr));
1354         skb_reset_network_header(skb);
1355         hdr = skb->nh.ipv6h;
1356
1357         *(__be32*)hdr = fl->fl6_flowlabel |
1358                      htonl(0x60000000 | ((int)np->cork.tclass << 20));
1359
1360         if (skb->len <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN)
1361                 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
1362         else
1363                 hdr->payload_len = 0;
1364         hdr->hop_limit = np->cork.hop_limit;
1365         hdr->nexthdr = proto;
1366         ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
1367         ipv6_addr_copy(&hdr->daddr, final_dst);
1368
1369         skb->priority = sk->sk_priority;
1370
1371         skb->dst = dst_clone(&rt->u.dst);
1372         IP6_INC_STATS(rt->rt6i_idev, IPSTATS_MIB_OUTREQUESTS);
1373         err = NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, skb->dst->dev, dst_output);
1374         if (err) {
1375                 if (err > 0)
1376                         err = np->recverr ? net_xmit_errno(err) : 0;
1377                 if (err)
1378                         goto error;
1379         }
1380
1381 out:
1382         inet->cork.flags &= ~IPCORK_OPT;
1383         kfree(np->cork.opt);
1384         np->cork.opt = NULL;
1385         if (np->cork.rt) {
1386                 dst_release(&np->cork.rt->u.dst);
1387                 np->cork.rt = NULL;
1388                 inet->cork.flags &= ~IPCORK_ALLFRAG;
1389         }
1390         memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1391         return err;
1392 error:
1393         goto out;
1394 }
1395
1396 void ip6_flush_pending_frames(struct sock *sk)
1397 {
1398         struct inet_sock *inet = inet_sk(sk);
1399         struct ipv6_pinfo *np = inet6_sk(sk);
1400         struct sk_buff *skb;
1401
1402         while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1403                 IP6_INC_STATS(ip6_dst_idev(skb->dst),
1404                               IPSTATS_MIB_OUTDISCARDS);
1405                 kfree_skb(skb);
1406         }
1407
1408         inet->cork.flags &= ~IPCORK_OPT;
1409
1410         kfree(np->cork.opt);
1411         np->cork.opt = NULL;
1412         if (np->cork.rt) {
1413                 dst_release(&np->cork.rt->u.dst);
1414                 np->cork.rt = NULL;
1415                 inet->cork.flags &= ~IPCORK_ALLFRAG;
1416         }
1417         memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1418 }