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
  42 #include <linux/netfilter.h>
  43 #include <linux/netfilter_ipv6.h>
  44
  45 #include <net/sock.h>
  46 #include <net/snmp.h>
  47
  48 #include <net/ipv6.h>
  49 #include <net/ndisc.h>
  50 #include <net/protocol.h>
  51 #include <net/ip6_route.h>
  52 #include <net/addrconf.h>
  53 #include <net/rawv6.h>
  54 #include <net/icmp.h>
  55 #include <net/xfrm.h>
  56 #include <net/checksum.h>
  57
  58 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *));
  59
  60 static __inline__ void ipv6_select_ident(struct sk_buff *skb, struct frag_hdr *fhdr)
  61 {
  62         static u32 ipv6_fragmentation_id = 1;
  63         static DEFINE_SPINLOCK(ip6_id_lock);
  64
  65         spin_lock_bh(&ip6_id_lock);
  66         fhdr->identification = htonl(ipv6_fragmentation_id);
  67         if (++ipv6_fragmentation_id == 0)
  68                 ipv6_fragmentation_id = 1;
  69         spin_unlock_bh(&ip6_id_lock);
  70 }
  71
  72 static inline int ip6_output_finish(struct sk_buff *skb)
  73 {
  74         struct dst_entry *dst = skb->dst;
  75
  76         if (dst->hh)
  77                 return neigh_hh_output(dst->hh, skb);
  78         else if (dst->neighbour)
  79                 return dst->neighbour->output(skb);
  80
  81         IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
  82         kfree_skb(skb);
  83         return -EINVAL;
  84
  85 }
  86
  87 /* dev_loopback_xmit for use with netfilter. */
  88 static int ip6_dev_loopback_xmit(struct sk_buff *newskb)
  89 {
  90         newskb->mac.raw = newskb->data;
  91         __skb_pull(newskb, newskb->nh.raw - newskb->data);
  92         newskb->pkt_type = PACKET_LOOPBACK;
  93         newskb->ip_summed = CHECKSUM_UNNECESSARY;
  94         BUG_TRAP(newskb->dst);
  95
  96         netif_rx(newskb);
  97         return 0;
  98 }
  99
 100
 101 static int ip6_output2(struct sk_buff *skb)
 102 {
 103         struct dst_entry *dst = skb->dst;
 104         struct net_device *dev = dst->dev;
 105
 106         skb->protocol = htons(ETH_P_IPV6);
 107         skb->dev = dev;
 108
 109         if (ipv6_addr_is_multicast(&skb->nh.ipv6h->daddr)) {
 110                 struct ipv6_pinfo* np = skb->sk ? inet6_sk(skb->sk) : NULL;
 111                 struct inet6_dev *idev = ip6_dst_idev(skb->dst);
 112
 113                 if (!(dev->flags & IFF_LOOPBACK) && (!np || np->mc_loop) &&
 114                     ipv6_chk_mcast_addr(dev, &skb->nh.ipv6h->daddr,
 115                                 &skb->nh.ipv6h->saddr)) {
 116                         struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
 117
 118                         /* Do not check for IFF_ALLMULTI; multicast routing
 119                            is not supported in any case.
 120                          */
 121                         if (newskb)
 122                                 NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, newskb, NULL,
 123                                         newskb->dev,
 124                                         ip6_dev_loopback_xmit);
 125
 126                         if (skb->nh.ipv6h->hop_limit == 0) {
 127                                 IP6_INC_STATS(idev, IPSTATS_MIB_OUTDISCARDS);
 128                                 kfree_skb(skb);
 129                                 return 0;
 130                         }
 131                 }
 132
 133                 IP6_INC_STATS(idev, IPSTATS_MIB_OUTMCASTPKTS);
 134         }
 135
 136         return NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, skb,NULL, skb->dev,ip6_output_finish);
 137 }
 138
 139 int ip6_output(struct sk_buff *skb)
 140 {
 141         if ((skb->len > dst_mtu(skb->dst) && !skb_is_gso(skb)) ||
 142                                 dst_allfrag(skb->dst))
 143                 return ip6_fragment(skb, ip6_output2);
 144         else
 145                 return ip6_output2(skb);
 146 }
 147
 148 /*
 149  *      xmit an sk_buff (used by TCP)
 150  */
 151
 152 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
 153              struct ipv6_txoptions *opt, int ipfragok)
 154 {
 155         struct ipv6_pinfo *np = inet6_sk(sk);
 156         struct in6_addr *first_hop = &fl->fl6_dst;
 157         struct dst_entry *dst = skb->dst;
 158         struct ipv6hdr *hdr;
 159         u8  proto = fl->proto;
 160         int seg_len = skb->len;
 161         int hlimit, tclass;
 162         u32 mtu;
 163
 164         if (opt) {
 165                 int head_room;
 166
 167                 /* First: exthdrs may take lots of space (~8K for now)
 168                    MAX_HEADER is not enough.
 169                  */
 170                 head_room = opt->opt_nflen + opt->opt_flen;
 171                 seg_len += head_room;
 172                 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
 173
 174                 if (skb_headroom(skb) < head_room) {
 175                         struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
 176                         if (skb2 == NULL) {
 177                                 IP6_INC_STATS(ip6_dst_idev(skb->dst),
 178                                               IPSTATS_MIB_OUTDISCARDS);
 179                                 kfree_skb(skb);
 180                                 return -ENOBUFS;
 181                         }
 182                         kfree_skb(skb);
 183                         skb = skb2;
 184                         if (sk)
 185                                 skb_set_owner_w(skb, sk);
 186                 }
 187                 if (opt->opt_flen)
 188                         ipv6_push_frag_opts(skb, opt, &proto);
 189                 if (opt->opt_nflen)
 190                         ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
 191         }
 192
 193         hdr = skb->nh.ipv6h = (struct ipv6hdr*)skb_push(skb, sizeof(struct ipv6hdr));
 194
 195         /*
 196          *      Fill in the IPv6 header
 197          */
 198
 199         hlimit = -1;
 200         if (np)
 201                 hlimit = np->hop_limit;
 202         if (hlimit < 0)
 203                 hlimit = dst_metric(dst, RTAX_HOPLIMIT);
 204         if (hlimit < 0)
 205                 hlimit = ipv6_get_hoplimit(dst->dev);
 206
 207         tclass = -1;
 208         if (np)
 209                 tclass = np->tclass;
 210         if (tclass < 0)
 211                 tclass = 0;
 212
 213         *(__be32 *)hdr = htonl(0x60000000 | (tclass << 20)) | fl->fl6_flowlabel;
 214
 215         hdr->payload_len = htons(seg_len);
 216         hdr->nexthdr = proto;
 217         hdr->hop_limit = hlimit;
 218
 219         ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
 220         ipv6_addr_copy(&hdr->daddr, first_hop);
 221
 222         skb->priority = sk->sk_priority;
 223
 224         mtu = dst_mtu(dst);
 225         if ((skb->len <= mtu) || ipfragok || skb_is_gso(skb)) {
 226                 IP6_INC_STATS(ip6_dst_idev(skb->dst),
 227                               IPSTATS_MIB_OUTREQUESTS);
 228                 return NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, dst->dev,
 229                                 dst_output);
 230         }
 231
 232         if (net_ratelimit())
 233                 printk(KERN_DEBUG "IPv6: sending pkt_too_big to self\n");
 234         skb->dev = dst->dev;
 235         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev);
 236         IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGFAILS);
 237         kfree_skb(skb);
 238         return -EMSGSIZE;
 239 }
 240
 241 /*
 242  *      To avoid extra problems ND packets are send through this
 243  *      routine. It's code duplication but I really want to avoid
 244  *      extra checks since ipv6_build_header is used by TCP (which
 245  *      is for us performance critical)
 246  */
 247
 248 int ip6_nd_hdr(struct sock *sk, struct sk_buff *skb, struct net_device *dev,
 249                struct in6_addr *saddr, struct in6_addr *daddr,
 250                int proto, int len)
 251 {
 252         struct ipv6_pinfo *np = inet6_sk(sk);
 253         struct ipv6hdr *hdr;
 254         int totlen;
 255
 256         skb->protocol = htons(ETH_P_IPV6);
 257         skb->dev = dev;
 258
 259         totlen = len + sizeof(struct ipv6hdr);
 260
 261         hdr = (struct ipv6hdr *) skb_put(skb, sizeof(struct ipv6hdr));
 262         skb->nh.ipv6h = hdr;
 263
 264         *(__be32*)hdr = htonl(0x60000000);
 265
 266         hdr->payload_len = htons(len);
 267         hdr->nexthdr = proto;
 268         hdr->hop_limit = np->hop_limit;
 269
 270         ipv6_addr_copy(&hdr->saddr, saddr);
 271         ipv6_addr_copy(&hdr->daddr, daddr);
 272
 273         return 0;
 274 }
 275
 276 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
 277 {
 278         struct ip6_ra_chain *ra;
 279         struct sock *last = NULL;
 280
 281         read_lock(&ip6_ra_lock);
 282         for (ra = ip6_ra_chain; ra; ra = ra->next) {
 283                 struct sock *sk = ra->sk;
 284                 if (sk && ra->sel == sel &&
 285                     (!sk->sk_bound_dev_if ||
 286                      sk->sk_bound_dev_if == skb->dev->ifindex)) {
 287                         if (last) {
 288                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
 289                                 if (skb2)
 290                                         rawv6_rcv(last, skb2);
 291                         }
 292                         last = sk;
 293                 }
 294         }
 295
 296         if (last) {
 297                 rawv6_rcv(last, skb);
 298                 read_unlock(&ip6_ra_lock);
 299                 return 1;
 300         }
 301         read_unlock(&ip6_ra_lock);
 302         return 0;
 303 }
 304
 305 static int ip6_forward_proxy_check(struct sk_buff *skb)
 306 {
 307         struct ipv6hdr *hdr = skb->nh.ipv6h;
 308         u8 nexthdr = hdr->nexthdr;
 309         int offset;
 310
 311         if (ipv6_ext_hdr(nexthdr)) {
 312                 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr);
 313                 if (offset < 0)
 314                         return 0;
 315         } else
 316                 offset = sizeof(struct ipv6hdr);
 317
 318         if (nexthdr == IPPROTO_ICMPV6) {
 319                 struct icmp6hdr *icmp6;
 320
 321                 if (!pskb_may_pull(skb, skb->nh.raw + offset + 1 - skb->data))
 322                         return 0;
 323
 324                 icmp6 = (struct icmp6hdr *)(skb->nh.raw + offset);
 325
 326                 switch (icmp6->icmp6_type) {
 327                 case NDISC_ROUTER_SOLICITATION:
 328                 case NDISC_ROUTER_ADVERTISEMENT:
 329                 case NDISC_NEIGHBOUR_SOLICITATION:
 330                 case NDISC_NEIGHBOUR_ADVERTISEMENT:
 331                 case NDISC_REDIRECT:
 332                         /* For reaction involving unicast neighbor discovery
 333                          * message destined to the proxied address, pass it to
 334                          * input function.
 335                          */
 336                         return 1;
 337                 default:
 338                         break;
 339                 }
 340         }
 341
 342         /*
 343          * The proxying router can't forward traffic sent to a link-local
 344          * address, so signal the sender and discard the packet. This
 345          * behavior is clarified by the MIPv6 specification.
 346          */
 347         if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
 348                 dst_link_failure(skb);
 349                 return -1;
 350         }
 351
 352         return 0;
 353 }
 354
 355 static inline int ip6_forward_finish(struct sk_buff *skb)
 356 {
 357         return dst_output(skb);
 358 }
 359
 360 int ip6_forward(struct sk_buff *skb)
 361 {
 362         struct dst_entry *dst = skb->dst;
 363         struct ipv6hdr *hdr = skb->nh.ipv6h;
 364         struct inet6_skb_parm *opt = IP6CB(skb);
 365
 366         if (ipv6_devconf.forwarding == 0)
 367                 goto error;
 368
 369         if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
 370                 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
 371                 goto drop;
 372         }
 373
 374         skb->ip_summed = CHECKSUM_NONE;
 375
 376         /*
 377          *      We DO NOT make any processing on
 378          *      RA packets, pushing them to user level AS IS
 379          *      without ane WARRANTY that application will be able
 380          *      to interpret them. The reason is that we
 381          *      cannot make anything clever here.
 382          *
 383          *      We are not end-node, so that if packet contains
 384          *      AH/ESP, we cannot make anything.
 385          *      Defragmentation also would be mistake, RA packets
 386          *      cannot be fragmented, because there is no warranty
 387          *      that different fragments will go along one path. --ANK
 388          */
 389         if (opt->ra) {
 390                 u8 *ptr = skb->nh.raw + opt->ra;
 391                 if (ip6_call_ra_chain(skb, (ptr[2]<<8) + ptr[3]))
 392                         return 0;
 393         }
 394
 395         /*
 396          *      check and decrement ttl
 397          */
 398         if (hdr->hop_limit <= 1) {
 399                 /* Force OUTPUT device used as source address */
 400                 skb->dev = dst->dev;
 401                 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT,
 402                             0, skb->dev);
 403                 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INHDRERRORS);
 404
 405                 kfree_skb(skb);
 406                 return -ETIMEDOUT;
 407         }
 408
 409         /* XXX: idev->cnf.proxy_ndp? */
 410         if (ipv6_devconf.proxy_ndp &&
 411             pneigh_lookup(&nd_tbl, &hdr->daddr, skb->dev, 0)) {
 412                 int proxied = ip6_forward_proxy_check(skb);
 413                 if (proxied > 0)
 414                         return ip6_input(skb);
 415                 else if (proxied < 0) {
 416                         IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
 417                         goto drop;
 418                 }
 419         }
 420
 421         if (!xfrm6_route_forward(skb)) {
 422                 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
 423                 goto drop;
 424         }
 425         dst = skb->dst;
 426
 427         /* IPv6 specs say nothing about it, but it is clear that we cannot
 428            send redirects to source routed frames.
 429          */
 430         if (skb->dev == dst->dev && dst->neighbour && opt->srcrt == 0) {
 431                 struct in6_addr *target = NULL;
 432                 struct rt6_info *rt;
 433                 struct neighbour *n = dst->neighbour;
 434
 435                 /*
 436                  *      incoming and outgoing devices are the same
 437                  *      send a redirect.
 438                  */
 439
 440                 rt = (struct rt6_info *) dst;
 441                 if ((rt->rt6i_flags & RTF_GATEWAY))
 442                         target = (struct in6_addr*)&n->primary_key;
 443                 else
 444                         target = &hdr->daddr;
 445
 446                 /* Limit redirects both by destination (here)
 447                    and by source (inside ndisc_send_redirect)
 448                  */
 449                 if (xrlim_allow(dst, 1*HZ))
 450                         ndisc_send_redirect(skb, n, target);
 451         } else if (ipv6_addr_type(&hdr->saddr)&(IPV6_ADDR_MULTICAST|IPV6_ADDR_LOOPBACK
 452                                                 |IPV6_ADDR_LINKLOCAL)) {
 453                 /* This check is security critical. */
 454                 goto error;
 455         }
 456
 457         if (skb->len > dst_mtu(dst)) {
 458                 /* Again, force OUTPUT device used as source address */
 459                 skb->dev = dst->dev;
 460                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, dst_mtu(dst), skb->dev);
 461                 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INTOOBIGERRORS);
 462                 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_FRAGFAILS);
 463                 kfree_skb(skb);
 464                 return -EMSGSIZE;
 465         }
 466
 467         if (skb_cow(skb, dst->dev->hard_header_len)) {
 468                 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_OUTDISCARDS);
 469                 goto drop;
 470         }
 471
 472         hdr = skb->nh.ipv6h;
 473
 474         /* Mangling hops number delayed to point after skb COW */
 475
 476         hdr->hop_limit--;
 477
 478         IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
 479         return NF_HOOK(PF_INET6,NF_IP6_FORWARD, skb, skb->dev, dst->dev, ip6_forward_finish);
 480
 481 error:
 482         IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
 483 drop:
 484         kfree_skb(skb);
 485         return -EINVAL;
 486 }
 487
 488 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
 489 {
 490         to->pkt_type = from->pkt_type;
 491         to->priority = from->priority;
 492         to->protocol = from->protocol;
 493         dst_release(to->dst);
 494         to->dst = dst_clone(from->dst);
 495         to->dev = from->dev;
 496         to->mark = from->mark;
 497
 498 #ifdef CONFIG_NET_SCHED
 499         to->tc_index = from->tc_index;
 500 #endif
 501 #ifdef CONFIG_NETFILTER
 502         /* Connection association is same as pre-frag packet */
 503         nf_conntrack_put(to->nfct);
 504         to->nfct = from->nfct;
 505         nf_conntrack_get(to->nfct);
 506         to->nfctinfo = from->nfctinfo;
 507 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
 508         nf_conntrack_put_reasm(to->nfct_reasm);
 509         to->nfct_reasm = from->nfct_reasm;
 510         nf_conntrack_get_reasm(to->nfct_reasm);
 511 #endif
 512 #ifdef CONFIG_BRIDGE_NETFILTER
 513         nf_bridge_put(to->nf_bridge);
 514         to->nf_bridge = from->nf_bridge;
 515         nf_bridge_get(to->nf_bridge);
 516 #endif
 517 #endif
 518         skb_copy_secmark(to, from);
 519 }
 520
 521 int ip6_find_1stfragopt(struct sk_buff *skb, u8 **nexthdr)
 522 {
 523         u16 offset = sizeof(struct ipv6hdr);
 524         struct ipv6_opt_hdr *exthdr = (struct ipv6_opt_hdr*)(skb->nh.ipv6h + 1);
 525         unsigned int packet_len = skb->tail - skb->nh.raw;
 526         int found_rhdr = 0;
 527         *nexthdr = &skb->nh.ipv6h->nexthdr;
 528
 529         while (offset + 1 <= packet_len) {
 530
 531                 switch (**nexthdr) {
 532
 533                 case NEXTHDR_HOP:
 534                         break;
 535                 case NEXTHDR_ROUTING:
 536                         found_rhdr = 1;
 537                         break;
 538                 case NEXTHDR_DEST:
 539 #ifdef CONFIG_IPV6_MIP6
 540                         if (ipv6_find_tlv(skb, offset, IPV6_TLV_HAO) >= 0)
 541                                 break;
 542 #endif
 543                         if (found_rhdr)
 544                                 return offset;
 545                         break;
 546                 default :
 547                         return offset;
 548                 }
 549
 550                 offset += ipv6_optlen(exthdr);
 551                 *nexthdr = &exthdr->nexthdr;
 552                 exthdr = (struct ipv6_opt_hdr*)(skb->nh.raw + offset);
 553         }
 554
 555         return offset;
 556 }
 557 EXPORT_SYMBOL_GPL(ip6_find_1stfragopt);
 558
 559 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
 560 {
 561         struct net_device *dev;
 562         struct sk_buff *frag;
 563         struct rt6_info *rt = (struct rt6_info*)skb->dst;
 564         struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
 565         struct ipv6hdr *tmp_hdr;
 566         struct frag_hdr *fh;
 567         unsigned int mtu, hlen, left, len;
 568         __be32 frag_id = 0;
 569         int ptr, offset = 0, err=0;
 570         u8 *prevhdr, nexthdr = 0;
 571
 572         dev = rt->u.dst.dev;
 573         hlen = ip6_find_1stfragopt(skb, &prevhdr);
 574         nexthdr = *prevhdr;
 575
 576         mtu = dst_mtu(&rt->u.dst);
 577         if (np && np->frag_size < mtu) {
 578                 if (np->frag_size)
 579                         mtu = np->frag_size;
 580         }
 581         mtu -= hlen + sizeof(struct frag_hdr);
 582
 583         if (skb_shinfo(skb)->frag_list) {
 584                 int first_len = skb_pagelen(skb);
 585
 586                 if (first_len - hlen > mtu ||
 587                     ((first_len - hlen) & 7) ||
 588                     skb_cloned(skb))
 589                         goto slow_path;
 590
 591                 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
 592                         /* Correct geometry. */
 593                         if (frag->len > mtu ||
 594                             ((frag->len & 7) && frag->next) ||
 595                             skb_headroom(frag) < hlen)
 596                             goto slow_path;
 597
 598                         /* Partially cloned skb? */
 599                         if (skb_shared(frag))
 600                                 goto slow_path;
 601
 602                         BUG_ON(frag->sk);
 603                         if (skb->sk) {
 604                                 sock_hold(skb->sk);
 605                                 frag->sk = skb->sk;
 606                                 frag->destructor = sock_wfree;
 607                                 skb->truesize -= frag->truesize;
 608                         }
 609                 }
 610
 611                 err = 0;
 612                 offset = 0;
 613                 frag = skb_shinfo(skb)->frag_list;
 614                 skb_shinfo(skb)->frag_list = NULL;
 615                 /* BUILD HEADER */
 616
 617                 *prevhdr = NEXTHDR_FRAGMENT;
 618                 tmp_hdr = kmemdup(skb->nh.raw, hlen, GFP_ATOMIC);
 619                 if (!tmp_hdr) {
 620                         IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGFAILS);
 621                         return -ENOMEM;
 622                 }
 623
 624                 __skb_pull(skb, hlen);
 625                 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
 626                 skb->nh.raw = __skb_push(skb, hlen);
 627                 memcpy(skb->nh.raw, tmp_hdr, hlen);
 628
 629                 ipv6_select_ident(skb, fh);
 630                 fh->nexthdr = nexthdr;
 631                 fh->reserved = 0;
 632                 fh->frag_off = htons(IP6_MF);
 633                 frag_id = fh->identification;
 634
 635                 first_len = skb_pagelen(skb);
 636                 skb->data_len = first_len - skb_headlen(skb);
 637                 skb->len = first_len;
 638                 skb->nh.ipv6h->payload_len = htons(first_len - sizeof(struct ipv6hdr));
 639
 640                 dst_hold(&rt->u.dst);
 641
 642                 for (;;) {
 643                         /* Prepare header of the next frame,
 644                          * before previous one went down. */
 645                         if (frag) {
 646                                 frag->ip_summed = CHECKSUM_NONE;
 647                                 frag->h.raw = frag->data;
 648                                 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
 649                                 frag->nh.raw = __skb_push(frag, hlen);
 650                                 memcpy(frag->nh.raw, tmp_hdr, hlen);
 651                                 offset += skb->len - hlen - sizeof(struct frag_hdr);
 652                                 fh->nexthdr = nexthdr;
 653                                 fh->reserved = 0;
 654                                 fh->frag_off = htons(offset);
 655                                 if (frag->next != NULL)
 656                                         fh->frag_off |= htons(IP6_MF);
 657                                 fh->identification = frag_id;
 658                                 frag->nh.ipv6h->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
 659                                 ip6_copy_metadata(frag, skb);
 660                         }
 661
 662                         err = output(skb);
 663                         if(!err)
 664                                 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGCREATES);
 665
 666                         if (err || !frag)
 667                                 break;
 668
 669                         skb = frag;
 670                         frag = skb->next;
 671                         skb->next = NULL;
 672                 }
 673
 674                 kfree(tmp_hdr);
 675
 676                 if (err == 0) {
 677                         IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGOKS);
 678                         dst_release(&rt->u.dst);
 679                         return 0;
 680                 }
 681
 682                 while (frag) {
 683                         skb = frag->next;
 684                         kfree_skb(frag);
 685                         frag = skb;
 686                 }
 687
 688                 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGFAILS);
 689                 dst_release(&rt->u.dst);
 690                 return err;
 691         }
 692
 693 slow_path:
 694         left = skb->len - hlen;         /* Space per frame */
 695         ptr = hlen;                     /* Where to start from */
 696
 697         /*
 698          *      Fragment the datagram.
 699          */
 700
 701         *prevhdr = NEXTHDR_FRAGMENT;
 702
 703         /*
 704          *      Keep copying data until we run out.
 705          */
 706         while(left > 0) {
 707                 len = left;
 708                 /* IF: it doesn't fit, use 'mtu' - the data space left */
 709                 if (len > mtu)
 710                         len = mtu;
 711                 /* IF: we are not sending upto and including the packet end
 712                    then align the next start on an eight byte boundary */
 713                 if (len < left) {
 714                         len &= ~7;
 715                 }
 716                 /*
 717                  *      Allocate buffer.
 718                  */
 719
 720                 if ((frag = alloc_skb(len+hlen+sizeof(struct frag_hdr)+LL_RESERVED_SPACE(rt->u.dst.dev), GFP_ATOMIC)) == NULL) {
 721                         NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
 722                         IP6_INC_STATS(ip6_dst_idev(skb->dst),
 723                                       IPSTATS_MIB_FRAGFAILS);
 724                         err = -ENOMEM;
 725                         goto fail;
 726                 }
 727
 728                 /*
 729                  *      Set up data on packet
 730                  */
 731
 732                 ip6_copy_metadata(frag, skb);
 733                 skb_reserve(frag, LL_RESERVED_SPACE(rt->u.dst.dev));
 734                 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
 735                 frag->nh.raw = frag->data;
 736                 fh = (struct frag_hdr*)(frag->data + hlen);
 737                 frag->h.raw = frag->data + hlen + sizeof(struct frag_hdr);
 738
 739                 /*
 740                  *      Charge the memory for the fragment to any owner
 741                  *      it might possess
 742                  */
 743                 if (skb->sk)
 744                         skb_set_owner_w(frag, skb->sk);
 745
 746                 /*
 747                  *      Copy the packet header into the new buffer.
 748                  */
 749                 memcpy(frag->nh.raw, skb->data, hlen);
 750
 751                 /*
 752                  *      Build fragment header.
 753                  */
 754                 fh->nexthdr = nexthdr;
 755                 fh->reserved = 0;
 756                 if (!frag_id) {
 757                         ipv6_select_ident(skb, fh);
 758                         frag_id = fh->identification;
 759                 } else
 760                         fh->identification = frag_id;
 761
 762                 /*
 763                  *      Copy a block of the IP datagram.
 764                  */
 765                 if (skb_copy_bits(skb, ptr, frag->h.raw, len))
 766                         BUG();
 767                 left -= len;
 768
 769                 fh->frag_off = htons(offset);
 770                 if (left > 0)
 771                         fh->frag_off |= htons(IP6_MF);
 772                 frag->nh.ipv6h->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
 773
 774                 ptr += len;
 775                 offset += len;
 776
 777                 /*
 778                  *      Put this fragment into the sending queue.
 779                  */
 780                 err = output(frag);
 781                 if (err)
 782                         goto fail;
 783
 784                 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGCREATES);
 785         }
 786         IP6_INC_STATS(ip6_dst_idev(skb->dst),
 787                       IPSTATS_MIB_FRAGOKS);
 788         kfree_skb(skb);
 789         return err;
 790
 791 fail:
 792         IP6_INC_STATS(ip6_dst_idev(skb->dst),
 793                       IPSTATS_MIB_FRAGFAILS);
 794         kfree_skb(skb);
 795         return err;
 796 }
 797
 798 static inline int ip6_rt_check(struct rt6key *rt_key,
 799                                struct in6_addr *fl_addr,
 800                                struct in6_addr *addr_cache)
 801 {
 802         return ((rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
 803                 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache)));
 804 }
 805
 806 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
 807                                           struct dst_entry *dst,
 808                                           struct flowi *fl)
 809 {
 810         struct ipv6_pinfo *np = inet6_sk(sk);
 811         struct rt6_info *rt = (struct rt6_info *)dst;
 812
 813         if (!dst)
 814                 goto out;
 815
 816         /* Yes, checking route validity in not connected
 817          * case is not very simple. Take into account,
 818          * that we do not support routing by source, TOS,
 819          * and MSG_DONTROUTE            --ANK (980726)
 820          *
 821          * 1. ip6_rt_check(): If route was host route,
 822          *    check that cached destination is current.
 823          *    If it is network route, we still may
 824          *    check its validity using saved pointer
 825          *    to the last used address: daddr_cache.
 826          *    We do not want to save whole address now,
 827          *    (because main consumer of this service
 828          *    is tcp, which has not this problem),
 829          *    so that the last trick works only on connected
 830          *    sockets.
 831          * 2. oif also should be the same.
 832          */
 833         if (ip6_rt_check(&rt->rt6i_dst, &fl->fl6_dst, np->daddr_cache) ||
 834 #ifdef CONFIG_IPV6_SUBTREES
 835             ip6_rt_check(&rt->rt6i_src, &fl->fl6_src, np->saddr_cache) ||
 836 #endif
 837             (fl->oif && fl->oif != dst->dev->ifindex)) {
 838                 dst_release(dst);
 839                 dst = NULL;
 840         }
 841
 842 out:
 843         return dst;
 844 }
 845
 846 static int ip6_dst_lookup_tail(struct sock *sk,
 847                                struct dst_entry **dst, struct flowi *fl)
 848 {
 849         int err;
 850
 851         if (*dst == NULL)
 852                 *dst = ip6_route_output(sk, fl);
 853
 854         if ((err = (*dst)->error))
 855                 goto out_err_release;
 856
 857         if (ipv6_addr_any(&fl->fl6_src)) {
 858                 err = ipv6_get_saddr(*dst, &fl->fl6_dst, &fl->fl6_src, sk->sk_nx_info);
 859
 860                 if (err)
 861                         goto out_err_release;
 862         }
 863
 864         return 0;
 865
 866 out_err_release:
 867         dst_release(*dst);
 868         *dst = NULL;
 869         return err;
 870 }
 871
 872 /**
 873  *      ip6_dst_lookup - perform route lookup on flow
 874  *      @sk: socket which provides route info
 875  *      @dst: pointer to dst_entry * for result
 876  *      @fl: flow to lookup
 877  *
 878  *      This function performs a route lookup on the given flow.
 879  *
 880  *      It returns zero on success, or a standard errno code on error.
 881  */
 882 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
 883 {
 884         *dst = NULL;
 885         return ip6_dst_lookup_tail(sk, dst, fl);
 886 }
 887 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
 888
 889 /**
 890  *      ip6_sk_dst_lookup - perform socket cached route lookup on flow
 891  *      @sk: socket which provides the dst cache and route info
 892  *      @dst: pointer to dst_entry * for result
 893  *      @fl: flow to lookup
 894  *
 895  *      This function performs a route lookup on the given flow with the
 896  *      possibility of using the cached route in the socket if it is valid.
 897  *      It will take the socket dst lock when operating on the dst cache.
 898  *      As a result, this function can only be used in process context.
 899  *
 900  *      It returns zero on success, or a standard errno code on error.
 901  */
 902 int ip6_sk_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
 903 {
 904         *dst = NULL;
 905         if (sk) {
 906                 *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
 907                 *dst = ip6_sk_dst_check(sk, *dst, fl);
 908         }
 909
 910         return ip6_dst_lookup_tail(sk, dst, fl);
 911 }
 912 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup);
 913
 914 static inline int ip6_ufo_append_data(struct sock *sk,
 915                         int getfrag(void *from, char *to, int offset, int len,
 916                         int odd, struct sk_buff *skb),
 917                         void *from, int length, int hh_len, int fragheaderlen,
 918                         int transhdrlen, int mtu,unsigned int flags)
 919
 920 {
 921         struct sk_buff *skb;
 922         int err;
 923
 924         /* There is support for UDP large send offload by network
 925          * device, so create one single skb packet containing complete
 926          * udp datagram
 927          */
 928         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
 929                 skb = sock_alloc_send_skb(sk,
 930                         hh_len + fragheaderlen + transhdrlen + 20,
 931                         (flags & MSG_DONTWAIT), &err);
 932                 if (skb == NULL)
 933                         return -ENOMEM;
 934
 935                 /* reserve space for Hardware header */
 936                 skb_reserve(skb, hh_len);
 937
 938                 /* create space for UDP/IP header */
 939                 skb_put(skb,fragheaderlen + transhdrlen);
 940
 941                 /* initialize network header pointer */
 942                 skb->nh.raw = skb->data;
 943
 944                 /* initialize protocol header pointer */
 945                 skb->h.raw = skb->data + fragheaderlen;
 946
 947                 skb->ip_summed = CHECKSUM_PARTIAL;
 948                 skb->csum = 0;
 949                 sk->sk_sndmsg_off = 0;
 950         }
 951
 952         err = skb_append_datato_frags(sk,skb, getfrag, from,
 953                                       (length - transhdrlen));
 954         if (!err) {
 955                 struct frag_hdr fhdr;
 956
 957                 /* specify the length of each IP datagram fragment*/
 958                 skb_shinfo(skb)->gso_size = mtu - fragheaderlen -
 959                                             sizeof(struct frag_hdr);
 960                 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
 961                 ipv6_select_ident(skb, &fhdr);
 962                 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
 963                 __skb_queue_tail(&sk->sk_write_queue, skb);
 964
 965                 return 0;
 966         }
 967         /* There is not enough support do UPD LSO,
 968          * so follow normal path
 969          */
 970         kfree_skb(skb);
 971
 972         return err;
 973 }
 974
 975 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
 976         int offset, int len, int odd, struct sk_buff *skb),
 977         void *from, int length, int transhdrlen,
 978         int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi *fl,
 979         struct rt6_info *rt, unsigned int flags)
 980 {
 981         struct inet_sock *inet = inet_sk(sk);
 982         struct ipv6_pinfo *np = inet6_sk(sk);
 983         struct sk_buff *skb;
 984         unsigned int maxfraglen, fragheaderlen;
 985         int exthdrlen;
 986         int hh_len;
 987         int mtu;
 988         int copy;
 989         int err;
 990         int offset = 0;
 991         int csummode = CHECKSUM_NONE;
 992
 993         if (flags&MSG_PROBE)
 994                 return 0;
 995         if (skb_queue_empty(&sk->sk_write_queue)) {
 996                 /*
 997                  * setup for corking
 998                  */
 999                 if (opt) {
1000                         if (np->cork.opt == NULL) {
1001                                 np->cork.opt = kmalloc(opt->tot_len,
1002                                                        sk->sk_allocation);
1003                                 if (unlikely(np->cork.opt == NULL))
1004                                         return -ENOBUFS;
1005                         } else if (np->cork.opt->tot_len < opt->tot_len) {
1006                                 printk(KERN_DEBUG "ip6_append_data: invalid option length\n");
1007                                 return -EINVAL;
1008                         }
1009                         memcpy(np->cork.opt, opt, opt->tot_len);
1010                         inet->cork.flags |= IPCORK_OPT;
1011                         /* need source address above miyazawa*/
1012                 }
1013                 dst_hold(&rt->u.dst);
1014                 np->cork.rt = rt;
1015                 inet->cork.fl = *fl;
1016                 np->cork.hop_limit = hlimit;
1017                 np->cork.tclass = tclass;
1018                 mtu = dst_mtu(rt->u.dst.path);
1019                 if (np->frag_size < mtu) {
1020                         if (np->frag_size)
1021                                 mtu = np->frag_size;
1022                 }
1023                 inet->cork.fragsize = mtu;
1024                 if (dst_allfrag(rt->u.dst.path))
1025                         inet->cork.flags |= IPCORK_ALLFRAG;
1026                 inet->cork.length = 0;
1027                 sk->sk_sndmsg_page = NULL;
1028                 sk->sk_sndmsg_off = 0;
1029                 exthdrlen = rt->u.dst.header_len + (opt ? opt->opt_flen : 0);
1030                 length += exthdrlen;
1031                 transhdrlen += exthdrlen;
1032         } else {
1033                 rt = np->cork.rt;
1034                 fl = &inet->cork.fl;
1035                 if (inet->cork.flags & IPCORK_OPT)
1036                         opt = np->cork.opt;
1037                 transhdrlen = 0;
1038                 exthdrlen = 0;
1039                 mtu = inet->cork.fragsize;
1040         }
1041
1042         hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1043
1044         fragheaderlen = sizeof(struct ipv6hdr) + rt->u.dst.nfheader_len + (opt ? opt->opt_nflen : 0);
1045         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
1046
1047         if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1048                 if (inet->cork.length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
1049                         ipv6_local_error(sk, EMSGSIZE, fl, mtu-exthdrlen);
1050                         return -EMSGSIZE;
1051                 }
1052         }
1053
1054         /*
1055          * Let's try using as much space as possible.
1056          * Use MTU if total length of the message fits into the MTU.
1057          * Otherwise, we need to reserve fragment header and
1058          * fragment alignment (= 8-15 octects, in total).
1059          *
1060          * Note that we may need to "move" the data from the tail of
1061          * of the buffer to the new fragment when we split
1062          * the message.
1063          *
1064          * FIXME: It may be fragmented into multiple chunks
1065          *        at once if non-fragmentable extension headers
1066          *        are too large.
1067          * --yoshfuji
1068          */
1069
1070         inet->cork.length += length;
1071         if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
1072             (rt->u.dst.dev->features & NETIF_F_UFO)) {
1073
1074                 err = ip6_ufo_append_data(sk, getfrag, from, length, hh_len,
1075                                           fragheaderlen, transhdrlen, mtu,
1076                                           flags);
1077                 if (err)
1078                         goto error;
1079                 return 0;
1080         }
1081
1082         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1083                 goto alloc_new_skb;
1084
1085         while (length > 0) {
1086                 /* Check if the remaining data fits into current packet. */
1087                 copy = (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1088                 if (copy < length)
1089                         copy = maxfraglen - skb->len;
1090
1091                 if (copy <= 0) {
1092                         char *data;
1093                         unsigned int datalen;
1094                         unsigned int fraglen;
1095                         unsigned int fraggap;
1096                         unsigned int alloclen;
1097                         struct sk_buff *skb_prev;
1098 alloc_new_skb:
1099                         skb_prev = skb;
1100
1101                         /* There's no room in the current skb */
1102                         if (skb_prev)
1103                                 fraggap = skb_prev->len - maxfraglen;
1104                         else
1105                                 fraggap = 0;
1106
1107                         /*
1108                          * If remaining data exceeds the mtu,
1109                          * we know we need more fragment(s).
1110                          */
1111                         datalen = length + fraggap;
1112                         if (datalen > (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1113                                 datalen = maxfraglen - fragheaderlen;
1114
1115                         fraglen = datalen + fragheaderlen;
1116                         if ((flags & MSG_MORE) &&
1117                             !(rt->u.dst.dev->features&NETIF_F_SG))
1118                                 alloclen = mtu;
1119                         else
1120                                 alloclen = datalen + fragheaderlen;
1121
1122                         /*
1123                          * The last fragment gets additional space at tail.
1124                          * Note: we overallocate on fragments with MSG_MODE
1125                          * because we have no idea if we're the last one.
1126                          */
1127                         if (datalen == length + fraggap)
1128                                 alloclen += rt->u.dst.trailer_len;
1129
1130                         /*
1131                          * We just reserve space for fragment header.
1132                          * Note: this may be overallocation if the message
1133                          * (without MSG_MORE) fits into the MTU.
1134                          */
1135                         alloclen += sizeof(struct frag_hdr);
1136
1137                         if (transhdrlen) {
1138                                 skb = sock_alloc_send_skb(sk,
1139                                                 alloclen + hh_len,
1140                                                 (flags & MSG_DONTWAIT), &err);
1141                         } else {
1142                                 skb = NULL;
1143                                 if (atomic_read(&sk->sk_wmem_alloc) <=
1144                                     2 * sk->sk_sndbuf)
1145                                         skb = sock_wmalloc(sk,
1146                                                            alloclen + hh_len, 1,
1147                                                            sk->sk_allocation);
1148                                 if (unlikely(skb == NULL))
1149                                         err = -ENOBUFS;
1150                         }
1151                         if (skb == NULL)
1152                                 goto error;
1153                         /*
1154                          *      Fill in the control structures
1155                          */
1156                         skb->ip_summed = csummode;
1157                         skb->csum = 0;
1158                         /* reserve for fragmentation */
1159                         skb_reserve(skb, hh_len+sizeof(struct frag_hdr));
1160
1161                         /*
1162                          *      Find where to start putting bytes
1163                          */
1164                         data = skb_put(skb, fraglen);
1165                         skb->nh.raw = data + exthdrlen;
1166                         data += fragheaderlen;
1167                         skb->h.raw = data + exthdrlen;
1168
1169                         if (fraggap) {
1170                                 skb->csum = skb_copy_and_csum_bits(
1171                                         skb_prev, maxfraglen,
1172                                         data + transhdrlen, fraggap, 0);
1173                                 skb_prev->csum = csum_sub(skb_prev->csum,
1174                                                           skb->csum);
1175                                 data += fraggap;
1176                                 pskb_trim_unique(skb_prev, maxfraglen);
1177                         }
1178                         copy = datalen - transhdrlen - fraggap;
1179                         if (copy < 0) {
1180                                 err = -EINVAL;
1181                                 kfree_skb(skb);
1182                                 goto error;
1183                         } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1184                                 err = -EFAULT;
1185                                 kfree_skb(skb);
1186                                 goto error;
1187                         }
1188
1189                         offset += copy;
1190                         length -= datalen - fraggap;
1191                         transhdrlen = 0;
1192                         exthdrlen = 0;
1193                         csummode = CHECKSUM_NONE;
1194
1195                         /*
1196                          * Put the packet on the pending queue
1197                          */
1198                         __skb_queue_tail(&sk->sk_write_queue, skb);
1199                         continue;
1200                 }
1201
1202                 if (copy > length)
1203                         copy = length;
1204
1205                 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
1206                         unsigned int off;
1207
1208                         off = skb->len;
1209                         if (getfrag(from, skb_put(skb, copy),
1210                                                 offset, copy, off, skb) < 0) {
1211                                 __skb_trim(skb, off);
1212                                 err = -EFAULT;
1213                                 goto error;
1214                         }
1215                 } else {
1216                         int i = skb_shinfo(skb)->nr_frags;
1217                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1218                         struct page *page = sk->sk_sndmsg_page;
1219                         int off = sk->sk_sndmsg_off;
1220                         unsigned int left;
1221
1222                         if (page && (left = PAGE_SIZE - off) > 0) {
1223                                 if (copy >= left)
1224                                         copy = left;
1225                                 if (page != frag->page) {
1226                                         if (i == MAX_SKB_FRAGS) {
1227                                                 err = -EMSGSIZE;
1228                                                 goto error;
1229                                         }
1230                                         get_page(page);
1231                                         skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1232                                         frag = &skb_shinfo(skb)->frags[i];
1233                                 }
1234                         } else if(i < MAX_SKB_FRAGS) {
1235                                 if (copy > PAGE_SIZE)
1236                                         copy = PAGE_SIZE;
1237                                 page = alloc_pages(sk->sk_allocation, 0);
1238                                 if (page == NULL) {
1239                                         err = -ENOMEM;
1240                                         goto error;
1241                                 }
1242                                 sk->sk_sndmsg_page = page;
1243                                 sk->sk_sndmsg_off = 0;
1244
1245                                 skb_fill_page_desc(skb, i, page, 0, 0);
1246                                 frag = &skb_shinfo(skb)->frags[i];
1247                                 skb->truesize += PAGE_SIZE;
1248                                 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1249                         } else {
1250                                 err = -EMSGSIZE;
1251                                 goto error;
1252                         }
1253                         if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1254                                 err = -EFAULT;
1255                                 goto error;
1256                         }
1257                         sk->sk_sndmsg_off += copy;
1258                         frag->size += copy;
1259                         skb->len += copy;
1260                         skb->data_len += copy;
1261                 }
1262                 offset += copy;
1263                 length -= copy;
1264         }
1265         return 0;
1266 error:
1267         inet->cork.length -= length;
1268         IP6_INC_STATS(rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1269         return err;
1270 }
1271
1272 int ip6_push_pending_frames(struct sock *sk)
1273 {
1274         struct sk_buff *skb, *tmp_skb;
1275         struct sk_buff **tail_skb;
1276         struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1277         struct inet_sock *inet = inet_sk(sk);
1278         struct ipv6_pinfo *np = inet6_sk(sk);
1279         struct ipv6hdr *hdr;
1280         struct ipv6_txoptions *opt = np->cork.opt;
1281         struct rt6_info *rt = np->cork.rt;
1282         struct flowi *fl = &inet->cork.fl;
1283         unsigned char proto = fl->proto;
1284         int err = 0;
1285
1286         if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1287                 goto out;
1288         tail_skb = &(skb_shinfo(skb)->frag_list);
1289
1290         /* move skb->data to ip header from ext header */
1291         if (skb->data < skb->nh.raw)
1292                 __skb_pull(skb, skb->nh.raw - skb->data);
1293         while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1294                 __skb_pull(tmp_skb, skb->h.raw - skb->nh.raw);
1295                 *tail_skb = tmp_skb;
1296                 tail_skb = &(tmp_skb->next);
1297                 skb->len += tmp_skb->len;
1298                 skb->data_len += tmp_skb->len;
1299                 skb->truesize += tmp_skb->truesize;
1300                 __sock_put(tmp_skb->sk);
1301                 tmp_skb->destructor = NULL;
1302                 tmp_skb->sk = NULL;
1303         }
1304
1305         ipv6_addr_copy(final_dst, &fl->fl6_dst);
1306         __skb_pull(skb, skb->h.raw - skb->nh.raw);
1307         if (opt && opt->opt_flen)
1308                 ipv6_push_frag_opts(skb, opt, &proto);
1309         if (opt && opt->opt_nflen)
1310                 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1311
1312         skb->nh.ipv6h = hdr = (struct ipv6hdr*) skb_push(skb, sizeof(struct ipv6hdr));
1313
1314         *(__be32*)hdr = fl->fl6_flowlabel |
1315                      htonl(0x60000000 | ((int)np->cork.tclass << 20));
1316
1317         if (skb->len <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN)
1318                 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
1319         else
1320                 hdr->payload_len = 0;
1321         hdr->hop_limit = np->cork.hop_limit;
1322         hdr->nexthdr = proto;
1323         ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
1324         ipv6_addr_copy(&hdr->daddr, final_dst);
1325
1326         skb->priority = sk->sk_priority;
1327
1328         skb->dst = dst_clone(&rt->u.dst);
1329         IP6_INC_STATS(rt->rt6i_idev, IPSTATS_MIB_OUTREQUESTS);
1330         err = NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, skb->dst->dev, dst_output);
1331         if (err) {
1332                 if (err > 0)
1333                         err = np->recverr ? net_xmit_errno(err) : 0;
1334                 if (err)
1335                         goto error;
1336         }
1337
1338 out:
1339         inet->cork.flags &= ~IPCORK_OPT;
1340         kfree(np->cork.opt);
1341         np->cork.opt = NULL;
1342         if (np->cork.rt) {
1343                 dst_release(&np->cork.rt->u.dst);
1344                 np->cork.rt = NULL;
1345                 inet->cork.flags &= ~IPCORK_ALLFRAG;
1346         }
1347         memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1348         return err;
1349 error:
1350         goto out;
1351 }
1352
1353 void ip6_flush_pending_frames(struct sock *sk)
1354 {
1355         struct inet_sock *inet = inet_sk(sk);
1356         struct ipv6_pinfo *np = inet6_sk(sk);
1357         struct sk_buff *skb;
1358
1359         while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1360                 IP6_INC_STATS(ip6_dst_idev(skb->dst),
1361                               IPSTATS_MIB_OUTDISCARDS);
1362                 kfree_skb(skb);
1363         }
1364
1365         inet->cork.flags &= ~IPCORK_OPT;
1366
1367         kfree(np->cork.opt);
1368         np->cork.opt = NULL;
1369         if (np->cork.rt) {
1370                 dst_release(&np->cork.rt->u.dst);
1371                 np->cork.rt = NULL;
1372                 inet->cork.flags &= ~IPCORK_ALLFRAG;
1373         }
1374         memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1375 }