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  *      Based on linux/net/ipv4/ip_output.c
   9  *
  10  *      This program is free software; you can redistribute it and/or
  11  *      modify it under the terms of the GNU General Public License
  12  *      as published by the Free Software Foundation; either version
  13  *      2 of the License, or (at your option) any later version.
  14  *
  15  *      Changes:
  16  *      A.N.Kuznetsov   :       airthmetics in fragmentation.
  17  *                              extension headers are implemented.
  18  *                              route changes now work.
  19  *                              ip6_forward does not confuse sniffers.
  20  *                              etc.
  21  *
  22  *      H. von Brand    :       Added missing #include <linux/string.h>
  23  *      Imran Patel     :       frag id should be in NBO
  24  *      Kazunori MIYAZAWA @USAGI
  25  *                      :       add ip6_append_data and related functions
  26  *                              for datagram xmit
  27  */
  28
  29 #include <linux/errno.h>
  30 #include <linux/kernel.h>
  31 #include <linux/string.h>
  32 #include <linux/socket.h>
  33 #include <linux/net.h>
  34 #include <linux/netdevice.h>
  35 #include <linux/if_arp.h>
  36 #include <linux/in6.h>
  37 #include <linux/tcp.h>
  38 #include <linux/route.h>
  39 #include <linux/module.h>
  40 #include <linux/slab.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 #include <linux/mroute6.h>
  58
  59 static int ip6_finish_output2(struct sock *sk, struct sk_buff *skb)
  60 {
  61         struct dst_entry *dst = skb_dst(skb);
  62         struct net_device *dev = dst->dev;
  63         struct neighbour *neigh;
  64         struct in6_addr *nexthop;
  65         int ret;
  66
  67         skb->protocol = htons(ETH_P_IPV6);
  68         skb->dev = dev;
  69
  70         if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
  71                 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
  72
  73                 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(sk) &&
  74                     ((mroute6_socket(dev_net(dev), skb) &&
  75                      !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
  76                      ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
  77                                          &ipv6_hdr(skb)->saddr))) {
  78                         struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
  79
  80                         /* Do not check for IFF_ALLMULTI; multicast routing
  81                            is not supported in any case.
  82                          */
  83                         if (newskb)
  84                                 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
  85                                         sk, newskb, NULL, newskb->dev,
  86                                         dev_loopback_xmit);
  87
  88                         if (ipv6_hdr(skb)->hop_limit == 0) {
  89                                 IP6_INC_STATS(dev_net(dev), idev,
  90                                               IPSTATS_MIB_OUTDISCARDS);
  91                                 kfree_skb(skb);
  92                                 return 0;
  93                         }
  94                 }
  95
  96                 IP6_UPD_PO_STATS(dev_net(dev), idev, IPSTATS_MIB_OUTMCAST,
  97                                 skb->len);
  98
  99                 if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb)->daddr) <=
 100                     IPV6_ADDR_SCOPE_NODELOCAL &&
 101                     !(dev->flags & IFF_LOOPBACK)) {
 102                         kfree_skb(skb);
 103                         return 0;
 104                 }
 105         }
 106
 107         rcu_read_lock_bh();
 108         nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
 109         neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
 110         if (unlikely(!neigh))
 111                 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
 112         if (!IS_ERR(neigh)) {
 113                 ret = dst_neigh_output(dst, neigh, skb);
 114                 rcu_read_unlock_bh();
 115                 return ret;
 116         }
 117         rcu_read_unlock_bh();
 118
 119         IP6_INC_STATS(dev_net(dst->dev),
 120                       ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
 121         kfree_skb(skb);
 122         return -EINVAL;
 123 }
 124
 125 static int ip6_finish_output(struct sock *sk, struct sk_buff *skb)
 126 {
 127         if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
 128             dst_allfrag(skb_dst(skb)) ||
 129             (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
 130                 return ip6_fragment(sk, skb, ip6_finish_output2);
 131         else
 132                 return ip6_finish_output2(sk, skb);
 133 }
 134
 135 int ip6_output(struct sock *sk, struct sk_buff *skb)
 136 {
 137         struct net_device *dev = skb_dst(skb)->dev;
 138         struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
 139         if (unlikely(idev->cnf.disable_ipv6)) {
 140                 IP6_INC_STATS(dev_net(dev), idev,
 141                               IPSTATS_MIB_OUTDISCARDS);
 142                 kfree_skb(skb);
 143                 return 0;
 144         }
 145
 146         return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, sk, skb,
 147                             NULL, dev,
 148                             ip6_finish_output,
 149                             !(IP6CB(skb)->flags & IP6SKB_REROUTED));
 150 }
 151
 152 /*
 153  *      xmit an sk_buff (used by TCP, SCTP and DCCP)
 154  */
 155
 156 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
 157              struct ipv6_txoptions *opt, int tclass)
 158 {
 159         struct net *net = sock_net(sk);
 160         struct ipv6_pinfo *np = inet6_sk(sk);
 161         struct in6_addr *first_hop = &fl6->daddr;
 162         struct dst_entry *dst = skb_dst(skb);
 163         struct ipv6hdr *hdr;
 164         u8  proto = fl6->flowi6_proto;
 165         int seg_len = skb->len;
 166         int hlimit = -1;
 167         u32 mtu;
 168
 169         if (opt) {
 170                 unsigned int head_room;
 171
 172                 /* First: exthdrs may take lots of space (~8K for now)
 173                    MAX_HEADER is not enough.
 174                  */
 175                 head_room = opt->opt_nflen + opt->opt_flen;
 176                 seg_len += head_room;
 177                 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
 178
 179                 if (skb_headroom(skb) < head_room) {
 180                         struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
 181                         if (!skb2) {
 182                                 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 183                                               IPSTATS_MIB_OUTDISCARDS);
 184                                 kfree_skb(skb);
 185                                 return -ENOBUFS;
 186                         }
 187                         consume_skb(skb);
 188                         skb = skb2;
 189                         skb_set_owner_w(skb, sk);
 190                 }
 191                 if (opt->opt_flen)
 192                         ipv6_push_frag_opts(skb, opt, &proto);
 193                 if (opt->opt_nflen)
 194                         ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
 195         }
 196
 197         skb_push(skb, sizeof(struct ipv6hdr));
 198         skb_reset_network_header(skb);
 199         hdr = ipv6_hdr(skb);
 200
 201         /*
 202          *      Fill in the IPv6 header
 203          */
 204         if (np)
 205                 hlimit = np->hop_limit;
 206         if (hlimit < 0)
 207                 hlimit = ip6_dst_hoplimit(dst);
 208
 209         ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
 210                                                      np->autoflowlabel));
 211
 212         hdr->payload_len = htons(seg_len);
 213         hdr->nexthdr = proto;
 214         hdr->hop_limit = hlimit;
 215
 216         hdr->saddr = fl6->saddr;
 217         hdr->daddr = *first_hop;
 218
 219         skb->protocol = htons(ETH_P_IPV6);
 220         skb->priority = sk->sk_priority;
 221         skb->mark = sk->sk_mark;
 222
 223         mtu = dst_mtu(dst);
 224         if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
 225                 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
 226                               IPSTATS_MIB_OUT, skb->len);
 227                 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, sk, skb,
 228                                NULL, dst->dev, dst_output_sk);
 229         }
 230
 231         skb->dev = dst->dev;
 232         ipv6_local_error(sk, EMSGSIZE, fl6, mtu);
 233         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
 234         kfree_skb(skb);
 235         return -EMSGSIZE;
 236 }
 237 EXPORT_SYMBOL(ip6_xmit);
 238
 239 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
 240 {
 241         struct ip6_ra_chain *ra;
 242         struct sock *last = NULL;
 243
 244         read_lock(&ip6_ra_lock);
 245         for (ra = ip6_ra_chain; ra; ra = ra->next) {
 246                 struct sock *sk = ra->sk;
 247                 if (sk && ra->sel == sel &&
 248                     (!sk->sk_bound_dev_if ||
 249                      sk->sk_bound_dev_if == skb->dev->ifindex)) {
 250                         if (last) {
 251                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
 252                                 if (skb2)
 253                                         rawv6_rcv(last, skb2);
 254                         }
 255                         last = sk;
 256                 }
 257         }
 258
 259         if (last) {
 260                 rawv6_rcv(last, skb);
 261                 read_unlock(&ip6_ra_lock);
 262                 return 1;
 263         }
 264         read_unlock(&ip6_ra_lock);
 265         return 0;
 266 }
 267
 268 static int ip6_forward_proxy_check(struct sk_buff *skb)
 269 {
 270         struct ipv6hdr *hdr = ipv6_hdr(skb);
 271         u8 nexthdr = hdr->nexthdr;
 272         __be16 frag_off;
 273         int offset;
 274
 275         if (ipv6_ext_hdr(nexthdr)) {
 276                 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
 277                 if (offset < 0)
 278                         return 0;
 279         } else
 280                 offset = sizeof(struct ipv6hdr);
 281
 282         if (nexthdr == IPPROTO_ICMPV6) {
 283                 struct icmp6hdr *icmp6;
 284
 285                 if (!pskb_may_pull(skb, (skb_network_header(skb) +
 286                                          offset + 1 - skb->data)))
 287                         return 0;
 288
 289                 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
 290
 291                 switch (icmp6->icmp6_type) {
 292                 case NDISC_ROUTER_SOLICITATION:
 293                 case NDISC_ROUTER_ADVERTISEMENT:
 294                 case NDISC_NEIGHBOUR_SOLICITATION:
 295                 case NDISC_NEIGHBOUR_ADVERTISEMENT:
 296                 case NDISC_REDIRECT:
 297                         /* For reaction involving unicast neighbor discovery
 298                          * message destined to the proxied address, pass it to
 299                          * input function.
 300                          */
 301                         return 1;
 302                 default:
 303                         break;
 304                 }
 305         }
 306
 307         /*
 308          * The proxying router can't forward traffic sent to a link-local
 309          * address, so signal the sender and discard the packet. This
 310          * behavior is clarified by the MIPv6 specification.
 311          */
 312         if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
 313                 dst_link_failure(skb);
 314                 return -1;
 315         }
 316
 317         return 0;
 318 }
 319
 320 static inline int ip6_forward_finish(struct sock *sk, struct sk_buff *skb)
 321 {
 322         skb_sender_cpu_clear(skb);
 323         return dst_output_sk(sk, skb);
 324 }
 325
 326 static unsigned int ip6_dst_mtu_forward(const struct dst_entry *dst)
 327 {
 328         unsigned int mtu;
 329         struct inet6_dev *idev;
 330
 331         if (dst_metric_locked(dst, RTAX_MTU)) {
 332                 mtu = dst_metric_raw(dst, RTAX_MTU);
 333                 if (mtu)
 334                         return mtu;
 335         }
 336
 337         mtu = IPV6_MIN_MTU;
 338         rcu_read_lock();
 339         idev = __in6_dev_get(dst->dev);
 340         if (idev)
 341                 mtu = idev->cnf.mtu6;
 342         rcu_read_unlock();
 343
 344         return mtu;
 345 }
 346
 347 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
 348 {
 349         if (skb->len <= mtu)
 350                 return false;
 351
 352         /* ipv6 conntrack defrag sets max_frag_size + ignore_df */
 353         if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
 354                 return true;
 355
 356         if (skb->ignore_df)
 357                 return false;
 358
 359         if (skb_is_gso(skb) && skb_gso_network_seglen(skb) <= mtu)
 360                 return false;
 361
 362         return true;
 363 }
 364
 365 int ip6_forward(struct sk_buff *skb)
 366 {
 367         struct dst_entry *dst = skb_dst(skb);
 368         struct ipv6hdr *hdr = ipv6_hdr(skb);
 369         struct inet6_skb_parm *opt = IP6CB(skb);
 370         struct net *net = dev_net(dst->dev);
 371         u32 mtu;
 372
 373         if (net->ipv6.devconf_all->forwarding == 0)
 374                 goto error;
 375
 376         if (skb->pkt_type != PACKET_HOST)
 377                 goto drop;
 378
 379         if (skb_warn_if_lro(skb))
 380                 goto drop;
 381
 382         if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
 383                 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
 384                                  IPSTATS_MIB_INDISCARDS);
 385                 goto drop;
 386         }
 387
 388         skb_forward_csum(skb);
 389
 390         /*
 391          *      We DO NOT make any processing on
 392          *      RA packets, pushing them to user level AS IS
 393          *      without ane WARRANTY that application will be able
 394          *      to interpret them. The reason is that we
 395          *      cannot make anything clever here.
 396          *
 397          *      We are not end-node, so that if packet contains
 398          *      AH/ESP, we cannot make anything.
 399          *      Defragmentation also would be mistake, RA packets
 400          *      cannot be fragmented, because there is no warranty
 401          *      that different fragments will go along one path. --ANK
 402          */
 403         if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
 404                 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
 405                         return 0;
 406         }
 407
 408         /*
 409          *      check and decrement ttl
 410          */
 411         if (hdr->hop_limit <= 1) {
 412                 /* Force OUTPUT device used as source address */
 413                 skb->dev = dst->dev;
 414                 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
 415                 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
 416                                  IPSTATS_MIB_INHDRERRORS);
 417
 418                 kfree_skb(skb);
 419                 return -ETIMEDOUT;
 420         }
 421
 422         /* XXX: idev->cnf.proxy_ndp? */
 423         if (net->ipv6.devconf_all->proxy_ndp &&
 424             pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
 425                 int proxied = ip6_forward_proxy_check(skb);
 426                 if (proxied > 0)
 427                         return ip6_input(skb);
 428                 else if (proxied < 0) {
 429                         IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
 430                                          IPSTATS_MIB_INDISCARDS);
 431                         goto drop;
 432                 }
 433         }
 434
 435         if (!xfrm6_route_forward(skb)) {
 436                 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
 437                                  IPSTATS_MIB_INDISCARDS);
 438                 goto drop;
 439         }
 440         dst = skb_dst(skb);
 441
 442         /* IPv6 specs say nothing about it, but it is clear that we cannot
 443            send redirects to source routed frames.
 444            We don't send redirects to frames decapsulated from IPsec.
 445          */
 446         if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
 447                 struct in6_addr *target = NULL;
 448                 struct inet_peer *peer;
 449                 struct rt6_info *rt;
 450
 451                 /*
 452                  *      incoming and outgoing devices are the same
 453                  *      send a redirect.
 454                  */
 455
 456                 rt = (struct rt6_info *) dst;
 457                 if (rt->rt6i_flags & RTF_GATEWAY)
 458                         target = &rt->rt6i_gateway;
 459                 else
 460                         target = &hdr->daddr;
 461
 462                 peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
 463
 464                 /* Limit redirects both by destination (here)
 465                    and by source (inside ndisc_send_redirect)
 466                  */
 467                 if (inet_peer_xrlim_allow(peer, 1*HZ))
 468                         ndisc_send_redirect(skb, target);
 469                 if (peer)
 470                         inet_putpeer(peer);
 471         } else {
 472                 int addrtype = ipv6_addr_type(&hdr->saddr);
 473
 474                 /* This check is security critical. */
 475                 if (addrtype == IPV6_ADDR_ANY ||
 476                     addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
 477                         goto error;
 478                 if (addrtype & IPV6_ADDR_LINKLOCAL) {
 479                         icmpv6_send(skb, ICMPV6_DEST_UNREACH,
 480                                     ICMPV6_NOT_NEIGHBOUR, 0);
 481                         goto error;
 482                 }
 483         }
 484
 485         mtu = ip6_dst_mtu_forward(dst);
 486         if (mtu < IPV6_MIN_MTU)
 487                 mtu = IPV6_MIN_MTU;
 488
 489         if (ip6_pkt_too_big(skb, mtu)) {
 490                 /* Again, force OUTPUT device used as source address */
 491                 skb->dev = dst->dev;
 492                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
 493                 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
 494                                  IPSTATS_MIB_INTOOBIGERRORS);
 495                 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
 496                                  IPSTATS_MIB_FRAGFAILS);
 497                 kfree_skb(skb);
 498                 return -EMSGSIZE;
 499         }
 500
 501         if (skb_cow(skb, dst->dev->hard_header_len)) {
 502                 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
 503                                  IPSTATS_MIB_OUTDISCARDS);
 504                 goto drop;
 505         }
 506
 507         hdr = ipv6_hdr(skb);
 508
 509         /* Mangling hops number delayed to point after skb COW */
 510
 511         hdr->hop_limit--;
 512
 513         IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
 514         IP6_ADD_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
 515         return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD, NULL, skb,
 516                        skb->dev, dst->dev,
 517                        ip6_forward_finish);
 518
 519 error:
 520         IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
 521 drop:
 522         kfree_skb(skb);
 523         return -EINVAL;
 524 }
 525
 526 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
 527 {
 528         to->pkt_type = from->pkt_type;
 529         to->priority = from->priority;
 530         to->protocol = from->protocol;
 531         skb_dst_drop(to);
 532         skb_dst_set(to, dst_clone(skb_dst(from)));
 533         to->dev = from->dev;
 534         to->mark = from->mark;
 535
 536 #ifdef CONFIG_NET_SCHED
 537         to->tc_index = from->tc_index;
 538 #endif
 539         nf_copy(to, from);
 540         skb_copy_secmark(to, from);
 541 }
 542
 543 int ip6_fragment(struct sock *sk, struct sk_buff *skb,
 544                  int (*output)(struct sock *, struct sk_buff *))
 545 {
 546         struct sk_buff *frag;
 547         struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
 548         struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
 549                                 inet6_sk(skb->sk) : NULL;
 550         struct ipv6hdr *tmp_hdr;
 551         struct frag_hdr *fh;
 552         unsigned int mtu, hlen, left, len;
 553         int hroom, troom;
 554         __be32 frag_id;
 555         int ptr, offset = 0, err = 0;
 556         u8 *prevhdr, nexthdr = 0;
 557         struct net *net = dev_net(skb_dst(skb)->dev);
 558
 559         hlen = ip6_find_1stfragopt(skb, &prevhdr);
 560         nexthdr = *prevhdr;
 561
 562         mtu = ip6_skb_dst_mtu(skb);
 563
 564         /* We must not fragment if the socket is set to force MTU discovery
 565          * or if the skb it not generated by a local socket.
 566          */
 567         if (unlikely(!skb->ignore_df && skb->len > mtu))
 568                 goto fail_toobig;
 569
 570         if (IP6CB(skb)->frag_max_size) {
 571                 if (IP6CB(skb)->frag_max_size > mtu)
 572                         goto fail_toobig;
 573
 574                 /* don't send fragments larger than what we received */
 575                 mtu = IP6CB(skb)->frag_max_size;
 576                 if (mtu < IPV6_MIN_MTU)
 577                         mtu = IPV6_MIN_MTU;
 578         }
 579
 580         if (np && np->frag_size < mtu) {
 581                 if (np->frag_size)
 582                         mtu = np->frag_size;
 583         }
 584         mtu -= hlen + sizeof(struct frag_hdr);
 585
 586         frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
 587                                     &ipv6_hdr(skb)->saddr);
 588
 589         if (skb_has_frag_list(skb)) {
 590                 int first_len = skb_pagelen(skb);
 591                 struct sk_buff *frag2;
 592
 593                 if (first_len - hlen > mtu ||
 594                     ((first_len - hlen) & 7) ||
 595                     skb_cloned(skb))
 596                         goto slow_path;
 597
 598                 skb_walk_frags(skb, frag) {
 599                         /* Correct geometry. */
 600                         if (frag->len > mtu ||
 601                             ((frag->len & 7) && frag->next) ||
 602                             skb_headroom(frag) < hlen)
 603                                 goto slow_path_clean;
 604
 605                         /* Partially cloned skb? */
 606                         if (skb_shared(frag))
 607                                 goto slow_path_clean;
 608
 609                         BUG_ON(frag->sk);
 610                         if (skb->sk) {
 611                                 frag->sk = skb->sk;
 612                                 frag->destructor = sock_wfree;
 613                         }
 614                         skb->truesize -= frag->truesize;
 615                 }
 616
 617                 err = 0;
 618                 offset = 0;
 619                 frag = skb_shinfo(skb)->frag_list;
 620                 skb_frag_list_init(skb);
 621                 /* BUILD HEADER */
 622
 623                 *prevhdr = NEXTHDR_FRAGMENT;
 624                 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
 625                 if (!tmp_hdr) {
 626                         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 627                                       IPSTATS_MIB_FRAGFAILS);
 628                         return -ENOMEM;
 629                 }
 630
 631                 __skb_pull(skb, hlen);
 632                 fh = (struct frag_hdr *)__skb_push(skb, sizeof(struct frag_hdr));
 633                 __skb_push(skb, hlen);
 634                 skb_reset_network_header(skb);
 635                 memcpy(skb_network_header(skb), tmp_hdr, hlen);
 636
 637                 fh->nexthdr = nexthdr;
 638                 fh->reserved = 0;
 639                 fh->frag_off = htons(IP6_MF);
 640                 fh->identification = frag_id;
 641
 642                 first_len = skb_pagelen(skb);
 643                 skb->data_len = first_len - skb_headlen(skb);
 644                 skb->len = first_len;
 645                 ipv6_hdr(skb)->payload_len = htons(first_len -
 646                                                    sizeof(struct ipv6hdr));
 647
 648                 dst_hold(&rt->dst);
 649
 650                 for (;;) {
 651                         /* Prepare header of the next frame,
 652                          * before previous one went down. */
 653                         if (frag) {
 654                                 frag->ip_summed = CHECKSUM_NONE;
 655                                 skb_reset_transport_header(frag);
 656                                 fh = (struct frag_hdr *)__skb_push(frag, sizeof(struct frag_hdr));
 657                                 __skb_push(frag, hlen);
 658                                 skb_reset_network_header(frag);
 659                                 memcpy(skb_network_header(frag), tmp_hdr,
 660                                        hlen);
 661                                 offset += skb->len - hlen - sizeof(struct frag_hdr);
 662                                 fh->nexthdr = nexthdr;
 663                                 fh->reserved = 0;
 664                                 fh->frag_off = htons(offset);
 665                                 if (frag->next)
 666                                         fh->frag_off |= htons(IP6_MF);
 667                                 fh->identification = frag_id;
 668                                 ipv6_hdr(frag)->payload_len =
 669                                                 htons(frag->len -
 670                                                       sizeof(struct ipv6hdr));
 671                                 ip6_copy_metadata(frag, skb);
 672                         }
 673
 674                         err = output(sk, skb);
 675                         if (!err)
 676                                 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
 677                                               IPSTATS_MIB_FRAGCREATES);
 678
 679                         if (err || !frag)
 680                                 break;
 681
 682                         skb = frag;
 683                         frag = skb->next;
 684                         skb->next = NULL;
 685                 }
 686
 687                 kfree(tmp_hdr);
 688
 689                 if (err == 0) {
 690                         IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
 691                                       IPSTATS_MIB_FRAGOKS);
 692                         ip6_rt_put(rt);
 693                         return 0;
 694                 }
 695
 696                 kfree_skb_list(frag);
 697
 698                 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
 699                               IPSTATS_MIB_FRAGFAILS);
 700                 ip6_rt_put(rt);
 701                 return err;
 702
 703 slow_path_clean:
 704                 skb_walk_frags(skb, frag2) {
 705                         if (frag2 == frag)
 706                                 break;
 707                         frag2->sk = NULL;
 708                         frag2->destructor = NULL;
 709                         skb->truesize += frag2->truesize;
 710                 }
 711         }
 712
 713 slow_path:
 714         if ((skb->ip_summed == CHECKSUM_PARTIAL) &&
 715             skb_checksum_help(skb))
 716                 goto fail;
 717
 718         left = skb->len - hlen;         /* Space per frame */
 719         ptr = hlen;                     /* Where to start from */
 720
 721         /*
 722          *      Fragment the datagram.
 723          */
 724
 725         *prevhdr = NEXTHDR_FRAGMENT;
 726         hroom = LL_RESERVED_SPACE(rt->dst.dev);
 727         troom = rt->dst.dev->needed_tailroom;
 728
 729         /*
 730          *      Keep copying data until we run out.
 731          */
 732         while (left > 0)        {
 733                 len = left;
 734                 /* IF: it doesn't fit, use 'mtu' - the data space left */
 735                 if (len > mtu)
 736                         len = mtu;
 737                 /* IF: we are not sending up to and including the packet end
 738                    then align the next start on an eight byte boundary */
 739                 if (len < left) {
 740                         len &= ~7;
 741                 }
 742
 743                 /* Allocate buffer */
 744                 frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
 745                                  hroom + troom, GFP_ATOMIC);
 746                 if (!frag) {
 747                         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 748                                       IPSTATS_MIB_FRAGFAILS);
 749                         err = -ENOMEM;
 750                         goto fail;
 751                 }
 752
 753                 /*
 754                  *      Set up data on packet
 755                  */
 756
 757                 ip6_copy_metadata(frag, skb);
 758                 skb_reserve(frag, hroom);
 759                 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
 760                 skb_reset_network_header(frag);
 761                 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
 762                 frag->transport_header = (frag->network_header + hlen +
 763                                           sizeof(struct frag_hdr));
 764
 765                 /*
 766                  *      Charge the memory for the fragment to any owner
 767                  *      it might possess
 768                  */
 769                 if (skb->sk)
 770                         skb_set_owner_w(frag, skb->sk);
 771
 772                 /*
 773                  *      Copy the packet header into the new buffer.
 774                  */
 775                 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
 776
 777                 /*
 778                  *      Build fragment header.
 779                  */
 780                 fh->nexthdr = nexthdr;
 781                 fh->reserved = 0;
 782                 fh->identification = frag_id;
 783
 784                 /*
 785                  *      Copy a block of the IP datagram.
 786                  */
 787                 BUG_ON(skb_copy_bits(skb, ptr, skb_transport_header(frag),
 788                                      len));
 789                 left -= len;
 790
 791                 fh->frag_off = htons(offset);
 792                 if (left > 0)
 793                         fh->frag_off |= htons(IP6_MF);
 794                 ipv6_hdr(frag)->payload_len = htons(frag->len -
 795                                                     sizeof(struct ipv6hdr));
 796
 797                 ptr += len;
 798                 offset += len;
 799
 800                 /*
 801                  *      Put this fragment into the sending queue.
 802                  */
 803                 err = output(sk, frag);
 804                 if (err)
 805                         goto fail;
 806
 807                 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 808                               IPSTATS_MIB_FRAGCREATES);
 809         }
 810         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 811                       IPSTATS_MIB_FRAGOKS);
 812         consume_skb(skb);
 813         return err;
 814
 815 fail_toobig:
 816         if (skb->sk && dst_allfrag(skb_dst(skb)))
 817                 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
 818
 819         skb->dev = skb_dst(skb)->dev;
 820         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
 821         err = -EMSGSIZE;
 822
 823 fail:
 824         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 825                       IPSTATS_MIB_FRAGFAILS);
 826         kfree_skb(skb);
 827         return err;
 828 }
 829
 830 static inline int ip6_rt_check(const struct rt6key *rt_key,
 831                                const struct in6_addr *fl_addr,
 832                                const struct in6_addr *addr_cache)
 833 {
 834         return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
 835                 (!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
 836 }
 837
 838 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
 839                                           struct dst_entry *dst,
 840                                           const struct flowi6 *fl6)
 841 {
 842         struct ipv6_pinfo *np = inet6_sk(sk);
 843         struct rt6_info *rt;
 844
 845         if (!dst)
 846                 goto out;
 847
 848         if (dst->ops->family != AF_INET6) {
 849                 dst_release(dst);
 850                 return NULL;
 851         }
 852
 853         rt = (struct rt6_info *)dst;
 854         /* Yes, checking route validity in not connected
 855          * case is not very simple. Take into account,
 856          * that we do not support routing by source, TOS,
 857          * and MSG_DONTROUTE            --ANK (980726)
 858          *
 859          * 1. ip6_rt_check(): If route was host route,
 860          *    check that cached destination is current.
 861          *    If it is network route, we still may
 862          *    check its validity using saved pointer
 863          *    to the last used address: daddr_cache.
 864          *    We do not want to save whole address now,
 865          *    (because main consumer of this service
 866          *    is tcp, which has not this problem),
 867          *    so that the last trick works only on connected
 868          *    sockets.
 869          * 2. oif also should be the same.
 870          */
 871         if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
 872 #ifdef CONFIG_IPV6_SUBTREES
 873             ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
 874 #endif
 875             (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
 876                 dst_release(dst);
 877                 dst = NULL;
 878         }
 879
 880 out:
 881         return dst;
 882 }
 883
 884 static int ip6_dst_lookup_tail(struct sock *sk,
 885                                struct dst_entry **dst, struct flowi6 *fl6)
 886 {
 887         struct net *net = sock_net(sk);
 888 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
 889         struct neighbour *n;
 890         struct rt6_info *rt;
 891 #endif
 892         int err;
 893
 894         /* The correct way to handle this would be to do
 895          * ip6_route_get_saddr, and then ip6_route_output; however,
 896          * the route-specific preferred source forces the
 897          * ip6_route_output call _before_ ip6_route_get_saddr.
 898          *
 899          * In source specific routing (no src=any default route),
 900          * ip6_route_output will fail given src=any saddr, though, so
 901          * that's why we try it again later.
 902          */
 903         if (ipv6_addr_any(&fl6->saddr) && (!*dst || !(*dst)->error)) {
 904                 struct rt6_info *rt;
 905                 bool had_dst = *dst != NULL;
 906
 907                 if (!had_dst)
 908                         *dst = ip6_route_output(net, sk, fl6);
 909                 rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
 910                 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
 911                                           sk ? inet6_sk(sk)->srcprefs : 0,
 912                                           &fl6->saddr);
 913                 if (err)
 914                         goto out_err_release;
 915
 916                 /* If we had an erroneous initial result, pretend it
 917                  * never existed and let the SA-enabled version take
 918                  * over.
 919                  */
 920                 if (!had_dst && (*dst)->error) {
 921                         dst_release(*dst);
 922                         *dst = NULL;
 923                 }
 924         }
 925
 926         if (!*dst)
 927                 *dst = ip6_route_output(net, sk, fl6);
 928
 929         err = (*dst)->error;
 930         if (err)
 931                 goto out_err_release;
 932
 933 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
 934         /*
 935          * Here if the dst entry we've looked up
 936          * has a neighbour entry that is in the INCOMPLETE
 937          * state and the src address from the flow is
 938          * marked as OPTIMISTIC, we release the found
 939          * dst entry and replace it instead with the
 940          * dst entry of the nexthop router
 941          */
 942         rt = (struct rt6_info *) *dst;
 943         rcu_read_lock_bh();
 944         n = __ipv6_neigh_lookup_noref(rt->dst.dev,
 945                                       rt6_nexthop(rt, &fl6->daddr));
 946         err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
 947         rcu_read_unlock_bh();
 948
 949         if (err) {
 950                 struct inet6_ifaddr *ifp;
 951                 struct flowi6 fl_gw6;
 952                 int redirect;
 953
 954                 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
 955                                       (*dst)->dev, 1);
 956
 957                 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
 958                 if (ifp)
 959                         in6_ifa_put(ifp);
 960
 961                 if (redirect) {
 962                         /*
 963                          * We need to get the dst entry for the
 964                          * default router instead
 965                          */
 966                         dst_release(*dst);
 967                         memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
 968                         memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
 969                         *dst = ip6_route_output(net, sk, &fl_gw6);
 970                         err = (*dst)->error;
 971                         if (err)
 972                                 goto out_err_release;
 973                 }
 974         }
 975 #endif
 976
 977         return 0;
 978
 979 out_err_release:
 980         if (err == -ENETUNREACH)
 981                 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
 982         dst_release(*dst);
 983         *dst = NULL;
 984         return err;
 985 }
 986
 987 /**
 988  *      ip6_dst_lookup - perform route lookup on flow
 989  *      @sk: socket which provides route info
 990  *      @dst: pointer to dst_entry * for result
 991  *      @fl6: flow to lookup
 992  *
 993  *      This function performs a route lookup on the given flow.
 994  *
 995  *      It returns zero on success, or a standard errno code on error.
 996  */
 997 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi6 *fl6)
 998 {
 999         *dst = NULL;
1000         return ip6_dst_lookup_tail(sk, dst, fl6);
1001 }
1002 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1003
1004 /**
1005  *      ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1006  *      @sk: socket which provides route info
1007  *      @fl6: flow to lookup
1008  *      @final_dst: final destination address for ipsec lookup
1009  *
1010  *      This function performs a route lookup on the given flow.
1011  *
1012  *      It returns a valid dst pointer on success, or a pointer encoded
1013  *      error code.
1014  */
1015 struct dst_entry *ip6_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1016                                       const struct in6_addr *final_dst)
1017 {
1018         struct dst_entry *dst = NULL;
1019         int err;
1020
1021         err = ip6_dst_lookup_tail(sk, &dst, fl6);
1022         if (err)
1023                 return ERR_PTR(err);
1024         if (final_dst)
1025                 fl6->daddr = *final_dst;
1026
1027         return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1028 }
1029 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1030
1031 /**
1032  *      ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1033  *      @sk: socket which provides the dst cache and route info
1034  *      @fl6: flow to lookup
1035  *      @final_dst: final destination address for ipsec lookup
1036  *
1037  *      This function performs a route lookup on the given flow with the
1038  *      possibility of using the cached route in the socket if it is valid.
1039  *      It will take the socket dst lock when operating on the dst cache.
1040  *      As a result, this function can only be used in process context.
1041  *
1042  *      It returns a valid dst pointer on success, or a pointer encoded
1043  *      error code.
1044  */
1045 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1046                                          const struct in6_addr *final_dst)
1047 {
1048         struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1049         int err;
1050
1051         dst = ip6_sk_dst_check(sk, dst, fl6);
1052
1053         err = ip6_dst_lookup_tail(sk, &dst, fl6);
1054         if (err)
1055                 return ERR_PTR(err);
1056         if (final_dst)
1057                 fl6->daddr = *final_dst;
1058
1059         return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1060 }
1061 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1062
1063 static inline int ip6_ufo_append_data(struct sock *sk,
1064                         struct sk_buff_head *queue,
1065                         int getfrag(void *from, char *to, int offset, int len,
1066                         int odd, struct sk_buff *skb),
1067                         void *from, int length, int hh_len, int fragheaderlen,
1068                         int transhdrlen, int mtu, unsigned int flags,
1069                         const struct flowi6 *fl6)
1070
1071 {
1072         struct sk_buff *skb;
1073         int err;
1074
1075         /* There is support for UDP large send offload by network
1076          * device, so create one single skb packet containing complete
1077          * udp datagram
1078          */
1079         skb = skb_peek_tail(queue);
1080         if (!skb) {
1081                 skb = sock_alloc_send_skb(sk,
1082                         hh_len + fragheaderlen + transhdrlen + 20,
1083                         (flags & MSG_DONTWAIT), &err);
1084                 if (!skb)
1085                         return err;
1086
1087                 /* reserve space for Hardware header */
1088                 skb_reserve(skb, hh_len);
1089
1090                 /* create space for UDP/IP header */
1091                 skb_put(skb, fragheaderlen + transhdrlen);
1092
1093                 /* initialize network header pointer */
1094                 skb_reset_network_header(skb);
1095
1096                 /* initialize protocol header pointer */
1097                 skb->transport_header = skb->network_header + fragheaderlen;
1098
1099                 skb->protocol = htons(ETH_P_IPV6);
1100                 skb->csum = 0;
1101
1102                 __skb_queue_tail(queue, skb);
1103         } else if (skb_is_gso(skb)) {
1104                 goto append;
1105         }
1106
1107         skb->ip_summed = CHECKSUM_PARTIAL;
1108         /* Specify the length of each IPv6 datagram fragment.
1109          * It has to be a multiple of 8.
1110          */
1111         skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1112                                      sizeof(struct frag_hdr)) & ~7;
1113         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1114         skb_shinfo(skb)->ip6_frag_id = ipv6_select_ident(sock_net(sk),
1115                                                          &fl6->daddr,
1116                                                          &fl6->saddr);
1117
1118 append:
1119         return skb_append_datato_frags(sk, skb, getfrag, from,
1120                                        (length - transhdrlen));
1121 }
1122
1123 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1124                                                gfp_t gfp)
1125 {
1126         return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1127 }
1128
1129 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1130                                                 gfp_t gfp)
1131 {
1132         return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1133 }
1134
1135 static void ip6_append_data_mtu(unsigned int *mtu,
1136                                 int *maxfraglen,
1137                                 unsigned int fragheaderlen,
1138                                 struct sk_buff *skb,
1139                                 struct rt6_info *rt,
1140                                 unsigned int orig_mtu)
1141 {
1142         if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1143                 if (!skb) {
1144                         /* first fragment, reserve header_len */
1145                         *mtu = orig_mtu - rt->dst.header_len;
1146
1147                 } else {
1148                         /*
1149                          * this fragment is not first, the headers
1150                          * space is regarded as data space.
1151                          */
1152                         *mtu = orig_mtu;
1153                 }
1154                 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1155                               + fragheaderlen - sizeof(struct frag_hdr);
1156         }
1157 }
1158
1159 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1160                           struct inet6_cork *v6_cork,
1161                           int hlimit, int tclass, struct ipv6_txoptions *opt,
1162                           struct rt6_info *rt, struct flowi6 *fl6)
1163 {
1164         struct ipv6_pinfo *np = inet6_sk(sk);
1165         unsigned int mtu;
1166
1167         /*
1168          * setup for corking
1169          */
1170         if (opt) {
1171                 if (WARN_ON(v6_cork->opt))
1172                         return -EINVAL;
1173
1174                 v6_cork->opt = kzalloc(opt->tot_len, sk->sk_allocation);
1175                 if (unlikely(!v6_cork->opt))
1176                         return -ENOBUFS;
1177
1178                 v6_cork->opt->tot_len = opt->tot_len;
1179                 v6_cork->opt->opt_flen = opt->opt_flen;
1180                 v6_cork->opt->opt_nflen = opt->opt_nflen;
1181
1182                 v6_cork->opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1183                                                     sk->sk_allocation);
1184                 if (opt->dst0opt && !v6_cork->opt->dst0opt)
1185                         return -ENOBUFS;
1186
1187                 v6_cork->opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1188                                                     sk->sk_allocation);
1189                 if (opt->dst1opt && !v6_cork->opt->dst1opt)
1190                         return -ENOBUFS;
1191
1192                 v6_cork->opt->hopopt = ip6_opt_dup(opt->hopopt,
1193                                                    sk->sk_allocation);
1194                 if (opt->hopopt && !v6_cork->opt->hopopt)
1195                         return -ENOBUFS;
1196
1197                 v6_cork->opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1198                                                     sk->sk_allocation);
1199                 if (opt->srcrt && !v6_cork->opt->srcrt)
1200                         return -ENOBUFS;
1201
1202                 /* need source address above miyazawa*/
1203         }
1204         dst_hold(&rt->dst);
1205         cork->base.dst = &rt->dst;
1206         cork->fl.u.ip6 = *fl6;
1207         v6_cork->hop_limit = hlimit;
1208         v6_cork->tclass = tclass;
1209         if (rt->dst.flags & DST_XFRM_TUNNEL)
1210                 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1211                       rt->dst.dev->mtu : dst_mtu(&rt->dst);
1212         else
1213                 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1214                       rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1215         if (np->frag_size < mtu) {
1216                 if (np->frag_size)
1217                         mtu = np->frag_size;
1218         }
1219         cork->base.fragsize = mtu;
1220         if (dst_allfrag(rt->dst.path))
1221                 cork->base.flags |= IPCORK_ALLFRAG;
1222         cork->base.length = 0;
1223
1224         return 0;
1225 }
1226
1227 static int __ip6_append_data(struct sock *sk,
1228                              struct flowi6 *fl6,
1229                              struct sk_buff_head *queue,
1230                              struct inet_cork *cork,
1231                              struct inet6_cork *v6_cork,
1232                              struct page_frag *pfrag,
1233                              int getfrag(void *from, char *to, int offset,
1234                                          int len, int odd, struct sk_buff *skb),
1235                              void *from, int length, int transhdrlen,
1236                              unsigned int flags, int dontfrag)
1237 {
1238         struct sk_buff *skb, *skb_prev = NULL;
1239         unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu;
1240         int exthdrlen = 0;
1241         int dst_exthdrlen = 0;
1242         int hh_len;
1243         int copy;
1244         int err;
1245         int offset = 0;
1246         __u8 tx_flags = 0;
1247         u32 tskey = 0;
1248         struct rt6_info *rt = (struct rt6_info *)cork->dst;
1249         struct ipv6_txoptions *opt = v6_cork->opt;
1250         int csummode = CHECKSUM_NONE;
1251
1252         skb = skb_peek_tail(queue);
1253         if (!skb) {
1254                 exthdrlen = opt ? opt->opt_flen : 0;
1255                 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1256         }
1257
1258         mtu = cork->fragsize;
1259         orig_mtu = mtu;
1260
1261         hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1262
1263         fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1264                         (opt ? opt->opt_nflen : 0);
1265         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1266                      sizeof(struct frag_hdr);
1267
1268         if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1269                 unsigned int maxnonfragsize, headersize;
1270
1271                 headersize = sizeof(struct ipv6hdr) +
1272                              (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1273                              (dst_allfrag(&rt->dst) ?
1274                               sizeof(struct frag_hdr) : 0) +
1275                              rt->rt6i_nfheader_len;
1276
1277                 if (ip6_sk_ignore_df(sk))
1278                         maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1279                 else
1280                         maxnonfragsize = mtu;
1281
1282                 /* dontfrag active */
1283                 if ((cork->length + length > mtu - headersize) && dontfrag &&
1284                     (sk->sk_protocol == IPPROTO_UDP ||
1285                      sk->sk_protocol == IPPROTO_RAW)) {
1286                         ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1287                                                    sizeof(struct ipv6hdr));
1288                         goto emsgsize;
1289                 }
1290
1291                 if (cork->length + length > maxnonfragsize - headersize) {
1292 emsgsize:
1293                         ipv6_local_error(sk, EMSGSIZE, fl6,
1294                                          mtu - headersize +
1295                                          sizeof(struct ipv6hdr));
1296                         return -EMSGSIZE;
1297                 }
1298         }
1299
1300         if (sk->sk_type == SOCK_DGRAM || sk->sk_type == SOCK_RAW) {
1301                 sock_tx_timestamp(sk, &tx_flags);
1302                 if (tx_flags & SKBTX_ANY_SW_TSTAMP &&
1303                     sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1304                         tskey = sk->sk_tskey++;
1305         }
1306
1307         /* If this is the first and only packet and device
1308          * supports checksum offloading, let's use it.
1309          * Use transhdrlen, same as IPv4, because partial
1310          * sums only work when transhdrlen is set.
1311          */
1312         if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1313             length + fragheaderlen < mtu &&
1314             rt->dst.dev->features & NETIF_F_V6_CSUM &&
1315             !exthdrlen)
1316                 csummode = CHECKSUM_PARTIAL;
1317         /*
1318          * Let's try using as much space as possible.
1319          * Use MTU if total length of the message fits into the MTU.
1320          * Otherwise, we need to reserve fragment header and
1321          * fragment alignment (= 8-15 octects, in total).
1322          *
1323          * Note that we may need to "move" the data from the tail of
1324          * of the buffer to the new fragment when we split
1325          * the message.
1326          *
1327          * FIXME: It may be fragmented into multiple chunks
1328          *        at once if non-fragmentable extension headers
1329          *        are too large.
1330          * --yoshfuji
1331          */
1332
1333         cork->length += length;
1334         if (((length > mtu) ||
1335              (skb && skb_is_gso(skb))) &&
1336             (sk->sk_protocol == IPPROTO_UDP) &&
1337             (rt->dst.dev->features & NETIF_F_UFO) &&
1338             (sk->sk_type == SOCK_DGRAM)) {
1339                 err = ip6_ufo_append_data(sk, queue, getfrag, from, length,
1340                                           hh_len, fragheaderlen,
1341                                           transhdrlen, mtu, flags, fl6);
1342                 if (err)
1343                         goto error;
1344                 return 0;
1345         }
1346
1347         if (!skb)
1348                 goto alloc_new_skb;
1349
1350         while (length > 0) {
1351                 /* Check if the remaining data fits into current packet. */
1352                 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1353                 if (copy < length)
1354                         copy = maxfraglen - skb->len;
1355
1356                 if (copy <= 0) {
1357                         char *data;
1358                         unsigned int datalen;
1359                         unsigned int fraglen;
1360                         unsigned int fraggap;
1361                         unsigned int alloclen;
1362 alloc_new_skb:
1363                         /* There's no room in the current skb */
1364                         if (skb)
1365                                 fraggap = skb->len - maxfraglen;
1366                         else
1367                                 fraggap = 0;
1368                         /* update mtu and maxfraglen if necessary */
1369                         if (!skb || !skb_prev)
1370                                 ip6_append_data_mtu(&mtu, &maxfraglen,
1371                                                     fragheaderlen, skb, rt,
1372                                                     orig_mtu);
1373
1374                         skb_prev = skb;
1375
1376                         /*
1377                          * If remaining data exceeds the mtu,
1378                          * we know we need more fragment(s).
1379                          */
1380                         datalen = length + fraggap;
1381
1382                         if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1383                                 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1384                         if ((flags & MSG_MORE) &&
1385                             !(rt->dst.dev->features&NETIF_F_SG))
1386                                 alloclen = mtu;
1387                         else
1388                                 alloclen = datalen + fragheaderlen;
1389
1390                         alloclen += dst_exthdrlen;
1391
1392                         if (datalen != length + fraggap) {
1393                                 /*
1394                                  * this is not the last fragment, the trailer
1395                                  * space is regarded as data space.
1396                                  */
1397                                 datalen += rt->dst.trailer_len;
1398                         }
1399
1400                         alloclen += rt->dst.trailer_len;
1401                         fraglen = datalen + fragheaderlen;
1402
1403                         /*
1404                          * We just reserve space for fragment header.
1405                          * Note: this may be overallocation if the message
1406                          * (without MSG_MORE) fits into the MTU.
1407                          */
1408                         alloclen += sizeof(struct frag_hdr);
1409
1410                         if (transhdrlen) {
1411                                 skb = sock_alloc_send_skb(sk,
1412                                                 alloclen + hh_len,
1413                                                 (flags & MSG_DONTWAIT), &err);
1414                         } else {
1415                                 skb = NULL;
1416                                 if (atomic_read(&sk->sk_wmem_alloc) <=
1417                                     2 * sk->sk_sndbuf)
1418                                         skb = sock_wmalloc(sk,
1419                                                            alloclen + hh_len, 1,
1420                                                            sk->sk_allocation);
1421                                 if (unlikely(!skb))
1422                                         err = -ENOBUFS;
1423                         }
1424                         if (!skb)
1425                                 goto error;
1426                         /*
1427                          *      Fill in the control structures
1428                          */
1429                         skb->protocol = htons(ETH_P_IPV6);
1430                         skb->ip_summed = csummode;
1431                         skb->csum = 0;
1432                         /* reserve for fragmentation and ipsec header */
1433                         skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1434                                     dst_exthdrlen);
1435
1436                         /* Only the initial fragment is time stamped */
1437                         skb_shinfo(skb)->tx_flags = tx_flags;
1438                         tx_flags = 0;
1439                         skb_shinfo(skb)->tskey = tskey;
1440                         tskey = 0;
1441
1442                         /*
1443                          *      Find where to start putting bytes
1444                          */
1445                         data = skb_put(skb, fraglen);
1446                         skb_set_network_header(skb, exthdrlen);
1447                         data += fragheaderlen;
1448                         skb->transport_header = (skb->network_header +
1449                                                  fragheaderlen);
1450                         if (fraggap) {
1451                                 skb->csum = skb_copy_and_csum_bits(
1452                                         skb_prev, maxfraglen,
1453                                         data + transhdrlen, fraggap, 0);
1454                                 skb_prev->csum = csum_sub(skb_prev->csum,
1455                                                           skb->csum);
1456                                 data += fraggap;
1457                                 pskb_trim_unique(skb_prev, maxfraglen);
1458                         }
1459                         copy = datalen - transhdrlen - fraggap;
1460
1461                         if (copy < 0) {
1462                                 err = -EINVAL;
1463                                 kfree_skb(skb);
1464                                 goto error;
1465                         } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1466                                 err = -EFAULT;
1467                                 kfree_skb(skb);
1468                                 goto error;
1469                         }
1470
1471                         offset += copy;
1472                         length -= datalen - fraggap;
1473                         transhdrlen = 0;
1474                         exthdrlen = 0;
1475                         dst_exthdrlen = 0;
1476
1477                         /*
1478                          * Put the packet on the pending queue
1479                          */
1480                         __skb_queue_tail(queue, skb);
1481                         continue;
1482                 }
1483
1484                 if (copy > length)
1485                         copy = length;
1486
1487                 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1488                         unsigned int off;
1489
1490                         off = skb->len;
1491                         if (getfrag(from, skb_put(skb, copy),
1492                                                 offset, copy, off, skb) < 0) {
1493                                 __skb_trim(skb, off);
1494                                 err = -EFAULT;
1495                                 goto error;
1496                         }
1497                 } else {
1498                         int i = skb_shinfo(skb)->nr_frags;
1499
1500                         err = -ENOMEM;
1501                         if (!sk_page_frag_refill(sk, pfrag))
1502                                 goto error;
1503
1504                         if (!skb_can_coalesce(skb, i, pfrag->page,
1505                                               pfrag->offset)) {
1506                                 err = -EMSGSIZE;
1507                                 if (i == MAX_SKB_FRAGS)
1508                                         goto error;
1509
1510                                 __skb_fill_page_desc(skb, i, pfrag->page,
1511                                                      pfrag->offset, 0);
1512                                 skb_shinfo(skb)->nr_frags = ++i;
1513                                 get_page(pfrag->page);
1514                         }
1515                         copy = min_t(int, copy, pfrag->size - pfrag->offset);
1516                         if (getfrag(from,
1517                                     page_address(pfrag->page) + pfrag->offset,
1518                                     offset, copy, skb->len, skb) < 0)
1519                                 goto error_efault;
1520
1521                         pfrag->offset += copy;
1522                         skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1523                         skb->len += copy;
1524                         skb->data_len += copy;
1525                         skb->truesize += copy;
1526                         atomic_add(copy, &sk->sk_wmem_alloc);
1527                 }
1528                 offset += copy;
1529                 length -= copy;
1530         }
1531
1532         return 0;
1533
1534 error_efault:
1535         err = -EFAULT;
1536 error:
1537         cork->length -= length;
1538         IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1539         return err;
1540 }
1541
1542 int ip6_append_data(struct sock *sk,
1543                     int getfrag(void *from, char *to, int offset, int len,
1544                                 int odd, struct sk_buff *skb),
1545                     void *from, int length, int transhdrlen, int hlimit,
1546                     int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1547                     struct rt6_info *rt, unsigned int flags, int dontfrag)
1548 {
1549         struct inet_sock *inet = inet_sk(sk);
1550         struct ipv6_pinfo *np = inet6_sk(sk);
1551         int exthdrlen;
1552         int err;
1553
1554         if (flags&MSG_PROBE)
1555                 return 0;
1556         if (skb_queue_empty(&sk->sk_write_queue)) {
1557                 /*
1558                  * setup for corking
1559                  */
1560                 err = ip6_setup_cork(sk, &inet->cork, &np->cork, hlimit,
1561                                      tclass, opt, rt, fl6);
1562                 if (err)
1563                         return err;
1564
1565                 exthdrlen = (opt ? opt->opt_flen : 0);
1566                 length += exthdrlen;
1567                 transhdrlen += exthdrlen;
1568         } else {
1569                 fl6 = &inet->cork.fl.u.ip6;
1570                 transhdrlen = 0;
1571         }
1572
1573         return __ip6_append_data(sk, fl6, &sk->sk_write_queue, &inet->cork.base,
1574                                  &np->cork, sk_page_frag(sk), getfrag,
1575                                  from, length, transhdrlen, flags, dontfrag);
1576 }
1577 EXPORT_SYMBOL_GPL(ip6_append_data);
1578
1579 static void ip6_cork_release(struct inet_cork_full *cork,
1580                              struct inet6_cork *v6_cork)
1581 {
1582         if (v6_cork->opt) {
1583                 kfree(v6_cork->opt->dst0opt);
1584                 kfree(v6_cork->opt->dst1opt);
1585                 kfree(v6_cork->opt->hopopt);
1586                 kfree(v6_cork->opt->srcrt);
1587                 kfree(v6_cork->opt);
1588                 v6_cork->opt = NULL;
1589         }
1590
1591         if (cork->base.dst) {
1592                 dst_release(cork->base.dst);
1593                 cork->base.dst = NULL;
1594                 cork->base.flags &= ~IPCORK_ALLFRAG;
1595         }
1596         memset(&cork->fl, 0, sizeof(cork->fl));
1597 }
1598
1599 struct sk_buff *__ip6_make_skb(struct sock *sk,
1600                                struct sk_buff_head *queue,
1601                                struct inet_cork_full *cork,
1602                                struct inet6_cork *v6_cork)
1603 {
1604         struct sk_buff *skb, *tmp_skb;
1605         struct sk_buff **tail_skb;
1606         struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1607         struct ipv6_pinfo *np = inet6_sk(sk);
1608         struct net *net = sock_net(sk);
1609         struct ipv6hdr *hdr;
1610         struct ipv6_txoptions *opt = v6_cork->opt;
1611         struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1612         struct flowi6 *fl6 = &cork->fl.u.ip6;
1613         unsigned char proto = fl6->flowi6_proto;
1614
1615         skb = __skb_dequeue(queue);
1616         if (!skb)
1617                 goto out;
1618         tail_skb = &(skb_shinfo(skb)->frag_list);
1619
1620         /* move skb->data to ip header from ext header */
1621         if (skb->data < skb_network_header(skb))
1622                 __skb_pull(skb, skb_network_offset(skb));
1623         while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1624                 __skb_pull(tmp_skb, skb_network_header_len(skb));
1625                 *tail_skb = tmp_skb;
1626                 tail_skb = &(tmp_skb->next);
1627                 skb->len += tmp_skb->len;
1628                 skb->data_len += tmp_skb->len;
1629                 skb->truesize += tmp_skb->truesize;
1630                 tmp_skb->destructor = NULL;
1631                 tmp_skb->sk = NULL;
1632         }
1633
1634         /* Allow local fragmentation. */
1635         skb->ignore_df = ip6_sk_ignore_df(sk);
1636
1637         *final_dst = fl6->daddr;
1638         __skb_pull(skb, skb_network_header_len(skb));
1639         if (opt && opt->opt_flen)
1640                 ipv6_push_frag_opts(skb, opt, &proto);
1641         if (opt && opt->opt_nflen)
1642                 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1643
1644         skb_push(skb, sizeof(struct ipv6hdr));
1645         skb_reset_network_header(skb);
1646         hdr = ipv6_hdr(skb);
1647
1648         ip6_flow_hdr(hdr, v6_cork->tclass,
1649                      ip6_make_flowlabel(net, skb, fl6->flowlabel,
1650                                         np->autoflowlabel));
1651         hdr->hop_limit = v6_cork->hop_limit;
1652         hdr->nexthdr = proto;
1653         hdr->saddr = fl6->saddr;
1654         hdr->daddr = *final_dst;
1655
1656         skb->priority = sk->sk_priority;
1657         skb->mark = sk->sk_mark;
1658
1659         skb_dst_set(skb, dst_clone(&rt->dst));
1660         IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1661         if (proto == IPPROTO_ICMPV6) {
1662                 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1663
1664                 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1665                 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1666         }
1667
1668         ip6_cork_release(cork, v6_cork);
1669 out:
1670         return skb;
1671 }
1672
1673 int ip6_send_skb(struct sk_buff *skb)
1674 {
1675         struct net *net = sock_net(skb->sk);
1676         struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1677         int err;
1678
1679         err = ip6_local_out(skb);
1680         if (err) {
1681                 if (err > 0)
1682                         err = net_xmit_errno(err);
1683                 if (err)
1684                         IP6_INC_STATS(net, rt->rt6i_idev,
1685                                       IPSTATS_MIB_OUTDISCARDS);
1686         }
1687
1688         return err;
1689 }
1690
1691 int ip6_push_pending_frames(struct sock *sk)
1692 {
1693         struct sk_buff *skb;
1694
1695         skb = ip6_finish_skb(sk);
1696         if (!skb)
1697                 return 0;
1698
1699         return ip6_send_skb(skb);
1700 }
1701 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1702
1703 static void __ip6_flush_pending_frames(struct sock *sk,
1704                                        struct sk_buff_head *queue,
1705                                        struct inet_cork_full *cork,
1706                                        struct inet6_cork *v6_cork)
1707 {
1708         struct sk_buff *skb;
1709
1710         while ((skb = __skb_dequeue_tail(queue)) != NULL) {
1711                 if (skb_dst(skb))
1712                         IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1713                                       IPSTATS_MIB_OUTDISCARDS);
1714                 kfree_skb(skb);
1715         }
1716
1717         ip6_cork_release(cork, v6_cork);
1718 }
1719
1720 void ip6_flush_pending_frames(struct sock *sk)
1721 {
1722         __ip6_flush_pending_frames(sk, &sk->sk_write_queue,
1723                                    &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
1724 }
1725 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1726
1727 struct sk_buff *ip6_make_skb(struct sock *sk,
1728                              int getfrag(void *from, char *to, int offset,
1729                                          int len, int odd, struct sk_buff *skb),
1730                              void *from, int length, int transhdrlen,
1731                              int hlimit, int tclass,
1732                              struct ipv6_txoptions *opt, struct flowi6 *fl6,
1733                              struct rt6_info *rt, unsigned int flags,
1734                              int dontfrag)
1735 {
1736         struct inet_cork_full cork;
1737         struct inet6_cork v6_cork;
1738         struct sk_buff_head queue;
1739         int exthdrlen = (opt ? opt->opt_flen : 0);
1740         int err;
1741
1742         if (flags & MSG_PROBE)
1743                 return NULL;
1744
1745         __skb_queue_head_init(&queue);
1746
1747         cork.base.flags = 0;
1748         cork.base.addr = 0;
1749         cork.base.opt = NULL;
1750         v6_cork.opt = NULL;
1751         err = ip6_setup_cork(sk, &cork, &v6_cork, hlimit, tclass, opt, rt, fl6);
1752         if (err)
1753                 return ERR_PTR(err);
1754
1755         if (dontfrag < 0)
1756                 dontfrag = inet6_sk(sk)->dontfrag;
1757
1758         err = __ip6_append_data(sk, fl6, &queue, &cork.base, &v6_cork,
1759                                 &current->task_frag, getfrag, from,
1760                                 length + exthdrlen, transhdrlen + exthdrlen,
1761                                 flags, dontfrag);
1762         if (err) {
1763                 __ip6_flush_pending_frames(sk, &queue, &cork, &v6_cork);
1764                 return ERR_PTR(err);
1765         }
1766
1767         return __ip6_make_skb(sk, &queue, &cork, &v6_cork);
1768 }