git.oblomov.eu Git - linux-2.6/blob - net/ipv4/ip_output.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              The Internet Protocol (IP) output module.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Donald Becker, <becker@super.org>
  11  *              Alan Cox, <Alan.Cox@linux.org>
  12  *              Richard Underwood
  13  *              Stefan Becker, <stefanb@yello.ping.de>
  14  *              Jorge Cwik, <jorge@laser.satlink.net>
  15  *              Arnt Gulbrandsen, <agulbra@nvg.unit.no>
  16  *              Hirokazu Takahashi, <taka@valinux.co.jp>
  17  *
  18  *      See ip_input.c for original log
  19  *
  20  *      Fixes:
  21  *              Alan Cox        :       Missing nonblock feature in ip_build_xmit.
  22  *              Mike Kilburn    :       htons() missing in ip_build_xmit.
  23  *              Bradford Johnson:       Fix faulty handling of some frames when
  24  *                                      no route is found.
  25  *              Alexander Demenshin:    Missing sk/skb free in ip_queue_xmit
  26  *                                      (in case if packet not accepted by
  27  *                                      output firewall rules)
  28  *              Mike McLagan    :       Routing by source
  29  *              Alexey Kuznetsov:       use new route cache
  30  *              Andi Kleen:             Fix broken PMTU recovery and remove
  31  *                                      some redundant tests.
  32  *      Vitaly E. Lavrov        :       Transparent proxy revived after year coma.
  33  *              Andi Kleen      :       Replace ip_reply with ip_send_reply.
  34  *              Andi Kleen      :       Split fast and slow ip_build_xmit path
  35  *                                      for decreased register pressure on x86
  36  *                                      and more readibility.
  37  *              Marc Boucher    :       When call_out_firewall returns FW_QUEUE,
  38  *                                      silently drop skb instead of failing with -EPERM.
  39  *              Detlev Wengorz  :       Copy protocol for fragments.
  40  *              Hirokazu Takahashi:     HW checksumming for outgoing UDP
  41  *                                      datagrams.
  42  *              Hirokazu Takahashi:     sendfile() on UDP works now.
  43  */
  44
  45 #include <asm/uaccess.h>
  46 #include <asm/system.h>
  47 #include <linux/module.h>
  48 #include <linux/types.h>
  49 #include <linux/kernel.h>
  50 #include <linux/mm.h>
  51 #include <linux/string.h>
  52 #include <linux/errno.h>
  53 #include <linux/highmem.h>
  54
  55 #include <linux/socket.h>
  56 #include <linux/sockios.h>
  57 #include <linux/in.h>
  58 #include <linux/inet.h>
  59 #include <linux/netdevice.h>
  60 #include <linux/etherdevice.h>
  61 #include <linux/proc_fs.h>
  62 #include <linux/stat.h>
  63 #include <linux/init.h>
  64
  65 #include <net/snmp.h>
  66 #include <net/ip.h>
  67 #include <net/protocol.h>
  68 #include <net/route.h>
  69 #include <net/xfrm.h>
  70 #include <linux/skbuff.h>
  71 #include <net/sock.h>
  72 #include <net/arp.h>
  73 #include <net/icmp.h>
  74 #include <net/checksum.h>
  75 #include <net/inetpeer.h>
  76 #include <linux/igmp.h>
  77 #include <linux/netfilter_ipv4.h>
  78 #include <linux/netfilter_bridge.h>
  79 #include <linux/mroute.h>
  80 #include <linux/netlink.h>
  81 #include <linux/tcp.h>
  82
  83 int sysctl_ip_default_ttl __read_mostly = IPDEFTTL;
  84
  85 /* Generate a checksum for an outgoing IP datagram. */
  86 __inline__ void ip_send_check(struct iphdr *iph)
  87 {
  88         iph->check = 0;
  89         iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
  90 }
  91
  92 int __ip_local_out(struct sk_buff *skb)
  93 {
  94         struct iphdr *iph = ip_hdr(skb);
  95
  96         iph->tot_len = htons(skb->len);
  97         ip_send_check(iph);
  98         return nf_hook(PF_INET, NF_INET_LOCAL_OUT, skb, NULL, skb->dst->dev,
  99                        dst_output);
 100 }
 101
 102 int ip_local_out(struct sk_buff *skb)
 103 {
 104         int err;
 105
 106         err = __ip_local_out(skb);
 107         if (likely(err == 1))
 108                 err = dst_output(skb);
 109
 110         return err;
 111 }
 112 EXPORT_SYMBOL_GPL(ip_local_out);
 113
 114 /* dev_loopback_xmit for use with netfilter. */
 115 static int ip_dev_loopback_xmit(struct sk_buff *newskb)
 116 {
 117         skb_reset_mac_header(newskb);
 118         __skb_pull(newskb, skb_network_offset(newskb));
 119         newskb->pkt_type = PACKET_LOOPBACK;
 120         newskb->ip_summed = CHECKSUM_UNNECESSARY;
 121         WARN_ON(!newskb->dst);
 122         netif_rx(newskb);
 123         return 0;
 124 }
 125
 126 static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
 127 {
 128         int ttl = inet->uc_ttl;
 129
 130         if (ttl < 0)
 131                 ttl = dst_metric(dst, RTAX_HOPLIMIT);
 132         return ttl;
 133 }
 134
 135 /*
 136  *              Add an ip header to a skbuff and send it out.
 137  *
 138  */
 139 int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk,
 140                           __be32 saddr, __be32 daddr, struct ip_options *opt)
 141 {
 142         struct inet_sock *inet = inet_sk(sk);
 143         struct rtable *rt = skb->rtable;
 144         struct iphdr *iph;
 145
 146         /* Build the IP header. */
 147         skb_push(skb, sizeof(struct iphdr) + (opt ? opt->optlen : 0));
 148         skb_reset_network_header(skb);
 149         iph = ip_hdr(skb);
 150         iph->version  = 4;
 151         iph->ihl      = 5;
 152         iph->tos      = inet->tos;
 153         if (ip_dont_fragment(sk, &rt->u.dst))
 154                 iph->frag_off = htons(IP_DF);
 155         else
 156                 iph->frag_off = 0;
 157         iph->ttl      = ip_select_ttl(inet, &rt->u.dst);
 158         iph->daddr    = rt->rt_dst;
 159         iph->saddr    = rt->rt_src;
 160         iph->protocol = sk->sk_protocol;
 161         ip_select_ident(iph, &rt->u.dst, sk);
 162
 163         if (opt && opt->optlen) {
 164                 iph->ihl += opt->optlen>>2;
 165                 ip_options_build(skb, opt, daddr, rt, 0);
 166         }
 167
 168         skb->priority = sk->sk_priority;
 169         skb->mark = sk->sk_mark;
 170
 171         /* Send it out. */
 172         return ip_local_out(skb);
 173 }
 174
 175 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
 176
 177 static inline int ip_finish_output2(struct sk_buff *skb)
 178 {
 179         struct dst_entry *dst = skb->dst;
 180         struct rtable *rt = (struct rtable *)dst;
 181         struct net_device *dev = dst->dev;
 182         unsigned int hh_len = LL_RESERVED_SPACE(dev);
 183
 184         if (rt->rt_type == RTN_MULTICAST)
 185                 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_OUTMCASTPKTS);
 186         else if (rt->rt_type == RTN_BROADCAST)
 187                 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_OUTBCASTPKTS);
 188
 189         /* Be paranoid, rather than too clever. */
 190         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
 191                 struct sk_buff *skb2;
 192
 193                 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
 194                 if (skb2 == NULL) {
 195                         kfree_skb(skb);
 196                         return -ENOMEM;
 197                 }
 198                 if (skb->sk)
 199                         skb_set_owner_w(skb2, skb->sk);
 200                 kfree_skb(skb);
 201                 skb = skb2;
 202         }
 203
 204         if (dst->hh)
 205                 return neigh_hh_output(dst->hh, skb);
 206         else if (dst->neighbour)
 207                 return dst->neighbour->output(skb);
 208
 209         if (net_ratelimit())
 210                 printk(KERN_DEBUG "ip_finish_output2: No header cache and no neighbour!\n");
 211         kfree_skb(skb);
 212         return -EINVAL;
 213 }
 214
 215 static inline int ip_skb_dst_mtu(struct sk_buff *skb)
 216 {
 217         struct inet_sock *inet = skb->sk ? inet_sk(skb->sk) : NULL;
 218
 219         return (inet && inet->pmtudisc == IP_PMTUDISC_PROBE) ?
 220                skb->dst->dev->mtu : dst_mtu(skb->dst);
 221 }
 222
 223 static int ip_finish_output(struct sk_buff *skb)
 224 {
 225 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
 226         /* Policy lookup after SNAT yielded a new policy */
 227         if (skb->dst->xfrm != NULL) {
 228                 IPCB(skb)->flags |= IPSKB_REROUTED;
 229                 return dst_output(skb);
 230         }
 231 #endif
 232         if (skb->len > ip_skb_dst_mtu(skb) && !skb_is_gso(skb))
 233                 return ip_fragment(skb, ip_finish_output2);
 234         else
 235                 return ip_finish_output2(skb);
 236 }
 237
 238 int ip_mc_output(struct sk_buff *skb)
 239 {
 240         struct sock *sk = skb->sk;
 241         struct rtable *rt = skb->rtable;
 242         struct net_device *dev = rt->u.dst.dev;
 243
 244         /*
 245          *      If the indicated interface is up and running, send the packet.
 246          */
 247         IP_INC_STATS(dev_net(dev), IPSTATS_MIB_OUTREQUESTS);
 248
 249         skb->dev = dev;
 250         skb->protocol = htons(ETH_P_IP);
 251
 252         /*
 253          *      Multicasts are looped back for other local users
 254          */
 255
 256         if (rt->rt_flags&RTCF_MULTICAST) {
 257                 if ((!sk || inet_sk(sk)->mc_loop)
 258 #ifdef CONFIG_IP_MROUTE
 259                 /* Small optimization: do not loopback not local frames,
 260                    which returned after forwarding; they will be  dropped
 261                    by ip_mr_input in any case.
 262                    Note, that local frames are looped back to be delivered
 263                    to local recipients.
 264
 265                    This check is duplicated in ip_mr_input at the moment.
 266                  */
 267                     && ((rt->rt_flags&RTCF_LOCAL) || !(IPCB(skb)->flags&IPSKB_FORWARDED))
 268 #endif
 269                 ) {
 270                         struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
 271                         if (newskb)
 272                                 NF_HOOK(PF_INET, NF_INET_POST_ROUTING, newskb,
 273                                         NULL, newskb->dev,
 274                                         ip_dev_loopback_xmit);
 275                 }
 276
 277                 /* Multicasts with ttl 0 must not go beyond the host */
 278
 279                 if (ip_hdr(skb)->ttl == 0) {
 280                         kfree_skb(skb);
 281                         return 0;
 282                 }
 283         }
 284
 285         if (rt->rt_flags&RTCF_BROADCAST) {
 286                 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
 287                 if (newskb)
 288                         NF_HOOK(PF_INET, NF_INET_POST_ROUTING, newskb, NULL,
 289                                 newskb->dev, ip_dev_loopback_xmit);
 290         }
 291
 292         return NF_HOOK_COND(PF_INET, NF_INET_POST_ROUTING, skb, NULL, skb->dev,
 293                             ip_finish_output,
 294                             !(IPCB(skb)->flags & IPSKB_REROUTED));
 295 }
 296
 297 int ip_output(struct sk_buff *skb)
 298 {
 299         struct net_device *dev = skb->dst->dev;
 300
 301         IP_INC_STATS(dev_net(dev), IPSTATS_MIB_OUTREQUESTS);
 302
 303         skb->dev = dev;
 304         skb->protocol = htons(ETH_P_IP);
 305
 306         return NF_HOOK_COND(PF_INET, NF_INET_POST_ROUTING, skb, NULL, dev,
 307                             ip_finish_output,
 308                             !(IPCB(skb)->flags & IPSKB_REROUTED));
 309 }
 310
 311 int ip_queue_xmit(struct sk_buff *skb, int ipfragok)
 312 {
 313         struct sock *sk = skb->sk;
 314         struct inet_sock *inet = inet_sk(sk);
 315         struct ip_options *opt = inet->opt;
 316         struct rtable *rt;
 317         struct iphdr *iph;
 318
 319         /* Skip all of this if the packet is already routed,
 320          * f.e. by something like SCTP.
 321          */
 322         rt = skb->rtable;
 323         if (rt != NULL)
 324                 goto packet_routed;
 325
 326         /* Make sure we can route this packet. */
 327         rt = (struct rtable *)__sk_dst_check(sk, 0);
 328         if (rt == NULL) {
 329                 __be32 daddr;
 330
 331                 /* Use correct destination address if we have options. */
 332                 daddr = inet->daddr;
 333                 if(opt && opt->srr)
 334                         daddr = opt->faddr;
 335
 336                 {
 337                         struct flowi fl = { .oif = sk->sk_bound_dev_if,
 338                                             .nl_u = { .ip4_u =
 339                                                       { .daddr = daddr,
 340                                                         .saddr = inet->saddr,
 341                                                         .tos = RT_CONN_FLAGS(sk) } },
 342                                             .proto = sk->sk_protocol,
 343                                             .flags = inet_sk_flowi_flags(sk),
 344                                             .uli_u = { .ports =
 345                                                        { .sport = inet->sport,
 346                                                          .dport = inet->dport } } };
 347
 348                         /* If this fails, retransmit mechanism of transport layer will
 349                          * keep trying until route appears or the connection times
 350                          * itself out.
 351                          */
 352                         security_sk_classify_flow(sk, &fl);
 353                         if (ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0))
 354                                 goto no_route;
 355                 }
 356                 sk_setup_caps(sk, &rt->u.dst);
 357         }
 358         skb->dst = dst_clone(&rt->u.dst);
 359
 360 packet_routed:
 361         if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
 362                 goto no_route;
 363
 364         /* OK, we know where to send it, allocate and build IP header. */
 365         skb_push(skb, sizeof(struct iphdr) + (opt ? opt->optlen : 0));
 366         skb_reset_network_header(skb);
 367         iph = ip_hdr(skb);
 368         *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
 369         if (ip_dont_fragment(sk, &rt->u.dst) && !ipfragok)
 370                 iph->frag_off = htons(IP_DF);
 371         else
 372                 iph->frag_off = 0;
 373         iph->ttl      = ip_select_ttl(inet, &rt->u.dst);
 374         iph->protocol = sk->sk_protocol;
 375         iph->saddr    = rt->rt_src;
 376         iph->daddr    = rt->rt_dst;
 377         /* Transport layer set skb->h.foo itself. */
 378
 379         if (opt && opt->optlen) {
 380                 iph->ihl += opt->optlen >> 2;
 381                 ip_options_build(skb, opt, inet->daddr, rt, 0);
 382         }
 383
 384         ip_select_ident_more(iph, &rt->u.dst, sk,
 385                              (skb_shinfo(skb)->gso_segs ?: 1) - 1);
 386
 387         skb->priority = sk->sk_priority;
 388         skb->mark = sk->sk_mark;
 389
 390         return ip_local_out(skb);
 391
 392 no_route:
 393         IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
 394         kfree_skb(skb);
 395         return -EHOSTUNREACH;
 396 }
 397
 398
 399 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
 400 {
 401         to->pkt_type = from->pkt_type;
 402         to->priority = from->priority;
 403         to->protocol = from->protocol;
 404         dst_release(to->dst);
 405         to->dst = dst_clone(from->dst);
 406         to->dev = from->dev;
 407         to->mark = from->mark;
 408
 409         /* Copy the flags to each fragment. */
 410         IPCB(to)->flags = IPCB(from)->flags;
 411
 412 #ifdef CONFIG_NET_SCHED
 413         to->tc_index = from->tc_index;
 414 #endif
 415         nf_copy(to, from);
 416 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
 417     defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
 418         to->nf_trace = from->nf_trace;
 419 #endif
 420 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
 421         to->ipvs_property = from->ipvs_property;
 422 #endif
 423         skb_copy_secmark(to, from);
 424 }
 425
 426 /*
 427  *      This IP datagram is too large to be sent in one piece.  Break it up into
 428  *      smaller pieces (each of size equal to IP header plus
 429  *      a block of the data of the original IP data part) that will yet fit in a
 430  *      single device frame, and queue such a frame for sending.
 431  */
 432
 433 int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff*))
 434 {
 435         struct iphdr *iph;
 436         int raw = 0;
 437         int ptr;
 438         struct net_device *dev;
 439         struct sk_buff *skb2;
 440         unsigned int mtu, hlen, left, len, ll_rs, pad;
 441         int offset;
 442         __be16 not_last_frag;
 443         struct rtable *rt = skb->rtable;
 444         int err = 0;
 445
 446         dev = rt->u.dst.dev;
 447
 448         /*
 449          *      Point into the IP datagram header.
 450          */
 451
 452         iph = ip_hdr(skb);
 453
 454         if (unlikely((iph->frag_off & htons(IP_DF)) && !skb->local_df)) {
 455                 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
 456                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
 457                           htonl(ip_skb_dst_mtu(skb)));
 458                 kfree_skb(skb);
 459                 return -EMSGSIZE;
 460         }
 461
 462         /*
 463          *      Setup starting values.
 464          */
 465
 466         hlen = iph->ihl * 4;
 467         mtu = dst_mtu(&rt->u.dst) - hlen;       /* Size of data space */
 468         IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
 469
 470         /* When frag_list is given, use it. First, check its validity:
 471          * some transformers could create wrong frag_list or break existing
 472          * one, it is not prohibited. In this case fall back to copying.
 473          *
 474          * LATER: this step can be merged to real generation of fragments,
 475          * we can switch to copy when see the first bad fragment.
 476          */
 477         if (skb_shinfo(skb)->frag_list) {
 478                 struct sk_buff *frag;
 479                 int first_len = skb_pagelen(skb);
 480                 int truesizes = 0;
 481
 482                 if (first_len - hlen > mtu ||
 483                     ((first_len - hlen) & 7) ||
 484                     (iph->frag_off & htons(IP_MF|IP_OFFSET)) ||
 485                     skb_cloned(skb))
 486                         goto slow_path;
 487
 488                 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
 489                         /* Correct geometry. */
 490                         if (frag->len > mtu ||
 491                             ((frag->len & 7) && frag->next) ||
 492                             skb_headroom(frag) < hlen)
 493                             goto slow_path;
 494
 495                         /* Partially cloned skb? */
 496                         if (skb_shared(frag))
 497                                 goto slow_path;
 498
 499                         BUG_ON(frag->sk);
 500                         if (skb->sk) {
 501                                 sock_hold(skb->sk);
 502                                 frag->sk = skb->sk;
 503                                 frag->destructor = sock_wfree;
 504                                 truesizes += frag->truesize;
 505                         }
 506                 }
 507
 508                 /* Everything is OK. Generate! */
 509
 510                 err = 0;
 511                 offset = 0;
 512                 frag = skb_shinfo(skb)->frag_list;
 513                 skb_shinfo(skb)->frag_list = NULL;
 514                 skb->data_len = first_len - skb_headlen(skb);
 515                 skb->truesize -= truesizes;
 516                 skb->len = first_len;
 517                 iph->tot_len = htons(first_len);
 518                 iph->frag_off = htons(IP_MF);
 519                 ip_send_check(iph);
 520
 521                 for (;;) {
 522                         /* Prepare header of the next frame,
 523                          * before previous one went down. */
 524                         if (frag) {
 525                                 frag->ip_summed = CHECKSUM_NONE;
 526                                 skb_reset_transport_header(frag);
 527                                 __skb_push(frag, hlen);
 528                                 skb_reset_network_header(frag);
 529                                 memcpy(skb_network_header(frag), iph, hlen);
 530                                 iph = ip_hdr(frag);
 531                                 iph->tot_len = htons(frag->len);
 532                                 ip_copy_metadata(frag, skb);
 533                                 if (offset == 0)
 534                                         ip_options_fragment(frag);
 535                                 offset += skb->len - hlen;
 536                                 iph->frag_off = htons(offset>>3);
 537                                 if (frag->next != NULL)
 538                                         iph->frag_off |= htons(IP_MF);
 539                                 /* Ready, complete checksum */
 540                                 ip_send_check(iph);
 541                         }
 542
 543                         err = output(skb);
 544
 545                         if (!err)
 546                                 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
 547                         if (err || !frag)
 548                                 break;
 549
 550                         skb = frag;
 551                         frag = skb->next;
 552                         skb->next = NULL;
 553                 }
 554
 555                 if (err == 0) {
 556                         IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
 557                         return 0;
 558                 }
 559
 560                 while (frag) {
 561                         skb = frag->next;
 562                         kfree_skb(frag);
 563                         frag = skb;
 564                 }
 565                 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
 566                 return err;
 567         }
 568
 569 slow_path:
 570         left = skb->len - hlen;         /* Space per frame */
 571         ptr = raw + hlen;               /* Where to start from */
 572
 573         /* for bridged IP traffic encapsulated inside f.e. a vlan header,
 574          * we need to make room for the encapsulating header
 575          */
 576         pad = nf_bridge_pad(skb);
 577         ll_rs = LL_RESERVED_SPACE_EXTRA(rt->u.dst.dev, pad);
 578         mtu -= pad;
 579
 580         /*
 581          *      Fragment the datagram.
 582          */
 583
 584         offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
 585         not_last_frag = iph->frag_off & htons(IP_MF);
 586
 587         /*
 588          *      Keep copying data until we run out.
 589          */
 590
 591         while (left > 0) {
 592                 len = left;
 593                 /* IF: it doesn't fit, use 'mtu' - the data space left */
 594                 if (len > mtu)
 595                         len = mtu;
 596                 /* IF: we are not sending upto and including the packet end
 597                    then align the next start on an eight byte boundary */
 598                 if (len < left) {
 599                         len &= ~7;
 600                 }
 601                 /*
 602                  *      Allocate buffer.
 603                  */
 604
 605                 if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) {
 606                         NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n");
 607                         err = -ENOMEM;
 608                         goto fail;
 609                 }
 610
 611                 /*
 612                  *      Set up data on packet
 613                  */
 614
 615                 ip_copy_metadata(skb2, skb);
 616                 skb_reserve(skb2, ll_rs);
 617                 skb_put(skb2, len + hlen);
 618                 skb_reset_network_header(skb2);
 619                 skb2->transport_header = skb2->network_header + hlen;
 620
 621                 /*
 622                  *      Charge the memory for the fragment to any owner
 623                  *      it might possess
 624                  */
 625
 626                 if (skb->sk)
 627                         skb_set_owner_w(skb2, skb->sk);
 628
 629                 /*
 630                  *      Copy the packet header into the new buffer.
 631                  */
 632
 633                 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
 634
 635                 /*
 636                  *      Copy a block of the IP datagram.
 637                  */
 638                 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
 639                         BUG();
 640                 left -= len;
 641
 642                 /*
 643                  *      Fill in the new header fields.
 644                  */
 645                 iph = ip_hdr(skb2);
 646                 iph->frag_off = htons((offset >> 3));
 647
 648                 /* ANK: dirty, but effective trick. Upgrade options only if
 649                  * the segment to be fragmented was THE FIRST (otherwise,
 650                  * options are already fixed) and make it ONCE
 651                  * on the initial skb, so that all the following fragments
 652                  * will inherit fixed options.
 653                  */
 654                 if (offset == 0)
 655                         ip_options_fragment(skb);
 656
 657                 /*
 658                  *      Added AC : If we are fragmenting a fragment that's not the
 659                  *                 last fragment then keep MF on each bit
 660                  */
 661                 if (left > 0 || not_last_frag)
 662                         iph->frag_off |= htons(IP_MF);
 663                 ptr += len;
 664                 offset += len;
 665
 666                 /*
 667                  *      Put this fragment into the sending queue.
 668                  */
 669                 iph->tot_len = htons(len + hlen);
 670
 671                 ip_send_check(iph);
 672
 673                 err = output(skb2);
 674                 if (err)
 675                         goto fail;
 676
 677                 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
 678         }
 679         kfree_skb(skb);
 680         IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
 681         return err;
 682
 683 fail:
 684         kfree_skb(skb);
 685         IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
 686         return err;
 687 }
 688
 689 EXPORT_SYMBOL(ip_fragment);
 690
 691 int
 692 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
 693 {
 694         struct iovec *iov = from;
 695
 696         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 697                 if (memcpy_fromiovecend(to, iov, offset, len) < 0)
 698                         return -EFAULT;
 699         } else {
 700                 __wsum csum = 0;
 701                 if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0)
 702                         return -EFAULT;
 703                 skb->csum = csum_block_add(skb->csum, csum, odd);
 704         }
 705         return 0;
 706 }
 707
 708 static inline __wsum
 709 csum_page(struct page *page, int offset, int copy)
 710 {
 711         char *kaddr;
 712         __wsum csum;
 713         kaddr = kmap(page);
 714         csum = csum_partial(kaddr + offset, copy, 0);
 715         kunmap(page);
 716         return csum;
 717 }
 718
 719 static inline int ip_ufo_append_data(struct sock *sk,
 720                         int getfrag(void *from, char *to, int offset, int len,
 721                                int odd, struct sk_buff *skb),
 722                         void *from, int length, int hh_len, int fragheaderlen,
 723                         int transhdrlen, int mtu,unsigned int flags)
 724 {
 725         struct sk_buff *skb;
 726         int err;
 727
 728         /* There is support for UDP fragmentation offload by network
 729          * device, so create one single skb packet containing complete
 730          * udp datagram
 731          */
 732         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
 733                 skb = sock_alloc_send_skb(sk,
 734                         hh_len + fragheaderlen + transhdrlen + 20,
 735                         (flags & MSG_DONTWAIT), &err);
 736
 737                 if (skb == NULL)
 738                         return err;
 739
 740                 /* reserve space for Hardware header */
 741                 skb_reserve(skb, hh_len);
 742
 743                 /* create space for UDP/IP header */
 744                 skb_put(skb,fragheaderlen + transhdrlen);
 745
 746                 /* initialize network header pointer */
 747                 skb_reset_network_header(skb);
 748
 749                 /* initialize protocol header pointer */
 750                 skb->transport_header = skb->network_header + fragheaderlen;
 751
 752                 skb->ip_summed = CHECKSUM_PARTIAL;
 753                 skb->csum = 0;
 754                 sk->sk_sndmsg_off = 0;
 755
 756                 /* specify the length of each IP datagram fragment */
 757                 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
 758                 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
 759                 __skb_queue_tail(&sk->sk_write_queue, skb);
 760         }
 761
 762         return skb_append_datato_frags(sk, skb, getfrag, from,
 763                                        (length - transhdrlen));
 764 }
 765
 766 /*
 767  *      ip_append_data() and ip_append_page() can make one large IP datagram
 768  *      from many pieces of data. Each pieces will be holded on the socket
 769  *      until ip_push_pending_frames() is called. Each piece can be a page
 770  *      or non-page data.
 771  *
 772  *      Not only UDP, other transport protocols - e.g. raw sockets - can use
 773  *      this interface potentially.
 774  *
 775  *      LATER: length must be adjusted by pad at tail, when it is required.
 776  */
 777 int ip_append_data(struct sock *sk,
 778                    int getfrag(void *from, char *to, int offset, int len,
 779                                int odd, struct sk_buff *skb),
 780                    void *from, int length, int transhdrlen,
 781                    struct ipcm_cookie *ipc, struct rtable *rt,
 782                    unsigned int flags)
 783 {
 784         struct inet_sock *inet = inet_sk(sk);
 785         struct sk_buff *skb;
 786
 787         struct ip_options *opt = NULL;
 788         int hh_len;
 789         int exthdrlen;
 790         int mtu;
 791         int copy;
 792         int err;
 793         int offset = 0;
 794         unsigned int maxfraglen, fragheaderlen;
 795         int csummode = CHECKSUM_NONE;
 796
 797         if (flags&MSG_PROBE)
 798                 return 0;
 799
 800         if (skb_queue_empty(&sk->sk_write_queue)) {
 801                 /*
 802                  * setup for corking.
 803                  */
 804                 opt = ipc->opt;
 805                 if (opt) {
 806                         if (inet->cork.opt == NULL) {
 807                                 inet->cork.opt = kmalloc(sizeof(struct ip_options) + 40, sk->sk_allocation);
 808                                 if (unlikely(inet->cork.opt == NULL))
 809                                         return -ENOBUFS;
 810                         }
 811                         memcpy(inet->cork.opt, opt, sizeof(struct ip_options)+opt->optlen);
 812                         inet->cork.flags |= IPCORK_OPT;
 813                         inet->cork.addr = ipc->addr;
 814                 }
 815                 dst_hold(&rt->u.dst);
 816                 inet->cork.fragsize = mtu = inet->pmtudisc == IP_PMTUDISC_PROBE ?
 817                                             rt->u.dst.dev->mtu :
 818                                             dst_mtu(rt->u.dst.path);
 819                 inet->cork.dst = &rt->u.dst;
 820                 inet->cork.length = 0;
 821                 sk->sk_sndmsg_page = NULL;
 822                 sk->sk_sndmsg_off = 0;
 823                 if ((exthdrlen = rt->u.dst.header_len) != 0) {
 824                         length += exthdrlen;
 825                         transhdrlen += exthdrlen;
 826                 }
 827         } else {
 828                 rt = (struct rtable *)inet->cork.dst;
 829                 if (inet->cork.flags & IPCORK_OPT)
 830                         opt = inet->cork.opt;
 831
 832                 transhdrlen = 0;
 833                 exthdrlen = 0;
 834                 mtu = inet->cork.fragsize;
 835         }
 836         hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
 837
 838         fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
 839         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
 840
 841         if (inet->cork.length + length > 0xFFFF - fragheaderlen) {
 842                 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu-exthdrlen);
 843                 return -EMSGSIZE;
 844         }
 845
 846         /*
 847          * transhdrlen > 0 means that this is the first fragment and we wish
 848          * it won't be fragmented in the future.
 849          */
 850         if (transhdrlen &&
 851             length + fragheaderlen <= mtu &&
 852             rt->u.dst.dev->features & NETIF_F_V4_CSUM &&
 853             !exthdrlen)
 854                 csummode = CHECKSUM_PARTIAL;
 855
 856         inet->cork.length += length;
 857         if (((length> mtu) || !skb_queue_empty(&sk->sk_write_queue)) &&
 858             (sk->sk_protocol == IPPROTO_UDP) &&
 859             (rt->u.dst.dev->features & NETIF_F_UFO)) {
 860                 err = ip_ufo_append_data(sk, getfrag, from, length, hh_len,
 861                                          fragheaderlen, transhdrlen, mtu,
 862                                          flags);
 863                 if (err)
 864                         goto error;
 865                 return 0;
 866         }
 867
 868         /* So, what's going on in the loop below?
 869          *
 870          * We use calculated fragment length to generate chained skb,
 871          * each of segments is IP fragment ready for sending to network after
 872          * adding appropriate IP header.
 873          */
 874
 875         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
 876                 goto alloc_new_skb;
 877
 878         while (length > 0) {
 879                 /* Check if the remaining data fits into current packet. */
 880                 copy = mtu - skb->len;
 881                 if (copy < length)
 882                         copy = maxfraglen - skb->len;
 883                 if (copy <= 0) {
 884                         char *data;
 885                         unsigned int datalen;
 886                         unsigned int fraglen;
 887                         unsigned int fraggap;
 888                         unsigned int alloclen;
 889                         struct sk_buff *skb_prev;
 890 alloc_new_skb:
 891                         skb_prev = skb;
 892                         if (skb_prev)
 893                                 fraggap = skb_prev->len - maxfraglen;
 894                         else
 895                                 fraggap = 0;
 896
 897                         /*
 898                          * If remaining data exceeds the mtu,
 899                          * we know we need more fragment(s).
 900                          */
 901                         datalen = length + fraggap;
 902                         if (datalen > mtu - fragheaderlen)
 903                                 datalen = maxfraglen - fragheaderlen;
 904                         fraglen = datalen + fragheaderlen;
 905
 906                         if ((flags & MSG_MORE) &&
 907                             !(rt->u.dst.dev->features&NETIF_F_SG))
 908                                 alloclen = mtu;
 909                         else
 910                                 alloclen = datalen + fragheaderlen;
 911
 912                         /* The last fragment gets additional space at tail.
 913                          * Note, with MSG_MORE we overallocate on fragments,
 914                          * because we have no idea what fragment will be
 915                          * the last.
 916                          */
 917                         if (datalen == length + fraggap)
 918                                 alloclen += rt->u.dst.trailer_len;
 919
 920                         if (transhdrlen) {
 921                                 skb = sock_alloc_send_skb(sk,
 922                                                 alloclen + hh_len + 15,
 923                                                 (flags & MSG_DONTWAIT), &err);
 924                         } else {
 925                                 skb = NULL;
 926                                 if (atomic_read(&sk->sk_wmem_alloc) <=
 927                                     2 * sk->sk_sndbuf)
 928                                         skb = sock_wmalloc(sk,
 929                                                            alloclen + hh_len + 15, 1,
 930                                                            sk->sk_allocation);
 931                                 if (unlikely(skb == NULL))
 932                                         err = -ENOBUFS;
 933                         }
 934                         if (skb == NULL)
 935                                 goto error;
 936
 937                         /*
 938                          *      Fill in the control structures
 939                          */
 940                         skb->ip_summed = csummode;
 941                         skb->csum = 0;
 942                         skb_reserve(skb, hh_len);
 943
 944                         /*
 945                          *      Find where to start putting bytes.
 946                          */
 947                         data = skb_put(skb, fraglen);
 948                         skb_set_network_header(skb, exthdrlen);
 949                         skb->transport_header = (skb->network_header +
 950                                                  fragheaderlen);
 951                         data += fragheaderlen;
 952
 953                         if (fraggap) {
 954                                 skb->csum = skb_copy_and_csum_bits(
 955                                         skb_prev, maxfraglen,
 956                                         data + transhdrlen, fraggap, 0);
 957                                 skb_prev->csum = csum_sub(skb_prev->csum,
 958                                                           skb->csum);
 959                                 data += fraggap;
 960                                 pskb_trim_unique(skb_prev, maxfraglen);
 961                         }
 962
 963                         copy = datalen - transhdrlen - fraggap;
 964                         if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
 965                                 err = -EFAULT;
 966                                 kfree_skb(skb);
 967                                 goto error;
 968                         }
 969
 970                         offset += copy;
 971                         length -= datalen - fraggap;
 972                         transhdrlen = 0;
 973                         exthdrlen = 0;
 974                         csummode = CHECKSUM_NONE;
 975
 976                         /*
 977                          * Put the packet on the pending queue.
 978                          */
 979                         __skb_queue_tail(&sk->sk_write_queue, skb);
 980                         continue;
 981                 }
 982
 983                 if (copy > length)
 984                         copy = length;
 985
 986                 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
 987                         unsigned int off;
 988
 989                         off = skb->len;
 990                         if (getfrag(from, skb_put(skb, copy),
 991                                         offset, copy, off, skb) < 0) {
 992                                 __skb_trim(skb, off);
 993                                 err = -EFAULT;
 994                                 goto error;
 995                         }
 996                 } else {
 997                         int i = skb_shinfo(skb)->nr_frags;
 998                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
 999                         struct page *page = sk->sk_sndmsg_page;
1000                         int off = sk->sk_sndmsg_off;
1001                         unsigned int left;
1002
1003                         if (page && (left = PAGE_SIZE - off) > 0) {
1004                                 if (copy >= left)
1005                                         copy = left;
1006                                 if (page != frag->page) {
1007                                         if (i == MAX_SKB_FRAGS) {
1008                                                 err = -EMSGSIZE;
1009                                                 goto error;
1010                                         }
1011                                         get_page(page);
1012                                         skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1013                                         frag = &skb_shinfo(skb)->frags[i];
1014                                 }
1015                         } else if (i < MAX_SKB_FRAGS) {
1016                                 if (copy > PAGE_SIZE)
1017                                         copy = PAGE_SIZE;
1018                                 page = alloc_pages(sk->sk_allocation, 0);
1019                                 if (page == NULL)  {
1020                                         err = -ENOMEM;
1021                                         goto error;
1022                                 }
1023                                 sk->sk_sndmsg_page = page;
1024                                 sk->sk_sndmsg_off = 0;
1025
1026                                 skb_fill_page_desc(skb, i, page, 0, 0);
1027                                 frag = &skb_shinfo(skb)->frags[i];
1028                         } else {
1029                                 err = -EMSGSIZE;
1030                                 goto error;
1031                         }
1032                         if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1033                                 err = -EFAULT;
1034                                 goto error;
1035                         }
1036                         sk->sk_sndmsg_off += copy;
1037                         frag->size += copy;
1038                         skb->len += copy;
1039                         skb->data_len += copy;
1040                         skb->truesize += copy;
1041                         atomic_add(copy, &sk->sk_wmem_alloc);
1042                 }
1043                 offset += copy;
1044                 length -= copy;
1045         }
1046
1047         return 0;
1048
1049 error:
1050         inet->cork.length -= length;
1051         IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1052         return err;
1053 }
1054
1055 ssize_t ip_append_page(struct sock *sk, struct page *page,
1056                        int offset, size_t size, int flags)
1057 {
1058         struct inet_sock *inet = inet_sk(sk);
1059         struct sk_buff *skb;
1060         struct rtable *rt;
1061         struct ip_options *opt = NULL;
1062         int hh_len;
1063         int mtu;
1064         int len;
1065         int err;
1066         unsigned int maxfraglen, fragheaderlen, fraggap;
1067
1068         if (inet->hdrincl)
1069                 return -EPERM;
1070
1071         if (flags&MSG_PROBE)
1072                 return 0;
1073
1074         if (skb_queue_empty(&sk->sk_write_queue))
1075                 return -EINVAL;
1076
1077         rt = (struct rtable *)inet->cork.dst;
1078         if (inet->cork.flags & IPCORK_OPT)
1079                 opt = inet->cork.opt;
1080
1081         if (!(rt->u.dst.dev->features&NETIF_F_SG))
1082                 return -EOPNOTSUPP;
1083
1084         hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1085         mtu = inet->cork.fragsize;
1086
1087         fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1088         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1089
1090         if (inet->cork.length + size > 0xFFFF - fragheaderlen) {
1091                 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu);
1092                 return -EMSGSIZE;
1093         }
1094
1095         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1096                 return -EINVAL;
1097
1098         inet->cork.length += size;
1099         if ((sk->sk_protocol == IPPROTO_UDP) &&
1100             (rt->u.dst.dev->features & NETIF_F_UFO)) {
1101                 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
1102                 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1103         }
1104
1105
1106         while (size > 0) {
1107                 int i;
1108
1109                 if (skb_is_gso(skb))
1110                         len = size;
1111                 else {
1112
1113                         /* Check if the remaining data fits into current packet. */
1114                         len = mtu - skb->len;
1115                         if (len < size)
1116                                 len = maxfraglen - skb->len;
1117                 }
1118                 if (len <= 0) {
1119                         struct sk_buff *skb_prev;
1120                         int alloclen;
1121
1122                         skb_prev = skb;
1123                         fraggap = skb_prev->len - maxfraglen;
1124
1125                         alloclen = fragheaderlen + hh_len + fraggap + 15;
1126                         skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1127                         if (unlikely(!skb)) {
1128                                 err = -ENOBUFS;
1129                                 goto error;
1130                         }
1131
1132                         /*
1133                          *      Fill in the control structures
1134                          */
1135                         skb->ip_summed = CHECKSUM_NONE;
1136                         skb->csum = 0;
1137                         skb_reserve(skb, hh_len);
1138
1139                         /*
1140                          *      Find where to start putting bytes.
1141                          */
1142                         skb_put(skb, fragheaderlen + fraggap);
1143                         skb_reset_network_header(skb);
1144                         skb->transport_header = (skb->network_header +
1145                                                  fragheaderlen);
1146                         if (fraggap) {
1147                                 skb->csum = skb_copy_and_csum_bits(skb_prev,
1148                                                                    maxfraglen,
1149                                                     skb_transport_header(skb),
1150                                                                    fraggap, 0);
1151                                 skb_prev->csum = csum_sub(skb_prev->csum,
1152                                                           skb->csum);
1153                                 pskb_trim_unique(skb_prev, maxfraglen);
1154                         }
1155
1156                         /*
1157                          * Put the packet on the pending queue.
1158                          */
1159                         __skb_queue_tail(&sk->sk_write_queue, skb);
1160                         continue;
1161                 }
1162
1163                 i = skb_shinfo(skb)->nr_frags;
1164                 if (len > size)
1165                         len = size;
1166                 if (skb_can_coalesce(skb, i, page, offset)) {
1167                         skb_shinfo(skb)->frags[i-1].size += len;
1168                 } else if (i < MAX_SKB_FRAGS) {
1169                         get_page(page);
1170                         skb_fill_page_desc(skb, i, page, offset, len);
1171                 } else {
1172                         err = -EMSGSIZE;
1173                         goto error;
1174                 }
1175
1176                 if (skb->ip_summed == CHECKSUM_NONE) {
1177                         __wsum csum;
1178                         csum = csum_page(page, offset, len);
1179                         skb->csum = csum_block_add(skb->csum, csum, skb->len);
1180                 }
1181
1182                 skb->len += len;
1183                 skb->data_len += len;
1184                 skb->truesize += len;
1185                 atomic_add(len, &sk->sk_wmem_alloc);
1186                 offset += len;
1187                 size -= len;
1188         }
1189         return 0;
1190
1191 error:
1192         inet->cork.length -= size;
1193         IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1194         return err;
1195 }
1196
1197 static void ip_cork_release(struct inet_sock *inet)
1198 {
1199         inet->cork.flags &= ~IPCORK_OPT;
1200         kfree(inet->cork.opt);
1201         inet->cork.opt = NULL;
1202         dst_release(inet->cork.dst);
1203         inet->cork.dst = NULL;
1204 }
1205
1206 /*
1207  *      Combined all pending IP fragments on the socket as one IP datagram
1208  *      and push them out.
1209  */
1210 int ip_push_pending_frames(struct sock *sk)
1211 {
1212         struct sk_buff *skb, *tmp_skb;
1213         struct sk_buff **tail_skb;
1214         struct inet_sock *inet = inet_sk(sk);
1215         struct net *net = sock_net(sk);
1216         struct ip_options *opt = NULL;
1217         struct rtable *rt = (struct rtable *)inet->cork.dst;
1218         struct iphdr *iph;
1219         __be16 df = 0;
1220         __u8 ttl;
1221         int err = 0;
1222
1223         if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1224                 goto out;
1225         tail_skb = &(skb_shinfo(skb)->frag_list);
1226
1227         /* move skb->data to ip header from ext header */
1228         if (skb->data < skb_network_header(skb))
1229                 __skb_pull(skb, skb_network_offset(skb));
1230         while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1231                 __skb_pull(tmp_skb, skb_network_header_len(skb));
1232                 *tail_skb = tmp_skb;
1233                 tail_skb = &(tmp_skb->next);
1234                 skb->len += tmp_skb->len;
1235                 skb->data_len += tmp_skb->len;
1236                 skb->truesize += tmp_skb->truesize;
1237                 __sock_put(tmp_skb->sk);
1238                 tmp_skb->destructor = NULL;
1239                 tmp_skb->sk = NULL;
1240         }
1241
1242         /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1243          * to fragment the frame generated here. No matter, what transforms
1244          * how transforms change size of the packet, it will come out.
1245          */
1246         if (inet->pmtudisc < IP_PMTUDISC_DO)
1247                 skb->local_df = 1;
1248
1249         /* DF bit is set when we want to see DF on outgoing frames.
1250          * If local_df is set too, we still allow to fragment this frame
1251          * locally. */
1252         if (inet->pmtudisc >= IP_PMTUDISC_DO ||
1253             (skb->len <= dst_mtu(&rt->u.dst) &&
1254              ip_dont_fragment(sk, &rt->u.dst)))
1255                 df = htons(IP_DF);
1256
1257         if (inet->cork.flags & IPCORK_OPT)
1258                 opt = inet->cork.opt;
1259
1260         if (rt->rt_type == RTN_MULTICAST)
1261                 ttl = inet->mc_ttl;
1262         else
1263                 ttl = ip_select_ttl(inet, &rt->u.dst);
1264
1265         iph = (struct iphdr *)skb->data;
1266         iph->version = 4;
1267         iph->ihl = 5;
1268         if (opt) {
1269                 iph->ihl += opt->optlen>>2;
1270                 ip_options_build(skb, opt, inet->cork.addr, rt, 0);
1271         }
1272         iph->tos = inet->tos;
1273         iph->frag_off = df;
1274         ip_select_ident(iph, &rt->u.dst, sk);
1275         iph->ttl = ttl;
1276         iph->protocol = sk->sk_protocol;
1277         iph->saddr = rt->rt_src;
1278         iph->daddr = rt->rt_dst;
1279
1280         skb->priority = sk->sk_priority;
1281         skb->mark = sk->sk_mark;
1282         skb->dst = dst_clone(&rt->u.dst);
1283
1284         if (iph->protocol == IPPROTO_ICMP)
1285                 icmp_out_count(net, ((struct icmphdr *)
1286                         skb_transport_header(skb))->type);
1287
1288         /* Netfilter gets whole the not fragmented skb. */
1289         err = ip_local_out(skb);
1290         if (err) {
1291                 if (err > 0)
1292                         err = inet->recverr ? net_xmit_errno(err) : 0;
1293                 if (err)
1294                         goto error;
1295         }
1296
1297 out:
1298         ip_cork_release(inet);
1299         return err;
1300
1301 error:
1302         IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
1303         goto out;
1304 }
1305
1306 /*
1307  *      Throw away all pending data on the socket.
1308  */
1309 void ip_flush_pending_frames(struct sock *sk)
1310 {
1311         struct sk_buff *skb;
1312
1313         while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL)
1314                 kfree_skb(skb);
1315
1316         ip_cork_release(inet_sk(sk));
1317 }
1318
1319
1320 /*
1321  *      Fetch data from kernel space and fill in checksum if needed.
1322  */
1323 static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1324                               int len, int odd, struct sk_buff *skb)
1325 {
1326         __wsum csum;
1327
1328         csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1329         skb->csum = csum_block_add(skb->csum, csum, odd);
1330         return 0;
1331 }
1332
1333 /*
1334  *      Generic function to send a packet as reply to another packet.
1335  *      Used to send TCP resets so far. ICMP should use this function too.
1336  *
1337  *      Should run single threaded per socket because it uses the sock
1338  *      structure to pass arguments.
1339  */
1340 void ip_send_reply(struct sock *sk, struct sk_buff *skb, struct ip_reply_arg *arg,
1341                    unsigned int len)
1342 {
1343         struct inet_sock *inet = inet_sk(sk);
1344         struct {
1345                 struct ip_options       opt;
1346                 char                    data[40];
1347         } replyopts;
1348         struct ipcm_cookie ipc;
1349         __be32 daddr;
1350         struct rtable *rt = skb->rtable;
1351
1352         if (ip_options_echo(&replyopts.opt, skb))
1353                 return;
1354
1355         daddr = ipc.addr = rt->rt_src;
1356         ipc.opt = NULL;
1357
1358         if (replyopts.opt.optlen) {
1359                 ipc.opt = &replyopts.opt;
1360
1361                 if (ipc.opt->srr)
1362                         daddr = replyopts.opt.faddr;
1363         }
1364
1365         {
1366                 struct flowi fl = { .oif = arg->bound_dev_if,
1367                                     .nl_u = { .ip4_u =
1368                                               { .daddr = daddr,
1369                                                 .saddr = rt->rt_spec_dst,
1370                                                 .tos = RT_TOS(ip_hdr(skb)->tos) } },
1371                                     /* Not quite clean, but right. */
1372                                     .uli_u = { .ports =
1373                                                { .sport = tcp_hdr(skb)->dest,
1374                                                  .dport = tcp_hdr(skb)->source } },
1375                                     .proto = sk->sk_protocol,
1376                                     .flags = ip_reply_arg_flowi_flags(arg) };
1377                 security_skb_classify_flow(skb, &fl);
1378                 if (ip_route_output_key(sock_net(sk), &rt, &fl))
1379                         return;
1380         }
1381
1382         /* And let IP do all the hard work.
1383
1384            This chunk is not reenterable, hence spinlock.
1385            Note that it uses the fact, that this function is called
1386            with locally disabled BH and that sk cannot be already spinlocked.
1387          */
1388         bh_lock_sock(sk);
1389         inet->tos = ip_hdr(skb)->tos;
1390         sk->sk_priority = skb->priority;
1391         sk->sk_protocol = ip_hdr(skb)->protocol;
1392         sk->sk_bound_dev_if = arg->bound_dev_if;
1393         ip_append_data(sk, ip_reply_glue_bits, arg->iov->iov_base, len, 0,
1394                        &ipc, rt, MSG_DONTWAIT);
1395         if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
1396                 if (arg->csumoffset >= 0)
1397                         *((__sum16 *)skb_transport_header(skb) +
1398                           arg->csumoffset) = csum_fold(csum_add(skb->csum,
1399                                                                 arg->csum));
1400                 skb->ip_summed = CHECKSUM_NONE;
1401                 ip_push_pending_frames(sk);
1402         }
1403
1404         bh_unlock_sock(sk);
1405
1406         ip_rt_put(rt);
1407 }
1408
1409 void __init ip_init(void)
1410 {
1411         ip_rt_init();
1412         inet_initpeers();
1413
1414 #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1415         igmp_mc_proc_init();
1416 #endif
1417 }
1418
1419 EXPORT_SYMBOL(ip_generic_getfrag);
1420 EXPORT_SYMBOL(ip_queue_xmit);
1421 EXPORT_SYMBOL(ip_send_check);