err.no 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  * Version:     $Id: ip_output.c,v 1.100 2002/02/01 22:01:03 davem Exp $
   9  *
  10  * Authors:     Ross Biro
  11  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  12  *              Donald Becker, <becker@super.org>
  13  *              Alan Cox, <Alan.Cox@linux.org>
  14  *              Richard Underwood
  15  *              Stefan Becker, <stefanb@yello.ping.de>
  16  *              Jorge Cwik, <jorge@laser.satlink.net>
  17  *              Arnt Gulbrandsen, <agulbra@nvg.unit.no>
  18  *              Hirokazu Takahashi, <taka@valinux.co.jp>
  19  *
  20  *      See ip_input.c for original log
  21  *
  22  *      Fixes:
  23  *              Alan Cox        :       Missing nonblock feature in ip_build_xmit.
  24  *              Mike Kilburn    :       htons() missing in ip_build_xmit.
  25  *              Bradford Johnson:       Fix faulty handling of some frames when
  26  *                                      no route is found.
  27  *              Alexander Demenshin:    Missing sk/skb free in ip_queue_xmit
  28  *                                      (in case if packet not accepted by
  29  *                                      output firewall rules)
  30  *              Mike McLagan    :       Routing by source
  31  *              Alexey Kuznetsov:       use new route cache
  32  *              Andi Kleen:             Fix broken PMTU recovery and remove
  33  *                                      some redundant tests.
  34  *      Vitaly E. Lavrov        :       Transparent proxy revived after year coma.
  35  *              Andi Kleen      :       Replace ip_reply with ip_send_reply.
  36  *              Andi Kleen      :       Split fast and slow ip_build_xmit path
  37  *                                      for decreased register pressure on x86
  38  *                                      and more readibility.
  39  *              Marc Boucher    :       When call_out_firewall returns FW_QUEUE,
  40  *                                      silently drop skb instead of failing with -EPERM.
  41  *              Detlev Wengorz  :       Copy protocol for fragments.
  42  *              Hirokazu Takahashi:     HW checksumming for outgoing UDP
  43  *                                      datagrams.
  44  *              Hirokazu Takahashi:     sendfile() on UDP works now.
  45  */
  46
  47 #include <asm/uaccess.h>
  48 #include <asm/system.h>
  49 #include <linux/module.h>
  50 #include <linux/types.h>
  51 #include <linux/kernel.h>
  52 #include <linux/mm.h>
  53 #include <linux/string.h>
  54 #include <linux/errno.h>
  55 #include <linux/highmem.h>
  56
  57 #include <linux/socket.h>
  58 #include <linux/sockios.h>
  59 #include <linux/in.h>
  60 #include <linux/inet.h>
  61 #include <linux/netdevice.h>
  62 #include <linux/etherdevice.h>
  63 #include <linux/proc_fs.h>
  64 #include <linux/stat.h>
  65 #include <linux/init.h>
  66
  67 #include <net/snmp.h>
  68 #include <net/ip.h>
  69 #include <net/protocol.h>
  70 #include <net/route.h>
  71 #include <net/xfrm.h>
  72 #include <linux/skbuff.h>
  73 #include <net/sock.h>
  74 #include <net/arp.h>
  75 #include <net/icmp.h>
  76 #include <net/checksum.h>
  77 #include <net/inetpeer.h>
  78 #include <linux/igmp.h>
  79 #include <linux/netfilter_ipv4.h>
  80 #include <linux/netfilter_bridge.h>
  81 #include <linux/mroute.h>
  82 #include <linux/netlink.h>
  83 #include <linux/tcp.h>
  84
  85 int sysctl_ip_default_ttl __read_mostly = IPDEFTTL;
  86
  87 /* Generate a checksum for an outgoing IP datagram. */
  88 __inline__ void ip_send_check(struct iphdr *iph)
  89 {
  90         iph->check = 0;
  91         iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
  92 }
  93
  94 int __ip_local_out(struct sk_buff *skb)
  95 {
  96         struct iphdr *iph = ip_hdr(skb);
  97
  98         iph->tot_len = htons(skb->len);
  99         ip_send_check(iph);
 100         return nf_hook(PF_INET, NF_INET_LOCAL_OUT, skb, NULL, skb->dst->dev,
 101                        dst_output);
 102 }
 103
 104 int ip_local_out(struct sk_buff *skb)
 105 {
 106         int err;
 107
 108         err = __ip_local_out(skb);
 109         if (likely(err == 1))
 110                 err = dst_output(skb);
 111
 112         return err;
 113 }
 114 EXPORT_SYMBOL_GPL(ip_local_out);
 115
 116 /* dev_loopback_xmit for use with netfilter. */
 117 static int ip_dev_loopback_xmit(struct sk_buff *newskb)
 118 {
 119         skb_reset_mac_header(newskb);
 120         __skb_pull(newskb, skb_network_offset(newskb));
 121         newskb->pkt_type = PACKET_LOOPBACK;
 122         newskb->ip_summed = CHECKSUM_UNNECESSARY;
 123         BUG_TRAP(newskb->dst);
 124         netif_rx(newskb);
 125         return 0;
 126 }
 127
 128 static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
 129 {
 130         int ttl = inet->uc_ttl;
 131
 132         if (ttl < 0)
 133                 ttl = dst_metric(dst, RTAX_HOPLIMIT);
 134         return ttl;
 135 }
 136
 137 /*
 138  *              Add an ip header to a skbuff and send it out.
 139  *
 140  */
 141 int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk,
 142                           __be32 saddr, __be32 daddr, struct ip_options *opt)
 143 {
 144         struct inet_sock *inet = inet_sk(sk);
 145         struct rtable *rt = skb->rtable;
 146         struct iphdr *iph;
 147
 148         /* Build the IP header. */
 149         skb_push(skb, sizeof(struct iphdr) + (opt ? opt->optlen : 0));
 150         skb_reset_network_header(skb);
 151         iph = ip_hdr(skb);
 152         iph->version  = 4;
 153         iph->ihl      = 5;
 154         iph->tos      = inet->tos;
 155         if (ip_dont_fragment(sk, &rt->u.dst))
 156                 iph->frag_off = htons(IP_DF);
 157         else
 158                 iph->frag_off = 0;
 159         iph->ttl      = ip_select_ttl(inet, &rt->u.dst);
 160         iph->daddr    = rt->rt_dst;
 161         iph->saddr    = rt->rt_src;
 162         iph->protocol = sk->sk_protocol;
 163         ip_select_ident(iph, &rt->u.dst, sk);
 164
 165         if (opt && opt->optlen) {
 166                 iph->ihl += opt->optlen>>2;
 167                 ip_options_build(skb, opt, daddr, rt, 0);
 168         }
 169
 170         skb->priority = sk->sk_priority;
 171         skb->mark = sk->sk_mark;
 172
 173         /* Send it out. */
 174         return ip_local_out(skb);
 175 }
 176
 177 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
 178
 179 static inline int ip_finish_output2(struct sk_buff *skb)
 180 {
 181         struct dst_entry *dst = skb->dst;
 182         struct rtable *rt = (struct rtable *)dst;
 183         struct net_device *dev = dst->dev;
 184         unsigned int hh_len = LL_RESERVED_SPACE(dev);
 185
 186         if (rt->rt_type == RTN_MULTICAST)
 187                 IP_INC_STATS(IPSTATS_MIB_OUTMCASTPKTS);
 188         else if (rt->rt_type == RTN_BROADCAST)
 189                 IP_INC_STATS(IPSTATS_MIB_OUTBCASTPKTS);
 190
 191         /* Be paranoid, rather than too clever. */
 192         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
 193                 struct sk_buff *skb2;
 194
 195                 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
 196                 if (skb2 == NULL) {
 197                         kfree_skb(skb);
 198                         return -ENOMEM;
 199                 }
 200                 if (skb->sk)
 201                         skb_set_owner_w(skb2, skb->sk);
 202                 kfree_skb(skb);
 203                 skb = skb2;
 204         }
 205
 206         if (dst->hh)
 207                 return neigh_hh_output(dst->hh, skb);
 208         else if (dst->neighbour)
 209                 return dst->neighbour->output(skb);
 210
 211         if (net_ratelimit())
 212                 printk(KERN_DEBUG "ip_finish_output2: No header cache and no neighbour!\n");
 213         kfree_skb(skb);
 214         return -EINVAL;
 215 }
 216
 217 static inline int ip_skb_dst_mtu(struct sk_buff *skb)
 218 {
 219         struct inet_sock *inet = skb->sk ? inet_sk(skb->sk) : NULL;
 220
 221         return (inet && inet->pmtudisc == IP_PMTUDISC_PROBE) ?
 222                skb->dst->dev->mtu : dst_mtu(skb->dst);
 223 }
 224
 225 static int ip_finish_output(struct sk_buff *skb)
 226 {
 227 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
 228         /* Policy lookup after SNAT yielded a new policy */
 229         if (skb->dst->xfrm != NULL) {
 230                 IPCB(skb)->flags |= IPSKB_REROUTED;
 231                 return dst_output(skb);
 232         }
 233 #endif
 234         if (skb->len > ip_skb_dst_mtu(skb) && !skb_is_gso(skb))
 235                 return ip_fragment(skb, ip_finish_output2);
 236         else
 237                 return ip_finish_output2(skb);
 238 }
 239
 240 int ip_mc_output(struct sk_buff *skb)
 241 {
 242         struct sock *sk = skb->sk;
 243         struct rtable *rt = skb->rtable;
 244         struct net_device *dev = rt->u.dst.dev;
 245
 246         /*
 247          *      If the indicated interface is up and running, send the packet.
 248          */
 249         IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
 250
 251         skb->dev = dev;
 252         skb->protocol = htons(ETH_P_IP);
 253
 254         /*
 255          *      Multicasts are looped back for other local users
 256          */
 257
 258         if (rt->rt_flags&RTCF_MULTICAST) {
 259                 if ((!sk || inet_sk(sk)->mc_loop)
 260 #ifdef CONFIG_IP_MROUTE
 261                 /* Small optimization: do not loopback not local frames,
 262                    which returned after forwarding; they will be  dropped
 263                    by ip_mr_input in any case.
 264                    Note, that local frames are looped back to be delivered
 265                    to local recipients.
 266
 267                    This check is duplicated in ip_mr_input at the moment.
 268                  */
 269                     && ((rt->rt_flags&RTCF_LOCAL) || !(IPCB(skb)->flags&IPSKB_FORWARDED))
 270 #endif
 271                 ) {
 272                         struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
 273                         if (newskb)
 274                                 NF_HOOK(PF_INET, NF_INET_POST_ROUTING, newskb,
 275                                         NULL, newskb->dev,
 276                                         ip_dev_loopback_xmit);
 277                 }
 278
 279                 /* Multicasts with ttl 0 must not go beyond the host */
 280
 281                 if (ip_hdr(skb)->ttl == 0) {
 282                         kfree_skb(skb);
 283                         return 0;
 284                 }
 285         }
 286
 287         if (rt->rt_flags&RTCF_BROADCAST) {
 288                 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
 289                 if (newskb)
 290                         NF_HOOK(PF_INET, NF_INET_POST_ROUTING, newskb, NULL,
 291                                 newskb->dev, ip_dev_loopback_xmit);
 292         }
 293
 294         return NF_HOOK_COND(PF_INET, NF_INET_POST_ROUTING, skb, NULL, skb->dev,
 295                             ip_finish_output,
 296                             !(IPCB(skb)->flags & IPSKB_REROUTED));
 297 }
 298
 299 int ip_output(struct sk_buff *skb)
 300 {
 301         struct net_device *dev = skb->dst->dev;
 302
 303         IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
 304
 305         skb->dev = dev;
 306         skb->protocol = htons(ETH_P_IP);
 307
 308         return NF_HOOK_COND(PF_INET, NF_INET_POST_ROUTING, skb, NULL, dev,
 309                             ip_finish_output,
 310                             !(IPCB(skb)->flags & IPSKB_REROUTED));
 311 }
 312
 313 int ip_queue_xmit(struct sk_buff *skb, int ipfragok)
 314 {
 315         struct sock *sk = skb->sk;
 316         struct inet_sock *inet = inet_sk(sk);
 317         struct ip_options *opt = inet->opt;
 318         struct rtable *rt;
 319         struct iphdr *iph;
 320
 321         /* Skip all of this if the packet is already routed,
 322          * f.e. by something like SCTP.
 323          */
 324         rt = skb->rtable;
 325         if (rt != NULL)
 326                 goto packet_routed;
 327
 328         /* Make sure we can route this packet. */
 329         rt = (struct rtable *)__sk_dst_check(sk, 0);
 330         if (rt == NULL) {
 331                 __be32 daddr;
 332
 333                 /* Use correct destination address if we have options. */
 334                 daddr = inet->daddr;
 335                 if(opt && opt->srr)
 336                         daddr = opt->faddr;
 337
 338                 {
 339                         struct flowi fl = { .oif = sk->sk_bound_dev_if,
 340                                             .nl_u = { .ip4_u =
 341                                                       { .daddr = daddr,
 342                                                         .saddr = inet->saddr,
 343                                                         .tos = RT_CONN_FLAGS(sk) } },
 344                                             .proto = sk->sk_protocol,
 345                                             .uli_u = { .ports =
 346                                                        { .sport = inet->sport,
 347                                                          .dport = inet->dport } } };
 348
 349                         /* If this fails, retransmit mechanism of transport layer will
 350                          * keep trying until route appears or the connection times
 351                          * itself out.
 352                          */
 353                         security_sk_classify_flow(sk, &fl);
 354                         if (ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0))
 355                                 goto no_route;
 356                 }
 357                 sk_setup_caps(sk, &rt->u.dst);
 358         }
 359         skb->dst = dst_clone(&rt->u.dst);
 360
 361 packet_routed:
 362         if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
 363                 goto no_route;
 364
 365         /* OK, we know where to send it, allocate and build IP header. */
 366         skb_push(skb, sizeof(struct iphdr) + (opt ? opt->optlen : 0));
 367         skb_reset_network_header(skb);
 368         iph = ip_hdr(skb);
 369         *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
 370         if (ip_dont_fragment(sk, &rt->u.dst) && !ipfragok)
 371                 iph->frag_off = htons(IP_DF);
 372         else
 373                 iph->frag_off = 0;
 374         iph->ttl      = ip_select_ttl(inet, &rt->u.dst);
 375         iph->protocol = sk->sk_protocol;
 376         iph->saddr    = rt->rt_src;
 377         iph->daddr    = rt->rt_dst;
 378         /* Transport layer set skb->h.foo itself. */
 379
 380         if (opt && opt->optlen) {
 381                 iph->ihl += opt->optlen >> 2;
 382                 ip_options_build(skb, opt, inet->daddr, rt, 0);
 383         }
 384
 385         ip_select_ident_more(iph, &rt->u.dst, sk,
 386                              (skb_shinfo(skb)->gso_segs ?: 1) - 1);
 387
 388         skb->priority = sk->sk_priority;
 389         skb->mark = sk->sk_mark;
 390
 391         return ip_local_out(skb);
 392
 393 no_route:
 394         IP_INC_STATS(IPSTATS_MIB_OUTNOROUTES);
 395         kfree_skb(skb);
 396         return -EHOSTUNREACH;
 397 }
 398
 399
 400 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
 401 {
 402         to->pkt_type = from->pkt_type;
 403         to->priority = from->priority;
 404         to->protocol = from->protocol;
 405         dst_release(to->dst);
 406         to->dst = dst_clone(from->dst);
 407         to->dev = from->dev;
 408         to->mark = from->mark;
 409
 410         /* Copy the flags to each fragment. */
 411         IPCB(to)->flags = IPCB(from)->flags;
 412
 413 #ifdef CONFIG_NET_SCHED
 414         to->tc_index = from->tc_index;
 415 #endif
 416         nf_copy(to, from);
 417 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
 418     defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
 419         to->nf_trace = from->nf_trace;
 420 #endif
 421 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
 422         to->ipvs_property = from->ipvs_property;
 423 #endif
 424         skb_copy_secmark(to, from);
 425 }
 426
 427 /*
 428  *      This IP datagram is too large to be sent in one piece.  Break it up into
 429  *      smaller pieces (each of size equal to IP header plus
 430  *      a block of the data of the original IP data part) that will yet fit in a
 431  *      single device frame, and queue such a frame for sending.
 432  */
 433
 434 int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff*))
 435 {
 436         struct iphdr *iph;
 437         int raw = 0;
 438         int ptr;
 439         struct net_device *dev;
 440         struct sk_buff *skb2;
 441         unsigned int mtu, hlen, left, len, ll_rs, pad;
 442         int offset;
 443         __be16 not_last_frag;
 444         struct rtable *rt = skb->rtable;
 445         int err = 0;
 446
 447         dev = rt->u.dst.dev;
 448
 449         /*
 450          *      Point into the IP datagram header.
 451          */
 452
 453         iph = ip_hdr(skb);
 454
 455         if (unlikely((iph->frag_off & htons(IP_DF)) && !skb->local_df)) {
 456                 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS);
 457                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
 458                           htonl(ip_skb_dst_mtu(skb)));
 459                 kfree_skb(skb);
 460                 return -EMSGSIZE;
 461         }
 462
 463         /*
 464          *      Setup starting values.
 465          */
 466
 467         hlen = iph->ihl * 4;
 468         mtu = dst_mtu(&rt->u.dst) - hlen;       /* Size of data space */
 469         IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
 470
 471         /* When frag_list is given, use it. First, check its validity:
 472          * some transformers could create wrong frag_list or break existing
 473          * one, it is not prohibited. In this case fall back to copying.
 474          *
 475          * LATER: this step can be merged to real generation of fragments,
 476          * we can switch to copy when see the first bad fragment.
 477          */
 478         if (skb_shinfo(skb)->frag_list) {
 479                 struct sk_buff *frag;
 480                 int first_len = skb_pagelen(skb);
 481                 int truesizes = 0;
 482
 483                 if (first_len - hlen > mtu ||
 484                     ((first_len - hlen) & 7) ||
 485                     (iph->frag_off & htons(IP_MF|IP_OFFSET)) ||
 486                     skb_cloned(skb))
 487                         goto slow_path;
 488
 489                 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
 490                         /* Correct geometry. */
 491                         if (frag->len > mtu ||
 492                             ((frag->len & 7) && frag->next) ||
 493                             skb_headroom(frag) < hlen)
 494                             goto slow_path;
 495
 496                         /* Partially cloned skb? */
 497                         if (skb_shared(frag))
 498                                 goto slow_path;
 499
 500                         BUG_ON(frag->sk);
 501                         if (skb->sk) {
 502                                 sock_hold(skb->sk);
 503                                 frag->sk = skb->sk;
 504                                 frag->destructor = sock_wfree;
 505                                 truesizes += frag->truesize;
 506                         }
 507                 }
 508
 509                 /* Everything is OK. Generate! */
 510
 511                 err = 0;
 512                 offset = 0;
 513                 frag = skb_shinfo(skb)->frag_list;
 514                 skb_shinfo(skb)->frag_list = NULL;
 515                 skb->data_len = first_len - skb_headlen(skb);
 516                 skb->truesize -= truesizes;
 517                 skb->len = first_len;
 518                 iph->tot_len = htons(first_len);
 519                 iph->frag_off = htons(IP_MF);
 520                 ip_send_check(iph);
 521
 522                 for (;;) {
 523                         /* Prepare header of the next frame,
 524                          * before previous one went down. */
 525                         if (frag) {
 526                                 frag->ip_summed = CHECKSUM_NONE;
 527                                 skb_reset_transport_header(frag);
 528                                 __skb_push(frag, hlen);
 529                                 skb_reset_network_header(frag);
 530                                 memcpy(skb_network_header(frag), iph, hlen);
 531                                 iph = ip_hdr(frag);
 532                                 iph->tot_len = htons(frag->len);
 533                                 ip_copy_metadata(frag, skb);
 534                                 if (offset == 0)
 535                                         ip_options_fragment(frag);
 536                                 offset += skb->len - hlen;
 537                                 iph->frag_off = htons(offset>>3);
 538                                 if (frag->next != NULL)
 539                                         iph->frag_off |= htons(IP_MF);
 540                                 /* Ready, complete checksum */
 541                                 ip_send_check(iph);
 542                         }
 543
 544                         err = output(skb);
 545
 546                         if (!err)
 547                                 IP_INC_STATS(IPSTATS_MIB_FRAGCREATES);
 548                         if (err || !frag)
 549                                 break;
 550
 551                         skb = frag;
 552                         frag = skb->next;
 553                         skb->next = NULL;
 554                 }
 555
 556                 if (err == 0) {
 557                         IP_INC_STATS(IPSTATS_MIB_FRAGOKS);
 558                         return 0;
 559                 }
 560
 561                 while (frag) {
 562                         skb = frag->next;
 563                         kfree_skb(frag);
 564                         frag = skb;
 565                 }
 566                 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS);
 567                 return err;
 568         }
 569
 570 slow_path:
 571         left = skb->len - hlen;         /* Space per frame */
 572         ptr = raw + hlen;               /* Where to start from */
 573
 574         /* for bridged IP traffic encapsulated inside f.e. a vlan header,
 575          * we need to make room for the encapsulating header
 576          */
 577         pad = nf_bridge_pad(skb);
 578         ll_rs = LL_RESERVED_SPACE_EXTRA(rt->u.dst.dev, pad);
 579         mtu -= pad;
 580
 581         /*
 582          *      Fragment the datagram.
 583          */
 584
 585         offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
 586         not_last_frag = iph->frag_off & htons(IP_MF);
 587
 588         /*
 589          *      Keep copying data until we run out.
 590          */
 591
 592         while (left > 0) {
 593                 len = left;
 594                 /* IF: it doesn't fit, use 'mtu' - the data space left */
 595                 if (len > mtu)
 596                         len = mtu;
 597                 /* IF: we are not sending upto and including the packet end
 598                    then align the next start on an eight byte boundary */
 599                 if (len < left) {
 600                         len &= ~7;
 601                 }
 602                 /*
 603                  *      Allocate buffer.
 604                  */
 605
 606                 if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) {
 607                         NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n");
 608                         err = -ENOMEM;
 609                         goto fail;
 610                 }
 611
 612                 /*
 613                  *      Set up data on packet
 614                  */
 615
 616                 ip_copy_metadata(skb2, skb);
 617                 skb_reserve(skb2, ll_rs);
 618                 skb_put(skb2, len + hlen);
 619                 skb_reset_network_header(skb2);
 620                 skb2->transport_header = skb2->network_header + hlen;
 621
 622                 /*
 623                  *      Charge the memory for the fragment to any owner
 624                  *      it might possess
 625                  */
 626
 627                 if (skb->sk)
 628                         skb_set_owner_w(skb2, skb->sk);
 629
 630                 /*
 631                  *      Copy the packet header into the new buffer.
 632                  */
 633
 634                 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
 635
 636                 /*
 637                  *      Copy a block of the IP datagram.
 638                  */
 639                 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
 640                         BUG();
 641                 left -= len;
 642
 643                 /*
 644                  *      Fill in the new header fields.
 645                  */
 646                 iph = ip_hdr(skb2);
 647                 iph->frag_off = htons((offset >> 3));
 648
 649                 /* ANK: dirty, but effective trick. Upgrade options only if
 650                  * the segment to be fragmented was THE FIRST (otherwise,
 651                  * options are already fixed) and make it ONCE
 652                  * on the initial skb, so that all the following fragments
 653                  * will inherit fixed options.
 654                  */
 655                 if (offset == 0)
 656                         ip_options_fragment(skb);
 657
 658                 /*
 659                  *      Added AC : If we are fragmenting a fragment that's not the
 660                  *                 last fragment then keep MF on each bit
 661                  */
 662                 if (left > 0 || not_last_frag)
 663                         iph->frag_off |= htons(IP_MF);
 664                 ptr += len;
 665                 offset += len;
 666
 667                 /*
 668                  *      Put this fragment into the sending queue.
 669                  */
 670                 iph->tot_len = htons(len + hlen);
 671
 672                 ip_send_check(iph);
 673
 674                 err = output(skb2);
 675                 if (err)
 676                         goto fail;
 677
 678                 IP_INC_STATS(IPSTATS_MIB_FRAGCREATES);
 679         }
 680         kfree_skb(skb);
 681         IP_INC_STATS(IPSTATS_MIB_FRAGOKS);
 682         return err;
 683
 684 fail:
 685         kfree_skb(skb);
 686         IP_INC_STATS(IPSTATS_MIB_FRAGFAILS);
 687         return err;
 688 }
 689
 690 EXPORT_SYMBOL(ip_fragment);
 691
 692 int
 693 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
 694 {
 695         struct iovec *iov = from;
 696
 697         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 698                 if (memcpy_fromiovecend(to, iov, offset, len) < 0)
 699                         return -EFAULT;
 700         } else {
 701                 __wsum csum = 0;
 702                 if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0)
 703                         return -EFAULT;
 704                 skb->csum = csum_block_add(skb->csum, csum, odd);
 705         }
 706         return 0;
 707 }
 708
 709 static inline __wsum
 710 csum_page(struct page *page, int offset, int copy)
 711 {
 712         char *kaddr;
 713         __wsum csum;
 714         kaddr = kmap(page);
 715         csum = csum_partial(kaddr + offset, copy, 0);
 716         kunmap(page);
 717         return csum;
 718 }
 719
 720 static inline int ip_ufo_append_data(struct sock *sk,
 721                         int getfrag(void *from, char *to, int offset, int len,
 722                                int odd, struct sk_buff *skb),
 723                         void *from, int length, int hh_len, int fragheaderlen,
 724                         int transhdrlen, int mtu,unsigned int flags)
 725 {
 726         struct sk_buff *skb;
 727         int err;
 728
 729         /* There is support for UDP fragmentation offload by network
 730          * device, so create one single skb packet containing complete
 731          * udp datagram
 732          */
 733         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
 734                 skb = sock_alloc_send_skb(sk,
 735                         hh_len + fragheaderlen + transhdrlen + 20,
 736                         (flags & MSG_DONTWAIT), &err);
 737
 738                 if (skb == NULL)
 739                         return err;
 740
 741                 /* reserve space for Hardware header */
 742                 skb_reserve(skb, hh_len);
 743
 744                 /* create space for UDP/IP header */
 745                 skb_put(skb,fragheaderlen + transhdrlen);
 746
 747                 /* initialize network header pointer */
 748                 skb_reset_network_header(skb);
 749
 750                 /* initialize protocol header pointer */
 751                 skb->transport_header = skb->network_header + fragheaderlen;
 752
 753                 skb->ip_summed = CHECKSUM_PARTIAL;
 754                 skb->csum = 0;
 755                 sk->sk_sndmsg_off = 0;
 756
 757                 /* specify the length of each IP datagram fragment */
 758                 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
 759                 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
 760                 __skb_queue_tail(&sk->sk_write_queue, skb);
 761         }
 762
 763         return skb_append_datato_frags(sk, skb, getfrag, from,
 764                                        (length - transhdrlen));
 765 }
 766
 767 /*
 768  *      ip_append_data() and ip_append_page() can make one large IP datagram
 769  *      from many pieces of data. Each pieces will be holded on the socket
 770  *      until ip_push_pending_frames() is called. Each piece can be a page
 771  *      or non-page data.
 772  *
 773  *      Not only UDP, other transport protocols - e.g. raw sockets - can use
 774  *      this interface potentially.
 775  *
 776  *      LATER: length must be adjusted by pad at tail, when it is required.
 777  */
 778 int ip_append_data(struct sock *sk,
 779                    int getfrag(void *from, char *to, int offset, int len,
 780                                int odd, struct sk_buff *skb),
 781                    void *from, int length, int transhdrlen,
 782                    struct ipcm_cookie *ipc, struct rtable *rt,
 783                    unsigned int flags)
 784 {
 785         struct inet_sock *inet = inet_sk(sk);
 786         struct sk_buff *skb;
 787
 788         struct ip_options *opt = NULL;
 789         int hh_len;
 790         int exthdrlen;
 791         int mtu;
 792         int copy;
 793         int err;
 794         int offset = 0;
 795         unsigned int maxfraglen, fragheaderlen;
 796         int csummode = CHECKSUM_NONE;
 797
 798         if (flags&MSG_PROBE)
 799                 return 0;
 800
 801         if (skb_queue_empty(&sk->sk_write_queue)) {
 802                 /*
 803                  * setup for corking.
 804                  */
 805                 opt = ipc->opt;
 806                 if (opt) {
 807                         if (inet->cork.opt == NULL) {
 808                                 inet->cork.opt = kmalloc(sizeof(struct ip_options) + 40, sk->sk_allocation);
 809                                 if (unlikely(inet->cork.opt == NULL))
 810                                         return -ENOBUFS;
 811                         }
 812                         memcpy(inet->cork.opt, opt, sizeof(struct ip_options)+opt->optlen);
 813                         inet->cork.flags |= IPCORK_OPT;
 814                         inet->cork.addr = ipc->addr;
 815                 }
 816                 dst_hold(&rt->u.dst);
 817                 inet->cork.fragsize = mtu = inet->pmtudisc == IP_PMTUDISC_PROBE ?
 818                                             rt->u.dst.dev->mtu :
 819                                             dst_mtu(rt->u.dst.path);
 820                 inet->cork.dst = &rt->u.dst;
 821                 inet->cork.length = 0;
 822                 sk->sk_sndmsg_page = NULL;
 823                 sk->sk_sndmsg_off = 0;
 824                 if ((exthdrlen = rt->u.dst.header_len) != 0) {
 825                         length += exthdrlen;
 826                         transhdrlen += exthdrlen;
 827                 }
 828         } else {
 829                 rt = (struct rtable *)inet->cork.dst;
 830                 if (inet->cork.flags & IPCORK_OPT)
 831                         opt = inet->cork.opt;
 832
 833                 transhdrlen = 0;
 834                 exthdrlen = 0;
 835                 mtu = inet->cork.fragsize;
 836         }
 837         hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
 838
 839         fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
 840         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
 841
 842         if (inet->cork.length + length > 0xFFFF - fragheaderlen) {
 843                 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu-exthdrlen);
 844                 return -EMSGSIZE;
 845         }
 846
 847         /*
 848          * transhdrlen > 0 means that this is the first fragment and we wish
 849          * it won't be fragmented in the future.
 850          */
 851         if (transhdrlen &&
 852             length + fragheaderlen <= mtu &&
 853             rt->u.dst.dev->features & NETIF_F_V4_CSUM &&
 854             !exthdrlen)
 855                 csummode = CHECKSUM_PARTIAL;
 856
 857         inet->cork.length += length;
 858         if (((length> mtu) || !skb_queue_empty(&sk->sk_write_queue)) &&
 859             (sk->sk_protocol == IPPROTO_UDP) &&
 860             (rt->u.dst.dev->features & NETIF_F_UFO)) {
 861                 err = ip_ufo_append_data(sk, getfrag, from, length, hh_len,
 862                                          fragheaderlen, transhdrlen, mtu,
 863                                          flags);
 864                 if (err)
 865                         goto error;
 866                 return 0;
 867         }
 868
 869         /* So, what's going on in the loop below?
 870          *
 871          * We use calculated fragment length to generate chained skb,
 872          * each of segments is IP fragment ready for sending to network after
 873          * adding appropriate IP header.
 874          */
 875
 876         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
 877                 goto alloc_new_skb;
 878
 879         while (length > 0) {
 880                 /* Check if the remaining data fits into current packet. */
 881                 copy = mtu - skb->len;
 882                 if (copy < length)
 883                         copy = maxfraglen - skb->len;
 884                 if (copy <= 0) {
 885                         char *data;
 886                         unsigned int datalen;
 887                         unsigned int fraglen;
 888                         unsigned int fraggap;
 889                         unsigned int alloclen;
 890                         struct sk_buff *skb_prev;
 891 alloc_new_skb:
 892                         skb_prev = skb;
 893                         if (skb_prev)
 894                                 fraggap = skb_prev->len - maxfraglen;
 895                         else
 896                                 fraggap = 0;
 897
 898                         /*
 899                          * If remaining data exceeds the mtu,
 900                          * we know we need more fragment(s).
 901                          */
 902                         datalen = length + fraggap;
 903                         if (datalen > mtu - fragheaderlen)
 904                                 datalen = maxfraglen - fragheaderlen;
 905                         fraglen = datalen + fragheaderlen;
 906
 907                         if ((flags & MSG_MORE) &&
 908                             !(rt->u.dst.dev->features&NETIF_F_SG))
 909                                 alloclen = mtu;
 910                         else
 911                                 alloclen = datalen + fragheaderlen;
 912
 913                         /* The last fragment gets additional space at tail.
 914                          * Note, with MSG_MORE we overallocate on fragments,
 915                          * because we have no idea what fragment will be
 916                          * the last.
 917                          */
 918                         if (datalen == length + fraggap)
 919                                 alloclen += rt->u.dst.trailer_len;
 920
 921                         if (transhdrlen) {
 922                                 skb = sock_alloc_send_skb(sk,
 923                                                 alloclen + hh_len + 15,
 924                                                 (flags & MSG_DONTWAIT), &err);
 925                         } else {
 926                                 skb = NULL;
 927                                 if (atomic_read(&sk->sk_wmem_alloc) <=
 928                                     2 * sk->sk_sndbuf)
 929                                         skb = sock_wmalloc(sk,
 930                                                            alloclen + hh_len + 15, 1,
 931                                                            sk->sk_allocation);
 932                                 if (unlikely(skb == NULL))
 933                                         err = -ENOBUFS;
 934                         }
 935                         if (skb == NULL)
 936                                 goto error;
 937
 938                         /*
 939                          *      Fill in the control structures
 940                          */
 941                         skb->ip_summed = csummode;
 942                         skb->csum = 0;
 943                         skb_reserve(skb, hh_len);
 944
 945                         /*
 946                          *      Find where to start putting bytes.
 947                          */
 948                         data = skb_put(skb, fraglen);
 949                         skb_set_network_header(skb, exthdrlen);
 950                         skb->transport_header = (skb->network_header +
 951                                                  fragheaderlen);
 952                         data += fragheaderlen;
 953
 954                         if (fraggap) {
 955                                 skb->csum = skb_copy_and_csum_bits(
 956                                         skb_prev, maxfraglen,
 957                                         data + transhdrlen, fraggap, 0);
 958                                 skb_prev->csum = csum_sub(skb_prev->csum,
 959                                                           skb->csum);
 960                                 data += fraggap;
 961                                 pskb_trim_unique(skb_prev, maxfraglen);
 962                         }
 963
 964                         copy = datalen - transhdrlen - fraggap;
 965                         if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
 966                                 err = -EFAULT;
 967                                 kfree_skb(skb);
 968                                 goto error;
 969                         }
 970
 971                         offset += copy;
 972                         length -= datalen - fraggap;
 973                         transhdrlen = 0;
 974                         exthdrlen = 0;
 975                         csummode = CHECKSUM_NONE;
 976
 977                         /*
 978                          * Put the packet on the pending queue.
 979                          */
 980                         __skb_queue_tail(&sk->sk_write_queue, skb);
 981                         continue;
 982                 }
 983
 984                 if (copy > length)
 985                         copy = length;
 986
 987                 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
 988                         unsigned int off;
 989
 990                         off = skb->len;
 991                         if (getfrag(from, skb_put(skb, copy),
 992                                         offset, copy, off, skb) < 0) {
 993                                 __skb_trim(skb, off);
 994                                 err = -EFAULT;
 995                                 goto error;
 996                         }
 997                 } else {
 998                         int i = skb_shinfo(skb)->nr_frags;
 999                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1000                         struct page *page = sk->sk_sndmsg_page;
1001                         int off = sk->sk_sndmsg_off;
1002                         unsigned int left;
1003
1004                         if (page && (left = PAGE_SIZE - off) > 0) {
1005                                 if (copy >= left)
1006                                         copy = left;
1007                                 if (page != frag->page) {
1008                                         if (i == MAX_SKB_FRAGS) {
1009                                                 err = -EMSGSIZE;
1010                                                 goto error;
1011                                         }
1012                                         get_page(page);
1013                                         skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1014                                         frag = &skb_shinfo(skb)->frags[i];
1015                                 }
1016                         } else if (i < MAX_SKB_FRAGS) {
1017                                 if (copy > PAGE_SIZE)
1018                                         copy = PAGE_SIZE;
1019                                 page = alloc_pages(sk->sk_allocation, 0);
1020                                 if (page == NULL)  {
1021                                         err = -ENOMEM;
1022                                         goto error;
1023                                 }
1024                                 sk->sk_sndmsg_page = page;
1025                                 sk->sk_sndmsg_off = 0;
1026
1027                                 skb_fill_page_desc(skb, i, page, 0, 0);
1028                                 frag = &skb_shinfo(skb)->frags[i];
1029                         } else {
1030                                 err = -EMSGSIZE;
1031                                 goto error;
1032                         }
1033                         if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1034                                 err = -EFAULT;
1035                                 goto error;
1036                         }
1037                         sk->sk_sndmsg_off += copy;
1038                         frag->size += copy;
1039                         skb->len += copy;
1040                         skb->data_len += copy;
1041                         skb->truesize += copy;
1042                         atomic_add(copy, &sk->sk_wmem_alloc);
1043                 }
1044                 offset += copy;
1045                 length -= copy;
1046         }
1047
1048         return 0;
1049
1050 error:
1051         inet->cork.length -= length;
1052         IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1053         return err;
1054 }
1055
1056 ssize_t ip_append_page(struct sock *sk, struct page *page,
1057                        int offset, size_t size, int flags)
1058 {
1059         struct inet_sock *inet = inet_sk(sk);
1060         struct sk_buff *skb;
1061         struct rtable *rt;
1062         struct ip_options *opt = NULL;
1063         int hh_len;
1064         int mtu;
1065         int len;
1066         int err;
1067         unsigned int maxfraglen, fragheaderlen, fraggap;
1068
1069         if (inet->hdrincl)
1070                 return -EPERM;
1071
1072         if (flags&MSG_PROBE)
1073                 return 0;
1074
1075         if (skb_queue_empty(&sk->sk_write_queue))
1076                 return -EINVAL;
1077
1078         rt = (struct rtable *)inet->cork.dst;
1079         if (inet->cork.flags & IPCORK_OPT)
1080                 opt = inet->cork.opt;
1081
1082         if (!(rt->u.dst.dev->features&NETIF_F_SG))
1083                 return -EOPNOTSUPP;
1084
1085         hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1086         mtu = inet->cork.fragsize;
1087
1088         fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1089         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1090
1091         if (inet->cork.length + size > 0xFFFF - fragheaderlen) {
1092                 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu);
1093                 return -EMSGSIZE;
1094         }
1095
1096         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1097                 return -EINVAL;
1098
1099         inet->cork.length += size;
1100         if ((sk->sk_protocol == IPPROTO_UDP) &&
1101             (rt->u.dst.dev->features & NETIF_F_UFO)) {
1102                 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
1103                 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1104         }
1105
1106
1107         while (size > 0) {
1108                 int i;
1109
1110                 if (skb_is_gso(skb))
1111                         len = size;
1112                 else {
1113
1114                         /* Check if the remaining data fits into current packet. */
1115                         len = mtu - skb->len;
1116                         if (len < size)
1117                                 len = maxfraglen - skb->len;
1118                 }
1119                 if (len <= 0) {
1120                         struct sk_buff *skb_prev;
1121                         int alloclen;
1122
1123                         skb_prev = skb;
1124                         fraggap = skb_prev->len - maxfraglen;
1125
1126                         alloclen = fragheaderlen + hh_len + fraggap + 15;
1127                         skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1128                         if (unlikely(!skb)) {
1129                                 err = -ENOBUFS;
1130                                 goto error;
1131                         }
1132
1133                         /*
1134                          *      Fill in the control structures
1135                          */
1136                         skb->ip_summed = CHECKSUM_NONE;
1137                         skb->csum = 0;
1138                         skb_reserve(skb, hh_len);
1139
1140                         /*
1141                          *      Find where to start putting bytes.
1142                          */
1143                         skb_put(skb, fragheaderlen + fraggap);
1144                         skb_reset_network_header(skb);
1145                         skb->transport_header = (skb->network_header +
1146                                                  fragheaderlen);
1147                         if (fraggap) {
1148                                 skb->csum = skb_copy_and_csum_bits(skb_prev,
1149                                                                    maxfraglen,
1150                                                     skb_transport_header(skb),
1151                                                                    fraggap, 0);
1152                                 skb_prev->csum = csum_sub(skb_prev->csum,
1153                                                           skb->csum);
1154                                 pskb_trim_unique(skb_prev, maxfraglen);
1155                         }
1156
1157                         /*
1158                          * Put the packet on the pending queue.
1159                          */
1160                         __skb_queue_tail(&sk->sk_write_queue, skb);
1161                         continue;
1162                 }
1163
1164                 i = skb_shinfo(skb)->nr_frags;
1165                 if (len > size)
1166                         len = size;
1167                 if (skb_can_coalesce(skb, i, page, offset)) {
1168                         skb_shinfo(skb)->frags[i-1].size += len;
1169                 } else if (i < MAX_SKB_FRAGS) {
1170                         get_page(page);
1171                         skb_fill_page_desc(skb, i, page, offset, len);
1172                 } else {
1173                         err = -EMSGSIZE;
1174                         goto error;
1175                 }
1176
1177                 if (skb->ip_summed == CHECKSUM_NONE) {
1178                         __wsum csum;
1179                         csum = csum_page(page, offset, len);
1180                         skb->csum = csum_block_add(skb->csum, csum, skb->len);
1181                 }
1182
1183                 skb->len += len;
1184                 skb->data_len += len;
1185                 skb->truesize += len;
1186                 atomic_add(len, &sk->sk_wmem_alloc);
1187                 offset += len;
1188                 size -= len;
1189         }
1190         return 0;
1191
1192 error:
1193         inet->cork.length -= size;
1194         IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1195         return err;
1196 }
1197
1198 static void ip_cork_release(struct inet_sock *inet)
1199 {
1200         inet->cork.flags &= ~IPCORK_OPT;
1201         kfree(inet->cork.opt);
1202         inet->cork.opt = NULL;
1203         dst_release(inet->cork.dst);
1204         inet->cork.dst = NULL;
1205 }
1206
1207 /*
1208  *      Combined all pending IP fragments on the socket as one IP datagram
1209  *      and push them out.
1210  */
1211 int ip_push_pending_frames(struct sock *sk)
1212 {
1213         struct sk_buff *skb, *tmp_skb;
1214         struct sk_buff **tail_skb;
1215         struct inet_sock *inet = inet_sk(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(((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(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                 security_skb_classify_flow(skb, &fl);
1377                 if (ip_route_output_key(sock_net(sk), &rt, &fl))
1378                         return;
1379         }
1380
1381         /* And let IP do all the hard work.
1382
1383            This chunk is not reenterable, hence spinlock.
1384            Note that it uses the fact, that this function is called
1385            with locally disabled BH and that sk cannot be already spinlocked.
1386          */
1387         bh_lock_sock(sk);
1388         inet->tos = ip_hdr(skb)->tos;
1389         sk->sk_priority = skb->priority;
1390         sk->sk_protocol = ip_hdr(skb)->protocol;
1391         sk->sk_bound_dev_if = arg->bound_dev_if;
1392         ip_append_data(sk, ip_reply_glue_bits, arg->iov->iov_base, len, 0,
1393                        &ipc, rt, MSG_DONTWAIT);
1394         if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
1395                 if (arg->csumoffset >= 0)
1396                         *((__sum16 *)skb_transport_header(skb) +
1397                           arg->csumoffset) = csum_fold(csum_add(skb->csum,
1398                                                                 arg->csum));
1399                 skb->ip_summed = CHECKSUM_NONE;
1400                 ip_push_pending_frames(sk);
1401         }
1402
1403         bh_unlock_sock(sk);
1404
1405         ip_rt_put(rt);
1406 }
1407
1408 void __init ip_init(void)
1409 {
1410         ip_rt_init();
1411         inet_initpeers();
1412
1413 #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1414         igmp_mc_proc_init();
1415 #endif
1416 }
1417
1418 EXPORT_SYMBOL(ip_generic_getfrag);
1419 EXPORT_SYMBOL(ip_queue_xmit);
1420 EXPORT_SYMBOL(ip_send_check);