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(sk->sk_net, &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
 758         err = skb_append_datato_frags(sk,skb, getfrag, from,
 759                                (length - transhdrlen));
 760         if (!err) {
 761                 /* specify the length of each IP datagram fragment*/
 762                 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
 763                 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
 764                 __skb_queue_tail(&sk->sk_write_queue, skb);
 765
 766                 return 0;
 767         }
 768         /* There is not enough support do UFO ,
 769          * so follow normal path
 770          */
 771         kfree_skb(skb);
 772         return err;
 773 }
 774
 775 /*
 776  *      ip_append_data() and ip_append_page() can make one large IP datagram
 777  *      from many pieces of data. Each pieces will be holded on the socket
 778  *      until ip_push_pending_frames() is called. Each piece can be a page
 779  *      or non-page data.
 780  *
 781  *      Not only UDP, other transport protocols - e.g. raw sockets - can use
 782  *      this interface potentially.
 783  *
 784  *      LATER: length must be adjusted by pad at tail, when it is required.
 785  */
 786 int ip_append_data(struct sock *sk,
 787                    int getfrag(void *from, char *to, int offset, int len,
 788                                int odd, struct sk_buff *skb),
 789                    void *from, int length, int transhdrlen,
 790                    struct ipcm_cookie *ipc, struct rtable *rt,
 791                    unsigned int flags)
 792 {
 793         struct inet_sock *inet = inet_sk(sk);
 794         struct sk_buff *skb;
 795
 796         struct ip_options *opt = NULL;
 797         int hh_len;
 798         int exthdrlen;
 799         int mtu;
 800         int copy;
 801         int err;
 802         int offset = 0;
 803         unsigned int maxfraglen, fragheaderlen;
 804         int csummode = CHECKSUM_NONE;
 805
 806         if (flags&MSG_PROBE)
 807                 return 0;
 808
 809         if (skb_queue_empty(&sk->sk_write_queue)) {
 810                 /*
 811                  * setup for corking.
 812                  */
 813                 opt = ipc->opt;
 814                 if (opt) {
 815                         if (inet->cork.opt == NULL) {
 816                                 inet->cork.opt = kmalloc(sizeof(struct ip_options) + 40, sk->sk_allocation);
 817                                 if (unlikely(inet->cork.opt == NULL))
 818                                         return -ENOBUFS;
 819                         }
 820                         memcpy(inet->cork.opt, opt, sizeof(struct ip_options)+opt->optlen);
 821                         inet->cork.flags |= IPCORK_OPT;
 822                         inet->cork.addr = ipc->addr;
 823                 }
 824                 dst_hold(&rt->u.dst);
 825                 inet->cork.fragsize = mtu = inet->pmtudisc == IP_PMTUDISC_PROBE ?
 826                                             rt->u.dst.dev->mtu :
 827                                             dst_mtu(rt->u.dst.path);
 828                 inet->cork.dst = &rt->u.dst;
 829                 inet->cork.length = 0;
 830                 sk->sk_sndmsg_page = NULL;
 831                 sk->sk_sndmsg_off = 0;
 832                 if ((exthdrlen = rt->u.dst.header_len) != 0) {
 833                         length += exthdrlen;
 834                         transhdrlen += exthdrlen;
 835                 }
 836         } else {
 837                 rt = (struct rtable *)inet->cork.dst;
 838                 if (inet->cork.flags & IPCORK_OPT)
 839                         opt = inet->cork.opt;
 840
 841                 transhdrlen = 0;
 842                 exthdrlen = 0;
 843                 mtu = inet->cork.fragsize;
 844         }
 845         hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
 846
 847         fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
 848         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
 849
 850         if (inet->cork.length + length > 0xFFFF - fragheaderlen) {
 851                 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu-exthdrlen);
 852                 return -EMSGSIZE;
 853         }
 854
 855         /*
 856          * transhdrlen > 0 means that this is the first fragment and we wish
 857          * it won't be fragmented in the future.
 858          */
 859         if (transhdrlen &&
 860             length + fragheaderlen <= mtu &&
 861             rt->u.dst.dev->features & NETIF_F_V4_CSUM &&
 862             !exthdrlen)
 863                 csummode = CHECKSUM_PARTIAL;
 864
 865         inet->cork.length += length;
 866         if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
 867                         (rt->u.dst.dev->features & NETIF_F_UFO)) {
 868
 869                 err = ip_ufo_append_data(sk, getfrag, from, length, hh_len,
 870                                          fragheaderlen, transhdrlen, mtu,
 871                                          flags);
 872                 if (err)
 873                         goto error;
 874                 return 0;
 875         }
 876
 877         /* So, what's going on in the loop below?
 878          *
 879          * We use calculated fragment length to generate chained skb,
 880          * each of segments is IP fragment ready for sending to network after
 881          * adding appropriate IP header.
 882          */
 883
 884         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
 885                 goto alloc_new_skb;
 886
 887         while (length > 0) {
 888                 /* Check if the remaining data fits into current packet. */
 889                 copy = mtu - skb->len;
 890                 if (copy < length)
 891                         copy = maxfraglen - skb->len;
 892                 if (copy <= 0) {
 893                         char *data;
 894                         unsigned int datalen;
 895                         unsigned int fraglen;
 896                         unsigned int fraggap;
 897                         unsigned int alloclen;
 898                         struct sk_buff *skb_prev;
 899 alloc_new_skb:
 900                         skb_prev = skb;
 901                         if (skb_prev)
 902                                 fraggap = skb_prev->len - maxfraglen;
 903                         else
 904                                 fraggap = 0;
 905
 906                         /*
 907                          * If remaining data exceeds the mtu,
 908                          * we know we need more fragment(s).
 909                          */
 910                         datalen = length + fraggap;
 911                         if (datalen > mtu - fragheaderlen)
 912                                 datalen = maxfraglen - fragheaderlen;
 913                         fraglen = datalen + fragheaderlen;
 914
 915                         if ((flags & MSG_MORE) &&
 916                             !(rt->u.dst.dev->features&NETIF_F_SG))
 917                                 alloclen = mtu;
 918                         else
 919                                 alloclen = datalen + fragheaderlen;
 920
 921                         /* The last fragment gets additional space at tail.
 922                          * Note, with MSG_MORE we overallocate on fragments,
 923                          * because we have no idea what fragment will be
 924                          * the last.
 925                          */
 926                         if (datalen == length + fraggap)
 927                                 alloclen += rt->u.dst.trailer_len;
 928
 929                         if (transhdrlen) {
 930                                 skb = sock_alloc_send_skb(sk,
 931                                                 alloclen + hh_len + 15,
 932                                                 (flags & MSG_DONTWAIT), &err);
 933                         } else {
 934                                 skb = NULL;
 935                                 if (atomic_read(&sk->sk_wmem_alloc) <=
 936                                     2 * sk->sk_sndbuf)
 937                                         skb = sock_wmalloc(sk,
 938                                                            alloclen + hh_len + 15, 1,
 939                                                            sk->sk_allocation);
 940                                 if (unlikely(skb == NULL))
 941                                         err = -ENOBUFS;
 942                         }
 943                         if (skb == NULL)
 944                                 goto error;
 945
 946                         /*
 947                          *      Fill in the control structures
 948                          */
 949                         skb->ip_summed = csummode;
 950                         skb->csum = 0;
 951                         skb_reserve(skb, hh_len);
 952
 953                         /*
 954                          *      Find where to start putting bytes.
 955                          */
 956                         data = skb_put(skb, fraglen);
 957                         skb_set_network_header(skb, exthdrlen);
 958                         skb->transport_header = (skb->network_header +
 959                                                  fragheaderlen);
 960                         data += fragheaderlen;
 961
 962                         if (fraggap) {
 963                                 skb->csum = skb_copy_and_csum_bits(
 964                                         skb_prev, maxfraglen,
 965                                         data + transhdrlen, fraggap, 0);
 966                                 skb_prev->csum = csum_sub(skb_prev->csum,
 967                                                           skb->csum);
 968                                 data += fraggap;
 969                                 pskb_trim_unique(skb_prev, maxfraglen);
 970                         }
 971
 972                         copy = datalen - transhdrlen - fraggap;
 973                         if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
 974                                 err = -EFAULT;
 975                                 kfree_skb(skb);
 976                                 goto error;
 977                         }
 978
 979                         offset += copy;
 980                         length -= datalen - fraggap;
 981                         transhdrlen = 0;
 982                         exthdrlen = 0;
 983                         csummode = CHECKSUM_NONE;
 984
 985                         /*
 986                          * Put the packet on the pending queue.
 987                          */
 988                         __skb_queue_tail(&sk->sk_write_queue, skb);
 989                         continue;
 990                 }
 991
 992                 if (copy > length)
 993                         copy = length;
 994
 995                 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
 996                         unsigned int off;
 997
 998                         off = skb->len;
 999                         if (getfrag(from, skb_put(skb, copy),
1000                                         offset, copy, off, skb) < 0) {
1001                                 __skb_trim(skb, off);
1002                                 err = -EFAULT;
1003                                 goto error;
1004                         }
1005                 } else {
1006                         int i = skb_shinfo(skb)->nr_frags;
1007                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1008                         struct page *page = sk->sk_sndmsg_page;
1009                         int off = sk->sk_sndmsg_off;
1010                         unsigned int left;
1011
1012                         if (page && (left = PAGE_SIZE - off) > 0) {
1013                                 if (copy >= left)
1014                                         copy = left;
1015                                 if (page != frag->page) {
1016                                         if (i == MAX_SKB_FRAGS) {
1017                                                 err = -EMSGSIZE;
1018                                                 goto error;
1019                                         }
1020                                         get_page(page);
1021                                         skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1022                                         frag = &skb_shinfo(skb)->frags[i];
1023                                 }
1024                         } else if (i < MAX_SKB_FRAGS) {
1025                                 if (copy > PAGE_SIZE)
1026                                         copy = PAGE_SIZE;
1027                                 page = alloc_pages(sk->sk_allocation, 0);
1028                                 if (page == NULL)  {
1029                                         err = -ENOMEM;
1030                                         goto error;
1031                                 }
1032                                 sk->sk_sndmsg_page = page;
1033                                 sk->sk_sndmsg_off = 0;
1034
1035                                 skb_fill_page_desc(skb, i, page, 0, 0);
1036                                 frag = &skb_shinfo(skb)->frags[i];
1037                         } else {
1038                                 err = -EMSGSIZE;
1039                                 goto error;
1040                         }
1041                         if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1042                                 err = -EFAULT;
1043                                 goto error;
1044                         }
1045                         sk->sk_sndmsg_off += copy;
1046                         frag->size += copy;
1047                         skb->len += copy;
1048                         skb->data_len += copy;
1049                         skb->truesize += copy;
1050                         atomic_add(copy, &sk->sk_wmem_alloc);
1051                 }
1052                 offset += copy;
1053                 length -= copy;
1054         }
1055
1056         return 0;
1057
1058 error:
1059         inet->cork.length -= length;
1060         IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1061         return err;
1062 }
1063
1064 ssize_t ip_append_page(struct sock *sk, struct page *page,
1065                        int offset, size_t size, int flags)
1066 {
1067         struct inet_sock *inet = inet_sk(sk);
1068         struct sk_buff *skb;
1069         struct rtable *rt;
1070         struct ip_options *opt = NULL;
1071         int hh_len;
1072         int mtu;
1073         int len;
1074         int err;
1075         unsigned int maxfraglen, fragheaderlen, fraggap;
1076
1077         if (inet->hdrincl)
1078                 return -EPERM;
1079
1080         if (flags&MSG_PROBE)
1081                 return 0;
1082
1083         if (skb_queue_empty(&sk->sk_write_queue))
1084                 return -EINVAL;
1085
1086         rt = (struct rtable *)inet->cork.dst;
1087         if (inet->cork.flags & IPCORK_OPT)
1088                 opt = inet->cork.opt;
1089
1090         if (!(rt->u.dst.dev->features&NETIF_F_SG))
1091                 return -EOPNOTSUPP;
1092
1093         hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1094         mtu = inet->cork.fragsize;
1095
1096         fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1097         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1098
1099         if (inet->cork.length + size > 0xFFFF - fragheaderlen) {
1100                 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu);
1101                 return -EMSGSIZE;
1102         }
1103
1104         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1105                 return -EINVAL;
1106
1107         inet->cork.length += size;
1108         if ((sk->sk_protocol == IPPROTO_UDP) &&
1109             (rt->u.dst.dev->features & NETIF_F_UFO)) {
1110                 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
1111                 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1112         }
1113
1114
1115         while (size > 0) {
1116                 int i;
1117
1118                 if (skb_is_gso(skb))
1119                         len = size;
1120                 else {
1121
1122                         /* Check if the remaining data fits into current packet. */
1123                         len = mtu - skb->len;
1124                         if (len < size)
1125                                 len = maxfraglen - skb->len;
1126                 }
1127                 if (len <= 0) {
1128                         struct sk_buff *skb_prev;
1129                         int alloclen;
1130
1131                         skb_prev = skb;
1132                         fraggap = skb_prev->len - maxfraglen;
1133
1134                         alloclen = fragheaderlen + hh_len + fraggap + 15;
1135                         skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1136                         if (unlikely(!skb)) {
1137                                 err = -ENOBUFS;
1138                                 goto error;
1139                         }
1140
1141                         /*
1142                          *      Fill in the control structures
1143                          */
1144                         skb->ip_summed = CHECKSUM_NONE;
1145                         skb->csum = 0;
1146                         skb_reserve(skb, hh_len);
1147
1148                         /*
1149                          *      Find where to start putting bytes.
1150                          */
1151                         skb_put(skb, fragheaderlen + fraggap);
1152                         skb_reset_network_header(skb);
1153                         skb->transport_header = (skb->network_header +
1154                                                  fragheaderlen);
1155                         if (fraggap) {
1156                                 skb->csum = skb_copy_and_csum_bits(skb_prev,
1157                                                                    maxfraglen,
1158                                                     skb_transport_header(skb),
1159                                                                    fraggap, 0);
1160                                 skb_prev->csum = csum_sub(skb_prev->csum,
1161                                                           skb->csum);
1162                                 pskb_trim_unique(skb_prev, maxfraglen);
1163                         }
1164
1165                         /*
1166                          * Put the packet on the pending queue.
1167                          */
1168                         __skb_queue_tail(&sk->sk_write_queue, skb);
1169                         continue;
1170                 }
1171
1172                 i = skb_shinfo(skb)->nr_frags;
1173                 if (len > size)
1174                         len = size;
1175                 if (skb_can_coalesce(skb, i, page, offset)) {
1176                         skb_shinfo(skb)->frags[i-1].size += len;
1177                 } else if (i < MAX_SKB_FRAGS) {
1178                         get_page(page);
1179                         skb_fill_page_desc(skb, i, page, offset, len);
1180                 } else {
1181                         err = -EMSGSIZE;
1182                         goto error;
1183                 }
1184
1185                 if (skb->ip_summed == CHECKSUM_NONE) {
1186                         __wsum csum;
1187                         csum = csum_page(page, offset, len);
1188                         skb->csum = csum_block_add(skb->csum, csum, skb->len);
1189                 }
1190
1191                 skb->len += len;
1192                 skb->data_len += len;
1193                 skb->truesize += len;
1194                 atomic_add(len, &sk->sk_wmem_alloc);
1195                 offset += len;
1196                 size -= len;
1197         }
1198         return 0;
1199
1200 error:
1201         inet->cork.length -= size;
1202         IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1203         return err;
1204 }
1205
1206 static void ip_cork_release(struct inet_sock *inet)
1207 {
1208         inet->cork.flags &= ~IPCORK_OPT;
1209         kfree(inet->cork.opt);
1210         inet->cork.opt = NULL;
1211         dst_release(inet->cork.dst);
1212         inet->cork.dst = NULL;
1213 }
1214
1215 /*
1216  *      Combined all pending IP fragments on the socket as one IP datagram
1217  *      and push them out.
1218  */
1219 int ip_push_pending_frames(struct sock *sk)
1220 {
1221         struct sk_buff *skb, *tmp_skb;
1222         struct sk_buff **tail_skb;
1223         struct inet_sock *inet = inet_sk(sk);
1224         struct ip_options *opt = NULL;
1225         struct rtable *rt = (struct rtable *)inet->cork.dst;
1226         struct iphdr *iph;
1227         __be16 df = 0;
1228         __u8 ttl;
1229         int err = 0;
1230
1231         if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1232                 goto out;
1233         tail_skb = &(skb_shinfo(skb)->frag_list);
1234
1235         /* move skb->data to ip header from ext header */
1236         if (skb->data < skb_network_header(skb))
1237                 __skb_pull(skb, skb_network_offset(skb));
1238         while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1239                 __skb_pull(tmp_skb, skb_network_header_len(skb));
1240                 *tail_skb = tmp_skb;
1241                 tail_skb = &(tmp_skb->next);
1242                 skb->len += tmp_skb->len;
1243                 skb->data_len += tmp_skb->len;
1244                 skb->truesize += tmp_skb->truesize;
1245                 __sock_put(tmp_skb->sk);
1246                 tmp_skb->destructor = NULL;
1247                 tmp_skb->sk = NULL;
1248         }
1249
1250         /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1251          * to fragment the frame generated here. No matter, what transforms
1252          * how transforms change size of the packet, it will come out.
1253          */
1254         if (inet->pmtudisc < IP_PMTUDISC_DO)
1255                 skb->local_df = 1;
1256
1257         /* DF bit is set when we want to see DF on outgoing frames.
1258          * If local_df is set too, we still allow to fragment this frame
1259          * locally. */
1260         if (inet->pmtudisc >= IP_PMTUDISC_DO ||
1261             (skb->len <= dst_mtu(&rt->u.dst) &&
1262              ip_dont_fragment(sk, &rt->u.dst)))
1263                 df = htons(IP_DF);
1264
1265         if (inet->cork.flags & IPCORK_OPT)
1266                 opt = inet->cork.opt;
1267
1268         if (rt->rt_type == RTN_MULTICAST)
1269                 ttl = inet->mc_ttl;
1270         else
1271                 ttl = ip_select_ttl(inet, &rt->u.dst);
1272
1273         iph = (struct iphdr *)skb->data;
1274         iph->version = 4;
1275         iph->ihl = 5;
1276         if (opt) {
1277                 iph->ihl += opt->optlen>>2;
1278                 ip_options_build(skb, opt, inet->cork.addr, rt, 0);
1279         }
1280         iph->tos = inet->tos;
1281         iph->frag_off = df;
1282         ip_select_ident(iph, &rt->u.dst, sk);
1283         iph->ttl = ttl;
1284         iph->protocol = sk->sk_protocol;
1285         iph->saddr = rt->rt_src;
1286         iph->daddr = rt->rt_dst;
1287
1288         skb->priority = sk->sk_priority;
1289         skb->mark = sk->sk_mark;
1290         skb->dst = dst_clone(&rt->u.dst);
1291
1292         if (iph->protocol == IPPROTO_ICMP)
1293                 icmp_out_count(((struct icmphdr *)
1294                         skb_transport_header(skb))->type);
1295
1296         /* Netfilter gets whole the not fragmented skb. */
1297         err = ip_local_out(skb);
1298         if (err) {
1299                 if (err > 0)
1300                         err = inet->recverr ? net_xmit_errno(err) : 0;
1301                 if (err)
1302                         goto error;
1303         }
1304
1305 out:
1306         ip_cork_release(inet);
1307         return err;
1308
1309 error:
1310         IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1311         goto out;
1312 }
1313
1314 /*
1315  *      Throw away all pending data on the socket.
1316  */
1317 void ip_flush_pending_frames(struct sock *sk)
1318 {
1319         struct sk_buff *skb;
1320
1321         while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL)
1322                 kfree_skb(skb);
1323
1324         ip_cork_release(inet_sk(sk));
1325 }
1326
1327
1328 /*
1329  *      Fetch data from kernel space and fill in checksum if needed.
1330  */
1331 static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1332                               int len, int odd, struct sk_buff *skb)
1333 {
1334         __wsum csum;
1335
1336         csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1337         skb->csum = csum_block_add(skb->csum, csum, odd);
1338         return 0;
1339 }
1340
1341 /*
1342  *      Generic function to send a packet as reply to another packet.
1343  *      Used to send TCP resets so far. ICMP should use this function too.
1344  *
1345  *      Should run single threaded per socket because it uses the sock
1346  *      structure to pass arguments.
1347  */
1348 void ip_send_reply(struct sock *sk, struct sk_buff *skb, struct ip_reply_arg *arg,
1349                    unsigned int len)
1350 {
1351         struct inet_sock *inet = inet_sk(sk);
1352         struct {
1353                 struct ip_options       opt;
1354                 char                    data[40];
1355         } replyopts;
1356         struct ipcm_cookie ipc;
1357         __be32 daddr;
1358         struct rtable *rt = skb->rtable;
1359
1360         if (ip_options_echo(&replyopts.opt, skb))
1361                 return;
1362
1363         daddr = ipc.addr = rt->rt_src;
1364         ipc.opt = NULL;
1365
1366         if (replyopts.opt.optlen) {
1367                 ipc.opt = &replyopts.opt;
1368
1369                 if (ipc.opt->srr)
1370                         daddr = replyopts.opt.faddr;
1371         }
1372
1373         {
1374                 struct flowi fl = { .oif = arg->bound_dev_if,
1375                                     .nl_u = { .ip4_u =
1376                                               { .daddr = daddr,
1377                                                 .saddr = rt->rt_spec_dst,
1378                                                 .tos = RT_TOS(ip_hdr(skb)->tos) } },
1379                                     /* Not quite clean, but right. */
1380                                     .uli_u = { .ports =
1381                                                { .sport = tcp_hdr(skb)->dest,
1382                                                  .dport = tcp_hdr(skb)->source } },
1383                                     .proto = sk->sk_protocol };
1384                 security_skb_classify_flow(skb, &fl);
1385                 if (ip_route_output_key(sk->sk_net, &rt, &fl))
1386                         return;
1387         }
1388
1389         /* And let IP do all the hard work.
1390
1391            This chunk is not reenterable, hence spinlock.
1392            Note that it uses the fact, that this function is called
1393            with locally disabled BH and that sk cannot be already spinlocked.
1394          */
1395         bh_lock_sock(sk);
1396         inet->tos = ip_hdr(skb)->tos;
1397         sk->sk_priority = skb->priority;
1398         sk->sk_protocol = ip_hdr(skb)->protocol;
1399         sk->sk_bound_dev_if = arg->bound_dev_if;
1400         ip_append_data(sk, ip_reply_glue_bits, arg->iov->iov_base, len, 0,
1401                        &ipc, rt, MSG_DONTWAIT);
1402         if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
1403                 if (arg->csumoffset >= 0)
1404                         *((__sum16 *)skb_transport_header(skb) +
1405                           arg->csumoffset) = csum_fold(csum_add(skb->csum,
1406                                                                 arg->csum));
1407                 skb->ip_summed = CHECKSUM_NONE;
1408                 ip_push_pending_frames(sk);
1409         }
1410
1411         bh_unlock_sock(sk);
1412
1413         ip_rt_put(rt);
1414 }
1415
1416 void __init ip_init(void)
1417 {
1418         ip_rt_init();
1419         inet_initpeers();
1420
1421 #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1422         igmp_mc_proc_init();
1423 #endif
1424 }
1425
1426 EXPORT_SYMBOL(ip_generic_getfrag);
1427 EXPORT_SYMBOL(ip_queue_xmit);
1428 EXPORT_SYMBOL(ip_send_check);