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/sched.h>
  53 #include <linux/mm.h>
  54 #include <linux/string.h>
  55 #include <linux/errno.h>
  56 #include <linux/config.h>
  57
  58 #include <linux/socket.h>
  59 #include <linux/sockios.h>
  60 #include <linux/in.h>
  61 #include <linux/inet.h>
  62 #include <linux/netdevice.h>
  63 #include <linux/etherdevice.h>
  64 #include <linux/proc_fs.h>
  65 #include <linux/stat.h>
  66 #include <linux/init.h>
  67
  68 #include <net/snmp.h>
  69 #include <net/ip.h>
  70 #include <net/protocol.h>
  71 #include <net/route.h>
  72 #include <net/xfrm.h>
  73 #include <linux/skbuff.h>
  74 #include <net/sock.h>
  75 #include <net/arp.h>
  76 #include <net/icmp.h>
  77 #include <net/checksum.h>
  78 #include <net/inetpeer.h>
  79 #include <net/checksum.h>
  80 #include <linux/igmp.h>
  81 #include <linux/netfilter_ipv4.h>
  82 #include <linux/netfilter_bridge.h>
  83 #include <linux/mroute.h>
  84 #include <linux/netlink.h>
  85 #include <linux/tcp.h>
  86
  87 int sysctl_ip_default_ttl = IPDEFTTL;
  88
  89 /* Generate a checksum for an outgoing IP datagram. */
  90 __inline__ void ip_send_check(struct iphdr *iph)
  91 {
  92         iph->check = 0;
  93         iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
  94 }
  95
  96 /* dev_loopback_xmit for use with netfilter. */
  97 static int ip_dev_loopback_xmit(struct sk_buff *newskb)
  98 {
  99         newskb->mac.raw = newskb->data;
 100         __skb_pull(newskb, newskb->nh.raw - newskb->data);
 101         newskb->pkt_type = PACKET_LOOPBACK;
 102         newskb->ip_summed = CHECKSUM_UNNECESSARY;
 103         BUG_TRAP(newskb->dst);
 104         netif_rx(newskb);
 105         return 0;
 106 }
 107
 108 static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
 109 {
 110         int ttl = inet->uc_ttl;
 111
 112         if (ttl < 0)
 113                 ttl = dst_metric(dst, RTAX_HOPLIMIT);
 114         return ttl;
 115 }
 116
 117 /*
 118  *              Add an ip header to a skbuff and send it out.
 119  *
 120  */
 121 int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk,
 122                           u32 saddr, u32 daddr, struct ip_options *opt)
 123 {
 124         struct inet_sock *inet = inet_sk(sk);
 125         struct rtable *rt = (struct rtable *)skb->dst;
 126         struct iphdr *iph;
 127
 128         /* Build the IP header. */
 129         if (opt)
 130                 iph=(struct iphdr *)skb_push(skb,sizeof(struct iphdr) + opt->optlen);
 131         else
 132                 iph=(struct iphdr *)skb_push(skb,sizeof(struct iphdr));
 133
 134         iph->version  = 4;
 135         iph->ihl      = 5;
 136         iph->tos      = inet->tos;
 137         if (ip_dont_fragment(sk, &rt->u.dst))
 138                 iph->frag_off = htons(IP_DF);
 139         else
 140                 iph->frag_off = 0;
 141         iph->ttl      = ip_select_ttl(inet, &rt->u.dst);
 142         iph->daddr    = rt->rt_dst;
 143         iph->saddr    = rt->rt_src;
 144         iph->protocol = sk->sk_protocol;
 145         iph->tot_len  = htons(skb->len);
 146         ip_select_ident(iph, &rt->u.dst, sk);
 147         skb->nh.iph   = iph;
 148
 149         if (opt && opt->optlen) {
 150                 iph->ihl += opt->optlen>>2;
 151                 ip_options_build(skb, opt, daddr, rt, 0);
 152         }
 153         ip_send_check(iph);
 154
 155         skb->priority = sk->sk_priority;
 156
 157         /* Send it out. */
 158         return NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL, rt->u.dst.dev,
 159                        dst_output);
 160 }
 161
 162 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
 163
 164 static inline int ip_finish_output2(struct sk_buff *skb)
 165 {
 166         struct dst_entry *dst = skb->dst;
 167         struct hh_cache *hh = dst->hh;
 168         struct net_device *dev = dst->dev;
 169         int hh_len = LL_RESERVED_SPACE(dev);
 170
 171         /* Be paranoid, rather than too clever. */
 172         if (unlikely(skb_headroom(skb) < hh_len && dev->hard_header)) {
 173                 struct sk_buff *skb2;
 174
 175                 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
 176                 if (skb2 == NULL) {
 177                         kfree_skb(skb);
 178                         return -ENOMEM;
 179                 }
 180                 if (skb->sk)
 181                         skb_set_owner_w(skb2, skb->sk);
 182                 kfree_skb(skb);
 183                 skb = skb2;
 184         }
 185
 186         if (hh) {
 187                 int hh_alen;
 188
 189                 read_lock_bh(&hh->hh_lock);
 190                 hh_alen = HH_DATA_ALIGN(hh->hh_len);
 191                 memcpy(skb->data - hh_alen, hh->hh_data, hh_alen);
 192                 read_unlock_bh(&hh->hh_lock);
 193                 skb_push(skb, hh->hh_len);
 194                 return hh->hh_output(skb);
 195         } else if (dst->neighbour)
 196                 return dst->neighbour->output(skb);
 197
 198         if (net_ratelimit())
 199                 printk(KERN_DEBUG "ip_finish_output2: No header cache and no neighbour!\n");
 200         kfree_skb(skb);
 201         return -EINVAL;
 202 }
 203
 204 static inline int ip_finish_output(struct sk_buff *skb)
 205 {
 206 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
 207         /* Policy lookup after SNAT yielded a new policy */
 208         if (skb->dst->xfrm != NULL) {
 209                 IPCB(skb)->flags |= IPSKB_REROUTED;
 210                 return dst_output(skb);
 211         }
 212 #endif
 213         if (skb->len > dst_mtu(skb->dst) &&
 214             !(skb_shinfo(skb)->ufo_size || skb_shinfo(skb)->tso_size))
 215                 return ip_fragment(skb, ip_finish_output2);
 216         else
 217                 return ip_finish_output2(skb);
 218 }
 219
 220 int ip_mc_output(struct sk_buff *skb)
 221 {
 222         struct sock *sk = skb->sk;
 223         struct rtable *rt = (struct rtable*)skb->dst;
 224         struct net_device *dev = rt->u.dst.dev;
 225
 226         /*
 227          *      If the indicated interface is up and running, send the packet.
 228          */
 229         IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
 230
 231         skb->dev = dev;
 232         skb->protocol = htons(ETH_P_IP);
 233
 234         /*
 235          *      Multicasts are looped back for other local users
 236          */
 237
 238         if (rt->rt_flags&RTCF_MULTICAST) {
 239                 if ((!sk || inet_sk(sk)->mc_loop)
 240 #ifdef CONFIG_IP_MROUTE
 241                 /* Small optimization: do not loopback not local frames,
 242                    which returned after forwarding; they will be  dropped
 243                    by ip_mr_input in any case.
 244                    Note, that local frames are looped back to be delivered
 245                    to local recipients.
 246
 247                    This check is duplicated in ip_mr_input at the moment.
 248                  */
 249                     && ((rt->rt_flags&RTCF_LOCAL) || !(IPCB(skb)->flags&IPSKB_FORWARDED))
 250 #endif
 251                 ) {
 252                         struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
 253                         if (newskb)
 254                                 NF_HOOK(PF_INET, NF_IP_POST_ROUTING, newskb, NULL,
 255                                         newskb->dev,
 256                                         ip_dev_loopback_xmit);
 257                 }
 258
 259                 /* Multicasts with ttl 0 must not go beyond the host */
 260
 261                 if (skb->nh.iph->ttl == 0) {
 262                         kfree_skb(skb);
 263                         return 0;
 264                 }
 265         }
 266
 267         if (rt->rt_flags&RTCF_BROADCAST) {
 268                 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
 269                 if (newskb)
 270                         NF_HOOK(PF_INET, NF_IP_POST_ROUTING, newskb, NULL,
 271                                 newskb->dev, ip_dev_loopback_xmit);
 272         }
 273
 274         return NF_HOOK_COND(PF_INET, NF_IP_POST_ROUTING, skb, NULL, skb->dev,
 275                             ip_finish_output,
 276                             !(IPCB(skb)->flags & IPSKB_REROUTED));
 277 }
 278
 279 int ip_output(struct sk_buff *skb)
 280 {
 281         struct net_device *dev = skb->dst->dev;
 282
 283         IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
 284
 285         skb->dev = dev;
 286         skb->protocol = htons(ETH_P_IP);
 287
 288         return NF_HOOK_COND(PF_INET, NF_IP_POST_ROUTING, skb, NULL, dev,
 289                             ip_finish_output,
 290                             !(IPCB(skb)->flags & IPSKB_REROUTED));
 291 }
 292
 293 int ip_queue_xmit(struct sk_buff *skb, int ipfragok)
 294 {
 295         struct sock *sk = skb->sk;
 296         struct inet_sock *inet = inet_sk(sk);
 297         struct ip_options *opt = inet->opt;
 298         struct rtable *rt;
 299         struct iphdr *iph;
 300
 301         /* Skip all of this if the packet is already routed,
 302          * f.e. by something like SCTP.
 303          */
 304         rt = (struct rtable *) skb->dst;
 305         if (rt != NULL)
 306                 goto packet_routed;
 307
 308         /* Make sure we can route this packet. */
 309         rt = (struct rtable *)__sk_dst_check(sk, 0);
 310         if (rt == NULL) {
 311                 u32 daddr;
 312
 313                 /* Use correct destination address if we have options. */
 314                 daddr = inet->daddr;
 315                 if(opt && opt->srr)
 316                         daddr = opt->faddr;
 317
 318                 {
 319                         struct flowi fl = { .oif = sk->sk_bound_dev_if,
 320                                             .nl_u = { .ip4_u =
 321                                                       { .daddr = daddr,
 322                                                         .saddr = inet->saddr,
 323                                                         .tos = RT_CONN_FLAGS(sk) } },
 324                                             .proto = sk->sk_protocol,
 325                                             .uli_u = { .ports =
 326                                                        { .sport = inet->sport,
 327                                                          .dport = inet->dport } } };
 328
 329                         /* If this fails, retransmit mechanism of transport layer will
 330                          * keep trying until route appears or the connection times
 331                          * itself out.
 332                          */
 333                         if (ip_route_output_flow(&rt, &fl, sk, 0))
 334                                 goto no_route;
 335                 }
 336                 sk_setup_caps(sk, &rt->u.dst);
 337         }
 338         skb->dst = dst_clone(&rt->u.dst);
 339
 340 packet_routed:
 341         if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
 342                 goto no_route;
 343
 344         /* OK, we know where to send it, allocate and build IP header. */
 345         iph = (struct iphdr *) skb_push(skb, sizeof(struct iphdr) + (opt ? opt->optlen : 0));
 346         *((__u16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
 347         iph->tot_len = htons(skb->len);
 348         if (ip_dont_fragment(sk, &rt->u.dst) && !ipfragok)
 349                 iph->frag_off = htons(IP_DF);
 350         else
 351                 iph->frag_off = 0;
 352         iph->ttl      = ip_select_ttl(inet, &rt->u.dst);
 353         iph->protocol = sk->sk_protocol;
 354         iph->saddr    = rt->rt_src;
 355         iph->daddr    = rt->rt_dst;
 356         skb->nh.iph   = iph;
 357         /* Transport layer set skb->h.foo itself. */
 358
 359         if (opt && opt->optlen) {
 360                 iph->ihl += opt->optlen >> 2;
 361                 ip_options_build(skb, opt, inet->daddr, rt, 0);
 362         }
 363
 364         ip_select_ident_more(iph, &rt->u.dst, sk,
 365                              (skb_shinfo(skb)->tso_segs ?: 1) - 1);
 366
 367         /* Add an IP checksum. */
 368         ip_send_check(iph);
 369
 370         skb->priority = sk->sk_priority;
 371
 372         return NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL, rt->u.dst.dev,
 373                        dst_output);
 374
 375 no_route:
 376         IP_INC_STATS(IPSTATS_MIB_OUTNOROUTES);
 377         kfree_skb(skb);
 378         return -EHOSTUNREACH;
 379 }
 380
 381
 382 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
 383 {
 384         to->pkt_type = from->pkt_type;
 385         to->priority = from->priority;
 386         to->protocol = from->protocol;
 387         dst_release(to->dst);
 388         to->dst = dst_clone(from->dst);
 389         to->dev = from->dev;
 390
 391         /* Copy the flags to each fragment. */
 392         IPCB(to)->flags = IPCB(from)->flags;
 393
 394 #ifdef CONFIG_NET_SCHED
 395         to->tc_index = from->tc_index;
 396 #endif
 397 #ifdef CONFIG_NETFILTER
 398         to->nfmark = from->nfmark;
 399         /* Connection association is same as pre-frag packet */
 400         nf_conntrack_put(to->nfct);
 401         to->nfct = from->nfct;
 402         nf_conntrack_get(to->nfct);
 403         to->nfctinfo = from->nfctinfo;
 404 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
 405         to->ipvs_property = from->ipvs_property;
 406 #endif
 407 #ifdef CONFIG_BRIDGE_NETFILTER
 408         nf_bridge_put(to->nf_bridge);
 409         to->nf_bridge = from->nf_bridge;
 410         nf_bridge_get(to->nf_bridge);
 411 #endif
 412 #endif
 413 }
 414
 415 /*
 416  *      This IP datagram is too large to be sent in one piece.  Break it up into
 417  *      smaller pieces (each of size equal to IP header plus
 418  *      a block of the data of the original IP data part) that will yet fit in a
 419  *      single device frame, and queue such a frame for sending.
 420  */
 421
 422 int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff*))
 423 {
 424         struct iphdr *iph;
 425         int raw = 0;
 426         int ptr;
 427         struct net_device *dev;
 428         struct sk_buff *skb2;
 429         unsigned int mtu, hlen, left, len, ll_rs;
 430         int offset;
 431         __be16 not_last_frag;
 432         struct rtable *rt = (struct rtable*)skb->dst;
 433         int err = 0;
 434
 435         dev = rt->u.dst.dev;
 436
 437         /*
 438          *      Point into the IP datagram header.
 439          */
 440
 441         iph = skb->nh.iph;
 442
 443         if (unlikely((iph->frag_off & htons(IP_DF)) && !skb->local_df)) {
 444                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
 445                           htonl(dst_mtu(&rt->u.dst)));
 446                 kfree_skb(skb);
 447                 return -EMSGSIZE;
 448         }
 449
 450         /*
 451          *      Setup starting values.
 452          */
 453
 454         hlen = iph->ihl * 4;
 455         mtu = dst_mtu(&rt->u.dst) - hlen;       /* Size of data space */
 456         IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
 457
 458         /* When frag_list is given, use it. First, check its validity:
 459          * some transformers could create wrong frag_list or break existing
 460          * one, it is not prohibited. In this case fall back to copying.
 461          *
 462          * LATER: this step can be merged to real generation of fragments,
 463          * we can switch to copy when see the first bad fragment.
 464          */
 465         if (skb_shinfo(skb)->frag_list) {
 466                 struct sk_buff *frag;
 467                 int first_len = skb_pagelen(skb);
 468
 469                 if (first_len - hlen > mtu ||
 470                     ((first_len - hlen) & 7) ||
 471                     (iph->frag_off & htons(IP_MF|IP_OFFSET)) ||
 472                     skb_cloned(skb))
 473                         goto slow_path;
 474
 475                 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
 476                         /* Correct geometry. */
 477                         if (frag->len > mtu ||
 478                             ((frag->len & 7) && frag->next) ||
 479                             skb_headroom(frag) < hlen)
 480                             goto slow_path;
 481
 482                         /* Partially cloned skb? */
 483                         if (skb_shared(frag))
 484                                 goto slow_path;
 485
 486                         BUG_ON(frag->sk);
 487                         if (skb->sk) {
 488                                 sock_hold(skb->sk);
 489                                 frag->sk = skb->sk;
 490                                 frag->destructor = sock_wfree;
 491                                 skb->truesize -= frag->truesize;
 492                         }
 493                 }
 494
 495                 /* Everything is OK. Generate! */
 496
 497                 err = 0;
 498                 offset = 0;
 499                 frag = skb_shinfo(skb)->frag_list;
 500                 skb_shinfo(skb)->frag_list = NULL;
 501                 skb->data_len = first_len - skb_headlen(skb);
 502                 skb->len = first_len;
 503                 iph->tot_len = htons(first_len);
 504                 iph->frag_off = htons(IP_MF);
 505                 ip_send_check(iph);
 506
 507                 for (;;) {
 508                         /* Prepare header of the next frame,
 509                          * before previous one went down. */
 510                         if (frag) {
 511                                 frag->ip_summed = CHECKSUM_NONE;
 512                                 frag->h.raw = frag->data;
 513                                 frag->nh.raw = __skb_push(frag, hlen);
 514                                 memcpy(frag->nh.raw, iph, hlen);
 515                                 iph = frag->nh.iph;
 516                                 iph->tot_len = htons(frag->len);
 517                                 ip_copy_metadata(frag, skb);
 518                                 if (offset == 0)
 519                                         ip_options_fragment(frag);
 520                                 offset += skb->len - hlen;
 521                                 iph->frag_off = htons(offset>>3);
 522                                 if (frag->next != NULL)
 523                                         iph->frag_off |= htons(IP_MF);
 524                                 /* Ready, complete checksum */
 525                                 ip_send_check(iph);
 526                         }
 527
 528                         err = output(skb);
 529
 530                         if (err || !frag)
 531                                 break;
 532
 533                         skb = frag;
 534                         frag = skb->next;
 535                         skb->next = NULL;
 536                 }
 537
 538                 if (err == 0) {
 539                         IP_INC_STATS(IPSTATS_MIB_FRAGOKS);
 540                         return 0;
 541                 }
 542
 543                 while (frag) {
 544                         skb = frag->next;
 545                         kfree_skb(frag);
 546                         frag = skb;
 547                 }
 548                 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS);
 549                 return err;
 550         }
 551
 552 slow_path:
 553         left = skb->len - hlen;         /* Space per frame */
 554         ptr = raw + hlen;               /* Where to start from */
 555
 556 #ifdef CONFIG_BRIDGE_NETFILTER
 557         /* for bridged IP traffic encapsulated inside f.e. a vlan header,
 558          * we need to make room for the encapsulating header */
 559         ll_rs = LL_RESERVED_SPACE_EXTRA(rt->u.dst.dev, nf_bridge_pad(skb));
 560         mtu -= nf_bridge_pad(skb);
 561 #else
 562         ll_rs = LL_RESERVED_SPACE(rt->u.dst.dev);
 563 #endif
 564         /*
 565          *      Fragment the datagram.
 566          */
 567
 568         offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
 569         not_last_frag = iph->frag_off & htons(IP_MF);
 570
 571         /*
 572          *      Keep copying data until we run out.
 573          */
 574
 575         while(left > 0) {
 576                 len = left;
 577                 /* IF: it doesn't fit, use 'mtu' - the data space left */
 578                 if (len > mtu)
 579                         len = mtu;
 580                 /* IF: we are not sending upto and including the packet end
 581                    then align the next start on an eight byte boundary */
 582                 if (len < left) {
 583                         len &= ~7;
 584                 }
 585                 /*
 586                  *      Allocate buffer.
 587                  */
 588
 589                 if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) {
 590                         NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n");
 591                         err = -ENOMEM;
 592                         goto fail;
 593                 }
 594
 595                 /*
 596                  *      Set up data on packet
 597                  */
 598
 599                 ip_copy_metadata(skb2, skb);
 600                 skb_reserve(skb2, ll_rs);
 601                 skb_put(skb2, len + hlen);
 602                 skb2->nh.raw = skb2->data;
 603                 skb2->h.raw = skb2->data + hlen;
 604
 605                 /*
 606                  *      Charge the memory for the fragment to any owner
 607                  *      it might possess
 608                  */
 609
 610                 if (skb->sk)
 611                         skb_set_owner_w(skb2, skb->sk);
 612
 613                 /*
 614                  *      Copy the packet header into the new buffer.
 615                  */
 616
 617                 memcpy(skb2->nh.raw, skb->data, hlen);
 618
 619                 /*
 620                  *      Copy a block of the IP datagram.
 621                  */
 622                 if (skb_copy_bits(skb, ptr, skb2->h.raw, len))
 623                         BUG();
 624                 left -= len;
 625
 626                 /*
 627                  *      Fill in the new header fields.
 628                  */
 629                 iph = skb2->nh.iph;
 630                 iph->frag_off = htons((offset >> 3));
 631
 632                 /* ANK: dirty, but effective trick. Upgrade options only if
 633                  * the segment to be fragmented was THE FIRST (otherwise,
 634                  * options are already fixed) and make it ONCE
 635                  * on the initial skb, so that all the following fragments
 636                  * will inherit fixed options.
 637                  */
 638                 if (offset == 0)
 639                         ip_options_fragment(skb);
 640
 641                 /*
 642                  *      Added AC : If we are fragmenting a fragment that's not the
 643                  *                 last fragment then keep MF on each bit
 644                  */
 645                 if (left > 0 || not_last_frag)
 646                         iph->frag_off |= htons(IP_MF);
 647                 ptr += len;
 648                 offset += len;
 649
 650                 /*
 651                  *      Put this fragment into the sending queue.
 652                  */
 653
 654                 IP_INC_STATS(IPSTATS_MIB_FRAGCREATES);
 655
 656                 iph->tot_len = htons(len + hlen);
 657
 658                 ip_send_check(iph);
 659
 660                 err = output(skb2);
 661                 if (err)
 662                         goto fail;
 663         }
 664         kfree_skb(skb);
 665         IP_INC_STATS(IPSTATS_MIB_FRAGOKS);
 666         return err;
 667
 668 fail:
 669         kfree_skb(skb);
 670         IP_INC_STATS(IPSTATS_MIB_FRAGFAILS);
 671         return err;
 672 }
 673
 674 EXPORT_SYMBOL(ip_fragment);
 675
 676 int
 677 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
 678 {
 679         struct iovec *iov = from;
 680
 681         if (skb->ip_summed == CHECKSUM_HW) {
 682                 if (memcpy_fromiovecend(to, iov, offset, len) < 0)
 683                         return -EFAULT;
 684         } else {
 685                 unsigned int csum = 0;
 686                 if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0)
 687                         return -EFAULT;
 688                 skb->csum = csum_block_add(skb->csum, csum, odd);
 689         }
 690         return 0;
 691 }
 692
 693 static inline unsigned int
 694 csum_page(struct page *page, int offset, int copy)
 695 {
 696         char *kaddr;
 697         unsigned int csum;
 698         kaddr = kmap(page);
 699         csum = csum_partial(kaddr + offset, copy, 0);
 700         kunmap(page);
 701         return csum;
 702 }
 703
 704 static inline int ip_ufo_append_data(struct sock *sk,
 705                         int getfrag(void *from, char *to, int offset, int len,
 706                                int odd, struct sk_buff *skb),
 707                         void *from, int length, int hh_len, int fragheaderlen,
 708                         int transhdrlen, int mtu,unsigned int flags)
 709 {
 710         struct sk_buff *skb;
 711         int err;
 712
 713         /* There is support for UDP fragmentation offload by network
 714          * device, so create one single skb packet containing complete
 715          * udp datagram
 716          */
 717         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
 718                 skb = sock_alloc_send_skb(sk,
 719                         hh_len + fragheaderlen + transhdrlen + 20,
 720                         (flags & MSG_DONTWAIT), &err);
 721
 722                 if (skb == NULL)
 723                         return err;
 724
 725                 /* reserve space for Hardware header */
 726                 skb_reserve(skb, hh_len);
 727
 728                 /* create space for UDP/IP header */
 729                 skb_put(skb,fragheaderlen + transhdrlen);
 730
 731                 /* initialize network header pointer */
 732                 skb->nh.raw = skb->data;
 733
 734                 /* initialize protocol header pointer */
 735                 skb->h.raw = skb->data + fragheaderlen;
 736
 737                 skb->ip_summed = CHECKSUM_HW;
 738                 skb->csum = 0;
 739                 sk->sk_sndmsg_off = 0;
 740         }
 741
 742         err = skb_append_datato_frags(sk,skb, getfrag, from,
 743                                (length - transhdrlen));
 744         if (!err) {
 745                 /* specify the length of each IP datagram fragment*/
 746                 skb_shinfo(skb)->ufo_size = (mtu - fragheaderlen);
 747                 __skb_queue_tail(&sk->sk_write_queue, skb);
 748
 749                 return 0;
 750         }
 751         /* There is not enough support do UFO ,
 752          * so follow normal path
 753          */
 754         kfree_skb(skb);
 755         return err;
 756 }
 757
 758 /*
 759  *      ip_append_data() and ip_append_page() can make one large IP datagram
 760  *      from many pieces of data. Each pieces will be holded on the socket
 761  *      until ip_push_pending_frames() is called. Each piece can be a page
 762  *      or non-page data.
 763  *
 764  *      Not only UDP, other transport protocols - e.g. raw sockets - can use
 765  *      this interface potentially.
 766  *
 767  *      LATER: length must be adjusted by pad at tail, when it is required.
 768  */
 769 int ip_append_data(struct sock *sk,
 770                    int getfrag(void *from, char *to, int offset, int len,
 771                                int odd, struct sk_buff *skb),
 772                    void *from, int length, int transhdrlen,
 773                    struct ipcm_cookie *ipc, struct rtable *rt,
 774                    unsigned int flags)
 775 {
 776         struct inet_sock *inet = inet_sk(sk);
 777         struct sk_buff *skb;
 778
 779         struct ip_options *opt = NULL;
 780         int hh_len;
 781         int exthdrlen;
 782         int mtu;
 783         int copy;
 784         int err;
 785         int offset = 0;
 786         unsigned int maxfraglen, fragheaderlen;
 787         int csummode = CHECKSUM_NONE;
 788
 789         if (flags&MSG_PROBE)
 790                 return 0;
 791
 792         if (skb_queue_empty(&sk->sk_write_queue)) {
 793                 /*
 794                  * setup for corking.
 795                  */
 796                 opt = ipc->opt;
 797                 if (opt) {
 798                         if (inet->cork.opt == NULL) {
 799                                 inet->cork.opt = kmalloc(sizeof(struct ip_options) + 40, sk->sk_allocation);
 800                                 if (unlikely(inet->cork.opt == NULL))
 801                                         return -ENOBUFS;
 802                         }
 803                         memcpy(inet->cork.opt, opt, sizeof(struct ip_options)+opt->optlen);
 804                         inet->cork.flags |= IPCORK_OPT;
 805                         inet->cork.addr = ipc->addr;
 806                 }
 807                 dst_hold(&rt->u.dst);
 808                 inet->cork.fragsize = mtu = dst_mtu(rt->u.dst.path);
 809                 inet->cork.rt = rt;
 810                 inet->cork.length = 0;
 811                 sk->sk_sndmsg_page = NULL;
 812                 sk->sk_sndmsg_off = 0;
 813                 if ((exthdrlen = rt->u.dst.header_len) != 0) {
 814                         length += exthdrlen;
 815                         transhdrlen += exthdrlen;
 816                 }
 817         } else {
 818                 rt = inet->cork.rt;
 819                 if (inet->cork.flags & IPCORK_OPT)
 820                         opt = inet->cork.opt;
 821
 822                 transhdrlen = 0;
 823                 exthdrlen = 0;
 824                 mtu = inet->cork.fragsize;
 825         }
 826         hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
 827
 828         fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
 829         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
 830
 831         if (inet->cork.length + length > 0xFFFF - fragheaderlen) {
 832                 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu-exthdrlen);
 833                 return -EMSGSIZE;
 834         }
 835
 836         /*
 837          * transhdrlen > 0 means that this is the first fragment and we wish
 838          * it won't be fragmented in the future.
 839          */
 840         if (transhdrlen &&
 841             length + fragheaderlen <= mtu &&
 842             rt->u.dst.dev->features&(NETIF_F_IP_CSUM|NETIF_F_NO_CSUM|NETIF_F_HW_CSUM) &&
 843             !exthdrlen)
 844                 csummode = CHECKSUM_HW;
 845
 846         inet->cork.length += length;
 847         if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
 848                         (rt->u.dst.dev->features & NETIF_F_UFO)) {
 849
 850                 err = ip_ufo_append_data(sk, getfrag, from, length, hh_len,
 851                                          fragheaderlen, transhdrlen, mtu,
 852                                          flags);
 853                 if (err)
 854                         goto error;
 855                 return 0;
 856         }
 857
 858         /* So, what's going on in the loop below?
 859          *
 860          * We use calculated fragment length to generate chained skb,
 861          * each of segments is IP fragment ready for sending to network after
 862          * adding appropriate IP header.
 863          */
 864
 865         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
 866                 goto alloc_new_skb;
 867
 868         while (length > 0) {
 869                 /* Check if the remaining data fits into current packet. */
 870                 copy = mtu - skb->len;
 871                 if (copy < length)
 872                         copy = maxfraglen - skb->len;
 873                 if (copy <= 0) {
 874                         char *data;
 875                         unsigned int datalen;
 876                         unsigned int fraglen;
 877                         unsigned int fraggap;
 878                         unsigned int alloclen;
 879                         struct sk_buff *skb_prev;
 880 alloc_new_skb:
 881                         skb_prev = skb;
 882                         if (skb_prev)
 883                                 fraggap = skb_prev->len - maxfraglen;
 884                         else
 885                                 fraggap = 0;
 886
 887                         /*
 888                          * If remaining data exceeds the mtu,
 889                          * we know we need more fragment(s).
 890                          */
 891                         datalen = length + fraggap;
 892                         if (datalen > mtu - fragheaderlen)
 893                                 datalen = maxfraglen - fragheaderlen;
 894                         fraglen = datalen + fragheaderlen;
 895
 896                         if ((flags & MSG_MORE) &&
 897                             !(rt->u.dst.dev->features&NETIF_F_SG))
 898                                 alloclen = mtu;
 899                         else
 900                                 alloclen = datalen + fragheaderlen;
 901
 902                         /* The last fragment gets additional space at tail.
 903                          * Note, with MSG_MORE we overallocate on fragments,
 904                          * because we have no idea what fragment will be
 905                          * the last.
 906                          */
 907                         if (datalen == length + fraggap)
 908                                 alloclen += rt->u.dst.trailer_len;
 909
 910                         if (transhdrlen) {
 911                                 skb = sock_alloc_send_skb(sk,
 912                                                 alloclen + hh_len + 15,
 913                                                 (flags & MSG_DONTWAIT), &err);
 914                         } else {
 915                                 skb = NULL;
 916                                 if (atomic_read(&sk->sk_wmem_alloc) <=
 917                                     2 * sk->sk_sndbuf)
 918                                         skb = sock_wmalloc(sk,
 919                                                            alloclen + hh_len + 15, 1,
 920                                                            sk->sk_allocation);
 921                                 if (unlikely(skb == NULL))
 922                                         err = -ENOBUFS;
 923                         }
 924                         if (skb == NULL)
 925                                 goto error;
 926
 927                         /*
 928                          *      Fill in the control structures
 929                          */
 930                         skb->ip_summed = csummode;
 931                         skb->csum = 0;
 932                         skb_reserve(skb, hh_len);
 933
 934                         /*
 935                          *      Find where to start putting bytes.
 936                          */
 937                         data = skb_put(skb, fraglen);
 938                         skb->nh.raw = data + exthdrlen;
 939                         data += fragheaderlen;
 940                         skb->h.raw = data + exthdrlen;
 941
 942                         if (fraggap) {
 943                                 skb->csum = skb_copy_and_csum_bits(
 944                                         skb_prev, maxfraglen,
 945                                         data + transhdrlen, fraggap, 0);
 946                                 skb_prev->csum = csum_sub(skb_prev->csum,
 947                                                           skb->csum);
 948                                 data += fraggap;
 949                                 skb_trim(skb_prev, maxfraglen);
 950                         }
 951
 952                         copy = datalen - transhdrlen - fraggap;
 953                         if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
 954                                 err = -EFAULT;
 955                                 kfree_skb(skb);
 956                                 goto error;
 957                         }
 958
 959                         offset += copy;
 960                         length -= datalen - fraggap;
 961                         transhdrlen = 0;
 962                         exthdrlen = 0;
 963                         csummode = CHECKSUM_NONE;
 964
 965                         /*
 966                          * Put the packet on the pending queue.
 967                          */
 968                         __skb_queue_tail(&sk->sk_write_queue, skb);
 969                         continue;
 970                 }
 971
 972                 if (copy > length)
 973                         copy = length;
 974
 975                 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
 976                         unsigned int off;
 977
 978                         off = skb->len;
 979                         if (getfrag(from, skb_put(skb, copy),
 980                                         offset, copy, off, skb) < 0) {
 981                                 __skb_trim(skb, off);
 982                                 err = -EFAULT;
 983                                 goto error;
 984                         }
 985                 } else {
 986                         int i = skb_shinfo(skb)->nr_frags;
 987                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
 988                         struct page *page = sk->sk_sndmsg_page;
 989                         int off = sk->sk_sndmsg_off;
 990                         unsigned int left;
 991
 992                         if (page && (left = PAGE_SIZE - off) > 0) {
 993                                 if (copy >= left)
 994                                         copy = left;
 995                                 if (page != frag->page) {
 996                                         if (i == MAX_SKB_FRAGS) {
 997                                                 err = -EMSGSIZE;
 998                                                 goto error;
 999                                         }
1000                                         get_page(page);
1001                                         skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1002                                         frag = &skb_shinfo(skb)->frags[i];
1003                                 }
1004                         } else if (i < MAX_SKB_FRAGS) {
1005                                 if (copy > PAGE_SIZE)
1006                                         copy = PAGE_SIZE;
1007                                 page = alloc_pages(sk->sk_allocation, 0);
1008                                 if (page == NULL)  {
1009                                         err = -ENOMEM;
1010                                         goto error;
1011                                 }
1012                                 sk->sk_sndmsg_page = page;
1013                                 sk->sk_sndmsg_off = 0;
1014
1015                                 skb_fill_page_desc(skb, i, page, 0, 0);
1016                                 frag = &skb_shinfo(skb)->frags[i];
1017                                 skb->truesize += PAGE_SIZE;
1018                                 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1019                         } else {
1020                                 err = -EMSGSIZE;
1021                                 goto error;
1022                         }
1023                         if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1024                                 err = -EFAULT;
1025                                 goto error;
1026                         }
1027                         sk->sk_sndmsg_off += copy;
1028                         frag->size += copy;
1029                         skb->len += copy;
1030                         skb->data_len += copy;
1031                 }
1032                 offset += copy;
1033                 length -= copy;
1034         }
1035
1036         return 0;
1037
1038 error:
1039         inet->cork.length -= length;
1040         IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1041         return err;
1042 }
1043
1044 ssize_t ip_append_page(struct sock *sk, struct page *page,
1045                        int offset, size_t size, int flags)
1046 {
1047         struct inet_sock *inet = inet_sk(sk);
1048         struct sk_buff *skb;
1049         struct rtable *rt;
1050         struct ip_options *opt = NULL;
1051         int hh_len;
1052         int mtu;
1053         int len;
1054         int err;
1055         unsigned int maxfraglen, fragheaderlen, fraggap;
1056
1057         if (inet->hdrincl)
1058                 return -EPERM;
1059
1060         if (flags&MSG_PROBE)
1061                 return 0;
1062
1063         if (skb_queue_empty(&sk->sk_write_queue))
1064                 return -EINVAL;
1065
1066         rt = inet->cork.rt;
1067         if (inet->cork.flags & IPCORK_OPT)
1068                 opt = inet->cork.opt;
1069
1070         if (!(rt->u.dst.dev->features&NETIF_F_SG))
1071                 return -EOPNOTSUPP;
1072
1073         hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1074         mtu = inet->cork.fragsize;
1075
1076         fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1077         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1078
1079         if (inet->cork.length + size > 0xFFFF - fragheaderlen) {
1080                 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu);
1081                 return -EMSGSIZE;
1082         }
1083
1084         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1085                 return -EINVAL;
1086
1087         inet->cork.length += size;
1088         if ((sk->sk_protocol == IPPROTO_UDP) &&
1089             (rt->u.dst.dev->features & NETIF_F_UFO))
1090                 skb_shinfo(skb)->ufo_size = (mtu - fragheaderlen);
1091
1092
1093         while (size > 0) {
1094                 int i;
1095
1096                 if (skb_shinfo(skb)->ufo_size)
1097                         len = size;
1098                 else {
1099
1100                         /* Check if the remaining data fits into current packet. */
1101                         len = mtu - skb->len;
1102                         if (len < size)
1103                                 len = maxfraglen - skb->len;
1104                 }
1105                 if (len <= 0) {
1106                         struct sk_buff *skb_prev;
1107                         char *data;
1108                         struct iphdr *iph;
1109                         int alloclen;
1110
1111                         skb_prev = skb;
1112                         fraggap = skb_prev->len - maxfraglen;
1113
1114                         alloclen = fragheaderlen + hh_len + fraggap + 15;
1115                         skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1116                         if (unlikely(!skb)) {
1117                                 err = -ENOBUFS;
1118                                 goto error;
1119                         }
1120
1121                         /*
1122                          *      Fill in the control structures
1123                          */
1124                         skb->ip_summed = CHECKSUM_NONE;
1125                         skb->csum = 0;
1126                         skb_reserve(skb, hh_len);
1127
1128                         /*
1129                          *      Find where to start putting bytes.
1130                          */
1131                         data = skb_put(skb, fragheaderlen + fraggap);
1132                         skb->nh.iph = iph = (struct iphdr *)data;
1133                         data += fragheaderlen;
1134                         skb->h.raw = data;
1135
1136                         if (fraggap) {
1137                                 skb->csum = skb_copy_and_csum_bits(
1138                                         skb_prev, maxfraglen,
1139                                         data, fraggap, 0);
1140                                 skb_prev->csum = csum_sub(skb_prev->csum,
1141                                                           skb->csum);
1142                                 skb_trim(skb_prev, maxfraglen);
1143                         }
1144
1145                         /*
1146                          * Put the packet on the pending queue.
1147                          */
1148                         __skb_queue_tail(&sk->sk_write_queue, skb);
1149                         continue;
1150                 }
1151
1152                 i = skb_shinfo(skb)->nr_frags;
1153                 if (len > size)
1154                         len = size;
1155                 if (skb_can_coalesce(skb, i, page, offset)) {
1156                         skb_shinfo(skb)->frags[i-1].size += len;
1157                 } else if (i < MAX_SKB_FRAGS) {
1158                         get_page(page);
1159                         skb_fill_page_desc(skb, i, page, offset, len);
1160                 } else {
1161                         err = -EMSGSIZE;
1162                         goto error;
1163                 }
1164
1165                 if (skb->ip_summed == CHECKSUM_NONE) {
1166                         unsigned int csum;
1167                         csum = csum_page(page, offset, len);
1168                         skb->csum = csum_block_add(skb->csum, csum, skb->len);
1169                 }
1170
1171                 skb->len += len;
1172                 skb->data_len += len;
1173                 offset += len;
1174                 size -= len;
1175         }
1176         return 0;
1177
1178 error:
1179         inet->cork.length -= size;
1180         IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1181         return err;
1182 }
1183
1184 /*
1185  *      Combined all pending IP fragments on the socket as one IP datagram
1186  *      and push them out.
1187  */
1188 int ip_push_pending_frames(struct sock *sk)
1189 {
1190         struct sk_buff *skb, *tmp_skb;
1191         struct sk_buff **tail_skb;
1192         struct inet_sock *inet = inet_sk(sk);
1193         struct ip_options *opt = NULL;
1194         struct rtable *rt = inet->cork.rt;
1195         struct iphdr *iph;
1196         __be16 df = 0;
1197         __u8 ttl;
1198         int err = 0;
1199
1200         if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1201                 goto out;
1202         tail_skb = &(skb_shinfo(skb)->frag_list);
1203
1204         /* move skb->data to ip header from ext header */
1205         if (skb->data < skb->nh.raw)
1206                 __skb_pull(skb, skb->nh.raw - skb->data);
1207         while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1208                 __skb_pull(tmp_skb, skb->h.raw - skb->nh.raw);
1209                 *tail_skb = tmp_skb;
1210                 tail_skb = &(tmp_skb->next);
1211                 skb->len += tmp_skb->len;
1212                 skb->data_len += tmp_skb->len;
1213                 skb->truesize += tmp_skb->truesize;
1214                 __sock_put(tmp_skb->sk);
1215                 tmp_skb->destructor = NULL;
1216                 tmp_skb->sk = NULL;
1217         }
1218
1219         /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1220          * to fragment the frame generated here. No matter, what transforms
1221          * how transforms change size of the packet, it will come out.
1222          */
1223         if (inet->pmtudisc != IP_PMTUDISC_DO)
1224                 skb->local_df = 1;
1225
1226         /* DF bit is set when we want to see DF on outgoing frames.
1227          * If local_df is set too, we still allow to fragment this frame
1228          * locally. */
1229         if (inet->pmtudisc == IP_PMTUDISC_DO ||
1230             (skb->len <= dst_mtu(&rt->u.dst) &&
1231              ip_dont_fragment(sk, &rt->u.dst)))
1232                 df = htons(IP_DF);
1233
1234         if (inet->cork.flags & IPCORK_OPT)
1235                 opt = inet->cork.opt;
1236
1237         if (rt->rt_type == RTN_MULTICAST)
1238                 ttl = inet->mc_ttl;
1239         else
1240                 ttl = ip_select_ttl(inet, &rt->u.dst);
1241
1242         iph = (struct iphdr *)skb->data;
1243         iph->version = 4;
1244         iph->ihl = 5;
1245         if (opt) {
1246                 iph->ihl += opt->optlen>>2;
1247                 ip_options_build(skb, opt, inet->cork.addr, rt, 0);
1248         }
1249         iph->tos = inet->tos;
1250         iph->tot_len = htons(skb->len);
1251         iph->frag_off = df;
1252         ip_select_ident(iph, &rt->u.dst, sk);
1253         iph->ttl = ttl;
1254         iph->protocol = sk->sk_protocol;
1255         iph->saddr = rt->rt_src;
1256         iph->daddr = rt->rt_dst;
1257         ip_send_check(iph);
1258
1259         skb->priority = sk->sk_priority;
1260         skb->dst = dst_clone(&rt->u.dst);
1261
1262         /* Netfilter gets whole the not fragmented skb. */
1263         err = NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL,
1264                       skb->dst->dev, dst_output);
1265         if (err) {
1266                 if (err > 0)
1267                         err = inet->recverr ? net_xmit_errno(err) : 0;
1268                 if (err)
1269                         goto error;
1270         }
1271
1272 out:
1273         inet->cork.flags &= ~IPCORK_OPT;
1274         kfree(inet->cork.opt);
1275         inet->cork.opt = NULL;
1276         if (inet->cork.rt) {
1277                 ip_rt_put(inet->cork.rt);
1278                 inet->cork.rt = NULL;
1279         }
1280         return err;
1281
1282 error:
1283         IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1284         goto out;
1285 }
1286
1287 /*
1288  *      Throw away all pending data on the socket.
1289  */
1290 void ip_flush_pending_frames(struct sock *sk)
1291 {
1292         struct inet_sock *inet = inet_sk(sk);
1293         struct sk_buff *skb;
1294
1295         while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL)
1296                 kfree_skb(skb);
1297
1298         inet->cork.flags &= ~IPCORK_OPT;
1299         kfree(inet->cork.opt);
1300         inet->cork.opt = NULL;
1301         if (inet->cork.rt) {
1302                 ip_rt_put(inet->cork.rt);
1303                 inet->cork.rt = NULL;
1304         }
1305 }
1306
1307
1308 /*
1309  *      Fetch data from kernel space and fill in checksum if needed.
1310  */
1311 static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1312                               int len, int odd, struct sk_buff *skb)
1313 {
1314         unsigned int csum;
1315
1316         csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1317         skb->csum = csum_block_add(skb->csum, csum, odd);
1318         return 0;
1319 }
1320
1321 /*
1322  *      Generic function to send a packet as reply to another packet.
1323  *      Used to send TCP resets so far. ICMP should use this function too.
1324  *
1325  *      Should run single threaded per socket because it uses the sock
1326  *      structure to pass arguments.
1327  *
1328  *      LATER: switch from ip_build_xmit to ip_append_*
1329  */
1330 void ip_send_reply(struct sock *sk, struct sk_buff *skb, struct ip_reply_arg *arg,
1331                    unsigned int len)
1332 {
1333         struct inet_sock *inet = inet_sk(sk);
1334         struct {
1335                 struct ip_options       opt;
1336                 char                    data[40];
1337         } replyopts;
1338         struct ipcm_cookie ipc;
1339         u32 daddr;
1340         struct rtable *rt = (struct rtable*)skb->dst;
1341
1342         if (ip_options_echo(&replyopts.opt, skb))
1343                 return;
1344
1345         daddr = ipc.addr = rt->rt_src;
1346         ipc.opt = NULL;
1347
1348         if (replyopts.opt.optlen) {
1349                 ipc.opt = &replyopts.opt;
1350
1351                 if (ipc.opt->srr)
1352                         daddr = replyopts.opt.faddr;
1353         }
1354
1355         {
1356                 struct flowi fl = { .nl_u = { .ip4_u =
1357                                               { .daddr = daddr,
1358                                                 .saddr = rt->rt_spec_dst,
1359                                                 .tos = RT_TOS(skb->nh.iph->tos) } },
1360                                     /* Not quite clean, but right. */
1361                                     .uli_u = { .ports =
1362                                                { .sport = skb->h.th->dest,
1363                                                  .dport = skb->h.th->source } },
1364                                     .proto = sk->sk_protocol };
1365                 if (ip_route_output_key(&rt, &fl))
1366                         return;
1367         }
1368
1369         /* And let IP do all the hard work.
1370
1371            This chunk is not reenterable, hence spinlock.
1372            Note that it uses the fact, that this function is called
1373            with locally disabled BH and that sk cannot be already spinlocked.
1374          */
1375         bh_lock_sock(sk);
1376         inet->tos = skb->nh.iph->tos;
1377         sk->sk_priority = skb->priority;
1378         sk->sk_protocol = skb->nh.iph->protocol;
1379         ip_append_data(sk, ip_reply_glue_bits, arg->iov->iov_base, len, 0,
1380                        &ipc, rt, MSG_DONTWAIT);
1381         if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
1382                 if (arg->csumoffset >= 0)
1383                         *((u16 *)skb->h.raw + arg->csumoffset) = csum_fold(csum_add(skb->csum, arg->csum));
1384                 skb->ip_summed = CHECKSUM_NONE;
1385                 ip_push_pending_frames(sk);
1386         }
1387
1388         bh_unlock_sock(sk);
1389
1390         ip_rt_put(rt);
1391 }
1392
1393 void __init ip_init(void)
1394 {
1395         ip_rt_init();
1396         inet_initpeers();
1397
1398 #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1399         igmp_mc_proc_init();
1400 #endif
1401 }
1402
1403 EXPORT_SYMBOL(ip_generic_getfrag);
1404 EXPORT_SYMBOL(ip_queue_xmit);
1405 EXPORT_SYMBOL(ip_send_check);