Pileus Git - ~andy/linux/blob - net/ipv4/ip_output.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              The Internet Protocol (IP) output module.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Donald Becker, <becker@super.org>
  11  *              Alan Cox, <Alan.Cox@linux.org>
  12  *              Richard Underwood
  13  *              Stefan Becker, <stefanb@yello.ping.de>
  14  *              Jorge Cwik, <jorge@laser.satlink.net>
  15  *              Arnt Gulbrandsen, <agulbra@nvg.unit.no>
  16  *              Hirokazu Takahashi, <taka@valinux.co.jp>
  17  *
  18  *      See ip_input.c for original log
  19  *
  20  *      Fixes:
  21  *              Alan Cox        :       Missing nonblock feature in ip_build_xmit.
  22  *              Mike Kilburn    :       htons() missing in ip_build_xmit.
  23  *              Bradford Johnson:       Fix faulty handling of some frames when
  24  *                                      no route is found.
  25  *              Alexander Demenshin:    Missing sk/skb free in ip_queue_xmit
  26  *                                      (in case if packet not accepted by
  27  *                                      output firewall rules)
  28  *              Mike McLagan    :       Routing by source
  29  *              Alexey Kuznetsov:       use new route cache
  30  *              Andi Kleen:             Fix broken PMTU recovery and remove
  31  *                                      some redundant tests.
  32  *      Vitaly E. Lavrov        :       Transparent proxy revived after year coma.
  33  *              Andi Kleen      :       Replace ip_reply with ip_send_reply.
  34  *              Andi Kleen      :       Split fast and slow ip_build_xmit path
  35  *                                      for decreased register pressure on x86
  36  *                                      and more readibility.
  37  *              Marc Boucher    :       When call_out_firewall returns FW_QUEUE,
  38  *                                      silently drop skb instead of failing with -EPERM.
  39  *              Detlev Wengorz  :       Copy protocol for fragments.
  40  *              Hirokazu Takahashi:     HW checksumming for outgoing UDP
  41  *                                      datagrams.
  42  *              Hirokazu Takahashi:     sendfile() on UDP works now.
  43  */
  44
  45 #include <asm/uaccess.h>
  46 #include <asm/system.h>
  47 #include <linux/module.h>
  48 #include <linux/types.h>
  49 #include <linux/kernel.h>
  50 #include <linux/mm.h>
  51 #include <linux/string.h>
  52 #include <linux/errno.h>
  53 #include <linux/highmem.h>
  54 #include <linux/slab.h>
  55
  56 #include <linux/socket.h>
  57 #include <linux/sockios.h>
  58 #include <linux/in.h>
  59 #include <linux/inet.h>
  60 #include <linux/netdevice.h>
  61 #include <linux/etherdevice.h>
  62 #include <linux/proc_fs.h>
  63 #include <linux/stat.h>
  64 #include <linux/init.h>
  65
  66 #include <net/snmp.h>
  67 #include <net/ip.h>
  68 #include <net/protocol.h>
  69 #include <net/route.h>
  70 #include <net/xfrm.h>
  71 #include <linux/skbuff.h>
  72 #include <net/sock.h>
  73 #include <net/arp.h>
  74 #include <net/icmp.h>
  75 #include <net/checksum.h>
  76 #include <net/inetpeer.h>
  77 #include <linux/igmp.h>
  78 #include <linux/netfilter_ipv4.h>
  79 #include <linux/netfilter_bridge.h>
  80 #include <linux/mroute.h>
  81 #include <linux/netlink.h>
  82 #include <linux/tcp.h>
  83
  84 int sysctl_ip_default_ttl __read_mostly = IPDEFTTL;
  85 EXPORT_SYMBOL(sysctl_ip_default_ttl);
  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 EXPORT_SYMBOL(ip_send_check);
  94
  95 int __ip_local_out(struct sk_buff *skb)
  96 {
  97         struct iphdr *iph = ip_hdr(skb);
  98
  99         iph->tot_len = htons(skb->len);
 100         ip_send_check(iph);
 101         return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, skb, NULL,
 102                        skb_dst(skb)->dev, dst_output);
 103 }
 104
 105 int ip_local_out(struct sk_buff *skb)
 106 {
 107         int err;
 108
 109         err = __ip_local_out(skb);
 110         if (likely(err == 1))
 111                 err = dst_output(skb);
 112
 113         return err;
 114 }
 115 EXPORT_SYMBOL_GPL(ip_local_out);
 116
 117 /* dev_loopback_xmit for use with netfilter. */
 118 static int ip_dev_loopback_xmit(struct sk_buff *newskb)
 119 {
 120         skb_reset_mac_header(newskb);
 121         __skb_pull(newskb, skb_network_offset(newskb));
 122         newskb->pkt_type = PACKET_LOOPBACK;
 123         newskb->ip_summed = CHECKSUM_UNNECESSARY;
 124         WARN_ON(!skb_dst(newskb));
 125         skb_dst_force(newskb);
 126         netif_rx_ni(newskb);
 127         return 0;
 128 }
 129
 130 static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
 131 {
 132         int ttl = inet->uc_ttl;
 133
 134         if (ttl < 0)
 135                 ttl = ip4_dst_hoplimit(dst);
 136         return ttl;
 137 }
 138
 139 /*
 140  *              Add an ip header to a skbuff and send it out.
 141  *
 142  */
 143 int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk,
 144                           __be32 saddr, __be32 daddr, struct ip_options_rcu *opt)
 145 {
 146         struct inet_sock *inet = inet_sk(sk);
 147         struct rtable *rt = skb_rtable(skb);
 148         struct iphdr *iph;
 149
 150         /* Build the IP header. */
 151         skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
 152         skb_reset_network_header(skb);
 153         iph = ip_hdr(skb);
 154         iph->version  = 4;
 155         iph->ihl      = 5;
 156         iph->tos      = inet->tos;
 157         if (ip_dont_fragment(sk, &rt->dst))
 158                 iph->frag_off = htons(IP_DF);
 159         else
 160                 iph->frag_off = 0;
 161         iph->ttl      = ip_select_ttl(inet, &rt->dst);
 162         iph->daddr    = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
 163         iph->saddr    = saddr;
 164         iph->protocol = sk->sk_protocol;
 165         ip_select_ident(iph, &rt->dst, sk);
 166
 167         if (opt && opt->opt.optlen) {
 168                 iph->ihl += opt->opt.optlen>>2;
 169                 ip_options_build(skb, &opt->opt, daddr, rt, 0);
 170         }
 171
 172         skb->priority = sk->sk_priority;
 173         skb->mark = sk->sk_mark;
 174
 175         /* Send it out. */
 176         return ip_local_out(skb);
 177 }
 178 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
 179
 180 static inline int ip_finish_output2(struct sk_buff *skb)
 181 {
 182         struct dst_entry *dst = skb_dst(skb);
 183         struct rtable *rt = (struct rtable *)dst;
 184         struct net_device *dev = dst->dev;
 185         unsigned int hh_len = LL_RESERVED_SPACE(dev);
 186         struct neighbour *neigh;
 187
 188         if (rt->rt_type == RTN_MULTICAST) {
 189                 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTMCAST, skb->len);
 190         } else if (rt->rt_type == RTN_BROADCAST)
 191                 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTBCAST, skb->len);
 192
 193         /* Be paranoid, rather than too clever. */
 194         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
 195                 struct sk_buff *skb2;
 196
 197                 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
 198                 if (skb2 == NULL) {
 199                         kfree_skb(skb);
 200                         return -ENOMEM;
 201                 }
 202                 if (skb->sk)
 203                         skb_set_owner_w(skb2, skb->sk);
 204                 kfree_skb(skb);
 205                 skb = skb2;
 206         }
 207
 208         rcu_read_lock();
 209         neigh = dst_get_neighbour_noref(dst);
 210         if (neigh) {
 211                 int res = neigh_output(neigh, skb);
 212
 213                 rcu_read_unlock();
 214                 return res;
 215         }
 216         rcu_read_unlock();
 217
 218         if (net_ratelimit())
 219                 printk(KERN_DEBUG "ip_finish_output2: No header cache and no neighbour!\n");
 220         kfree_skb(skb);
 221         return -EINVAL;
 222 }
 223
 224 static inline int ip_skb_dst_mtu(struct sk_buff *skb)
 225 {
 226         struct inet_sock *inet = skb->sk ? inet_sk(skb->sk) : NULL;
 227
 228         return (inet && inet->pmtudisc == IP_PMTUDISC_PROBE) ?
 229                skb_dst(skb)->dev->mtu : dst_mtu(skb_dst(skb));
 230 }
 231
 232 static int ip_finish_output(struct sk_buff *skb)
 233 {
 234 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
 235         /* Policy lookup after SNAT yielded a new policy */
 236         if (skb_dst(skb)->xfrm != NULL) {
 237                 IPCB(skb)->flags |= IPSKB_REROUTED;
 238                 return dst_output(skb);
 239         }
 240 #endif
 241         if (skb->len > ip_skb_dst_mtu(skb) && !skb_is_gso(skb))
 242                 return ip_fragment(skb, ip_finish_output2);
 243         else
 244                 return ip_finish_output2(skb);
 245 }
 246
 247 int ip_mc_output(struct sk_buff *skb)
 248 {
 249         struct sock *sk = skb->sk;
 250         struct rtable *rt = skb_rtable(skb);
 251         struct net_device *dev = rt->dst.dev;
 252
 253         /*
 254          *      If the indicated interface is up and running, send the packet.
 255          */
 256         IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
 257
 258         skb->dev = dev;
 259         skb->protocol = htons(ETH_P_IP);
 260
 261         /*
 262          *      Multicasts are looped back for other local users
 263          */
 264
 265         if (rt->rt_flags&RTCF_MULTICAST) {
 266                 if (sk_mc_loop(sk)
 267 #ifdef CONFIG_IP_MROUTE
 268                 /* Small optimization: do not loopback not local frames,
 269                    which returned after forwarding; they will be  dropped
 270                    by ip_mr_input in any case.
 271                    Note, that local frames are looped back to be delivered
 272                    to local recipients.
 273
 274                    This check is duplicated in ip_mr_input at the moment.
 275                  */
 276                     &&
 277                     ((rt->rt_flags & RTCF_LOCAL) ||
 278                      !(IPCB(skb)->flags & IPSKB_FORWARDED))
 279 #endif
 280                    ) {
 281                         struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
 282                         if (newskb)
 283                                 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
 284                                         newskb, NULL, newskb->dev,
 285                                         ip_dev_loopback_xmit);
 286                 }
 287
 288                 /* Multicasts with ttl 0 must not go beyond the host */
 289
 290                 if (ip_hdr(skb)->ttl == 0) {
 291                         kfree_skb(skb);
 292                         return 0;
 293                 }
 294         }
 295
 296         if (rt->rt_flags&RTCF_BROADCAST) {
 297                 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
 298                 if (newskb)
 299                         NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, newskb,
 300                                 NULL, newskb->dev, ip_dev_loopback_xmit);
 301         }
 302
 303         return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL,
 304                             skb->dev, ip_finish_output,
 305                             !(IPCB(skb)->flags & IPSKB_REROUTED));
 306 }
 307
 308 int ip_output(struct sk_buff *skb)
 309 {
 310         struct net_device *dev = skb_dst(skb)->dev;
 311
 312         IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
 313
 314         skb->dev = dev;
 315         skb->protocol = htons(ETH_P_IP);
 316
 317         return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL, dev,
 318                             ip_finish_output,
 319                             !(IPCB(skb)->flags & IPSKB_REROUTED));
 320 }
 321
 322 /*
 323  * copy saddr and daddr, possibly using 64bit load/stores
 324  * Equivalent to :
 325  *   iph->saddr = fl4->saddr;
 326  *   iph->daddr = fl4->daddr;
 327  */
 328 static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
 329 {
 330         BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
 331                      offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
 332         memcpy(&iph->saddr, &fl4->saddr,
 333                sizeof(fl4->saddr) + sizeof(fl4->daddr));
 334 }
 335
 336 int ip_queue_xmit(struct sk_buff *skb, struct flowi *fl)
 337 {
 338         struct sock *sk = skb->sk;
 339         struct inet_sock *inet = inet_sk(sk);
 340         struct ip_options_rcu *inet_opt;
 341         struct flowi4 *fl4;
 342         struct rtable *rt;
 343         struct iphdr *iph;
 344         int res;
 345
 346         /* Skip all of this if the packet is already routed,
 347          * f.e. by something like SCTP.
 348          */
 349         rcu_read_lock();
 350         inet_opt = rcu_dereference(inet->inet_opt);
 351         fl4 = &fl->u.ip4;
 352         rt = skb_rtable(skb);
 353         if (rt != NULL)
 354                 goto packet_routed;
 355
 356         /* Make sure we can route this packet. */
 357         rt = (struct rtable *)__sk_dst_check(sk, 0);
 358         if (rt == NULL) {
 359                 __be32 daddr;
 360
 361                 /* Use correct destination address if we have options. */
 362                 daddr = inet->inet_daddr;
 363                 if (inet_opt && inet_opt->opt.srr)
 364                         daddr = inet_opt->opt.faddr;
 365
 366                 /* If this fails, retransmit mechanism of transport layer will
 367                  * keep trying until route appears or the connection times
 368                  * itself out.
 369                  */
 370                 rt = ip_route_output_ports(sock_net(sk), fl4, sk,
 371                                            daddr, inet->inet_saddr,
 372                                            inet->inet_dport,
 373                                            inet->inet_sport,
 374                                            sk->sk_protocol,
 375                                            RT_CONN_FLAGS(sk),
 376                                            sk->sk_bound_dev_if);
 377                 if (IS_ERR(rt))
 378                         goto no_route;
 379                 sk_setup_caps(sk, &rt->dst);
 380         }
 381         skb_dst_set_noref(skb, &rt->dst);
 382
 383 packet_routed:
 384         if (inet_opt && inet_opt->opt.is_strictroute && fl4->daddr != rt->rt_gateway)
 385                 goto no_route;
 386
 387         /* OK, we know where to send it, allocate and build IP header. */
 388         skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
 389         skb_reset_network_header(skb);
 390         iph = ip_hdr(skb);
 391         *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
 392         if (ip_dont_fragment(sk, &rt->dst) && !skb->local_df)
 393                 iph->frag_off = htons(IP_DF);
 394         else
 395                 iph->frag_off = 0;
 396         iph->ttl      = ip_select_ttl(inet, &rt->dst);
 397         iph->protocol = sk->sk_protocol;
 398         ip_copy_addrs(iph, fl4);
 399
 400         /* Transport layer set skb->h.foo itself. */
 401
 402         if (inet_opt && inet_opt->opt.optlen) {
 403                 iph->ihl += inet_opt->opt.optlen >> 2;
 404                 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0);
 405         }
 406
 407         ip_select_ident_more(iph, &rt->dst, sk,
 408                              (skb_shinfo(skb)->gso_segs ?: 1) - 1);
 409
 410         skb->priority = sk->sk_priority;
 411         skb->mark = sk->sk_mark;
 412
 413         res = ip_local_out(skb);
 414         rcu_read_unlock();
 415         return res;
 416
 417 no_route:
 418         rcu_read_unlock();
 419         IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
 420         kfree_skb(skb);
 421         return -EHOSTUNREACH;
 422 }
 423 EXPORT_SYMBOL(ip_queue_xmit);
 424
 425
 426 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
 427 {
 428         to->pkt_type = from->pkt_type;
 429         to->priority = from->priority;
 430         to->protocol = from->protocol;
 431         skb_dst_drop(to);
 432         skb_dst_copy(to, from);
 433         to->dev = from->dev;
 434         to->mark = from->mark;
 435
 436         /* Copy the flags to each fragment. */
 437         IPCB(to)->flags = IPCB(from)->flags;
 438
 439 #ifdef CONFIG_NET_SCHED
 440         to->tc_index = from->tc_index;
 441 #endif
 442         nf_copy(to, from);
 443 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
 444     defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
 445         to->nf_trace = from->nf_trace;
 446 #endif
 447 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
 448         to->ipvs_property = from->ipvs_property;
 449 #endif
 450         skb_copy_secmark(to, from);
 451 }
 452
 453 /*
 454  *      This IP datagram is too large to be sent in one piece.  Break it up into
 455  *      smaller pieces (each of size equal to IP header plus
 456  *      a block of the data of the original IP data part) that will yet fit in a
 457  *      single device frame, and queue such a frame for sending.
 458  */
 459
 460 int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
 461 {
 462         struct iphdr *iph;
 463         int ptr;
 464         struct net_device *dev;
 465         struct sk_buff *skb2;
 466         unsigned int mtu, hlen, left, len, ll_rs;
 467         int offset;
 468         __be16 not_last_frag;
 469         struct rtable *rt = skb_rtable(skb);
 470         int err = 0;
 471
 472         dev = rt->dst.dev;
 473
 474         /*
 475          *      Point into the IP datagram header.
 476          */
 477
 478         iph = ip_hdr(skb);
 479
 480         if (unlikely((iph->frag_off & htons(IP_DF)) && !skb->local_df)) {
 481                 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
 482                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
 483                           htonl(ip_skb_dst_mtu(skb)));
 484                 kfree_skb(skb);
 485                 return -EMSGSIZE;
 486         }
 487
 488         /*
 489          *      Setup starting values.
 490          */
 491
 492         hlen = iph->ihl * 4;
 493         mtu = dst_mtu(&rt->dst) - hlen; /* Size of data space */
 494 #ifdef CONFIG_BRIDGE_NETFILTER
 495         if (skb->nf_bridge)
 496                 mtu -= nf_bridge_mtu_reduction(skb);
 497 #endif
 498         IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
 499
 500         /* When frag_list is given, use it. First, check its validity:
 501          * some transformers could create wrong frag_list or break existing
 502          * one, it is not prohibited. In this case fall back to copying.
 503          *
 504          * LATER: this step can be merged to real generation of fragments,
 505          * we can switch to copy when see the first bad fragment.
 506          */
 507         if (skb_has_frag_list(skb)) {
 508                 struct sk_buff *frag, *frag2;
 509                 int first_len = skb_pagelen(skb);
 510
 511                 if (first_len - hlen > mtu ||
 512                     ((first_len - hlen) & 7) ||
 513                     ip_is_fragment(iph) ||
 514                     skb_cloned(skb))
 515                         goto slow_path;
 516
 517                 skb_walk_frags(skb, frag) {
 518                         /* Correct geometry. */
 519                         if (frag->len > mtu ||
 520                             ((frag->len & 7) && frag->next) ||
 521                             skb_headroom(frag) < hlen)
 522                                 goto slow_path_clean;
 523
 524                         /* Partially cloned skb? */
 525                         if (skb_shared(frag))
 526                                 goto slow_path_clean;
 527
 528                         BUG_ON(frag->sk);
 529                         if (skb->sk) {
 530                                 frag->sk = skb->sk;
 531                                 frag->destructor = sock_wfree;
 532                         }
 533                         skb->truesize -= frag->truesize;
 534                 }
 535
 536                 /* Everything is OK. Generate! */
 537
 538                 err = 0;
 539                 offset = 0;
 540                 frag = skb_shinfo(skb)->frag_list;
 541                 skb_frag_list_init(skb);
 542                 skb->data_len = first_len - skb_headlen(skb);
 543                 skb->len = first_len;
 544                 iph->tot_len = htons(first_len);
 545                 iph->frag_off = htons(IP_MF);
 546                 ip_send_check(iph);
 547
 548                 for (;;) {
 549                         /* Prepare header of the next frame,
 550                          * before previous one went down. */
 551                         if (frag) {
 552                                 frag->ip_summed = CHECKSUM_NONE;
 553                                 skb_reset_transport_header(frag);
 554                                 __skb_push(frag, hlen);
 555                                 skb_reset_network_header(frag);
 556                                 memcpy(skb_network_header(frag), iph, hlen);
 557                                 iph = ip_hdr(frag);
 558                                 iph->tot_len = htons(frag->len);
 559                                 ip_copy_metadata(frag, skb);
 560                                 if (offset == 0)
 561                                         ip_options_fragment(frag);
 562                                 offset += skb->len - hlen;
 563                                 iph->frag_off = htons(offset>>3);
 564                                 if (frag->next != NULL)
 565                                         iph->frag_off |= htons(IP_MF);
 566                                 /* Ready, complete checksum */
 567                                 ip_send_check(iph);
 568                         }
 569
 570                         err = output(skb);
 571
 572                         if (!err)
 573                                 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
 574                         if (err || !frag)
 575                                 break;
 576
 577                         skb = frag;
 578                         frag = skb->next;
 579                         skb->next = NULL;
 580                 }
 581
 582                 if (err == 0) {
 583                         IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
 584                         return 0;
 585                 }
 586
 587                 while (frag) {
 588                         skb = frag->next;
 589                         kfree_skb(frag);
 590                         frag = skb;
 591                 }
 592                 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
 593                 return err;
 594
 595 slow_path_clean:
 596                 skb_walk_frags(skb, frag2) {
 597                         if (frag2 == frag)
 598                                 break;
 599                         frag2->sk = NULL;
 600                         frag2->destructor = NULL;
 601                         skb->truesize += frag2->truesize;
 602                 }
 603         }
 604
 605 slow_path:
 606         left = skb->len - hlen;         /* Space per frame */
 607         ptr = hlen;             /* Where to start from */
 608
 609         /* for bridged IP traffic encapsulated inside f.e. a vlan header,
 610          * we need to make room for the encapsulating header
 611          */
 612         ll_rs = LL_RESERVED_SPACE_EXTRA(rt->dst.dev, nf_bridge_pad(skb));
 613
 614         /*
 615          *      Fragment the datagram.
 616          */
 617
 618         offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
 619         not_last_frag = iph->frag_off & htons(IP_MF);
 620
 621         /*
 622          *      Keep copying data until we run out.
 623          */
 624
 625         while (left > 0) {
 626                 len = left;
 627                 /* IF: it doesn't fit, use 'mtu' - the data space left */
 628                 if (len > mtu)
 629                         len = mtu;
 630                 /* IF: we are not sending up to and including the packet end
 631                    then align the next start on an eight byte boundary */
 632                 if (len < left) {
 633                         len &= ~7;
 634                 }
 635                 /*
 636                  *      Allocate buffer.
 637                  */
 638
 639                 if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) {
 640                         NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n");
 641                         err = -ENOMEM;
 642                         goto fail;
 643                 }
 644
 645                 /*
 646                  *      Set up data on packet
 647                  */
 648
 649                 ip_copy_metadata(skb2, skb);
 650                 skb_reserve(skb2, ll_rs);
 651                 skb_put(skb2, len + hlen);
 652                 skb_reset_network_header(skb2);
 653                 skb2->transport_header = skb2->network_header + hlen;
 654
 655                 /*
 656                  *      Charge the memory for the fragment to any owner
 657                  *      it might possess
 658                  */
 659
 660                 if (skb->sk)
 661                         skb_set_owner_w(skb2, skb->sk);
 662
 663                 /*
 664                  *      Copy the packet header into the new buffer.
 665                  */
 666
 667                 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
 668
 669                 /*
 670                  *      Copy a block of the IP datagram.
 671                  */
 672                 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
 673                         BUG();
 674                 left -= len;
 675
 676                 /*
 677                  *      Fill in the new header fields.
 678                  */
 679                 iph = ip_hdr(skb2);
 680                 iph->frag_off = htons((offset >> 3));
 681
 682                 /* ANK: dirty, but effective trick. Upgrade options only if
 683                  * the segment to be fragmented was THE FIRST (otherwise,
 684                  * options are already fixed) and make it ONCE
 685                  * on the initial skb, so that all the following fragments
 686                  * will inherit fixed options.
 687                  */
 688                 if (offset == 0)
 689                         ip_options_fragment(skb);
 690
 691                 /*
 692                  *      Added AC : If we are fragmenting a fragment that's not the
 693                  *                 last fragment then keep MF on each bit
 694                  */
 695                 if (left > 0 || not_last_frag)
 696                         iph->frag_off |= htons(IP_MF);
 697                 ptr += len;
 698                 offset += len;
 699
 700                 /*
 701                  *      Put this fragment into the sending queue.
 702                  */
 703                 iph->tot_len = htons(len + hlen);
 704
 705                 ip_send_check(iph);
 706
 707                 err = output(skb2);
 708                 if (err)
 709                         goto fail;
 710
 711                 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
 712         }
 713         kfree_skb(skb);
 714         IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
 715         return err;
 716
 717 fail:
 718         kfree_skb(skb);
 719         IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
 720         return err;
 721 }
 722 EXPORT_SYMBOL(ip_fragment);
 723
 724 int
 725 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
 726 {
 727         struct iovec *iov = from;
 728
 729         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 730                 if (memcpy_fromiovecend(to, iov, offset, len) < 0)
 731                         return -EFAULT;
 732         } else {
 733                 __wsum csum = 0;
 734                 if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0)
 735                         return -EFAULT;
 736                 skb->csum = csum_block_add(skb->csum, csum, odd);
 737         }
 738         return 0;
 739 }
 740 EXPORT_SYMBOL(ip_generic_getfrag);
 741
 742 static inline __wsum
 743 csum_page(struct page *page, int offset, int copy)
 744 {
 745         char *kaddr;
 746         __wsum csum;
 747         kaddr = kmap(page);
 748         csum = csum_partial(kaddr + offset, copy, 0);
 749         kunmap(page);
 750         return csum;
 751 }
 752
 753 static inline int ip_ufo_append_data(struct sock *sk,
 754                         struct sk_buff_head *queue,
 755                         int getfrag(void *from, char *to, int offset, int len,
 756                                int odd, struct sk_buff *skb),
 757                         void *from, int length, int hh_len, int fragheaderlen,
 758                         int transhdrlen, int maxfraglen, unsigned int flags)
 759 {
 760         struct sk_buff *skb;
 761         int err;
 762
 763         /* There is support for UDP fragmentation offload by network
 764          * device, so create one single skb packet containing complete
 765          * udp datagram
 766          */
 767         if ((skb = skb_peek_tail(queue)) == NULL) {
 768                 skb = sock_alloc_send_skb(sk,
 769                         hh_len + fragheaderlen + transhdrlen + 20,
 770                         (flags & MSG_DONTWAIT), &err);
 771
 772                 if (skb == NULL)
 773                         return err;
 774
 775                 /* reserve space for Hardware header */
 776                 skb_reserve(skb, hh_len);
 777
 778                 /* create space for UDP/IP header */
 779                 skb_put(skb, fragheaderlen + transhdrlen);
 780
 781                 /* initialize network header pointer */
 782                 skb_reset_network_header(skb);
 783
 784                 /* initialize protocol header pointer */
 785                 skb->transport_header = skb->network_header + fragheaderlen;
 786
 787                 skb->ip_summed = CHECKSUM_PARTIAL;
 788                 skb->csum = 0;
 789
 790                 /* specify the length of each IP datagram fragment */
 791                 skb_shinfo(skb)->gso_size = maxfraglen - fragheaderlen;
 792                 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
 793                 __skb_queue_tail(queue, skb);
 794         }
 795
 796         return skb_append_datato_frags(sk, skb, getfrag, from,
 797                                        (length - transhdrlen));
 798 }
 799
 800 static int __ip_append_data(struct sock *sk,
 801                             struct flowi4 *fl4,
 802                             struct sk_buff_head *queue,
 803                             struct inet_cork *cork,
 804                             int getfrag(void *from, char *to, int offset,
 805                                         int len, int odd, struct sk_buff *skb),
 806                             void *from, int length, int transhdrlen,
 807                             unsigned int flags)
 808 {
 809         struct inet_sock *inet = inet_sk(sk);
 810         struct sk_buff *skb;
 811
 812         struct ip_options *opt = cork->opt;
 813         int hh_len;
 814         int exthdrlen;
 815         int mtu;
 816         int copy;
 817         int err;
 818         int offset = 0;
 819         unsigned int maxfraglen, fragheaderlen;
 820         int csummode = CHECKSUM_NONE;
 821         struct rtable *rt = (struct rtable *)cork->dst;
 822
 823         skb = skb_peek_tail(queue);
 824
 825         exthdrlen = !skb ? rt->dst.header_len : 0;
 826         mtu = cork->fragsize;
 827
 828         hh_len = LL_RESERVED_SPACE(rt->dst.dev);
 829
 830         fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
 831         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
 832
 833         if (cork->length + length > 0xFFFF - fragheaderlen) {
 834                 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
 835                                mtu-exthdrlen);
 836                 return -EMSGSIZE;
 837         }
 838
 839         /*
 840          * transhdrlen > 0 means that this is the first fragment and we wish
 841          * it won't be fragmented in the future.
 842          */
 843         if (transhdrlen &&
 844             length + fragheaderlen <= mtu &&
 845             rt->dst.dev->features & NETIF_F_V4_CSUM &&
 846             !exthdrlen)
 847                 csummode = CHECKSUM_PARTIAL;
 848
 849         cork->length += length;
 850         if (((length > mtu) || (skb && skb_is_gso(skb))) &&
 851             (sk->sk_protocol == IPPROTO_UDP) &&
 852             (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len) {
 853                 err = ip_ufo_append_data(sk, queue, getfrag, from, length,
 854                                          hh_len, fragheaderlen, transhdrlen,
 855                                          maxfraglen, flags);
 856                 if (err)
 857                         goto error;
 858                 return 0;
 859         }
 860
 861         /* So, what's going on in the loop below?
 862          *
 863          * We use calculated fragment length to generate chained skb,
 864          * each of segments is IP fragment ready for sending to network after
 865          * adding appropriate IP header.
 866          */
 867
 868         if (!skb)
 869                 goto alloc_new_skb;
 870
 871         while (length > 0) {
 872                 /* Check if the remaining data fits into current packet. */
 873                 copy = mtu - skb->len;
 874                 if (copy < length)
 875                         copy = maxfraglen - skb->len;
 876                 if (copy <= 0) {
 877                         char *data;
 878                         unsigned int datalen;
 879                         unsigned int fraglen;
 880                         unsigned int fraggap;
 881                         unsigned int alloclen;
 882                         struct sk_buff *skb_prev;
 883 alloc_new_skb:
 884                         skb_prev = skb;
 885                         if (skb_prev)
 886                                 fraggap = skb_prev->len - maxfraglen;
 887                         else
 888                                 fraggap = 0;
 889
 890                         /*
 891                          * If remaining data exceeds the mtu,
 892                          * we know we need more fragment(s).
 893                          */
 894                         datalen = length + fraggap;
 895                         if (datalen > mtu - fragheaderlen)
 896                                 datalen = maxfraglen - fragheaderlen;
 897                         fraglen = datalen + fragheaderlen;
 898
 899                         if ((flags & MSG_MORE) &&
 900                             !(rt->dst.dev->features&NETIF_F_SG))
 901                                 alloclen = mtu;
 902                         else
 903                                 alloclen = fraglen;
 904
 905                         alloclen += exthdrlen;
 906
 907                         /* The last fragment gets additional space at tail.
 908                          * Note, with MSG_MORE we overallocate on fragments,
 909                          * because we have no idea what fragment will be
 910                          * the last.
 911                          */
 912                         if (datalen == length + fraggap)
 913                                 alloclen += rt->dst.trailer_len;
 914
 915                         if (transhdrlen) {
 916                                 skb = sock_alloc_send_skb(sk,
 917                                                 alloclen + hh_len + 15,
 918                                                 (flags & MSG_DONTWAIT), &err);
 919                         } else {
 920                                 skb = NULL;
 921                                 if (atomic_read(&sk->sk_wmem_alloc) <=
 922                                     2 * sk->sk_sndbuf)
 923                                         skb = sock_wmalloc(sk,
 924                                                            alloclen + hh_len + 15, 1,
 925                                                            sk->sk_allocation);
 926                                 if (unlikely(skb == NULL))
 927                                         err = -ENOBUFS;
 928                                 else
 929                                         /* only the initial fragment is
 930                                            time stamped */
 931                                         cork->tx_flags = 0;
 932                         }
 933                         if (skb == NULL)
 934                                 goto error;
 935
 936                         /*
 937                          *      Fill in the control structures
 938                          */
 939                         skb->ip_summed = csummode;
 940                         skb->csum = 0;
 941                         skb_reserve(skb, hh_len);
 942                         skb_shinfo(skb)->tx_flags = cork->tx_flags;
 943
 944                         /*
 945                          *      Find where to start putting bytes.
 946                          */
 947                         data = skb_put(skb, fraglen + exthdrlen);
 948                         skb_set_network_header(skb, exthdrlen);
 949                         skb->transport_header = (skb->network_header +
 950                                                  fragheaderlen);
 951                         data += fragheaderlen + exthdrlen;
 952
 953                         if (fraggap) {
 954                                 skb->csum = skb_copy_and_csum_bits(
 955                                         skb_prev, maxfraglen,
 956                                         data + transhdrlen, fraggap, 0);
 957                                 skb_prev->csum = csum_sub(skb_prev->csum,
 958                                                           skb->csum);
 959                                 data += fraggap;
 960                                 pskb_trim_unique(skb_prev, maxfraglen);
 961                         }
 962
 963                         copy = datalen - transhdrlen - fraggap;
 964                         if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
 965                                 err = -EFAULT;
 966                                 kfree_skb(skb);
 967                                 goto error;
 968                         }
 969
 970                         offset += copy;
 971                         length -= datalen - fraggap;
 972                         transhdrlen = 0;
 973                         exthdrlen = 0;
 974                         csummode = CHECKSUM_NONE;
 975
 976                         /*
 977                          * Put the packet on the pending queue.
 978                          */
 979                         __skb_queue_tail(queue, skb);
 980                         continue;
 981                 }
 982
 983                 if (copy > length)
 984                         copy = length;
 985
 986                 if (!(rt->dst.dev->features&NETIF_F_SG)) {
 987                         unsigned int off;
 988
 989                         off = skb->len;
 990                         if (getfrag(from, skb_put(skb, copy),
 991                                         offset, copy, off, skb) < 0) {
 992                                 __skb_trim(skb, off);
 993                                 err = -EFAULT;
 994                                 goto error;
 995                         }
 996                 } else {
 997                         int i = skb_shinfo(skb)->nr_frags;
 998                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
 999                         struct page *page = cork->page;
1000                         int off = cork->off;
1001                         unsigned int left;
1002
1003                         if (page && (left = PAGE_SIZE - off) > 0) {
1004                                 if (copy >= left)
1005                                         copy = left;
1006                                 if (page != skb_frag_page(frag)) {
1007                                         if (i == MAX_SKB_FRAGS) {
1008                                                 err = -EMSGSIZE;
1009                                                 goto error;
1010                                         }
1011                                         skb_fill_page_desc(skb, i, page, off, 0);
1012                                         skb_frag_ref(skb, i);
1013                                         frag = &skb_shinfo(skb)->frags[i];
1014                                 }
1015                         } else if (i < MAX_SKB_FRAGS) {
1016                                 if (copy > PAGE_SIZE)
1017                                         copy = PAGE_SIZE;
1018                                 page = alloc_pages(sk->sk_allocation, 0);
1019                                 if (page == NULL)  {
1020                                         err = -ENOMEM;
1021                                         goto error;
1022                                 }
1023                                 cork->page = page;
1024                                 cork->off = 0;
1025
1026                                 skb_fill_page_desc(skb, i, page, 0, 0);
1027                                 frag = &skb_shinfo(skb)->frags[i];
1028                         } else {
1029                                 err = -EMSGSIZE;
1030                                 goto error;
1031                         }
1032                         if (getfrag(from, skb_frag_address(frag)+skb_frag_size(frag),
1033                                     offset, copy, skb->len, skb) < 0) {
1034                                 err = -EFAULT;
1035                                 goto error;
1036                         }
1037                         cork->off += copy;
1038                         skb_frag_size_add(frag, copy);
1039                         skb->len += copy;
1040                         skb->data_len += copy;
1041                         skb->truesize += copy;
1042                         atomic_add(copy, &sk->sk_wmem_alloc);
1043                 }
1044                 offset += copy;
1045                 length -= copy;
1046         }
1047
1048         return 0;
1049
1050 error:
1051         cork->length -= length;
1052         IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1053         return err;
1054 }
1055
1056 static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
1057                          struct ipcm_cookie *ipc, struct rtable **rtp)
1058 {
1059         struct inet_sock *inet = inet_sk(sk);
1060         struct ip_options_rcu *opt;
1061         struct rtable *rt;
1062
1063         /*
1064          * setup for corking.
1065          */
1066         opt = ipc->opt;
1067         if (opt) {
1068                 if (cork->opt == NULL) {
1069                         cork->opt = kmalloc(sizeof(struct ip_options) + 40,
1070                                             sk->sk_allocation);
1071                         if (unlikely(cork->opt == NULL))
1072                                 return -ENOBUFS;
1073                 }
1074                 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
1075                 cork->flags |= IPCORK_OPT;
1076                 cork->addr = ipc->addr;
1077         }
1078         rt = *rtp;
1079         if (unlikely(!rt))
1080                 return -EFAULT;
1081         /*
1082          * We steal reference to this route, caller should not release it
1083          */
1084         *rtp = NULL;
1085         cork->fragsize = inet->pmtudisc == IP_PMTUDISC_PROBE ?
1086                          rt->dst.dev->mtu : dst_mtu(&rt->dst);
1087         cork->dst = &rt->dst;
1088         cork->length = 0;
1089         cork->tx_flags = ipc->tx_flags;
1090         cork->page = NULL;
1091         cork->off = 0;
1092
1093         return 0;
1094 }
1095
1096 /*
1097  *      ip_append_data() and ip_append_page() can make one large IP datagram
1098  *      from many pieces of data. Each pieces will be holded on the socket
1099  *      until ip_push_pending_frames() is called. Each piece can be a page
1100  *      or non-page data.
1101  *
1102  *      Not only UDP, other transport protocols - e.g. raw sockets - can use
1103  *      this interface potentially.
1104  *
1105  *      LATER: length must be adjusted by pad at tail, when it is required.
1106  */
1107 int ip_append_data(struct sock *sk, struct flowi4 *fl4,
1108                    int getfrag(void *from, char *to, int offset, int len,
1109                                int odd, struct sk_buff *skb),
1110                    void *from, int length, int transhdrlen,
1111                    struct ipcm_cookie *ipc, struct rtable **rtp,
1112                    unsigned int flags)
1113 {
1114         struct inet_sock *inet = inet_sk(sk);
1115         int err;
1116
1117         if (flags&MSG_PROBE)
1118                 return 0;
1119
1120         if (skb_queue_empty(&sk->sk_write_queue)) {
1121                 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
1122                 if (err)
1123                         return err;
1124         } else {
1125                 transhdrlen = 0;
1126         }
1127
1128         return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base, getfrag,
1129                                 from, length, transhdrlen, flags);
1130 }
1131
1132 ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
1133                        int offset, size_t size, int flags)
1134 {
1135         struct inet_sock *inet = inet_sk(sk);
1136         struct sk_buff *skb;
1137         struct rtable *rt;
1138         struct ip_options *opt = NULL;
1139         struct inet_cork *cork;
1140         int hh_len;
1141         int mtu;
1142         int len;
1143         int err;
1144         unsigned int maxfraglen, fragheaderlen, fraggap;
1145
1146         if (inet->hdrincl)
1147                 return -EPERM;
1148
1149         if (flags&MSG_PROBE)
1150                 return 0;
1151
1152         if (skb_queue_empty(&sk->sk_write_queue))
1153                 return -EINVAL;
1154
1155         cork = &inet->cork.base;
1156         rt = (struct rtable *)cork->dst;
1157         if (cork->flags & IPCORK_OPT)
1158                 opt = cork->opt;
1159
1160         if (!(rt->dst.dev->features&NETIF_F_SG))
1161                 return -EOPNOTSUPP;
1162
1163         hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1164         mtu = cork->fragsize;
1165
1166         fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1167         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1168
1169         if (cork->length + size > 0xFFFF - fragheaderlen) {
1170                 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, mtu);
1171                 return -EMSGSIZE;
1172         }
1173
1174         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1175                 return -EINVAL;
1176
1177         cork->length += size;
1178         if ((size + skb->len > mtu) &&
1179             (sk->sk_protocol == IPPROTO_UDP) &&
1180             (rt->dst.dev->features & NETIF_F_UFO)) {
1181                 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
1182                 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1183         }
1184
1185
1186         while (size > 0) {
1187                 int i;
1188
1189                 if (skb_is_gso(skb))
1190                         len = size;
1191                 else {
1192
1193                         /* Check if the remaining data fits into current packet. */
1194                         len = mtu - skb->len;
1195                         if (len < size)
1196                                 len = maxfraglen - skb->len;
1197                 }
1198                 if (len <= 0) {
1199                         struct sk_buff *skb_prev;
1200                         int alloclen;
1201
1202                         skb_prev = skb;
1203                         fraggap = skb_prev->len - maxfraglen;
1204
1205                         alloclen = fragheaderlen + hh_len + fraggap + 15;
1206                         skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1207                         if (unlikely(!skb)) {
1208                                 err = -ENOBUFS;
1209                                 goto error;
1210                         }
1211
1212                         /*
1213                          *      Fill in the control structures
1214                          */
1215                         skb->ip_summed = CHECKSUM_NONE;
1216                         skb->csum = 0;
1217                         skb_reserve(skb, hh_len);
1218
1219                         /*
1220                          *      Find where to start putting bytes.
1221                          */
1222                         skb_put(skb, fragheaderlen + fraggap);
1223                         skb_reset_network_header(skb);
1224                         skb->transport_header = (skb->network_header +
1225                                                  fragheaderlen);
1226                         if (fraggap) {
1227                                 skb->csum = skb_copy_and_csum_bits(skb_prev,
1228                                                                    maxfraglen,
1229                                                     skb_transport_header(skb),
1230                                                                    fraggap, 0);
1231                                 skb_prev->csum = csum_sub(skb_prev->csum,
1232                                                           skb->csum);
1233                                 pskb_trim_unique(skb_prev, maxfraglen);
1234                         }
1235
1236                         /*
1237                          * Put the packet on the pending queue.
1238                          */
1239                         __skb_queue_tail(&sk->sk_write_queue, skb);
1240                         continue;
1241                 }
1242
1243                 i = skb_shinfo(skb)->nr_frags;
1244                 if (len > size)
1245                         len = size;
1246                 if (skb_can_coalesce(skb, i, page, offset)) {
1247                         skb_frag_size_add(&skb_shinfo(skb)->frags[i-1], len);
1248                 } else if (i < MAX_SKB_FRAGS) {
1249                         get_page(page);
1250                         skb_fill_page_desc(skb, i, page, offset, len);
1251                 } else {
1252                         err = -EMSGSIZE;
1253                         goto error;
1254                 }
1255
1256                 if (skb->ip_summed == CHECKSUM_NONE) {
1257                         __wsum csum;
1258                         csum = csum_page(page, offset, len);
1259                         skb->csum = csum_block_add(skb->csum, csum, skb->len);
1260                 }
1261
1262                 skb->len += len;
1263                 skb->data_len += len;
1264                 skb->truesize += len;
1265                 atomic_add(len, &sk->sk_wmem_alloc);
1266                 offset += len;
1267                 size -= len;
1268         }
1269         return 0;
1270
1271 error:
1272         cork->length -= size;
1273         IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1274         return err;
1275 }
1276
1277 static void ip_cork_release(struct inet_cork *cork)
1278 {
1279         cork->flags &= ~IPCORK_OPT;
1280         kfree(cork->opt);
1281         cork->opt = NULL;
1282         dst_release(cork->dst);
1283         cork->dst = NULL;
1284 }
1285
1286 /*
1287  *      Combined all pending IP fragments on the socket as one IP datagram
1288  *      and push them out.
1289  */
1290 struct sk_buff *__ip_make_skb(struct sock *sk,
1291                               struct flowi4 *fl4,
1292                               struct sk_buff_head *queue,
1293                               struct inet_cork *cork)
1294 {
1295         struct sk_buff *skb, *tmp_skb;
1296         struct sk_buff **tail_skb;
1297         struct inet_sock *inet = inet_sk(sk);
1298         struct net *net = sock_net(sk);
1299         struct ip_options *opt = NULL;
1300         struct rtable *rt = (struct rtable *)cork->dst;
1301         struct iphdr *iph;
1302         __be16 df = 0;
1303         __u8 ttl;
1304
1305         if ((skb = __skb_dequeue(queue)) == NULL)
1306                 goto out;
1307         tail_skb = &(skb_shinfo(skb)->frag_list);
1308
1309         /* move skb->data to ip header from ext header */
1310         if (skb->data < skb_network_header(skb))
1311                 __skb_pull(skb, skb_network_offset(skb));
1312         while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1313                 __skb_pull(tmp_skb, skb_network_header_len(skb));
1314                 *tail_skb = tmp_skb;
1315                 tail_skb = &(tmp_skb->next);
1316                 skb->len += tmp_skb->len;
1317                 skb->data_len += tmp_skb->len;
1318                 skb->truesize += tmp_skb->truesize;
1319                 tmp_skb->destructor = NULL;
1320                 tmp_skb->sk = NULL;
1321         }
1322
1323         /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1324          * to fragment the frame generated here. No matter, what transforms
1325          * how transforms change size of the packet, it will come out.
1326          */
1327         if (inet->pmtudisc < IP_PMTUDISC_DO)
1328                 skb->local_df = 1;
1329
1330         /* DF bit is set when we want to see DF on outgoing frames.
1331          * If local_df is set too, we still allow to fragment this frame
1332          * locally. */
1333         if (inet->pmtudisc >= IP_PMTUDISC_DO ||
1334             (skb->len <= dst_mtu(&rt->dst) &&
1335              ip_dont_fragment(sk, &rt->dst)))
1336                 df = htons(IP_DF);
1337
1338         if (cork->flags & IPCORK_OPT)
1339                 opt = cork->opt;
1340
1341         if (rt->rt_type == RTN_MULTICAST)
1342                 ttl = inet->mc_ttl;
1343         else
1344                 ttl = ip_select_ttl(inet, &rt->dst);
1345
1346         iph = (struct iphdr *)skb->data;
1347         iph->version = 4;
1348         iph->ihl = 5;
1349         iph->tos = inet->tos;
1350         iph->frag_off = df;
1351         ip_select_ident(iph, &rt->dst, sk);
1352         iph->ttl = ttl;
1353         iph->protocol = sk->sk_protocol;
1354         ip_copy_addrs(iph, fl4);
1355
1356         if (opt) {
1357                 iph->ihl += opt->optlen>>2;
1358                 ip_options_build(skb, opt, cork->addr, rt, 0);
1359         }
1360
1361         skb->priority = sk->sk_priority;
1362         skb->mark = sk->sk_mark;
1363         /*
1364          * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1365          * on dst refcount
1366          */
1367         cork->dst = NULL;
1368         skb_dst_set(skb, &rt->dst);
1369
1370         if (iph->protocol == IPPROTO_ICMP)
1371                 icmp_out_count(net, ((struct icmphdr *)
1372                         skb_transport_header(skb))->type);
1373
1374         ip_cork_release(cork);
1375 out:
1376         return skb;
1377 }
1378
1379 int ip_send_skb(struct sk_buff *skb)
1380 {
1381         struct net *net = sock_net(skb->sk);
1382         int err;
1383
1384         err = ip_local_out(skb);
1385         if (err) {
1386                 if (err > 0)
1387                         err = net_xmit_errno(err);
1388                 if (err)
1389                         IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
1390         }
1391
1392         return err;
1393 }
1394
1395 int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
1396 {
1397         struct sk_buff *skb;
1398
1399         skb = ip_finish_skb(sk, fl4);
1400         if (!skb)
1401                 return 0;
1402
1403         /* Netfilter gets whole the not fragmented skb. */
1404         return ip_send_skb(skb);
1405 }
1406
1407 /*
1408  *      Throw away all pending data on the socket.
1409  */
1410 static void __ip_flush_pending_frames(struct sock *sk,
1411                                       struct sk_buff_head *queue,
1412                                       struct inet_cork *cork)
1413 {
1414         struct sk_buff *skb;
1415
1416         while ((skb = __skb_dequeue_tail(queue)) != NULL)
1417                 kfree_skb(skb);
1418
1419         ip_cork_release(cork);
1420 }
1421
1422 void ip_flush_pending_frames(struct sock *sk)
1423 {
1424         __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
1425 }
1426
1427 struct sk_buff *ip_make_skb(struct sock *sk,
1428                             struct flowi4 *fl4,
1429                             int getfrag(void *from, char *to, int offset,
1430                                         int len, int odd, struct sk_buff *skb),
1431                             void *from, int length, int transhdrlen,
1432                             struct ipcm_cookie *ipc, struct rtable **rtp,
1433                             unsigned int flags)
1434 {
1435         struct inet_cork cork;
1436         struct sk_buff_head queue;
1437         int err;
1438
1439         if (flags & MSG_PROBE)
1440                 return NULL;
1441
1442         __skb_queue_head_init(&queue);
1443
1444         cork.flags = 0;
1445         cork.addr = 0;
1446         cork.opt = NULL;
1447         err = ip_setup_cork(sk, &cork, ipc, rtp);
1448         if (err)
1449                 return ERR_PTR(err);
1450
1451         err = __ip_append_data(sk, fl4, &queue, &cork, getfrag,
1452                                from, length, transhdrlen, flags);
1453         if (err) {
1454                 __ip_flush_pending_frames(sk, &queue, &cork);
1455                 return ERR_PTR(err);
1456         }
1457
1458         return __ip_make_skb(sk, fl4, &queue, &cork);
1459 }
1460
1461 /*
1462  *      Fetch data from kernel space and fill in checksum if needed.
1463  */
1464 static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1465                               int len, int odd, struct sk_buff *skb)
1466 {
1467         __wsum csum;
1468
1469         csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1470         skb->csum = csum_block_add(skb->csum, csum, odd);
1471         return 0;
1472 }
1473
1474 /*
1475  *      Generic function to send a packet as reply to another packet.
1476  *      Used to send TCP resets so far. ICMP should use this function too.
1477  *
1478  *      Should run single threaded per socket because it uses the sock
1479  *      structure to pass arguments.
1480  */
1481 void ip_send_reply(struct sock *sk, struct sk_buff *skb, __be32 daddr,
1482                    const struct ip_reply_arg *arg, unsigned int len)
1483 {
1484         struct inet_sock *inet = inet_sk(sk);
1485         struct ip_options_data replyopts;
1486         struct ipcm_cookie ipc;
1487         struct flowi4 fl4;
1488         struct rtable *rt = skb_rtable(skb);
1489
1490         if (ip_options_echo(&replyopts.opt.opt, skb))
1491                 return;
1492
1493         ipc.addr = daddr;
1494         ipc.opt = NULL;
1495         ipc.tx_flags = 0;
1496
1497         if (replyopts.opt.opt.optlen) {
1498                 ipc.opt = &replyopts.opt;
1499
1500                 if (replyopts.opt.opt.srr)
1501                         daddr = replyopts.opt.opt.faddr;
1502         }
1503
1504         flowi4_init_output(&fl4, arg->bound_dev_if, 0,
1505                            RT_TOS(arg->tos),
1506                            RT_SCOPE_UNIVERSE, sk->sk_protocol,
1507                            ip_reply_arg_flowi_flags(arg),
1508                            daddr, rt->rt_spec_dst,
1509                            tcp_hdr(skb)->source, tcp_hdr(skb)->dest);
1510         security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
1511         rt = ip_route_output_key(sock_net(sk), &fl4);
1512         if (IS_ERR(rt))
1513                 return;
1514
1515         /* And let IP do all the hard work.
1516
1517            This chunk is not reenterable, hence spinlock.
1518            Note that it uses the fact, that this function is called
1519            with locally disabled BH and that sk cannot be already spinlocked.
1520          */
1521         bh_lock_sock(sk);
1522         inet->tos = arg->tos;
1523         sk->sk_priority = skb->priority;
1524         sk->sk_protocol = ip_hdr(skb)->protocol;
1525         sk->sk_bound_dev_if = arg->bound_dev_if;
1526         ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base, len, 0,
1527                        &ipc, &rt, MSG_DONTWAIT);
1528         if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
1529                 if (arg->csumoffset >= 0)
1530                         *((__sum16 *)skb_transport_header(skb) +
1531                           arg->csumoffset) = csum_fold(csum_add(skb->csum,
1532                                                                 arg->csum));
1533                 skb->ip_summed = CHECKSUM_NONE;
1534                 ip_push_pending_frames(sk, &fl4);
1535         }
1536
1537         bh_unlock_sock(sk);
1538
1539         ip_rt_put(rt);
1540 }
1541
1542 void __init ip_init(void)
1543 {
1544         ip_rt_init();
1545         inet_initpeers();
1546
1547 #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1548         igmp_mc_proc_init();
1549 #endif
1550 }