24a29a39e9a885dfa96300067fd37154b0875f53
[cascardo/linux.git] / 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 <linux/module.h>
47 #include <linux/types.h>
48 #include <linux/kernel.h>
49 #include <linux/mm.h>
50 #include <linux/string.h>
51 #include <linux/errno.h>
52 #include <linux/highmem.h>
53 #include <linux/slab.h>
54
55 #include <linux/socket.h>
56 #include <linux/sockios.h>
57 #include <linux/in.h>
58 #include <linux/inet.h>
59 #include <linux/netdevice.h>
60 #include <linux/etherdevice.h>
61 #include <linux/proc_fs.h>
62 #include <linux/stat.h>
63 #include <linux/init.h>
64
65 #include <net/snmp.h>
66 #include <net/ip.h>
67 #include <net/protocol.h>
68 #include <net/route.h>
69 #include <net/xfrm.h>
70 #include <linux/skbuff.h>
71 #include <net/sock.h>
72 #include <net/arp.h>
73 #include <net/icmp.h>
74 #include <net/checksum.h>
75 #include <net/inetpeer.h>
76 #include <linux/igmp.h>
77 #include <linux/netfilter_ipv4.h>
78 #include <linux/netfilter_bridge.h>
79 #include <linux/mroute.h>
80 #include <linux/netlink.h>
81 #include <linux/tcp.h>
82
83 int sysctl_ip_default_ttl __read_mostly = IPDEFTTL;
84 EXPORT_SYMBOL(sysctl_ip_default_ttl);
85
86 /* Generate a checksum for an outgoing IP datagram. */
87 __inline__ void ip_send_check(struct iphdr *iph)
88 {
89         iph->check = 0;
90         iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
91 }
92 EXPORT_SYMBOL(ip_send_check);
93
94 int __ip_local_out(struct sk_buff *skb)
95 {
96         struct iphdr *iph = ip_hdr(skb);
97
98         iph->tot_len = htons(skb->len);
99         ip_send_check(iph);
100         return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, skb, NULL,
101                        skb_dst(skb)->dev, dst_output);
102 }
103
104 int ip_local_out(struct sk_buff *skb)
105 {
106         int err;
107
108         err = __ip_local_out(skb);
109         if (likely(err == 1))
110                 err = dst_output(skb);
111
112         return err;
113 }
114 EXPORT_SYMBOL_GPL(ip_local_out);
115
116 static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
117 {
118         int ttl = inet->uc_ttl;
119
120         if (ttl < 0)
121                 ttl = ip4_dst_hoplimit(dst);
122         return ttl;
123 }
124
125 /*
126  *              Add an ip header to a skbuff and send it out.
127  *
128  */
129 int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk,
130                           __be32 saddr, __be32 daddr, struct ip_options_rcu *opt)
131 {
132         struct inet_sock *inet = inet_sk(sk);
133         struct rtable *rt = skb_rtable(skb);
134         struct iphdr *iph;
135
136         /* Build the IP header. */
137         skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
138         skb_reset_network_header(skb);
139         iph = ip_hdr(skb);
140         iph->version  = 4;
141         iph->ihl      = 5;
142         iph->tos      = inet->tos;
143         if (ip_dont_fragment(sk, &rt->dst))
144                 iph->frag_off = htons(IP_DF);
145         else
146                 iph->frag_off = 0;
147         iph->ttl      = ip_select_ttl(inet, &rt->dst);
148         iph->daddr    = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
149         iph->saddr    = saddr;
150         iph->protocol = sk->sk_protocol;
151         ip_select_ident(iph, &rt->dst, sk);
152
153         if (opt && opt->opt.optlen) {
154                 iph->ihl += opt->opt.optlen>>2;
155                 ip_options_build(skb, &opt->opt, daddr, rt, 0);
156         }
157
158         skb->priority = sk->sk_priority;
159         skb->mark = sk->sk_mark;
160
161         /* Send it out. */
162         return ip_local_out(skb);
163 }
164 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
165
166 static inline int ip_finish_output2(struct sk_buff *skb)
167 {
168         struct dst_entry *dst = skb_dst(skb);
169         struct rtable *rt = (struct rtable *)dst;
170         struct net_device *dev = dst->dev;
171         unsigned int hh_len = LL_RESERVED_SPACE(dev);
172         struct neighbour *neigh;
173         u32 nexthop;
174
175         if (rt->rt_type == RTN_MULTICAST) {
176                 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTMCAST, skb->len);
177         } else if (rt->rt_type == RTN_BROADCAST)
178                 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTBCAST, skb->len);
179
180         /* Be paranoid, rather than too clever. */
181         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
182                 struct sk_buff *skb2;
183
184                 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
185                 if (skb2 == NULL) {
186                         kfree_skb(skb);
187                         return -ENOMEM;
188                 }
189                 if (skb->sk)
190                         skb_set_owner_w(skb2, skb->sk);
191                 consume_skb(skb);
192                 skb = skb2;
193         }
194
195         rcu_read_lock_bh();
196         nexthop = rt->rt_gateway ? rt->rt_gateway : ip_hdr(skb)->daddr;
197         neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
198         if (unlikely(!neigh))
199                 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
200         if (!IS_ERR(neigh)) {
201                 int res = dst_neigh_output(dst, neigh, skb);
202
203                 rcu_read_unlock_bh();
204                 return res;
205         }
206         rcu_read_unlock_bh();
207
208         net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
209                             __func__);
210         kfree_skb(skb);
211         return -EINVAL;
212 }
213
214 static inline int ip_skb_dst_mtu(struct sk_buff *skb)
215 {
216         struct inet_sock *inet = skb->sk ? inet_sk(skb->sk) : NULL;
217
218         return (inet && inet->pmtudisc == IP_PMTUDISC_PROBE) ?
219                skb_dst(skb)->dev->mtu : dst_mtu(skb_dst(skb));
220 }
221
222 static int ip_finish_output(struct sk_buff *skb)
223 {
224 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
225         /* Policy lookup after SNAT yielded a new policy */
226         if (skb_dst(skb)->xfrm != NULL) {
227                 IPCB(skb)->flags |= IPSKB_REROUTED;
228                 return dst_output(skb);
229         }
230 #endif
231         if (skb->len > ip_skb_dst_mtu(skb) && !skb_is_gso(skb))
232                 return ip_fragment(skb, ip_finish_output2);
233         else
234                 return ip_finish_output2(skb);
235 }
236
237 int ip_mc_output(struct sk_buff *skb)
238 {
239         struct sock *sk = skb->sk;
240         struct rtable *rt = skb_rtable(skb);
241         struct net_device *dev = rt->dst.dev;
242
243         /*
244          *      If the indicated interface is up and running, send the packet.
245          */
246         IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
247
248         skb->dev = dev;
249         skb->protocol = htons(ETH_P_IP);
250
251         /*
252          *      Multicasts are looped back for other local users
253          */
254
255         if (rt->rt_flags&RTCF_MULTICAST) {
256                 if (sk_mc_loop(sk)
257 #ifdef CONFIG_IP_MROUTE
258                 /* Small optimization: do not loopback not local frames,
259                    which returned after forwarding; they will be  dropped
260                    by ip_mr_input in any case.
261                    Note, that local frames are looped back to be delivered
262                    to local recipients.
263
264                    This check is duplicated in ip_mr_input at the moment.
265                  */
266                     &&
267                     ((rt->rt_flags & RTCF_LOCAL) ||
268                      !(IPCB(skb)->flags & IPSKB_FORWARDED))
269 #endif
270                    ) {
271                         struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
272                         if (newskb)
273                                 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
274                                         newskb, NULL, newskb->dev,
275                                         dev_loopback_xmit);
276                 }
277
278                 /* Multicasts with ttl 0 must not go beyond the host */
279
280                 if (ip_hdr(skb)->ttl == 0) {
281                         kfree_skb(skb);
282                         return 0;
283                 }
284         }
285
286         if (rt->rt_flags&RTCF_BROADCAST) {
287                 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
288                 if (newskb)
289                         NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, newskb,
290                                 NULL, newskb->dev, dev_loopback_xmit);
291         }
292
293         return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL,
294                             skb->dev, ip_finish_output,
295                             !(IPCB(skb)->flags & IPSKB_REROUTED));
296 }
297
298 int ip_output(struct sk_buff *skb)
299 {
300         struct net_device *dev = skb_dst(skb)->dev;
301
302         IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
303
304         skb->dev = dev;
305         skb->protocol = htons(ETH_P_IP);
306
307         return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL, dev,
308                             ip_finish_output,
309                             !(IPCB(skb)->flags & IPSKB_REROUTED));
310 }
311
312 /*
313  * copy saddr and daddr, possibly using 64bit load/stores
314  * Equivalent to :
315  *   iph->saddr = fl4->saddr;
316  *   iph->daddr = fl4->daddr;
317  */
318 static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
319 {
320         BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
321                      offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
322         memcpy(&iph->saddr, &fl4->saddr,
323                sizeof(fl4->saddr) + sizeof(fl4->daddr));
324 }
325
326 int ip_queue_xmit(struct sk_buff *skb, struct flowi *fl)
327 {
328         struct sock *sk = skb->sk;
329         struct inet_sock *inet = inet_sk(sk);
330         struct ip_options_rcu *inet_opt;
331         struct flowi4 *fl4;
332         struct rtable *rt;
333         struct iphdr *iph;
334         int res;
335
336         /* Skip all of this if the packet is already routed,
337          * f.e. by something like SCTP.
338          */
339         rcu_read_lock();
340         inet_opt = rcu_dereference(inet->inet_opt);
341         fl4 = &fl->u.ip4;
342         rt = skb_rtable(skb);
343         if (rt != NULL)
344                 goto packet_routed;
345
346         /* Make sure we can route this packet. */
347         rt = (struct rtable *)__sk_dst_check(sk, 0);
348         if (rt == NULL) {
349                 __be32 daddr;
350
351                 /* Use correct destination address if we have options. */
352                 daddr = inet->inet_daddr;
353                 if (inet_opt && inet_opt->opt.srr)
354                         daddr = inet_opt->opt.faddr;
355
356                 /* If this fails, retransmit mechanism of transport layer will
357                  * keep trying until route appears or the connection times
358                  * itself out.
359                  */
360                 rt = ip_route_output_ports(sock_net(sk), fl4, sk,
361                                            daddr, inet->inet_saddr,
362                                            inet->inet_dport,
363                                            inet->inet_sport,
364                                            sk->sk_protocol,
365                                            RT_CONN_FLAGS(sk),
366                                            sk->sk_bound_dev_if);
367                 if (IS_ERR(rt))
368                         goto no_route;
369                 sk_setup_caps(sk, &rt->dst);
370         }
371         skb_dst_set_noref(skb, &rt->dst);
372
373 packet_routed:
374         if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_gateway)
375                 goto no_route;
376
377         /* OK, we know where to send it, allocate and build IP header. */
378         skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
379         skb_reset_network_header(skb);
380         iph = ip_hdr(skb);
381         *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
382         if (ip_dont_fragment(sk, &rt->dst) && !skb->local_df)
383                 iph->frag_off = htons(IP_DF);
384         else
385                 iph->frag_off = 0;
386         iph->ttl      = ip_select_ttl(inet, &rt->dst);
387         iph->protocol = sk->sk_protocol;
388         ip_copy_addrs(iph, fl4);
389
390         /* Transport layer set skb->h.foo itself. */
391
392         if (inet_opt && inet_opt->opt.optlen) {
393                 iph->ihl += inet_opt->opt.optlen >> 2;
394                 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0);
395         }
396
397         ip_select_ident_more(iph, &rt->dst, sk,
398                              (skb_shinfo(skb)->gso_segs ?: 1) - 1);
399
400         skb->priority = sk->sk_priority;
401         skb->mark = sk->sk_mark;
402
403         res = ip_local_out(skb);
404         rcu_read_unlock();
405         return res;
406
407 no_route:
408         rcu_read_unlock();
409         IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
410         kfree_skb(skb);
411         return -EHOSTUNREACH;
412 }
413 EXPORT_SYMBOL(ip_queue_xmit);
414
415
416 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
417 {
418         to->pkt_type = from->pkt_type;
419         to->priority = from->priority;
420         to->protocol = from->protocol;
421         skb_dst_drop(to);
422         skb_dst_copy(to, from);
423         to->dev = from->dev;
424         to->mark = from->mark;
425
426         /* Copy the flags to each fragment. */
427         IPCB(to)->flags = IPCB(from)->flags;
428
429 #ifdef CONFIG_NET_SCHED
430         to->tc_index = from->tc_index;
431 #endif
432         nf_copy(to, from);
433 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
434     defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
435         to->nf_trace = from->nf_trace;
436 #endif
437 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
438         to->ipvs_property = from->ipvs_property;
439 #endif
440         skb_copy_secmark(to, from);
441 }
442
443 /*
444  *      This IP datagram is too large to be sent in one piece.  Break it up into
445  *      smaller pieces (each of size equal to IP header plus
446  *      a block of the data of the original IP data part) that will yet fit in a
447  *      single device frame, and queue such a frame for sending.
448  */
449
450 int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
451 {
452         struct iphdr *iph;
453         int ptr;
454         struct net_device *dev;
455         struct sk_buff *skb2;
456         unsigned int mtu, hlen, left, len, ll_rs;
457         int offset;
458         __be16 not_last_frag;
459         struct rtable *rt = skb_rtable(skb);
460         int err = 0;
461
462         dev = rt->dst.dev;
463
464         /*
465          *      Point into the IP datagram header.
466          */
467
468         iph = ip_hdr(skb);
469
470         if (unlikely(((iph->frag_off & htons(IP_DF)) && !skb->local_df) ||
471                      (IPCB(skb)->frag_max_size &&
472                       IPCB(skb)->frag_max_size > dst_mtu(&rt->dst)))) {
473                 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
474                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
475                           htonl(ip_skb_dst_mtu(skb)));
476                 kfree_skb(skb);
477                 return -EMSGSIZE;
478         }
479
480         /*
481          *      Setup starting values.
482          */
483
484         hlen = iph->ihl * 4;
485         mtu = dst_mtu(&rt->dst) - hlen; /* Size of data space */
486 #ifdef CONFIG_BRIDGE_NETFILTER
487         if (skb->nf_bridge)
488                 mtu -= nf_bridge_mtu_reduction(skb);
489 #endif
490         IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
491
492         /* When frag_list is given, use it. First, check its validity:
493          * some transformers could create wrong frag_list or break existing
494          * one, it is not prohibited. In this case fall back to copying.
495          *
496          * LATER: this step can be merged to real generation of fragments,
497          * we can switch to copy when see the first bad fragment.
498          */
499         if (skb_has_frag_list(skb)) {
500                 struct sk_buff *frag, *frag2;
501                 int first_len = skb_pagelen(skb);
502
503                 if (first_len - hlen > mtu ||
504                     ((first_len - hlen) & 7) ||
505                     ip_is_fragment(iph) ||
506                     skb_cloned(skb))
507                         goto slow_path;
508
509                 skb_walk_frags(skb, frag) {
510                         /* Correct geometry. */
511                         if (frag->len > mtu ||
512                             ((frag->len & 7) && frag->next) ||
513                             skb_headroom(frag) < hlen)
514                                 goto slow_path_clean;
515
516                         /* Partially cloned skb? */
517                         if (skb_shared(frag))
518                                 goto slow_path_clean;
519
520                         BUG_ON(frag->sk);
521                         if (skb->sk) {
522                                 frag->sk = skb->sk;
523                                 frag->destructor = sock_wfree;
524                         }
525                         skb->truesize -= frag->truesize;
526                 }
527
528                 /* Everything is OK. Generate! */
529
530                 err = 0;
531                 offset = 0;
532                 frag = skb_shinfo(skb)->frag_list;
533                 skb_frag_list_init(skb);
534                 skb->data_len = first_len - skb_headlen(skb);
535                 skb->len = first_len;
536                 iph->tot_len = htons(first_len);
537                 iph->frag_off = htons(IP_MF);
538                 ip_send_check(iph);
539
540                 for (;;) {
541                         /* Prepare header of the next frame,
542                          * before previous one went down. */
543                         if (frag) {
544                                 frag->ip_summed = CHECKSUM_NONE;
545                                 skb_reset_transport_header(frag);
546                                 __skb_push(frag, hlen);
547                                 skb_reset_network_header(frag);
548                                 memcpy(skb_network_header(frag), iph, hlen);
549                                 iph = ip_hdr(frag);
550                                 iph->tot_len = htons(frag->len);
551                                 ip_copy_metadata(frag, skb);
552                                 if (offset == 0)
553                                         ip_options_fragment(frag);
554                                 offset += skb->len - hlen;
555                                 iph->frag_off = htons(offset>>3);
556                                 if (frag->next != NULL)
557                                         iph->frag_off |= htons(IP_MF);
558                                 /* Ready, complete checksum */
559                                 ip_send_check(iph);
560                         }
561
562                         err = output(skb);
563
564                         if (!err)
565                                 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
566                         if (err || !frag)
567                                 break;
568
569                         skb = frag;
570                         frag = skb->next;
571                         skb->next = NULL;
572                 }
573
574                 if (err == 0) {
575                         IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
576                         return 0;
577                 }
578
579                 while (frag) {
580                         skb = frag->next;
581                         kfree_skb(frag);
582                         frag = skb;
583                 }
584                 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
585                 return err;
586
587 slow_path_clean:
588                 skb_walk_frags(skb, frag2) {
589                         if (frag2 == frag)
590                                 break;
591                         frag2->sk = NULL;
592                         frag2->destructor = NULL;
593                         skb->truesize += frag2->truesize;
594                 }
595         }
596
597 slow_path:
598         left = skb->len - hlen;         /* Space per frame */
599         ptr = hlen;             /* Where to start from */
600
601         /* for bridged IP traffic encapsulated inside f.e. a vlan header,
602          * we need to make room for the encapsulating header
603          */
604         ll_rs = LL_RESERVED_SPACE_EXTRA(rt->dst.dev, nf_bridge_pad(skb));
605
606         /*
607          *      Fragment the datagram.
608          */
609
610         offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
611         not_last_frag = iph->frag_off & htons(IP_MF);
612
613         /*
614          *      Keep copying data until we run out.
615          */
616
617         while (left > 0) {
618                 len = left;
619                 /* IF: it doesn't fit, use 'mtu' - the data space left */
620                 if (len > mtu)
621                         len = mtu;
622                 /* IF: we are not sending up to and including the packet end
623                    then align the next start on an eight byte boundary */
624                 if (len < left) {
625                         len &= ~7;
626                 }
627                 /*
628                  *      Allocate buffer.
629                  */
630
631                 if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) {
632                         NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n");
633                         err = -ENOMEM;
634                         goto fail;
635                 }
636
637                 /*
638                  *      Set up data on packet
639                  */
640
641                 ip_copy_metadata(skb2, skb);
642                 skb_reserve(skb2, ll_rs);
643                 skb_put(skb2, len + hlen);
644                 skb_reset_network_header(skb2);
645                 skb2->transport_header = skb2->network_header + hlen;
646
647                 /*
648                  *      Charge the memory for the fragment to any owner
649                  *      it might possess
650                  */
651
652                 if (skb->sk)
653                         skb_set_owner_w(skb2, skb->sk);
654
655                 /*
656                  *      Copy the packet header into the new buffer.
657                  */
658
659                 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
660
661                 /*
662                  *      Copy a block of the IP datagram.
663                  */
664                 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
665                         BUG();
666                 left -= len;
667
668                 /*
669                  *      Fill in the new header fields.
670                  */
671                 iph = ip_hdr(skb2);
672                 iph->frag_off = htons((offset >> 3));
673
674                 /* ANK: dirty, but effective trick. Upgrade options only if
675                  * the segment to be fragmented was THE FIRST (otherwise,
676                  * options are already fixed) and make it ONCE
677                  * on the initial skb, so that all the following fragments
678                  * will inherit fixed options.
679                  */
680                 if (offset == 0)
681                         ip_options_fragment(skb);
682
683                 /*
684                  *      Added AC : If we are fragmenting a fragment that's not the
685                  *                 last fragment then keep MF on each bit
686                  */
687                 if (left > 0 || not_last_frag)
688                         iph->frag_off |= htons(IP_MF);
689                 ptr += len;
690                 offset += len;
691
692                 /*
693                  *      Put this fragment into the sending queue.
694                  */
695                 iph->tot_len = htons(len + hlen);
696
697                 ip_send_check(iph);
698
699                 err = output(skb2);
700                 if (err)
701                         goto fail;
702
703                 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
704         }
705         consume_skb(skb);
706         IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
707         return err;
708
709 fail:
710         kfree_skb(skb);
711         IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
712         return err;
713 }
714 EXPORT_SYMBOL(ip_fragment);
715
716 int
717 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
718 {
719         struct iovec *iov = from;
720
721         if (skb->ip_summed == CHECKSUM_PARTIAL) {
722                 if (memcpy_fromiovecend(to, iov, offset, len) < 0)
723                         return -EFAULT;
724         } else {
725                 __wsum csum = 0;
726                 if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0)
727                         return -EFAULT;
728                 skb->csum = csum_block_add(skb->csum, csum, odd);
729         }
730         return 0;
731 }
732 EXPORT_SYMBOL(ip_generic_getfrag);
733
734 static inline __wsum
735 csum_page(struct page *page, int offset, int copy)
736 {
737         char *kaddr;
738         __wsum csum;
739         kaddr = kmap(page);
740         csum = csum_partial(kaddr + offset, copy, 0);
741         kunmap(page);
742         return csum;
743 }
744
745 static inline int ip_ufo_append_data(struct sock *sk,
746                         struct sk_buff_head *queue,
747                         int getfrag(void *from, char *to, int offset, int len,
748                                int odd, struct sk_buff *skb),
749                         void *from, int length, int hh_len, int fragheaderlen,
750                         int transhdrlen, int maxfraglen, unsigned int flags)
751 {
752         struct sk_buff *skb;
753         int err;
754
755         /* There is support for UDP fragmentation offload by network
756          * device, so create one single skb packet containing complete
757          * udp datagram
758          */
759         if ((skb = skb_peek_tail(queue)) == NULL) {
760                 skb = sock_alloc_send_skb(sk,
761                         hh_len + fragheaderlen + transhdrlen + 20,
762                         (flags & MSG_DONTWAIT), &err);
763
764                 if (skb == NULL)
765                         return err;
766
767                 /* reserve space for Hardware header */
768                 skb_reserve(skb, hh_len);
769
770                 /* create space for UDP/IP header */
771                 skb_put(skb, fragheaderlen + transhdrlen);
772
773                 /* initialize network header pointer */
774                 skb_reset_network_header(skb);
775
776                 /* initialize protocol header pointer */
777                 skb->transport_header = skb->network_header + fragheaderlen;
778
779                 skb->ip_summed = CHECKSUM_PARTIAL;
780                 skb->csum = 0;
781
782                 /* specify the length of each IP datagram fragment */
783                 skb_shinfo(skb)->gso_size = maxfraglen - fragheaderlen;
784                 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
785                 __skb_queue_tail(queue, skb);
786         }
787
788         return skb_append_datato_frags(sk, skb, getfrag, from,
789                                        (length - transhdrlen));
790 }
791
792 static int __ip_append_data(struct sock *sk,
793                             struct flowi4 *fl4,
794                             struct sk_buff_head *queue,
795                             struct inet_cork *cork,
796                             struct page_frag *pfrag,
797                             int getfrag(void *from, char *to, int offset,
798                                         int len, int odd, struct sk_buff *skb),
799                             void *from, int length, int transhdrlen,
800                             unsigned int flags)
801 {
802         struct inet_sock *inet = inet_sk(sk);
803         struct sk_buff *skb;
804
805         struct ip_options *opt = cork->opt;
806         int hh_len;
807         int exthdrlen;
808         int mtu;
809         int copy;
810         int err;
811         int offset = 0;
812         unsigned int maxfraglen, fragheaderlen;
813         int csummode = CHECKSUM_NONE;
814         struct rtable *rt = (struct rtable *)cork->dst;
815
816         skb = skb_peek_tail(queue);
817
818         exthdrlen = !skb ? rt->dst.header_len : 0;
819         mtu = cork->fragsize;
820
821         hh_len = LL_RESERVED_SPACE(rt->dst.dev);
822
823         fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
824         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
825
826         if (cork->length + length > 0xFFFF - fragheaderlen) {
827                 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
828                                mtu-exthdrlen);
829                 return -EMSGSIZE;
830         }
831
832         /*
833          * transhdrlen > 0 means that this is the first fragment and we wish
834          * it won't be fragmented in the future.
835          */
836         if (transhdrlen &&
837             length + fragheaderlen <= mtu &&
838             rt->dst.dev->features & NETIF_F_V4_CSUM &&
839             !exthdrlen)
840                 csummode = CHECKSUM_PARTIAL;
841
842         cork->length += length;
843         if (((length > mtu) || (skb && skb_is_gso(skb))) &&
844             (sk->sk_protocol == IPPROTO_UDP) &&
845             (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len) {
846                 err = ip_ufo_append_data(sk, queue, getfrag, from, length,
847                                          hh_len, fragheaderlen, transhdrlen,
848                                          maxfraglen, flags);
849                 if (err)
850                         goto error;
851                 return 0;
852         }
853
854         /* So, what's going on in the loop below?
855          *
856          * We use calculated fragment length to generate chained skb,
857          * each of segments is IP fragment ready for sending to network after
858          * adding appropriate IP header.
859          */
860
861         if (!skb)
862                 goto alloc_new_skb;
863
864         while (length > 0) {
865                 /* Check if the remaining data fits into current packet. */
866                 copy = mtu - skb->len;
867                 if (copy < length)
868                         copy = maxfraglen - skb->len;
869                 if (copy <= 0) {
870                         char *data;
871                         unsigned int datalen;
872                         unsigned int fraglen;
873                         unsigned int fraggap;
874                         unsigned int alloclen;
875                         struct sk_buff *skb_prev;
876 alloc_new_skb:
877                         skb_prev = skb;
878                         if (skb_prev)
879                                 fraggap = skb_prev->len - maxfraglen;
880                         else
881                                 fraggap = 0;
882
883                         /*
884                          * If remaining data exceeds the mtu,
885                          * we know we need more fragment(s).
886                          */
887                         datalen = length + fraggap;
888                         if (datalen > mtu - fragheaderlen)
889                                 datalen = maxfraglen - fragheaderlen;
890                         fraglen = datalen + fragheaderlen;
891
892                         if ((flags & MSG_MORE) &&
893                             !(rt->dst.dev->features&NETIF_F_SG))
894                                 alloclen = mtu;
895                         else
896                                 alloclen = fraglen;
897
898                         alloclen += exthdrlen;
899
900                         /* The last fragment gets additional space at tail.
901                          * Note, with MSG_MORE we overallocate on fragments,
902                          * because we have no idea what fragment will be
903                          * the last.
904                          */
905                         if (datalen == length + fraggap)
906                                 alloclen += rt->dst.trailer_len;
907
908                         if (transhdrlen) {
909                                 skb = sock_alloc_send_skb(sk,
910                                                 alloclen + hh_len + 15,
911                                                 (flags & MSG_DONTWAIT), &err);
912                         } else {
913                                 skb = NULL;
914                                 if (atomic_read(&sk->sk_wmem_alloc) <=
915                                     2 * sk->sk_sndbuf)
916                                         skb = sock_wmalloc(sk,
917                                                            alloclen + hh_len + 15, 1,
918                                                            sk->sk_allocation);
919                                 if (unlikely(skb == NULL))
920                                         err = -ENOBUFS;
921                                 else
922                                         /* only the initial fragment is
923                                            time stamped */
924                                         cork->tx_flags = 0;
925                         }
926                         if (skb == NULL)
927                                 goto error;
928
929                         /*
930                          *      Fill in the control structures
931                          */
932                         skb->ip_summed = csummode;
933                         skb->csum = 0;
934                         skb_reserve(skb, hh_len);
935                         skb_shinfo(skb)->tx_flags = cork->tx_flags;
936
937                         /*
938                          *      Find where to start putting bytes.
939                          */
940                         data = skb_put(skb, fraglen + exthdrlen);
941                         skb_set_network_header(skb, exthdrlen);
942                         skb->transport_header = (skb->network_header +
943                                                  fragheaderlen);
944                         data += fragheaderlen + exthdrlen;
945
946                         if (fraggap) {
947                                 skb->csum = skb_copy_and_csum_bits(
948                                         skb_prev, maxfraglen,
949                                         data + transhdrlen, fraggap, 0);
950                                 skb_prev->csum = csum_sub(skb_prev->csum,
951                                                           skb->csum);
952                                 data += fraggap;
953                                 pskb_trim_unique(skb_prev, maxfraglen);
954                         }
955
956                         copy = datalen - transhdrlen - fraggap;
957                         if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
958                                 err = -EFAULT;
959                                 kfree_skb(skb);
960                                 goto error;
961                         }
962
963                         offset += copy;
964                         length -= datalen - fraggap;
965                         transhdrlen = 0;
966                         exthdrlen = 0;
967                         csummode = CHECKSUM_NONE;
968
969                         /*
970                          * Put the packet on the pending queue.
971                          */
972                         __skb_queue_tail(queue, skb);
973                         continue;
974                 }
975
976                 if (copy > length)
977                         copy = length;
978
979                 if (!(rt->dst.dev->features&NETIF_F_SG)) {
980                         unsigned int off;
981
982                         off = skb->len;
983                         if (getfrag(from, skb_put(skb, copy),
984                                         offset, copy, off, skb) < 0) {
985                                 __skb_trim(skb, off);
986                                 err = -EFAULT;
987                                 goto error;
988                         }
989                 } else {
990                         int i = skb_shinfo(skb)->nr_frags;
991
992                         err = -ENOMEM;
993                         if (!sk_page_frag_refill(sk, pfrag))
994                                 goto error;
995
996                         if (!skb_can_coalesce(skb, i, pfrag->page,
997                                               pfrag->offset)) {
998                                 err = -EMSGSIZE;
999                                 if (i == MAX_SKB_FRAGS)
1000                                         goto error;
1001
1002                                 __skb_fill_page_desc(skb, i, pfrag->page,
1003                                                      pfrag->offset, 0);
1004                                 skb_shinfo(skb)->nr_frags = ++i;
1005                                 get_page(pfrag->page);
1006                         }
1007                         copy = min_t(int, copy, pfrag->size - pfrag->offset);
1008                         if (getfrag(from,
1009                                     page_address(pfrag->page) + pfrag->offset,
1010                                     offset, copy, skb->len, skb) < 0)
1011                                 goto error_efault;
1012
1013                         pfrag->offset += copy;
1014                         skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1015                         skb->len += copy;
1016                         skb->data_len += copy;
1017                         skb->truesize += copy;
1018                         atomic_add(copy, &sk->sk_wmem_alloc);
1019                 }
1020                 offset += copy;
1021                 length -= copy;
1022         }
1023
1024         return 0;
1025
1026 error_efault:
1027         err = -EFAULT;
1028 error:
1029         cork->length -= length;
1030         IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1031         return err;
1032 }
1033
1034 static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
1035                          struct ipcm_cookie *ipc, struct rtable **rtp)
1036 {
1037         struct inet_sock *inet = inet_sk(sk);
1038         struct ip_options_rcu *opt;
1039         struct rtable *rt;
1040
1041         /*
1042          * setup for corking.
1043          */
1044         opt = ipc->opt;
1045         if (opt) {
1046                 if (cork->opt == NULL) {
1047                         cork->opt = kmalloc(sizeof(struct ip_options) + 40,
1048                                             sk->sk_allocation);
1049                         if (unlikely(cork->opt == NULL))
1050                                 return -ENOBUFS;
1051                 }
1052                 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
1053                 cork->flags |= IPCORK_OPT;
1054                 cork->addr = ipc->addr;
1055         }
1056         rt = *rtp;
1057         if (unlikely(!rt))
1058                 return -EFAULT;
1059         /*
1060          * We steal reference to this route, caller should not release it
1061          */
1062         *rtp = NULL;
1063         cork->fragsize = inet->pmtudisc == IP_PMTUDISC_PROBE ?
1064                          rt->dst.dev->mtu : dst_mtu(&rt->dst);
1065         cork->dst = &rt->dst;
1066         cork->length = 0;
1067         cork->tx_flags = ipc->tx_flags;
1068
1069         return 0;
1070 }
1071
1072 /*
1073  *      ip_append_data() and ip_append_page() can make one large IP datagram
1074  *      from many pieces of data. Each pieces will be holded on the socket
1075  *      until ip_push_pending_frames() is called. Each piece can be a page
1076  *      or non-page data.
1077  *
1078  *      Not only UDP, other transport protocols - e.g. raw sockets - can use
1079  *      this interface potentially.
1080  *
1081  *      LATER: length must be adjusted by pad at tail, when it is required.
1082  */
1083 int ip_append_data(struct sock *sk, struct flowi4 *fl4,
1084                    int getfrag(void *from, char *to, int offset, int len,
1085                                int odd, struct sk_buff *skb),
1086                    void *from, int length, int transhdrlen,
1087                    struct ipcm_cookie *ipc, struct rtable **rtp,
1088                    unsigned int flags)
1089 {
1090         struct inet_sock *inet = inet_sk(sk);
1091         int err;
1092
1093         if (flags&MSG_PROBE)
1094                 return 0;
1095
1096         if (skb_queue_empty(&sk->sk_write_queue)) {
1097                 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
1098                 if (err)
1099                         return err;
1100         } else {
1101                 transhdrlen = 0;
1102         }
1103
1104         return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
1105                                 sk_page_frag(sk), getfrag,
1106                                 from, length, transhdrlen, flags);
1107 }
1108
1109 ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
1110                        int offset, size_t size, int flags)
1111 {
1112         struct inet_sock *inet = inet_sk(sk);
1113         struct sk_buff *skb;
1114         struct rtable *rt;
1115         struct ip_options *opt = NULL;
1116         struct inet_cork *cork;
1117         int hh_len;
1118         int mtu;
1119         int len;
1120         int err;
1121         unsigned int maxfraglen, fragheaderlen, fraggap;
1122
1123         if (inet->hdrincl)
1124                 return -EPERM;
1125
1126         if (flags&MSG_PROBE)
1127                 return 0;
1128
1129         if (skb_queue_empty(&sk->sk_write_queue))
1130                 return -EINVAL;
1131
1132         cork = &inet->cork.base;
1133         rt = (struct rtable *)cork->dst;
1134         if (cork->flags & IPCORK_OPT)
1135                 opt = cork->opt;
1136
1137         if (!(rt->dst.dev->features&NETIF_F_SG))
1138                 return -EOPNOTSUPP;
1139
1140         hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1141         mtu = cork->fragsize;
1142
1143         fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1144         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1145
1146         if (cork->length + size > 0xFFFF - fragheaderlen) {
1147                 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, mtu);
1148                 return -EMSGSIZE;
1149         }
1150
1151         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1152                 return -EINVAL;
1153
1154         cork->length += size;
1155         if ((size + skb->len > mtu) &&
1156             (sk->sk_protocol == IPPROTO_UDP) &&
1157             (rt->dst.dev->features & NETIF_F_UFO)) {
1158                 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
1159                 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1160         }
1161
1162
1163         while (size > 0) {
1164                 int i;
1165
1166                 if (skb_is_gso(skb))
1167                         len = size;
1168                 else {
1169
1170                         /* Check if the remaining data fits into current packet. */
1171                         len = mtu - skb->len;
1172                         if (len < size)
1173                                 len = maxfraglen - skb->len;
1174                 }
1175                 if (len <= 0) {
1176                         struct sk_buff *skb_prev;
1177                         int alloclen;
1178
1179                         skb_prev = skb;
1180                         fraggap = skb_prev->len - maxfraglen;
1181
1182                         alloclen = fragheaderlen + hh_len + fraggap + 15;
1183                         skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1184                         if (unlikely(!skb)) {
1185                                 err = -ENOBUFS;
1186                                 goto error;
1187                         }
1188
1189                         /*
1190                          *      Fill in the control structures
1191                          */
1192                         skb->ip_summed = CHECKSUM_NONE;
1193                         skb->csum = 0;
1194                         skb_reserve(skb, hh_len);
1195
1196                         /*
1197                          *      Find where to start putting bytes.
1198                          */
1199                         skb_put(skb, fragheaderlen + fraggap);
1200                         skb_reset_network_header(skb);
1201                         skb->transport_header = (skb->network_header +
1202                                                  fragheaderlen);
1203                         if (fraggap) {
1204                                 skb->csum = skb_copy_and_csum_bits(skb_prev,
1205                                                                    maxfraglen,
1206                                                     skb_transport_header(skb),
1207                                                                    fraggap, 0);
1208                                 skb_prev->csum = csum_sub(skb_prev->csum,
1209                                                           skb->csum);
1210                                 pskb_trim_unique(skb_prev, maxfraglen);
1211                         }
1212
1213                         /*
1214                          * Put the packet on the pending queue.
1215                          */
1216                         __skb_queue_tail(&sk->sk_write_queue, skb);
1217                         continue;
1218                 }
1219
1220                 i = skb_shinfo(skb)->nr_frags;
1221                 if (len > size)
1222                         len = size;
1223                 if (skb_can_coalesce(skb, i, page, offset)) {
1224                         skb_frag_size_add(&skb_shinfo(skb)->frags[i-1], len);
1225                 } else if (i < MAX_SKB_FRAGS) {
1226                         get_page(page);
1227                         skb_fill_page_desc(skb, i, page, offset, len);
1228                 } else {
1229                         err = -EMSGSIZE;
1230                         goto error;
1231                 }
1232
1233                 if (skb->ip_summed == CHECKSUM_NONE) {
1234                         __wsum csum;
1235                         csum = csum_page(page, offset, len);
1236                         skb->csum = csum_block_add(skb->csum, csum, skb->len);
1237                 }
1238
1239                 skb->len += len;
1240                 skb->data_len += len;
1241                 skb->truesize += len;
1242                 atomic_add(len, &sk->sk_wmem_alloc);
1243                 offset += len;
1244                 size -= len;
1245         }
1246         return 0;
1247
1248 error:
1249         cork->length -= size;
1250         IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1251         return err;
1252 }
1253
1254 static void ip_cork_release(struct inet_cork *cork)
1255 {
1256         cork->flags &= ~IPCORK_OPT;
1257         kfree(cork->opt);
1258         cork->opt = NULL;
1259         dst_release(cork->dst);
1260         cork->dst = NULL;
1261 }
1262
1263 /*
1264  *      Combined all pending IP fragments on the socket as one IP datagram
1265  *      and push them out.
1266  */
1267 struct sk_buff *__ip_make_skb(struct sock *sk,
1268                               struct flowi4 *fl4,
1269                               struct sk_buff_head *queue,
1270                               struct inet_cork *cork)
1271 {
1272         struct sk_buff *skb, *tmp_skb;
1273         struct sk_buff **tail_skb;
1274         struct inet_sock *inet = inet_sk(sk);
1275         struct net *net = sock_net(sk);
1276         struct ip_options *opt = NULL;
1277         struct rtable *rt = (struct rtable *)cork->dst;
1278         struct iphdr *iph;
1279         __be16 df = 0;
1280         __u8 ttl;
1281
1282         if ((skb = __skb_dequeue(queue)) == NULL)
1283                 goto out;
1284         tail_skb = &(skb_shinfo(skb)->frag_list);
1285
1286         /* move skb->data to ip header from ext header */
1287         if (skb->data < skb_network_header(skb))
1288                 __skb_pull(skb, skb_network_offset(skb));
1289         while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1290                 __skb_pull(tmp_skb, skb_network_header_len(skb));
1291                 *tail_skb = tmp_skb;
1292                 tail_skb = &(tmp_skb->next);
1293                 skb->len += tmp_skb->len;
1294                 skb->data_len += tmp_skb->len;
1295                 skb->truesize += tmp_skb->truesize;
1296                 tmp_skb->destructor = NULL;
1297                 tmp_skb->sk = NULL;
1298         }
1299
1300         /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1301          * to fragment the frame generated here. No matter, what transforms
1302          * how transforms change size of the packet, it will come out.
1303          */
1304         if (inet->pmtudisc < IP_PMTUDISC_DO)
1305                 skb->local_df = 1;
1306
1307         /* DF bit is set when we want to see DF on outgoing frames.
1308          * If local_df is set too, we still allow to fragment this frame
1309          * locally. */
1310         if (inet->pmtudisc >= IP_PMTUDISC_DO ||
1311             (skb->len <= dst_mtu(&rt->dst) &&
1312              ip_dont_fragment(sk, &rt->dst)))
1313                 df = htons(IP_DF);
1314
1315         if (cork->flags & IPCORK_OPT)
1316                 opt = cork->opt;
1317
1318         if (rt->rt_type == RTN_MULTICAST)
1319                 ttl = inet->mc_ttl;
1320         else
1321                 ttl = ip_select_ttl(inet, &rt->dst);
1322
1323         iph = (struct iphdr *)skb->data;
1324         iph->version = 4;
1325         iph->ihl = 5;
1326         iph->tos = inet->tos;
1327         iph->frag_off = df;
1328         iph->ttl = ttl;
1329         iph->protocol = sk->sk_protocol;
1330         ip_copy_addrs(iph, fl4);
1331         ip_select_ident(iph, &rt->dst, sk);
1332
1333         if (opt) {
1334                 iph->ihl += opt->optlen>>2;
1335                 ip_options_build(skb, opt, cork->addr, rt, 0);
1336         }
1337
1338         skb->priority = sk->sk_priority;
1339         skb->mark = sk->sk_mark;
1340         /*
1341          * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1342          * on dst refcount
1343          */
1344         cork->dst = NULL;
1345         skb_dst_set(skb, &rt->dst);
1346
1347         if (iph->protocol == IPPROTO_ICMP)
1348                 icmp_out_count(net, ((struct icmphdr *)
1349                         skb_transport_header(skb))->type);
1350
1351         ip_cork_release(cork);
1352 out:
1353         return skb;
1354 }
1355
1356 int ip_send_skb(struct net *net, struct sk_buff *skb)
1357 {
1358         int err;
1359
1360         err = ip_local_out(skb);
1361         if (err) {
1362                 if (err > 0)
1363                         err = net_xmit_errno(err);
1364                 if (err)
1365                         IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
1366         }
1367
1368         return err;
1369 }
1370
1371 int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
1372 {
1373         struct sk_buff *skb;
1374
1375         skb = ip_finish_skb(sk, fl4);
1376         if (!skb)
1377                 return 0;
1378
1379         /* Netfilter gets whole the not fragmented skb. */
1380         return ip_send_skb(sock_net(sk), skb);
1381 }
1382
1383 /*
1384  *      Throw away all pending data on the socket.
1385  */
1386 static void __ip_flush_pending_frames(struct sock *sk,
1387                                       struct sk_buff_head *queue,
1388                                       struct inet_cork *cork)
1389 {
1390         struct sk_buff *skb;
1391
1392         while ((skb = __skb_dequeue_tail(queue)) != NULL)
1393                 kfree_skb(skb);
1394
1395         ip_cork_release(cork);
1396 }
1397
1398 void ip_flush_pending_frames(struct sock *sk)
1399 {
1400         __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
1401 }
1402
1403 struct sk_buff *ip_make_skb(struct sock *sk,
1404                             struct flowi4 *fl4,
1405                             int getfrag(void *from, char *to, int offset,
1406                                         int len, int odd, struct sk_buff *skb),
1407                             void *from, int length, int transhdrlen,
1408                             struct ipcm_cookie *ipc, struct rtable **rtp,
1409                             unsigned int flags)
1410 {
1411         struct inet_cork cork;
1412         struct sk_buff_head queue;
1413         int err;
1414
1415         if (flags & MSG_PROBE)
1416                 return NULL;
1417
1418         __skb_queue_head_init(&queue);
1419
1420         cork.flags = 0;
1421         cork.addr = 0;
1422         cork.opt = NULL;
1423         err = ip_setup_cork(sk, &cork, ipc, rtp);
1424         if (err)
1425                 return ERR_PTR(err);
1426
1427         err = __ip_append_data(sk, fl4, &queue, &cork,
1428                                &current->task_frag, getfrag,
1429                                from, length, transhdrlen, flags);
1430         if (err) {
1431                 __ip_flush_pending_frames(sk, &queue, &cork);
1432                 return ERR_PTR(err);
1433         }
1434
1435         return __ip_make_skb(sk, fl4, &queue, &cork);
1436 }
1437
1438 /*
1439  *      Fetch data from kernel space and fill in checksum if needed.
1440  */
1441 static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1442                               int len, int odd, struct sk_buff *skb)
1443 {
1444         __wsum csum;
1445
1446         csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1447         skb->csum = csum_block_add(skb->csum, csum, odd);
1448         return 0;
1449 }
1450
1451 /*
1452  *      Generic function to send a packet as reply to another packet.
1453  *      Used to send some TCP resets/acks so far.
1454  *
1455  *      Use a fake percpu inet socket to avoid false sharing and contention.
1456  */
1457 static DEFINE_PER_CPU(struct inet_sock, unicast_sock) = {
1458         .sk = {
1459                 .__sk_common = {
1460                         .skc_refcnt = ATOMIC_INIT(1),
1461                 },
1462                 .sk_wmem_alloc  = ATOMIC_INIT(1),
1463                 .sk_allocation  = GFP_ATOMIC,
1464                 .sk_flags       = (1UL << SOCK_USE_WRITE_QUEUE),
1465         },
1466         .pmtudisc       = IP_PMTUDISC_WANT,
1467         .uc_ttl         = -1,
1468 };
1469
1470 void ip_send_unicast_reply(struct net *net, struct sk_buff *skb, __be32 daddr,
1471                            __be32 saddr, const struct ip_reply_arg *arg,
1472                            unsigned int len)
1473 {
1474         struct ip_options_data replyopts;
1475         struct ipcm_cookie ipc;
1476         struct flowi4 fl4;
1477         struct rtable *rt = skb_rtable(skb);
1478         struct sk_buff *nskb;
1479         struct sock *sk;
1480         struct inet_sock *inet;
1481
1482         if (ip_options_echo(&replyopts.opt.opt, skb))
1483                 return;
1484
1485         ipc.addr = daddr;
1486         ipc.opt = NULL;
1487         ipc.tx_flags = 0;
1488
1489         if (replyopts.opt.opt.optlen) {
1490                 ipc.opt = &replyopts.opt;
1491
1492                 if (replyopts.opt.opt.srr)
1493                         daddr = replyopts.opt.opt.faddr;
1494         }
1495
1496         flowi4_init_output(&fl4, arg->bound_dev_if, 0,
1497                            RT_TOS(arg->tos),
1498                            RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
1499                            ip_reply_arg_flowi_flags(arg),
1500                            daddr, saddr,
1501                            tcp_hdr(skb)->source, tcp_hdr(skb)->dest);
1502         security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
1503         rt = ip_route_output_key(net, &fl4);
1504         if (IS_ERR(rt))
1505                 return;
1506
1507         inet = &get_cpu_var(unicast_sock);
1508
1509         inet->tos = arg->tos;
1510         sk = &inet->sk;
1511         sk->sk_priority = skb->priority;
1512         sk->sk_protocol = ip_hdr(skb)->protocol;
1513         sk->sk_bound_dev_if = arg->bound_dev_if;
1514         sock_net_set(sk, net);
1515         __skb_queue_head_init(&sk->sk_write_queue);
1516         sk->sk_sndbuf = sysctl_wmem_default;
1517         ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base, len, 0,
1518                        &ipc, &rt, MSG_DONTWAIT);
1519         nskb = skb_peek(&sk->sk_write_queue);
1520         if (nskb) {
1521                 if (arg->csumoffset >= 0)
1522                         *((__sum16 *)skb_transport_header(nskb) +
1523                           arg->csumoffset) = csum_fold(csum_add(nskb->csum,
1524                                                                 arg->csum));
1525                 nskb->ip_summed = CHECKSUM_NONE;
1526                 skb_orphan(nskb);
1527                 skb_set_queue_mapping(nskb, skb_get_queue_mapping(skb));
1528                 ip_push_pending_frames(sk, &fl4);
1529         }
1530
1531         put_cpu_var(unicast_sock);
1532
1533         ip_rt_put(rt);
1534 }
1535
1536 void __init ip_init(void)
1537 {
1538         ip_rt_init();
1539         inet_initpeers();
1540
1541 #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1542         igmp_mc_proc_init();
1543 #endif
1544 }