Merge tag 'wireless-drivers-next-for-davem-2015-05-21' of git://git.kernel.org/pub...
[linux-2.6-block.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 static int ip_fragment(struct sock *sk, struct sk_buff *skb,
87                        int (*output)(struct sock *, struct sk_buff *));
88
89 /* Generate a checksum for an outgoing IP datagram. */
90 void ip_send_check(struct iphdr *iph)
91 {
92         iph->check = 0;
93         iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
94 }
95 EXPORT_SYMBOL(ip_send_check);
96
97 static int __ip_local_out_sk(struct sock *sk, struct sk_buff *skb)
98 {
99         struct iphdr *iph = ip_hdr(skb);
100
101         iph->tot_len = htons(skb->len);
102         ip_send_check(iph);
103         return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, sk, skb, NULL,
104                        skb_dst(skb)->dev, dst_output_sk);
105 }
106
107 int __ip_local_out(struct sk_buff *skb)
108 {
109         return __ip_local_out_sk(skb->sk, skb);
110 }
111
112 int ip_local_out_sk(struct sock *sk, struct sk_buff *skb)
113 {
114         int err;
115
116         err = __ip_local_out(skb);
117         if (likely(err == 1))
118                 err = dst_output_sk(sk, skb);
119
120         return err;
121 }
122 EXPORT_SYMBOL_GPL(ip_local_out_sk);
123
124 static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
125 {
126         int ttl = inet->uc_ttl;
127
128         if (ttl < 0)
129                 ttl = ip4_dst_hoplimit(dst);
130         return ttl;
131 }
132
133 /*
134  *              Add an ip header to a skbuff and send it out.
135  *
136  */
137 int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk,
138                           __be32 saddr, __be32 daddr, struct ip_options_rcu *opt)
139 {
140         struct inet_sock *inet = inet_sk(sk);
141         struct rtable *rt = skb_rtable(skb);
142         struct iphdr *iph;
143
144         /* Build the IP header. */
145         skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
146         skb_reset_network_header(skb);
147         iph = ip_hdr(skb);
148         iph->version  = 4;
149         iph->ihl      = 5;
150         iph->tos      = inet->tos;
151         if (ip_dont_fragment(sk, &rt->dst))
152                 iph->frag_off = htons(IP_DF);
153         else
154                 iph->frag_off = 0;
155         iph->ttl      = ip_select_ttl(inet, &rt->dst);
156         iph->daddr    = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
157         iph->saddr    = saddr;
158         iph->protocol = sk->sk_protocol;
159         ip_select_ident(sock_net(sk), skb, sk);
160
161         if (opt && opt->opt.optlen) {
162                 iph->ihl += opt->opt.optlen>>2;
163                 ip_options_build(skb, &opt->opt, daddr, rt, 0);
164         }
165
166         skb->priority = sk->sk_priority;
167         skb->mark = sk->sk_mark;
168
169         /* Send it out. */
170         return ip_local_out(skb);
171 }
172 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
173
174 static inline int ip_finish_output2(struct sock *sk, struct sk_buff *skb)
175 {
176         struct dst_entry *dst = skb_dst(skb);
177         struct rtable *rt = (struct rtable *)dst;
178         struct net_device *dev = dst->dev;
179         unsigned int hh_len = LL_RESERVED_SPACE(dev);
180         struct neighbour *neigh;
181         u32 nexthop;
182
183         if (rt->rt_type == RTN_MULTICAST) {
184                 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTMCAST, skb->len);
185         } else if (rt->rt_type == RTN_BROADCAST)
186                 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTBCAST, skb->len);
187
188         /* Be paranoid, rather than too clever. */
189         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
190                 struct sk_buff *skb2;
191
192                 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
193                 if (!skb2) {
194                         kfree_skb(skb);
195                         return -ENOMEM;
196                 }
197                 if (skb->sk)
198                         skb_set_owner_w(skb2, skb->sk);
199                 consume_skb(skb);
200                 skb = skb2;
201         }
202
203         rcu_read_lock_bh();
204         nexthop = (__force u32) rt_nexthop(rt, ip_hdr(skb)->daddr);
205         neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
206         if (unlikely(!neigh))
207                 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
208         if (!IS_ERR(neigh)) {
209                 int res = dst_neigh_output(dst, neigh, skb);
210
211                 rcu_read_unlock_bh();
212                 return res;
213         }
214         rcu_read_unlock_bh();
215
216         net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
217                             __func__);
218         kfree_skb(skb);
219         return -EINVAL;
220 }
221
222 static int ip_finish_output_gso(struct sock *sk, struct sk_buff *skb)
223 {
224         netdev_features_t features;
225         struct sk_buff *segs;
226         int ret = 0;
227
228         /* common case: locally created skb or seglen is <= mtu */
229         if (((IPCB(skb)->flags & IPSKB_FORWARDED) == 0) ||
230               skb_gso_network_seglen(skb) <= ip_skb_dst_mtu(skb))
231                 return ip_finish_output2(sk, skb);
232
233         /* Slowpath -  GSO segment length is exceeding the dst MTU.
234          *
235          * This can happen in two cases:
236          * 1) TCP GRO packet, DF bit not set
237          * 2) skb arrived via virtio-net, we thus get TSO/GSO skbs directly
238          * from host network stack.
239          */
240         features = netif_skb_features(skb);
241         segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
242         if (IS_ERR_OR_NULL(segs)) {
243                 kfree_skb(skb);
244                 return -ENOMEM;
245         }
246
247         consume_skb(skb);
248
249         do {
250                 struct sk_buff *nskb = segs->next;
251                 int err;
252
253                 segs->next = NULL;
254                 err = ip_fragment(sk, segs, ip_finish_output2);
255
256                 if (err && ret == 0)
257                         ret = err;
258                 segs = nskb;
259         } while (segs);
260
261         return ret;
262 }
263
264 static int ip_finish_output(struct sock *sk, struct sk_buff *skb)
265 {
266 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
267         /* Policy lookup after SNAT yielded a new policy */
268         if (skb_dst(skb)->xfrm) {
269                 IPCB(skb)->flags |= IPSKB_REROUTED;
270                 return dst_output_sk(sk, skb);
271         }
272 #endif
273         if (skb_is_gso(skb))
274                 return ip_finish_output_gso(sk, skb);
275
276         if (skb->len > ip_skb_dst_mtu(skb))
277                 return ip_fragment(sk, skb, ip_finish_output2);
278
279         return ip_finish_output2(sk, skb);
280 }
281
282 int ip_mc_output(struct sock *sk, struct sk_buff *skb)
283 {
284         struct rtable *rt = skb_rtable(skb);
285         struct net_device *dev = rt->dst.dev;
286
287         /*
288          *      If the indicated interface is up and running, send the packet.
289          */
290         IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
291
292         skb->dev = dev;
293         skb->protocol = htons(ETH_P_IP);
294
295         /*
296          *      Multicasts are looped back for other local users
297          */
298
299         if (rt->rt_flags&RTCF_MULTICAST) {
300                 if (sk_mc_loop(sk)
301 #ifdef CONFIG_IP_MROUTE
302                 /* Small optimization: do not loopback not local frames,
303                    which returned after forwarding; they will be  dropped
304                    by ip_mr_input in any case.
305                    Note, that local frames are looped back to be delivered
306                    to local recipients.
307
308                    This check is duplicated in ip_mr_input at the moment.
309                  */
310                     &&
311                     ((rt->rt_flags & RTCF_LOCAL) ||
312                      !(IPCB(skb)->flags & IPSKB_FORWARDED))
313 #endif
314                    ) {
315                         struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
316                         if (newskb)
317                                 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
318                                         sk, newskb, NULL, newskb->dev,
319                                         dev_loopback_xmit);
320                 }
321
322                 /* Multicasts with ttl 0 must not go beyond the host */
323
324                 if (ip_hdr(skb)->ttl == 0) {
325                         kfree_skb(skb);
326                         return 0;
327                 }
328         }
329
330         if (rt->rt_flags&RTCF_BROADCAST) {
331                 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
332                 if (newskb)
333                         NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, sk, newskb,
334                                 NULL, newskb->dev, dev_loopback_xmit);
335         }
336
337         return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, sk, skb, NULL,
338                             skb->dev, ip_finish_output,
339                             !(IPCB(skb)->flags & IPSKB_REROUTED));
340 }
341
342 int ip_output(struct sock *sk, struct sk_buff *skb)
343 {
344         struct net_device *dev = skb_dst(skb)->dev;
345
346         IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
347
348         skb->dev = dev;
349         skb->protocol = htons(ETH_P_IP);
350
351         return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, sk, skb,
352                             NULL, dev,
353                             ip_finish_output,
354                             !(IPCB(skb)->flags & IPSKB_REROUTED));
355 }
356
357 /*
358  * copy saddr and daddr, possibly using 64bit load/stores
359  * Equivalent to :
360  *   iph->saddr = fl4->saddr;
361  *   iph->daddr = fl4->daddr;
362  */
363 static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
364 {
365         BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
366                      offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
367         memcpy(&iph->saddr, &fl4->saddr,
368                sizeof(fl4->saddr) + sizeof(fl4->daddr));
369 }
370
371 /* Note: skb->sk can be different from sk, in case of tunnels */
372 int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl)
373 {
374         struct inet_sock *inet = inet_sk(sk);
375         struct ip_options_rcu *inet_opt;
376         struct flowi4 *fl4;
377         struct rtable *rt;
378         struct iphdr *iph;
379         int res;
380
381         /* Skip all of this if the packet is already routed,
382          * f.e. by something like SCTP.
383          */
384         rcu_read_lock();
385         inet_opt = rcu_dereference(inet->inet_opt);
386         fl4 = &fl->u.ip4;
387         rt = skb_rtable(skb);
388         if (rt)
389                 goto packet_routed;
390
391         /* Make sure we can route this packet. */
392         rt = (struct rtable *)__sk_dst_check(sk, 0);
393         if (!rt) {
394                 __be32 daddr;
395
396                 /* Use correct destination address if we have options. */
397                 daddr = inet->inet_daddr;
398                 if (inet_opt && inet_opt->opt.srr)
399                         daddr = inet_opt->opt.faddr;
400
401                 /* If this fails, retransmit mechanism of transport layer will
402                  * keep trying until route appears or the connection times
403                  * itself out.
404                  */
405                 rt = ip_route_output_ports(sock_net(sk), fl4, sk,
406                                            daddr, inet->inet_saddr,
407                                            inet->inet_dport,
408                                            inet->inet_sport,
409                                            sk->sk_protocol,
410                                            RT_CONN_FLAGS(sk),
411                                            sk->sk_bound_dev_if);
412                 if (IS_ERR(rt))
413                         goto no_route;
414                 sk_setup_caps(sk, &rt->dst);
415         }
416         skb_dst_set_noref(skb, &rt->dst);
417
418 packet_routed:
419         if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
420                 goto no_route;
421
422         /* OK, we know where to send it, allocate and build IP header. */
423         skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
424         skb_reset_network_header(skb);
425         iph = ip_hdr(skb);
426         *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
427         if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df)
428                 iph->frag_off = htons(IP_DF);
429         else
430                 iph->frag_off = 0;
431         iph->ttl      = ip_select_ttl(inet, &rt->dst);
432         iph->protocol = sk->sk_protocol;
433         ip_copy_addrs(iph, fl4);
434
435         /* Transport layer set skb->h.foo itself. */
436
437         if (inet_opt && inet_opt->opt.optlen) {
438                 iph->ihl += inet_opt->opt.optlen >> 2;
439                 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0);
440         }
441
442         ip_select_ident_segs(sock_net(sk), skb, sk,
443                              skb_shinfo(skb)->gso_segs ?: 1);
444
445         /* TODO : should we use skb->sk here instead of sk ? */
446         skb->priority = sk->sk_priority;
447         skb->mark = sk->sk_mark;
448
449         res = ip_local_out(skb);
450         rcu_read_unlock();
451         return res;
452
453 no_route:
454         rcu_read_unlock();
455         IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
456         kfree_skb(skb);
457         return -EHOSTUNREACH;
458 }
459 EXPORT_SYMBOL(ip_queue_xmit);
460
461 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
462 {
463         to->pkt_type = from->pkt_type;
464         to->priority = from->priority;
465         to->protocol = from->protocol;
466         skb_dst_drop(to);
467         skb_dst_copy(to, from);
468         to->dev = from->dev;
469         to->mark = from->mark;
470
471         /* Copy the flags to each fragment. */
472         IPCB(to)->flags = IPCB(from)->flags;
473
474 #ifdef CONFIG_NET_SCHED
475         to->tc_index = from->tc_index;
476 #endif
477         nf_copy(to, from);
478 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
479         to->ipvs_property = from->ipvs_property;
480 #endif
481         skb_copy_secmark(to, from);
482 }
483
484 static int ip_fragment(struct sock *sk, struct sk_buff *skb,
485                        int (*output)(struct sock *, struct sk_buff *))
486 {
487         struct iphdr *iph = ip_hdr(skb);
488         unsigned int mtu = ip_skb_dst_mtu(skb);
489
490         if (unlikely(((iph->frag_off & htons(IP_DF)) && !skb->ignore_df) ||
491                      (IPCB(skb)->frag_max_size &&
492                       IPCB(skb)->frag_max_size > mtu))) {
493                 struct rtable *rt = skb_rtable(skb);
494                 struct net_device *dev = rt->dst.dev;
495
496                 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
497                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
498                           htonl(mtu));
499                 kfree_skb(skb);
500                 return -EMSGSIZE;
501         }
502
503         return ip_do_fragment(sk, skb, output);
504 }
505
506 /*
507  *      This IP datagram is too large to be sent in one piece.  Break it up into
508  *      smaller pieces (each of size equal to IP header plus
509  *      a block of the data of the original IP data part) that will yet fit in a
510  *      single device frame, and queue such a frame for sending.
511  */
512
513 int ip_do_fragment(struct sock *sk, struct sk_buff *skb,
514                    int (*output)(struct sock *, struct sk_buff *))
515 {
516         struct iphdr *iph;
517         int ptr;
518         struct net_device *dev;
519         struct sk_buff *skb2;
520         unsigned int mtu, hlen, left, len, ll_rs;
521         int offset;
522         __be16 not_last_frag;
523         struct rtable *rt = skb_rtable(skb);
524         int err = 0;
525
526         dev = rt->dst.dev;
527
528         /*
529          *      Point into the IP datagram header.
530          */
531
532         iph = ip_hdr(skb);
533
534         mtu = ip_skb_dst_mtu(skb);
535
536         /*
537          *      Setup starting values.
538          */
539
540         hlen = iph->ihl * 4;
541         mtu = mtu - hlen;       /* Size of data space */
542 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
543         if (skb->nf_bridge)
544                 mtu -= nf_bridge_mtu_reduction(skb);
545 #endif
546         IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
547
548         /* When frag_list is given, use it. First, check its validity:
549          * some transformers could create wrong frag_list or break existing
550          * one, it is not prohibited. In this case fall back to copying.
551          *
552          * LATER: this step can be merged to real generation of fragments,
553          * we can switch to copy when see the first bad fragment.
554          */
555         if (skb_has_frag_list(skb)) {
556                 struct sk_buff *frag, *frag2;
557                 int first_len = skb_pagelen(skb);
558
559                 if (first_len - hlen > mtu ||
560                     ((first_len - hlen) & 7) ||
561                     ip_is_fragment(iph) ||
562                     skb_cloned(skb))
563                         goto slow_path;
564
565                 skb_walk_frags(skb, frag) {
566                         /* Correct geometry. */
567                         if (frag->len > mtu ||
568                             ((frag->len & 7) && frag->next) ||
569                             skb_headroom(frag) < hlen)
570                                 goto slow_path_clean;
571
572                         /* Partially cloned skb? */
573                         if (skb_shared(frag))
574                                 goto slow_path_clean;
575
576                         BUG_ON(frag->sk);
577                         if (skb->sk) {
578                                 frag->sk = skb->sk;
579                                 frag->destructor = sock_wfree;
580                         }
581                         skb->truesize -= frag->truesize;
582                 }
583
584                 /* Everything is OK. Generate! */
585
586                 err = 0;
587                 offset = 0;
588                 frag = skb_shinfo(skb)->frag_list;
589                 skb_frag_list_init(skb);
590                 skb->data_len = first_len - skb_headlen(skb);
591                 skb->len = first_len;
592                 iph->tot_len = htons(first_len);
593                 iph->frag_off = htons(IP_MF);
594                 ip_send_check(iph);
595
596                 for (;;) {
597                         /* Prepare header of the next frame,
598                          * before previous one went down. */
599                         if (frag) {
600                                 frag->ip_summed = CHECKSUM_NONE;
601                                 skb_reset_transport_header(frag);
602                                 __skb_push(frag, hlen);
603                                 skb_reset_network_header(frag);
604                                 memcpy(skb_network_header(frag), iph, hlen);
605                                 iph = ip_hdr(frag);
606                                 iph->tot_len = htons(frag->len);
607                                 ip_copy_metadata(frag, skb);
608                                 if (offset == 0)
609                                         ip_options_fragment(frag);
610                                 offset += skb->len - hlen;
611                                 iph->frag_off = htons(offset>>3);
612                                 if (frag->next)
613                                         iph->frag_off |= htons(IP_MF);
614                                 /* Ready, complete checksum */
615                                 ip_send_check(iph);
616                         }
617
618                         err = output(sk, skb);
619
620                         if (!err)
621                                 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
622                         if (err || !frag)
623                                 break;
624
625                         skb = frag;
626                         frag = skb->next;
627                         skb->next = NULL;
628                 }
629
630                 if (err == 0) {
631                         IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
632                         return 0;
633                 }
634
635                 while (frag) {
636                         skb = frag->next;
637                         kfree_skb(frag);
638                         frag = skb;
639                 }
640                 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
641                 return err;
642
643 slow_path_clean:
644                 skb_walk_frags(skb, frag2) {
645                         if (frag2 == frag)
646                                 break;
647                         frag2->sk = NULL;
648                         frag2->destructor = NULL;
649                         skb->truesize += frag2->truesize;
650                 }
651         }
652
653 slow_path:
654         /* for offloaded checksums cleanup checksum before fragmentation */
655         if ((skb->ip_summed == CHECKSUM_PARTIAL) && skb_checksum_help(skb))
656                 goto fail;
657         iph = ip_hdr(skb);
658
659         left = skb->len - hlen;         /* Space per frame */
660         ptr = hlen;             /* Where to start from */
661
662         ll_rs = LL_RESERVED_SPACE(rt->dst.dev);
663
664         /*
665          *      Fragment the datagram.
666          */
667
668         offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
669         not_last_frag = iph->frag_off & htons(IP_MF);
670
671         /*
672          *      Keep copying data until we run out.
673          */
674
675         while (left > 0) {
676                 len = left;
677                 /* IF: it doesn't fit, use 'mtu' - the data space left */
678                 if (len > mtu)
679                         len = mtu;
680                 /* IF: we are not sending up to and including the packet end
681                    then align the next start on an eight byte boundary */
682                 if (len < left) {
683                         len &= ~7;
684                 }
685
686                 /* Allocate buffer */
687                 skb2 = alloc_skb(len + hlen + ll_rs, GFP_ATOMIC);
688                 if (!skb2) {
689                         err = -ENOMEM;
690                         goto fail;
691                 }
692
693                 /*
694                  *      Set up data on packet
695                  */
696
697                 ip_copy_metadata(skb2, skb);
698                 skb_reserve(skb2, ll_rs);
699                 skb_put(skb2, len + hlen);
700                 skb_reset_network_header(skb2);
701                 skb2->transport_header = skb2->network_header + hlen;
702
703                 /*
704                  *      Charge the memory for the fragment to any owner
705                  *      it might possess
706                  */
707
708                 if (skb->sk)
709                         skb_set_owner_w(skb2, skb->sk);
710
711                 /*
712                  *      Copy the packet header into the new buffer.
713                  */
714
715                 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
716
717                 /*
718                  *      Copy a block of the IP datagram.
719                  */
720                 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
721                         BUG();
722                 left -= len;
723
724                 /*
725                  *      Fill in the new header fields.
726                  */
727                 iph = ip_hdr(skb2);
728                 iph->frag_off = htons((offset >> 3));
729
730                 /* ANK: dirty, but effective trick. Upgrade options only if
731                  * the segment to be fragmented was THE FIRST (otherwise,
732                  * options are already fixed) and make it ONCE
733                  * on the initial skb, so that all the following fragments
734                  * will inherit fixed options.
735                  */
736                 if (offset == 0)
737                         ip_options_fragment(skb);
738
739                 /*
740                  *      Added AC : If we are fragmenting a fragment that's not the
741                  *                 last fragment then keep MF on each bit
742                  */
743                 if (left > 0 || not_last_frag)
744                         iph->frag_off |= htons(IP_MF);
745                 ptr += len;
746                 offset += len;
747
748                 /*
749                  *      Put this fragment into the sending queue.
750                  */
751                 iph->tot_len = htons(len + hlen);
752
753                 ip_send_check(iph);
754
755                 err = output(sk, skb2);
756                 if (err)
757                         goto fail;
758
759                 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
760         }
761         consume_skb(skb);
762         IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
763         return err;
764
765 fail:
766         kfree_skb(skb);
767         IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
768         return err;
769 }
770 EXPORT_SYMBOL(ip_do_fragment);
771
772 int
773 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
774 {
775         struct msghdr *msg = from;
776
777         if (skb->ip_summed == CHECKSUM_PARTIAL) {
778                 if (copy_from_iter(to, len, &msg->msg_iter) != len)
779                         return -EFAULT;
780         } else {
781                 __wsum csum = 0;
782                 if (csum_and_copy_from_iter(to, len, &csum, &msg->msg_iter) != len)
783                         return -EFAULT;
784                 skb->csum = csum_block_add(skb->csum, csum, odd);
785         }
786         return 0;
787 }
788 EXPORT_SYMBOL(ip_generic_getfrag);
789
790 static inline __wsum
791 csum_page(struct page *page, int offset, int copy)
792 {
793         char *kaddr;
794         __wsum csum;
795         kaddr = kmap(page);
796         csum = csum_partial(kaddr + offset, copy, 0);
797         kunmap(page);
798         return csum;
799 }
800
801 static inline int ip_ufo_append_data(struct sock *sk,
802                         struct sk_buff_head *queue,
803                         int getfrag(void *from, char *to, int offset, int len,
804                                int odd, struct sk_buff *skb),
805                         void *from, int length, int hh_len, int fragheaderlen,
806                         int transhdrlen, int maxfraglen, unsigned int flags)
807 {
808         struct sk_buff *skb;
809         int err;
810
811         /* There is support for UDP fragmentation offload by network
812          * device, so create one single skb packet containing complete
813          * udp datagram
814          */
815         skb = skb_peek_tail(queue);
816         if (!skb) {
817                 skb = sock_alloc_send_skb(sk,
818                         hh_len + fragheaderlen + transhdrlen + 20,
819                         (flags & MSG_DONTWAIT), &err);
820
821                 if (!skb)
822                         return err;
823
824                 /* reserve space for Hardware header */
825                 skb_reserve(skb, hh_len);
826
827                 /* create space for UDP/IP header */
828                 skb_put(skb, fragheaderlen + transhdrlen);
829
830                 /* initialize network header pointer */
831                 skb_reset_network_header(skb);
832
833                 /* initialize protocol header pointer */
834                 skb->transport_header = skb->network_header + fragheaderlen;
835
836                 skb->csum = 0;
837
838                 __skb_queue_tail(queue, skb);
839         } else if (skb_is_gso(skb)) {
840                 goto append;
841         }
842
843         skb->ip_summed = CHECKSUM_PARTIAL;
844         /* specify the length of each IP datagram fragment */
845         skb_shinfo(skb)->gso_size = maxfraglen - fragheaderlen;
846         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
847
848 append:
849         return skb_append_datato_frags(sk, skb, getfrag, from,
850                                        (length - transhdrlen));
851 }
852
853 static int __ip_append_data(struct sock *sk,
854                             struct flowi4 *fl4,
855                             struct sk_buff_head *queue,
856                             struct inet_cork *cork,
857                             struct page_frag *pfrag,
858                             int getfrag(void *from, char *to, int offset,
859                                         int len, int odd, struct sk_buff *skb),
860                             void *from, int length, int transhdrlen,
861                             unsigned int flags)
862 {
863         struct inet_sock *inet = inet_sk(sk);
864         struct sk_buff *skb;
865
866         struct ip_options *opt = cork->opt;
867         int hh_len;
868         int exthdrlen;
869         int mtu;
870         int copy;
871         int err;
872         int offset = 0;
873         unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
874         int csummode = CHECKSUM_NONE;
875         struct rtable *rt = (struct rtable *)cork->dst;
876         u32 tskey = 0;
877
878         skb = skb_peek_tail(queue);
879
880         exthdrlen = !skb ? rt->dst.header_len : 0;
881         mtu = cork->fragsize;
882         if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
883             sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
884                 tskey = sk->sk_tskey++;
885
886         hh_len = LL_RESERVED_SPACE(rt->dst.dev);
887
888         fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
889         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
890         maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
891
892         if (cork->length + length > maxnonfragsize - fragheaderlen) {
893                 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
894                                mtu - (opt ? opt->optlen : 0));
895                 return -EMSGSIZE;
896         }
897
898         /*
899          * transhdrlen > 0 means that this is the first fragment and we wish
900          * it won't be fragmented in the future.
901          */
902         if (transhdrlen &&
903             length + fragheaderlen <= mtu &&
904             rt->dst.dev->features & NETIF_F_V4_CSUM &&
905             !exthdrlen)
906                 csummode = CHECKSUM_PARTIAL;
907
908         cork->length += length;
909         if (((length > mtu) || (skb && skb_is_gso(skb))) &&
910             (sk->sk_protocol == IPPROTO_UDP) &&
911             (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len &&
912             (sk->sk_type == SOCK_DGRAM)) {
913                 err = ip_ufo_append_data(sk, queue, getfrag, from, length,
914                                          hh_len, fragheaderlen, transhdrlen,
915                                          maxfraglen, flags);
916                 if (err)
917                         goto error;
918                 return 0;
919         }
920
921         /* So, what's going on in the loop below?
922          *
923          * We use calculated fragment length to generate chained skb,
924          * each of segments is IP fragment ready for sending to network after
925          * adding appropriate IP header.
926          */
927
928         if (!skb)
929                 goto alloc_new_skb;
930
931         while (length > 0) {
932                 /* Check if the remaining data fits into current packet. */
933                 copy = mtu - skb->len;
934                 if (copy < length)
935                         copy = maxfraglen - skb->len;
936                 if (copy <= 0) {
937                         char *data;
938                         unsigned int datalen;
939                         unsigned int fraglen;
940                         unsigned int fraggap;
941                         unsigned int alloclen;
942                         struct sk_buff *skb_prev;
943 alloc_new_skb:
944                         skb_prev = skb;
945                         if (skb_prev)
946                                 fraggap = skb_prev->len - maxfraglen;
947                         else
948                                 fraggap = 0;
949
950                         /*
951                          * If remaining data exceeds the mtu,
952                          * we know we need more fragment(s).
953                          */
954                         datalen = length + fraggap;
955                         if (datalen > mtu - fragheaderlen)
956                                 datalen = maxfraglen - fragheaderlen;
957                         fraglen = datalen + fragheaderlen;
958
959                         if ((flags & MSG_MORE) &&
960                             !(rt->dst.dev->features&NETIF_F_SG))
961                                 alloclen = mtu;
962                         else
963                                 alloclen = fraglen;
964
965                         alloclen += exthdrlen;
966
967                         /* The last fragment gets additional space at tail.
968                          * Note, with MSG_MORE we overallocate on fragments,
969                          * because we have no idea what fragment will be
970                          * the last.
971                          */
972                         if (datalen == length + fraggap)
973                                 alloclen += rt->dst.trailer_len;
974
975                         if (transhdrlen) {
976                                 skb = sock_alloc_send_skb(sk,
977                                                 alloclen + hh_len + 15,
978                                                 (flags & MSG_DONTWAIT), &err);
979                         } else {
980                                 skb = NULL;
981                                 if (atomic_read(&sk->sk_wmem_alloc) <=
982                                     2 * sk->sk_sndbuf)
983                                         skb = sock_wmalloc(sk,
984                                                            alloclen + hh_len + 15, 1,
985                                                            sk->sk_allocation);
986                                 if (unlikely(!skb))
987                                         err = -ENOBUFS;
988                         }
989                         if (!skb)
990                                 goto error;
991
992                         /*
993                          *      Fill in the control structures
994                          */
995                         skb->ip_summed = csummode;
996                         skb->csum = 0;
997                         skb_reserve(skb, hh_len);
998
999                         /* only the initial fragment is time stamped */
1000                         skb_shinfo(skb)->tx_flags = cork->tx_flags;
1001                         cork->tx_flags = 0;
1002                         skb_shinfo(skb)->tskey = tskey;
1003                         tskey = 0;
1004
1005                         /*
1006                          *      Find where to start putting bytes.
1007                          */
1008                         data = skb_put(skb, fraglen + exthdrlen);
1009                         skb_set_network_header(skb, exthdrlen);
1010                         skb->transport_header = (skb->network_header +
1011                                                  fragheaderlen);
1012                         data += fragheaderlen + exthdrlen;
1013
1014                         if (fraggap) {
1015                                 skb->csum = skb_copy_and_csum_bits(
1016                                         skb_prev, maxfraglen,
1017                                         data + transhdrlen, fraggap, 0);
1018                                 skb_prev->csum = csum_sub(skb_prev->csum,
1019                                                           skb->csum);
1020                                 data += fraggap;
1021                                 pskb_trim_unique(skb_prev, maxfraglen);
1022                         }
1023
1024                         copy = datalen - transhdrlen - fraggap;
1025                         if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1026                                 err = -EFAULT;
1027                                 kfree_skb(skb);
1028                                 goto error;
1029                         }
1030
1031                         offset += copy;
1032                         length -= datalen - fraggap;
1033                         transhdrlen = 0;
1034                         exthdrlen = 0;
1035                         csummode = CHECKSUM_NONE;
1036
1037                         /*
1038                          * Put the packet on the pending queue.
1039                          */
1040                         __skb_queue_tail(queue, skb);
1041                         continue;
1042                 }
1043
1044                 if (copy > length)
1045                         copy = length;
1046
1047                 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1048                         unsigned int off;
1049
1050                         off = skb->len;
1051                         if (getfrag(from, skb_put(skb, copy),
1052                                         offset, copy, off, skb) < 0) {
1053                                 __skb_trim(skb, off);
1054                                 err = -EFAULT;
1055                                 goto error;
1056                         }
1057                 } else {
1058                         int i = skb_shinfo(skb)->nr_frags;
1059
1060                         err = -ENOMEM;
1061                         if (!sk_page_frag_refill(sk, pfrag))
1062                                 goto error;
1063
1064                         if (!skb_can_coalesce(skb, i, pfrag->page,
1065                                               pfrag->offset)) {
1066                                 err = -EMSGSIZE;
1067                                 if (i == MAX_SKB_FRAGS)
1068                                         goto error;
1069
1070                                 __skb_fill_page_desc(skb, i, pfrag->page,
1071                                                      pfrag->offset, 0);
1072                                 skb_shinfo(skb)->nr_frags = ++i;
1073                                 get_page(pfrag->page);
1074                         }
1075                         copy = min_t(int, copy, pfrag->size - pfrag->offset);
1076                         if (getfrag(from,
1077                                     page_address(pfrag->page) + pfrag->offset,
1078                                     offset, copy, skb->len, skb) < 0)
1079                                 goto error_efault;
1080
1081                         pfrag->offset += copy;
1082                         skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1083                         skb->len += copy;
1084                         skb->data_len += copy;
1085                         skb->truesize += copy;
1086                         atomic_add(copy, &sk->sk_wmem_alloc);
1087                 }
1088                 offset += copy;
1089                 length -= copy;
1090         }
1091
1092         return 0;
1093
1094 error_efault:
1095         err = -EFAULT;
1096 error:
1097         cork->length -= length;
1098         IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1099         return err;
1100 }
1101
1102 static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
1103                          struct ipcm_cookie *ipc, struct rtable **rtp)
1104 {
1105         struct ip_options_rcu *opt;
1106         struct rtable *rt;
1107
1108         /*
1109          * setup for corking.
1110          */
1111         opt = ipc->opt;
1112         if (opt) {
1113                 if (!cork->opt) {
1114                         cork->opt = kmalloc(sizeof(struct ip_options) + 40,
1115                                             sk->sk_allocation);
1116                         if (unlikely(!cork->opt))
1117                                 return -ENOBUFS;
1118                 }
1119                 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
1120                 cork->flags |= IPCORK_OPT;
1121                 cork->addr = ipc->addr;
1122         }
1123         rt = *rtp;
1124         if (unlikely(!rt))
1125                 return -EFAULT;
1126         /*
1127          * We steal reference to this route, caller should not release it
1128          */
1129         *rtp = NULL;
1130         cork->fragsize = ip_sk_use_pmtu(sk) ?
1131                          dst_mtu(&rt->dst) : rt->dst.dev->mtu;
1132         cork->dst = &rt->dst;
1133         cork->length = 0;
1134         cork->ttl = ipc->ttl;
1135         cork->tos = ipc->tos;
1136         cork->priority = ipc->priority;
1137         cork->tx_flags = ipc->tx_flags;
1138
1139         return 0;
1140 }
1141
1142 /*
1143  *      ip_append_data() and ip_append_page() can make one large IP datagram
1144  *      from many pieces of data. Each pieces will be holded on the socket
1145  *      until ip_push_pending_frames() is called. Each piece can be a page
1146  *      or non-page data.
1147  *
1148  *      Not only UDP, other transport protocols - e.g. raw sockets - can use
1149  *      this interface potentially.
1150  *
1151  *      LATER: length must be adjusted by pad at tail, when it is required.
1152  */
1153 int ip_append_data(struct sock *sk, struct flowi4 *fl4,
1154                    int getfrag(void *from, char *to, int offset, int len,
1155                                int odd, struct sk_buff *skb),
1156                    void *from, int length, int transhdrlen,
1157                    struct ipcm_cookie *ipc, struct rtable **rtp,
1158                    unsigned int flags)
1159 {
1160         struct inet_sock *inet = inet_sk(sk);
1161         int err;
1162
1163         if (flags&MSG_PROBE)
1164                 return 0;
1165
1166         if (skb_queue_empty(&sk->sk_write_queue)) {
1167                 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
1168                 if (err)
1169                         return err;
1170         } else {
1171                 transhdrlen = 0;
1172         }
1173
1174         return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
1175                                 sk_page_frag(sk), getfrag,
1176                                 from, length, transhdrlen, flags);
1177 }
1178
1179 ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
1180                        int offset, size_t size, int flags)
1181 {
1182         struct inet_sock *inet = inet_sk(sk);
1183         struct sk_buff *skb;
1184         struct rtable *rt;
1185         struct ip_options *opt = NULL;
1186         struct inet_cork *cork;
1187         int hh_len;
1188         int mtu;
1189         int len;
1190         int err;
1191         unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize;
1192
1193         if (inet->hdrincl)
1194                 return -EPERM;
1195
1196         if (flags&MSG_PROBE)
1197                 return 0;
1198
1199         if (skb_queue_empty(&sk->sk_write_queue))
1200                 return -EINVAL;
1201
1202         cork = &inet->cork.base;
1203         rt = (struct rtable *)cork->dst;
1204         if (cork->flags & IPCORK_OPT)
1205                 opt = cork->opt;
1206
1207         if (!(rt->dst.dev->features&NETIF_F_SG))
1208                 return -EOPNOTSUPP;
1209
1210         hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1211         mtu = cork->fragsize;
1212
1213         fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1214         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1215         maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
1216
1217         if (cork->length + size > maxnonfragsize - fragheaderlen) {
1218                 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
1219                                mtu - (opt ? opt->optlen : 0));
1220                 return -EMSGSIZE;
1221         }
1222
1223         skb = skb_peek_tail(&sk->sk_write_queue);
1224         if (!skb)
1225                 return -EINVAL;
1226
1227         cork->length += size;
1228         if ((size + skb->len > mtu) &&
1229             (sk->sk_protocol == IPPROTO_UDP) &&
1230             (rt->dst.dev->features & NETIF_F_UFO)) {
1231                 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
1232                 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1233         }
1234
1235         while (size > 0) {
1236                 int i;
1237
1238                 if (skb_is_gso(skb))
1239                         len = size;
1240                 else {
1241
1242                         /* Check if the remaining data fits into current packet. */
1243                         len = mtu - skb->len;
1244                         if (len < size)
1245                                 len = maxfraglen - skb->len;
1246                 }
1247                 if (len <= 0) {
1248                         struct sk_buff *skb_prev;
1249                         int alloclen;
1250
1251                         skb_prev = skb;
1252                         fraggap = skb_prev->len - maxfraglen;
1253
1254                         alloclen = fragheaderlen + hh_len + fraggap + 15;
1255                         skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1256                         if (unlikely(!skb)) {
1257                                 err = -ENOBUFS;
1258                                 goto error;
1259                         }
1260
1261                         /*
1262                          *      Fill in the control structures
1263                          */
1264                         skb->ip_summed = CHECKSUM_NONE;
1265                         skb->csum = 0;
1266                         skb_reserve(skb, hh_len);
1267
1268                         /*
1269                          *      Find where to start putting bytes.
1270                          */
1271                         skb_put(skb, fragheaderlen + fraggap);
1272                         skb_reset_network_header(skb);
1273                         skb->transport_header = (skb->network_header +
1274                                                  fragheaderlen);
1275                         if (fraggap) {
1276                                 skb->csum = skb_copy_and_csum_bits(skb_prev,
1277                                                                    maxfraglen,
1278                                                     skb_transport_header(skb),
1279                                                                    fraggap, 0);
1280                                 skb_prev->csum = csum_sub(skb_prev->csum,
1281                                                           skb->csum);
1282                                 pskb_trim_unique(skb_prev, maxfraglen);
1283                         }
1284
1285                         /*
1286                          * Put the packet on the pending queue.
1287                          */
1288                         __skb_queue_tail(&sk->sk_write_queue, skb);
1289                         continue;
1290                 }
1291
1292                 i = skb_shinfo(skb)->nr_frags;
1293                 if (len > size)
1294                         len = size;
1295                 if (skb_can_coalesce(skb, i, page, offset)) {
1296                         skb_frag_size_add(&skb_shinfo(skb)->frags[i-1], len);
1297                 } else if (i < MAX_SKB_FRAGS) {
1298                         get_page(page);
1299                         skb_fill_page_desc(skb, i, page, offset, len);
1300                 } else {
1301                         err = -EMSGSIZE;
1302                         goto error;
1303                 }
1304
1305                 if (skb->ip_summed == CHECKSUM_NONE) {
1306                         __wsum csum;
1307                         csum = csum_page(page, offset, len);
1308                         skb->csum = csum_block_add(skb->csum, csum, skb->len);
1309                 }
1310
1311                 skb->len += len;
1312                 skb->data_len += len;
1313                 skb->truesize += len;
1314                 atomic_add(len, &sk->sk_wmem_alloc);
1315                 offset += len;
1316                 size -= len;
1317         }
1318         return 0;
1319
1320 error:
1321         cork->length -= size;
1322         IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1323         return err;
1324 }
1325
1326 static void ip_cork_release(struct inet_cork *cork)
1327 {
1328         cork->flags &= ~IPCORK_OPT;
1329         kfree(cork->opt);
1330         cork->opt = NULL;
1331         dst_release(cork->dst);
1332         cork->dst = NULL;
1333 }
1334
1335 /*
1336  *      Combined all pending IP fragments on the socket as one IP datagram
1337  *      and push them out.
1338  */
1339 struct sk_buff *__ip_make_skb(struct sock *sk,
1340                               struct flowi4 *fl4,
1341                               struct sk_buff_head *queue,
1342                               struct inet_cork *cork)
1343 {
1344         struct sk_buff *skb, *tmp_skb;
1345         struct sk_buff **tail_skb;
1346         struct inet_sock *inet = inet_sk(sk);
1347         struct net *net = sock_net(sk);
1348         struct ip_options *opt = NULL;
1349         struct rtable *rt = (struct rtable *)cork->dst;
1350         struct iphdr *iph;
1351         __be16 df = 0;
1352         __u8 ttl;
1353
1354         skb = __skb_dequeue(queue);
1355         if (!skb)
1356                 goto out;
1357         tail_skb = &(skb_shinfo(skb)->frag_list);
1358
1359         /* move skb->data to ip header from ext header */
1360         if (skb->data < skb_network_header(skb))
1361                 __skb_pull(skb, skb_network_offset(skb));
1362         while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1363                 __skb_pull(tmp_skb, skb_network_header_len(skb));
1364                 *tail_skb = tmp_skb;
1365                 tail_skb = &(tmp_skb->next);
1366                 skb->len += tmp_skb->len;
1367                 skb->data_len += tmp_skb->len;
1368                 skb->truesize += tmp_skb->truesize;
1369                 tmp_skb->destructor = NULL;
1370                 tmp_skb->sk = NULL;
1371         }
1372
1373         /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1374          * to fragment the frame generated here. No matter, what transforms
1375          * how transforms change size of the packet, it will come out.
1376          */
1377         skb->ignore_df = ip_sk_ignore_df(sk);
1378
1379         /* DF bit is set when we want to see DF on outgoing frames.
1380          * If ignore_df is set too, we still allow to fragment this frame
1381          * locally. */
1382         if (inet->pmtudisc == IP_PMTUDISC_DO ||
1383             inet->pmtudisc == IP_PMTUDISC_PROBE ||
1384             (skb->len <= dst_mtu(&rt->dst) &&
1385              ip_dont_fragment(sk, &rt->dst)))
1386                 df = htons(IP_DF);
1387
1388         if (cork->flags & IPCORK_OPT)
1389                 opt = cork->opt;
1390
1391         if (cork->ttl != 0)
1392                 ttl = cork->ttl;
1393         else if (rt->rt_type == RTN_MULTICAST)
1394                 ttl = inet->mc_ttl;
1395         else
1396                 ttl = ip_select_ttl(inet, &rt->dst);
1397
1398         iph = ip_hdr(skb);
1399         iph->version = 4;
1400         iph->ihl = 5;
1401         iph->tos = (cork->tos != -1) ? cork->tos : inet->tos;
1402         iph->frag_off = df;
1403         iph->ttl = ttl;
1404         iph->protocol = sk->sk_protocol;
1405         ip_copy_addrs(iph, fl4);
1406         ip_select_ident(net, skb, sk);
1407
1408         if (opt) {
1409                 iph->ihl += opt->optlen>>2;
1410                 ip_options_build(skb, opt, cork->addr, rt, 0);
1411         }
1412
1413         skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority;
1414         skb->mark = sk->sk_mark;
1415         /*
1416          * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1417          * on dst refcount
1418          */
1419         cork->dst = NULL;
1420         skb_dst_set(skb, &rt->dst);
1421
1422         if (iph->protocol == IPPROTO_ICMP)
1423                 icmp_out_count(net, ((struct icmphdr *)
1424                         skb_transport_header(skb))->type);
1425
1426         ip_cork_release(cork);
1427 out:
1428         return skb;
1429 }
1430
1431 int ip_send_skb(struct net *net, struct sk_buff *skb)
1432 {
1433         int err;
1434
1435         err = ip_local_out(skb);
1436         if (err) {
1437                 if (err > 0)
1438                         err = net_xmit_errno(err);
1439                 if (err)
1440                         IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
1441         }
1442
1443         return err;
1444 }
1445
1446 int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
1447 {
1448         struct sk_buff *skb;
1449
1450         skb = ip_finish_skb(sk, fl4);
1451         if (!skb)
1452                 return 0;
1453
1454         /* Netfilter gets whole the not fragmented skb. */
1455         return ip_send_skb(sock_net(sk), skb);
1456 }
1457
1458 /*
1459  *      Throw away all pending data on the socket.
1460  */
1461 static void __ip_flush_pending_frames(struct sock *sk,
1462                                       struct sk_buff_head *queue,
1463                                       struct inet_cork *cork)
1464 {
1465         struct sk_buff *skb;
1466
1467         while ((skb = __skb_dequeue_tail(queue)) != NULL)
1468                 kfree_skb(skb);
1469
1470         ip_cork_release(cork);
1471 }
1472
1473 void ip_flush_pending_frames(struct sock *sk)
1474 {
1475         __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
1476 }
1477
1478 struct sk_buff *ip_make_skb(struct sock *sk,
1479                             struct flowi4 *fl4,
1480                             int getfrag(void *from, char *to, int offset,
1481                                         int len, int odd, struct sk_buff *skb),
1482                             void *from, int length, int transhdrlen,
1483                             struct ipcm_cookie *ipc, struct rtable **rtp,
1484                             unsigned int flags)
1485 {
1486         struct inet_cork cork;
1487         struct sk_buff_head queue;
1488         int err;
1489
1490         if (flags & MSG_PROBE)
1491                 return NULL;
1492
1493         __skb_queue_head_init(&queue);
1494
1495         cork.flags = 0;
1496         cork.addr = 0;
1497         cork.opt = NULL;
1498         err = ip_setup_cork(sk, &cork, ipc, rtp);
1499         if (err)
1500                 return ERR_PTR(err);
1501
1502         err = __ip_append_data(sk, fl4, &queue, &cork,
1503                                &current->task_frag, getfrag,
1504                                from, length, transhdrlen, flags);
1505         if (err) {
1506                 __ip_flush_pending_frames(sk, &queue, &cork);
1507                 return ERR_PTR(err);
1508         }
1509
1510         return __ip_make_skb(sk, fl4, &queue, &cork);
1511 }
1512
1513 /*
1514  *      Fetch data from kernel space and fill in checksum if needed.
1515  */
1516 static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1517                               int len, int odd, struct sk_buff *skb)
1518 {
1519         __wsum csum;
1520
1521         csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1522         skb->csum = csum_block_add(skb->csum, csum, odd);
1523         return 0;
1524 }
1525
1526 /*
1527  *      Generic function to send a packet as reply to another packet.
1528  *      Used to send some TCP resets/acks so far.
1529  */
1530 void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,
1531                            const struct ip_options *sopt,
1532                            __be32 daddr, __be32 saddr,
1533                            const struct ip_reply_arg *arg,
1534                            unsigned int len)
1535 {
1536         struct ip_options_data replyopts;
1537         struct ipcm_cookie ipc;
1538         struct flowi4 fl4;
1539         struct rtable *rt = skb_rtable(skb);
1540         struct net *net = sock_net(sk);
1541         struct sk_buff *nskb;
1542         int err;
1543
1544         if (__ip_options_echo(&replyopts.opt.opt, skb, sopt))
1545                 return;
1546
1547         ipc.addr = daddr;
1548         ipc.opt = NULL;
1549         ipc.tx_flags = 0;
1550         ipc.ttl = 0;
1551         ipc.tos = -1;
1552
1553         if (replyopts.opt.opt.optlen) {
1554                 ipc.opt = &replyopts.opt;
1555
1556                 if (replyopts.opt.opt.srr)
1557                         daddr = replyopts.opt.opt.faddr;
1558         }
1559
1560         flowi4_init_output(&fl4, arg->bound_dev_if,
1561                            IP4_REPLY_MARK(net, skb->mark),
1562                            RT_TOS(arg->tos),
1563                            RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
1564                            ip_reply_arg_flowi_flags(arg),
1565                            daddr, saddr,
1566                            tcp_hdr(skb)->source, tcp_hdr(skb)->dest);
1567         security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
1568         rt = ip_route_output_key(net, &fl4);
1569         if (IS_ERR(rt))
1570                 return;
1571
1572         inet_sk(sk)->tos = arg->tos;
1573
1574         sk->sk_priority = skb->priority;
1575         sk->sk_protocol = ip_hdr(skb)->protocol;
1576         sk->sk_bound_dev_if = arg->bound_dev_if;
1577         sk->sk_sndbuf = sysctl_wmem_default;
1578         err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base,
1579                              len, 0, &ipc, &rt, MSG_DONTWAIT);
1580         if (unlikely(err)) {
1581                 ip_flush_pending_frames(sk);
1582                 goto out;
1583         }
1584
1585         nskb = skb_peek(&sk->sk_write_queue);
1586         if (nskb) {
1587                 if (arg->csumoffset >= 0)
1588                         *((__sum16 *)skb_transport_header(nskb) +
1589                           arg->csumoffset) = csum_fold(csum_add(nskb->csum,
1590                                                                 arg->csum));
1591                 nskb->ip_summed = CHECKSUM_NONE;
1592                 skb_set_queue_mapping(nskb, skb_get_queue_mapping(skb));
1593                 ip_push_pending_frames(sk, &fl4);
1594         }
1595 out:
1596         ip_rt_put(rt);
1597 }
1598
1599 void __init ip_init(void)
1600 {
1601         ip_rt_init();
1602         inet_initpeers();
1603
1604 #if defined(CONFIG_IP_MULTICAST)
1605         igmp_mc_init();
1606 #endif
1607 }