[IPSEC] xfrm: Abstract out encapsulation modes
[linux-2.6.git] / net / ipv6 / ip6_output.c
1 /*
2  *      IPv6 output functions
3  *      Linux INET6 implementation 
4  *
5  *      Authors:
6  *      Pedro Roque             <roque@di.fc.ul.pt>     
7  *
8  *      $Id: ip6_output.c,v 1.34 2002/02/01 22:01:04 davem Exp $
9  *
10  *      Based on linux/net/ipv4/ip_output.c
11  *
12  *      This program is free software; you can redistribute it and/or
13  *      modify it under the terms of the GNU General Public License
14  *      as published by the Free Software Foundation; either version
15  *      2 of the License, or (at your option) any later version.
16  *
17  *      Changes:
18  *      A.N.Kuznetsov   :       airthmetics in fragmentation.
19  *                              extension headers are implemented.
20  *                              route changes now work.
21  *                              ip6_forward does not confuse sniffers.
22  *                              etc.
23  *
24  *      H. von Brand    :       Added missing #include <linux/string.h>
25  *      Imran Patel     :       frag id should be in NBO
26  *      Kazunori MIYAZAWA @USAGI
27  *                      :       add ip6_append_data and related functions
28  *                              for datagram xmit
29  */
30
31 #include <linux/config.h>
32 #include <linux/errno.h>
33 #include <linux/types.h>
34 #include <linux/string.h>
35 #include <linux/socket.h>
36 #include <linux/net.h>
37 #include <linux/netdevice.h>
38 #include <linux/if_arp.h>
39 #include <linux/in6.h>
40 #include <linux/tcp.h>
41 #include <linux/route.h>
42 #include <linux/module.h>
43
44 #include <linux/netfilter.h>
45 #include <linux/netfilter_ipv6.h>
46
47 #include <net/sock.h>
48 #include <net/snmp.h>
49
50 #include <net/ipv6.h>
51 #include <net/ndisc.h>
52 #include <net/protocol.h>
53 #include <net/ip6_route.h>
54 #include <net/addrconf.h>
55 #include <net/rawv6.h>
56 #include <net/icmp.h>
57 #include <net/xfrm.h>
58 #include <net/checksum.h>
59
60 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *));
61
62 static __inline__ void ipv6_select_ident(struct sk_buff *skb, struct frag_hdr *fhdr)
63 {
64         static u32 ipv6_fragmentation_id = 1;
65         static DEFINE_SPINLOCK(ip6_id_lock);
66
67         spin_lock_bh(&ip6_id_lock);
68         fhdr->identification = htonl(ipv6_fragmentation_id);
69         if (++ipv6_fragmentation_id == 0)
70                 ipv6_fragmentation_id = 1;
71         spin_unlock_bh(&ip6_id_lock);
72 }
73
74 static inline int ip6_output_finish(struct sk_buff *skb)
75 {
76
77         struct dst_entry *dst = skb->dst;
78         struct hh_cache *hh = dst->hh;
79
80         if (hh) {
81                 int hh_alen;
82
83                 read_lock_bh(&hh->hh_lock);
84                 hh_alen = HH_DATA_ALIGN(hh->hh_len);
85                 memcpy(skb->data - hh_alen, hh->hh_data, hh_alen);
86                 read_unlock_bh(&hh->hh_lock);
87                 skb_push(skb, hh->hh_len);
88                 return hh->hh_output(skb);
89         } else if (dst->neighbour)
90                 return dst->neighbour->output(skb);
91
92         IP6_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
93         kfree_skb(skb);
94         return -EINVAL;
95
96 }
97
98 /* dev_loopback_xmit for use with netfilter. */
99 static int ip6_dev_loopback_xmit(struct sk_buff *newskb)
100 {
101         newskb->mac.raw = newskb->data;
102         __skb_pull(newskb, newskb->nh.raw - newskb->data);
103         newskb->pkt_type = PACKET_LOOPBACK;
104         newskb->ip_summed = CHECKSUM_UNNECESSARY;
105         BUG_TRAP(newskb->dst);
106
107         netif_rx(newskb);
108         return 0;
109 }
110
111
112 static int ip6_output2(struct sk_buff *skb)
113 {
114         struct dst_entry *dst = skb->dst;
115         struct net_device *dev = dst->dev;
116
117         skb->protocol = htons(ETH_P_IPV6);
118         skb->dev = dev;
119
120         if (ipv6_addr_is_multicast(&skb->nh.ipv6h->daddr)) {
121                 struct ipv6_pinfo* np = skb->sk ? inet6_sk(skb->sk) : NULL;
122
123                 if (!(dev->flags & IFF_LOOPBACK) && (!np || np->mc_loop) &&
124                     ipv6_chk_mcast_addr(dev, &skb->nh.ipv6h->daddr,
125                                 &skb->nh.ipv6h->saddr)) {
126                         struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
127
128                         /* Do not check for IFF_ALLMULTI; multicast routing
129                            is not supported in any case.
130                          */
131                         if (newskb)
132                                 NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, newskb, NULL,
133                                         newskb->dev,
134                                         ip6_dev_loopback_xmit);
135
136                         if (skb->nh.ipv6h->hop_limit == 0) {
137                                 IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
138                                 kfree_skb(skb);
139                                 return 0;
140                         }
141                 }
142
143                 IP6_INC_STATS(IPSTATS_MIB_OUTMCASTPKTS);
144         }
145
146         return NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, skb,NULL, skb->dev,ip6_output_finish);
147 }
148
149 int ip6_output(struct sk_buff *skb)
150 {
151         if ((skb->len > dst_mtu(skb->dst) && !skb_shinfo(skb)->ufo_size) ||
152                                 dst_allfrag(skb->dst))
153                 return ip6_fragment(skb, ip6_output2);
154         else
155                 return ip6_output2(skb);
156 }
157
158 /*
159  *      xmit an sk_buff (used by TCP)
160  */
161
162 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
163              struct ipv6_txoptions *opt, int ipfragok)
164 {
165         struct ipv6_pinfo *np = inet6_sk(sk);
166         struct in6_addr *first_hop = &fl->fl6_dst;
167         struct dst_entry *dst = skb->dst;
168         struct ipv6hdr *hdr;
169         u8  proto = fl->proto;
170         int seg_len = skb->len;
171         int hlimit, tclass;
172         u32 mtu;
173
174         if (opt) {
175                 int head_room;
176
177                 /* First: exthdrs may take lots of space (~8K for now)
178                    MAX_HEADER is not enough.
179                  */
180                 head_room = opt->opt_nflen + opt->opt_flen;
181                 seg_len += head_room;
182                 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
183
184                 if (skb_headroom(skb) < head_room) {
185                         struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
186                         kfree_skb(skb);
187                         skb = skb2;
188                         if (skb == NULL) {      
189                                 IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
190                                 return -ENOBUFS;
191                         }
192                         if (sk)
193                                 skb_set_owner_w(skb, sk);
194                 }
195                 if (opt->opt_flen)
196                         ipv6_push_frag_opts(skb, opt, &proto);
197                 if (opt->opt_nflen)
198                         ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
199         }
200
201         hdr = skb->nh.ipv6h = (struct ipv6hdr*)skb_push(skb, sizeof(struct ipv6hdr));
202
203         /*
204          *      Fill in the IPv6 header
205          */
206
207         hlimit = -1;
208         if (np)
209                 hlimit = np->hop_limit;
210         if (hlimit < 0)
211                 hlimit = dst_metric(dst, RTAX_HOPLIMIT);
212         if (hlimit < 0)
213                 hlimit = ipv6_get_hoplimit(dst->dev);
214
215         tclass = -1;
216         if (np)
217                 tclass = np->tclass;
218         if (tclass < 0)
219                 tclass = 0;
220
221         *(u32 *)hdr = htonl(0x60000000 | (tclass << 20)) | fl->fl6_flowlabel;
222
223         hdr->payload_len = htons(seg_len);
224         hdr->nexthdr = proto;
225         hdr->hop_limit = hlimit;
226
227         ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
228         ipv6_addr_copy(&hdr->daddr, first_hop);
229
230         skb->priority = sk->sk_priority;
231
232         mtu = dst_mtu(dst);
233         if ((skb->len <= mtu) || ipfragok) {
234                 IP6_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
235                 return NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, dst->dev,
236                                 dst_output);
237         }
238
239         if (net_ratelimit())
240                 printk(KERN_DEBUG "IPv6: sending pkt_too_big to self\n");
241         skb->dev = dst->dev;
242         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev);
243         IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
244         kfree_skb(skb);
245         return -EMSGSIZE;
246 }
247
248 /*
249  *      To avoid extra problems ND packets are send through this
250  *      routine. It's code duplication but I really want to avoid
251  *      extra checks since ipv6_build_header is used by TCP (which
252  *      is for us performance critical)
253  */
254
255 int ip6_nd_hdr(struct sock *sk, struct sk_buff *skb, struct net_device *dev,
256                struct in6_addr *saddr, struct in6_addr *daddr,
257                int proto, int len)
258 {
259         struct ipv6_pinfo *np = inet6_sk(sk);
260         struct ipv6hdr *hdr;
261         int totlen;
262
263         skb->protocol = htons(ETH_P_IPV6);
264         skb->dev = dev;
265
266         totlen = len + sizeof(struct ipv6hdr);
267
268         hdr = (struct ipv6hdr *) skb_put(skb, sizeof(struct ipv6hdr));
269         skb->nh.ipv6h = hdr;
270
271         *(u32*)hdr = htonl(0x60000000);
272
273         hdr->payload_len = htons(len);
274         hdr->nexthdr = proto;
275         hdr->hop_limit = np->hop_limit;
276
277         ipv6_addr_copy(&hdr->saddr, saddr);
278         ipv6_addr_copy(&hdr->daddr, daddr);
279
280         return 0;
281 }
282
283 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
284 {
285         struct ip6_ra_chain *ra;
286         struct sock *last = NULL;
287
288         read_lock(&ip6_ra_lock);
289         for (ra = ip6_ra_chain; ra; ra = ra->next) {
290                 struct sock *sk = ra->sk;
291                 if (sk && ra->sel == sel &&
292                     (!sk->sk_bound_dev_if ||
293                      sk->sk_bound_dev_if == skb->dev->ifindex)) {
294                         if (last) {
295                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
296                                 if (skb2)
297                                         rawv6_rcv(last, skb2);
298                         }
299                         last = sk;
300                 }
301         }
302
303         if (last) {
304                 rawv6_rcv(last, skb);
305                 read_unlock(&ip6_ra_lock);
306                 return 1;
307         }
308         read_unlock(&ip6_ra_lock);
309         return 0;
310 }
311
312 static inline int ip6_forward_finish(struct sk_buff *skb)
313 {
314         return dst_output(skb);
315 }
316
317 int ip6_forward(struct sk_buff *skb)
318 {
319         struct dst_entry *dst = skb->dst;
320         struct ipv6hdr *hdr = skb->nh.ipv6h;
321         struct inet6_skb_parm *opt = IP6CB(skb);
322         
323         if (ipv6_devconf.forwarding == 0)
324                 goto error;
325
326         if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
327                 IP6_INC_STATS(IPSTATS_MIB_INDISCARDS);
328                 goto drop;
329         }
330
331         skb->ip_summed = CHECKSUM_NONE;
332
333         /*
334          *      We DO NOT make any processing on
335          *      RA packets, pushing them to user level AS IS
336          *      without ane WARRANTY that application will be able
337          *      to interpret them. The reason is that we
338          *      cannot make anything clever here.
339          *
340          *      We are not end-node, so that if packet contains
341          *      AH/ESP, we cannot make anything.
342          *      Defragmentation also would be mistake, RA packets
343          *      cannot be fragmented, because there is no warranty
344          *      that different fragments will go along one path. --ANK
345          */
346         if (opt->ra) {
347                 u8 *ptr = skb->nh.raw + opt->ra;
348                 if (ip6_call_ra_chain(skb, (ptr[2]<<8) + ptr[3]))
349                         return 0;
350         }
351
352         /*
353          *      check and decrement ttl
354          */
355         if (hdr->hop_limit <= 1) {
356                 /* Force OUTPUT device used as source address */
357                 skb->dev = dst->dev;
358                 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT,
359                             0, skb->dev);
360
361                 kfree_skb(skb);
362                 return -ETIMEDOUT;
363         }
364
365         if (!xfrm6_route_forward(skb)) {
366                 IP6_INC_STATS(IPSTATS_MIB_INDISCARDS);
367                 goto drop;
368         }
369         dst = skb->dst;
370
371         /* IPv6 specs say nothing about it, but it is clear that we cannot
372            send redirects to source routed frames.
373          */
374         if (skb->dev == dst->dev && dst->neighbour && opt->srcrt == 0) {
375                 struct in6_addr *target = NULL;
376                 struct rt6_info *rt;
377                 struct neighbour *n = dst->neighbour;
378
379                 /*
380                  *      incoming and outgoing devices are the same
381                  *      send a redirect.
382                  */
383
384                 rt = (struct rt6_info *) dst;
385                 if ((rt->rt6i_flags & RTF_GATEWAY))
386                         target = (struct in6_addr*)&n->primary_key;
387                 else
388                         target = &hdr->daddr;
389
390                 /* Limit redirects both by destination (here)
391                    and by source (inside ndisc_send_redirect)
392                  */
393                 if (xrlim_allow(dst, 1*HZ))
394                         ndisc_send_redirect(skb, n, target);
395         } else if (ipv6_addr_type(&hdr->saddr)&(IPV6_ADDR_MULTICAST|IPV6_ADDR_LOOPBACK
396                                                 |IPV6_ADDR_LINKLOCAL)) {
397                 /* This check is security critical. */
398                 goto error;
399         }
400
401         if (skb->len > dst_mtu(dst)) {
402                 /* Again, force OUTPUT device used as source address */
403                 skb->dev = dst->dev;
404                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, dst_mtu(dst), skb->dev);
405                 IP6_INC_STATS_BH(IPSTATS_MIB_INTOOBIGERRORS);
406                 IP6_INC_STATS_BH(IPSTATS_MIB_FRAGFAILS);
407                 kfree_skb(skb);
408                 return -EMSGSIZE;
409         }
410
411         if (skb_cow(skb, dst->dev->hard_header_len)) {
412                 IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
413                 goto drop;
414         }
415
416         hdr = skb->nh.ipv6h;
417
418         /* Mangling hops number delayed to point after skb COW */
419  
420         hdr->hop_limit--;
421
422         IP6_INC_STATS_BH(IPSTATS_MIB_OUTFORWDATAGRAMS);
423         return NF_HOOK(PF_INET6,NF_IP6_FORWARD, skb, skb->dev, dst->dev, ip6_forward_finish);
424
425 error:
426         IP6_INC_STATS_BH(IPSTATS_MIB_INADDRERRORS);
427 drop:
428         kfree_skb(skb);
429         return -EINVAL;
430 }
431
432 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
433 {
434         to->pkt_type = from->pkt_type;
435         to->priority = from->priority;
436         to->protocol = from->protocol;
437         dst_release(to->dst);
438         to->dst = dst_clone(from->dst);
439         to->dev = from->dev;
440
441 #ifdef CONFIG_NET_SCHED
442         to->tc_index = from->tc_index;
443 #endif
444 #ifdef CONFIG_NETFILTER
445         to->nfmark = from->nfmark;
446         /* Connection association is same as pre-frag packet */
447         nf_conntrack_put(to->nfct);
448         to->nfct = from->nfct;
449         nf_conntrack_get(to->nfct);
450         to->nfctinfo = from->nfctinfo;
451 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
452         nf_conntrack_put_reasm(to->nfct_reasm);
453         to->nfct_reasm = from->nfct_reasm;
454         nf_conntrack_get_reasm(to->nfct_reasm);
455 #endif
456 #ifdef CONFIG_BRIDGE_NETFILTER
457         nf_bridge_put(to->nf_bridge);
458         to->nf_bridge = from->nf_bridge;
459         nf_bridge_get(to->nf_bridge);
460 #endif
461 #endif
462 }
463
464 int ip6_find_1stfragopt(struct sk_buff *skb, u8 **nexthdr)
465 {
466         u16 offset = sizeof(struct ipv6hdr);
467         struct ipv6_opt_hdr *exthdr = (struct ipv6_opt_hdr*)(skb->nh.ipv6h + 1);
468         unsigned int packet_len = skb->tail - skb->nh.raw;
469         int found_rhdr = 0;
470         *nexthdr = &skb->nh.ipv6h->nexthdr;
471
472         while (offset + 1 <= packet_len) {
473
474                 switch (**nexthdr) {
475
476                 case NEXTHDR_HOP:
477                 case NEXTHDR_ROUTING:
478                 case NEXTHDR_DEST:
479                         if (**nexthdr == NEXTHDR_ROUTING) found_rhdr = 1;
480                         if (**nexthdr == NEXTHDR_DEST && found_rhdr) return offset;
481                         offset += ipv6_optlen(exthdr);
482                         *nexthdr = &exthdr->nexthdr;
483                         exthdr = (struct ipv6_opt_hdr*)(skb->nh.raw + offset);
484                         break;
485                 default :
486                         return offset;
487                 }
488         }
489
490         return offset;
491 }
492 EXPORT_SYMBOL_GPL(ip6_find_1stfragopt);
493
494 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
495 {
496         struct net_device *dev;
497         struct sk_buff *frag;
498         struct rt6_info *rt = (struct rt6_info*)skb->dst;
499         struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
500         struct ipv6hdr *tmp_hdr;
501         struct frag_hdr *fh;
502         unsigned int mtu, hlen, left, len;
503         u32 frag_id = 0;
504         int ptr, offset = 0, err=0;
505         u8 *prevhdr, nexthdr = 0;
506
507         dev = rt->u.dst.dev;
508         hlen = ip6_find_1stfragopt(skb, &prevhdr);
509         nexthdr = *prevhdr;
510
511         mtu = dst_mtu(&rt->u.dst);
512         if (np && np->frag_size < mtu) {
513                 if (np->frag_size)
514                         mtu = np->frag_size;
515         }
516         mtu -= hlen + sizeof(struct frag_hdr);
517
518         if (skb_shinfo(skb)->frag_list) {
519                 int first_len = skb_pagelen(skb);
520
521                 if (first_len - hlen > mtu ||
522                     ((first_len - hlen) & 7) ||
523                     skb_cloned(skb))
524                         goto slow_path;
525
526                 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
527                         /* Correct geometry. */
528                         if (frag->len > mtu ||
529                             ((frag->len & 7) && frag->next) ||
530                             skb_headroom(frag) < hlen)
531                             goto slow_path;
532
533                         /* Partially cloned skb? */
534                         if (skb_shared(frag))
535                                 goto slow_path;
536
537                         BUG_ON(frag->sk);
538                         if (skb->sk) {
539                                 sock_hold(skb->sk);
540                                 frag->sk = skb->sk;
541                                 frag->destructor = sock_wfree;
542                                 skb->truesize -= frag->truesize;
543                         }
544                 }
545
546                 err = 0;
547                 offset = 0;
548                 frag = skb_shinfo(skb)->frag_list;
549                 skb_shinfo(skb)->frag_list = NULL;
550                 /* BUILD HEADER */
551
552                 tmp_hdr = kmalloc(hlen, GFP_ATOMIC);
553                 if (!tmp_hdr) {
554                         IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
555                         return -ENOMEM;
556                 }
557
558                 *prevhdr = NEXTHDR_FRAGMENT;
559                 memcpy(tmp_hdr, skb->nh.raw, hlen);
560                 __skb_pull(skb, hlen);
561                 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
562                 skb->nh.raw = __skb_push(skb, hlen);
563                 memcpy(skb->nh.raw, tmp_hdr, hlen);
564
565                 ipv6_select_ident(skb, fh);
566                 fh->nexthdr = nexthdr;
567                 fh->reserved = 0;
568                 fh->frag_off = htons(IP6_MF);
569                 frag_id = fh->identification;
570
571                 first_len = skb_pagelen(skb);
572                 skb->data_len = first_len - skb_headlen(skb);
573                 skb->len = first_len;
574                 skb->nh.ipv6h->payload_len = htons(first_len - sizeof(struct ipv6hdr));
575  
576
577                 for (;;) {
578                         /* Prepare header of the next frame,
579                          * before previous one went down. */
580                         if (frag) {
581                                 frag->ip_summed = CHECKSUM_NONE;
582                                 frag->h.raw = frag->data;
583                                 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
584                                 frag->nh.raw = __skb_push(frag, hlen);
585                                 memcpy(frag->nh.raw, tmp_hdr, hlen);
586                                 offset += skb->len - hlen - sizeof(struct frag_hdr);
587                                 fh->nexthdr = nexthdr;
588                                 fh->reserved = 0;
589                                 fh->frag_off = htons(offset);
590                                 if (frag->next != NULL)
591                                         fh->frag_off |= htons(IP6_MF);
592                                 fh->identification = frag_id;
593                                 frag->nh.ipv6h->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
594                                 ip6_copy_metadata(frag, skb);
595                         }
596                         
597                         err = output(skb);
598                         if (err || !frag)
599                                 break;
600
601                         skb = frag;
602                         frag = skb->next;
603                         skb->next = NULL;
604                 }
605
606                 kfree(tmp_hdr);
607
608                 if (err == 0) {
609                         IP6_INC_STATS(IPSTATS_MIB_FRAGOKS);
610                         return 0;
611                 }
612
613                 while (frag) {
614                         skb = frag->next;
615                         kfree_skb(frag);
616                         frag = skb;
617                 }
618
619                 IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
620                 return err;
621         }
622
623 slow_path:
624         left = skb->len - hlen;         /* Space per frame */
625         ptr = hlen;                     /* Where to start from */
626
627         /*
628          *      Fragment the datagram.
629          */
630
631         *prevhdr = NEXTHDR_FRAGMENT;
632
633         /*
634          *      Keep copying data until we run out.
635          */
636         while(left > 0) {
637                 len = left;
638                 /* IF: it doesn't fit, use 'mtu' - the data space left */
639                 if (len > mtu)
640                         len = mtu;
641                 /* IF: we are not sending upto and including the packet end
642                    then align the next start on an eight byte boundary */
643                 if (len < left) {
644                         len &= ~7;
645                 }
646                 /*
647                  *      Allocate buffer.
648                  */
649
650                 if ((frag = alloc_skb(len+hlen+sizeof(struct frag_hdr)+LL_RESERVED_SPACE(rt->u.dst.dev), GFP_ATOMIC)) == NULL) {
651                         NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
652                         IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
653                         err = -ENOMEM;
654                         goto fail;
655                 }
656
657                 /*
658                  *      Set up data on packet
659                  */
660
661                 ip6_copy_metadata(frag, skb);
662                 skb_reserve(frag, LL_RESERVED_SPACE(rt->u.dst.dev));
663                 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
664                 frag->nh.raw = frag->data;
665                 fh = (struct frag_hdr*)(frag->data + hlen);
666                 frag->h.raw = frag->data + hlen + sizeof(struct frag_hdr);
667
668                 /*
669                  *      Charge the memory for the fragment to any owner
670                  *      it might possess
671                  */
672                 if (skb->sk)
673                         skb_set_owner_w(frag, skb->sk);
674
675                 /*
676                  *      Copy the packet header into the new buffer.
677                  */
678                 memcpy(frag->nh.raw, skb->data, hlen);
679
680                 /*
681                  *      Build fragment header.
682                  */
683                 fh->nexthdr = nexthdr;
684                 fh->reserved = 0;
685                 if (!frag_id) {
686                         ipv6_select_ident(skb, fh);
687                         frag_id = fh->identification;
688                 } else
689                         fh->identification = frag_id;
690
691                 /*
692                  *      Copy a block of the IP datagram.
693                  */
694                 if (skb_copy_bits(skb, ptr, frag->h.raw, len))
695                         BUG();
696                 left -= len;
697
698                 fh->frag_off = htons(offset);
699                 if (left > 0)
700                         fh->frag_off |= htons(IP6_MF);
701                 frag->nh.ipv6h->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
702
703                 ptr += len;
704                 offset += len;
705
706                 /*
707                  *      Put this fragment into the sending queue.
708                  */
709
710                 IP6_INC_STATS(IPSTATS_MIB_FRAGCREATES);
711
712                 err = output(frag);
713                 if (err)
714                         goto fail;
715         }
716         kfree_skb(skb);
717         IP6_INC_STATS(IPSTATS_MIB_FRAGOKS);
718         return err;
719
720 fail:
721         kfree_skb(skb); 
722         IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
723         return err;
724 }
725
726 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
727 {
728         int err = 0;
729
730         *dst = NULL;
731         if (sk) {
732                 struct ipv6_pinfo *np = inet6_sk(sk);
733         
734                 *dst = sk_dst_check(sk, np->dst_cookie);
735                 if (*dst) {
736                         struct rt6_info *rt = (struct rt6_info*)*dst;
737         
738                         /* Yes, checking route validity in not connected
739                          * case is not very simple. Take into account,
740                          * that we do not support routing by source, TOS,
741                          * and MSG_DONTROUTE            --ANK (980726)
742                          *
743                          * 1. If route was host route, check that
744                          *    cached destination is current.
745                          *    If it is network route, we still may
746                          *    check its validity using saved pointer
747                          *    to the last used address: daddr_cache.
748                          *    We do not want to save whole address now,
749                          *    (because main consumer of this service
750                          *    is tcp, which has not this problem),
751                          *    so that the last trick works only on connected
752                          *    sockets.
753                          * 2. oif also should be the same.
754                          */
755                         if (((rt->rt6i_dst.plen != 128 ||
756                               !ipv6_addr_equal(&fl->fl6_dst,
757                                                &rt->rt6i_dst.addr))
758                              && (np->daddr_cache == NULL ||
759                                  !ipv6_addr_equal(&fl->fl6_dst,
760                                                   np->daddr_cache)))
761                             || (fl->oif && fl->oif != (*dst)->dev->ifindex)) {
762                                 dst_release(*dst);
763                                 *dst = NULL;
764                         }
765                 }
766         }
767
768         if (*dst == NULL)
769                 *dst = ip6_route_output(sk, fl);
770
771         if ((err = (*dst)->error))
772                 goto out_err_release;
773
774         if (ipv6_addr_any(&fl->fl6_src)) {
775                 err = ipv6_get_saddr(*dst, &fl->fl6_dst, &fl->fl6_src);
776
777                 if (err)
778                         goto out_err_release;
779         }
780
781         return 0;
782
783 out_err_release:
784         dst_release(*dst);
785         *dst = NULL;
786         return err;
787 }
788
789 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
790
791 static inline int ip6_ufo_append_data(struct sock *sk,
792                         int getfrag(void *from, char *to, int offset, int len,
793                         int odd, struct sk_buff *skb),
794                         void *from, int length, int hh_len, int fragheaderlen,
795                         int transhdrlen, int mtu,unsigned int flags)
796
797 {
798         struct sk_buff *skb;
799         int err;
800
801         /* There is support for UDP large send offload by network
802          * device, so create one single skb packet containing complete
803          * udp datagram
804          */
805         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
806                 skb = sock_alloc_send_skb(sk,
807                         hh_len + fragheaderlen + transhdrlen + 20,
808                         (flags & MSG_DONTWAIT), &err);
809                 if (skb == NULL)
810                         return -ENOMEM;
811
812                 /* reserve space for Hardware header */
813                 skb_reserve(skb, hh_len);
814
815                 /* create space for UDP/IP header */
816                 skb_put(skb,fragheaderlen + transhdrlen);
817
818                 /* initialize network header pointer */
819                 skb->nh.raw = skb->data;
820
821                 /* initialize protocol header pointer */
822                 skb->h.raw = skb->data + fragheaderlen;
823
824                 skb->ip_summed = CHECKSUM_HW;
825                 skb->csum = 0;
826                 sk->sk_sndmsg_off = 0;
827         }
828
829         err = skb_append_datato_frags(sk,skb, getfrag, from,
830                                       (length - transhdrlen));
831         if (!err) {
832                 struct frag_hdr fhdr;
833
834                 /* specify the length of each IP datagram fragment*/
835                 skb_shinfo(skb)->ufo_size = (mtu - fragheaderlen) - 
836                                                 sizeof(struct frag_hdr);
837                 ipv6_select_ident(skb, &fhdr);
838                 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
839                 __skb_queue_tail(&sk->sk_write_queue, skb);
840
841                 return 0;
842         }
843         /* There is not enough support do UPD LSO,
844          * so follow normal path
845          */
846         kfree_skb(skb);
847
848         return err;
849 }
850
851 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
852         int offset, int len, int odd, struct sk_buff *skb),
853         void *from, int length, int transhdrlen,
854         int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi *fl,
855         struct rt6_info *rt, unsigned int flags)
856 {
857         struct inet_sock *inet = inet_sk(sk);
858         struct ipv6_pinfo *np = inet6_sk(sk);
859         struct sk_buff *skb;
860         unsigned int maxfraglen, fragheaderlen;
861         int exthdrlen;
862         int hh_len;
863         int mtu;
864         int copy;
865         int err;
866         int offset = 0;
867         int csummode = CHECKSUM_NONE;
868
869         if (flags&MSG_PROBE)
870                 return 0;
871         if (skb_queue_empty(&sk->sk_write_queue)) {
872                 /*
873                  * setup for corking
874                  */
875                 if (opt) {
876                         if (np->cork.opt == NULL) {
877                                 np->cork.opt = kmalloc(opt->tot_len,
878                                                        sk->sk_allocation);
879                                 if (unlikely(np->cork.opt == NULL))
880                                         return -ENOBUFS;
881                         } else if (np->cork.opt->tot_len < opt->tot_len) {
882                                 printk(KERN_DEBUG "ip6_append_data: invalid option length\n");
883                                 return -EINVAL;
884                         }
885                         memcpy(np->cork.opt, opt, opt->tot_len);
886                         inet->cork.flags |= IPCORK_OPT;
887                         /* need source address above miyazawa*/
888                 }
889                 dst_hold(&rt->u.dst);
890                 np->cork.rt = rt;
891                 inet->cork.fl = *fl;
892                 np->cork.hop_limit = hlimit;
893                 np->cork.tclass = tclass;
894                 mtu = dst_mtu(rt->u.dst.path);
895                 if (np->frag_size < mtu) {
896                         if (np->frag_size)
897                                 mtu = np->frag_size;
898                 }
899                 inet->cork.fragsize = mtu;
900                 if (dst_allfrag(rt->u.dst.path))
901                         inet->cork.flags |= IPCORK_ALLFRAG;
902                 inet->cork.length = 0;
903                 sk->sk_sndmsg_page = NULL;
904                 sk->sk_sndmsg_off = 0;
905                 exthdrlen = rt->u.dst.header_len + (opt ? opt->opt_flen : 0);
906                 length += exthdrlen;
907                 transhdrlen += exthdrlen;
908         } else {
909                 rt = np->cork.rt;
910                 fl = &inet->cork.fl;
911                 if (inet->cork.flags & IPCORK_OPT)
912                         opt = np->cork.opt;
913                 transhdrlen = 0;
914                 exthdrlen = 0;
915                 mtu = inet->cork.fragsize;
916         }
917
918         hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
919
920         fragheaderlen = sizeof(struct ipv6hdr) + (opt ? opt->opt_nflen : 0);
921         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
922
923         if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
924                 if (inet->cork.length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
925                         ipv6_local_error(sk, EMSGSIZE, fl, mtu-exthdrlen);
926                         return -EMSGSIZE;
927                 }
928         }
929
930         /*
931          * Let's try using as much space as possible.
932          * Use MTU if total length of the message fits into the MTU.
933          * Otherwise, we need to reserve fragment header and
934          * fragment alignment (= 8-15 octects, in total).
935          *
936          * Note that we may need to "move" the data from the tail of
937          * of the buffer to the new fragment when we split 
938          * the message.
939          *
940          * FIXME: It may be fragmented into multiple chunks 
941          *        at once if non-fragmentable extension headers
942          *        are too large.
943          * --yoshfuji 
944          */
945
946         inet->cork.length += length;
947         if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
948             (rt->u.dst.dev->features & NETIF_F_UFO)) {
949
950                 err = ip6_ufo_append_data(sk, getfrag, from, length, hh_len,
951                                           fragheaderlen, transhdrlen, mtu,
952                                           flags);
953                 if (err)
954                         goto error;
955                 return 0;
956         }
957
958         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
959                 goto alloc_new_skb;
960
961         while (length > 0) {
962                 /* Check if the remaining data fits into current packet. */
963                 copy = (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
964                 if (copy < length)
965                         copy = maxfraglen - skb->len;
966
967                 if (copy <= 0) {
968                         char *data;
969                         unsigned int datalen;
970                         unsigned int fraglen;
971                         unsigned int fraggap;
972                         unsigned int alloclen;
973                         struct sk_buff *skb_prev;
974 alloc_new_skb:
975                         skb_prev = skb;
976
977                         /* There's no room in the current skb */
978                         if (skb_prev)
979                                 fraggap = skb_prev->len - maxfraglen;
980                         else
981                                 fraggap = 0;
982
983                         /*
984                          * If remaining data exceeds the mtu,
985                          * we know we need more fragment(s).
986                          */
987                         datalen = length + fraggap;
988                         if (datalen > (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
989                                 datalen = maxfraglen - fragheaderlen;
990
991                         fraglen = datalen + fragheaderlen;
992                         if ((flags & MSG_MORE) &&
993                             !(rt->u.dst.dev->features&NETIF_F_SG))
994                                 alloclen = mtu;
995                         else
996                                 alloclen = datalen + fragheaderlen;
997
998                         /*
999                          * The last fragment gets additional space at tail.
1000                          * Note: we overallocate on fragments with MSG_MODE
1001                          * because we have no idea if we're the last one.
1002                          */
1003                         if (datalen == length + fraggap)
1004                                 alloclen += rt->u.dst.trailer_len;
1005
1006                         /*
1007                          * We just reserve space for fragment header.
1008                          * Note: this may be overallocation if the message 
1009                          * (without MSG_MORE) fits into the MTU.
1010                          */
1011                         alloclen += sizeof(struct frag_hdr);
1012
1013                         if (transhdrlen) {
1014                                 skb = sock_alloc_send_skb(sk,
1015                                                 alloclen + hh_len,
1016                                                 (flags & MSG_DONTWAIT), &err);
1017                         } else {
1018                                 skb = NULL;
1019                                 if (atomic_read(&sk->sk_wmem_alloc) <=
1020                                     2 * sk->sk_sndbuf)
1021                                         skb = sock_wmalloc(sk,
1022                                                            alloclen + hh_len, 1,
1023                                                            sk->sk_allocation);
1024                                 if (unlikely(skb == NULL))
1025                                         err = -ENOBUFS;
1026                         }
1027                         if (skb == NULL)
1028                                 goto error;
1029                         /*
1030                          *      Fill in the control structures
1031                          */
1032                         skb->ip_summed = csummode;
1033                         skb->csum = 0;
1034                         /* reserve for fragmentation */
1035                         skb_reserve(skb, hh_len+sizeof(struct frag_hdr));
1036
1037                         /*
1038                          *      Find where to start putting bytes
1039                          */
1040                         data = skb_put(skb, fraglen);
1041                         skb->nh.raw = data + exthdrlen;
1042                         data += fragheaderlen;
1043                         skb->h.raw = data + exthdrlen;
1044
1045                         if (fraggap) {
1046                                 skb->csum = skb_copy_and_csum_bits(
1047                                         skb_prev, maxfraglen,
1048                                         data + transhdrlen, fraggap, 0);
1049                                 skb_prev->csum = csum_sub(skb_prev->csum,
1050                                                           skb->csum);
1051                                 data += fraggap;
1052                                 skb_trim(skb_prev, maxfraglen);
1053                         }
1054                         copy = datalen - transhdrlen - fraggap;
1055                         if (copy < 0) {
1056                                 err = -EINVAL;
1057                                 kfree_skb(skb);
1058                                 goto error;
1059                         } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1060                                 err = -EFAULT;
1061                                 kfree_skb(skb);
1062                                 goto error;
1063                         }
1064
1065                         offset += copy;
1066                         length -= datalen - fraggap;
1067                         transhdrlen = 0;
1068                         exthdrlen = 0;
1069                         csummode = CHECKSUM_NONE;
1070
1071                         /*
1072                          * Put the packet on the pending queue
1073                          */
1074                         __skb_queue_tail(&sk->sk_write_queue, skb);
1075                         continue;
1076                 }
1077
1078                 if (copy > length)
1079                         copy = length;
1080
1081                 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
1082                         unsigned int off;
1083
1084                         off = skb->len;
1085                         if (getfrag(from, skb_put(skb, copy),
1086                                                 offset, copy, off, skb) < 0) {
1087                                 __skb_trim(skb, off);
1088                                 err = -EFAULT;
1089                                 goto error;
1090                         }
1091                 } else {
1092                         int i = skb_shinfo(skb)->nr_frags;
1093                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1094                         struct page *page = sk->sk_sndmsg_page;
1095                         int off = sk->sk_sndmsg_off;
1096                         unsigned int left;
1097
1098                         if (page && (left = PAGE_SIZE - off) > 0) {
1099                                 if (copy >= left)
1100                                         copy = left;
1101                                 if (page != frag->page) {
1102                                         if (i == MAX_SKB_FRAGS) {
1103                                                 err = -EMSGSIZE;
1104                                                 goto error;
1105                                         }
1106                                         get_page(page);
1107                                         skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1108                                         frag = &skb_shinfo(skb)->frags[i];
1109                                 }
1110                         } else if(i < MAX_SKB_FRAGS) {
1111                                 if (copy > PAGE_SIZE)
1112                                         copy = PAGE_SIZE;
1113                                 page = alloc_pages(sk->sk_allocation, 0);
1114                                 if (page == NULL) {
1115                                         err = -ENOMEM;
1116                                         goto error;
1117                                 }
1118                                 sk->sk_sndmsg_page = page;
1119                                 sk->sk_sndmsg_off = 0;
1120
1121                                 skb_fill_page_desc(skb, i, page, 0, 0);
1122                                 frag = &skb_shinfo(skb)->frags[i];
1123                                 skb->truesize += PAGE_SIZE;
1124                                 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1125                         } else {
1126                                 err = -EMSGSIZE;
1127                                 goto error;
1128                         }
1129                         if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1130                                 err = -EFAULT;
1131                                 goto error;
1132                         }
1133                         sk->sk_sndmsg_off += copy;
1134                         frag->size += copy;
1135                         skb->len += copy;
1136                         skb->data_len += copy;
1137                 }
1138                 offset += copy;
1139                 length -= copy;
1140         }
1141         return 0;
1142 error:
1143         inet->cork.length -= length;
1144         IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1145         return err;
1146 }
1147
1148 int ip6_push_pending_frames(struct sock *sk)
1149 {
1150         struct sk_buff *skb, *tmp_skb;
1151         struct sk_buff **tail_skb;
1152         struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1153         struct inet_sock *inet = inet_sk(sk);
1154         struct ipv6_pinfo *np = inet6_sk(sk);
1155         struct ipv6hdr *hdr;
1156         struct ipv6_txoptions *opt = np->cork.opt;
1157         struct rt6_info *rt = np->cork.rt;
1158         struct flowi *fl = &inet->cork.fl;
1159         unsigned char proto = fl->proto;
1160         int err = 0;
1161
1162         if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1163                 goto out;
1164         tail_skb = &(skb_shinfo(skb)->frag_list);
1165
1166         /* move skb->data to ip header from ext header */
1167         if (skb->data < skb->nh.raw)
1168                 __skb_pull(skb, skb->nh.raw - skb->data);
1169         while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1170                 __skb_pull(tmp_skb, skb->h.raw - skb->nh.raw);
1171                 *tail_skb = tmp_skb;
1172                 tail_skb = &(tmp_skb->next);
1173                 skb->len += tmp_skb->len;
1174                 skb->data_len += tmp_skb->len;
1175                 skb->truesize += tmp_skb->truesize;
1176                 __sock_put(tmp_skb->sk);
1177                 tmp_skb->destructor = NULL;
1178                 tmp_skb->sk = NULL;
1179         }
1180
1181         ipv6_addr_copy(final_dst, &fl->fl6_dst);
1182         __skb_pull(skb, skb->h.raw - skb->nh.raw);
1183         if (opt && opt->opt_flen)
1184                 ipv6_push_frag_opts(skb, opt, &proto);
1185         if (opt && opt->opt_nflen)
1186                 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1187
1188         skb->nh.ipv6h = hdr = (struct ipv6hdr*) skb_push(skb, sizeof(struct ipv6hdr));
1189         
1190         *(u32*)hdr = fl->fl6_flowlabel |
1191                      htonl(0x60000000 | ((int)np->cork.tclass << 20));
1192
1193         if (skb->len <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN)
1194                 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
1195         else
1196                 hdr->payload_len = 0;
1197         hdr->hop_limit = np->cork.hop_limit;
1198         hdr->nexthdr = proto;
1199         ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
1200         ipv6_addr_copy(&hdr->daddr, final_dst);
1201
1202         skb->priority = sk->sk_priority;
1203
1204         skb->dst = dst_clone(&rt->u.dst);
1205         IP6_INC_STATS(IPSTATS_MIB_OUTREQUESTS); 
1206         err = NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, skb->dst->dev, dst_output);
1207         if (err) {
1208                 if (err > 0)
1209                         err = np->recverr ? net_xmit_errno(err) : 0;
1210                 if (err)
1211                         goto error;
1212         }
1213
1214 out:
1215         inet->cork.flags &= ~IPCORK_OPT;
1216         kfree(np->cork.opt);
1217         np->cork.opt = NULL;
1218         if (np->cork.rt) {
1219                 dst_release(&np->cork.rt->u.dst);
1220                 np->cork.rt = NULL;
1221                 inet->cork.flags &= ~IPCORK_ALLFRAG;
1222         }
1223         memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1224         return err;
1225 error:
1226         goto out;
1227 }
1228
1229 void ip6_flush_pending_frames(struct sock *sk)
1230 {
1231         struct inet_sock *inet = inet_sk(sk);
1232         struct ipv6_pinfo *np = inet6_sk(sk);
1233         struct sk_buff *skb;
1234
1235         while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1236                 IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1237                 kfree_skb(skb);
1238         }
1239
1240         inet->cork.flags &= ~IPCORK_OPT;
1241
1242         kfree(np->cork.opt);
1243         np->cork.opt = NULL;
1244         if (np->cork.rt) {
1245                 dst_release(&np->cork.rt->u.dst);
1246                 np->cork.rt = NULL;
1247                 inet->cork.flags &= ~IPCORK_ALLFRAG;
1248         }
1249         memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1250 }