[SECMARK]: Add secmark support to core networking.
[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         skb_copy_secmark(to, from);
463 }
464
465 int ip6_find_1stfragopt(struct sk_buff *skb, u8 **nexthdr)
466 {
467         u16 offset = sizeof(struct ipv6hdr);
468         struct ipv6_opt_hdr *exthdr = (struct ipv6_opt_hdr*)(skb->nh.ipv6h + 1);
469         unsigned int packet_len = skb->tail - skb->nh.raw;
470         int found_rhdr = 0;
471         *nexthdr = &skb->nh.ipv6h->nexthdr;
472
473         while (offset + 1 <= packet_len) {
474
475                 switch (**nexthdr) {
476
477                 case NEXTHDR_HOP:
478                 case NEXTHDR_ROUTING:
479                 case NEXTHDR_DEST:
480                         if (**nexthdr == NEXTHDR_ROUTING) found_rhdr = 1;
481                         if (**nexthdr == NEXTHDR_DEST && found_rhdr) return offset;
482                         offset += ipv6_optlen(exthdr);
483                         *nexthdr = &exthdr->nexthdr;
484                         exthdr = (struct ipv6_opt_hdr*)(skb->nh.raw + offset);
485                         break;
486                 default :
487                         return offset;
488                 }
489         }
490
491         return offset;
492 }
493 EXPORT_SYMBOL_GPL(ip6_find_1stfragopt);
494
495 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
496 {
497         struct net_device *dev;
498         struct sk_buff *frag;
499         struct rt6_info *rt = (struct rt6_info*)skb->dst;
500         struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
501         struct ipv6hdr *tmp_hdr;
502         struct frag_hdr *fh;
503         unsigned int mtu, hlen, left, len;
504         u32 frag_id = 0;
505         int ptr, offset = 0, err=0;
506         u8 *prevhdr, nexthdr = 0;
507
508         dev = rt->u.dst.dev;
509         hlen = ip6_find_1stfragopt(skb, &prevhdr);
510         nexthdr = *prevhdr;
511
512         mtu = dst_mtu(&rt->u.dst);
513         if (np && np->frag_size < mtu) {
514                 if (np->frag_size)
515                         mtu = np->frag_size;
516         }
517         mtu -= hlen + sizeof(struct frag_hdr);
518
519         if (skb_shinfo(skb)->frag_list) {
520                 int first_len = skb_pagelen(skb);
521
522                 if (first_len - hlen > mtu ||
523                     ((first_len - hlen) & 7) ||
524                     skb_cloned(skb))
525                         goto slow_path;
526
527                 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
528                         /* Correct geometry. */
529                         if (frag->len > mtu ||
530                             ((frag->len & 7) && frag->next) ||
531                             skb_headroom(frag) < hlen)
532                             goto slow_path;
533
534                         /* Partially cloned skb? */
535                         if (skb_shared(frag))
536                                 goto slow_path;
537
538                         BUG_ON(frag->sk);
539                         if (skb->sk) {
540                                 sock_hold(skb->sk);
541                                 frag->sk = skb->sk;
542                                 frag->destructor = sock_wfree;
543                                 skb->truesize -= frag->truesize;
544                         }
545                 }
546
547                 err = 0;
548                 offset = 0;
549                 frag = skb_shinfo(skb)->frag_list;
550                 skb_shinfo(skb)->frag_list = NULL;
551                 /* BUILD HEADER */
552
553                 tmp_hdr = kmalloc(hlen, GFP_ATOMIC);
554                 if (!tmp_hdr) {
555                         IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
556                         return -ENOMEM;
557                 }
558
559                 *prevhdr = NEXTHDR_FRAGMENT;
560                 memcpy(tmp_hdr, skb->nh.raw, hlen);
561                 __skb_pull(skb, hlen);
562                 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
563                 skb->nh.raw = __skb_push(skb, hlen);
564                 memcpy(skb->nh.raw, tmp_hdr, hlen);
565
566                 ipv6_select_ident(skb, fh);
567                 fh->nexthdr = nexthdr;
568                 fh->reserved = 0;
569                 fh->frag_off = htons(IP6_MF);
570                 frag_id = fh->identification;
571
572                 first_len = skb_pagelen(skb);
573                 skb->data_len = first_len - skb_headlen(skb);
574                 skb->len = first_len;
575                 skb->nh.ipv6h->payload_len = htons(first_len - sizeof(struct ipv6hdr));
576  
577
578                 for (;;) {
579                         /* Prepare header of the next frame,
580                          * before previous one went down. */
581                         if (frag) {
582                                 frag->ip_summed = CHECKSUM_NONE;
583                                 frag->h.raw = frag->data;
584                                 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
585                                 frag->nh.raw = __skb_push(frag, hlen);
586                                 memcpy(frag->nh.raw, tmp_hdr, hlen);
587                                 offset += skb->len - hlen - sizeof(struct frag_hdr);
588                                 fh->nexthdr = nexthdr;
589                                 fh->reserved = 0;
590                                 fh->frag_off = htons(offset);
591                                 if (frag->next != NULL)
592                                         fh->frag_off |= htons(IP6_MF);
593                                 fh->identification = frag_id;
594                                 frag->nh.ipv6h->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
595                                 ip6_copy_metadata(frag, skb);
596                         }
597                         
598                         err = output(skb);
599                         if (err || !frag)
600                                 break;
601
602                         skb = frag;
603                         frag = skb->next;
604                         skb->next = NULL;
605                 }
606
607                 kfree(tmp_hdr);
608
609                 if (err == 0) {
610                         IP6_INC_STATS(IPSTATS_MIB_FRAGOKS);
611                         return 0;
612                 }
613
614                 while (frag) {
615                         skb = frag->next;
616                         kfree_skb(frag);
617                         frag = skb;
618                 }
619
620                 IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
621                 return err;
622         }
623
624 slow_path:
625         left = skb->len - hlen;         /* Space per frame */
626         ptr = hlen;                     /* Where to start from */
627
628         /*
629          *      Fragment the datagram.
630          */
631
632         *prevhdr = NEXTHDR_FRAGMENT;
633
634         /*
635          *      Keep copying data until we run out.
636          */
637         while(left > 0) {
638                 len = left;
639                 /* IF: it doesn't fit, use 'mtu' - the data space left */
640                 if (len > mtu)
641                         len = mtu;
642                 /* IF: we are not sending upto and including the packet end
643                    then align the next start on an eight byte boundary */
644                 if (len < left) {
645                         len &= ~7;
646                 }
647                 /*
648                  *      Allocate buffer.
649                  */
650
651                 if ((frag = alloc_skb(len+hlen+sizeof(struct frag_hdr)+LL_RESERVED_SPACE(rt->u.dst.dev), GFP_ATOMIC)) == NULL) {
652                         NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
653                         IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
654                         err = -ENOMEM;
655                         goto fail;
656                 }
657
658                 /*
659                  *      Set up data on packet
660                  */
661
662                 ip6_copy_metadata(frag, skb);
663                 skb_reserve(frag, LL_RESERVED_SPACE(rt->u.dst.dev));
664                 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
665                 frag->nh.raw = frag->data;
666                 fh = (struct frag_hdr*)(frag->data + hlen);
667                 frag->h.raw = frag->data + hlen + sizeof(struct frag_hdr);
668
669                 /*
670                  *      Charge the memory for the fragment to any owner
671                  *      it might possess
672                  */
673                 if (skb->sk)
674                         skb_set_owner_w(frag, skb->sk);
675
676                 /*
677                  *      Copy the packet header into the new buffer.
678                  */
679                 memcpy(frag->nh.raw, skb->data, hlen);
680
681                 /*
682                  *      Build fragment header.
683                  */
684                 fh->nexthdr = nexthdr;
685                 fh->reserved = 0;
686                 if (!frag_id) {
687                         ipv6_select_ident(skb, fh);
688                         frag_id = fh->identification;
689                 } else
690                         fh->identification = frag_id;
691
692                 /*
693                  *      Copy a block of the IP datagram.
694                  */
695                 if (skb_copy_bits(skb, ptr, frag->h.raw, len))
696                         BUG();
697                 left -= len;
698
699                 fh->frag_off = htons(offset);
700                 if (left > 0)
701                         fh->frag_off |= htons(IP6_MF);
702                 frag->nh.ipv6h->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
703
704                 ptr += len;
705                 offset += len;
706
707                 /*
708                  *      Put this fragment into the sending queue.
709                  */
710
711                 IP6_INC_STATS(IPSTATS_MIB_FRAGCREATES);
712
713                 err = output(frag);
714                 if (err)
715                         goto fail;
716         }
717         kfree_skb(skb);
718         IP6_INC_STATS(IPSTATS_MIB_FRAGOKS);
719         return err;
720
721 fail:
722         kfree_skb(skb); 
723         IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
724         return err;
725 }
726
727 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
728 {
729         int err = 0;
730
731         *dst = NULL;
732         if (sk) {
733                 struct ipv6_pinfo *np = inet6_sk(sk);
734         
735                 *dst = sk_dst_check(sk, np->dst_cookie);
736                 if (*dst) {
737                         struct rt6_info *rt = (struct rt6_info*)*dst;
738         
739                         /* Yes, checking route validity in not connected
740                          * case is not very simple. Take into account,
741                          * that we do not support routing by source, TOS,
742                          * and MSG_DONTROUTE            --ANK (980726)
743                          *
744                          * 1. If route was host route, check that
745                          *    cached destination is current.
746                          *    If it is network route, we still may
747                          *    check its validity using saved pointer
748                          *    to the last used address: daddr_cache.
749                          *    We do not want to save whole address now,
750                          *    (because main consumer of this service
751                          *    is tcp, which has not this problem),
752                          *    so that the last trick works only on connected
753                          *    sockets.
754                          * 2. oif also should be the same.
755                          */
756                         if (((rt->rt6i_dst.plen != 128 ||
757                               !ipv6_addr_equal(&fl->fl6_dst,
758                                                &rt->rt6i_dst.addr))
759                              && (np->daddr_cache == NULL ||
760                                  !ipv6_addr_equal(&fl->fl6_dst,
761                                                   np->daddr_cache)))
762                             || (fl->oif && fl->oif != (*dst)->dev->ifindex)) {
763                                 dst_release(*dst);
764                                 *dst = NULL;
765                         }
766                 }
767         }
768
769         if (*dst == NULL)
770                 *dst = ip6_route_output(sk, fl);
771
772         if ((err = (*dst)->error))
773                 goto out_err_release;
774
775         if (ipv6_addr_any(&fl->fl6_src)) {
776                 err = ipv6_get_saddr(*dst, &fl->fl6_dst, &fl->fl6_src);
777
778                 if (err)
779                         goto out_err_release;
780         }
781
782         return 0;
783
784 out_err_release:
785         dst_release(*dst);
786         *dst = NULL;
787         return err;
788 }
789
790 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
791
792 static inline int ip6_ufo_append_data(struct sock *sk,
793                         int getfrag(void *from, char *to, int offset, int len,
794                         int odd, struct sk_buff *skb),
795                         void *from, int length, int hh_len, int fragheaderlen,
796                         int transhdrlen, int mtu,unsigned int flags)
797
798 {
799         struct sk_buff *skb;
800         int err;
801
802         /* There is support for UDP large send offload by network
803          * device, so create one single skb packet containing complete
804          * udp datagram
805          */
806         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
807                 skb = sock_alloc_send_skb(sk,
808                         hh_len + fragheaderlen + transhdrlen + 20,
809                         (flags & MSG_DONTWAIT), &err);
810                 if (skb == NULL)
811                         return -ENOMEM;
812
813                 /* reserve space for Hardware header */
814                 skb_reserve(skb, hh_len);
815
816                 /* create space for UDP/IP header */
817                 skb_put(skb,fragheaderlen + transhdrlen);
818
819                 /* initialize network header pointer */
820                 skb->nh.raw = skb->data;
821
822                 /* initialize protocol header pointer */
823                 skb->h.raw = skb->data + fragheaderlen;
824
825                 skb->ip_summed = CHECKSUM_HW;
826                 skb->csum = 0;
827                 sk->sk_sndmsg_off = 0;
828         }
829
830         err = skb_append_datato_frags(sk,skb, getfrag, from,
831                                       (length - transhdrlen));
832         if (!err) {
833                 struct frag_hdr fhdr;
834
835                 /* specify the length of each IP datagram fragment*/
836                 skb_shinfo(skb)->ufo_size = (mtu - fragheaderlen) - 
837                                                 sizeof(struct frag_hdr);
838                 ipv6_select_ident(skb, &fhdr);
839                 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
840                 __skb_queue_tail(&sk->sk_write_queue, skb);
841
842                 return 0;
843         }
844         /* There is not enough support do UPD LSO,
845          * so follow normal path
846          */
847         kfree_skb(skb);
848
849         return err;
850 }
851
852 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
853         int offset, int len, int odd, struct sk_buff *skb),
854         void *from, int length, int transhdrlen,
855         int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi *fl,
856         struct rt6_info *rt, unsigned int flags)
857 {
858         struct inet_sock *inet = inet_sk(sk);
859         struct ipv6_pinfo *np = inet6_sk(sk);
860         struct sk_buff *skb;
861         unsigned int maxfraglen, fragheaderlen;
862         int exthdrlen;
863         int hh_len;
864         int mtu;
865         int copy;
866         int err;
867         int offset = 0;
868         int csummode = CHECKSUM_NONE;
869
870         if (flags&MSG_PROBE)
871                 return 0;
872         if (skb_queue_empty(&sk->sk_write_queue)) {
873                 /*
874                  * setup for corking
875                  */
876                 if (opt) {
877                         if (np->cork.opt == NULL) {
878                                 np->cork.opt = kmalloc(opt->tot_len,
879                                                        sk->sk_allocation);
880                                 if (unlikely(np->cork.opt == NULL))
881                                         return -ENOBUFS;
882                         } else if (np->cork.opt->tot_len < opt->tot_len) {
883                                 printk(KERN_DEBUG "ip6_append_data: invalid option length\n");
884                                 return -EINVAL;
885                         }
886                         memcpy(np->cork.opt, opt, opt->tot_len);
887                         inet->cork.flags |= IPCORK_OPT;
888                         /* need source address above miyazawa*/
889                 }
890                 dst_hold(&rt->u.dst);
891                 np->cork.rt = rt;
892                 inet->cork.fl = *fl;
893                 np->cork.hop_limit = hlimit;
894                 np->cork.tclass = tclass;
895                 mtu = dst_mtu(rt->u.dst.path);
896                 if (np->frag_size < mtu) {
897                         if (np->frag_size)
898                                 mtu = np->frag_size;
899                 }
900                 inet->cork.fragsize = mtu;
901                 if (dst_allfrag(rt->u.dst.path))
902                         inet->cork.flags |= IPCORK_ALLFRAG;
903                 inet->cork.length = 0;
904                 sk->sk_sndmsg_page = NULL;
905                 sk->sk_sndmsg_off = 0;
906                 exthdrlen = rt->u.dst.header_len + (opt ? opt->opt_flen : 0);
907                 length += exthdrlen;
908                 transhdrlen += exthdrlen;
909         } else {
910                 rt = np->cork.rt;
911                 fl = &inet->cork.fl;
912                 if (inet->cork.flags & IPCORK_OPT)
913                         opt = np->cork.opt;
914                 transhdrlen = 0;
915                 exthdrlen = 0;
916                 mtu = inet->cork.fragsize;
917         }
918
919         hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
920
921         fragheaderlen = sizeof(struct ipv6hdr) + (opt ? opt->opt_nflen : 0);
922         maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
923
924         if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
925                 if (inet->cork.length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
926                         ipv6_local_error(sk, EMSGSIZE, fl, mtu-exthdrlen);
927                         return -EMSGSIZE;
928                 }
929         }
930
931         /*
932          * Let's try using as much space as possible.
933          * Use MTU if total length of the message fits into the MTU.
934          * Otherwise, we need to reserve fragment header and
935          * fragment alignment (= 8-15 octects, in total).
936          *
937          * Note that we may need to "move" the data from the tail of
938          * of the buffer to the new fragment when we split 
939          * the message.
940          *
941          * FIXME: It may be fragmented into multiple chunks 
942          *        at once if non-fragmentable extension headers
943          *        are too large.
944          * --yoshfuji 
945          */
946
947         inet->cork.length += length;
948         if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
949             (rt->u.dst.dev->features & NETIF_F_UFO)) {
950
951                 err = ip6_ufo_append_data(sk, getfrag, from, length, hh_len,
952                                           fragheaderlen, transhdrlen, mtu,
953                                           flags);
954                 if (err)
955                         goto error;
956                 return 0;
957         }
958
959         if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
960                 goto alloc_new_skb;
961
962         while (length > 0) {
963                 /* Check if the remaining data fits into current packet. */
964                 copy = (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
965                 if (copy < length)
966                         copy = maxfraglen - skb->len;
967
968                 if (copy <= 0) {
969                         char *data;
970                         unsigned int datalen;
971                         unsigned int fraglen;
972                         unsigned int fraggap;
973                         unsigned int alloclen;
974                         struct sk_buff *skb_prev;
975 alloc_new_skb:
976                         skb_prev = skb;
977
978                         /* There's no room in the current skb */
979                         if (skb_prev)
980                                 fraggap = skb_prev->len - maxfraglen;
981                         else
982                                 fraggap = 0;
983
984                         /*
985                          * If remaining data exceeds the mtu,
986                          * we know we need more fragment(s).
987                          */
988                         datalen = length + fraggap;
989                         if (datalen > (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
990                                 datalen = maxfraglen - fragheaderlen;
991
992                         fraglen = datalen + fragheaderlen;
993                         if ((flags & MSG_MORE) &&
994                             !(rt->u.dst.dev->features&NETIF_F_SG))
995                                 alloclen = mtu;
996                         else
997                                 alloclen = datalen + fragheaderlen;
998
999                         /*
1000                          * The last fragment gets additional space at tail.
1001                          * Note: we overallocate on fragments with MSG_MODE
1002                          * because we have no idea if we're the last one.
1003                          */
1004                         if (datalen == length + fraggap)
1005                                 alloclen += rt->u.dst.trailer_len;
1006
1007                         /*
1008                          * We just reserve space for fragment header.
1009                          * Note: this may be overallocation if the message 
1010                          * (without MSG_MORE) fits into the MTU.
1011                          */
1012                         alloclen += sizeof(struct frag_hdr);
1013
1014                         if (transhdrlen) {
1015                                 skb = sock_alloc_send_skb(sk,
1016                                                 alloclen + hh_len,
1017                                                 (flags & MSG_DONTWAIT), &err);
1018                         } else {
1019                                 skb = NULL;
1020                                 if (atomic_read(&sk->sk_wmem_alloc) <=
1021                                     2 * sk->sk_sndbuf)
1022                                         skb = sock_wmalloc(sk,
1023                                                            alloclen + hh_len, 1,
1024                                                            sk->sk_allocation);
1025                                 if (unlikely(skb == NULL))
1026                                         err = -ENOBUFS;
1027                         }
1028                         if (skb == NULL)
1029                                 goto error;
1030                         /*
1031                          *      Fill in the control structures
1032                          */
1033                         skb->ip_summed = csummode;
1034                         skb->csum = 0;
1035                         /* reserve for fragmentation */
1036                         skb_reserve(skb, hh_len+sizeof(struct frag_hdr));
1037
1038                         /*
1039                          *      Find where to start putting bytes
1040                          */
1041                         data = skb_put(skb, fraglen);
1042                         skb->nh.raw = data + exthdrlen;
1043                         data += fragheaderlen;
1044                         skb->h.raw = data + exthdrlen;
1045
1046                         if (fraggap) {
1047                                 skb->csum = skb_copy_and_csum_bits(
1048                                         skb_prev, maxfraglen,
1049                                         data + transhdrlen, fraggap, 0);
1050                                 skb_prev->csum = csum_sub(skb_prev->csum,
1051                                                           skb->csum);
1052                                 data += fraggap;
1053                                 skb_trim(skb_prev, maxfraglen);
1054                         }
1055                         copy = datalen - transhdrlen - fraggap;
1056                         if (copy < 0) {
1057                                 err = -EINVAL;
1058                                 kfree_skb(skb);
1059                                 goto error;
1060                         } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1061                                 err = -EFAULT;
1062                                 kfree_skb(skb);
1063                                 goto error;
1064                         }
1065
1066                         offset += copy;
1067                         length -= datalen - fraggap;
1068                         transhdrlen = 0;
1069                         exthdrlen = 0;
1070                         csummode = CHECKSUM_NONE;
1071
1072                         /*
1073                          * Put the packet on the pending queue
1074                          */
1075                         __skb_queue_tail(&sk->sk_write_queue, skb);
1076                         continue;
1077                 }
1078
1079                 if (copy > length)
1080                         copy = length;
1081
1082                 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
1083                         unsigned int off;
1084
1085                         off = skb->len;
1086                         if (getfrag(from, skb_put(skb, copy),
1087                                                 offset, copy, off, skb) < 0) {
1088                                 __skb_trim(skb, off);
1089                                 err = -EFAULT;
1090                                 goto error;
1091                         }
1092                 } else {
1093                         int i = skb_shinfo(skb)->nr_frags;
1094                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1095                         struct page *page = sk->sk_sndmsg_page;
1096                         int off = sk->sk_sndmsg_off;
1097                         unsigned int left;
1098
1099                         if (page && (left = PAGE_SIZE - off) > 0) {
1100                                 if (copy >= left)
1101                                         copy = left;
1102                                 if (page != frag->page) {
1103                                         if (i == MAX_SKB_FRAGS) {
1104                                                 err = -EMSGSIZE;
1105                                                 goto error;
1106                                         }
1107                                         get_page(page);
1108                                         skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1109                                         frag = &skb_shinfo(skb)->frags[i];
1110                                 }
1111                         } else if(i < MAX_SKB_FRAGS) {
1112                                 if (copy > PAGE_SIZE)
1113                                         copy = PAGE_SIZE;
1114                                 page = alloc_pages(sk->sk_allocation, 0);
1115                                 if (page == NULL) {
1116                                         err = -ENOMEM;
1117                                         goto error;
1118                                 }
1119                                 sk->sk_sndmsg_page = page;
1120                                 sk->sk_sndmsg_off = 0;
1121
1122                                 skb_fill_page_desc(skb, i, page, 0, 0);
1123                                 frag = &skb_shinfo(skb)->frags[i];
1124                                 skb->truesize += PAGE_SIZE;
1125                                 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1126                         } else {
1127                                 err = -EMSGSIZE;
1128                                 goto error;
1129                         }
1130                         if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1131                                 err = -EFAULT;
1132                                 goto error;
1133                         }
1134                         sk->sk_sndmsg_off += copy;
1135                         frag->size += copy;
1136                         skb->len += copy;
1137                         skb->data_len += copy;
1138                 }
1139                 offset += copy;
1140                 length -= copy;
1141         }
1142         return 0;
1143 error:
1144         inet->cork.length -= length;
1145         IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1146         return err;
1147 }
1148
1149 int ip6_push_pending_frames(struct sock *sk)
1150 {
1151         struct sk_buff *skb, *tmp_skb;
1152         struct sk_buff **tail_skb;
1153         struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1154         struct inet_sock *inet = inet_sk(sk);
1155         struct ipv6_pinfo *np = inet6_sk(sk);
1156         struct ipv6hdr *hdr;
1157         struct ipv6_txoptions *opt = np->cork.opt;
1158         struct rt6_info *rt = np->cork.rt;
1159         struct flowi *fl = &inet->cork.fl;
1160         unsigned char proto = fl->proto;
1161         int err = 0;
1162
1163         if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1164                 goto out;
1165         tail_skb = &(skb_shinfo(skb)->frag_list);
1166
1167         /* move skb->data to ip header from ext header */
1168         if (skb->data < skb->nh.raw)
1169                 __skb_pull(skb, skb->nh.raw - skb->data);
1170         while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1171                 __skb_pull(tmp_skb, skb->h.raw - skb->nh.raw);
1172                 *tail_skb = tmp_skb;
1173                 tail_skb = &(tmp_skb->next);
1174                 skb->len += tmp_skb->len;
1175                 skb->data_len += tmp_skb->len;
1176                 skb->truesize += tmp_skb->truesize;
1177                 __sock_put(tmp_skb->sk);
1178                 tmp_skb->destructor = NULL;
1179                 tmp_skb->sk = NULL;
1180         }
1181
1182         ipv6_addr_copy(final_dst, &fl->fl6_dst);
1183         __skb_pull(skb, skb->h.raw - skb->nh.raw);
1184         if (opt && opt->opt_flen)
1185                 ipv6_push_frag_opts(skb, opt, &proto);
1186         if (opt && opt->opt_nflen)
1187                 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1188
1189         skb->nh.ipv6h = hdr = (struct ipv6hdr*) skb_push(skb, sizeof(struct ipv6hdr));
1190         
1191         *(u32*)hdr = fl->fl6_flowlabel |
1192                      htonl(0x60000000 | ((int)np->cork.tclass << 20));
1193
1194         if (skb->len <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN)
1195                 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
1196         else
1197                 hdr->payload_len = 0;
1198         hdr->hop_limit = np->cork.hop_limit;
1199         hdr->nexthdr = proto;
1200         ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
1201         ipv6_addr_copy(&hdr->daddr, final_dst);
1202
1203         skb->priority = sk->sk_priority;
1204
1205         skb->dst = dst_clone(&rt->u.dst);
1206         IP6_INC_STATS(IPSTATS_MIB_OUTREQUESTS); 
1207         err = NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, skb->dst->dev, dst_output);
1208         if (err) {
1209                 if (err > 0)
1210                         err = np->recverr ? net_xmit_errno(err) : 0;
1211                 if (err)
1212                         goto error;
1213         }
1214
1215 out:
1216         inet->cork.flags &= ~IPCORK_OPT;
1217         kfree(np->cork.opt);
1218         np->cork.opt = NULL;
1219         if (np->cork.rt) {
1220                 dst_release(&np->cork.rt->u.dst);
1221                 np->cork.rt = NULL;
1222                 inet->cork.flags &= ~IPCORK_ALLFRAG;
1223         }
1224         memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1225         return err;
1226 error:
1227         goto out;
1228 }
1229
1230 void ip6_flush_pending_frames(struct sock *sk)
1231 {
1232         struct inet_sock *inet = inet_sk(sk);
1233         struct ipv6_pinfo *np = inet6_sk(sk);
1234         struct sk_buff *skb;
1235
1236         while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1237                 IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1238                 kfree_skb(skb);
1239         }
1240
1241         inet->cork.flags &= ~IPCORK_OPT;
1242
1243         kfree(np->cork.opt);
1244         np->cork.opt = NULL;
1245         if (np->cork.rt) {
1246                 dst_release(&np->cork.rt->u.dst);
1247                 np->cork.rt = NULL;
1248                 inet->cork.flags &= ~IPCORK_ALLFRAG;
1249         }
1250         memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1251 }