Merge master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6
[linux/fpc-iii.git] / net / ipv6 / ip6_output.c
blob66716911962eb967aa487b964895548c3542ea8c
1 /*
2 * IPv6 output functions
3 * Linux INET6 implementation
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
8 * $Id: ip6_output.c,v 1.34 2002/02/01 22:01:04 davem Exp $
10 * Based on linux/net/ipv4/ip_output.c
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.
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.
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
31 #include <linux/errno.h>
32 #include <linux/types.h>
33 #include <linux/string.h>
34 #include <linux/socket.h>
35 #include <linux/net.h>
36 #include <linux/netdevice.h>
37 #include <linux/if_arp.h>
38 #include <linux/in6.h>
39 #include <linux/tcp.h>
40 #include <linux/route.h>
41 #include <linux/module.h>
43 #include <linux/netfilter.h>
44 #include <linux/netfilter_ipv6.h>
46 #include <net/sock.h>
47 #include <net/snmp.h>
49 #include <net/ipv6.h>
50 #include <net/ndisc.h>
51 #include <net/protocol.h>
52 #include <net/ip6_route.h>
53 #include <net/addrconf.h>
54 #include <net/rawv6.h>
55 #include <net/icmp.h>
56 #include <net/xfrm.h>
57 #include <net/checksum.h>
59 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *));
61 static __inline__ void ipv6_select_ident(struct sk_buff *skb, struct frag_hdr *fhdr)
63 static u32 ipv6_fragmentation_id = 1;
64 static DEFINE_SPINLOCK(ip6_id_lock);
66 spin_lock_bh(&ip6_id_lock);
67 fhdr->identification = htonl(ipv6_fragmentation_id);
68 if (++ipv6_fragmentation_id == 0)
69 ipv6_fragmentation_id = 1;
70 spin_unlock_bh(&ip6_id_lock);
73 static inline int ip6_output_finish(struct sk_buff *skb)
76 struct dst_entry *dst = skb->dst;
77 struct hh_cache *hh = dst->hh;
79 if (hh) {
80 int hh_alen;
82 read_lock_bh(&hh->hh_lock);
83 hh_alen = HH_DATA_ALIGN(hh->hh_len);
84 memcpy(skb->data - hh_alen, hh->hh_data, hh_alen);
85 read_unlock_bh(&hh->hh_lock);
86 skb_push(skb, hh->hh_len);
87 return hh->hh_output(skb);
88 } else if (dst->neighbour)
89 return dst->neighbour->output(skb);
91 IP6_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
92 kfree_skb(skb);
93 return -EINVAL;
97 /* dev_loopback_xmit for use with netfilter. */
98 static int ip6_dev_loopback_xmit(struct sk_buff *newskb)
100 newskb->mac.raw = newskb->data;
101 __skb_pull(newskb, newskb->nh.raw - newskb->data);
102 newskb->pkt_type = PACKET_LOOPBACK;
103 newskb->ip_summed = CHECKSUM_UNNECESSARY;
104 BUG_TRAP(newskb->dst);
106 netif_rx(newskb);
107 return 0;
111 static int ip6_output2(struct sk_buff *skb)
113 struct dst_entry *dst = skb->dst;
114 struct net_device *dev = dst->dev;
116 skb->protocol = htons(ETH_P_IPV6);
117 skb->dev = dev;
119 if (ipv6_addr_is_multicast(&skb->nh.ipv6h->daddr)) {
120 struct ipv6_pinfo* np = skb->sk ? inet6_sk(skb->sk) : NULL;
122 if (!(dev->flags & IFF_LOOPBACK) && (!np || np->mc_loop) &&
123 ipv6_chk_mcast_addr(dev, &skb->nh.ipv6h->daddr,
124 &skb->nh.ipv6h->saddr)) {
125 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
127 /* Do not check for IFF_ALLMULTI; multicast routing
128 is not supported in any case.
130 if (newskb)
131 NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, newskb, NULL,
132 newskb->dev,
133 ip6_dev_loopback_xmit);
135 if (skb->nh.ipv6h->hop_limit == 0) {
136 IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
137 kfree_skb(skb);
138 return 0;
142 IP6_INC_STATS(IPSTATS_MIB_OUTMCASTPKTS);
145 return NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, skb,NULL, skb->dev,ip6_output_finish);
148 int ip6_output(struct sk_buff *skb)
150 if ((skb->len > dst_mtu(skb->dst) && !skb_is_gso(skb)) ||
151 dst_allfrag(skb->dst))
152 return ip6_fragment(skb, ip6_output2);
153 else
154 return ip6_output2(skb);
158 * xmit an sk_buff (used by TCP)
161 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
162 struct ipv6_txoptions *opt, int ipfragok)
164 struct ipv6_pinfo *np = inet6_sk(sk);
165 struct in6_addr *first_hop = &fl->fl6_dst;
166 struct dst_entry *dst = skb->dst;
167 struct ipv6hdr *hdr;
168 u8 proto = fl->proto;
169 int seg_len = skb->len;
170 int hlimit, tclass;
171 u32 mtu;
173 if (opt) {
174 int head_room;
176 /* First: exthdrs may take lots of space (~8K for now)
177 MAX_HEADER is not enough.
179 head_room = opt->opt_nflen + opt->opt_flen;
180 seg_len += head_room;
181 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
183 if (skb_headroom(skb) < head_room) {
184 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
185 kfree_skb(skb);
186 skb = skb2;
187 if (skb == NULL) {
188 IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
189 return -ENOBUFS;
191 if (sk)
192 skb_set_owner_w(skb, sk);
194 if (opt->opt_flen)
195 ipv6_push_frag_opts(skb, opt, &proto);
196 if (opt->opt_nflen)
197 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
200 hdr = skb->nh.ipv6h = (struct ipv6hdr*)skb_push(skb, sizeof(struct ipv6hdr));
203 * Fill in the IPv6 header
206 hlimit = -1;
207 if (np)
208 hlimit = np->hop_limit;
209 if (hlimit < 0)
210 hlimit = dst_metric(dst, RTAX_HOPLIMIT);
211 if (hlimit < 0)
212 hlimit = ipv6_get_hoplimit(dst->dev);
214 tclass = -1;
215 if (np)
216 tclass = np->tclass;
217 if (tclass < 0)
218 tclass = 0;
220 *(u32 *)hdr = htonl(0x60000000 | (tclass << 20)) | fl->fl6_flowlabel;
222 hdr->payload_len = htons(seg_len);
223 hdr->nexthdr = proto;
224 hdr->hop_limit = hlimit;
226 ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
227 ipv6_addr_copy(&hdr->daddr, first_hop);
229 skb->priority = sk->sk_priority;
231 mtu = dst_mtu(dst);
232 if ((skb->len <= mtu) || ipfragok || skb_is_gso(skb)) {
233 IP6_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
234 return NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, dst->dev,
235 dst_output);
238 if (net_ratelimit())
239 printk(KERN_DEBUG "IPv6: sending pkt_too_big to self\n");
240 skb->dev = dst->dev;
241 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev);
242 IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
243 kfree_skb(skb);
244 return -EMSGSIZE;
248 * To avoid extra problems ND packets are send through this
249 * routine. It's code duplication but I really want to avoid
250 * extra checks since ipv6_build_header is used by TCP (which
251 * is for us performance critical)
254 int ip6_nd_hdr(struct sock *sk, struct sk_buff *skb, struct net_device *dev,
255 struct in6_addr *saddr, struct in6_addr *daddr,
256 int proto, int len)
258 struct ipv6_pinfo *np = inet6_sk(sk);
259 struct ipv6hdr *hdr;
260 int totlen;
262 skb->protocol = htons(ETH_P_IPV6);
263 skb->dev = dev;
265 totlen = len + sizeof(struct ipv6hdr);
267 hdr = (struct ipv6hdr *) skb_put(skb, sizeof(struct ipv6hdr));
268 skb->nh.ipv6h = hdr;
270 *(u32*)hdr = htonl(0x60000000);
272 hdr->payload_len = htons(len);
273 hdr->nexthdr = proto;
274 hdr->hop_limit = np->hop_limit;
276 ipv6_addr_copy(&hdr->saddr, saddr);
277 ipv6_addr_copy(&hdr->daddr, daddr);
279 return 0;
282 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
284 struct ip6_ra_chain *ra;
285 struct sock *last = NULL;
287 read_lock(&ip6_ra_lock);
288 for (ra = ip6_ra_chain; ra; ra = ra->next) {
289 struct sock *sk = ra->sk;
290 if (sk && ra->sel == sel &&
291 (!sk->sk_bound_dev_if ||
292 sk->sk_bound_dev_if == skb->dev->ifindex)) {
293 if (last) {
294 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
295 if (skb2)
296 rawv6_rcv(last, skb2);
298 last = sk;
302 if (last) {
303 rawv6_rcv(last, skb);
304 read_unlock(&ip6_ra_lock);
305 return 1;
307 read_unlock(&ip6_ra_lock);
308 return 0;
311 static int ip6_forward_proxy_check(struct sk_buff *skb)
313 struct ipv6hdr *hdr = skb->nh.ipv6h;
314 u8 nexthdr = hdr->nexthdr;
315 int offset;
317 if (ipv6_ext_hdr(nexthdr)) {
318 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr);
319 if (offset < 0)
320 return 0;
321 } else
322 offset = sizeof(struct ipv6hdr);
324 if (nexthdr == IPPROTO_ICMPV6) {
325 struct icmp6hdr *icmp6;
327 if (!pskb_may_pull(skb, skb->nh.raw + offset + 1 - skb->data))
328 return 0;
330 icmp6 = (struct icmp6hdr *)(skb->nh.raw + offset);
332 switch (icmp6->icmp6_type) {
333 case NDISC_ROUTER_SOLICITATION:
334 case NDISC_ROUTER_ADVERTISEMENT:
335 case NDISC_NEIGHBOUR_SOLICITATION:
336 case NDISC_NEIGHBOUR_ADVERTISEMENT:
337 case NDISC_REDIRECT:
338 /* For reaction involving unicast neighbor discovery
339 * message destined to the proxied address, pass it to
340 * input function.
342 return 1;
343 default:
344 break;
349 * The proxying router can't forward traffic sent to a link-local
350 * address, so signal the sender and discard the packet. This
351 * behavior is clarified by the MIPv6 specification.
353 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
354 dst_link_failure(skb);
355 return -1;
358 return 0;
361 static inline int ip6_forward_finish(struct sk_buff *skb)
363 return dst_output(skb);
366 int ip6_forward(struct sk_buff *skb)
368 struct dst_entry *dst = skb->dst;
369 struct ipv6hdr *hdr = skb->nh.ipv6h;
370 struct inet6_skb_parm *opt = IP6CB(skb);
372 if (ipv6_devconf.forwarding == 0)
373 goto error;
375 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
376 IP6_INC_STATS(IPSTATS_MIB_INDISCARDS);
377 goto drop;
380 skb->ip_summed = CHECKSUM_NONE;
383 * We DO NOT make any processing on
384 * RA packets, pushing them to user level AS IS
385 * without ane WARRANTY that application will be able
386 * to interpret them. The reason is that we
387 * cannot make anything clever here.
389 * We are not end-node, so that if packet contains
390 * AH/ESP, we cannot make anything.
391 * Defragmentation also would be mistake, RA packets
392 * cannot be fragmented, because there is no warranty
393 * that different fragments will go along one path. --ANK
395 if (opt->ra) {
396 u8 *ptr = skb->nh.raw + opt->ra;
397 if (ip6_call_ra_chain(skb, (ptr[2]<<8) + ptr[3]))
398 return 0;
402 * check and decrement ttl
404 if (hdr->hop_limit <= 1) {
405 /* Force OUTPUT device used as source address */
406 skb->dev = dst->dev;
407 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT,
408 0, skb->dev);
409 IP6_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS);
411 kfree_skb(skb);
412 return -ETIMEDOUT;
415 /* XXX: idev->cnf.proxy_ndp? */
416 if (ipv6_devconf.proxy_ndp &&
417 pneigh_lookup(&nd_tbl, &hdr->daddr, skb->dev, 0)) {
418 int proxied = ip6_forward_proxy_check(skb);
419 if (proxied > 0)
420 return ip6_input(skb);
421 else if (proxied < 0) {
422 IP6_INC_STATS(IPSTATS_MIB_INDISCARDS);
423 goto drop;
427 if (!xfrm6_route_forward(skb)) {
428 IP6_INC_STATS(IPSTATS_MIB_INDISCARDS);
429 goto drop;
431 dst = skb->dst;
433 /* IPv6 specs say nothing about it, but it is clear that we cannot
434 send redirects to source routed frames.
436 if (skb->dev == dst->dev && dst->neighbour && opt->srcrt == 0) {
437 struct in6_addr *target = NULL;
438 struct rt6_info *rt;
439 struct neighbour *n = dst->neighbour;
442 * incoming and outgoing devices are the same
443 * send a redirect.
446 rt = (struct rt6_info *) dst;
447 if ((rt->rt6i_flags & RTF_GATEWAY))
448 target = (struct in6_addr*)&n->primary_key;
449 else
450 target = &hdr->daddr;
452 /* Limit redirects both by destination (here)
453 and by source (inside ndisc_send_redirect)
455 if (xrlim_allow(dst, 1*HZ))
456 ndisc_send_redirect(skb, n, target);
457 } else if (ipv6_addr_type(&hdr->saddr)&(IPV6_ADDR_MULTICAST|IPV6_ADDR_LOOPBACK
458 |IPV6_ADDR_LINKLOCAL)) {
459 /* This check is security critical. */
460 goto error;
463 if (skb->len > dst_mtu(dst)) {
464 /* Again, force OUTPUT device used as source address */
465 skb->dev = dst->dev;
466 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, dst_mtu(dst), skb->dev);
467 IP6_INC_STATS_BH(IPSTATS_MIB_INTOOBIGERRORS);
468 IP6_INC_STATS_BH(IPSTATS_MIB_FRAGFAILS);
469 kfree_skb(skb);
470 return -EMSGSIZE;
473 if (skb_cow(skb, dst->dev->hard_header_len)) {
474 IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
475 goto drop;
478 hdr = skb->nh.ipv6h;
480 /* Mangling hops number delayed to point after skb COW */
482 hdr->hop_limit--;
484 IP6_INC_STATS_BH(IPSTATS_MIB_OUTFORWDATAGRAMS);
485 return NF_HOOK(PF_INET6,NF_IP6_FORWARD, skb, skb->dev, dst->dev, ip6_forward_finish);
487 error:
488 IP6_INC_STATS_BH(IPSTATS_MIB_INADDRERRORS);
489 drop:
490 kfree_skb(skb);
491 return -EINVAL;
494 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
496 to->pkt_type = from->pkt_type;
497 to->priority = from->priority;
498 to->protocol = from->protocol;
499 dst_release(to->dst);
500 to->dst = dst_clone(from->dst);
501 to->dev = from->dev;
503 #ifdef CONFIG_NET_SCHED
504 to->tc_index = from->tc_index;
505 #endif
506 #ifdef CONFIG_NETFILTER
507 to->nfmark = from->nfmark;
508 /* Connection association is same as pre-frag packet */
509 nf_conntrack_put(to->nfct);
510 to->nfct = from->nfct;
511 nf_conntrack_get(to->nfct);
512 to->nfctinfo = from->nfctinfo;
513 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
514 nf_conntrack_put_reasm(to->nfct_reasm);
515 to->nfct_reasm = from->nfct_reasm;
516 nf_conntrack_get_reasm(to->nfct_reasm);
517 #endif
518 #ifdef CONFIG_BRIDGE_NETFILTER
519 nf_bridge_put(to->nf_bridge);
520 to->nf_bridge = from->nf_bridge;
521 nf_bridge_get(to->nf_bridge);
522 #endif
523 #endif
524 skb_copy_secmark(to, from);
527 int ip6_find_1stfragopt(struct sk_buff *skb, u8 **nexthdr)
529 u16 offset = sizeof(struct ipv6hdr);
530 struct ipv6_opt_hdr *exthdr = (struct ipv6_opt_hdr*)(skb->nh.ipv6h + 1);
531 unsigned int packet_len = skb->tail - skb->nh.raw;
532 int found_rhdr = 0;
533 *nexthdr = &skb->nh.ipv6h->nexthdr;
535 while (offset + 1 <= packet_len) {
537 switch (**nexthdr) {
539 case NEXTHDR_HOP:
540 break;
541 case NEXTHDR_ROUTING:
542 found_rhdr = 1;
543 break;
544 case NEXTHDR_DEST:
545 #ifdef CONFIG_IPV6_MIP6
546 if (ipv6_find_tlv(skb, offset, IPV6_TLV_HAO) >= 0)
547 break;
548 #endif
549 if (found_rhdr)
550 return offset;
551 break;
552 default :
553 return offset;
556 offset += ipv6_optlen(exthdr);
557 *nexthdr = &exthdr->nexthdr;
558 exthdr = (struct ipv6_opt_hdr*)(skb->nh.raw + offset);
561 return offset;
563 EXPORT_SYMBOL_GPL(ip6_find_1stfragopt);
565 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
567 struct net_device *dev;
568 struct sk_buff *frag;
569 struct rt6_info *rt = (struct rt6_info*)skb->dst;
570 struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
571 struct ipv6hdr *tmp_hdr;
572 struct frag_hdr *fh;
573 unsigned int mtu, hlen, left, len;
574 u32 frag_id = 0;
575 int ptr, offset = 0, err=0;
576 u8 *prevhdr, nexthdr = 0;
578 dev = rt->u.dst.dev;
579 hlen = ip6_find_1stfragopt(skb, &prevhdr);
580 nexthdr = *prevhdr;
582 mtu = dst_mtu(&rt->u.dst);
583 if (np && np->frag_size < mtu) {
584 if (np->frag_size)
585 mtu = np->frag_size;
587 mtu -= hlen + sizeof(struct frag_hdr);
589 if (skb_shinfo(skb)->frag_list) {
590 int first_len = skb_pagelen(skb);
592 if (first_len - hlen > mtu ||
593 ((first_len - hlen) & 7) ||
594 skb_cloned(skb))
595 goto slow_path;
597 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
598 /* Correct geometry. */
599 if (frag->len > mtu ||
600 ((frag->len & 7) && frag->next) ||
601 skb_headroom(frag) < hlen)
602 goto slow_path;
604 /* Partially cloned skb? */
605 if (skb_shared(frag))
606 goto slow_path;
608 BUG_ON(frag->sk);
609 if (skb->sk) {
610 sock_hold(skb->sk);
611 frag->sk = skb->sk;
612 frag->destructor = sock_wfree;
613 skb->truesize -= frag->truesize;
617 err = 0;
618 offset = 0;
619 frag = skb_shinfo(skb)->frag_list;
620 skb_shinfo(skb)->frag_list = NULL;
621 /* BUILD HEADER */
623 tmp_hdr = kmalloc(hlen, GFP_ATOMIC);
624 if (!tmp_hdr) {
625 IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
626 return -ENOMEM;
629 *prevhdr = NEXTHDR_FRAGMENT;
630 memcpy(tmp_hdr, skb->nh.raw, hlen);
631 __skb_pull(skb, hlen);
632 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
633 skb->nh.raw = __skb_push(skb, hlen);
634 memcpy(skb->nh.raw, tmp_hdr, hlen);
636 ipv6_select_ident(skb, fh);
637 fh->nexthdr = nexthdr;
638 fh->reserved = 0;
639 fh->frag_off = htons(IP6_MF);
640 frag_id = fh->identification;
642 first_len = skb_pagelen(skb);
643 skb->data_len = first_len - skb_headlen(skb);
644 skb->len = first_len;
645 skb->nh.ipv6h->payload_len = htons(first_len - sizeof(struct ipv6hdr));
648 for (;;) {
649 /* Prepare header of the next frame,
650 * before previous one went down. */
651 if (frag) {
652 frag->ip_summed = CHECKSUM_NONE;
653 frag->h.raw = frag->data;
654 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
655 frag->nh.raw = __skb_push(frag, hlen);
656 memcpy(frag->nh.raw, tmp_hdr, hlen);
657 offset += skb->len - hlen - sizeof(struct frag_hdr);
658 fh->nexthdr = nexthdr;
659 fh->reserved = 0;
660 fh->frag_off = htons(offset);
661 if (frag->next != NULL)
662 fh->frag_off |= htons(IP6_MF);
663 fh->identification = frag_id;
664 frag->nh.ipv6h->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
665 ip6_copy_metadata(frag, skb);
668 err = output(skb);
669 if(!err)
670 IP6_INC_STATS(IPSTATS_MIB_FRAGCREATES);
672 if (err || !frag)
673 break;
675 skb = frag;
676 frag = skb->next;
677 skb->next = NULL;
680 kfree(tmp_hdr);
682 if (err == 0) {
683 IP6_INC_STATS(IPSTATS_MIB_FRAGOKS);
684 return 0;
687 while (frag) {
688 skb = frag->next;
689 kfree_skb(frag);
690 frag = skb;
693 IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
694 return err;
697 slow_path:
698 left = skb->len - hlen; /* Space per frame */
699 ptr = hlen; /* Where to start from */
702 * Fragment the datagram.
705 *prevhdr = NEXTHDR_FRAGMENT;
708 * Keep copying data until we run out.
710 while(left > 0) {
711 len = left;
712 /* IF: it doesn't fit, use 'mtu' - the data space left */
713 if (len > mtu)
714 len = mtu;
715 /* IF: we are not sending upto and including the packet end
716 then align the next start on an eight byte boundary */
717 if (len < left) {
718 len &= ~7;
721 * Allocate buffer.
724 if ((frag = alloc_skb(len+hlen+sizeof(struct frag_hdr)+LL_RESERVED_SPACE(rt->u.dst.dev), GFP_ATOMIC)) == NULL) {
725 NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
726 IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
727 err = -ENOMEM;
728 goto fail;
732 * Set up data on packet
735 ip6_copy_metadata(frag, skb);
736 skb_reserve(frag, LL_RESERVED_SPACE(rt->u.dst.dev));
737 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
738 frag->nh.raw = frag->data;
739 fh = (struct frag_hdr*)(frag->data + hlen);
740 frag->h.raw = frag->data + hlen + sizeof(struct frag_hdr);
743 * Charge the memory for the fragment to any owner
744 * it might possess
746 if (skb->sk)
747 skb_set_owner_w(frag, skb->sk);
750 * Copy the packet header into the new buffer.
752 memcpy(frag->nh.raw, skb->data, hlen);
755 * Build fragment header.
757 fh->nexthdr = nexthdr;
758 fh->reserved = 0;
759 if (!frag_id) {
760 ipv6_select_ident(skb, fh);
761 frag_id = fh->identification;
762 } else
763 fh->identification = frag_id;
766 * Copy a block of the IP datagram.
768 if (skb_copy_bits(skb, ptr, frag->h.raw, len))
769 BUG();
770 left -= len;
772 fh->frag_off = htons(offset);
773 if (left > 0)
774 fh->frag_off |= htons(IP6_MF);
775 frag->nh.ipv6h->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
777 ptr += len;
778 offset += len;
781 * Put this fragment into the sending queue.
783 err = output(frag);
784 if (err)
785 goto fail;
787 IP6_INC_STATS(IPSTATS_MIB_FRAGCREATES);
789 kfree_skb(skb);
790 IP6_INC_STATS(IPSTATS_MIB_FRAGOKS);
791 return err;
793 fail:
794 kfree_skb(skb);
795 IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
796 return err;
799 static inline int ip6_rt_check(struct rt6key *rt_key,
800 struct in6_addr *fl_addr,
801 struct in6_addr *addr_cache)
803 return ((rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
804 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache)));
807 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
808 struct dst_entry *dst,
809 struct flowi *fl)
811 struct ipv6_pinfo *np = inet6_sk(sk);
812 struct rt6_info *rt = (struct rt6_info *)dst;
814 if (!dst)
815 goto out;
817 /* Yes, checking route validity in not connected
818 * case is not very simple. Take into account,
819 * that we do not support routing by source, TOS,
820 * and MSG_DONTROUTE --ANK (980726)
822 * 1. ip6_rt_check(): If route was host route,
823 * check that cached destination is current.
824 * If it is network route, we still may
825 * check its validity using saved pointer
826 * to the last used address: daddr_cache.
827 * We do not want to save whole address now,
828 * (because main consumer of this service
829 * is tcp, which has not this problem),
830 * so that the last trick works only on connected
831 * sockets.
832 * 2. oif also should be the same.
834 if (ip6_rt_check(&rt->rt6i_dst, &fl->fl6_dst, np->daddr_cache) ||
835 #ifdef CONFIG_IPV6_SUBTREES
836 ip6_rt_check(&rt->rt6i_src, &fl->fl6_src, np->saddr_cache) ||
837 #endif
838 (fl->oif && fl->oif != dst->dev->ifindex)) {
839 dst_release(dst);
840 dst = NULL;
843 out:
844 return dst;
847 static int ip6_dst_lookup_tail(struct sock *sk,
848 struct dst_entry **dst, struct flowi *fl)
850 int err;
852 if (*dst == NULL)
853 *dst = ip6_route_output(sk, fl);
855 if ((err = (*dst)->error))
856 goto out_err_release;
858 if (ipv6_addr_any(&fl->fl6_src)) {
859 err = ipv6_get_saddr(*dst, &fl->fl6_dst, &fl->fl6_src);
860 if (err)
861 goto out_err_release;
864 return 0;
866 out_err_release:
867 dst_release(*dst);
868 *dst = NULL;
869 return err;
873 * ip6_dst_lookup - perform route lookup on flow
874 * @sk: socket which provides route info
875 * @dst: pointer to dst_entry * for result
876 * @fl: flow to lookup
878 * This function performs a route lookup on the given flow.
880 * It returns zero on success, or a standard errno code on error.
882 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
884 *dst = NULL;
885 return ip6_dst_lookup_tail(sk, dst, fl);
887 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
890 * ip6_sk_dst_lookup - perform socket cached route lookup on flow
891 * @sk: socket which provides the dst cache and route info
892 * @dst: pointer to dst_entry * for result
893 * @fl: flow to lookup
895 * This function performs a route lookup on the given flow with the
896 * possibility of using the cached route in the socket if it is valid.
897 * It will take the socket dst lock when operating on the dst cache.
898 * As a result, this function can only be used in process context.
900 * It returns zero on success, or a standard errno code on error.
902 int ip6_sk_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
904 *dst = NULL;
905 if (sk) {
906 *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
907 *dst = ip6_sk_dst_check(sk, *dst, fl);
910 return ip6_dst_lookup_tail(sk, dst, fl);
912 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup);
914 static inline int ip6_ufo_append_data(struct sock *sk,
915 int getfrag(void *from, char *to, int offset, int len,
916 int odd, struct sk_buff *skb),
917 void *from, int length, int hh_len, int fragheaderlen,
918 int transhdrlen, int mtu,unsigned int flags)
921 struct sk_buff *skb;
922 int err;
924 /* There is support for UDP large send offload by network
925 * device, so create one single skb packet containing complete
926 * udp datagram
928 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
929 skb = sock_alloc_send_skb(sk,
930 hh_len + fragheaderlen + transhdrlen + 20,
931 (flags & MSG_DONTWAIT), &err);
932 if (skb == NULL)
933 return -ENOMEM;
935 /* reserve space for Hardware header */
936 skb_reserve(skb, hh_len);
938 /* create space for UDP/IP header */
939 skb_put(skb,fragheaderlen + transhdrlen);
941 /* initialize network header pointer */
942 skb->nh.raw = skb->data;
944 /* initialize protocol header pointer */
945 skb->h.raw = skb->data + fragheaderlen;
947 skb->ip_summed = CHECKSUM_PARTIAL;
948 skb->csum = 0;
949 sk->sk_sndmsg_off = 0;
952 err = skb_append_datato_frags(sk,skb, getfrag, from,
953 (length - transhdrlen));
954 if (!err) {
955 struct frag_hdr fhdr;
957 /* specify the length of each IP datagram fragment*/
958 skb_shinfo(skb)->gso_size = mtu - fragheaderlen -
959 sizeof(struct frag_hdr);
960 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
961 ipv6_select_ident(skb, &fhdr);
962 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
963 __skb_queue_tail(&sk->sk_write_queue, skb);
965 return 0;
967 /* There is not enough support do UPD LSO,
968 * so follow normal path
970 kfree_skb(skb);
972 return err;
975 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
976 int offset, int len, int odd, struct sk_buff *skb),
977 void *from, int length, int transhdrlen,
978 int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi *fl,
979 struct rt6_info *rt, unsigned int flags)
981 struct inet_sock *inet = inet_sk(sk);
982 struct ipv6_pinfo *np = inet6_sk(sk);
983 struct sk_buff *skb;
984 unsigned int maxfraglen, fragheaderlen;
985 int exthdrlen;
986 int hh_len;
987 int mtu;
988 int copy;
989 int err;
990 int offset = 0;
991 int csummode = CHECKSUM_NONE;
993 if (flags&MSG_PROBE)
994 return 0;
995 if (skb_queue_empty(&sk->sk_write_queue)) {
997 * setup for corking
999 if (opt) {
1000 if (np->cork.opt == NULL) {
1001 np->cork.opt = kmalloc(opt->tot_len,
1002 sk->sk_allocation);
1003 if (unlikely(np->cork.opt == NULL))
1004 return -ENOBUFS;
1005 } else if (np->cork.opt->tot_len < opt->tot_len) {
1006 printk(KERN_DEBUG "ip6_append_data: invalid option length\n");
1007 return -EINVAL;
1009 memcpy(np->cork.opt, opt, opt->tot_len);
1010 inet->cork.flags |= IPCORK_OPT;
1011 /* need source address above miyazawa*/
1013 dst_hold(&rt->u.dst);
1014 np->cork.rt = rt;
1015 inet->cork.fl = *fl;
1016 np->cork.hop_limit = hlimit;
1017 np->cork.tclass = tclass;
1018 mtu = dst_mtu(rt->u.dst.path);
1019 if (np->frag_size < mtu) {
1020 if (np->frag_size)
1021 mtu = np->frag_size;
1023 inet->cork.fragsize = mtu;
1024 if (dst_allfrag(rt->u.dst.path))
1025 inet->cork.flags |= IPCORK_ALLFRAG;
1026 inet->cork.length = 0;
1027 sk->sk_sndmsg_page = NULL;
1028 sk->sk_sndmsg_off = 0;
1029 exthdrlen = rt->u.dst.header_len + (opt ? opt->opt_flen : 0);
1030 length += exthdrlen;
1031 transhdrlen += exthdrlen;
1032 } else {
1033 rt = np->cork.rt;
1034 fl = &inet->cork.fl;
1035 if (inet->cork.flags & IPCORK_OPT)
1036 opt = np->cork.opt;
1037 transhdrlen = 0;
1038 exthdrlen = 0;
1039 mtu = inet->cork.fragsize;
1042 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1044 fragheaderlen = sizeof(struct ipv6hdr) + rt->u.dst.nfheader_len + (opt ? opt->opt_nflen : 0);
1045 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
1047 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1048 if (inet->cork.length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
1049 ipv6_local_error(sk, EMSGSIZE, fl, mtu-exthdrlen);
1050 return -EMSGSIZE;
1055 * Let's try using as much space as possible.
1056 * Use MTU if total length of the message fits into the MTU.
1057 * Otherwise, we need to reserve fragment header and
1058 * fragment alignment (= 8-15 octects, in total).
1060 * Note that we may need to "move" the data from the tail of
1061 * of the buffer to the new fragment when we split
1062 * the message.
1064 * FIXME: It may be fragmented into multiple chunks
1065 * at once if non-fragmentable extension headers
1066 * are too large.
1067 * --yoshfuji
1070 inet->cork.length += length;
1071 if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
1072 (rt->u.dst.dev->features & NETIF_F_UFO)) {
1074 err = ip6_ufo_append_data(sk, getfrag, from, length, hh_len,
1075 fragheaderlen, transhdrlen, mtu,
1076 flags);
1077 if (err)
1078 goto error;
1079 return 0;
1082 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1083 goto alloc_new_skb;
1085 while (length > 0) {
1086 /* Check if the remaining data fits into current packet. */
1087 copy = (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1088 if (copy < length)
1089 copy = maxfraglen - skb->len;
1091 if (copy <= 0) {
1092 char *data;
1093 unsigned int datalen;
1094 unsigned int fraglen;
1095 unsigned int fraggap;
1096 unsigned int alloclen;
1097 struct sk_buff *skb_prev;
1098 alloc_new_skb:
1099 skb_prev = skb;
1101 /* There's no room in the current skb */
1102 if (skb_prev)
1103 fraggap = skb_prev->len - maxfraglen;
1104 else
1105 fraggap = 0;
1108 * If remaining data exceeds the mtu,
1109 * we know we need more fragment(s).
1111 datalen = length + fraggap;
1112 if (datalen > (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1113 datalen = maxfraglen - fragheaderlen;
1115 fraglen = datalen + fragheaderlen;
1116 if ((flags & MSG_MORE) &&
1117 !(rt->u.dst.dev->features&NETIF_F_SG))
1118 alloclen = mtu;
1119 else
1120 alloclen = datalen + fragheaderlen;
1123 * The last fragment gets additional space at tail.
1124 * Note: we overallocate on fragments with MSG_MODE
1125 * because we have no idea if we're the last one.
1127 if (datalen == length + fraggap)
1128 alloclen += rt->u.dst.trailer_len;
1131 * We just reserve space for fragment header.
1132 * Note: this may be overallocation if the message
1133 * (without MSG_MORE) fits into the MTU.
1135 alloclen += sizeof(struct frag_hdr);
1137 if (transhdrlen) {
1138 skb = sock_alloc_send_skb(sk,
1139 alloclen + hh_len,
1140 (flags & MSG_DONTWAIT), &err);
1141 } else {
1142 skb = NULL;
1143 if (atomic_read(&sk->sk_wmem_alloc) <=
1144 2 * sk->sk_sndbuf)
1145 skb = sock_wmalloc(sk,
1146 alloclen + hh_len, 1,
1147 sk->sk_allocation);
1148 if (unlikely(skb == NULL))
1149 err = -ENOBUFS;
1151 if (skb == NULL)
1152 goto error;
1154 * Fill in the control structures
1156 skb->ip_summed = csummode;
1157 skb->csum = 0;
1158 /* reserve for fragmentation */
1159 skb_reserve(skb, hh_len+sizeof(struct frag_hdr));
1162 * Find where to start putting bytes
1164 data = skb_put(skb, fraglen);
1165 skb->nh.raw = data + exthdrlen;
1166 data += fragheaderlen;
1167 skb->h.raw = data + exthdrlen;
1169 if (fraggap) {
1170 skb->csum = skb_copy_and_csum_bits(
1171 skb_prev, maxfraglen,
1172 data + transhdrlen, fraggap, 0);
1173 skb_prev->csum = csum_sub(skb_prev->csum,
1174 skb->csum);
1175 data += fraggap;
1176 pskb_trim_unique(skb_prev, maxfraglen);
1178 copy = datalen - transhdrlen - fraggap;
1179 if (copy < 0) {
1180 err = -EINVAL;
1181 kfree_skb(skb);
1182 goto error;
1183 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1184 err = -EFAULT;
1185 kfree_skb(skb);
1186 goto error;
1189 offset += copy;
1190 length -= datalen - fraggap;
1191 transhdrlen = 0;
1192 exthdrlen = 0;
1193 csummode = CHECKSUM_NONE;
1196 * Put the packet on the pending queue
1198 __skb_queue_tail(&sk->sk_write_queue, skb);
1199 continue;
1202 if (copy > length)
1203 copy = length;
1205 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
1206 unsigned int off;
1208 off = skb->len;
1209 if (getfrag(from, skb_put(skb, copy),
1210 offset, copy, off, skb) < 0) {
1211 __skb_trim(skb, off);
1212 err = -EFAULT;
1213 goto error;
1215 } else {
1216 int i = skb_shinfo(skb)->nr_frags;
1217 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1218 struct page *page = sk->sk_sndmsg_page;
1219 int off = sk->sk_sndmsg_off;
1220 unsigned int left;
1222 if (page && (left = PAGE_SIZE - off) > 0) {
1223 if (copy >= left)
1224 copy = left;
1225 if (page != frag->page) {
1226 if (i == MAX_SKB_FRAGS) {
1227 err = -EMSGSIZE;
1228 goto error;
1230 get_page(page);
1231 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1232 frag = &skb_shinfo(skb)->frags[i];
1234 } else if(i < MAX_SKB_FRAGS) {
1235 if (copy > PAGE_SIZE)
1236 copy = PAGE_SIZE;
1237 page = alloc_pages(sk->sk_allocation, 0);
1238 if (page == NULL) {
1239 err = -ENOMEM;
1240 goto error;
1242 sk->sk_sndmsg_page = page;
1243 sk->sk_sndmsg_off = 0;
1245 skb_fill_page_desc(skb, i, page, 0, 0);
1246 frag = &skb_shinfo(skb)->frags[i];
1247 skb->truesize += PAGE_SIZE;
1248 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1249 } else {
1250 err = -EMSGSIZE;
1251 goto error;
1253 if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1254 err = -EFAULT;
1255 goto error;
1257 sk->sk_sndmsg_off += copy;
1258 frag->size += copy;
1259 skb->len += copy;
1260 skb->data_len += copy;
1262 offset += copy;
1263 length -= copy;
1265 return 0;
1266 error:
1267 inet->cork.length -= length;
1268 IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1269 return err;
1272 int ip6_push_pending_frames(struct sock *sk)
1274 struct sk_buff *skb, *tmp_skb;
1275 struct sk_buff **tail_skb;
1276 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1277 struct inet_sock *inet = inet_sk(sk);
1278 struct ipv6_pinfo *np = inet6_sk(sk);
1279 struct ipv6hdr *hdr;
1280 struct ipv6_txoptions *opt = np->cork.opt;
1281 struct rt6_info *rt = np->cork.rt;
1282 struct flowi *fl = &inet->cork.fl;
1283 unsigned char proto = fl->proto;
1284 int err = 0;
1286 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1287 goto out;
1288 tail_skb = &(skb_shinfo(skb)->frag_list);
1290 /* move skb->data to ip header from ext header */
1291 if (skb->data < skb->nh.raw)
1292 __skb_pull(skb, skb->nh.raw - skb->data);
1293 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1294 __skb_pull(tmp_skb, skb->h.raw - skb->nh.raw);
1295 *tail_skb = tmp_skb;
1296 tail_skb = &(tmp_skb->next);
1297 skb->len += tmp_skb->len;
1298 skb->data_len += tmp_skb->len;
1299 skb->truesize += tmp_skb->truesize;
1300 __sock_put(tmp_skb->sk);
1301 tmp_skb->destructor = NULL;
1302 tmp_skb->sk = NULL;
1305 ipv6_addr_copy(final_dst, &fl->fl6_dst);
1306 __skb_pull(skb, skb->h.raw - skb->nh.raw);
1307 if (opt && opt->opt_flen)
1308 ipv6_push_frag_opts(skb, opt, &proto);
1309 if (opt && opt->opt_nflen)
1310 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1312 skb->nh.ipv6h = hdr = (struct ipv6hdr*) skb_push(skb, sizeof(struct ipv6hdr));
1314 *(u32*)hdr = fl->fl6_flowlabel |
1315 htonl(0x60000000 | ((int)np->cork.tclass << 20));
1317 if (skb->len <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN)
1318 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
1319 else
1320 hdr->payload_len = 0;
1321 hdr->hop_limit = np->cork.hop_limit;
1322 hdr->nexthdr = proto;
1323 ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
1324 ipv6_addr_copy(&hdr->daddr, final_dst);
1326 skb->priority = sk->sk_priority;
1328 skb->dst = dst_clone(&rt->u.dst);
1329 IP6_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
1330 err = NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, skb->dst->dev, dst_output);
1331 if (err) {
1332 if (err > 0)
1333 err = np->recverr ? net_xmit_errno(err) : 0;
1334 if (err)
1335 goto error;
1338 out:
1339 inet->cork.flags &= ~IPCORK_OPT;
1340 kfree(np->cork.opt);
1341 np->cork.opt = NULL;
1342 if (np->cork.rt) {
1343 dst_release(&np->cork.rt->u.dst);
1344 np->cork.rt = NULL;
1345 inet->cork.flags &= ~IPCORK_ALLFRAG;
1347 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1348 return err;
1349 error:
1350 goto out;
1353 void ip6_flush_pending_frames(struct sock *sk)
1355 struct inet_sock *inet = inet_sk(sk);
1356 struct ipv6_pinfo *np = inet6_sk(sk);
1357 struct sk_buff *skb;
1359 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1360 IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1361 kfree_skb(skb);
1364 inet->cork.flags &= ~IPCORK_OPT;
1366 kfree(np->cork.opt);
1367 np->cork.opt = NULL;
1368 if (np->cork.rt) {
1369 dst_release(&np->cork.rt->u.dst);
1370 np->cork.rt = NULL;
1371 inet->cork.flags &= ~IPCORK_ALLFRAG;
1373 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));