2 * IPv6 output functions
3 * Linux INET6 implementation
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.
18 * A.N.Kuznetsov : airthmetics in fragmentation.
19 * extension headers are implemented.
20 * route changes now work.
21 * ip6_forward does not confuse sniffers.
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
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>
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>
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)
75 struct dst_entry *dst = skb->dst;
78 return neigh_hh_output(dst->hh, skb);
79 else if (dst->neighbour)
80 return dst->neighbour->output(skb);
82 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
88 /* dev_loopback_xmit for use with netfilter. */
89 static int ip6_dev_loopback_xmit(struct sk_buff *newskb)
91 skb_reset_mac_header(newskb);
92 __skb_pull(newskb, skb_network_offset(newskb));
93 newskb->pkt_type = PACKET_LOOPBACK;
94 newskb->ip_summed = CHECKSUM_UNNECESSARY;
95 BUG_TRAP(newskb->dst);
102 static int ip6_output2(struct sk_buff *skb)
104 struct dst_entry *dst = skb->dst;
105 struct net_device *dev = dst->dev;
107 skb->protocol = htons(ETH_P_IPV6);
110 if (ipv6_addr_is_multicast(&skb->nh.ipv6h->daddr)) {
111 struct ipv6_pinfo* np = skb->sk ? inet6_sk(skb->sk) : NULL;
112 struct inet6_dev *idev = ip6_dst_idev(skb->dst);
114 if (!(dev->flags & IFF_LOOPBACK) && (!np || np->mc_loop) &&
115 ipv6_chk_mcast_addr(dev, &skb->nh.ipv6h->daddr,
116 &skb->nh.ipv6h->saddr)) {
117 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
119 /* Do not check for IFF_ALLMULTI; multicast routing
120 is not supported in any case.
123 NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, newskb, NULL,
125 ip6_dev_loopback_xmit);
127 if (skb->nh.ipv6h->hop_limit == 0) {
128 IP6_INC_STATS(idev, IPSTATS_MIB_OUTDISCARDS);
134 IP6_INC_STATS(idev, IPSTATS_MIB_OUTMCASTPKTS);
137 return NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, skb,NULL, skb->dev,ip6_output_finish);
140 int ip6_output(struct sk_buff *skb)
142 if ((skb->len > dst_mtu(skb->dst) && !skb_is_gso(skb)) ||
143 dst_allfrag(skb->dst))
144 return ip6_fragment(skb, ip6_output2);
146 return ip6_output2(skb);
150 * xmit an sk_buff (used by TCP)
153 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
154 struct ipv6_txoptions *opt, int ipfragok)
156 struct ipv6_pinfo *np = inet6_sk(sk);
157 struct in6_addr *first_hop = &fl->fl6_dst;
158 struct dst_entry *dst = skb->dst;
160 u8 proto = fl->proto;
161 int seg_len = skb->len;
168 /* First: exthdrs may take lots of space (~8K for now)
169 MAX_HEADER is not enough.
171 head_room = opt->opt_nflen + opt->opt_flen;
172 seg_len += head_room;
173 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
175 if (skb_headroom(skb) < head_room) {
176 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
178 IP6_INC_STATS(ip6_dst_idev(skb->dst),
179 IPSTATS_MIB_OUTDISCARDS);
186 skb_set_owner_w(skb, sk);
189 ipv6_push_frag_opts(skb, opt, &proto);
191 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
194 skb_push(skb, sizeof(struct ipv6hdr));
195 skb_reset_network_header(skb);
199 * Fill in the IPv6 header
204 hlimit = np->hop_limit;
206 hlimit = dst_metric(dst, RTAX_HOPLIMIT);
208 hlimit = ipv6_get_hoplimit(dst->dev);
216 *(__be32 *)hdr = htonl(0x60000000 | (tclass << 20)) | fl->fl6_flowlabel;
218 hdr->payload_len = htons(seg_len);
219 hdr->nexthdr = proto;
220 hdr->hop_limit = hlimit;
222 ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
223 ipv6_addr_copy(&hdr->daddr, first_hop);
225 skb->priority = sk->sk_priority;
228 if ((skb->len <= mtu) || ipfragok || skb_is_gso(skb)) {
229 IP6_INC_STATS(ip6_dst_idev(skb->dst),
230 IPSTATS_MIB_OUTREQUESTS);
231 return NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, dst->dev,
236 printk(KERN_DEBUG "IPv6: sending pkt_too_big to self\n");
238 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev);
239 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGFAILS);
244 EXPORT_SYMBOL(ip6_xmit);
247 * To avoid extra problems ND packets are send through this
248 * routine. It's code duplication but I really want to avoid
249 * extra checks since ipv6_build_header is used by TCP (which
250 * is for us performance critical)
253 int ip6_nd_hdr(struct sock *sk, struct sk_buff *skb, struct net_device *dev,
254 struct in6_addr *saddr, struct in6_addr *daddr,
257 struct ipv6_pinfo *np = inet6_sk(sk);
261 skb->protocol = htons(ETH_P_IPV6);
264 totlen = len + sizeof(struct ipv6hdr);
266 hdr = (struct ipv6hdr *) skb_put(skb, sizeof(struct ipv6hdr));
269 *(__be32*)hdr = htonl(0x60000000);
271 hdr->payload_len = htons(len);
272 hdr->nexthdr = proto;
273 hdr->hop_limit = np->hop_limit;
275 ipv6_addr_copy(&hdr->saddr, saddr);
276 ipv6_addr_copy(&hdr->daddr, daddr);
281 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
283 struct ip6_ra_chain *ra;
284 struct sock *last = NULL;
286 read_lock(&ip6_ra_lock);
287 for (ra = ip6_ra_chain; ra; ra = ra->next) {
288 struct sock *sk = ra->sk;
289 if (sk && ra->sel == sel &&
290 (!sk->sk_bound_dev_if ||
291 sk->sk_bound_dev_if == skb->dev->ifindex)) {
293 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
295 rawv6_rcv(last, skb2);
302 rawv6_rcv(last, skb);
303 read_unlock(&ip6_ra_lock);
306 read_unlock(&ip6_ra_lock);
310 static int ip6_forward_proxy_check(struct sk_buff *skb)
312 struct ipv6hdr *hdr = skb->nh.ipv6h;
313 u8 nexthdr = hdr->nexthdr;
316 if (ipv6_ext_hdr(nexthdr)) {
317 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr);
321 offset = sizeof(struct ipv6hdr);
323 if (nexthdr == IPPROTO_ICMPV6) {
324 struct icmp6hdr *icmp6;
326 if (!pskb_may_pull(skb, (skb_network_header(skb) +
327 offset + 1 - skb->data)))
330 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + 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:
338 /* For reaction involving unicast neighbor discovery
339 * message destined to the proxied address, pass it to
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);
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)
375 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
376 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
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
396 u8 *ptr = skb_network_header(skb) + opt->ra;
397 if (ip6_call_ra_chain(skb, (ptr[2]<<8) + ptr[3]))
402 * check and decrement ttl
404 if (hdr->hop_limit <= 1) {
405 /* Force OUTPUT device used as source address */
407 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT,
409 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INHDRERRORS);
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);
420 return ip6_input(skb);
421 else if (proxied < 0) {
422 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
427 if (!xfrm6_route_forward(skb)) {
428 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
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;
439 struct neighbour *n = dst->neighbour;
442 * incoming and outgoing devices are the same
446 rt = (struct rt6_info *) dst;
447 if ((rt->rt6i_flags & RTF_GATEWAY))
448 target = (struct in6_addr*)&n->primary_key;
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. */
463 if (skb->len > dst_mtu(dst)) {
464 /* Again, force OUTPUT device used as source address */
466 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, dst_mtu(dst), skb->dev);
467 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INTOOBIGERRORS);
468 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_FRAGFAILS);
473 if (skb_cow(skb, dst->dev->hard_header_len)) {
474 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_OUTDISCARDS);
480 /* Mangling hops number delayed to point after skb COW */
484 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
485 return NF_HOOK(PF_INET6,NF_IP6_FORWARD, skb, skb->dev, dst->dev, ip6_forward_finish);
488 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
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);
502 to->mark = from->mark;
504 #ifdef CONFIG_NET_SCHED
505 to->tc_index = from->tc_index;
507 #ifdef CONFIG_NETFILTER
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);
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);
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_network_header(skb);
533 *nexthdr = &skb->nh.ipv6h->nexthdr;
535 while (offset + 1 <= packet_len) {
541 case NEXTHDR_ROUTING:
545 #ifdef CONFIG_IPV6_MIP6
546 if (ipv6_find_tlv(skb, offset, IPV6_TLV_HAO) >= 0)
556 offset += ipv6_optlen(exthdr);
557 *nexthdr = &exthdr->nexthdr;
558 exthdr = (struct ipv6_opt_hdr *)(skb_network_header(skb) +
564 EXPORT_SYMBOL_GPL(ip6_find_1stfragopt);
566 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
568 struct net_device *dev;
569 struct sk_buff *frag;
570 struct rt6_info *rt = (struct rt6_info*)skb->dst;
571 struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
572 struct ipv6hdr *tmp_hdr;
574 unsigned int mtu, hlen, left, len;
576 int ptr, offset = 0, err=0;
577 u8 *prevhdr, nexthdr = 0;
580 hlen = ip6_find_1stfragopt(skb, &prevhdr);
583 mtu = dst_mtu(&rt->u.dst);
584 if (np && np->frag_size < mtu) {
588 mtu -= hlen + sizeof(struct frag_hdr);
590 if (skb_shinfo(skb)->frag_list) {
591 int first_len = skb_pagelen(skb);
593 if (first_len - hlen > mtu ||
594 ((first_len - hlen) & 7) ||
598 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
599 /* Correct geometry. */
600 if (frag->len > mtu ||
601 ((frag->len & 7) && frag->next) ||
602 skb_headroom(frag) < hlen)
605 /* Partially cloned skb? */
606 if (skb_shared(frag))
613 frag->destructor = sock_wfree;
614 skb->truesize -= frag->truesize;
620 frag = skb_shinfo(skb)->frag_list;
621 skb_shinfo(skb)->frag_list = NULL;
624 *prevhdr = NEXTHDR_FRAGMENT;
625 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
627 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGFAILS);
631 __skb_pull(skb, hlen);
632 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
633 __skb_push(skb, hlen);
634 skb_reset_network_header(skb);
635 memcpy(skb_network_header(skb), tmp_hdr, hlen);
637 ipv6_select_ident(skb, fh);
638 fh->nexthdr = nexthdr;
640 fh->frag_off = htons(IP6_MF);
641 frag_id = fh->identification;
643 first_len = skb_pagelen(skb);
644 skb->data_len = first_len - skb_headlen(skb);
645 skb->len = first_len;
646 skb->nh.ipv6h->payload_len = htons(first_len - sizeof(struct ipv6hdr));
648 dst_hold(&rt->u.dst);
651 /* Prepare header of the next frame,
652 * before previous one went down. */
654 frag->ip_summed = CHECKSUM_NONE;
655 frag->h.raw = frag->data;
656 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
657 __skb_push(frag, hlen);
658 skb_reset_network_header(frag);
659 memcpy(skb_network_header(frag), tmp_hdr,
661 offset += skb->len - hlen - sizeof(struct frag_hdr);
662 fh->nexthdr = nexthdr;
664 fh->frag_off = htons(offset);
665 if (frag->next != NULL)
666 fh->frag_off |= htons(IP6_MF);
667 fh->identification = frag_id;
668 frag->nh.ipv6h->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
669 ip6_copy_metadata(frag, skb);
674 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGCREATES);
687 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGOKS);
688 dst_release(&rt->u.dst);
698 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGFAILS);
699 dst_release(&rt->u.dst);
704 left = skb->len - hlen; /* Space per frame */
705 ptr = hlen; /* Where to start from */
708 * Fragment the datagram.
711 *prevhdr = NEXTHDR_FRAGMENT;
714 * Keep copying data until we run out.
718 /* IF: it doesn't fit, use 'mtu' - the data space left */
721 /* IF: we are not sending upto and including the packet end
722 then align the next start on an eight byte boundary */
730 if ((frag = alloc_skb(len+hlen+sizeof(struct frag_hdr)+LL_RESERVED_SPACE(rt->u.dst.dev), GFP_ATOMIC)) == NULL) {
731 NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
732 IP6_INC_STATS(ip6_dst_idev(skb->dst),
733 IPSTATS_MIB_FRAGFAILS);
739 * Set up data on packet
742 ip6_copy_metadata(frag, skb);
743 skb_reserve(frag, LL_RESERVED_SPACE(rt->u.dst.dev));
744 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
745 skb_reset_network_header(frag);
746 fh = (struct frag_hdr*)(frag->data + hlen);
747 frag->h.raw = frag->data + hlen + sizeof(struct frag_hdr);
750 * Charge the memory for the fragment to any owner
754 skb_set_owner_w(frag, skb->sk);
757 * Copy the packet header into the new buffer.
759 memcpy(skb_network_header(frag), skb->data, hlen);
762 * Build fragment header.
764 fh->nexthdr = nexthdr;
767 ipv6_select_ident(skb, fh);
768 frag_id = fh->identification;
770 fh->identification = frag_id;
773 * Copy a block of the IP datagram.
775 if (skb_copy_bits(skb, ptr, frag->h.raw, len))
779 fh->frag_off = htons(offset);
781 fh->frag_off |= htons(IP6_MF);
782 frag->nh.ipv6h->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
788 * Put this fragment into the sending queue.
794 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGCREATES);
796 IP6_INC_STATS(ip6_dst_idev(skb->dst),
797 IPSTATS_MIB_FRAGOKS);
802 IP6_INC_STATS(ip6_dst_idev(skb->dst),
803 IPSTATS_MIB_FRAGFAILS);
808 static inline int ip6_rt_check(struct rt6key *rt_key,
809 struct in6_addr *fl_addr,
810 struct in6_addr *addr_cache)
812 return ((rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
813 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache)));
816 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
817 struct dst_entry *dst,
820 struct ipv6_pinfo *np = inet6_sk(sk);
821 struct rt6_info *rt = (struct rt6_info *)dst;
826 /* Yes, checking route validity in not connected
827 * case is not very simple. Take into account,
828 * that we do not support routing by source, TOS,
829 * and MSG_DONTROUTE --ANK (980726)
831 * 1. ip6_rt_check(): If route was host route,
832 * check that cached destination is current.
833 * If it is network route, we still may
834 * check its validity using saved pointer
835 * to the last used address: daddr_cache.
836 * We do not want to save whole address now,
837 * (because main consumer of this service
838 * is tcp, which has not this problem),
839 * so that the last trick works only on connected
841 * 2. oif also should be the same.
843 if (ip6_rt_check(&rt->rt6i_dst, &fl->fl6_dst, np->daddr_cache) ||
844 #ifdef CONFIG_IPV6_SUBTREES
845 ip6_rt_check(&rt->rt6i_src, &fl->fl6_src, np->saddr_cache) ||
847 (fl->oif && fl->oif != dst->dev->ifindex)) {
856 static int ip6_dst_lookup_tail(struct sock *sk,
857 struct dst_entry **dst, struct flowi *fl)
862 *dst = ip6_route_output(sk, fl);
864 if ((err = (*dst)->error))
865 goto out_err_release;
867 if (ipv6_addr_any(&fl->fl6_src)) {
868 err = ipv6_get_saddr(*dst, &fl->fl6_dst, &fl->fl6_src);
870 goto out_err_release;
873 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
875 * Here if the dst entry we've looked up
876 * has a neighbour entry that is in the INCOMPLETE
877 * state and the src address from the flow is
878 * marked as OPTIMISTIC, we release the found
879 * dst entry and replace it instead with the
880 * dst entry of the nexthop router
882 if (!((*dst)->neighbour->nud_state & NUD_VALID)) {
883 struct inet6_ifaddr *ifp;
887 ifp = ipv6_get_ifaddr(&fl->fl6_src, (*dst)->dev, 1);
889 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
895 * We need to get the dst entry for the
896 * default router instead
899 memcpy(&fl_gw, fl, sizeof(struct flowi));
900 memset(&fl_gw.fl6_dst, 0, sizeof(struct in6_addr));
901 *dst = ip6_route_output(sk, &fl_gw);
902 if ((err = (*dst)->error))
903 goto out_err_release;
917 * ip6_dst_lookup - perform route lookup on flow
918 * @sk: socket which provides route info
919 * @dst: pointer to dst_entry * for result
920 * @fl: flow to lookup
922 * This function performs a route lookup on the given flow.
924 * It returns zero on success, or a standard errno code on error.
926 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
929 return ip6_dst_lookup_tail(sk, dst, fl);
931 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
934 * ip6_sk_dst_lookup - perform socket cached route lookup on flow
935 * @sk: socket which provides the dst cache and route info
936 * @dst: pointer to dst_entry * for result
937 * @fl: flow to lookup
939 * This function performs a route lookup on the given flow with the
940 * possibility of using the cached route in the socket if it is valid.
941 * It will take the socket dst lock when operating on the dst cache.
942 * As a result, this function can only be used in process context.
944 * It returns zero on success, or a standard errno code on error.
946 int ip6_sk_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
950 *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
951 *dst = ip6_sk_dst_check(sk, *dst, fl);
954 return ip6_dst_lookup_tail(sk, dst, fl);
956 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup);
958 static inline int ip6_ufo_append_data(struct sock *sk,
959 int getfrag(void *from, char *to, int offset, int len,
960 int odd, struct sk_buff *skb),
961 void *from, int length, int hh_len, int fragheaderlen,
962 int transhdrlen, int mtu,unsigned int flags)
968 /* There is support for UDP large send offload by network
969 * device, so create one single skb packet containing complete
972 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
973 skb = sock_alloc_send_skb(sk,
974 hh_len + fragheaderlen + transhdrlen + 20,
975 (flags & MSG_DONTWAIT), &err);
979 /* reserve space for Hardware header */
980 skb_reserve(skb, hh_len);
982 /* create space for UDP/IP header */
983 skb_put(skb,fragheaderlen + transhdrlen);
985 /* initialize network header pointer */
986 skb_reset_network_header(skb);
988 /* initialize protocol header pointer */
989 skb->h.raw = skb->data + fragheaderlen;
991 skb->ip_summed = CHECKSUM_PARTIAL;
993 sk->sk_sndmsg_off = 0;
996 err = skb_append_datato_frags(sk,skb, getfrag, from,
997 (length - transhdrlen));
999 struct frag_hdr fhdr;
1001 /* specify the length of each IP datagram fragment*/
1002 skb_shinfo(skb)->gso_size = mtu - fragheaderlen -
1003 sizeof(struct frag_hdr);
1004 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1005 ipv6_select_ident(skb, &fhdr);
1006 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1007 __skb_queue_tail(&sk->sk_write_queue, skb);
1011 /* There is not enough support do UPD LSO,
1012 * so follow normal path
1019 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1020 int offset, int len, int odd, struct sk_buff *skb),
1021 void *from, int length, int transhdrlen,
1022 int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi *fl,
1023 struct rt6_info *rt, unsigned int flags)
1025 struct inet_sock *inet = inet_sk(sk);
1026 struct ipv6_pinfo *np = inet6_sk(sk);
1027 struct sk_buff *skb;
1028 unsigned int maxfraglen, fragheaderlen;
1035 int csummode = CHECKSUM_NONE;
1037 if (flags&MSG_PROBE)
1039 if (skb_queue_empty(&sk->sk_write_queue)) {
1044 if (np->cork.opt == NULL) {
1045 np->cork.opt = kmalloc(opt->tot_len,
1047 if (unlikely(np->cork.opt == NULL))
1049 } else if (np->cork.opt->tot_len < opt->tot_len) {
1050 printk(KERN_DEBUG "ip6_append_data: invalid option length\n");
1053 memcpy(np->cork.opt, opt, opt->tot_len);
1054 inet->cork.flags |= IPCORK_OPT;
1055 /* need source address above miyazawa*/
1057 dst_hold(&rt->u.dst);
1059 inet->cork.fl = *fl;
1060 np->cork.hop_limit = hlimit;
1061 np->cork.tclass = tclass;
1062 mtu = dst_mtu(rt->u.dst.path);
1063 if (np->frag_size < mtu) {
1065 mtu = np->frag_size;
1067 inet->cork.fragsize = mtu;
1068 if (dst_allfrag(rt->u.dst.path))
1069 inet->cork.flags |= IPCORK_ALLFRAG;
1070 inet->cork.length = 0;
1071 sk->sk_sndmsg_page = NULL;
1072 sk->sk_sndmsg_off = 0;
1073 exthdrlen = rt->u.dst.header_len + (opt ? opt->opt_flen : 0);
1074 length += exthdrlen;
1075 transhdrlen += exthdrlen;
1078 fl = &inet->cork.fl;
1079 if (inet->cork.flags & IPCORK_OPT)
1083 mtu = inet->cork.fragsize;
1086 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1088 fragheaderlen = sizeof(struct ipv6hdr) + rt->u.dst.nfheader_len + (opt ? opt->opt_nflen : 0);
1089 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
1091 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1092 if (inet->cork.length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
1093 ipv6_local_error(sk, EMSGSIZE, fl, mtu-exthdrlen);
1099 * Let's try using as much space as possible.
1100 * Use MTU if total length of the message fits into the MTU.
1101 * Otherwise, we need to reserve fragment header and
1102 * fragment alignment (= 8-15 octects, in total).
1104 * Note that we may need to "move" the data from the tail of
1105 * of the buffer to the new fragment when we split
1108 * FIXME: It may be fragmented into multiple chunks
1109 * at once if non-fragmentable extension headers
1114 inet->cork.length += length;
1115 if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
1116 (rt->u.dst.dev->features & NETIF_F_UFO)) {
1118 err = ip6_ufo_append_data(sk, getfrag, from, length, hh_len,
1119 fragheaderlen, transhdrlen, mtu,
1126 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1129 while (length > 0) {
1130 /* Check if the remaining data fits into current packet. */
1131 copy = (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1133 copy = maxfraglen - skb->len;
1137 unsigned int datalen;
1138 unsigned int fraglen;
1139 unsigned int fraggap;
1140 unsigned int alloclen;
1141 struct sk_buff *skb_prev;
1145 /* There's no room in the current skb */
1147 fraggap = skb_prev->len - maxfraglen;
1152 * If remaining data exceeds the mtu,
1153 * we know we need more fragment(s).
1155 datalen = length + fraggap;
1156 if (datalen > (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1157 datalen = maxfraglen - fragheaderlen;
1159 fraglen = datalen + fragheaderlen;
1160 if ((flags & MSG_MORE) &&
1161 !(rt->u.dst.dev->features&NETIF_F_SG))
1164 alloclen = datalen + fragheaderlen;
1167 * The last fragment gets additional space at tail.
1168 * Note: we overallocate on fragments with MSG_MODE
1169 * because we have no idea if we're the last one.
1171 if (datalen == length + fraggap)
1172 alloclen += rt->u.dst.trailer_len;
1175 * We just reserve space for fragment header.
1176 * Note: this may be overallocation if the message
1177 * (without MSG_MORE) fits into the MTU.
1179 alloclen += sizeof(struct frag_hdr);
1182 skb = sock_alloc_send_skb(sk,
1184 (flags & MSG_DONTWAIT), &err);
1187 if (atomic_read(&sk->sk_wmem_alloc) <=
1189 skb = sock_wmalloc(sk,
1190 alloclen + hh_len, 1,
1192 if (unlikely(skb == NULL))
1198 * Fill in the control structures
1200 skb->ip_summed = csummode;
1202 /* reserve for fragmentation */
1203 skb_reserve(skb, hh_len+sizeof(struct frag_hdr));
1206 * Find where to start putting bytes
1208 data = skb_put(skb, fraglen);
1209 skb_set_network_header(skb, exthdrlen);
1210 data += fragheaderlen;
1211 skb->h.raw = skb->nh.raw + fragheaderlen;
1214 skb->csum = skb_copy_and_csum_bits(
1215 skb_prev, maxfraglen,
1216 data + transhdrlen, fraggap, 0);
1217 skb_prev->csum = csum_sub(skb_prev->csum,
1220 pskb_trim_unique(skb_prev, maxfraglen);
1222 copy = datalen - transhdrlen - fraggap;
1227 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1234 length -= datalen - fraggap;
1237 csummode = CHECKSUM_NONE;
1240 * Put the packet on the pending queue
1242 __skb_queue_tail(&sk->sk_write_queue, skb);
1249 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
1253 if (getfrag(from, skb_put(skb, copy),
1254 offset, copy, off, skb) < 0) {
1255 __skb_trim(skb, off);
1260 int i = skb_shinfo(skb)->nr_frags;
1261 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1262 struct page *page = sk->sk_sndmsg_page;
1263 int off = sk->sk_sndmsg_off;
1266 if (page && (left = PAGE_SIZE - off) > 0) {
1269 if (page != frag->page) {
1270 if (i == MAX_SKB_FRAGS) {
1275 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1276 frag = &skb_shinfo(skb)->frags[i];
1278 } else if(i < MAX_SKB_FRAGS) {
1279 if (copy > PAGE_SIZE)
1281 page = alloc_pages(sk->sk_allocation, 0);
1286 sk->sk_sndmsg_page = page;
1287 sk->sk_sndmsg_off = 0;
1289 skb_fill_page_desc(skb, i, page, 0, 0);
1290 frag = &skb_shinfo(skb)->frags[i];
1291 skb->truesize += PAGE_SIZE;
1292 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1297 if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1301 sk->sk_sndmsg_off += copy;
1304 skb->data_len += copy;
1311 inet->cork.length -= length;
1312 IP6_INC_STATS(rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1316 int ip6_push_pending_frames(struct sock *sk)
1318 struct sk_buff *skb, *tmp_skb;
1319 struct sk_buff **tail_skb;
1320 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1321 struct inet_sock *inet = inet_sk(sk);
1322 struct ipv6_pinfo *np = inet6_sk(sk);
1323 struct ipv6hdr *hdr;
1324 struct ipv6_txoptions *opt = np->cork.opt;
1325 struct rt6_info *rt = np->cork.rt;
1326 struct flowi *fl = &inet->cork.fl;
1327 unsigned char proto = fl->proto;
1330 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1332 tail_skb = &(skb_shinfo(skb)->frag_list);
1334 /* move skb->data to ip header from ext header */
1335 if (skb->data < skb_network_header(skb))
1336 __skb_pull(skb, skb_network_offset(skb));
1337 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1338 __skb_pull(tmp_skb, skb->h.raw - skb->nh.raw);
1339 *tail_skb = tmp_skb;
1340 tail_skb = &(tmp_skb->next);
1341 skb->len += tmp_skb->len;
1342 skb->data_len += tmp_skb->len;
1343 skb->truesize += tmp_skb->truesize;
1344 __sock_put(tmp_skb->sk);
1345 tmp_skb->destructor = NULL;
1349 ipv6_addr_copy(final_dst, &fl->fl6_dst);
1350 __skb_pull(skb, skb->h.raw - skb->nh.raw);
1351 if (opt && opt->opt_flen)
1352 ipv6_push_frag_opts(skb, opt, &proto);
1353 if (opt && opt->opt_nflen)
1354 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1356 skb_push(skb, sizeof(struct ipv6hdr));
1357 skb_reset_network_header(skb);
1358 hdr = skb->nh.ipv6h;
1360 *(__be32*)hdr = fl->fl6_flowlabel |
1361 htonl(0x60000000 | ((int)np->cork.tclass << 20));
1363 if (skb->len <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN)
1364 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
1366 hdr->payload_len = 0;
1367 hdr->hop_limit = np->cork.hop_limit;
1368 hdr->nexthdr = proto;
1369 ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
1370 ipv6_addr_copy(&hdr->daddr, final_dst);
1372 skb->priority = sk->sk_priority;
1374 skb->dst = dst_clone(&rt->u.dst);
1375 IP6_INC_STATS(rt->rt6i_idev, IPSTATS_MIB_OUTREQUESTS);
1376 err = NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, skb->dst->dev, dst_output);
1379 err = np->recverr ? net_xmit_errno(err) : 0;
1385 inet->cork.flags &= ~IPCORK_OPT;
1386 kfree(np->cork.opt);
1387 np->cork.opt = NULL;
1389 dst_release(&np->cork.rt->u.dst);
1391 inet->cork.flags &= ~IPCORK_ALLFRAG;
1393 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1399 void ip6_flush_pending_frames(struct sock *sk)
1401 struct inet_sock *inet = inet_sk(sk);
1402 struct ipv6_pinfo *np = inet6_sk(sk);
1403 struct sk_buff *skb;
1405 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1406 IP6_INC_STATS(ip6_dst_idev(skb->dst),
1407 IPSTATS_MIB_OUTDISCARDS);
1411 inet->cork.flags &= ~IPCORK_OPT;
1413 kfree(np->cork.opt);
1414 np->cork.opt = NULL;
1416 dst_release(&np->cork.rt->u.dst);
1418 inet->cork.flags &= ~IPCORK_ALLFRAG;
1420 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));