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(&ipv6_hdr(skb)->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, &ipv6_hdr(skb)->daddr,
116 &ipv6_hdr(skb)->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 (ipv6_hdr(skb)->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 skb_reset_network_header(skb);
267 skb_put(skb, sizeof(struct ipv6hdr));
270 *(__be32*)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);
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)) {
294 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
296 rawv6_rcv(last, skb2);
303 rawv6_rcv(last, skb);
304 read_unlock(&ip6_ra_lock);
307 read_unlock(&ip6_ra_lock);
311 static int ip6_forward_proxy_check(struct sk_buff *skb)
313 struct ipv6hdr *hdr = ipv6_hdr(skb);
314 u8 nexthdr = hdr->nexthdr;
317 if (ipv6_ext_hdr(nexthdr)) {
318 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr);
322 offset = sizeof(struct ipv6hdr);
324 if (nexthdr == IPPROTO_ICMPV6) {
325 struct icmp6hdr *icmp6;
327 if (!pskb_may_pull(skb, (skb_network_header(skb) +
328 offset + 1 - skb->data)))
331 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
333 switch (icmp6->icmp6_type) {
334 case NDISC_ROUTER_SOLICITATION:
335 case NDISC_ROUTER_ADVERTISEMENT:
336 case NDISC_NEIGHBOUR_SOLICITATION:
337 case NDISC_NEIGHBOUR_ADVERTISEMENT:
339 /* For reaction involving unicast neighbor discovery
340 * message destined to the proxied address, pass it to
350 * The proxying router can't forward traffic sent to a link-local
351 * address, so signal the sender and discard the packet. This
352 * behavior is clarified by the MIPv6 specification.
354 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
355 dst_link_failure(skb);
362 static inline int ip6_forward_finish(struct sk_buff *skb)
364 return dst_output(skb);
367 int ip6_forward(struct sk_buff *skb)
369 struct dst_entry *dst = skb->dst;
370 struct ipv6hdr *hdr = ipv6_hdr(skb);
371 struct inet6_skb_parm *opt = IP6CB(skb);
373 if (ipv6_devconf.forwarding == 0)
376 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
377 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
381 skb->ip_summed = CHECKSUM_NONE;
384 * We DO NOT make any processing on
385 * RA packets, pushing them to user level AS IS
386 * without ane WARRANTY that application will be able
387 * to interpret them. The reason is that we
388 * cannot make anything clever here.
390 * We are not end-node, so that if packet contains
391 * AH/ESP, we cannot make anything.
392 * Defragmentation also would be mistake, RA packets
393 * cannot be fragmented, because there is no warranty
394 * that different fragments will go along one path. --ANK
397 u8 *ptr = skb_network_header(skb) + opt->ra;
398 if (ip6_call_ra_chain(skb, (ptr[2]<<8) + ptr[3]))
403 * check and decrement ttl
405 if (hdr->hop_limit <= 1) {
406 /* Force OUTPUT device used as source address */
408 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT,
410 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INHDRERRORS);
416 /* XXX: idev->cnf.proxy_ndp? */
417 if (ipv6_devconf.proxy_ndp &&
418 pneigh_lookup(&nd_tbl, &hdr->daddr, skb->dev, 0)) {
419 int proxied = ip6_forward_proxy_check(skb);
421 return ip6_input(skb);
422 else if (proxied < 0) {
423 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
428 if (!xfrm6_route_forward(skb)) {
429 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
434 /* IPv6 specs say nothing about it, but it is clear that we cannot
435 send redirects to source routed frames.
437 if (skb->dev == dst->dev && dst->neighbour && opt->srcrt == 0) {
438 struct in6_addr *target = NULL;
440 struct neighbour *n = dst->neighbour;
443 * incoming and outgoing devices are the same
447 rt = (struct rt6_info *) dst;
448 if ((rt->rt6i_flags & RTF_GATEWAY))
449 target = (struct in6_addr*)&n->primary_key;
451 target = &hdr->daddr;
453 /* Limit redirects both by destination (here)
454 and by source (inside ndisc_send_redirect)
456 if (xrlim_allow(dst, 1*HZ))
457 ndisc_send_redirect(skb, n, target);
458 } else if (ipv6_addr_type(&hdr->saddr)&(IPV6_ADDR_MULTICAST|IPV6_ADDR_LOOPBACK
459 |IPV6_ADDR_LINKLOCAL)) {
460 /* This check is security critical. */
464 if (skb->len > dst_mtu(dst)) {
465 /* Again, force OUTPUT device used as source address */
467 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, dst_mtu(dst), skb->dev);
468 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INTOOBIGERRORS);
469 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_FRAGFAILS);
474 if (skb_cow(skb, dst->dev->hard_header_len)) {
475 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_OUTDISCARDS);
481 /* Mangling hops number delayed to point after skb COW */
485 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
486 return NF_HOOK(PF_INET6,NF_IP6_FORWARD, skb, skb->dev, dst->dev, ip6_forward_finish);
489 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
495 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
497 to->pkt_type = from->pkt_type;
498 to->priority = from->priority;
499 to->protocol = from->protocol;
500 dst_release(to->dst);
501 to->dst = dst_clone(from->dst);
503 to->mark = from->mark;
505 #ifdef CONFIG_NET_SCHED
506 to->tc_index = from->tc_index;
509 skb_copy_secmark(to, from);
512 int ip6_find_1stfragopt(struct sk_buff *skb, u8 **nexthdr)
514 u16 offset = sizeof(struct ipv6hdr);
515 struct ipv6_opt_hdr *exthdr =
516 (struct ipv6_opt_hdr *)(ipv6_hdr(skb) + 1);
517 unsigned int packet_len = skb->tail - skb->network_header;
519 *nexthdr = &ipv6_hdr(skb)->nexthdr;
521 while (offset + 1 <= packet_len) {
527 case NEXTHDR_ROUTING:
531 #ifdef CONFIG_IPV6_MIP6
532 if (ipv6_find_tlv(skb, offset, IPV6_TLV_HAO) >= 0)
542 offset += ipv6_optlen(exthdr);
543 *nexthdr = &exthdr->nexthdr;
544 exthdr = (struct ipv6_opt_hdr *)(skb_network_header(skb) +
550 EXPORT_SYMBOL_GPL(ip6_find_1stfragopt);
552 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
554 struct net_device *dev;
555 struct sk_buff *frag;
556 struct rt6_info *rt = (struct rt6_info*)skb->dst;
557 struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
558 struct ipv6hdr *tmp_hdr;
560 unsigned int mtu, hlen, left, len;
562 int ptr, offset = 0, err=0;
563 u8 *prevhdr, nexthdr = 0;
566 hlen = ip6_find_1stfragopt(skb, &prevhdr);
569 mtu = dst_mtu(&rt->u.dst);
570 if (np && np->frag_size < mtu) {
574 mtu -= hlen + sizeof(struct frag_hdr);
576 if (skb_shinfo(skb)->frag_list) {
577 int first_len = skb_pagelen(skb);
579 if (first_len - hlen > mtu ||
580 ((first_len - hlen) & 7) ||
584 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
585 /* Correct geometry. */
586 if (frag->len > mtu ||
587 ((frag->len & 7) && frag->next) ||
588 skb_headroom(frag) < hlen)
591 /* Partially cloned skb? */
592 if (skb_shared(frag))
599 frag->destructor = sock_wfree;
600 skb->truesize -= frag->truesize;
606 frag = skb_shinfo(skb)->frag_list;
607 skb_shinfo(skb)->frag_list = NULL;
610 *prevhdr = NEXTHDR_FRAGMENT;
611 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
613 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGFAILS);
617 __skb_pull(skb, hlen);
618 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
619 __skb_push(skb, hlen);
620 skb_reset_network_header(skb);
621 memcpy(skb_network_header(skb), tmp_hdr, hlen);
623 ipv6_select_ident(skb, fh);
624 fh->nexthdr = nexthdr;
626 fh->frag_off = htons(IP6_MF);
627 frag_id = fh->identification;
629 first_len = skb_pagelen(skb);
630 skb->data_len = first_len - skb_headlen(skb);
631 skb->len = first_len;
632 ipv6_hdr(skb)->payload_len = htons(first_len -
633 sizeof(struct ipv6hdr));
635 dst_hold(&rt->u.dst);
638 /* Prepare header of the next frame,
639 * before previous one went down. */
641 frag->ip_summed = CHECKSUM_NONE;
642 skb_reset_transport_header(frag);
643 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
644 __skb_push(frag, hlen);
645 skb_reset_network_header(frag);
646 memcpy(skb_network_header(frag), tmp_hdr,
648 offset += skb->len - hlen - sizeof(struct frag_hdr);
649 fh->nexthdr = nexthdr;
651 fh->frag_off = htons(offset);
652 if (frag->next != NULL)
653 fh->frag_off |= htons(IP6_MF);
654 fh->identification = frag_id;
655 ipv6_hdr(frag)->payload_len =
657 sizeof(struct ipv6hdr));
658 ip6_copy_metadata(frag, skb);
663 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGCREATES);
676 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGOKS);
677 dst_release(&rt->u.dst);
687 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGFAILS);
688 dst_release(&rt->u.dst);
693 left = skb->len - hlen; /* Space per frame */
694 ptr = hlen; /* Where to start from */
697 * Fragment the datagram.
700 *prevhdr = NEXTHDR_FRAGMENT;
703 * Keep copying data until we run out.
707 /* IF: it doesn't fit, use 'mtu' - the data space left */
710 /* IF: we are not sending upto and including the packet end
711 then align the next start on an eight byte boundary */
719 if ((frag = alloc_skb(len+hlen+sizeof(struct frag_hdr)+LL_RESERVED_SPACE(rt->u.dst.dev), GFP_ATOMIC)) == NULL) {
720 NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
721 IP6_INC_STATS(ip6_dst_idev(skb->dst),
722 IPSTATS_MIB_FRAGFAILS);
728 * Set up data on packet
731 ip6_copy_metadata(frag, skb);
732 skb_reserve(frag, LL_RESERVED_SPACE(rt->u.dst.dev));
733 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
734 skb_reset_network_header(frag);
735 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
736 frag->transport_header = (frag->network_header + hlen +
737 sizeof(struct frag_hdr));
740 * Charge the memory for the fragment to any owner
744 skb_set_owner_w(frag, skb->sk);
747 * Copy the packet header into the new buffer.
749 memcpy(skb_network_header(frag), skb->data, hlen);
752 * Build fragment header.
754 fh->nexthdr = nexthdr;
757 ipv6_select_ident(skb, fh);
758 frag_id = fh->identification;
760 fh->identification = frag_id;
763 * Copy a block of the IP datagram.
765 if (skb_copy_bits(skb, ptr, skb_transport_header(skb), len))
769 fh->frag_off = htons(offset);
771 fh->frag_off |= htons(IP6_MF);
772 ipv6_hdr(frag)->payload_len = htons(frag->len -
773 sizeof(struct ipv6hdr));
779 * Put this fragment into the sending queue.
785 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGCREATES);
787 IP6_INC_STATS(ip6_dst_idev(skb->dst),
788 IPSTATS_MIB_FRAGOKS);
793 IP6_INC_STATS(ip6_dst_idev(skb->dst),
794 IPSTATS_MIB_FRAGFAILS);
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,
811 struct ipv6_pinfo *np = inet6_sk(sk);
812 struct rt6_info *rt = (struct rt6_info *)dst;
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
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) ||
838 (fl->oif && fl->oif != dst->dev->ifindex)) {
847 static int ip6_dst_lookup_tail(struct sock *sk,
848 struct dst_entry **dst, struct flowi *fl)
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);
861 goto out_err_release;
864 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
866 * Here if the dst entry we've looked up
867 * has a neighbour entry that is in the INCOMPLETE
868 * state and the src address from the flow is
869 * marked as OPTIMISTIC, we release the found
870 * dst entry and replace it instead with the
871 * dst entry of the nexthop router
873 if (!((*dst)->neighbour->nud_state & NUD_VALID)) {
874 struct inet6_ifaddr *ifp;
878 ifp = ipv6_get_ifaddr(&fl->fl6_src, (*dst)->dev, 1);
880 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
886 * We need to get the dst entry for the
887 * default router instead
890 memcpy(&fl_gw, fl, sizeof(struct flowi));
891 memset(&fl_gw.fl6_dst, 0, sizeof(struct in6_addr));
892 *dst = ip6_route_output(sk, &fl_gw);
893 if ((err = (*dst)->error))
894 goto out_err_release;
908 * ip6_dst_lookup - perform route lookup on flow
909 * @sk: socket which provides route info
910 * @dst: pointer to dst_entry * for result
911 * @fl: flow to lookup
913 * This function performs a route lookup on the given flow.
915 * It returns zero on success, or a standard errno code on error.
917 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
920 return ip6_dst_lookup_tail(sk, dst, fl);
922 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
925 * ip6_sk_dst_lookup - perform socket cached route lookup on flow
926 * @sk: socket which provides the dst cache and route info
927 * @dst: pointer to dst_entry * for result
928 * @fl: flow to lookup
930 * This function performs a route lookup on the given flow with the
931 * possibility of using the cached route in the socket if it is valid.
932 * It will take the socket dst lock when operating on the dst cache.
933 * As a result, this function can only be used in process context.
935 * It returns zero on success, or a standard errno code on error.
937 int ip6_sk_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
941 *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
942 *dst = ip6_sk_dst_check(sk, *dst, fl);
945 return ip6_dst_lookup_tail(sk, dst, fl);
947 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup);
949 static inline int ip6_ufo_append_data(struct sock *sk,
950 int getfrag(void *from, char *to, int offset, int len,
951 int odd, struct sk_buff *skb),
952 void *from, int length, int hh_len, int fragheaderlen,
953 int transhdrlen, int mtu,unsigned int flags)
959 /* There is support for UDP large send offload by network
960 * device, so create one single skb packet containing complete
963 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
964 skb = sock_alloc_send_skb(sk,
965 hh_len + fragheaderlen + transhdrlen + 20,
966 (flags & MSG_DONTWAIT), &err);
970 /* reserve space for Hardware header */
971 skb_reserve(skb, hh_len);
973 /* create space for UDP/IP header */
974 skb_put(skb,fragheaderlen + transhdrlen);
976 /* initialize network header pointer */
977 skb_reset_network_header(skb);
979 /* initialize protocol header pointer */
980 skb->transport_header = skb->network_header + fragheaderlen;
982 skb->ip_summed = CHECKSUM_PARTIAL;
984 sk->sk_sndmsg_off = 0;
987 err = skb_append_datato_frags(sk,skb, getfrag, from,
988 (length - transhdrlen));
990 struct frag_hdr fhdr;
992 /* specify the length of each IP datagram fragment*/
993 skb_shinfo(skb)->gso_size = mtu - fragheaderlen -
994 sizeof(struct frag_hdr);
995 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
996 ipv6_select_ident(skb, &fhdr);
997 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
998 __skb_queue_tail(&sk->sk_write_queue, skb);
1002 /* There is not enough support do UPD LSO,
1003 * so follow normal path
1010 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1011 int offset, int len, int odd, struct sk_buff *skb),
1012 void *from, int length, int transhdrlen,
1013 int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi *fl,
1014 struct rt6_info *rt, unsigned int flags)
1016 struct inet_sock *inet = inet_sk(sk);
1017 struct ipv6_pinfo *np = inet6_sk(sk);
1018 struct sk_buff *skb;
1019 unsigned int maxfraglen, fragheaderlen;
1026 int csummode = CHECKSUM_NONE;
1028 if (flags&MSG_PROBE)
1030 if (skb_queue_empty(&sk->sk_write_queue)) {
1035 if (np->cork.opt == NULL) {
1036 np->cork.opt = kmalloc(opt->tot_len,
1038 if (unlikely(np->cork.opt == NULL))
1040 } else if (np->cork.opt->tot_len < opt->tot_len) {
1041 printk(KERN_DEBUG "ip6_append_data: invalid option length\n");
1044 memcpy(np->cork.opt, opt, opt->tot_len);
1045 inet->cork.flags |= IPCORK_OPT;
1046 /* need source address above miyazawa*/
1048 dst_hold(&rt->u.dst);
1050 inet->cork.fl = *fl;
1051 np->cork.hop_limit = hlimit;
1052 np->cork.tclass = tclass;
1053 mtu = dst_mtu(rt->u.dst.path);
1054 if (np->frag_size < mtu) {
1056 mtu = np->frag_size;
1058 inet->cork.fragsize = mtu;
1059 if (dst_allfrag(rt->u.dst.path))
1060 inet->cork.flags |= IPCORK_ALLFRAG;
1061 inet->cork.length = 0;
1062 sk->sk_sndmsg_page = NULL;
1063 sk->sk_sndmsg_off = 0;
1064 exthdrlen = rt->u.dst.header_len + (opt ? opt->opt_flen : 0);
1065 length += exthdrlen;
1066 transhdrlen += exthdrlen;
1069 fl = &inet->cork.fl;
1070 if (inet->cork.flags & IPCORK_OPT)
1074 mtu = inet->cork.fragsize;
1077 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1079 fragheaderlen = sizeof(struct ipv6hdr) + rt->u.dst.nfheader_len + (opt ? opt->opt_nflen : 0);
1080 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
1082 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1083 if (inet->cork.length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
1084 ipv6_local_error(sk, EMSGSIZE, fl, mtu-exthdrlen);
1090 * Let's try using as much space as possible.
1091 * Use MTU if total length of the message fits into the MTU.
1092 * Otherwise, we need to reserve fragment header and
1093 * fragment alignment (= 8-15 octects, in total).
1095 * Note that we may need to "move" the data from the tail of
1096 * of the buffer to the new fragment when we split
1099 * FIXME: It may be fragmented into multiple chunks
1100 * at once if non-fragmentable extension headers
1105 inet->cork.length += length;
1106 if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
1107 (rt->u.dst.dev->features & NETIF_F_UFO)) {
1109 err = ip6_ufo_append_data(sk, getfrag, from, length, hh_len,
1110 fragheaderlen, transhdrlen, mtu,
1117 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1120 while (length > 0) {
1121 /* Check if the remaining data fits into current packet. */
1122 copy = (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1124 copy = maxfraglen - skb->len;
1128 unsigned int datalen;
1129 unsigned int fraglen;
1130 unsigned int fraggap;
1131 unsigned int alloclen;
1132 struct sk_buff *skb_prev;
1136 /* There's no room in the current skb */
1138 fraggap = skb_prev->len - maxfraglen;
1143 * If remaining data exceeds the mtu,
1144 * we know we need more fragment(s).
1146 datalen = length + fraggap;
1147 if (datalen > (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1148 datalen = maxfraglen - fragheaderlen;
1150 fraglen = datalen + fragheaderlen;
1151 if ((flags & MSG_MORE) &&
1152 !(rt->u.dst.dev->features&NETIF_F_SG))
1155 alloclen = datalen + fragheaderlen;
1158 * The last fragment gets additional space at tail.
1159 * Note: we overallocate on fragments with MSG_MODE
1160 * because we have no idea if we're the last one.
1162 if (datalen == length + fraggap)
1163 alloclen += rt->u.dst.trailer_len;
1166 * We just reserve space for fragment header.
1167 * Note: this may be overallocation if the message
1168 * (without MSG_MORE) fits into the MTU.
1170 alloclen += sizeof(struct frag_hdr);
1173 skb = sock_alloc_send_skb(sk,
1175 (flags & MSG_DONTWAIT), &err);
1178 if (atomic_read(&sk->sk_wmem_alloc) <=
1180 skb = sock_wmalloc(sk,
1181 alloclen + hh_len, 1,
1183 if (unlikely(skb == NULL))
1189 * Fill in the control structures
1191 skb->ip_summed = csummode;
1193 /* reserve for fragmentation */
1194 skb_reserve(skb, hh_len+sizeof(struct frag_hdr));
1197 * Find where to start putting bytes
1199 data = skb_put(skb, fraglen);
1200 skb_set_network_header(skb, exthdrlen);
1201 data += fragheaderlen;
1202 skb->transport_header = (skb->network_header +
1205 skb->csum = skb_copy_and_csum_bits(
1206 skb_prev, maxfraglen,
1207 data + transhdrlen, fraggap, 0);
1208 skb_prev->csum = csum_sub(skb_prev->csum,
1211 pskb_trim_unique(skb_prev, maxfraglen);
1213 copy = datalen - transhdrlen - fraggap;
1218 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1225 length -= datalen - fraggap;
1228 csummode = CHECKSUM_NONE;
1231 * Put the packet on the pending queue
1233 __skb_queue_tail(&sk->sk_write_queue, skb);
1240 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
1244 if (getfrag(from, skb_put(skb, copy),
1245 offset, copy, off, skb) < 0) {
1246 __skb_trim(skb, off);
1251 int i = skb_shinfo(skb)->nr_frags;
1252 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1253 struct page *page = sk->sk_sndmsg_page;
1254 int off = sk->sk_sndmsg_off;
1257 if (page && (left = PAGE_SIZE - off) > 0) {
1260 if (page != frag->page) {
1261 if (i == MAX_SKB_FRAGS) {
1266 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1267 frag = &skb_shinfo(skb)->frags[i];
1269 } else if(i < MAX_SKB_FRAGS) {
1270 if (copy > PAGE_SIZE)
1272 page = alloc_pages(sk->sk_allocation, 0);
1277 sk->sk_sndmsg_page = page;
1278 sk->sk_sndmsg_off = 0;
1280 skb_fill_page_desc(skb, i, page, 0, 0);
1281 frag = &skb_shinfo(skb)->frags[i];
1282 skb->truesize += PAGE_SIZE;
1283 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1288 if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1292 sk->sk_sndmsg_off += copy;
1295 skb->data_len += copy;
1302 inet->cork.length -= length;
1303 IP6_INC_STATS(rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1307 int ip6_push_pending_frames(struct sock *sk)
1309 struct sk_buff *skb, *tmp_skb;
1310 struct sk_buff **tail_skb;
1311 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1312 struct inet_sock *inet = inet_sk(sk);
1313 struct ipv6_pinfo *np = inet6_sk(sk);
1314 struct ipv6hdr *hdr;
1315 struct ipv6_txoptions *opt = np->cork.opt;
1316 struct rt6_info *rt = np->cork.rt;
1317 struct flowi *fl = &inet->cork.fl;
1318 unsigned char proto = fl->proto;
1321 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1323 tail_skb = &(skb_shinfo(skb)->frag_list);
1325 /* move skb->data to ip header from ext header */
1326 if (skb->data < skb_network_header(skb))
1327 __skb_pull(skb, skb_network_offset(skb));
1328 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1329 __skb_pull(tmp_skb, skb_network_header_len(skb));
1330 *tail_skb = tmp_skb;
1331 tail_skb = &(tmp_skb->next);
1332 skb->len += tmp_skb->len;
1333 skb->data_len += tmp_skb->len;
1334 skb->truesize += tmp_skb->truesize;
1335 __sock_put(tmp_skb->sk);
1336 tmp_skb->destructor = NULL;
1340 ipv6_addr_copy(final_dst, &fl->fl6_dst);
1341 __skb_pull(skb, skb_network_header_len(skb));
1342 if (opt && opt->opt_flen)
1343 ipv6_push_frag_opts(skb, opt, &proto);
1344 if (opt && opt->opt_nflen)
1345 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1347 skb_push(skb, sizeof(struct ipv6hdr));
1348 skb_reset_network_header(skb);
1349 hdr = ipv6_hdr(skb);
1351 *(__be32*)hdr = fl->fl6_flowlabel |
1352 htonl(0x60000000 | ((int)np->cork.tclass << 20));
1354 if (skb->len <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN)
1355 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
1357 hdr->payload_len = 0;
1358 hdr->hop_limit = np->cork.hop_limit;
1359 hdr->nexthdr = proto;
1360 ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
1361 ipv6_addr_copy(&hdr->daddr, final_dst);
1363 skb->priority = sk->sk_priority;
1365 skb->dst = dst_clone(&rt->u.dst);
1366 IP6_INC_STATS(rt->rt6i_idev, IPSTATS_MIB_OUTREQUESTS);
1367 err = NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, skb->dst->dev, dst_output);
1370 err = np->recverr ? net_xmit_errno(err) : 0;
1376 inet->cork.flags &= ~IPCORK_OPT;
1377 kfree(np->cork.opt);
1378 np->cork.opt = NULL;
1380 dst_release(&np->cork.rt->u.dst);
1382 inet->cork.flags &= ~IPCORK_ALLFRAG;
1384 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1390 void ip6_flush_pending_frames(struct sock *sk)
1392 struct inet_sock *inet = inet_sk(sk);
1393 struct ipv6_pinfo *np = inet6_sk(sk);
1394 struct sk_buff *skb;
1396 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1397 IP6_INC_STATS(ip6_dst_idev(skb->dst),
1398 IPSTATS_MIB_OUTDISCARDS);
1402 inet->cork.flags &= ~IPCORK_OPT;
1404 kfree(np->cork.opt);
1405 np->cork.opt = NULL;
1407 dst_release(&np->cork.rt->u.dst);
1409 inet->cork.flags &= ~IPCORK_ALLFRAG;
1411 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));