2 * IPv6 output functions
3 * Linux INET6 implementation
6 * Pedro Roque <roque@di.fc.ul.pt>
8 * Based on linux/net/ipv4/ip_output.c
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
16 * A.N.Kuznetsov : airthmetics in fragmentation.
17 * extension headers are implemented.
18 * route changes now work.
19 * ip6_forward does not confuse sniffers.
22 * H. von Brand : Added missing #include <linux/string.h>
23 * Imran Patel : frag id should be in NBO
24 * Kazunori MIYAZAWA @USAGI
25 * : add ip6_append_data and related functions
29 #include <linux/errno.h>
30 #include <linux/kernel.h>
31 #include <linux/string.h>
32 #include <linux/socket.h>
33 #include <linux/net.h>
34 #include <linux/netdevice.h>
35 #include <linux/if_arp.h>
36 #include <linux/in6.h>
37 #include <linux/tcp.h>
38 #include <linux/route.h>
39 #include <linux/module.h>
40 #include <linux/slab.h>
42 #include <linux/netfilter.h>
43 #include <linux/netfilter_ipv6.h>
49 #include <net/ndisc.h>
50 #include <net/protocol.h>
51 #include <net/ip6_route.h>
52 #include <net/addrconf.h>
53 #include <net/rawv6.h>
56 #include <net/checksum.h>
57 #include <linux/mroute6.h>
59 static int ip6_finish_output2(struct sock *sk, struct sk_buff *skb)
61 struct dst_entry *dst = skb_dst(skb);
62 struct net_device *dev = dst->dev;
63 struct neighbour *neigh;
64 struct in6_addr *nexthop;
67 skb->protocol = htons(ETH_P_IPV6);
70 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
71 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
73 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(sk) &&
74 ((mroute6_socket(dev_net(dev), skb) &&
75 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
76 ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
77 &ipv6_hdr(skb)->saddr))) {
78 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
80 /* Do not check for IFF_ALLMULTI; multicast routing
81 is not supported in any case.
84 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
85 sk, newskb, NULL, newskb->dev,
88 if (ipv6_hdr(skb)->hop_limit == 0) {
89 IP6_INC_STATS(dev_net(dev), idev,
90 IPSTATS_MIB_OUTDISCARDS);
96 IP6_UPD_PO_STATS(dev_net(dev), idev, IPSTATS_MIB_OUTMCAST,
99 if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb)->daddr) <=
100 IPV6_ADDR_SCOPE_NODELOCAL &&
101 !(dev->flags & IFF_LOOPBACK)) {
108 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
109 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
110 if (unlikely(!neigh))
111 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
112 if (!IS_ERR(neigh)) {
113 ret = dst_neigh_output(dst, neigh, skb);
114 rcu_read_unlock_bh();
117 rcu_read_unlock_bh();
119 IP6_INC_STATS(dev_net(dst->dev),
120 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
125 static int ip6_finish_output(struct sock *sk, struct sk_buff *skb)
127 if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
128 dst_allfrag(skb_dst(skb)) ||
129 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
130 return ip6_fragment(sk, skb, ip6_finish_output2);
132 return ip6_finish_output2(sk, skb);
135 int ip6_output(struct sock *sk, struct sk_buff *skb)
137 struct net_device *dev = skb_dst(skb)->dev;
138 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
139 if (unlikely(idev->cnf.disable_ipv6)) {
140 IP6_INC_STATS(dev_net(dev), idev,
141 IPSTATS_MIB_OUTDISCARDS);
146 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, sk, skb,
149 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
153 * xmit an sk_buff (used by TCP, SCTP and DCCP)
156 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
157 struct ipv6_txoptions *opt, int tclass)
159 struct net *net = sock_net(sk);
160 struct ipv6_pinfo *np = inet6_sk(sk);
161 struct in6_addr *first_hop = &fl6->daddr;
162 struct dst_entry *dst = skb_dst(skb);
164 u8 proto = fl6->flowi6_proto;
165 int seg_len = skb->len;
170 unsigned int head_room;
172 /* First: exthdrs may take lots of space (~8K for now)
173 MAX_HEADER is not enough.
175 head_room = opt->opt_nflen + opt->opt_flen;
176 seg_len += head_room;
177 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
179 if (skb_headroom(skb) < head_room) {
180 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
182 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
183 IPSTATS_MIB_OUTDISCARDS);
189 skb_set_owner_w(skb, sk);
192 ipv6_push_frag_opts(skb, opt, &proto);
194 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
197 skb_push(skb, sizeof(struct ipv6hdr));
198 skb_reset_network_header(skb);
202 * Fill in the IPv6 header
205 hlimit = np->hop_limit;
207 hlimit = ip6_dst_hoplimit(dst);
209 ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
210 np->autoflowlabel, fl6));
212 hdr->payload_len = htons(seg_len);
213 hdr->nexthdr = proto;
214 hdr->hop_limit = hlimit;
216 hdr->saddr = fl6->saddr;
217 hdr->daddr = *first_hop;
219 skb->protocol = htons(ETH_P_IPV6);
220 skb->priority = sk->sk_priority;
221 skb->mark = sk->sk_mark;
224 if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
225 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
226 IPSTATS_MIB_OUT, skb->len);
227 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, sk, skb,
228 NULL, dst->dev, dst_output_sk);
232 ipv6_local_error(sk, EMSGSIZE, fl6, mtu);
233 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
237 EXPORT_SYMBOL(ip6_xmit);
239 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
241 struct ip6_ra_chain *ra;
242 struct sock *last = NULL;
244 read_lock(&ip6_ra_lock);
245 for (ra = ip6_ra_chain; ra; ra = ra->next) {
246 struct sock *sk = ra->sk;
247 if (sk && ra->sel == sel &&
248 (!sk->sk_bound_dev_if ||
249 sk->sk_bound_dev_if == skb->dev->ifindex)) {
251 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
253 rawv6_rcv(last, skb2);
260 rawv6_rcv(last, skb);
261 read_unlock(&ip6_ra_lock);
264 read_unlock(&ip6_ra_lock);
268 static int ip6_forward_proxy_check(struct sk_buff *skb)
270 struct ipv6hdr *hdr = ipv6_hdr(skb);
271 u8 nexthdr = hdr->nexthdr;
275 if (ipv6_ext_hdr(nexthdr)) {
276 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
280 offset = sizeof(struct ipv6hdr);
282 if (nexthdr == IPPROTO_ICMPV6) {
283 struct icmp6hdr *icmp6;
285 if (!pskb_may_pull(skb, (skb_network_header(skb) +
286 offset + 1 - skb->data)))
289 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
291 switch (icmp6->icmp6_type) {
292 case NDISC_ROUTER_SOLICITATION:
293 case NDISC_ROUTER_ADVERTISEMENT:
294 case NDISC_NEIGHBOUR_SOLICITATION:
295 case NDISC_NEIGHBOUR_ADVERTISEMENT:
297 /* For reaction involving unicast neighbor discovery
298 * message destined to the proxied address, pass it to
308 * The proxying router can't forward traffic sent to a link-local
309 * address, so signal the sender and discard the packet. This
310 * behavior is clarified by the MIPv6 specification.
312 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
313 dst_link_failure(skb);
320 static inline int ip6_forward_finish(struct sock *sk, struct sk_buff *skb)
322 skb_sender_cpu_clear(skb);
323 return dst_output_sk(sk, skb);
326 static unsigned int ip6_dst_mtu_forward(const struct dst_entry *dst)
329 struct inet6_dev *idev;
331 if (dst_metric_locked(dst, RTAX_MTU)) {
332 mtu = dst_metric_raw(dst, RTAX_MTU);
339 idev = __in6_dev_get(dst->dev);
341 mtu = idev->cnf.mtu6;
347 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
352 /* ipv6 conntrack defrag sets max_frag_size + ignore_df */
353 if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
359 if (skb_is_gso(skb) && skb_gso_network_seglen(skb) <= mtu)
365 int ip6_forward(struct sk_buff *skb)
367 struct dst_entry *dst = skb_dst(skb);
368 struct ipv6hdr *hdr = ipv6_hdr(skb);
369 struct inet6_skb_parm *opt = IP6CB(skb);
370 struct net *net = dev_net(dst->dev);
373 if (net->ipv6.devconf_all->forwarding == 0)
376 if (skb->pkt_type != PACKET_HOST)
379 if (unlikely(skb->sk))
382 if (skb_warn_if_lro(skb))
385 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
386 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
387 IPSTATS_MIB_INDISCARDS);
391 skb_forward_csum(skb);
394 * We DO NOT make any processing on
395 * RA packets, pushing them to user level AS IS
396 * without ane WARRANTY that application will be able
397 * to interpret them. The reason is that we
398 * cannot make anything clever here.
400 * We are not end-node, so that if packet contains
401 * AH/ESP, we cannot make anything.
402 * Defragmentation also would be mistake, RA packets
403 * cannot be fragmented, because there is no warranty
404 * that different fragments will go along one path. --ANK
406 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
407 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
412 * check and decrement ttl
414 if (hdr->hop_limit <= 1) {
415 /* Force OUTPUT device used as source address */
417 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
418 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
419 IPSTATS_MIB_INHDRERRORS);
425 /* XXX: idev->cnf.proxy_ndp? */
426 if (net->ipv6.devconf_all->proxy_ndp &&
427 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
428 int proxied = ip6_forward_proxy_check(skb);
430 return ip6_input(skb);
431 else if (proxied < 0) {
432 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
433 IPSTATS_MIB_INDISCARDS);
438 if (!xfrm6_route_forward(skb)) {
439 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
440 IPSTATS_MIB_INDISCARDS);
445 /* IPv6 specs say nothing about it, but it is clear that we cannot
446 send redirects to source routed frames.
447 We don't send redirects to frames decapsulated from IPsec.
449 if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
450 struct in6_addr *target = NULL;
451 struct inet_peer *peer;
455 * incoming and outgoing devices are the same
459 rt = (struct rt6_info *) dst;
460 if (rt->rt6i_flags & RTF_GATEWAY)
461 target = &rt->rt6i_gateway;
463 target = &hdr->daddr;
465 peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
467 /* Limit redirects both by destination (here)
468 and by source (inside ndisc_send_redirect)
470 if (inet_peer_xrlim_allow(peer, 1*HZ))
471 ndisc_send_redirect(skb, target);
475 int addrtype = ipv6_addr_type(&hdr->saddr);
477 /* This check is security critical. */
478 if (addrtype == IPV6_ADDR_ANY ||
479 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
481 if (addrtype & IPV6_ADDR_LINKLOCAL) {
482 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
483 ICMPV6_NOT_NEIGHBOUR, 0);
488 mtu = ip6_dst_mtu_forward(dst);
489 if (mtu < IPV6_MIN_MTU)
492 if (ip6_pkt_too_big(skb, mtu)) {
493 /* Again, force OUTPUT device used as source address */
495 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
496 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
497 IPSTATS_MIB_INTOOBIGERRORS);
498 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
499 IPSTATS_MIB_FRAGFAILS);
504 if (skb_cow(skb, dst->dev->hard_header_len)) {
505 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
506 IPSTATS_MIB_OUTDISCARDS);
512 /* Mangling hops number delayed to point after skb COW */
516 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
517 IP6_ADD_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
518 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD, NULL, skb,
523 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
529 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
531 to->pkt_type = from->pkt_type;
532 to->priority = from->priority;
533 to->protocol = from->protocol;
535 skb_dst_set(to, dst_clone(skb_dst(from)));
537 to->mark = from->mark;
539 #ifdef CONFIG_NET_SCHED
540 to->tc_index = from->tc_index;
543 skb_copy_secmark(to, from);
546 int ip6_fragment(struct sock *sk, struct sk_buff *skb,
547 int (*output)(struct sock *, struct sk_buff *))
549 struct sk_buff *frag;
550 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
551 struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
552 inet6_sk(skb->sk) : NULL;
553 struct ipv6hdr *tmp_hdr;
555 unsigned int mtu, hlen, left, len;
558 int ptr, offset = 0, err = 0;
559 u8 *prevhdr, nexthdr = 0;
560 struct net *net = dev_net(skb_dst(skb)->dev);
562 hlen = ip6_find_1stfragopt(skb, &prevhdr);
565 mtu = ip6_skb_dst_mtu(skb);
567 /* We must not fragment if the socket is set to force MTU discovery
568 * or if the skb it not generated by a local socket.
570 if (unlikely(!skb->ignore_df && skb->len > mtu))
573 if (IP6CB(skb)->frag_max_size) {
574 if (IP6CB(skb)->frag_max_size > mtu)
577 /* don't send fragments larger than what we received */
578 mtu = IP6CB(skb)->frag_max_size;
579 if (mtu < IPV6_MIN_MTU)
583 if (np && np->frag_size < mtu) {
587 mtu -= hlen + sizeof(struct frag_hdr);
589 frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
590 &ipv6_hdr(skb)->saddr);
592 hroom = LL_RESERVED_SPACE(rt->dst.dev);
593 if (skb_has_frag_list(skb)) {
594 int first_len = skb_pagelen(skb);
595 struct sk_buff *frag2;
597 if (first_len - hlen > mtu ||
598 ((first_len - hlen) & 7) ||
600 skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
603 skb_walk_frags(skb, frag) {
604 /* Correct geometry. */
605 if (frag->len > mtu ||
606 ((frag->len & 7) && frag->next) ||
607 skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
608 goto slow_path_clean;
610 /* Partially cloned skb? */
611 if (skb_shared(frag))
612 goto slow_path_clean;
617 frag->destructor = sock_wfree;
619 skb->truesize -= frag->truesize;
626 *prevhdr = NEXTHDR_FRAGMENT;
627 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
629 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
630 IPSTATS_MIB_FRAGFAILS);
634 frag = skb_shinfo(skb)->frag_list;
635 skb_frag_list_init(skb);
637 __skb_pull(skb, hlen);
638 fh = (struct frag_hdr *)__skb_push(skb, sizeof(struct frag_hdr));
639 __skb_push(skb, hlen);
640 skb_reset_network_header(skb);
641 memcpy(skb_network_header(skb), tmp_hdr, hlen);
643 fh->nexthdr = nexthdr;
645 fh->frag_off = htons(IP6_MF);
646 fh->identification = frag_id;
648 first_len = skb_pagelen(skb);
649 skb->data_len = first_len - skb_headlen(skb);
650 skb->len = first_len;
651 ipv6_hdr(skb)->payload_len = htons(first_len -
652 sizeof(struct ipv6hdr));
657 /* Prepare header of the next frame,
658 * before previous one went down. */
660 frag->ip_summed = CHECKSUM_NONE;
661 skb_reset_transport_header(frag);
662 fh = (struct frag_hdr *)__skb_push(frag, sizeof(struct frag_hdr));
663 __skb_push(frag, hlen);
664 skb_reset_network_header(frag);
665 memcpy(skb_network_header(frag), tmp_hdr,
667 offset += skb->len - hlen - sizeof(struct frag_hdr);
668 fh->nexthdr = nexthdr;
670 fh->frag_off = htons(offset);
672 fh->frag_off |= htons(IP6_MF);
673 fh->identification = frag_id;
674 ipv6_hdr(frag)->payload_len =
676 sizeof(struct ipv6hdr));
677 ip6_copy_metadata(frag, skb);
680 err = output(sk, skb);
682 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
683 IPSTATS_MIB_FRAGCREATES);
696 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
697 IPSTATS_MIB_FRAGOKS);
702 kfree_skb_list(frag);
704 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
705 IPSTATS_MIB_FRAGFAILS);
710 skb_walk_frags(skb, frag2) {
714 frag2->destructor = NULL;
715 skb->truesize += frag2->truesize;
720 if ((skb->ip_summed == CHECKSUM_PARTIAL) &&
721 skb_checksum_help(skb))
724 left = skb->len - hlen; /* Space per frame */
725 ptr = hlen; /* Where to start from */
728 * Fragment the datagram.
731 *prevhdr = NEXTHDR_FRAGMENT;
732 troom = rt->dst.dev->needed_tailroom;
735 * Keep copying data until we run out.
739 /* IF: it doesn't fit, use 'mtu' - the data space left */
742 /* IF: we are not sending up to and including the packet end
743 then align the next start on an eight byte boundary */
748 /* Allocate buffer */
749 frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
750 hroom + troom, GFP_ATOMIC);
752 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
753 IPSTATS_MIB_FRAGFAILS);
759 * Set up data on packet
762 ip6_copy_metadata(frag, skb);
763 skb_reserve(frag, hroom);
764 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
765 skb_reset_network_header(frag);
766 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
767 frag->transport_header = (frag->network_header + hlen +
768 sizeof(struct frag_hdr));
771 * Charge the memory for the fragment to any owner
775 skb_set_owner_w(frag, skb->sk);
778 * Copy the packet header into the new buffer.
780 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
783 * Build fragment header.
785 fh->nexthdr = nexthdr;
787 fh->identification = frag_id;
790 * Copy a block of the IP datagram.
792 BUG_ON(skb_copy_bits(skb, ptr, skb_transport_header(frag),
796 fh->frag_off = htons(offset);
798 fh->frag_off |= htons(IP6_MF);
799 ipv6_hdr(frag)->payload_len = htons(frag->len -
800 sizeof(struct ipv6hdr));
806 * Put this fragment into the sending queue.
808 err = output(sk, frag);
812 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
813 IPSTATS_MIB_FRAGCREATES);
815 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
816 IPSTATS_MIB_FRAGOKS);
821 if (skb->sk && dst_allfrag(skb_dst(skb)))
822 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
824 skb->dev = skb_dst(skb)->dev;
825 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
829 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
830 IPSTATS_MIB_FRAGFAILS);
835 static inline int ip6_rt_check(const struct rt6key *rt_key,
836 const struct in6_addr *fl_addr,
837 const struct in6_addr *addr_cache)
839 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
840 (!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
843 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
844 struct dst_entry *dst,
845 const struct flowi6 *fl6)
847 struct ipv6_pinfo *np = inet6_sk(sk);
853 if (dst->ops->family != AF_INET6) {
858 rt = (struct rt6_info *)dst;
859 /* Yes, checking route validity in not connected
860 * case is not very simple. Take into account,
861 * that we do not support routing by source, TOS,
862 * and MSG_DONTROUTE --ANK (980726)
864 * 1. ip6_rt_check(): If route was host route,
865 * check that cached destination is current.
866 * If it is network route, we still may
867 * check its validity using saved pointer
868 * to the last used address: daddr_cache.
869 * We do not want to save whole address now,
870 * (because main consumer of this service
871 * is tcp, which has not this problem),
872 * so that the last trick works only on connected
874 * 2. oif also should be the same.
876 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
877 #ifdef CONFIG_IPV6_SUBTREES
878 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
880 (!(fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF) &&
881 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex))) {
890 static int ip6_dst_lookup_tail(struct net *net, struct sock *sk,
891 struct dst_entry **dst, struct flowi6 *fl6)
893 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
899 /* The correct way to handle this would be to do
900 * ip6_route_get_saddr, and then ip6_route_output; however,
901 * the route-specific preferred source forces the
902 * ip6_route_output call _before_ ip6_route_get_saddr.
904 * In source specific routing (no src=any default route),
905 * ip6_route_output will fail given src=any saddr, though, so
906 * that's why we try it again later.
908 if (ipv6_addr_any(&fl6->saddr) && (!*dst || !(*dst)->error)) {
910 bool had_dst = *dst != NULL;
913 *dst = ip6_route_output(net, sk, fl6);
914 rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
915 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
916 sk ? inet6_sk(sk)->srcprefs : 0,
919 goto out_err_release;
921 /* If we had an erroneous initial result, pretend it
922 * never existed and let the SA-enabled version take
925 if (!had_dst && (*dst)->error) {
932 *dst = ip6_route_output(net, sk, fl6);
936 goto out_err_release;
938 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
940 * Here if the dst entry we've looked up
941 * has a neighbour entry that is in the INCOMPLETE
942 * state and the src address from the flow is
943 * marked as OPTIMISTIC, we release the found
944 * dst entry and replace it instead with the
945 * dst entry of the nexthop router
947 rt = (struct rt6_info *) *dst;
949 n = __ipv6_neigh_lookup_noref(rt->dst.dev,
950 rt6_nexthop(rt, &fl6->daddr));
951 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
952 rcu_read_unlock_bh();
955 struct inet6_ifaddr *ifp;
956 struct flowi6 fl_gw6;
959 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
962 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
968 * We need to get the dst entry for the
969 * default router instead
972 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
973 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
974 *dst = ip6_route_output(net, sk, &fl_gw6);
977 goto out_err_release;
985 if (err == -ENETUNREACH)
986 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
993 * ip6_dst_lookup - perform route lookup on flow
994 * @sk: socket which provides route info
995 * @dst: pointer to dst_entry * for result
996 * @fl6: flow to lookup
998 * This function performs a route lookup on the given flow.
1000 * It returns zero on success, or a standard errno code on error.
1002 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1006 return ip6_dst_lookup_tail(net, sk, dst, fl6);
1008 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1011 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1012 * @sk: socket which provides route info
1013 * @fl6: flow to lookup
1014 * @final_dst: final destination address for ipsec lookup
1016 * This function performs a route lookup on the given flow.
1018 * It returns a valid dst pointer on success, or a pointer encoded
1021 struct dst_entry *ip6_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1022 const struct in6_addr *final_dst)
1024 struct dst_entry *dst = NULL;
1027 err = ip6_dst_lookup_tail(sock_net(sk), sk, &dst, fl6);
1029 return ERR_PTR(err);
1031 fl6->daddr = *final_dst;
1032 if (!fl6->flowi6_oif)
1033 fl6->flowi6_oif = dst->dev->ifindex;
1035 return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1037 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1040 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1041 * @sk: socket which provides the dst cache and route info
1042 * @fl6: flow to lookup
1043 * @final_dst: final destination address for ipsec lookup
1045 * This function performs a route lookup on the given flow with the
1046 * possibility of using the cached route in the socket if it is valid.
1047 * It will take the socket dst lock when operating on the dst cache.
1048 * As a result, this function can only be used in process context.
1050 * It returns a valid dst pointer on success, or a pointer encoded
1053 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1054 const struct in6_addr *final_dst)
1056 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1059 dst = ip6_sk_dst_check(sk, dst, fl6);
1061 err = ip6_dst_lookup_tail(sock_net(sk), sk, &dst, fl6);
1063 return ERR_PTR(err);
1065 fl6->daddr = *final_dst;
1067 return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1069 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1071 static inline int ip6_ufo_append_data(struct sock *sk,
1072 struct sk_buff_head *queue,
1073 int getfrag(void *from, char *to, int offset, int len,
1074 int odd, struct sk_buff *skb),
1075 void *from, int length, int hh_len, int fragheaderlen,
1076 int transhdrlen, int mtu, unsigned int flags,
1077 const struct flowi6 *fl6)
1080 struct sk_buff *skb;
1083 /* There is support for UDP large send offload by network
1084 * device, so create one single skb packet containing complete
1087 skb = skb_peek_tail(queue);
1089 skb = sock_alloc_send_skb(sk,
1090 hh_len + fragheaderlen + transhdrlen + 20,
1091 (flags & MSG_DONTWAIT), &err);
1095 /* reserve space for Hardware header */
1096 skb_reserve(skb, hh_len);
1098 /* create space for UDP/IP header */
1099 skb_put(skb, fragheaderlen + transhdrlen);
1101 /* initialize network header pointer */
1102 skb_reset_network_header(skb);
1104 /* initialize protocol header pointer */
1105 skb->transport_header = skb->network_header + fragheaderlen;
1107 skb->protocol = htons(ETH_P_IPV6);
1110 __skb_queue_tail(queue, skb);
1111 } else if (skb_is_gso(skb)) {
1115 skb->ip_summed = CHECKSUM_PARTIAL;
1116 /* Specify the length of each IPv6 datagram fragment.
1117 * It has to be a multiple of 8.
1119 skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1120 sizeof(struct frag_hdr)) & ~7;
1121 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1122 skb_shinfo(skb)->ip6_frag_id = ipv6_select_ident(sock_net(sk),
1127 return skb_append_datato_frags(sk, skb, getfrag, from,
1128 (length - transhdrlen));
1131 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1134 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1137 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1140 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1143 static void ip6_append_data_mtu(unsigned int *mtu,
1145 unsigned int fragheaderlen,
1146 struct sk_buff *skb,
1147 struct rt6_info *rt,
1148 unsigned int orig_mtu)
1150 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1152 /* first fragment, reserve header_len */
1153 *mtu = orig_mtu - rt->dst.header_len;
1157 * this fragment is not first, the headers
1158 * space is regarded as data space.
1162 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1163 + fragheaderlen - sizeof(struct frag_hdr);
1167 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1168 struct inet6_cork *v6_cork,
1169 int hlimit, int tclass, struct ipv6_txoptions *opt,
1170 struct rt6_info *rt, struct flowi6 *fl6)
1172 struct ipv6_pinfo *np = inet6_sk(sk);
1179 if (WARN_ON(v6_cork->opt))
1182 v6_cork->opt = kzalloc(opt->tot_len, sk->sk_allocation);
1183 if (unlikely(!v6_cork->opt))
1186 v6_cork->opt->tot_len = opt->tot_len;
1187 v6_cork->opt->opt_flen = opt->opt_flen;
1188 v6_cork->opt->opt_nflen = opt->opt_nflen;
1190 v6_cork->opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1192 if (opt->dst0opt && !v6_cork->opt->dst0opt)
1195 v6_cork->opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1197 if (opt->dst1opt && !v6_cork->opt->dst1opt)
1200 v6_cork->opt->hopopt = ip6_opt_dup(opt->hopopt,
1202 if (opt->hopopt && !v6_cork->opt->hopopt)
1205 v6_cork->opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1207 if (opt->srcrt && !v6_cork->opt->srcrt)
1210 /* need source address above miyazawa*/
1213 cork->base.dst = &rt->dst;
1214 cork->fl.u.ip6 = *fl6;
1215 v6_cork->hop_limit = hlimit;
1216 v6_cork->tclass = tclass;
1217 if (rt->dst.flags & DST_XFRM_TUNNEL)
1218 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1219 rt->dst.dev->mtu : dst_mtu(&rt->dst);
1221 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1222 rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1223 if (np->frag_size < mtu) {
1225 mtu = np->frag_size;
1227 cork->base.fragsize = mtu;
1228 if (dst_allfrag(rt->dst.path))
1229 cork->base.flags |= IPCORK_ALLFRAG;
1230 cork->base.length = 0;
1235 static int __ip6_append_data(struct sock *sk,
1237 struct sk_buff_head *queue,
1238 struct inet_cork *cork,
1239 struct inet6_cork *v6_cork,
1240 struct page_frag *pfrag,
1241 int getfrag(void *from, char *to, int offset,
1242 int len, int odd, struct sk_buff *skb),
1243 void *from, int length, int transhdrlen,
1244 unsigned int flags, int dontfrag)
1246 struct sk_buff *skb, *skb_prev = NULL;
1247 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu;
1249 int dst_exthdrlen = 0;
1256 struct rt6_info *rt = (struct rt6_info *)cork->dst;
1257 struct ipv6_txoptions *opt = v6_cork->opt;
1258 int csummode = CHECKSUM_NONE;
1260 skb = skb_peek_tail(queue);
1262 exthdrlen = opt ? opt->opt_flen : 0;
1263 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1266 mtu = cork->fragsize;
1269 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1271 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1272 (opt ? opt->opt_nflen : 0);
1273 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1274 sizeof(struct frag_hdr);
1276 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1277 unsigned int maxnonfragsize, headersize;
1279 headersize = sizeof(struct ipv6hdr) +
1280 (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1281 (dst_allfrag(&rt->dst) ?
1282 sizeof(struct frag_hdr) : 0) +
1283 rt->rt6i_nfheader_len;
1285 if (ip6_sk_ignore_df(sk))
1286 maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1288 maxnonfragsize = mtu;
1290 /* dontfrag active */
1291 if ((cork->length + length > mtu - headersize) && dontfrag &&
1292 (sk->sk_protocol == IPPROTO_UDP ||
1293 sk->sk_protocol == IPPROTO_RAW)) {
1294 ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1295 sizeof(struct ipv6hdr));
1299 if (cork->length + length > maxnonfragsize - headersize) {
1301 ipv6_local_error(sk, EMSGSIZE, fl6,
1303 sizeof(struct ipv6hdr));
1308 if (sk->sk_type == SOCK_DGRAM || sk->sk_type == SOCK_RAW) {
1309 sock_tx_timestamp(sk, &tx_flags);
1310 if (tx_flags & SKBTX_ANY_SW_TSTAMP &&
1311 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1312 tskey = sk->sk_tskey++;
1315 /* If this is the first and only packet and device
1316 * supports checksum offloading, let's use it.
1317 * Use transhdrlen, same as IPv4, because partial
1318 * sums only work when transhdrlen is set.
1320 if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1321 length + fragheaderlen < mtu &&
1322 rt->dst.dev->features & NETIF_F_V6_CSUM &&
1324 csummode = CHECKSUM_PARTIAL;
1326 * Let's try using as much space as possible.
1327 * Use MTU if total length of the message fits into the MTU.
1328 * Otherwise, we need to reserve fragment header and
1329 * fragment alignment (= 8-15 octects, in total).
1331 * Note that we may need to "move" the data from the tail of
1332 * of the buffer to the new fragment when we split
1335 * FIXME: It may be fragmented into multiple chunks
1336 * at once if non-fragmentable extension headers
1341 cork->length += length;
1342 if (((length > mtu) ||
1343 (skb && skb_is_gso(skb))) &&
1344 (sk->sk_protocol == IPPROTO_UDP) &&
1345 (rt->dst.dev->features & NETIF_F_UFO) &&
1346 (sk->sk_type == SOCK_DGRAM)) {
1347 err = ip6_ufo_append_data(sk, queue, getfrag, from, length,
1348 hh_len, fragheaderlen,
1349 transhdrlen, mtu, flags, fl6);
1358 while (length > 0) {
1359 /* Check if the remaining data fits into current packet. */
1360 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1362 copy = maxfraglen - skb->len;
1366 unsigned int datalen;
1367 unsigned int fraglen;
1368 unsigned int fraggap;
1369 unsigned int alloclen;
1371 /* There's no room in the current skb */
1373 fraggap = skb->len - maxfraglen;
1376 /* update mtu and maxfraglen if necessary */
1377 if (!skb || !skb_prev)
1378 ip6_append_data_mtu(&mtu, &maxfraglen,
1379 fragheaderlen, skb, rt,
1385 * If remaining data exceeds the mtu,
1386 * we know we need more fragment(s).
1388 datalen = length + fraggap;
1390 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1391 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1392 if ((flags & MSG_MORE) &&
1393 !(rt->dst.dev->features&NETIF_F_SG))
1396 alloclen = datalen + fragheaderlen;
1398 alloclen += dst_exthdrlen;
1400 if (datalen != length + fraggap) {
1402 * this is not the last fragment, the trailer
1403 * space is regarded as data space.
1405 datalen += rt->dst.trailer_len;
1408 alloclen += rt->dst.trailer_len;
1409 fraglen = datalen + fragheaderlen;
1412 * We just reserve space for fragment header.
1413 * Note: this may be overallocation if the message
1414 * (without MSG_MORE) fits into the MTU.
1416 alloclen += sizeof(struct frag_hdr);
1419 skb = sock_alloc_send_skb(sk,
1421 (flags & MSG_DONTWAIT), &err);
1424 if (atomic_read(&sk->sk_wmem_alloc) <=
1426 skb = sock_wmalloc(sk,
1427 alloclen + hh_len, 1,
1435 * Fill in the control structures
1437 skb->protocol = htons(ETH_P_IPV6);
1438 skb->ip_summed = csummode;
1440 /* reserve for fragmentation and ipsec header */
1441 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1444 /* Only the initial fragment is time stamped */
1445 skb_shinfo(skb)->tx_flags = tx_flags;
1447 skb_shinfo(skb)->tskey = tskey;
1451 * Find where to start putting bytes
1453 data = skb_put(skb, fraglen);
1454 skb_set_network_header(skb, exthdrlen);
1455 data += fragheaderlen;
1456 skb->transport_header = (skb->network_header +
1459 skb->csum = skb_copy_and_csum_bits(
1460 skb_prev, maxfraglen,
1461 data + transhdrlen, fraggap, 0);
1462 skb_prev->csum = csum_sub(skb_prev->csum,
1465 pskb_trim_unique(skb_prev, maxfraglen);
1467 copy = datalen - transhdrlen - fraggap;
1473 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1480 length -= datalen - fraggap;
1486 * Put the packet on the pending queue
1488 __skb_queue_tail(queue, skb);
1495 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1499 if (getfrag(from, skb_put(skb, copy),
1500 offset, copy, off, skb) < 0) {
1501 __skb_trim(skb, off);
1506 int i = skb_shinfo(skb)->nr_frags;
1509 if (!sk_page_frag_refill(sk, pfrag))
1512 if (!skb_can_coalesce(skb, i, pfrag->page,
1515 if (i == MAX_SKB_FRAGS)
1518 __skb_fill_page_desc(skb, i, pfrag->page,
1520 skb_shinfo(skb)->nr_frags = ++i;
1521 get_page(pfrag->page);
1523 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1525 page_address(pfrag->page) + pfrag->offset,
1526 offset, copy, skb->len, skb) < 0)
1529 pfrag->offset += copy;
1530 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1532 skb->data_len += copy;
1533 skb->truesize += copy;
1534 atomic_add(copy, &sk->sk_wmem_alloc);
1545 cork->length -= length;
1546 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1550 int ip6_append_data(struct sock *sk,
1551 int getfrag(void *from, char *to, int offset, int len,
1552 int odd, struct sk_buff *skb),
1553 void *from, int length, int transhdrlen, int hlimit,
1554 int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1555 struct rt6_info *rt, unsigned int flags, int dontfrag)
1557 struct inet_sock *inet = inet_sk(sk);
1558 struct ipv6_pinfo *np = inet6_sk(sk);
1562 if (flags&MSG_PROBE)
1564 if (skb_queue_empty(&sk->sk_write_queue)) {
1568 err = ip6_setup_cork(sk, &inet->cork, &np->cork, hlimit,
1569 tclass, opt, rt, fl6);
1573 exthdrlen = (opt ? opt->opt_flen : 0);
1574 length += exthdrlen;
1575 transhdrlen += exthdrlen;
1577 fl6 = &inet->cork.fl.u.ip6;
1581 return __ip6_append_data(sk, fl6, &sk->sk_write_queue, &inet->cork.base,
1582 &np->cork, sk_page_frag(sk), getfrag,
1583 from, length, transhdrlen, flags, dontfrag);
1585 EXPORT_SYMBOL_GPL(ip6_append_data);
1587 static void ip6_cork_release(struct inet_cork_full *cork,
1588 struct inet6_cork *v6_cork)
1591 kfree(v6_cork->opt->dst0opt);
1592 kfree(v6_cork->opt->dst1opt);
1593 kfree(v6_cork->opt->hopopt);
1594 kfree(v6_cork->opt->srcrt);
1595 kfree(v6_cork->opt);
1596 v6_cork->opt = NULL;
1599 if (cork->base.dst) {
1600 dst_release(cork->base.dst);
1601 cork->base.dst = NULL;
1602 cork->base.flags &= ~IPCORK_ALLFRAG;
1604 memset(&cork->fl, 0, sizeof(cork->fl));
1607 struct sk_buff *__ip6_make_skb(struct sock *sk,
1608 struct sk_buff_head *queue,
1609 struct inet_cork_full *cork,
1610 struct inet6_cork *v6_cork)
1612 struct sk_buff *skb, *tmp_skb;
1613 struct sk_buff **tail_skb;
1614 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1615 struct ipv6_pinfo *np = inet6_sk(sk);
1616 struct net *net = sock_net(sk);
1617 struct ipv6hdr *hdr;
1618 struct ipv6_txoptions *opt = v6_cork->opt;
1619 struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1620 struct flowi6 *fl6 = &cork->fl.u.ip6;
1621 unsigned char proto = fl6->flowi6_proto;
1623 skb = __skb_dequeue(queue);
1626 tail_skb = &(skb_shinfo(skb)->frag_list);
1628 /* move skb->data to ip header from ext header */
1629 if (skb->data < skb_network_header(skb))
1630 __skb_pull(skb, skb_network_offset(skb));
1631 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1632 __skb_pull(tmp_skb, skb_network_header_len(skb));
1633 *tail_skb = tmp_skb;
1634 tail_skb = &(tmp_skb->next);
1635 skb->len += tmp_skb->len;
1636 skb->data_len += tmp_skb->len;
1637 skb->truesize += tmp_skb->truesize;
1638 tmp_skb->destructor = NULL;
1642 /* Allow local fragmentation. */
1643 skb->ignore_df = ip6_sk_ignore_df(sk);
1645 *final_dst = fl6->daddr;
1646 __skb_pull(skb, skb_network_header_len(skb));
1647 if (opt && opt->opt_flen)
1648 ipv6_push_frag_opts(skb, opt, &proto);
1649 if (opt && opt->opt_nflen)
1650 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1652 skb_push(skb, sizeof(struct ipv6hdr));
1653 skb_reset_network_header(skb);
1654 hdr = ipv6_hdr(skb);
1656 ip6_flow_hdr(hdr, v6_cork->tclass,
1657 ip6_make_flowlabel(net, skb, fl6->flowlabel,
1658 np->autoflowlabel, fl6));
1659 hdr->hop_limit = v6_cork->hop_limit;
1660 hdr->nexthdr = proto;
1661 hdr->saddr = fl6->saddr;
1662 hdr->daddr = *final_dst;
1664 skb->priority = sk->sk_priority;
1665 skb->mark = sk->sk_mark;
1667 skb_dst_set(skb, dst_clone(&rt->dst));
1668 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1669 if (proto == IPPROTO_ICMPV6) {
1670 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1672 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1673 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1676 ip6_cork_release(cork, v6_cork);
1681 int ip6_send_skb(struct sk_buff *skb)
1683 struct net *net = sock_net(skb->sk);
1684 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1687 err = ip6_local_out(skb);
1690 err = net_xmit_errno(err);
1692 IP6_INC_STATS(net, rt->rt6i_idev,
1693 IPSTATS_MIB_OUTDISCARDS);
1699 int ip6_push_pending_frames(struct sock *sk)
1701 struct sk_buff *skb;
1703 skb = ip6_finish_skb(sk);
1707 return ip6_send_skb(skb);
1709 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1711 static void __ip6_flush_pending_frames(struct sock *sk,
1712 struct sk_buff_head *queue,
1713 struct inet_cork_full *cork,
1714 struct inet6_cork *v6_cork)
1716 struct sk_buff *skb;
1718 while ((skb = __skb_dequeue_tail(queue)) != NULL) {
1720 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1721 IPSTATS_MIB_OUTDISCARDS);
1725 ip6_cork_release(cork, v6_cork);
1728 void ip6_flush_pending_frames(struct sock *sk)
1730 __ip6_flush_pending_frames(sk, &sk->sk_write_queue,
1731 &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
1733 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1735 struct sk_buff *ip6_make_skb(struct sock *sk,
1736 int getfrag(void *from, char *to, int offset,
1737 int len, int odd, struct sk_buff *skb),
1738 void *from, int length, int transhdrlen,
1739 int hlimit, int tclass,
1740 struct ipv6_txoptions *opt, struct flowi6 *fl6,
1741 struct rt6_info *rt, unsigned int flags,
1744 struct inet_cork_full cork;
1745 struct inet6_cork v6_cork;
1746 struct sk_buff_head queue;
1747 int exthdrlen = (opt ? opt->opt_flen : 0);
1750 if (flags & MSG_PROBE)
1753 __skb_queue_head_init(&queue);
1755 cork.base.flags = 0;
1757 cork.base.opt = NULL;
1759 err = ip6_setup_cork(sk, &cork, &v6_cork, hlimit, tclass, opt, rt, fl6);
1761 return ERR_PTR(err);
1764 dontfrag = inet6_sk(sk)->dontfrag;
1766 err = __ip6_append_data(sk, fl6, &queue, &cork.base, &v6_cork,
1767 ¤t->task_frag, getfrag, from,
1768 length + exthdrlen, transhdrlen + exthdrlen,
1771 __ip6_flush_pending_frames(sk, &queue, &cork, &v6_cork);
1772 return ERR_PTR(err);
1775 return __ip6_make_skb(sk, &queue, &cork, &v6_cork);