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>
58 #include <net/l3mdev.h>
60 static int ip6_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
62 struct dst_entry *dst = skb_dst(skb);
63 struct net_device *dev = dst->dev;
64 struct neighbour *neigh;
65 struct in6_addr *nexthop;
68 skb->protocol = htons(ETH_P_IPV6);
71 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
72 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
74 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(sk) &&
75 ((mroute6_socket(net, skb) &&
76 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
77 ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
78 &ipv6_hdr(skb)->saddr))) {
79 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
81 /* Do not check for IFF_ALLMULTI; multicast routing
82 is not supported in any case.
85 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
86 net, sk, newskb, NULL, newskb->dev,
89 if (ipv6_hdr(skb)->hop_limit == 0) {
90 IP6_INC_STATS(net, idev,
91 IPSTATS_MIB_OUTDISCARDS);
97 IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUTMCAST, skb->len);
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(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
124 static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
126 if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
127 dst_allfrag(skb_dst(skb)) ||
128 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
129 return ip6_fragment(net, sk, skb, ip6_finish_output2);
131 return ip6_finish_output2(net, sk, skb);
134 int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
136 struct net_device *dev = skb_dst(skb)->dev;
137 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
139 if (unlikely(idev->cnf.disable_ipv6)) {
140 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
145 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
146 net, sk, skb, NULL, dev,
148 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
152 * xmit an sk_buff (used by TCP, SCTP and DCCP)
153 * Note : socket lock is not held for SYNACK packets, but might be modified
154 * by calls to skb_set_owner_w() and ipv6_local_error(),
155 * which are using proper atomic operations or spinlocks.
157 int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
158 struct ipv6_txoptions *opt, int tclass)
160 struct net *net = sock_net(sk);
161 const struct ipv6_pinfo *np = inet6_sk(sk);
162 struct in6_addr *first_hop = &fl6->daddr;
163 struct dst_entry *dst = skb_dst(skb);
165 u8 proto = fl6->flowi6_proto;
166 int seg_len = skb->len;
171 unsigned int head_room;
173 /* First: exthdrs may take lots of space (~8K for now)
174 MAX_HEADER is not enough.
176 head_room = opt->opt_nflen + opt->opt_flen;
177 seg_len += head_room;
178 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
180 if (skb_headroom(skb) < head_room) {
181 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
183 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
184 IPSTATS_MIB_OUTDISCARDS);
190 /* skb_set_owner_w() changes sk->sk_wmem_alloc atomically,
191 * it is safe to call in our context (socket lock not held)
193 skb_set_owner_w(skb, (struct sock *)sk);
196 ipv6_push_frag_opts(skb, opt, &proto);
198 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
201 skb_push(skb, sizeof(struct ipv6hdr));
202 skb_reset_network_header(skb);
206 * Fill in the IPv6 header
209 hlimit = np->hop_limit;
211 hlimit = ip6_dst_hoplimit(dst);
213 ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
214 np->autoflowlabel, fl6));
216 hdr->payload_len = htons(seg_len);
217 hdr->nexthdr = proto;
218 hdr->hop_limit = hlimit;
220 hdr->saddr = fl6->saddr;
221 hdr->daddr = *first_hop;
223 skb->protocol = htons(ETH_P_IPV6);
224 skb->priority = sk->sk_priority;
225 skb->mark = sk->sk_mark;
228 if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
229 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
230 IPSTATS_MIB_OUT, skb->len);
231 /* hooks should never assume socket lock is held.
232 * we promote our socket to non const
234 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
235 net, (struct sock *)sk, skb, NULL, dst->dev,
240 /* ipv6_local_error() does not require socket lock,
241 * we promote our socket to non const
243 ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
245 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
249 EXPORT_SYMBOL(ip6_xmit);
251 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
253 struct ip6_ra_chain *ra;
254 struct sock *last = NULL;
256 read_lock(&ip6_ra_lock);
257 for (ra = ip6_ra_chain; ra; ra = ra->next) {
258 struct sock *sk = ra->sk;
259 if (sk && ra->sel == sel &&
260 (!sk->sk_bound_dev_if ||
261 sk->sk_bound_dev_if == skb->dev->ifindex)) {
263 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
265 rawv6_rcv(last, skb2);
272 rawv6_rcv(last, skb);
273 read_unlock(&ip6_ra_lock);
276 read_unlock(&ip6_ra_lock);
280 static int ip6_forward_proxy_check(struct sk_buff *skb)
282 struct ipv6hdr *hdr = ipv6_hdr(skb);
283 u8 nexthdr = hdr->nexthdr;
287 if (ipv6_ext_hdr(nexthdr)) {
288 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
292 offset = sizeof(struct ipv6hdr);
294 if (nexthdr == IPPROTO_ICMPV6) {
295 struct icmp6hdr *icmp6;
297 if (!pskb_may_pull(skb, (skb_network_header(skb) +
298 offset + 1 - skb->data)))
301 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
303 switch (icmp6->icmp6_type) {
304 case NDISC_ROUTER_SOLICITATION:
305 case NDISC_ROUTER_ADVERTISEMENT:
306 case NDISC_NEIGHBOUR_SOLICITATION:
307 case NDISC_NEIGHBOUR_ADVERTISEMENT:
309 /* For reaction involving unicast neighbor discovery
310 * message destined to the proxied address, pass it to
320 * The proxying router can't forward traffic sent to a link-local
321 * address, so signal the sender and discard the packet. This
322 * behavior is clarified by the MIPv6 specification.
324 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
325 dst_link_failure(skb);
332 static inline int ip6_forward_finish(struct net *net, struct sock *sk,
335 skb_sender_cpu_clear(skb);
336 return dst_output(net, sk, skb);
339 static unsigned int ip6_dst_mtu_forward(const struct dst_entry *dst)
342 struct inet6_dev *idev;
344 if (dst_metric_locked(dst, RTAX_MTU)) {
345 mtu = dst_metric_raw(dst, RTAX_MTU);
352 idev = __in6_dev_get(dst->dev);
354 mtu = idev->cnf.mtu6;
360 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
365 /* ipv6 conntrack defrag sets max_frag_size + ignore_df */
366 if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
372 if (skb_is_gso(skb) && skb_gso_network_seglen(skb) <= mtu)
378 int ip6_forward(struct sk_buff *skb)
380 struct dst_entry *dst = skb_dst(skb);
381 struct ipv6hdr *hdr = ipv6_hdr(skb);
382 struct inet6_skb_parm *opt = IP6CB(skb);
383 struct net *net = dev_net(dst->dev);
386 if (net->ipv6.devconf_all->forwarding == 0)
389 if (skb->pkt_type != PACKET_HOST)
392 if (unlikely(skb->sk))
395 if (skb_warn_if_lro(skb))
398 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
399 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
400 IPSTATS_MIB_INDISCARDS);
404 skb_forward_csum(skb);
407 * We DO NOT make any processing on
408 * RA packets, pushing them to user level AS IS
409 * without ane WARRANTY that application will be able
410 * to interpret them. The reason is that we
411 * cannot make anything clever here.
413 * We are not end-node, so that if packet contains
414 * AH/ESP, we cannot make anything.
415 * Defragmentation also would be mistake, RA packets
416 * cannot be fragmented, because there is no warranty
417 * that different fragments will go along one path. --ANK
419 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
420 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
425 * check and decrement ttl
427 if (hdr->hop_limit <= 1) {
428 /* Force OUTPUT device used as source address */
430 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
431 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
432 IPSTATS_MIB_INHDRERRORS);
438 /* XXX: idev->cnf.proxy_ndp? */
439 if (net->ipv6.devconf_all->proxy_ndp &&
440 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
441 int proxied = ip6_forward_proxy_check(skb);
443 return ip6_input(skb);
444 else if (proxied < 0) {
445 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
446 IPSTATS_MIB_INDISCARDS);
451 if (!xfrm6_route_forward(skb)) {
452 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
453 IPSTATS_MIB_INDISCARDS);
458 /* IPv6 specs say nothing about it, but it is clear that we cannot
459 send redirects to source routed frames.
460 We don't send redirects to frames decapsulated from IPsec.
462 if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
463 struct in6_addr *target = NULL;
464 struct inet_peer *peer;
468 * incoming and outgoing devices are the same
472 rt = (struct rt6_info *) dst;
473 if (rt->rt6i_flags & RTF_GATEWAY)
474 target = &rt->rt6i_gateway;
476 target = &hdr->daddr;
478 peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
480 /* Limit redirects both by destination (here)
481 and by source (inside ndisc_send_redirect)
483 if (inet_peer_xrlim_allow(peer, 1*HZ))
484 ndisc_send_redirect(skb, target);
488 int addrtype = ipv6_addr_type(&hdr->saddr);
490 /* This check is security critical. */
491 if (addrtype == IPV6_ADDR_ANY ||
492 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
494 if (addrtype & IPV6_ADDR_LINKLOCAL) {
495 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
496 ICMPV6_NOT_NEIGHBOUR, 0);
501 mtu = ip6_dst_mtu_forward(dst);
502 if (mtu < IPV6_MIN_MTU)
505 if (ip6_pkt_too_big(skb, mtu)) {
506 /* Again, force OUTPUT device used as source address */
508 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
509 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
510 IPSTATS_MIB_INTOOBIGERRORS);
511 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
512 IPSTATS_MIB_FRAGFAILS);
517 if (skb_cow(skb, dst->dev->hard_header_len)) {
518 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
519 IPSTATS_MIB_OUTDISCARDS);
525 /* Mangling hops number delayed to point after skb COW */
529 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
530 IP6_ADD_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
531 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
532 net, NULL, skb, skb->dev, dst->dev,
536 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
542 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
544 to->pkt_type = from->pkt_type;
545 to->priority = from->priority;
546 to->protocol = from->protocol;
548 skb_dst_set(to, dst_clone(skb_dst(from)));
550 to->mark = from->mark;
552 #ifdef CONFIG_NET_SCHED
553 to->tc_index = from->tc_index;
556 skb_copy_secmark(to, from);
559 int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
560 int (*output)(struct net *, struct sock *, struct sk_buff *))
562 struct sk_buff *frag;
563 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
564 struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
565 inet6_sk(skb->sk) : NULL;
566 struct ipv6hdr *tmp_hdr;
568 unsigned int mtu, hlen, left, len;
571 int ptr, offset = 0, err = 0;
572 u8 *prevhdr, nexthdr = 0;
574 hlen = ip6_find_1stfragopt(skb, &prevhdr);
577 mtu = ip6_skb_dst_mtu(skb);
579 /* We must not fragment if the socket is set to force MTU discovery
580 * or if the skb it not generated by a local socket.
582 if (unlikely(!skb->ignore_df && skb->len > mtu))
585 if (IP6CB(skb)->frag_max_size) {
586 if (IP6CB(skb)->frag_max_size > mtu)
589 /* don't send fragments larger than what we received */
590 mtu = IP6CB(skb)->frag_max_size;
591 if (mtu < IPV6_MIN_MTU)
595 if (np && np->frag_size < mtu) {
599 if (mtu < hlen + sizeof(struct frag_hdr) + 8)
601 mtu -= hlen + sizeof(struct frag_hdr);
603 frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
604 &ipv6_hdr(skb)->saddr);
606 hroom = LL_RESERVED_SPACE(rt->dst.dev);
607 if (skb_has_frag_list(skb)) {
608 int first_len = skb_pagelen(skb);
609 struct sk_buff *frag2;
611 if (first_len - hlen > mtu ||
612 ((first_len - hlen) & 7) ||
614 skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
617 skb_walk_frags(skb, frag) {
618 /* Correct geometry. */
619 if (frag->len > mtu ||
620 ((frag->len & 7) && frag->next) ||
621 skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
622 goto slow_path_clean;
624 /* Partially cloned skb? */
625 if (skb_shared(frag))
626 goto slow_path_clean;
631 frag->destructor = sock_wfree;
633 skb->truesize -= frag->truesize;
640 *prevhdr = NEXTHDR_FRAGMENT;
641 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
643 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
644 IPSTATS_MIB_FRAGFAILS);
648 frag = skb_shinfo(skb)->frag_list;
649 skb_frag_list_init(skb);
651 __skb_pull(skb, hlen);
652 fh = (struct frag_hdr *)__skb_push(skb, sizeof(struct frag_hdr));
653 __skb_push(skb, hlen);
654 skb_reset_network_header(skb);
655 memcpy(skb_network_header(skb), tmp_hdr, hlen);
657 fh->nexthdr = nexthdr;
659 fh->frag_off = htons(IP6_MF);
660 fh->identification = frag_id;
662 first_len = skb_pagelen(skb);
663 skb->data_len = first_len - skb_headlen(skb);
664 skb->len = first_len;
665 ipv6_hdr(skb)->payload_len = htons(first_len -
666 sizeof(struct ipv6hdr));
671 /* Prepare header of the next frame,
672 * before previous one went down. */
674 frag->ip_summed = CHECKSUM_NONE;
675 skb_reset_transport_header(frag);
676 fh = (struct frag_hdr *)__skb_push(frag, sizeof(struct frag_hdr));
677 __skb_push(frag, hlen);
678 skb_reset_network_header(frag);
679 memcpy(skb_network_header(frag), tmp_hdr,
681 offset += skb->len - hlen - sizeof(struct frag_hdr);
682 fh->nexthdr = nexthdr;
684 fh->frag_off = htons(offset);
686 fh->frag_off |= htons(IP6_MF);
687 fh->identification = frag_id;
688 ipv6_hdr(frag)->payload_len =
690 sizeof(struct ipv6hdr));
691 ip6_copy_metadata(frag, skb);
694 err = output(net, sk, skb);
696 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
697 IPSTATS_MIB_FRAGCREATES);
710 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
711 IPSTATS_MIB_FRAGOKS);
716 kfree_skb_list(frag);
718 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
719 IPSTATS_MIB_FRAGFAILS);
724 skb_walk_frags(skb, frag2) {
728 frag2->destructor = NULL;
729 skb->truesize += frag2->truesize;
734 if ((skb->ip_summed == CHECKSUM_PARTIAL) &&
735 skb_checksum_help(skb))
738 left = skb->len - hlen; /* Space per frame */
739 ptr = hlen; /* Where to start from */
742 * Fragment the datagram.
745 *prevhdr = NEXTHDR_FRAGMENT;
746 troom = rt->dst.dev->needed_tailroom;
749 * Keep copying data until we run out.
753 /* IF: it doesn't fit, use 'mtu' - the data space left */
756 /* IF: we are not sending up to and including the packet end
757 then align the next start on an eight byte boundary */
762 /* Allocate buffer */
763 frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
764 hroom + troom, GFP_ATOMIC);
766 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
767 IPSTATS_MIB_FRAGFAILS);
773 * Set up data on packet
776 ip6_copy_metadata(frag, skb);
777 skb_reserve(frag, hroom);
778 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
779 skb_reset_network_header(frag);
780 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
781 frag->transport_header = (frag->network_header + hlen +
782 sizeof(struct frag_hdr));
785 * Charge the memory for the fragment to any owner
789 skb_set_owner_w(frag, skb->sk);
792 * Copy the packet header into the new buffer.
794 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
797 * Build fragment header.
799 fh->nexthdr = nexthdr;
801 fh->identification = frag_id;
804 * Copy a block of the IP datagram.
806 BUG_ON(skb_copy_bits(skb, ptr, skb_transport_header(frag),
810 fh->frag_off = htons(offset);
812 fh->frag_off |= htons(IP6_MF);
813 ipv6_hdr(frag)->payload_len = htons(frag->len -
814 sizeof(struct ipv6hdr));
820 * Put this fragment into the sending queue.
822 err = output(net, sk, frag);
826 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
827 IPSTATS_MIB_FRAGCREATES);
829 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
830 IPSTATS_MIB_FRAGOKS);
835 if (skb->sk && dst_allfrag(skb_dst(skb)))
836 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
838 skb->dev = skb_dst(skb)->dev;
839 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
843 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
844 IPSTATS_MIB_FRAGFAILS);
849 static inline int ip6_rt_check(const struct rt6key *rt_key,
850 const struct in6_addr *fl_addr,
851 const struct in6_addr *addr_cache)
853 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
854 (!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
857 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
858 struct dst_entry *dst,
859 const struct flowi6 *fl6)
861 struct ipv6_pinfo *np = inet6_sk(sk);
867 if (dst->ops->family != AF_INET6) {
872 rt = (struct rt6_info *)dst;
873 /* Yes, checking route validity in not connected
874 * case is not very simple. Take into account,
875 * that we do not support routing by source, TOS,
876 * and MSG_DONTROUTE --ANK (980726)
878 * 1. ip6_rt_check(): If route was host route,
879 * check that cached destination is current.
880 * If it is network route, we still may
881 * check its validity using saved pointer
882 * to the last used address: daddr_cache.
883 * We do not want to save whole address now,
884 * (because main consumer of this service
885 * is tcp, which has not this problem),
886 * so that the last trick works only on connected
888 * 2. oif also should be the same.
890 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
891 #ifdef CONFIG_IPV6_SUBTREES
892 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
894 (!(fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF) &&
895 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex))) {
904 static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
905 struct dst_entry **dst, struct flowi6 *fl6)
907 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
913 /* The correct way to handle this would be to do
914 * ip6_route_get_saddr, and then ip6_route_output; however,
915 * the route-specific preferred source forces the
916 * ip6_route_output call _before_ ip6_route_get_saddr.
918 * In source specific routing (no src=any default route),
919 * ip6_route_output will fail given src=any saddr, though, so
920 * that's why we try it again later.
922 if (ipv6_addr_any(&fl6->saddr) && (!*dst || !(*dst)->error)) {
924 bool had_dst = *dst != NULL;
927 *dst = ip6_route_output(net, sk, fl6);
928 rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
929 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
930 sk ? inet6_sk(sk)->srcprefs : 0,
933 goto out_err_release;
935 /* If we had an erroneous initial result, pretend it
936 * never existed and let the SA-enabled version take
939 if (!had_dst && (*dst)->error) {
946 *dst = ip6_route_output(net, sk, fl6);
950 goto out_err_release;
952 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
954 * Here if the dst entry we've looked up
955 * has a neighbour entry that is in the INCOMPLETE
956 * state and the src address from the flow is
957 * marked as OPTIMISTIC, we release the found
958 * dst entry and replace it instead with the
959 * dst entry of the nexthop router
961 rt = (struct rt6_info *) *dst;
963 n = __ipv6_neigh_lookup_noref(rt->dst.dev,
964 rt6_nexthop(rt, &fl6->daddr));
965 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
966 rcu_read_unlock_bh();
969 struct inet6_ifaddr *ifp;
970 struct flowi6 fl_gw6;
973 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
976 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
982 * We need to get the dst entry for the
983 * default router instead
986 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
987 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
988 *dst = ip6_route_output(net, sk, &fl_gw6);
991 goto out_err_release;
999 if (err == -ENETUNREACH)
1000 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1007 * ip6_dst_lookup - perform route lookup on flow
1008 * @sk: socket which provides route info
1009 * @dst: pointer to dst_entry * for result
1010 * @fl6: flow to lookup
1012 * This function performs a route lookup on the given flow.
1014 * It returns zero on success, or a standard errno code on error.
1016 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1020 return ip6_dst_lookup_tail(net, sk, dst, fl6);
1022 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1025 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1026 * @sk: socket which provides route info
1027 * @fl6: flow to lookup
1028 * @final_dst: final destination address for ipsec lookup
1030 * This function performs a route lookup on the given flow.
1032 * It returns a valid dst pointer on success, or a pointer encoded
1035 struct dst_entry *ip6_dst_lookup_flow(const struct sock *sk, struct flowi6 *fl6,
1036 const struct in6_addr *final_dst)
1038 struct dst_entry *dst = NULL;
1041 err = ip6_dst_lookup_tail(sock_net(sk), sk, &dst, fl6);
1043 return ERR_PTR(err);
1045 fl6->daddr = *final_dst;
1046 if (!fl6->flowi6_oif)
1047 fl6->flowi6_oif = l3mdev_fib_oif(dst->dev);
1049 return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1051 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1054 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1055 * @sk: socket which provides the dst cache and route info
1056 * @fl6: flow to lookup
1057 * @final_dst: final destination address for ipsec lookup
1059 * This function performs a route lookup on the given flow with the
1060 * possibility of using the cached route in the socket if it is valid.
1061 * It will take the socket dst lock when operating on the dst cache.
1062 * As a result, this function can only be used in process context.
1064 * It returns a valid dst pointer on success, or a pointer encoded
1067 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1068 const struct in6_addr *final_dst)
1070 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1073 dst = ip6_sk_dst_check(sk, dst, fl6);
1075 err = ip6_dst_lookup_tail(sock_net(sk), sk, &dst, fl6);
1077 return ERR_PTR(err);
1079 fl6->daddr = *final_dst;
1081 return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1083 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1085 static inline int ip6_ufo_append_data(struct sock *sk,
1086 struct sk_buff_head *queue,
1087 int getfrag(void *from, char *to, int offset, int len,
1088 int odd, struct sk_buff *skb),
1089 void *from, int length, int hh_len, int fragheaderlen,
1090 int transhdrlen, int mtu, unsigned int flags,
1091 const struct flowi6 *fl6)
1094 struct sk_buff *skb;
1097 /* There is support for UDP large send offload by network
1098 * device, so create one single skb packet containing complete
1101 skb = skb_peek_tail(queue);
1103 skb = sock_alloc_send_skb(sk,
1104 hh_len + fragheaderlen + transhdrlen + 20,
1105 (flags & MSG_DONTWAIT), &err);
1109 /* reserve space for Hardware header */
1110 skb_reserve(skb, hh_len);
1112 /* create space for UDP/IP header */
1113 skb_put(skb, fragheaderlen + transhdrlen);
1115 /* initialize network header pointer */
1116 skb_reset_network_header(skb);
1118 /* initialize protocol header pointer */
1119 skb->transport_header = skb->network_header + fragheaderlen;
1121 skb->protocol = htons(ETH_P_IPV6);
1124 __skb_queue_tail(queue, skb);
1125 } else if (skb_is_gso(skb)) {
1129 skb->ip_summed = CHECKSUM_PARTIAL;
1130 /* Specify the length of each IPv6 datagram fragment.
1131 * It has to be a multiple of 8.
1133 skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1134 sizeof(struct frag_hdr)) & ~7;
1135 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1136 skb_shinfo(skb)->ip6_frag_id = ipv6_select_ident(sock_net(sk),
1141 return skb_append_datato_frags(sk, skb, getfrag, from,
1142 (length - transhdrlen));
1145 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1148 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1151 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1154 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1157 static void ip6_append_data_mtu(unsigned int *mtu,
1159 unsigned int fragheaderlen,
1160 struct sk_buff *skb,
1161 struct rt6_info *rt,
1162 unsigned int orig_mtu)
1164 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1166 /* first fragment, reserve header_len */
1167 *mtu = orig_mtu - rt->dst.header_len;
1171 * this fragment is not first, the headers
1172 * space is regarded as data space.
1176 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1177 + fragheaderlen - sizeof(struct frag_hdr);
1181 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1182 struct inet6_cork *v6_cork,
1183 int hlimit, int tclass, struct ipv6_txoptions *opt,
1184 struct rt6_info *rt, struct flowi6 *fl6)
1186 struct ipv6_pinfo *np = inet6_sk(sk);
1193 if (WARN_ON(v6_cork->opt))
1196 v6_cork->opt = kzalloc(opt->tot_len, sk->sk_allocation);
1197 if (unlikely(!v6_cork->opt))
1200 v6_cork->opt->tot_len = opt->tot_len;
1201 v6_cork->opt->opt_flen = opt->opt_flen;
1202 v6_cork->opt->opt_nflen = opt->opt_nflen;
1204 v6_cork->opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1206 if (opt->dst0opt && !v6_cork->opt->dst0opt)
1209 v6_cork->opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1211 if (opt->dst1opt && !v6_cork->opt->dst1opt)
1214 v6_cork->opt->hopopt = ip6_opt_dup(opt->hopopt,
1216 if (opt->hopopt && !v6_cork->opt->hopopt)
1219 v6_cork->opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1221 if (opt->srcrt && !v6_cork->opt->srcrt)
1224 /* need source address above miyazawa*/
1227 cork->base.dst = &rt->dst;
1228 cork->fl.u.ip6 = *fl6;
1229 v6_cork->hop_limit = hlimit;
1230 v6_cork->tclass = tclass;
1231 if (rt->dst.flags & DST_XFRM_TUNNEL)
1232 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1233 rt->dst.dev->mtu : dst_mtu(&rt->dst);
1235 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1236 rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1237 if (np->frag_size < mtu) {
1239 mtu = np->frag_size;
1241 cork->base.fragsize = mtu;
1242 if (dst_allfrag(rt->dst.path))
1243 cork->base.flags |= IPCORK_ALLFRAG;
1244 cork->base.length = 0;
1249 static int __ip6_append_data(struct sock *sk,
1251 struct sk_buff_head *queue,
1252 struct inet_cork *cork,
1253 struct inet6_cork *v6_cork,
1254 struct page_frag *pfrag,
1255 int getfrag(void *from, char *to, int offset,
1256 int len, int odd, struct sk_buff *skb),
1257 void *from, int length, int transhdrlen,
1258 unsigned int flags, int dontfrag)
1260 struct sk_buff *skb, *skb_prev = NULL;
1261 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu;
1263 int dst_exthdrlen = 0;
1270 struct rt6_info *rt = (struct rt6_info *)cork->dst;
1271 struct ipv6_txoptions *opt = v6_cork->opt;
1272 int csummode = CHECKSUM_NONE;
1274 skb = skb_peek_tail(queue);
1276 exthdrlen = opt ? opt->opt_flen : 0;
1277 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1280 mtu = cork->fragsize;
1283 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1285 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1286 (opt ? opt->opt_nflen : 0);
1287 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1288 sizeof(struct frag_hdr);
1290 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1291 unsigned int maxnonfragsize, headersize;
1293 headersize = sizeof(struct ipv6hdr) +
1294 (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1295 (dst_allfrag(&rt->dst) ?
1296 sizeof(struct frag_hdr) : 0) +
1297 rt->rt6i_nfheader_len;
1299 if (ip6_sk_ignore_df(sk))
1300 maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1302 maxnonfragsize = mtu;
1304 /* dontfrag active */
1305 if ((cork->length + length > mtu - headersize) && dontfrag &&
1306 (sk->sk_protocol == IPPROTO_UDP ||
1307 sk->sk_protocol == IPPROTO_RAW)) {
1308 ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1309 sizeof(struct ipv6hdr));
1313 if (cork->length + length > maxnonfragsize - headersize) {
1315 ipv6_local_error(sk, EMSGSIZE, fl6,
1317 sizeof(struct ipv6hdr));
1322 if (sk->sk_type == SOCK_DGRAM || sk->sk_type == SOCK_RAW) {
1323 sock_tx_timestamp(sk, &tx_flags);
1324 if (tx_flags & SKBTX_ANY_SW_TSTAMP &&
1325 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1326 tskey = sk->sk_tskey++;
1329 /* If this is the first and only packet and device
1330 * supports checksum offloading, let's use it.
1331 * Use transhdrlen, same as IPv4, because partial
1332 * sums only work when transhdrlen is set.
1334 if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1335 length + fragheaderlen < mtu &&
1336 rt->dst.dev->features & NETIF_F_V6_CSUM &&
1338 csummode = CHECKSUM_PARTIAL;
1340 * Let's try using as much space as possible.
1341 * Use MTU if total length of the message fits into the MTU.
1342 * Otherwise, we need to reserve fragment header and
1343 * fragment alignment (= 8-15 octects, in total).
1345 * Note that we may need to "move" the data from the tail of
1346 * of the buffer to the new fragment when we split
1349 * FIXME: It may be fragmented into multiple chunks
1350 * at once if non-fragmentable extension headers
1355 cork->length += length;
1356 if (((length > mtu) ||
1357 (skb && skb_is_gso(skb))) &&
1358 (sk->sk_protocol == IPPROTO_UDP) &&
1359 (rt->dst.dev->features & NETIF_F_UFO) &&
1360 (sk->sk_type == SOCK_DGRAM)) {
1361 err = ip6_ufo_append_data(sk, queue, getfrag, from, length,
1362 hh_len, fragheaderlen,
1363 transhdrlen, mtu, flags, fl6);
1372 while (length > 0) {
1373 /* Check if the remaining data fits into current packet. */
1374 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1376 copy = maxfraglen - skb->len;
1380 unsigned int datalen;
1381 unsigned int fraglen;
1382 unsigned int fraggap;
1383 unsigned int alloclen;
1385 /* There's no room in the current skb */
1387 fraggap = skb->len - maxfraglen;
1390 /* update mtu and maxfraglen if necessary */
1391 if (!skb || !skb_prev)
1392 ip6_append_data_mtu(&mtu, &maxfraglen,
1393 fragheaderlen, skb, rt,
1399 * If remaining data exceeds the mtu,
1400 * we know we need more fragment(s).
1402 datalen = length + fraggap;
1404 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1405 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1406 if ((flags & MSG_MORE) &&
1407 !(rt->dst.dev->features&NETIF_F_SG))
1410 alloclen = datalen + fragheaderlen;
1412 alloclen += dst_exthdrlen;
1414 if (datalen != length + fraggap) {
1416 * this is not the last fragment, the trailer
1417 * space is regarded as data space.
1419 datalen += rt->dst.trailer_len;
1422 alloclen += rt->dst.trailer_len;
1423 fraglen = datalen + fragheaderlen;
1426 * We just reserve space for fragment header.
1427 * Note: this may be overallocation if the message
1428 * (without MSG_MORE) fits into the MTU.
1430 alloclen += sizeof(struct frag_hdr);
1433 skb = sock_alloc_send_skb(sk,
1435 (flags & MSG_DONTWAIT), &err);
1438 if (atomic_read(&sk->sk_wmem_alloc) <=
1440 skb = sock_wmalloc(sk,
1441 alloclen + hh_len, 1,
1449 * Fill in the control structures
1451 skb->protocol = htons(ETH_P_IPV6);
1452 skb->ip_summed = csummode;
1454 /* reserve for fragmentation and ipsec header */
1455 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1458 /* Only the initial fragment is time stamped */
1459 skb_shinfo(skb)->tx_flags = tx_flags;
1461 skb_shinfo(skb)->tskey = tskey;
1465 * Find where to start putting bytes
1467 data = skb_put(skb, fraglen);
1468 skb_set_network_header(skb, exthdrlen);
1469 data += fragheaderlen;
1470 skb->transport_header = (skb->network_header +
1473 skb->csum = skb_copy_and_csum_bits(
1474 skb_prev, maxfraglen,
1475 data + transhdrlen, fraggap, 0);
1476 skb_prev->csum = csum_sub(skb_prev->csum,
1479 pskb_trim_unique(skb_prev, maxfraglen);
1481 copy = datalen - transhdrlen - fraggap;
1487 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1494 length -= datalen - fraggap;
1500 * Put the packet on the pending queue
1502 __skb_queue_tail(queue, skb);
1509 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1513 if (getfrag(from, skb_put(skb, copy),
1514 offset, copy, off, skb) < 0) {
1515 __skb_trim(skb, off);
1520 int i = skb_shinfo(skb)->nr_frags;
1523 if (!sk_page_frag_refill(sk, pfrag))
1526 if (!skb_can_coalesce(skb, i, pfrag->page,
1529 if (i == MAX_SKB_FRAGS)
1532 __skb_fill_page_desc(skb, i, pfrag->page,
1534 skb_shinfo(skb)->nr_frags = ++i;
1535 get_page(pfrag->page);
1537 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1539 page_address(pfrag->page) + pfrag->offset,
1540 offset, copy, skb->len, skb) < 0)
1543 pfrag->offset += copy;
1544 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1546 skb->data_len += copy;
1547 skb->truesize += copy;
1548 atomic_add(copy, &sk->sk_wmem_alloc);
1559 cork->length -= length;
1560 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1564 int ip6_append_data(struct sock *sk,
1565 int getfrag(void *from, char *to, int offset, int len,
1566 int odd, struct sk_buff *skb),
1567 void *from, int length, int transhdrlen, int hlimit,
1568 int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1569 struct rt6_info *rt, unsigned int flags, int dontfrag)
1571 struct inet_sock *inet = inet_sk(sk);
1572 struct ipv6_pinfo *np = inet6_sk(sk);
1576 if (flags&MSG_PROBE)
1578 if (skb_queue_empty(&sk->sk_write_queue)) {
1582 err = ip6_setup_cork(sk, &inet->cork, &np->cork, hlimit,
1583 tclass, opt, rt, fl6);
1587 exthdrlen = (opt ? opt->opt_flen : 0);
1588 length += exthdrlen;
1589 transhdrlen += exthdrlen;
1591 fl6 = &inet->cork.fl.u.ip6;
1595 return __ip6_append_data(sk, fl6, &sk->sk_write_queue, &inet->cork.base,
1596 &np->cork, sk_page_frag(sk), getfrag,
1597 from, length, transhdrlen, flags, dontfrag);
1599 EXPORT_SYMBOL_GPL(ip6_append_data);
1601 static void ip6_cork_release(struct inet_cork_full *cork,
1602 struct inet6_cork *v6_cork)
1605 kfree(v6_cork->opt->dst0opt);
1606 kfree(v6_cork->opt->dst1opt);
1607 kfree(v6_cork->opt->hopopt);
1608 kfree(v6_cork->opt->srcrt);
1609 kfree(v6_cork->opt);
1610 v6_cork->opt = NULL;
1613 if (cork->base.dst) {
1614 dst_release(cork->base.dst);
1615 cork->base.dst = NULL;
1616 cork->base.flags &= ~IPCORK_ALLFRAG;
1618 memset(&cork->fl, 0, sizeof(cork->fl));
1621 struct sk_buff *__ip6_make_skb(struct sock *sk,
1622 struct sk_buff_head *queue,
1623 struct inet_cork_full *cork,
1624 struct inet6_cork *v6_cork)
1626 struct sk_buff *skb, *tmp_skb;
1627 struct sk_buff **tail_skb;
1628 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1629 struct ipv6_pinfo *np = inet6_sk(sk);
1630 struct net *net = sock_net(sk);
1631 struct ipv6hdr *hdr;
1632 struct ipv6_txoptions *opt = v6_cork->opt;
1633 struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1634 struct flowi6 *fl6 = &cork->fl.u.ip6;
1635 unsigned char proto = fl6->flowi6_proto;
1637 skb = __skb_dequeue(queue);
1640 tail_skb = &(skb_shinfo(skb)->frag_list);
1642 /* move skb->data to ip header from ext header */
1643 if (skb->data < skb_network_header(skb))
1644 __skb_pull(skb, skb_network_offset(skb));
1645 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1646 __skb_pull(tmp_skb, skb_network_header_len(skb));
1647 *tail_skb = tmp_skb;
1648 tail_skb = &(tmp_skb->next);
1649 skb->len += tmp_skb->len;
1650 skb->data_len += tmp_skb->len;
1651 skb->truesize += tmp_skb->truesize;
1652 tmp_skb->destructor = NULL;
1656 /* Allow local fragmentation. */
1657 skb->ignore_df = ip6_sk_ignore_df(sk);
1659 *final_dst = fl6->daddr;
1660 __skb_pull(skb, skb_network_header_len(skb));
1661 if (opt && opt->opt_flen)
1662 ipv6_push_frag_opts(skb, opt, &proto);
1663 if (opt && opt->opt_nflen)
1664 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1666 skb_push(skb, sizeof(struct ipv6hdr));
1667 skb_reset_network_header(skb);
1668 hdr = ipv6_hdr(skb);
1670 ip6_flow_hdr(hdr, v6_cork->tclass,
1671 ip6_make_flowlabel(net, skb, fl6->flowlabel,
1672 np->autoflowlabel, fl6));
1673 hdr->hop_limit = v6_cork->hop_limit;
1674 hdr->nexthdr = proto;
1675 hdr->saddr = fl6->saddr;
1676 hdr->daddr = *final_dst;
1678 skb->priority = sk->sk_priority;
1679 skb->mark = sk->sk_mark;
1681 skb_dst_set(skb, dst_clone(&rt->dst));
1682 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1683 if (proto == IPPROTO_ICMPV6) {
1684 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1686 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1687 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1690 ip6_cork_release(cork, v6_cork);
1695 int ip6_send_skb(struct sk_buff *skb)
1697 struct net *net = sock_net(skb->sk);
1698 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1701 err = ip6_local_out(net, skb->sk, skb);
1704 err = net_xmit_errno(err);
1706 IP6_INC_STATS(net, rt->rt6i_idev,
1707 IPSTATS_MIB_OUTDISCARDS);
1713 int ip6_push_pending_frames(struct sock *sk)
1715 struct sk_buff *skb;
1717 skb = ip6_finish_skb(sk);
1721 return ip6_send_skb(skb);
1723 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1725 static void __ip6_flush_pending_frames(struct sock *sk,
1726 struct sk_buff_head *queue,
1727 struct inet_cork_full *cork,
1728 struct inet6_cork *v6_cork)
1730 struct sk_buff *skb;
1732 while ((skb = __skb_dequeue_tail(queue)) != NULL) {
1734 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1735 IPSTATS_MIB_OUTDISCARDS);
1739 ip6_cork_release(cork, v6_cork);
1742 void ip6_flush_pending_frames(struct sock *sk)
1744 __ip6_flush_pending_frames(sk, &sk->sk_write_queue,
1745 &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
1747 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1749 struct sk_buff *ip6_make_skb(struct sock *sk,
1750 int getfrag(void *from, char *to, int offset,
1751 int len, int odd, struct sk_buff *skb),
1752 void *from, int length, int transhdrlen,
1753 int hlimit, int tclass,
1754 struct ipv6_txoptions *opt, struct flowi6 *fl6,
1755 struct rt6_info *rt, unsigned int flags,
1758 struct inet_cork_full cork;
1759 struct inet6_cork v6_cork;
1760 struct sk_buff_head queue;
1761 int exthdrlen = (opt ? opt->opt_flen : 0);
1764 if (flags & MSG_PROBE)
1767 __skb_queue_head_init(&queue);
1769 cork.base.flags = 0;
1771 cork.base.opt = NULL;
1773 err = ip6_setup_cork(sk, &cork, &v6_cork, hlimit, tclass, opt, rt, fl6);
1775 return ERR_PTR(err);
1778 dontfrag = inet6_sk(sk)->dontfrag;
1780 err = __ip6_append_data(sk, fl6, &queue, &cork.base, &v6_cork,
1781 ¤t->task_frag, getfrag, from,
1782 length + exthdrlen, transhdrlen + exthdrlen,
1785 __ip6_flush_pending_frames(sk, &queue, &cork, &v6_cork);
1786 return ERR_PTR(err);
1789 return __ip6_make_skb(sk, &queue, &cork, &v6_cork);