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 net *net, 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(net, 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 net, sk, newskb, NULL, newskb->dev,
88 if (ipv6_hdr(skb)->hop_limit == 0) {
89 IP6_INC_STATS(net, idev,
90 IPSTATS_MIB_OUTDISCARDS);
96 IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUTMCAST, skb->len);
98 if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb)->daddr) <=
99 IPV6_ADDR_SCOPE_NODELOCAL &&
100 !(dev->flags & IFF_LOOPBACK)) {
107 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
108 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
109 if (unlikely(!neigh))
110 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
111 if (!IS_ERR(neigh)) {
112 ret = dst_neigh_output(dst, neigh, skb);
113 rcu_read_unlock_bh();
116 rcu_read_unlock_bh();
118 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
123 static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
125 if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
126 dst_allfrag(skb_dst(skb)) ||
127 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
128 return ip6_fragment(net, sk, skb, ip6_finish_output2);
130 return ip6_finish_output2(net, sk, skb);
133 int ip6_output(struct sock *sk, struct sk_buff *skb)
135 struct net_device *dev = skb_dst(skb)->dev;
136 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
137 struct net *net = dev_net(dev);
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 (skb_warn_if_lro(skb))
395 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
396 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
397 IPSTATS_MIB_INDISCARDS);
401 skb_forward_csum(skb);
404 * We DO NOT make any processing on
405 * RA packets, pushing them to user level AS IS
406 * without ane WARRANTY that application will be able
407 * to interpret them. The reason is that we
408 * cannot make anything clever here.
410 * We are not end-node, so that if packet contains
411 * AH/ESP, we cannot make anything.
412 * Defragmentation also would be mistake, RA packets
413 * cannot be fragmented, because there is no warranty
414 * that different fragments will go along one path. --ANK
416 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
417 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
422 * check and decrement ttl
424 if (hdr->hop_limit <= 1) {
425 /* Force OUTPUT device used as source address */
427 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
428 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
429 IPSTATS_MIB_INHDRERRORS);
435 /* XXX: idev->cnf.proxy_ndp? */
436 if (net->ipv6.devconf_all->proxy_ndp &&
437 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
438 int proxied = ip6_forward_proxy_check(skb);
440 return ip6_input(skb);
441 else if (proxied < 0) {
442 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
443 IPSTATS_MIB_INDISCARDS);
448 if (!xfrm6_route_forward(skb)) {
449 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
450 IPSTATS_MIB_INDISCARDS);
455 /* IPv6 specs say nothing about it, but it is clear that we cannot
456 send redirects to source routed frames.
457 We don't send redirects to frames decapsulated from IPsec.
459 if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
460 struct in6_addr *target = NULL;
461 struct inet_peer *peer;
465 * incoming and outgoing devices are the same
469 rt = (struct rt6_info *) dst;
470 if (rt->rt6i_flags & RTF_GATEWAY)
471 target = &rt->rt6i_gateway;
473 target = &hdr->daddr;
475 peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
477 /* Limit redirects both by destination (here)
478 and by source (inside ndisc_send_redirect)
480 if (inet_peer_xrlim_allow(peer, 1*HZ))
481 ndisc_send_redirect(skb, target);
485 int addrtype = ipv6_addr_type(&hdr->saddr);
487 /* This check is security critical. */
488 if (addrtype == IPV6_ADDR_ANY ||
489 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
491 if (addrtype & IPV6_ADDR_LINKLOCAL) {
492 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
493 ICMPV6_NOT_NEIGHBOUR, 0);
498 mtu = ip6_dst_mtu_forward(dst);
499 if (mtu < IPV6_MIN_MTU)
502 if (ip6_pkt_too_big(skb, mtu)) {
503 /* Again, force OUTPUT device used as source address */
505 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
506 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
507 IPSTATS_MIB_INTOOBIGERRORS);
508 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
509 IPSTATS_MIB_FRAGFAILS);
514 if (skb_cow(skb, dst->dev->hard_header_len)) {
515 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
516 IPSTATS_MIB_OUTDISCARDS);
522 /* Mangling hops number delayed to point after skb COW */
526 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
527 IP6_ADD_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
528 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
529 net, NULL, skb, skb->dev, dst->dev,
533 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
539 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
541 to->pkt_type = from->pkt_type;
542 to->priority = from->priority;
543 to->protocol = from->protocol;
545 skb_dst_set(to, dst_clone(skb_dst(from)));
547 to->mark = from->mark;
549 #ifdef CONFIG_NET_SCHED
550 to->tc_index = from->tc_index;
553 skb_copy_secmark(to, from);
556 int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
557 int (*output)(struct net *, struct sock *, struct sk_buff *))
559 struct sk_buff *frag;
560 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
561 struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
562 inet6_sk(skb->sk) : NULL;
563 struct ipv6hdr *tmp_hdr;
565 unsigned int mtu, hlen, left, len;
568 int ptr, offset = 0, err = 0;
569 u8 *prevhdr, nexthdr = 0;
571 hlen = ip6_find_1stfragopt(skb, &prevhdr);
574 mtu = ip6_skb_dst_mtu(skb);
576 /* We must not fragment if the socket is set to force MTU discovery
577 * or if the skb it not generated by a local socket.
579 if (unlikely(!skb->ignore_df && skb->len > mtu))
582 if (IP6CB(skb)->frag_max_size) {
583 if (IP6CB(skb)->frag_max_size > mtu)
586 /* don't send fragments larger than what we received */
587 mtu = IP6CB(skb)->frag_max_size;
588 if (mtu < IPV6_MIN_MTU)
592 if (np && np->frag_size < mtu) {
596 mtu -= hlen + sizeof(struct frag_hdr);
598 frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
599 &ipv6_hdr(skb)->saddr);
601 hroom = LL_RESERVED_SPACE(rt->dst.dev);
602 if (skb_has_frag_list(skb)) {
603 int first_len = skb_pagelen(skb);
604 struct sk_buff *frag2;
606 if (first_len - hlen > mtu ||
607 ((first_len - hlen) & 7) ||
609 skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
612 skb_walk_frags(skb, frag) {
613 /* Correct geometry. */
614 if (frag->len > mtu ||
615 ((frag->len & 7) && frag->next) ||
616 skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
617 goto slow_path_clean;
619 /* Partially cloned skb? */
620 if (skb_shared(frag))
621 goto slow_path_clean;
626 frag->destructor = sock_wfree;
628 skb->truesize -= frag->truesize;
635 *prevhdr = NEXTHDR_FRAGMENT;
636 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
638 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
639 IPSTATS_MIB_FRAGFAILS);
643 frag = skb_shinfo(skb)->frag_list;
644 skb_frag_list_init(skb);
646 __skb_pull(skb, hlen);
647 fh = (struct frag_hdr *)__skb_push(skb, sizeof(struct frag_hdr));
648 __skb_push(skb, hlen);
649 skb_reset_network_header(skb);
650 memcpy(skb_network_header(skb), tmp_hdr, hlen);
652 fh->nexthdr = nexthdr;
654 fh->frag_off = htons(IP6_MF);
655 fh->identification = frag_id;
657 first_len = skb_pagelen(skb);
658 skb->data_len = first_len - skb_headlen(skb);
659 skb->len = first_len;
660 ipv6_hdr(skb)->payload_len = htons(first_len -
661 sizeof(struct ipv6hdr));
666 /* Prepare header of the next frame,
667 * before previous one went down. */
669 frag->ip_summed = CHECKSUM_NONE;
670 skb_reset_transport_header(frag);
671 fh = (struct frag_hdr *)__skb_push(frag, sizeof(struct frag_hdr));
672 __skb_push(frag, hlen);
673 skb_reset_network_header(frag);
674 memcpy(skb_network_header(frag), tmp_hdr,
676 offset += skb->len - hlen - sizeof(struct frag_hdr);
677 fh->nexthdr = nexthdr;
679 fh->frag_off = htons(offset);
681 fh->frag_off |= htons(IP6_MF);
682 fh->identification = frag_id;
683 ipv6_hdr(frag)->payload_len =
685 sizeof(struct ipv6hdr));
686 ip6_copy_metadata(frag, skb);
689 err = output(net, sk, skb);
691 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
692 IPSTATS_MIB_FRAGCREATES);
705 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
706 IPSTATS_MIB_FRAGOKS);
711 kfree_skb_list(frag);
713 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
714 IPSTATS_MIB_FRAGFAILS);
719 skb_walk_frags(skb, frag2) {
723 frag2->destructor = NULL;
724 skb->truesize += frag2->truesize;
729 if ((skb->ip_summed == CHECKSUM_PARTIAL) &&
730 skb_checksum_help(skb))
733 left = skb->len - hlen; /* Space per frame */
734 ptr = hlen; /* Where to start from */
737 * Fragment the datagram.
740 *prevhdr = NEXTHDR_FRAGMENT;
741 troom = rt->dst.dev->needed_tailroom;
744 * Keep copying data until we run out.
748 /* IF: it doesn't fit, use 'mtu' - the data space left */
751 /* IF: we are not sending up to and including the packet end
752 then align the next start on an eight byte boundary */
757 /* Allocate buffer */
758 frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
759 hroom + troom, GFP_ATOMIC);
761 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
762 IPSTATS_MIB_FRAGFAILS);
768 * Set up data on packet
771 ip6_copy_metadata(frag, skb);
772 skb_reserve(frag, hroom);
773 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
774 skb_reset_network_header(frag);
775 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
776 frag->transport_header = (frag->network_header + hlen +
777 sizeof(struct frag_hdr));
780 * Charge the memory for the fragment to any owner
784 skb_set_owner_w(frag, skb->sk);
787 * Copy the packet header into the new buffer.
789 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
792 * Build fragment header.
794 fh->nexthdr = nexthdr;
796 fh->identification = frag_id;
799 * Copy a block of the IP datagram.
801 BUG_ON(skb_copy_bits(skb, ptr, skb_transport_header(frag),
805 fh->frag_off = htons(offset);
807 fh->frag_off |= htons(IP6_MF);
808 ipv6_hdr(frag)->payload_len = htons(frag->len -
809 sizeof(struct ipv6hdr));
815 * Put this fragment into the sending queue.
817 err = output(net, sk, frag);
821 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
822 IPSTATS_MIB_FRAGCREATES);
824 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
825 IPSTATS_MIB_FRAGOKS);
830 if (skb->sk && dst_allfrag(skb_dst(skb)))
831 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
833 skb->dev = skb_dst(skb)->dev;
834 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
838 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
839 IPSTATS_MIB_FRAGFAILS);
844 static inline int ip6_rt_check(const struct rt6key *rt_key,
845 const struct in6_addr *fl_addr,
846 const struct in6_addr *addr_cache)
848 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
849 (!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
852 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
853 struct dst_entry *dst,
854 const struct flowi6 *fl6)
856 struct ipv6_pinfo *np = inet6_sk(sk);
862 if (dst->ops->family != AF_INET6) {
867 rt = (struct rt6_info *)dst;
868 /* Yes, checking route validity in not connected
869 * case is not very simple. Take into account,
870 * that we do not support routing by source, TOS,
871 * and MSG_DONTROUTE --ANK (980726)
873 * 1. ip6_rt_check(): If route was host route,
874 * check that cached destination is current.
875 * If it is network route, we still may
876 * check its validity using saved pointer
877 * to the last used address: daddr_cache.
878 * We do not want to save whole address now,
879 * (because main consumer of this service
880 * is tcp, which has not this problem),
881 * so that the last trick works only on connected
883 * 2. oif also should be the same.
885 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
886 #ifdef CONFIG_IPV6_SUBTREES
887 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
889 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
898 static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
899 struct dst_entry **dst, struct flowi6 *fl6)
901 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
907 /* The correct way to handle this would be to do
908 * ip6_route_get_saddr, and then ip6_route_output; however,
909 * the route-specific preferred source forces the
910 * ip6_route_output call _before_ ip6_route_get_saddr.
912 * In source specific routing (no src=any default route),
913 * ip6_route_output will fail given src=any saddr, though, so
914 * that's why we try it again later.
916 if (ipv6_addr_any(&fl6->saddr) && (!*dst || !(*dst)->error)) {
918 bool had_dst = *dst != NULL;
921 *dst = ip6_route_output(net, sk, fl6);
922 rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
923 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
924 sk ? inet6_sk(sk)->srcprefs : 0,
927 goto out_err_release;
929 /* If we had an erroneous initial result, pretend it
930 * never existed and let the SA-enabled version take
933 if (!had_dst && (*dst)->error) {
940 *dst = ip6_route_output(net, sk, fl6);
944 goto out_err_release;
946 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
948 * Here if the dst entry we've looked up
949 * has a neighbour entry that is in the INCOMPLETE
950 * state and the src address from the flow is
951 * marked as OPTIMISTIC, we release the found
952 * dst entry and replace it instead with the
953 * dst entry of the nexthop router
955 rt = (struct rt6_info *) *dst;
957 n = __ipv6_neigh_lookup_noref(rt->dst.dev,
958 rt6_nexthop(rt, &fl6->daddr));
959 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
960 rcu_read_unlock_bh();
963 struct inet6_ifaddr *ifp;
964 struct flowi6 fl_gw6;
967 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
970 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
976 * We need to get the dst entry for the
977 * default router instead
980 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
981 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
982 *dst = ip6_route_output(net, sk, &fl_gw6);
985 goto out_err_release;
993 if (err == -ENETUNREACH)
994 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1001 * ip6_dst_lookup - perform route lookup on flow
1002 * @sk: socket which provides route info
1003 * @dst: pointer to dst_entry * for result
1004 * @fl6: flow to lookup
1006 * This function performs a route lookup on the given flow.
1008 * It returns zero on success, or a standard errno code on error.
1010 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1014 return ip6_dst_lookup_tail(net, sk, dst, fl6);
1016 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1019 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1020 * @sk: socket which provides route info
1021 * @fl6: flow to lookup
1022 * @final_dst: final destination address for ipsec lookup
1024 * This function performs a route lookup on the given flow.
1026 * It returns a valid dst pointer on success, or a pointer encoded
1029 struct dst_entry *ip6_dst_lookup_flow(const struct sock *sk, struct flowi6 *fl6,
1030 const struct in6_addr *final_dst)
1032 struct dst_entry *dst = NULL;
1035 err = ip6_dst_lookup_tail(sock_net(sk), sk, &dst, fl6);
1037 return ERR_PTR(err);
1039 fl6->daddr = *final_dst;
1040 if (!fl6->flowi6_oif)
1041 fl6->flowi6_oif = dst->dev->ifindex;
1043 return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1045 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1048 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1049 * @sk: socket which provides the dst cache and route info
1050 * @fl6: flow to lookup
1051 * @final_dst: final destination address for ipsec lookup
1053 * This function performs a route lookup on the given flow with the
1054 * possibility of using the cached route in the socket if it is valid.
1055 * It will take the socket dst lock when operating on the dst cache.
1056 * As a result, this function can only be used in process context.
1058 * It returns a valid dst pointer on success, or a pointer encoded
1061 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1062 const struct in6_addr *final_dst)
1064 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1067 dst = ip6_sk_dst_check(sk, dst, fl6);
1069 err = ip6_dst_lookup_tail(sock_net(sk), sk, &dst, fl6);
1071 return ERR_PTR(err);
1073 fl6->daddr = *final_dst;
1075 return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1077 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1079 static inline int ip6_ufo_append_data(struct sock *sk,
1080 struct sk_buff_head *queue,
1081 int getfrag(void *from, char *to, int offset, int len,
1082 int odd, struct sk_buff *skb),
1083 void *from, int length, int hh_len, int fragheaderlen,
1084 int transhdrlen, int mtu, unsigned int flags,
1085 const struct flowi6 *fl6)
1088 struct sk_buff *skb;
1091 /* There is support for UDP large send offload by network
1092 * device, so create one single skb packet containing complete
1095 skb = skb_peek_tail(queue);
1097 skb = sock_alloc_send_skb(sk,
1098 hh_len + fragheaderlen + transhdrlen + 20,
1099 (flags & MSG_DONTWAIT), &err);
1103 /* reserve space for Hardware header */
1104 skb_reserve(skb, hh_len);
1106 /* create space for UDP/IP header */
1107 skb_put(skb, fragheaderlen + transhdrlen);
1109 /* initialize network header pointer */
1110 skb_reset_network_header(skb);
1112 /* initialize protocol header pointer */
1113 skb->transport_header = skb->network_header + fragheaderlen;
1115 skb->protocol = htons(ETH_P_IPV6);
1118 __skb_queue_tail(queue, skb);
1119 } else if (skb_is_gso(skb)) {
1123 skb->ip_summed = CHECKSUM_PARTIAL;
1124 /* Specify the length of each IPv6 datagram fragment.
1125 * It has to be a multiple of 8.
1127 skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1128 sizeof(struct frag_hdr)) & ~7;
1129 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1130 skb_shinfo(skb)->ip6_frag_id = ipv6_select_ident(sock_net(sk),
1135 return skb_append_datato_frags(sk, skb, getfrag, from,
1136 (length - transhdrlen));
1139 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1142 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1145 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1148 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1151 static void ip6_append_data_mtu(unsigned int *mtu,
1153 unsigned int fragheaderlen,
1154 struct sk_buff *skb,
1155 struct rt6_info *rt,
1156 unsigned int orig_mtu)
1158 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1160 /* first fragment, reserve header_len */
1161 *mtu = orig_mtu - rt->dst.header_len;
1165 * this fragment is not first, the headers
1166 * space is regarded as data space.
1170 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1171 + fragheaderlen - sizeof(struct frag_hdr);
1175 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1176 struct inet6_cork *v6_cork,
1177 int hlimit, int tclass, struct ipv6_txoptions *opt,
1178 struct rt6_info *rt, struct flowi6 *fl6)
1180 struct ipv6_pinfo *np = inet6_sk(sk);
1187 if (WARN_ON(v6_cork->opt))
1190 v6_cork->opt = kzalloc(opt->tot_len, sk->sk_allocation);
1191 if (unlikely(!v6_cork->opt))
1194 v6_cork->opt->tot_len = opt->tot_len;
1195 v6_cork->opt->opt_flen = opt->opt_flen;
1196 v6_cork->opt->opt_nflen = opt->opt_nflen;
1198 v6_cork->opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1200 if (opt->dst0opt && !v6_cork->opt->dst0opt)
1203 v6_cork->opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1205 if (opt->dst1opt && !v6_cork->opt->dst1opt)
1208 v6_cork->opt->hopopt = ip6_opt_dup(opt->hopopt,
1210 if (opt->hopopt && !v6_cork->opt->hopopt)
1213 v6_cork->opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1215 if (opt->srcrt && !v6_cork->opt->srcrt)
1218 /* need source address above miyazawa*/
1221 cork->base.dst = &rt->dst;
1222 cork->fl.u.ip6 = *fl6;
1223 v6_cork->hop_limit = hlimit;
1224 v6_cork->tclass = tclass;
1225 if (rt->dst.flags & DST_XFRM_TUNNEL)
1226 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1227 rt->dst.dev->mtu : dst_mtu(&rt->dst);
1229 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1230 rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1231 if (np->frag_size < mtu) {
1233 mtu = np->frag_size;
1235 cork->base.fragsize = mtu;
1236 if (dst_allfrag(rt->dst.path))
1237 cork->base.flags |= IPCORK_ALLFRAG;
1238 cork->base.length = 0;
1243 static int __ip6_append_data(struct sock *sk,
1245 struct sk_buff_head *queue,
1246 struct inet_cork *cork,
1247 struct inet6_cork *v6_cork,
1248 struct page_frag *pfrag,
1249 int getfrag(void *from, char *to, int offset,
1250 int len, int odd, struct sk_buff *skb),
1251 void *from, int length, int transhdrlen,
1252 unsigned int flags, int dontfrag)
1254 struct sk_buff *skb, *skb_prev = NULL;
1255 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu;
1257 int dst_exthdrlen = 0;
1264 struct rt6_info *rt = (struct rt6_info *)cork->dst;
1265 struct ipv6_txoptions *opt = v6_cork->opt;
1266 int csummode = CHECKSUM_NONE;
1268 skb = skb_peek_tail(queue);
1270 exthdrlen = opt ? opt->opt_flen : 0;
1271 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1274 mtu = cork->fragsize;
1277 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1279 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1280 (opt ? opt->opt_nflen : 0);
1281 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1282 sizeof(struct frag_hdr);
1284 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1285 unsigned int maxnonfragsize, headersize;
1287 headersize = sizeof(struct ipv6hdr) +
1288 (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1289 (dst_allfrag(&rt->dst) ?
1290 sizeof(struct frag_hdr) : 0) +
1291 rt->rt6i_nfheader_len;
1293 if (ip6_sk_ignore_df(sk))
1294 maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1296 maxnonfragsize = mtu;
1298 /* dontfrag active */
1299 if ((cork->length + length > mtu - headersize) && dontfrag &&
1300 (sk->sk_protocol == IPPROTO_UDP ||
1301 sk->sk_protocol == IPPROTO_RAW)) {
1302 ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1303 sizeof(struct ipv6hdr));
1307 if (cork->length + length > maxnonfragsize - headersize) {
1309 ipv6_local_error(sk, EMSGSIZE, fl6,
1311 sizeof(struct ipv6hdr));
1316 if (sk->sk_type == SOCK_DGRAM || sk->sk_type == SOCK_RAW) {
1317 sock_tx_timestamp(sk, &tx_flags);
1318 if (tx_flags & SKBTX_ANY_SW_TSTAMP &&
1319 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1320 tskey = sk->sk_tskey++;
1323 /* If this is the first and only packet and device
1324 * supports checksum offloading, let's use it.
1325 * Use transhdrlen, same as IPv4, because partial
1326 * sums only work when transhdrlen is set.
1328 if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1329 length + fragheaderlen < mtu &&
1330 rt->dst.dev->features & NETIF_F_V6_CSUM &&
1332 csummode = CHECKSUM_PARTIAL;
1334 * Let's try using as much space as possible.
1335 * Use MTU if total length of the message fits into the MTU.
1336 * Otherwise, we need to reserve fragment header and
1337 * fragment alignment (= 8-15 octects, in total).
1339 * Note that we may need to "move" the data from the tail of
1340 * of the buffer to the new fragment when we split
1343 * FIXME: It may be fragmented into multiple chunks
1344 * at once if non-fragmentable extension headers
1349 cork->length += length;
1350 if (((length > mtu) ||
1351 (skb && skb_is_gso(skb))) &&
1352 (sk->sk_protocol == IPPROTO_UDP) &&
1353 (rt->dst.dev->features & NETIF_F_UFO) &&
1354 (sk->sk_type == SOCK_DGRAM)) {
1355 err = ip6_ufo_append_data(sk, queue, getfrag, from, length,
1356 hh_len, fragheaderlen,
1357 transhdrlen, mtu, flags, fl6);
1366 while (length > 0) {
1367 /* Check if the remaining data fits into current packet. */
1368 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1370 copy = maxfraglen - skb->len;
1374 unsigned int datalen;
1375 unsigned int fraglen;
1376 unsigned int fraggap;
1377 unsigned int alloclen;
1379 /* There's no room in the current skb */
1381 fraggap = skb->len - maxfraglen;
1384 /* update mtu and maxfraglen if necessary */
1385 if (!skb || !skb_prev)
1386 ip6_append_data_mtu(&mtu, &maxfraglen,
1387 fragheaderlen, skb, rt,
1393 * If remaining data exceeds the mtu,
1394 * we know we need more fragment(s).
1396 datalen = length + fraggap;
1398 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1399 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1400 if ((flags & MSG_MORE) &&
1401 !(rt->dst.dev->features&NETIF_F_SG))
1404 alloclen = datalen + fragheaderlen;
1406 alloclen += dst_exthdrlen;
1408 if (datalen != length + fraggap) {
1410 * this is not the last fragment, the trailer
1411 * space is regarded as data space.
1413 datalen += rt->dst.trailer_len;
1416 alloclen += rt->dst.trailer_len;
1417 fraglen = datalen + fragheaderlen;
1420 * We just reserve space for fragment header.
1421 * Note: this may be overallocation if the message
1422 * (without MSG_MORE) fits into the MTU.
1424 alloclen += sizeof(struct frag_hdr);
1427 skb = sock_alloc_send_skb(sk,
1429 (flags & MSG_DONTWAIT), &err);
1432 if (atomic_read(&sk->sk_wmem_alloc) <=
1434 skb = sock_wmalloc(sk,
1435 alloclen + hh_len, 1,
1443 * Fill in the control structures
1445 skb->protocol = htons(ETH_P_IPV6);
1446 skb->ip_summed = csummode;
1448 /* reserve for fragmentation and ipsec header */
1449 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1452 /* Only the initial fragment is time stamped */
1453 skb_shinfo(skb)->tx_flags = tx_flags;
1455 skb_shinfo(skb)->tskey = tskey;
1459 * Find where to start putting bytes
1461 data = skb_put(skb, fraglen);
1462 skb_set_network_header(skb, exthdrlen);
1463 data += fragheaderlen;
1464 skb->transport_header = (skb->network_header +
1467 skb->csum = skb_copy_and_csum_bits(
1468 skb_prev, maxfraglen,
1469 data + transhdrlen, fraggap, 0);
1470 skb_prev->csum = csum_sub(skb_prev->csum,
1473 pskb_trim_unique(skb_prev, maxfraglen);
1475 copy = datalen - transhdrlen - fraggap;
1481 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1488 length -= datalen - fraggap;
1494 * Put the packet on the pending queue
1496 __skb_queue_tail(queue, skb);
1503 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1507 if (getfrag(from, skb_put(skb, copy),
1508 offset, copy, off, skb) < 0) {
1509 __skb_trim(skb, off);
1514 int i = skb_shinfo(skb)->nr_frags;
1517 if (!sk_page_frag_refill(sk, pfrag))
1520 if (!skb_can_coalesce(skb, i, pfrag->page,
1523 if (i == MAX_SKB_FRAGS)
1526 __skb_fill_page_desc(skb, i, pfrag->page,
1528 skb_shinfo(skb)->nr_frags = ++i;
1529 get_page(pfrag->page);
1531 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1533 page_address(pfrag->page) + pfrag->offset,
1534 offset, copy, skb->len, skb) < 0)
1537 pfrag->offset += copy;
1538 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1540 skb->data_len += copy;
1541 skb->truesize += copy;
1542 atomic_add(copy, &sk->sk_wmem_alloc);
1553 cork->length -= length;
1554 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1558 int ip6_append_data(struct sock *sk,
1559 int getfrag(void *from, char *to, int offset, int len,
1560 int odd, struct sk_buff *skb),
1561 void *from, int length, int transhdrlen, int hlimit,
1562 int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1563 struct rt6_info *rt, unsigned int flags, int dontfrag)
1565 struct inet_sock *inet = inet_sk(sk);
1566 struct ipv6_pinfo *np = inet6_sk(sk);
1570 if (flags&MSG_PROBE)
1572 if (skb_queue_empty(&sk->sk_write_queue)) {
1576 err = ip6_setup_cork(sk, &inet->cork, &np->cork, hlimit,
1577 tclass, opt, rt, fl6);
1581 exthdrlen = (opt ? opt->opt_flen : 0);
1582 length += exthdrlen;
1583 transhdrlen += exthdrlen;
1585 fl6 = &inet->cork.fl.u.ip6;
1589 return __ip6_append_data(sk, fl6, &sk->sk_write_queue, &inet->cork.base,
1590 &np->cork, sk_page_frag(sk), getfrag,
1591 from, length, transhdrlen, flags, dontfrag);
1593 EXPORT_SYMBOL_GPL(ip6_append_data);
1595 static void ip6_cork_release(struct inet_cork_full *cork,
1596 struct inet6_cork *v6_cork)
1599 kfree(v6_cork->opt->dst0opt);
1600 kfree(v6_cork->opt->dst1opt);
1601 kfree(v6_cork->opt->hopopt);
1602 kfree(v6_cork->opt->srcrt);
1603 kfree(v6_cork->opt);
1604 v6_cork->opt = NULL;
1607 if (cork->base.dst) {
1608 dst_release(cork->base.dst);
1609 cork->base.dst = NULL;
1610 cork->base.flags &= ~IPCORK_ALLFRAG;
1612 memset(&cork->fl, 0, sizeof(cork->fl));
1615 struct sk_buff *__ip6_make_skb(struct sock *sk,
1616 struct sk_buff_head *queue,
1617 struct inet_cork_full *cork,
1618 struct inet6_cork *v6_cork)
1620 struct sk_buff *skb, *tmp_skb;
1621 struct sk_buff **tail_skb;
1622 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1623 struct ipv6_pinfo *np = inet6_sk(sk);
1624 struct net *net = sock_net(sk);
1625 struct ipv6hdr *hdr;
1626 struct ipv6_txoptions *opt = v6_cork->opt;
1627 struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1628 struct flowi6 *fl6 = &cork->fl.u.ip6;
1629 unsigned char proto = fl6->flowi6_proto;
1631 skb = __skb_dequeue(queue);
1634 tail_skb = &(skb_shinfo(skb)->frag_list);
1636 /* move skb->data to ip header from ext header */
1637 if (skb->data < skb_network_header(skb))
1638 __skb_pull(skb, skb_network_offset(skb));
1639 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1640 __skb_pull(tmp_skb, skb_network_header_len(skb));
1641 *tail_skb = tmp_skb;
1642 tail_skb = &(tmp_skb->next);
1643 skb->len += tmp_skb->len;
1644 skb->data_len += tmp_skb->len;
1645 skb->truesize += tmp_skb->truesize;
1646 tmp_skb->destructor = NULL;
1650 /* Allow local fragmentation. */
1651 skb->ignore_df = ip6_sk_ignore_df(sk);
1653 *final_dst = fl6->daddr;
1654 __skb_pull(skb, skb_network_header_len(skb));
1655 if (opt && opt->opt_flen)
1656 ipv6_push_frag_opts(skb, opt, &proto);
1657 if (opt && opt->opt_nflen)
1658 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1660 skb_push(skb, sizeof(struct ipv6hdr));
1661 skb_reset_network_header(skb);
1662 hdr = ipv6_hdr(skb);
1664 ip6_flow_hdr(hdr, v6_cork->tclass,
1665 ip6_make_flowlabel(net, skb, fl6->flowlabel,
1666 np->autoflowlabel, fl6));
1667 hdr->hop_limit = v6_cork->hop_limit;
1668 hdr->nexthdr = proto;
1669 hdr->saddr = fl6->saddr;
1670 hdr->daddr = *final_dst;
1672 skb->priority = sk->sk_priority;
1673 skb->mark = sk->sk_mark;
1675 skb_dst_set(skb, dst_clone(&rt->dst));
1676 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1677 if (proto == IPPROTO_ICMPV6) {
1678 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1680 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1681 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1684 ip6_cork_release(cork, v6_cork);
1689 int ip6_send_skb(struct sk_buff *skb)
1691 struct net *net = sock_net(skb->sk);
1692 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1695 err = ip6_local_out(skb->sk, skb);
1698 err = net_xmit_errno(err);
1700 IP6_INC_STATS(net, rt->rt6i_idev,
1701 IPSTATS_MIB_OUTDISCARDS);
1707 int ip6_push_pending_frames(struct sock *sk)
1709 struct sk_buff *skb;
1711 skb = ip6_finish_skb(sk);
1715 return ip6_send_skb(skb);
1717 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1719 static void __ip6_flush_pending_frames(struct sock *sk,
1720 struct sk_buff_head *queue,
1721 struct inet_cork_full *cork,
1722 struct inet6_cork *v6_cork)
1724 struct sk_buff *skb;
1726 while ((skb = __skb_dequeue_tail(queue)) != NULL) {
1728 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1729 IPSTATS_MIB_OUTDISCARDS);
1733 ip6_cork_release(cork, v6_cork);
1736 void ip6_flush_pending_frames(struct sock *sk)
1738 __ip6_flush_pending_frames(sk, &sk->sk_write_queue,
1739 &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
1741 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1743 struct sk_buff *ip6_make_skb(struct sock *sk,
1744 int getfrag(void *from, char *to, int offset,
1745 int len, int odd, struct sk_buff *skb),
1746 void *from, int length, int transhdrlen,
1747 int hlimit, int tclass,
1748 struct ipv6_txoptions *opt, struct flowi6 *fl6,
1749 struct rt6_info *rt, unsigned int flags,
1752 struct inet_cork_full cork;
1753 struct inet6_cork v6_cork;
1754 struct sk_buff_head queue;
1755 int exthdrlen = (opt ? opt->opt_flen : 0);
1758 if (flags & MSG_PROBE)
1761 __skb_queue_head_init(&queue);
1763 cork.base.flags = 0;
1765 cork.base.opt = NULL;
1767 err = ip6_setup_cork(sk, &cork, &v6_cork, hlimit, tclass, opt, rt, fl6);
1769 return ERR_PTR(err);
1772 dontfrag = inet6_sk(sk)->dontfrag;
1774 err = __ip6_append_data(sk, fl6, &queue, &cork.base, &v6_cork,
1775 ¤t->task_frag, getfrag, from,
1776 length + exthdrlen, transhdrlen + exthdrlen,
1779 __ip6_flush_pending_frames(sk, &queue, &cork, &v6_cork);
1780 return ERR_PTR(err);
1783 return __ip6_make_skb(sk, &queue, &cork, &v6_cork);