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/overflow-arith.h>
32 #include <linux/string.h>
33 #include <linux/socket.h>
34 #include <linux/net.h>
35 #include <linux/netdevice.h>
36 #include <linux/if_arp.h>
37 #include <linux/in6.h>
38 #include <linux/tcp.h>
39 #include <linux/route.h>
40 #include <linux/module.h>
41 #include <linux/slab.h>
43 #include <linux/netfilter.h>
44 #include <linux/netfilter_ipv6.h>
50 #include <net/ndisc.h>
51 #include <net/protocol.h>
52 #include <net/ip6_route.h>
53 #include <net/addrconf.h>
54 #include <net/rawv6.h>
57 #include <net/checksum.h>
58 #include <linux/mroute6.h>
59 #include <net/l3mdev.h>
61 static int ip6_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
63 struct dst_entry *dst = skb_dst(skb);
64 struct net_device *dev = dst->dev;
65 struct neighbour *neigh;
66 struct in6_addr *nexthop;
69 skb->protocol = htons(ETH_P_IPV6);
72 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
73 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
75 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(sk) &&
76 ((mroute6_socket(net, skb) &&
77 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
78 ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
79 &ipv6_hdr(skb)->saddr))) {
80 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
82 /* Do not check for IFF_ALLMULTI; multicast routing
83 is not supported in any case.
86 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
87 net, sk, newskb, NULL, newskb->dev,
90 if (ipv6_hdr(skb)->hop_limit == 0) {
91 IP6_INC_STATS(net, idev,
92 IPSTATS_MIB_OUTDISCARDS);
98 IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUTMCAST, skb->len);
100 if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb)->daddr) <=
101 IPV6_ADDR_SCOPE_NODELOCAL &&
102 !(dev->flags & IFF_LOOPBACK)) {
109 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
110 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
111 if (unlikely(!neigh))
112 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
113 if (!IS_ERR(neigh)) {
114 ret = dst_neigh_output(dst, neigh, skb);
115 rcu_read_unlock_bh();
118 rcu_read_unlock_bh();
120 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
125 static int ip6_finish_output(struct net *net, 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(net, sk, skb, ip6_finish_output2);
132 return ip6_finish_output2(net, sk, skb);
135 int ip6_output(struct net *net, 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));
140 if (unlikely(idev->cnf.disable_ipv6)) {
141 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
146 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
147 net, sk, skb, NULL, dev,
149 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
153 * xmit an sk_buff (used by TCP, SCTP and DCCP)
154 * Note : socket lock is not held for SYNACK packets, but might be modified
155 * by calls to skb_set_owner_w() and ipv6_local_error(),
156 * which are using proper atomic operations or spinlocks.
158 int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
159 struct ipv6_txoptions *opt, int tclass)
161 struct net *net = sock_net(sk);
162 const struct ipv6_pinfo *np = inet6_sk(sk);
163 struct in6_addr *first_hop = &fl6->daddr;
164 struct dst_entry *dst = skb_dst(skb);
166 u8 proto = fl6->flowi6_proto;
167 int seg_len = skb->len;
172 unsigned int head_room;
174 /* First: exthdrs may take lots of space (~8K for now)
175 MAX_HEADER is not enough.
177 head_room = opt->opt_nflen + opt->opt_flen;
178 seg_len += head_room;
179 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
181 if (skb_headroom(skb) < head_room) {
182 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
184 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
185 IPSTATS_MIB_OUTDISCARDS);
191 /* skb_set_owner_w() changes sk->sk_wmem_alloc atomically,
192 * it is safe to call in our context (socket lock not held)
194 skb_set_owner_w(skb, (struct sock *)sk);
197 ipv6_push_frag_opts(skb, opt, &proto);
199 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
202 skb_push(skb, sizeof(struct ipv6hdr));
203 skb_reset_network_header(skb);
207 * Fill in the IPv6 header
210 hlimit = np->hop_limit;
212 hlimit = ip6_dst_hoplimit(dst);
214 ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
215 np->autoflowlabel, fl6));
217 hdr->payload_len = htons(seg_len);
218 hdr->nexthdr = proto;
219 hdr->hop_limit = hlimit;
221 hdr->saddr = fl6->saddr;
222 hdr->daddr = *first_hop;
224 skb->protocol = htons(ETH_P_IPV6);
225 skb->priority = sk->sk_priority;
226 skb->mark = sk->sk_mark;
229 if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
230 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
231 IPSTATS_MIB_OUT, skb->len);
232 /* hooks should never assume socket lock is held.
233 * we promote our socket to non const
235 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
236 net, (struct sock *)sk, skb, NULL, dst->dev,
241 /* ipv6_local_error() does not require socket lock,
242 * we promote our socket to non const
244 ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
246 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
250 EXPORT_SYMBOL(ip6_xmit);
252 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
254 struct ip6_ra_chain *ra;
255 struct sock *last = NULL;
257 read_lock(&ip6_ra_lock);
258 for (ra = ip6_ra_chain; ra; ra = ra->next) {
259 struct sock *sk = ra->sk;
260 if (sk && ra->sel == sel &&
261 (!sk->sk_bound_dev_if ||
262 sk->sk_bound_dev_if == skb->dev->ifindex)) {
264 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
266 rawv6_rcv(last, skb2);
273 rawv6_rcv(last, skb);
274 read_unlock(&ip6_ra_lock);
277 read_unlock(&ip6_ra_lock);
281 static int ip6_forward_proxy_check(struct sk_buff *skb)
283 struct ipv6hdr *hdr = ipv6_hdr(skb);
284 u8 nexthdr = hdr->nexthdr;
288 if (ipv6_ext_hdr(nexthdr)) {
289 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
293 offset = sizeof(struct ipv6hdr);
295 if (nexthdr == IPPROTO_ICMPV6) {
296 struct icmp6hdr *icmp6;
298 if (!pskb_may_pull(skb, (skb_network_header(skb) +
299 offset + 1 - skb->data)))
302 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
304 switch (icmp6->icmp6_type) {
305 case NDISC_ROUTER_SOLICITATION:
306 case NDISC_ROUTER_ADVERTISEMENT:
307 case NDISC_NEIGHBOUR_SOLICITATION:
308 case NDISC_NEIGHBOUR_ADVERTISEMENT:
310 /* For reaction involving unicast neighbor discovery
311 * message destined to the proxied address, pass it to
321 * The proxying router can't forward traffic sent to a link-local
322 * address, so signal the sender and discard the packet. This
323 * behavior is clarified by the MIPv6 specification.
325 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
326 dst_link_failure(skb);
333 static inline int ip6_forward_finish(struct net *net, struct sock *sk,
336 skb_sender_cpu_clear(skb);
337 return dst_output(net, sk, skb);
340 static unsigned int ip6_dst_mtu_forward(const struct dst_entry *dst)
343 struct inet6_dev *idev;
345 if (dst_metric_locked(dst, RTAX_MTU)) {
346 mtu = dst_metric_raw(dst, RTAX_MTU);
353 idev = __in6_dev_get(dst->dev);
355 mtu = idev->cnf.mtu6;
361 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
366 /* ipv6 conntrack defrag sets max_frag_size + ignore_df */
367 if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
373 if (skb_is_gso(skb) && skb_gso_network_seglen(skb) <= mtu)
379 int ip6_forward(struct sk_buff *skb)
381 struct dst_entry *dst = skb_dst(skb);
382 struct ipv6hdr *hdr = ipv6_hdr(skb);
383 struct inet6_skb_parm *opt = IP6CB(skb);
384 struct net *net = dev_net(dst->dev);
387 if (net->ipv6.devconf_all->forwarding == 0)
390 if (skb->pkt_type != PACKET_HOST)
393 if (unlikely(skb->sk))
396 if (skb_warn_if_lro(skb))
399 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
400 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
401 IPSTATS_MIB_INDISCARDS);
405 skb_forward_csum(skb);
408 * We DO NOT make any processing on
409 * RA packets, pushing them to user level AS IS
410 * without ane WARRANTY that application will be able
411 * to interpret them. The reason is that we
412 * cannot make anything clever here.
414 * We are not end-node, so that if packet contains
415 * AH/ESP, we cannot make anything.
416 * Defragmentation also would be mistake, RA packets
417 * cannot be fragmented, because there is no warranty
418 * that different fragments will go along one path. --ANK
420 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
421 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
426 * check and decrement ttl
428 if (hdr->hop_limit <= 1) {
429 /* Force OUTPUT device used as source address */
431 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
432 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
433 IPSTATS_MIB_INHDRERRORS);
439 /* XXX: idev->cnf.proxy_ndp? */
440 if (net->ipv6.devconf_all->proxy_ndp &&
441 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
442 int proxied = ip6_forward_proxy_check(skb);
444 return ip6_input(skb);
445 else if (proxied < 0) {
446 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
447 IPSTATS_MIB_INDISCARDS);
452 if (!xfrm6_route_forward(skb)) {
453 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
454 IPSTATS_MIB_INDISCARDS);
459 /* IPv6 specs say nothing about it, but it is clear that we cannot
460 send redirects to source routed frames.
461 We don't send redirects to frames decapsulated from IPsec.
463 if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
464 struct in6_addr *target = NULL;
465 struct inet_peer *peer;
469 * incoming and outgoing devices are the same
473 rt = (struct rt6_info *) dst;
474 if (rt->rt6i_flags & RTF_GATEWAY)
475 target = &rt->rt6i_gateway;
477 target = &hdr->daddr;
479 peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
481 /* Limit redirects both by destination (here)
482 and by source (inside ndisc_send_redirect)
484 if (inet_peer_xrlim_allow(peer, 1*HZ))
485 ndisc_send_redirect(skb, target);
489 int addrtype = ipv6_addr_type(&hdr->saddr);
491 /* This check is security critical. */
492 if (addrtype == IPV6_ADDR_ANY ||
493 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
495 if (addrtype & IPV6_ADDR_LINKLOCAL) {
496 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
497 ICMPV6_NOT_NEIGHBOUR, 0);
502 mtu = ip6_dst_mtu_forward(dst);
503 if (mtu < IPV6_MIN_MTU)
506 if (ip6_pkt_too_big(skb, mtu)) {
507 /* Again, force OUTPUT device used as source address */
509 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
510 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
511 IPSTATS_MIB_INTOOBIGERRORS);
512 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
513 IPSTATS_MIB_FRAGFAILS);
518 if (skb_cow(skb, dst->dev->hard_header_len)) {
519 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
520 IPSTATS_MIB_OUTDISCARDS);
526 /* Mangling hops number delayed to point after skb COW */
530 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
531 IP6_ADD_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
532 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
533 net, NULL, skb, skb->dev, dst->dev,
537 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
543 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
545 to->pkt_type = from->pkt_type;
546 to->priority = from->priority;
547 to->protocol = from->protocol;
549 skb_dst_set(to, dst_clone(skb_dst(from)));
551 to->mark = from->mark;
553 #ifdef CONFIG_NET_SCHED
554 to->tc_index = from->tc_index;
557 skb_copy_secmark(to, from);
560 int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
561 int (*output)(struct net *, struct sock *, struct sk_buff *))
563 struct sk_buff *frag;
564 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
565 struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
566 inet6_sk(skb->sk) : NULL;
567 struct ipv6hdr *tmp_hdr;
569 unsigned int mtu, hlen, left, len;
572 int ptr, offset = 0, err = 0;
573 u8 *prevhdr, nexthdr = 0;
575 hlen = ip6_find_1stfragopt(skb, &prevhdr);
578 mtu = ip6_skb_dst_mtu(skb);
580 /* We must not fragment if the socket is set to force MTU discovery
581 * or if the skb it not generated by a local socket.
583 if (unlikely(!skb->ignore_df && skb->len > mtu))
586 if (IP6CB(skb)->frag_max_size) {
587 if (IP6CB(skb)->frag_max_size > mtu)
590 /* don't send fragments larger than what we received */
591 mtu = IP6CB(skb)->frag_max_size;
592 if (mtu < IPV6_MIN_MTU)
596 if (np && np->frag_size < mtu) {
601 if (overflow_usub(mtu, hlen + sizeof(struct frag_hdr), &mtu) ||
605 frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
606 &ipv6_hdr(skb)->saddr);
608 hroom = LL_RESERVED_SPACE(rt->dst.dev);
609 if (skb_has_frag_list(skb)) {
610 int first_len = skb_pagelen(skb);
611 struct sk_buff *frag2;
613 if (first_len - hlen > mtu ||
614 ((first_len - hlen) & 7) ||
616 skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
619 skb_walk_frags(skb, frag) {
620 /* Correct geometry. */
621 if (frag->len > mtu ||
622 ((frag->len & 7) && frag->next) ||
623 skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
624 goto slow_path_clean;
626 /* Partially cloned skb? */
627 if (skb_shared(frag))
628 goto slow_path_clean;
633 frag->destructor = sock_wfree;
635 skb->truesize -= frag->truesize;
642 *prevhdr = NEXTHDR_FRAGMENT;
643 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
645 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
646 IPSTATS_MIB_FRAGFAILS);
650 frag = skb_shinfo(skb)->frag_list;
651 skb_frag_list_init(skb);
653 __skb_pull(skb, hlen);
654 fh = (struct frag_hdr *)__skb_push(skb, sizeof(struct frag_hdr));
655 __skb_push(skb, hlen);
656 skb_reset_network_header(skb);
657 memcpy(skb_network_header(skb), tmp_hdr, hlen);
659 fh->nexthdr = nexthdr;
661 fh->frag_off = htons(IP6_MF);
662 fh->identification = frag_id;
664 first_len = skb_pagelen(skb);
665 skb->data_len = first_len - skb_headlen(skb);
666 skb->len = first_len;
667 ipv6_hdr(skb)->payload_len = htons(first_len -
668 sizeof(struct ipv6hdr));
673 /* Prepare header of the next frame,
674 * before previous one went down. */
676 frag->ip_summed = CHECKSUM_NONE;
677 skb_reset_transport_header(frag);
678 fh = (struct frag_hdr *)__skb_push(frag, sizeof(struct frag_hdr));
679 __skb_push(frag, hlen);
680 skb_reset_network_header(frag);
681 memcpy(skb_network_header(frag), tmp_hdr,
683 offset += skb->len - hlen - sizeof(struct frag_hdr);
684 fh->nexthdr = nexthdr;
686 fh->frag_off = htons(offset);
688 fh->frag_off |= htons(IP6_MF);
689 fh->identification = frag_id;
690 ipv6_hdr(frag)->payload_len =
692 sizeof(struct ipv6hdr));
693 ip6_copy_metadata(frag, skb);
696 err = output(net, sk, skb);
698 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
699 IPSTATS_MIB_FRAGCREATES);
712 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
713 IPSTATS_MIB_FRAGOKS);
718 kfree_skb_list(frag);
720 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
721 IPSTATS_MIB_FRAGFAILS);
726 skb_walk_frags(skb, frag2) {
730 frag2->destructor = NULL;
731 skb->truesize += frag2->truesize;
736 if ((skb->ip_summed == CHECKSUM_PARTIAL) &&
737 skb_checksum_help(skb))
740 left = skb->len - hlen; /* Space per frame */
741 ptr = hlen; /* Where to start from */
744 * Fragment the datagram.
747 *prevhdr = NEXTHDR_FRAGMENT;
748 troom = rt->dst.dev->needed_tailroom;
751 * Keep copying data until we run out.
755 /* IF: it doesn't fit, use 'mtu' - the data space left */
758 /* IF: we are not sending up to and including the packet end
759 then align the next start on an eight byte boundary */
764 /* Allocate buffer */
765 frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
766 hroom + troom, GFP_ATOMIC);
768 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
769 IPSTATS_MIB_FRAGFAILS);
775 * Set up data on packet
778 ip6_copy_metadata(frag, skb);
779 skb_reserve(frag, hroom);
780 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
781 skb_reset_network_header(frag);
782 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
783 frag->transport_header = (frag->network_header + hlen +
784 sizeof(struct frag_hdr));
787 * Charge the memory for the fragment to any owner
791 skb_set_owner_w(frag, skb->sk);
794 * Copy the packet header into the new buffer.
796 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
799 * Build fragment header.
801 fh->nexthdr = nexthdr;
803 fh->identification = frag_id;
806 * Copy a block of the IP datagram.
808 BUG_ON(skb_copy_bits(skb, ptr, skb_transport_header(frag),
812 fh->frag_off = htons(offset);
814 fh->frag_off |= htons(IP6_MF);
815 ipv6_hdr(frag)->payload_len = htons(frag->len -
816 sizeof(struct ipv6hdr));
822 * Put this fragment into the sending queue.
824 err = output(net, sk, frag);
828 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
829 IPSTATS_MIB_FRAGCREATES);
831 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
832 IPSTATS_MIB_FRAGOKS);
837 if (skb->sk && dst_allfrag(skb_dst(skb)))
838 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
840 skb->dev = skb_dst(skb)->dev;
841 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
845 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
846 IPSTATS_MIB_FRAGFAILS);
851 static inline int ip6_rt_check(const struct rt6key *rt_key,
852 const struct in6_addr *fl_addr,
853 const struct in6_addr *addr_cache)
855 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
856 (!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
859 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
860 struct dst_entry *dst,
861 const struct flowi6 *fl6)
863 struct ipv6_pinfo *np = inet6_sk(sk);
869 if (dst->ops->family != AF_INET6) {
874 rt = (struct rt6_info *)dst;
875 /* Yes, checking route validity in not connected
876 * case is not very simple. Take into account,
877 * that we do not support routing by source, TOS,
878 * and MSG_DONTROUTE --ANK (980726)
880 * 1. ip6_rt_check(): If route was host route,
881 * check that cached destination is current.
882 * If it is network route, we still may
883 * check its validity using saved pointer
884 * to the last used address: daddr_cache.
885 * We do not want to save whole address now,
886 * (because main consumer of this service
887 * is tcp, which has not this problem),
888 * so that the last trick works only on connected
890 * 2. oif also should be the same.
892 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
893 #ifdef CONFIG_IPV6_SUBTREES
894 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
896 (!(fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF) &&
897 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex))) {
906 static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
907 struct dst_entry **dst, struct flowi6 *fl6)
909 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
915 /* The correct way to handle this would be to do
916 * ip6_route_get_saddr, and then ip6_route_output; however,
917 * the route-specific preferred source forces the
918 * ip6_route_output call _before_ ip6_route_get_saddr.
920 * In source specific routing (no src=any default route),
921 * ip6_route_output will fail given src=any saddr, though, so
922 * that's why we try it again later.
924 if (ipv6_addr_any(&fl6->saddr) && (!*dst || !(*dst)->error)) {
926 bool had_dst = *dst != NULL;
929 *dst = ip6_route_output(net, sk, fl6);
930 rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
931 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
932 sk ? inet6_sk(sk)->srcprefs : 0,
935 goto out_err_release;
937 /* If we had an erroneous initial result, pretend it
938 * never existed and let the SA-enabled version take
941 if (!had_dst && (*dst)->error) {
948 *dst = ip6_route_output(net, sk, fl6);
952 goto out_err_release;
954 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
956 * Here if the dst entry we've looked up
957 * has a neighbour entry that is in the INCOMPLETE
958 * state and the src address from the flow is
959 * marked as OPTIMISTIC, we release the found
960 * dst entry and replace it instead with the
961 * dst entry of the nexthop router
963 rt = (struct rt6_info *) *dst;
965 n = __ipv6_neigh_lookup_noref(rt->dst.dev,
966 rt6_nexthop(rt, &fl6->daddr));
967 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
968 rcu_read_unlock_bh();
971 struct inet6_ifaddr *ifp;
972 struct flowi6 fl_gw6;
975 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
978 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
984 * We need to get the dst entry for the
985 * default router instead
988 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
989 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
990 *dst = ip6_route_output(net, sk, &fl_gw6);
993 goto out_err_release;
1001 if (err == -ENETUNREACH)
1002 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1009 * ip6_dst_lookup - perform route lookup on flow
1010 * @sk: socket which provides route info
1011 * @dst: pointer to dst_entry * for result
1012 * @fl6: flow to lookup
1014 * This function performs a route lookup on the given flow.
1016 * It returns zero on success, or a standard errno code on error.
1018 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1022 return ip6_dst_lookup_tail(net, sk, dst, fl6);
1024 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1027 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1028 * @sk: socket which provides route info
1029 * @fl6: flow to lookup
1030 * @final_dst: final destination address for ipsec lookup
1032 * This function performs a route lookup on the given flow.
1034 * It returns a valid dst pointer on success, or a pointer encoded
1037 struct dst_entry *ip6_dst_lookup_flow(const struct sock *sk, struct flowi6 *fl6,
1038 const struct in6_addr *final_dst)
1040 struct dst_entry *dst = NULL;
1043 err = ip6_dst_lookup_tail(sock_net(sk), sk, &dst, fl6);
1045 return ERR_PTR(err);
1047 fl6->daddr = *final_dst;
1048 if (!fl6->flowi6_oif)
1049 fl6->flowi6_oif = l3mdev_fib_oif(dst->dev);
1051 return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1053 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1056 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1057 * @sk: socket which provides the dst cache and route info
1058 * @fl6: flow to lookup
1059 * @final_dst: final destination address for ipsec lookup
1061 * This function performs a route lookup on the given flow with the
1062 * possibility of using the cached route in the socket if it is valid.
1063 * It will take the socket dst lock when operating on the dst cache.
1064 * As a result, this function can only be used in process context.
1066 * It returns a valid dst pointer on success, or a pointer encoded
1069 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1070 const struct in6_addr *final_dst)
1072 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1075 dst = ip6_sk_dst_check(sk, dst, fl6);
1077 err = ip6_dst_lookup_tail(sock_net(sk), sk, &dst, fl6);
1079 return ERR_PTR(err);
1081 fl6->daddr = *final_dst;
1083 return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1085 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1087 static inline int ip6_ufo_append_data(struct sock *sk,
1088 struct sk_buff_head *queue,
1089 int getfrag(void *from, char *to, int offset, int len,
1090 int odd, struct sk_buff *skb),
1091 void *from, int length, int hh_len, int fragheaderlen,
1092 int transhdrlen, int mtu, unsigned int flags,
1093 const struct flowi6 *fl6)
1096 struct sk_buff *skb;
1099 /* There is support for UDP large send offload by network
1100 * device, so create one single skb packet containing complete
1103 skb = skb_peek_tail(queue);
1105 skb = sock_alloc_send_skb(sk,
1106 hh_len + fragheaderlen + transhdrlen + 20,
1107 (flags & MSG_DONTWAIT), &err);
1111 /* reserve space for Hardware header */
1112 skb_reserve(skb, hh_len);
1114 /* create space for UDP/IP header */
1115 skb_put(skb, fragheaderlen + transhdrlen);
1117 /* initialize network header pointer */
1118 skb_reset_network_header(skb);
1120 /* initialize protocol header pointer */
1121 skb->transport_header = skb->network_header + fragheaderlen;
1123 skb->protocol = htons(ETH_P_IPV6);
1126 __skb_queue_tail(queue, skb);
1127 } else if (skb_is_gso(skb)) {
1131 skb->ip_summed = CHECKSUM_PARTIAL;
1132 /* Specify the length of each IPv6 datagram fragment.
1133 * It has to be a multiple of 8.
1135 skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1136 sizeof(struct frag_hdr)) & ~7;
1137 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1138 skb_shinfo(skb)->ip6_frag_id = ipv6_select_ident(sock_net(sk),
1143 return skb_append_datato_frags(sk, skb, getfrag, from,
1144 (length - transhdrlen));
1147 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1150 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1153 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1156 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1159 static void ip6_append_data_mtu(unsigned int *mtu,
1161 unsigned int fragheaderlen,
1162 struct sk_buff *skb,
1163 struct rt6_info *rt,
1164 unsigned int orig_mtu)
1166 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1168 /* first fragment, reserve header_len */
1169 *mtu = orig_mtu - rt->dst.header_len;
1173 * this fragment is not first, the headers
1174 * space is regarded as data space.
1178 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1179 + fragheaderlen - sizeof(struct frag_hdr);
1183 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1184 struct inet6_cork *v6_cork,
1185 int hlimit, int tclass, struct ipv6_txoptions *opt,
1186 struct rt6_info *rt, struct flowi6 *fl6)
1188 struct ipv6_pinfo *np = inet6_sk(sk);
1195 if (WARN_ON(v6_cork->opt))
1198 v6_cork->opt = kzalloc(opt->tot_len, sk->sk_allocation);
1199 if (unlikely(!v6_cork->opt))
1202 v6_cork->opt->tot_len = opt->tot_len;
1203 v6_cork->opt->opt_flen = opt->opt_flen;
1204 v6_cork->opt->opt_nflen = opt->opt_nflen;
1206 v6_cork->opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1208 if (opt->dst0opt && !v6_cork->opt->dst0opt)
1211 v6_cork->opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1213 if (opt->dst1opt && !v6_cork->opt->dst1opt)
1216 v6_cork->opt->hopopt = ip6_opt_dup(opt->hopopt,
1218 if (opt->hopopt && !v6_cork->opt->hopopt)
1221 v6_cork->opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1223 if (opt->srcrt && !v6_cork->opt->srcrt)
1226 /* need source address above miyazawa*/
1229 cork->base.dst = &rt->dst;
1230 cork->fl.u.ip6 = *fl6;
1231 v6_cork->hop_limit = hlimit;
1232 v6_cork->tclass = tclass;
1233 if (rt->dst.flags & DST_XFRM_TUNNEL)
1234 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1235 rt->dst.dev->mtu : dst_mtu(&rt->dst);
1237 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1238 rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1239 if (np->frag_size < mtu) {
1241 mtu = np->frag_size;
1243 cork->base.fragsize = mtu;
1244 if (dst_allfrag(rt->dst.path))
1245 cork->base.flags |= IPCORK_ALLFRAG;
1246 cork->base.length = 0;
1251 static int __ip6_append_data(struct sock *sk,
1253 struct sk_buff_head *queue,
1254 struct inet_cork *cork,
1255 struct inet6_cork *v6_cork,
1256 struct page_frag *pfrag,
1257 int getfrag(void *from, char *to, int offset,
1258 int len, int odd, struct sk_buff *skb),
1259 void *from, int length, int transhdrlen,
1260 unsigned int flags, int dontfrag)
1262 struct sk_buff *skb, *skb_prev = NULL;
1263 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu;
1265 int dst_exthdrlen = 0;
1272 struct rt6_info *rt = (struct rt6_info *)cork->dst;
1273 struct ipv6_txoptions *opt = v6_cork->opt;
1274 int csummode = CHECKSUM_NONE;
1276 skb = skb_peek_tail(queue);
1278 exthdrlen = opt ? opt->opt_flen : 0;
1279 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1282 mtu = cork->fragsize;
1285 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1287 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1288 (opt ? opt->opt_nflen : 0);
1289 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1290 sizeof(struct frag_hdr);
1292 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1293 unsigned int maxnonfragsize, headersize;
1295 headersize = sizeof(struct ipv6hdr) +
1296 (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1297 (dst_allfrag(&rt->dst) ?
1298 sizeof(struct frag_hdr) : 0) +
1299 rt->rt6i_nfheader_len;
1301 if (ip6_sk_ignore_df(sk))
1302 maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1304 maxnonfragsize = mtu;
1306 /* dontfrag active */
1307 if ((cork->length + length > mtu - headersize) && dontfrag &&
1308 (sk->sk_protocol == IPPROTO_UDP ||
1309 sk->sk_protocol == IPPROTO_RAW)) {
1310 ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1311 sizeof(struct ipv6hdr));
1315 if (cork->length + length > maxnonfragsize - headersize) {
1317 ipv6_local_error(sk, EMSGSIZE, fl6,
1319 sizeof(struct ipv6hdr));
1324 if (sk->sk_type == SOCK_DGRAM || sk->sk_type == SOCK_RAW) {
1325 sock_tx_timestamp(sk, &tx_flags);
1326 if (tx_flags & SKBTX_ANY_SW_TSTAMP &&
1327 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1328 tskey = sk->sk_tskey++;
1331 /* If this is the first and only packet and device
1332 * supports checksum offloading, let's use it.
1333 * Use transhdrlen, same as IPv4, because partial
1334 * sums only work when transhdrlen is set.
1336 if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1337 length + fragheaderlen < mtu &&
1338 rt->dst.dev->features & NETIF_F_V6_CSUM &&
1340 csummode = CHECKSUM_PARTIAL;
1342 * Let's try using as much space as possible.
1343 * Use MTU if total length of the message fits into the MTU.
1344 * Otherwise, we need to reserve fragment header and
1345 * fragment alignment (= 8-15 octects, in total).
1347 * Note that we may need to "move" the data from the tail of
1348 * of the buffer to the new fragment when we split
1351 * FIXME: It may be fragmented into multiple chunks
1352 * at once if non-fragmentable extension headers
1357 cork->length += length;
1358 if (((length > mtu) ||
1359 (skb && skb_is_gso(skb))) &&
1360 (sk->sk_protocol == IPPROTO_UDP) &&
1361 (rt->dst.dev->features & NETIF_F_UFO) &&
1362 (sk->sk_type == SOCK_DGRAM)) {
1363 err = ip6_ufo_append_data(sk, queue, getfrag, from, length,
1364 hh_len, fragheaderlen,
1365 transhdrlen, mtu, flags, fl6);
1374 while (length > 0) {
1375 /* Check if the remaining data fits into current packet. */
1376 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1378 copy = maxfraglen - skb->len;
1382 unsigned int datalen;
1383 unsigned int fraglen;
1384 unsigned int fraggap;
1385 unsigned int alloclen;
1387 /* There's no room in the current skb */
1389 fraggap = skb->len - maxfraglen;
1392 /* update mtu and maxfraglen if necessary */
1393 if (!skb || !skb_prev)
1394 ip6_append_data_mtu(&mtu, &maxfraglen,
1395 fragheaderlen, skb, rt,
1401 * If remaining data exceeds the mtu,
1402 * we know we need more fragment(s).
1404 datalen = length + fraggap;
1406 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1407 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1408 if ((flags & MSG_MORE) &&
1409 !(rt->dst.dev->features&NETIF_F_SG))
1412 alloclen = datalen + fragheaderlen;
1414 alloclen += dst_exthdrlen;
1416 if (datalen != length + fraggap) {
1418 * this is not the last fragment, the trailer
1419 * space is regarded as data space.
1421 datalen += rt->dst.trailer_len;
1424 alloclen += rt->dst.trailer_len;
1425 fraglen = datalen + fragheaderlen;
1428 * We just reserve space for fragment header.
1429 * Note: this may be overallocation if the message
1430 * (without MSG_MORE) fits into the MTU.
1432 alloclen += sizeof(struct frag_hdr);
1435 skb = sock_alloc_send_skb(sk,
1437 (flags & MSG_DONTWAIT), &err);
1440 if (atomic_read(&sk->sk_wmem_alloc) <=
1442 skb = sock_wmalloc(sk,
1443 alloclen + hh_len, 1,
1451 * Fill in the control structures
1453 skb->protocol = htons(ETH_P_IPV6);
1454 skb->ip_summed = csummode;
1456 /* reserve for fragmentation and ipsec header */
1457 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1460 /* Only the initial fragment is time stamped */
1461 skb_shinfo(skb)->tx_flags = tx_flags;
1463 skb_shinfo(skb)->tskey = tskey;
1467 * Find where to start putting bytes
1469 data = skb_put(skb, fraglen);
1470 skb_set_network_header(skb, exthdrlen);
1471 data += fragheaderlen;
1472 skb->transport_header = (skb->network_header +
1475 skb->csum = skb_copy_and_csum_bits(
1476 skb_prev, maxfraglen,
1477 data + transhdrlen, fraggap, 0);
1478 skb_prev->csum = csum_sub(skb_prev->csum,
1481 pskb_trim_unique(skb_prev, maxfraglen);
1483 copy = datalen - transhdrlen - fraggap;
1489 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1496 length -= datalen - fraggap;
1502 * Put the packet on the pending queue
1504 __skb_queue_tail(queue, skb);
1511 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1515 if (getfrag(from, skb_put(skb, copy),
1516 offset, copy, off, skb) < 0) {
1517 __skb_trim(skb, off);
1522 int i = skb_shinfo(skb)->nr_frags;
1525 if (!sk_page_frag_refill(sk, pfrag))
1528 if (!skb_can_coalesce(skb, i, pfrag->page,
1531 if (i == MAX_SKB_FRAGS)
1534 __skb_fill_page_desc(skb, i, pfrag->page,
1536 skb_shinfo(skb)->nr_frags = ++i;
1537 get_page(pfrag->page);
1539 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1541 page_address(pfrag->page) + pfrag->offset,
1542 offset, copy, skb->len, skb) < 0)
1545 pfrag->offset += copy;
1546 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1548 skb->data_len += copy;
1549 skb->truesize += copy;
1550 atomic_add(copy, &sk->sk_wmem_alloc);
1561 cork->length -= length;
1562 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1566 int ip6_append_data(struct sock *sk,
1567 int getfrag(void *from, char *to, int offset, int len,
1568 int odd, struct sk_buff *skb),
1569 void *from, int length, int transhdrlen, int hlimit,
1570 int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1571 struct rt6_info *rt, unsigned int flags, int dontfrag)
1573 struct inet_sock *inet = inet_sk(sk);
1574 struct ipv6_pinfo *np = inet6_sk(sk);
1578 if (flags&MSG_PROBE)
1580 if (skb_queue_empty(&sk->sk_write_queue)) {
1584 err = ip6_setup_cork(sk, &inet->cork, &np->cork, hlimit,
1585 tclass, opt, rt, fl6);
1589 exthdrlen = (opt ? opt->opt_flen : 0);
1590 length += exthdrlen;
1591 transhdrlen += exthdrlen;
1593 fl6 = &inet->cork.fl.u.ip6;
1597 return __ip6_append_data(sk, fl6, &sk->sk_write_queue, &inet->cork.base,
1598 &np->cork, sk_page_frag(sk), getfrag,
1599 from, length, transhdrlen, flags, dontfrag);
1601 EXPORT_SYMBOL_GPL(ip6_append_data);
1603 static void ip6_cork_release(struct inet_cork_full *cork,
1604 struct inet6_cork *v6_cork)
1607 kfree(v6_cork->opt->dst0opt);
1608 kfree(v6_cork->opt->dst1opt);
1609 kfree(v6_cork->opt->hopopt);
1610 kfree(v6_cork->opt->srcrt);
1611 kfree(v6_cork->opt);
1612 v6_cork->opt = NULL;
1615 if (cork->base.dst) {
1616 dst_release(cork->base.dst);
1617 cork->base.dst = NULL;
1618 cork->base.flags &= ~IPCORK_ALLFRAG;
1620 memset(&cork->fl, 0, sizeof(cork->fl));
1623 struct sk_buff *__ip6_make_skb(struct sock *sk,
1624 struct sk_buff_head *queue,
1625 struct inet_cork_full *cork,
1626 struct inet6_cork *v6_cork)
1628 struct sk_buff *skb, *tmp_skb;
1629 struct sk_buff **tail_skb;
1630 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1631 struct ipv6_pinfo *np = inet6_sk(sk);
1632 struct net *net = sock_net(sk);
1633 struct ipv6hdr *hdr;
1634 struct ipv6_txoptions *opt = v6_cork->opt;
1635 struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1636 struct flowi6 *fl6 = &cork->fl.u.ip6;
1637 unsigned char proto = fl6->flowi6_proto;
1639 skb = __skb_dequeue(queue);
1642 tail_skb = &(skb_shinfo(skb)->frag_list);
1644 /* move skb->data to ip header from ext header */
1645 if (skb->data < skb_network_header(skb))
1646 __skb_pull(skb, skb_network_offset(skb));
1647 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1648 __skb_pull(tmp_skb, skb_network_header_len(skb));
1649 *tail_skb = tmp_skb;
1650 tail_skb = &(tmp_skb->next);
1651 skb->len += tmp_skb->len;
1652 skb->data_len += tmp_skb->len;
1653 skb->truesize += tmp_skb->truesize;
1654 tmp_skb->destructor = NULL;
1658 /* Allow local fragmentation. */
1659 skb->ignore_df = ip6_sk_ignore_df(sk);
1661 *final_dst = fl6->daddr;
1662 __skb_pull(skb, skb_network_header_len(skb));
1663 if (opt && opt->opt_flen)
1664 ipv6_push_frag_opts(skb, opt, &proto);
1665 if (opt && opt->opt_nflen)
1666 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1668 skb_push(skb, sizeof(struct ipv6hdr));
1669 skb_reset_network_header(skb);
1670 hdr = ipv6_hdr(skb);
1672 ip6_flow_hdr(hdr, v6_cork->tclass,
1673 ip6_make_flowlabel(net, skb, fl6->flowlabel,
1674 np->autoflowlabel, fl6));
1675 hdr->hop_limit = v6_cork->hop_limit;
1676 hdr->nexthdr = proto;
1677 hdr->saddr = fl6->saddr;
1678 hdr->daddr = *final_dst;
1680 skb->priority = sk->sk_priority;
1681 skb->mark = sk->sk_mark;
1683 skb_dst_set(skb, dst_clone(&rt->dst));
1684 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1685 if (proto == IPPROTO_ICMPV6) {
1686 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1688 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1689 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1692 ip6_cork_release(cork, v6_cork);
1697 int ip6_send_skb(struct sk_buff *skb)
1699 struct net *net = sock_net(skb->sk);
1700 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1703 err = ip6_local_out(net, skb->sk, skb);
1706 err = net_xmit_errno(err);
1708 IP6_INC_STATS(net, rt->rt6i_idev,
1709 IPSTATS_MIB_OUTDISCARDS);
1715 int ip6_push_pending_frames(struct sock *sk)
1717 struct sk_buff *skb;
1719 skb = ip6_finish_skb(sk);
1723 return ip6_send_skb(skb);
1725 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1727 static void __ip6_flush_pending_frames(struct sock *sk,
1728 struct sk_buff_head *queue,
1729 struct inet_cork_full *cork,
1730 struct inet6_cork *v6_cork)
1732 struct sk_buff *skb;
1734 while ((skb = __skb_dequeue_tail(queue)) != NULL) {
1736 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1737 IPSTATS_MIB_OUTDISCARDS);
1741 ip6_cork_release(cork, v6_cork);
1744 void ip6_flush_pending_frames(struct sock *sk)
1746 __ip6_flush_pending_frames(sk, &sk->sk_write_queue,
1747 &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
1749 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1751 struct sk_buff *ip6_make_skb(struct sock *sk,
1752 int getfrag(void *from, char *to, int offset,
1753 int len, int odd, struct sk_buff *skb),
1754 void *from, int length, int transhdrlen,
1755 int hlimit, int tclass,
1756 struct ipv6_txoptions *opt, struct flowi6 *fl6,
1757 struct rt6_info *rt, unsigned int flags,
1760 struct inet_cork_full cork;
1761 struct inet6_cork v6_cork;
1762 struct sk_buff_head queue;
1763 int exthdrlen = (opt ? opt->opt_flen : 0);
1766 if (flags & MSG_PROBE)
1769 __skb_queue_head_init(&queue);
1771 cork.base.flags = 0;
1773 cork.base.opt = NULL;
1775 err = ip6_setup_cork(sk, &cork, &v6_cork, hlimit, tclass, opt, rt, fl6);
1777 return ERR_PTR(err);
1780 dontfrag = inet6_sk(sk)->dontfrag;
1782 err = __ip6_append_data(sk, fl6, &queue, &cork.base, &v6_cork,
1783 ¤t->task_frag, getfrag, from,
1784 length + exthdrlen, transhdrlen + exthdrlen,
1787 __ip6_flush_pending_frames(sk, &queue, &cork, &v6_cork);
1788 return ERR_PTR(err);
1791 return __ip6_make_skb(sk, &queue, &cork, &v6_cork);