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 int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *));
61 int __ip6_local_out(struct sk_buff *skb)
65 len = skb->len - sizeof(struct ipv6hdr);
66 if (len > IPV6_MAXPLEN)
68 ipv6_hdr(skb)->payload_len = htons(len);
70 return nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, skb, NULL,
71 skb_dst(skb)->dev, dst_output);
74 int ip6_local_out(struct sk_buff *skb)
78 err = __ip6_local_out(skb);
80 err = dst_output(skb);
84 EXPORT_SYMBOL_GPL(ip6_local_out);
86 /* dev_loopback_xmit for use with netfilter. */
87 static int ip6_dev_loopback_xmit(struct sk_buff *newskb)
89 skb_reset_mac_header(newskb);
90 __skb_pull(newskb, skb_network_offset(newskb));
91 newskb->pkt_type = PACKET_LOOPBACK;
92 newskb->ip_summed = CHECKSUM_UNNECESSARY;
93 WARN_ON(!skb_dst(newskb));
99 static int ip6_finish_output2(struct sk_buff *skb)
101 struct dst_entry *dst = skb_dst(skb);
102 struct net_device *dev = dst->dev;
103 struct neighbour *neigh;
105 skb->protocol = htons(ETH_P_IPV6);
108 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
109 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
111 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(skb->sk) &&
112 ((mroute6_socket(dev_net(dev), skb) &&
113 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
114 ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
115 &ipv6_hdr(skb)->saddr))) {
116 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
118 /* Do not check for IFF_ALLMULTI; multicast routing
119 is not supported in any case.
122 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
123 newskb, NULL, newskb->dev,
124 ip6_dev_loopback_xmit);
126 if (ipv6_hdr(skb)->hop_limit == 0) {
127 IP6_INC_STATS(dev_net(dev), idev,
128 IPSTATS_MIB_OUTDISCARDS);
134 IP6_UPD_PO_STATS(dev_net(dev), idev, IPSTATS_MIB_OUTMCAST,
139 neigh = dst_get_neighbour(dst);
141 int res = neigh_output(neigh, skb);
147 IP6_INC_STATS(dev_net(dst->dev),
148 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
153 static int ip6_finish_output(struct sk_buff *skb)
155 if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
156 dst_allfrag(skb_dst(skb)))
157 return ip6_fragment(skb, ip6_finish_output2);
159 return ip6_finish_output2(skb);
162 int ip6_output(struct sk_buff *skb)
164 struct net_device *dev = skb_dst(skb)->dev;
165 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
166 if (unlikely(idev->cnf.disable_ipv6)) {
167 IP6_INC_STATS(dev_net(dev), idev,
168 IPSTATS_MIB_OUTDISCARDS);
173 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, skb, NULL, dev,
175 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
179 * xmit an sk_buff (used by TCP, SCTP and DCCP)
182 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
183 struct ipv6_txoptions *opt, int tclass)
185 struct net *net = sock_net(sk);
186 struct ipv6_pinfo *np = inet6_sk(sk);
187 struct in6_addr *first_hop = &fl6->daddr;
188 struct dst_entry *dst = skb_dst(skb);
190 u8 proto = fl6->flowi6_proto;
191 int seg_len = skb->len;
196 unsigned int head_room;
198 /* First: exthdrs may take lots of space (~8K for now)
199 MAX_HEADER is not enough.
201 head_room = opt->opt_nflen + opt->opt_flen;
202 seg_len += head_room;
203 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
205 if (skb_headroom(skb) < head_room) {
206 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
208 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
209 IPSTATS_MIB_OUTDISCARDS);
215 skb_set_owner_w(skb, sk);
218 ipv6_push_frag_opts(skb, opt, &proto);
220 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
223 skb_push(skb, sizeof(struct ipv6hdr));
224 skb_reset_network_header(skb);
228 * Fill in the IPv6 header
231 hlimit = np->hop_limit;
233 hlimit = ip6_dst_hoplimit(dst);
235 *(__be32 *)hdr = htonl(0x60000000 | (tclass << 20)) | fl6->flowlabel;
237 hdr->payload_len = htons(seg_len);
238 hdr->nexthdr = proto;
239 hdr->hop_limit = hlimit;
241 ipv6_addr_copy(&hdr->saddr, &fl6->saddr);
242 ipv6_addr_copy(&hdr->daddr, first_hop);
244 skb->priority = sk->sk_priority;
245 skb->mark = sk->sk_mark;
248 if ((skb->len <= mtu) || skb->local_df || skb_is_gso(skb)) {
249 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
250 IPSTATS_MIB_OUT, skb->len);
251 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, skb, NULL,
252 dst->dev, dst_output);
256 printk(KERN_DEBUG "IPv6: sending pkt_too_big to self\n");
258 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
259 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
264 EXPORT_SYMBOL(ip6_xmit);
267 * To avoid extra problems ND packets are send through this
268 * routine. It's code duplication but I really want to avoid
269 * extra checks since ipv6_build_header is used by TCP (which
270 * is for us performance critical)
273 int ip6_nd_hdr(struct sock *sk, struct sk_buff *skb, struct net_device *dev,
274 const struct in6_addr *saddr, const struct in6_addr *daddr,
277 struct ipv6_pinfo *np = inet6_sk(sk);
280 skb->protocol = htons(ETH_P_IPV6);
283 skb_reset_network_header(skb);
284 skb_put(skb, sizeof(struct ipv6hdr));
287 *(__be32*)hdr = htonl(0x60000000);
289 hdr->payload_len = htons(len);
290 hdr->nexthdr = proto;
291 hdr->hop_limit = np->hop_limit;
293 ipv6_addr_copy(&hdr->saddr, saddr);
294 ipv6_addr_copy(&hdr->daddr, daddr);
299 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
301 struct ip6_ra_chain *ra;
302 struct sock *last = NULL;
304 read_lock(&ip6_ra_lock);
305 for (ra = ip6_ra_chain; ra; ra = ra->next) {
306 struct sock *sk = ra->sk;
307 if (sk && ra->sel == sel &&
308 (!sk->sk_bound_dev_if ||
309 sk->sk_bound_dev_if == skb->dev->ifindex)) {
311 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
313 rawv6_rcv(last, skb2);
320 rawv6_rcv(last, skb);
321 read_unlock(&ip6_ra_lock);
324 read_unlock(&ip6_ra_lock);
328 static int ip6_forward_proxy_check(struct sk_buff *skb)
330 struct ipv6hdr *hdr = ipv6_hdr(skb);
331 u8 nexthdr = hdr->nexthdr;
334 if (ipv6_ext_hdr(nexthdr)) {
335 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr);
339 offset = sizeof(struct ipv6hdr);
341 if (nexthdr == IPPROTO_ICMPV6) {
342 struct icmp6hdr *icmp6;
344 if (!pskb_may_pull(skb, (skb_network_header(skb) +
345 offset + 1 - skb->data)))
348 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
350 switch (icmp6->icmp6_type) {
351 case NDISC_ROUTER_SOLICITATION:
352 case NDISC_ROUTER_ADVERTISEMENT:
353 case NDISC_NEIGHBOUR_SOLICITATION:
354 case NDISC_NEIGHBOUR_ADVERTISEMENT:
356 /* For reaction involving unicast neighbor discovery
357 * message destined to the proxied address, pass it to
367 * The proxying router can't forward traffic sent to a link-local
368 * address, so signal the sender and discard the packet. This
369 * behavior is clarified by the MIPv6 specification.
371 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
372 dst_link_failure(skb);
379 static inline int ip6_forward_finish(struct sk_buff *skb)
381 return dst_output(skb);
384 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
386 if (skb->len <= mtu || skb->local_df)
389 if (skb_is_gso(skb) && skb_gso_network_seglen(skb) <= mtu)
395 int ip6_forward(struct sk_buff *skb)
397 struct dst_entry *dst = skb_dst(skb);
398 struct ipv6hdr *hdr = ipv6_hdr(skb);
399 struct inet6_skb_parm *opt = IP6CB(skb);
400 struct net *net = dev_net(dst->dev);
404 if (net->ipv6.devconf_all->forwarding == 0)
407 if (skb_warn_if_lro(skb))
410 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
411 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
415 if (skb->pkt_type != PACKET_HOST)
418 skb_forward_csum(skb);
421 * We DO NOT make any processing on
422 * RA packets, pushing them to user level AS IS
423 * without ane WARRANTY that application will be able
424 * to interpret them. The reason is that we
425 * cannot make anything clever here.
427 * We are not end-node, so that if packet contains
428 * AH/ESP, we cannot make anything.
429 * Defragmentation also would be mistake, RA packets
430 * cannot be fragmented, because there is no warranty
431 * that different fragments will go along one path. --ANK
434 u8 *ptr = skb_network_header(skb) + opt->ra;
435 if (ip6_call_ra_chain(skb, (ptr[2]<<8) + ptr[3]))
440 * check and decrement ttl
442 if (hdr->hop_limit <= 1) {
443 /* Force OUTPUT device used as source address */
445 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
446 IP6_INC_STATS_BH(net,
447 ip6_dst_idev(dst), IPSTATS_MIB_INHDRERRORS);
453 /* XXX: idev->cnf.proxy_ndp? */
454 if (net->ipv6.devconf_all->proxy_ndp &&
455 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
456 int proxied = ip6_forward_proxy_check(skb);
458 return ip6_input(skb);
459 else if (proxied < 0) {
460 IP6_INC_STATS(net, ip6_dst_idev(dst),
461 IPSTATS_MIB_INDISCARDS);
466 if (!xfrm6_route_forward(skb)) {
467 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
472 /* IPv6 specs say nothing about it, but it is clear that we cannot
473 send redirects to source routed frames.
474 We don't send redirects to frames decapsulated from IPsec.
476 n = dst_get_neighbour(dst);
477 if (skb->dev == dst->dev && n && opt->srcrt == 0 && !skb_sec_path(skb)) {
478 struct in6_addr *target = NULL;
482 * incoming and outgoing devices are the same
486 rt = (struct rt6_info *) dst;
487 if ((rt->rt6i_flags & RTF_GATEWAY))
488 target = (struct in6_addr*)&n->primary_key;
490 target = &hdr->daddr;
493 rt6_bind_peer(rt, 1);
495 /* Limit redirects both by destination (here)
496 and by source (inside ndisc_send_redirect)
498 if (inet_peer_xrlim_allow(rt->rt6i_peer, 1*HZ))
499 ndisc_send_redirect(skb, n, target);
501 int addrtype = ipv6_addr_type(&hdr->saddr);
503 /* This check is security critical. */
504 if (addrtype == IPV6_ADDR_ANY ||
505 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
507 if (addrtype & IPV6_ADDR_LINKLOCAL) {
508 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
509 ICMPV6_NOT_NEIGHBOUR, 0);
515 if (mtu < IPV6_MIN_MTU)
518 if (ip6_pkt_too_big(skb, mtu)) {
519 /* Again, force OUTPUT device used as source address */
521 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
522 IP6_INC_STATS_BH(net,
523 ip6_dst_idev(dst), IPSTATS_MIB_INTOOBIGERRORS);
524 IP6_INC_STATS_BH(net,
525 ip6_dst_idev(dst), IPSTATS_MIB_FRAGFAILS);
530 if (skb_cow(skb, dst->dev->hard_header_len)) {
531 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTDISCARDS);
537 /* Mangling hops number delayed to point after skb COW */
541 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
542 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD, skb, skb->dev, dst->dev,
546 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
552 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
554 to->pkt_type = from->pkt_type;
555 to->priority = from->priority;
556 to->protocol = from->protocol;
558 skb_dst_set(to, dst_clone(skb_dst(from)));
560 to->mark = from->mark;
562 #ifdef CONFIG_NET_SCHED
563 to->tc_index = from->tc_index;
566 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
567 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
568 to->nf_trace = from->nf_trace;
570 skb_copy_secmark(to, from);
573 int ip6_find_1stfragopt(struct sk_buff *skb, u8 **nexthdr)
575 u16 offset = sizeof(struct ipv6hdr);
576 struct ipv6_opt_hdr *exthdr =
577 (struct ipv6_opt_hdr *)(ipv6_hdr(skb) + 1);
578 unsigned int packet_len = skb->tail - skb->network_header;
580 *nexthdr = &ipv6_hdr(skb)->nexthdr;
582 while (offset + 1 <= packet_len) {
588 case NEXTHDR_ROUTING:
592 #if defined(CONFIG_IPV6_MIP6) || defined(CONFIG_IPV6_MIP6_MODULE)
593 if (ipv6_find_tlv(skb, offset, IPV6_TLV_HAO) >= 0)
603 offset += ipv6_optlen(exthdr);
604 *nexthdr = &exthdr->nexthdr;
605 exthdr = (struct ipv6_opt_hdr *)(skb_network_header(skb) +
612 void ipv6_select_ident(struct frag_hdr *fhdr, struct rt6_info *rt)
614 static atomic_t ipv6_fragmentation_id;
617 if (rt && !(rt->dst.flags & DST_NOPEER)) {
618 struct inet_peer *peer;
621 rt6_bind_peer(rt, 1);
622 peer = rt->rt6i_peer;
624 fhdr->identification = htonl(inet_getid(peer, 0));
628 ident = atomic_inc_return(&ipv6_fragmentation_id);
629 fhdr->identification = htonl(ident);
632 int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
634 struct sk_buff *frag;
635 struct rt6_info *rt = (struct rt6_info*)skb_dst(skb);
636 struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
637 struct ipv6hdr *tmp_hdr;
639 unsigned int mtu, hlen, left, len;
641 int ptr, offset = 0, err=0;
642 u8 *prevhdr, nexthdr = 0;
643 struct net *net = dev_net(skb_dst(skb)->dev);
645 hlen = ip6_find_1stfragopt(skb, &prevhdr);
648 mtu = ip6_skb_dst_mtu(skb);
650 /* We must not fragment if the socket is set to force MTU discovery
651 * or if the skb it not generated by a local socket.
653 if (!skb->local_df && skb->len > mtu) {
654 skb->dev = skb_dst(skb)->dev;
655 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
656 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
657 IPSTATS_MIB_FRAGFAILS);
662 if (np && np->frag_size < mtu) {
666 mtu -= hlen + sizeof(struct frag_hdr);
668 if (skb_has_frag_list(skb)) {
669 int first_len = skb_pagelen(skb);
670 struct sk_buff *frag2;
672 if (first_len - hlen > mtu ||
673 ((first_len - hlen) & 7) ||
677 skb_walk_frags(skb, frag) {
678 /* Correct geometry. */
679 if (frag->len > mtu ||
680 ((frag->len & 7) && frag->next) ||
681 skb_headroom(frag) < hlen)
682 goto slow_path_clean;
684 /* Partially cloned skb? */
685 if (skb_shared(frag))
686 goto slow_path_clean;
691 frag->destructor = sock_wfree;
693 skb->truesize -= frag->truesize;
698 frag = skb_shinfo(skb)->frag_list;
699 skb_frag_list_init(skb);
702 *prevhdr = NEXTHDR_FRAGMENT;
703 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
705 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
706 IPSTATS_MIB_FRAGFAILS);
710 __skb_pull(skb, hlen);
711 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
712 __skb_push(skb, hlen);
713 skb_reset_network_header(skb);
714 memcpy(skb_network_header(skb), tmp_hdr, hlen);
716 ipv6_select_ident(fh, rt);
717 fh->nexthdr = nexthdr;
719 fh->frag_off = htons(IP6_MF);
720 frag_id = fh->identification;
722 first_len = skb_pagelen(skb);
723 skb->data_len = first_len - skb_headlen(skb);
724 skb->len = first_len;
725 ipv6_hdr(skb)->payload_len = htons(first_len -
726 sizeof(struct ipv6hdr));
731 /* Prepare header of the next frame,
732 * before previous one went down. */
734 frag->ip_summed = CHECKSUM_NONE;
735 skb_reset_transport_header(frag);
736 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
737 __skb_push(frag, hlen);
738 skb_reset_network_header(frag);
739 memcpy(skb_network_header(frag), tmp_hdr,
741 offset += skb->len - hlen - sizeof(struct frag_hdr);
742 fh->nexthdr = nexthdr;
744 fh->frag_off = htons(offset);
745 if (frag->next != NULL)
746 fh->frag_off |= htons(IP6_MF);
747 fh->identification = frag_id;
748 ipv6_hdr(frag)->payload_len =
750 sizeof(struct ipv6hdr));
751 ip6_copy_metadata(frag, skb);
756 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
757 IPSTATS_MIB_FRAGCREATES);
770 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
771 IPSTATS_MIB_FRAGOKS);
772 dst_release(&rt->dst);
782 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
783 IPSTATS_MIB_FRAGFAILS);
784 dst_release(&rt->dst);
788 skb_walk_frags(skb, frag2) {
792 frag2->destructor = NULL;
793 skb->truesize += frag2->truesize;
798 left = skb->len - hlen; /* Space per frame */
799 ptr = hlen; /* Where to start from */
802 * Fragment the datagram.
805 *prevhdr = NEXTHDR_FRAGMENT;
808 * Keep copying data until we run out.
812 /* IF: it doesn't fit, use 'mtu' - the data space left */
815 /* IF: we are not sending up to and including the packet end
816 then align the next start on an eight byte boundary */
824 if ((frag = alloc_skb(len+hlen+sizeof(struct frag_hdr)+LL_ALLOCATED_SPACE(rt->dst.dev), GFP_ATOMIC)) == NULL) {
825 NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
826 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
827 IPSTATS_MIB_FRAGFAILS);
833 * Set up data on packet
836 ip6_copy_metadata(frag, skb);
837 skb_reserve(frag, LL_RESERVED_SPACE(rt->dst.dev));
838 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
839 skb_reset_network_header(frag);
840 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
841 frag->transport_header = (frag->network_header + hlen +
842 sizeof(struct frag_hdr));
845 * Charge the memory for the fragment to any owner
849 skb_set_owner_w(frag, skb->sk);
852 * Copy the packet header into the new buffer.
854 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
857 * Build fragment header.
859 fh->nexthdr = nexthdr;
862 ipv6_select_ident(fh, rt);
863 frag_id = fh->identification;
865 fh->identification = frag_id;
868 * Copy a block of the IP datagram.
870 if (skb_copy_bits(skb, ptr, skb_transport_header(frag), len))
874 fh->frag_off = htons(offset);
876 fh->frag_off |= htons(IP6_MF);
877 ipv6_hdr(frag)->payload_len = htons(frag->len -
878 sizeof(struct ipv6hdr));
884 * Put this fragment into the sending queue.
890 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
891 IPSTATS_MIB_FRAGCREATES);
893 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
894 IPSTATS_MIB_FRAGOKS);
899 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
900 IPSTATS_MIB_FRAGFAILS);
905 static inline int ip6_rt_check(const struct rt6key *rt_key,
906 const struct in6_addr *fl_addr,
907 const struct in6_addr *addr_cache)
909 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
910 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache));
913 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
914 struct dst_entry *dst,
915 const struct flowi6 *fl6)
917 struct ipv6_pinfo *np = inet6_sk(sk);
923 if (dst->ops->family != AF_INET6) {
928 rt = (struct rt6_info *)dst;
929 /* Yes, checking route validity in not connected
930 * case is not very simple. Take into account,
931 * that we do not support routing by source, TOS,
932 * and MSG_DONTROUTE --ANK (980726)
934 * 1. ip6_rt_check(): If route was host route,
935 * check that cached destination is current.
936 * If it is network route, we still may
937 * check its validity using saved pointer
938 * to the last used address: daddr_cache.
939 * We do not want to save whole address now,
940 * (because main consumer of this service
941 * is tcp, which has not this problem),
942 * so that the last trick works only on connected
944 * 2. oif also should be the same.
946 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
947 #ifdef CONFIG_IPV6_SUBTREES
948 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
950 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
959 static int ip6_dst_lookup_tail(struct sock *sk,
960 struct dst_entry **dst, struct flowi6 *fl6)
962 struct net *net = sock_net(sk);
963 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
969 *dst = ip6_route_output(net, sk, fl6);
971 if ((err = (*dst)->error))
972 goto out_err_release;
974 if (ipv6_addr_any(&fl6->saddr)) {
975 struct rt6_info *rt = (struct rt6_info *) *dst;
976 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
977 sk ? inet6_sk(sk)->srcprefs : 0,
980 goto out_err_release;
983 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
985 * Here if the dst entry we've looked up
986 * has a neighbour entry that is in the INCOMPLETE
987 * state and the src address from the flow is
988 * marked as OPTIMISTIC, we release the found
989 * dst entry and replace it instead with the
990 * dst entry of the nexthop router
993 n = dst_get_neighbour(*dst);
994 if (n && !(n->nud_state & NUD_VALID)) {
995 struct inet6_ifaddr *ifp;
996 struct flowi6 fl_gw6;
1000 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1003 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1009 * We need to get the dst entry for the
1010 * default router instead
1013 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1014 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1015 *dst = ip6_route_output(net, sk, &fl_gw6);
1016 if ((err = (*dst)->error))
1017 goto out_err_release;
1027 if (err == -ENETUNREACH)
1028 IP6_INC_STATS_BH(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1035 * ip6_dst_lookup - perform route lookup on flow
1036 * @sk: socket which provides route info
1037 * @dst: pointer to dst_entry * for result
1038 * @fl6: flow to lookup
1040 * This function performs a route lookup on the given flow.
1042 * It returns zero on success, or a standard errno code on error.
1044 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi6 *fl6)
1047 return ip6_dst_lookup_tail(sk, dst, fl6);
1049 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1052 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1053 * @sk: socket which provides route info
1054 * @fl6: flow to lookup
1055 * @final_dst: final destination address for ipsec lookup
1056 * @can_sleep: we are in a sleepable context
1058 * This function performs a route lookup on the given flow.
1060 * It returns a valid dst pointer on success, or a pointer encoded
1063 struct dst_entry *ip6_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1064 const struct in6_addr *final_dst,
1067 struct dst_entry *dst = NULL;
1070 err = ip6_dst_lookup_tail(sk, &dst, fl6);
1072 return ERR_PTR(err);
1074 ipv6_addr_copy(&fl6->daddr, final_dst);
1076 fl6->flowi6_flags |= FLOWI_FLAG_CAN_SLEEP;
1078 return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1080 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1083 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1084 * @sk: socket which provides the dst cache and route info
1085 * @fl6: flow to lookup
1086 * @final_dst: final destination address for ipsec lookup
1087 * @can_sleep: we are in a sleepable context
1089 * This function performs a route lookup on the given flow with the
1090 * possibility of using the cached route in the socket if it is valid.
1091 * It will take the socket dst lock when operating on the dst cache.
1092 * As a result, this function can only be used in process context.
1094 * It returns a valid dst pointer on success, or a pointer encoded
1097 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1098 const struct in6_addr *final_dst,
1101 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1104 dst = ip6_sk_dst_check(sk, dst, fl6);
1106 err = ip6_dst_lookup_tail(sk, &dst, fl6);
1108 return ERR_PTR(err);
1110 ipv6_addr_copy(&fl6->daddr, final_dst);
1112 fl6->flowi6_flags |= FLOWI_FLAG_CAN_SLEEP;
1114 return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1116 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1118 static inline int ip6_ufo_append_data(struct sock *sk,
1119 int getfrag(void *from, char *to, int offset, int len,
1120 int odd, struct sk_buff *skb),
1121 void *from, int length, int hh_len, int fragheaderlen,
1122 int transhdrlen, int mtu,unsigned int flags,
1123 struct rt6_info *rt)
1126 struct sk_buff *skb;
1129 /* There is support for UDP large send offload by network
1130 * device, so create one single skb packet containing complete
1133 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
1134 struct frag_hdr fhdr;
1136 skb = sock_alloc_send_skb(sk,
1137 hh_len + fragheaderlen + transhdrlen + 20,
1138 (flags & MSG_DONTWAIT), &err);
1142 /* reserve space for Hardware header */
1143 skb_reserve(skb, hh_len);
1145 /* create space for UDP/IP header */
1146 skb_put(skb,fragheaderlen + transhdrlen);
1148 /* initialize network header pointer */
1149 skb_reset_network_header(skb);
1151 /* initialize protocol header pointer */
1152 skb->transport_header = skb->network_header + fragheaderlen;
1154 skb->ip_summed = CHECKSUM_PARTIAL;
1157 /* Specify the length of each IPv6 datagram fragment.
1158 * It has to be a multiple of 8.
1160 skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1161 sizeof(struct frag_hdr)) & ~7;
1162 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1163 ipv6_select_ident(&fhdr, rt);
1164 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1165 __skb_queue_tail(&sk->sk_write_queue, skb);
1168 return skb_append_datato_frags(sk, skb, getfrag, from,
1169 (length - transhdrlen));
1172 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1175 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1178 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1181 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1184 static void ip6_append_data_mtu(unsigned int *mtu,
1186 unsigned int fragheaderlen,
1187 struct sk_buff *skb,
1188 struct rt6_info *rt,
1189 unsigned int orig_mtu)
1191 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1193 /* first fragment, reserve header_len */
1194 *mtu = orig_mtu - rt->dst.header_len;
1198 * this fragment is not first, the headers
1199 * space is regarded as data space.
1203 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1204 + fragheaderlen - sizeof(struct frag_hdr);
1208 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1209 int offset, int len, int odd, struct sk_buff *skb),
1210 void *from, int length, int transhdrlen,
1211 int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1212 struct rt6_info *rt, unsigned int flags, int dontfrag)
1214 struct inet_sock *inet = inet_sk(sk);
1215 struct ipv6_pinfo *np = inet6_sk(sk);
1216 struct inet_cork *cork;
1217 struct sk_buff *skb, *skb_prev = NULL;
1218 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu;
1225 int csummode = CHECKSUM_NONE;
1228 if (flags&MSG_PROBE)
1230 cork = &inet->cork.base;
1231 if (skb_queue_empty(&sk->sk_write_queue)) {
1236 if (WARN_ON(np->cork.opt))
1239 np->cork.opt = kzalloc(opt->tot_len, sk->sk_allocation);
1240 if (unlikely(np->cork.opt == NULL))
1243 np->cork.opt->tot_len = opt->tot_len;
1244 np->cork.opt->opt_flen = opt->opt_flen;
1245 np->cork.opt->opt_nflen = opt->opt_nflen;
1247 np->cork.opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1249 if (opt->dst0opt && !np->cork.opt->dst0opt)
1252 np->cork.opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1254 if (opt->dst1opt && !np->cork.opt->dst1opt)
1257 np->cork.opt->hopopt = ip6_opt_dup(opt->hopopt,
1259 if (opt->hopopt && !np->cork.opt->hopopt)
1262 np->cork.opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1264 if (opt->srcrt && !np->cork.opt->srcrt)
1267 /* need source address above miyazawa*/
1270 cork->dst = &rt->dst;
1271 inet->cork.fl.u.ip6 = *fl6;
1272 np->cork.hop_limit = hlimit;
1273 np->cork.tclass = tclass;
1274 if (rt->dst.flags & DST_XFRM_TUNNEL)
1275 mtu = np->pmtudisc == IPV6_PMTUDISC_PROBE ?
1276 rt->dst.dev->mtu : dst_mtu(&rt->dst);
1278 mtu = np->pmtudisc == IPV6_PMTUDISC_PROBE ?
1279 rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1280 if (np->frag_size < mtu) {
1282 mtu = np->frag_size;
1284 cork->fragsize = mtu;
1285 if (dst_allfrag(rt->dst.path))
1286 cork->flags |= IPCORK_ALLFRAG;
1288 sk->sk_sndmsg_page = NULL;
1289 sk->sk_sndmsg_off = 0;
1290 exthdrlen = (opt ? opt->opt_flen : 0);
1291 length += exthdrlen;
1292 transhdrlen += exthdrlen;
1293 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1295 rt = (struct rt6_info *)cork->dst;
1296 fl6 = &inet->cork.fl.u.ip6;
1301 mtu = cork->fragsize;
1305 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1307 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1308 (opt ? opt->opt_nflen : 0);
1309 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
1311 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1312 if (cork->length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
1313 ipv6_local_error(sk, EMSGSIZE, fl6, mtu-exthdrlen);
1318 /* For UDP, check if TX timestamp is enabled */
1319 if (sk->sk_type == SOCK_DGRAM) {
1320 err = sock_tx_timestamp(sk, &tx_flags);
1326 * Let's try using as much space as possible.
1327 * Use MTU if total length of the message fits into the MTU.
1328 * Otherwise, we need to reserve fragment header and
1329 * fragment alignment (= 8-15 octects, in total).
1331 * Note that we may need to "move" the data from the tail of
1332 * of the buffer to the new fragment when we split
1335 * FIXME: It may be fragmented into multiple chunks
1336 * at once if non-fragmentable extension headers
1341 if ((length > mtu) && dontfrag && (sk->sk_protocol == IPPROTO_UDP ||
1342 sk->sk_protocol == IPPROTO_RAW)) {
1343 ipv6_local_rxpmtu(sk, fl6, mtu-exthdrlen);
1347 skb = skb_peek_tail(&sk->sk_write_queue);
1348 cork->length += length;
1349 if (((length > mtu) ||
1350 (skb && skb_has_frags(skb))) &&
1351 (sk->sk_protocol == IPPROTO_UDP) &&
1352 (rt->dst.dev->features & NETIF_F_UFO)) {
1353 err = ip6_ufo_append_data(sk, getfrag, from, length,
1354 hh_len, fragheaderlen,
1355 transhdrlen, mtu, flags, rt);
1364 while (length > 0) {
1365 /* Check if the remaining data fits into current packet. */
1366 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1368 copy = maxfraglen - skb->len;
1372 unsigned int datalen;
1373 unsigned int fraglen;
1374 unsigned int fraggap;
1375 unsigned int alloclen;
1377 /* There's no room in the current skb */
1379 fraggap = skb->len - maxfraglen;
1382 /* update mtu and maxfraglen if necessary */
1383 if (skb == NULL || skb_prev == NULL)
1384 ip6_append_data_mtu(&mtu, &maxfraglen,
1385 fragheaderlen, skb, rt,
1391 * If remaining data exceeds the mtu,
1392 * we know we need more fragment(s).
1394 datalen = length + fraggap;
1396 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1397 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1398 if ((flags & MSG_MORE) &&
1399 !(rt->dst.dev->features&NETIF_F_SG))
1402 alloclen = datalen + fragheaderlen;
1404 alloclen += dst_exthdrlen;
1406 if (datalen != length + fraggap) {
1408 * this is not the last fragment, the trailer
1409 * space is regarded as data space.
1411 datalen += rt->dst.trailer_len;
1414 alloclen += rt->dst.trailer_len;
1415 fraglen = datalen + fragheaderlen;
1418 * We just reserve space for fragment header.
1419 * Note: this may be overallocation if the message
1420 * (without MSG_MORE) fits into the MTU.
1422 alloclen += sizeof(struct frag_hdr);
1425 skb = sock_alloc_send_skb(sk,
1427 (flags & MSG_DONTWAIT), &err);
1430 if (atomic_read(&sk->sk_wmem_alloc) <=
1432 skb = sock_wmalloc(sk,
1433 alloclen + hh_len, 1,
1435 if (unlikely(skb == NULL))
1438 /* Only the initial fragment
1447 * Fill in the control structures
1449 skb->ip_summed = csummode;
1451 /* reserve for fragmentation and ipsec header */
1452 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1455 if (sk->sk_type == SOCK_DGRAM)
1456 skb_shinfo(skb)->tx_flags = tx_flags;
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;
1492 csummode = CHECKSUM_NONE;
1495 * Put the packet on the pending queue
1497 __skb_queue_tail(&sk->sk_write_queue, skb);
1504 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1508 if (getfrag(from, skb_put(skb, copy),
1509 offset, copy, off, skb) < 0) {
1510 __skb_trim(skb, off);
1515 int i = skb_shinfo(skb)->nr_frags;
1516 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1517 struct page *page = sk->sk_sndmsg_page;
1518 int off = sk->sk_sndmsg_off;
1521 if (page && (left = PAGE_SIZE - off) > 0) {
1524 if (page != skb_frag_page(frag)) {
1525 if (i == MAX_SKB_FRAGS) {
1529 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1530 skb_frag_ref(skb, i);
1531 frag = &skb_shinfo(skb)->frags[i];
1533 } else if(i < MAX_SKB_FRAGS) {
1534 if (copy > PAGE_SIZE)
1536 page = alloc_pages(sk->sk_allocation, 0);
1541 sk->sk_sndmsg_page = page;
1542 sk->sk_sndmsg_off = 0;
1544 skb_fill_page_desc(skb, i, page, 0, 0);
1545 frag = &skb_shinfo(skb)->frags[i];
1551 skb_frag_address(frag) + skb_frag_size(frag),
1552 offset, copy, skb->len, skb) < 0) {
1556 sk->sk_sndmsg_off += copy;
1557 skb_frag_size_add(frag, copy);
1559 skb->data_len += copy;
1560 skb->truesize += copy;
1561 atomic_add(copy, &sk->sk_wmem_alloc);
1568 cork->length -= length;
1569 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1573 static void ip6_cork_release(struct inet_sock *inet, struct ipv6_pinfo *np)
1576 kfree(np->cork.opt->dst0opt);
1577 kfree(np->cork.opt->dst1opt);
1578 kfree(np->cork.opt->hopopt);
1579 kfree(np->cork.opt->srcrt);
1580 kfree(np->cork.opt);
1581 np->cork.opt = NULL;
1584 if (inet->cork.base.dst) {
1585 dst_release(inet->cork.base.dst);
1586 inet->cork.base.dst = NULL;
1587 inet->cork.base.flags &= ~IPCORK_ALLFRAG;
1589 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1592 int ip6_push_pending_frames(struct sock *sk)
1594 struct sk_buff *skb, *tmp_skb;
1595 struct sk_buff **tail_skb;
1596 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1597 struct inet_sock *inet = inet_sk(sk);
1598 struct ipv6_pinfo *np = inet6_sk(sk);
1599 struct net *net = sock_net(sk);
1600 struct ipv6hdr *hdr;
1601 struct ipv6_txoptions *opt = np->cork.opt;
1602 struct rt6_info *rt = (struct rt6_info *)inet->cork.base.dst;
1603 struct flowi6 *fl6 = &inet->cork.fl.u.ip6;
1604 unsigned char proto = fl6->flowi6_proto;
1607 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1609 tail_skb = &(skb_shinfo(skb)->frag_list);
1611 /* move skb->data to ip header from ext header */
1612 if (skb->data < skb_network_header(skb))
1613 __skb_pull(skb, skb_network_offset(skb));
1614 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1615 __skb_pull(tmp_skb, skb_network_header_len(skb));
1616 *tail_skb = tmp_skb;
1617 tail_skb = &(tmp_skb->next);
1618 skb->len += tmp_skb->len;
1619 skb->data_len += tmp_skb->len;
1620 skb->truesize += tmp_skb->truesize;
1621 tmp_skb->destructor = NULL;
1625 /* Allow local fragmentation. */
1626 if (np->pmtudisc < IPV6_PMTUDISC_DO)
1629 ipv6_addr_copy(final_dst, &fl6->daddr);
1630 __skb_pull(skb, skb_network_header_len(skb));
1631 if (opt && opt->opt_flen)
1632 ipv6_push_frag_opts(skb, opt, &proto);
1633 if (opt && opt->opt_nflen)
1634 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1636 skb_push(skb, sizeof(struct ipv6hdr));
1637 skb_reset_network_header(skb);
1638 hdr = ipv6_hdr(skb);
1640 *(__be32*)hdr = fl6->flowlabel |
1641 htonl(0x60000000 | ((int)np->cork.tclass << 20));
1643 hdr->hop_limit = np->cork.hop_limit;
1644 hdr->nexthdr = proto;
1645 ipv6_addr_copy(&hdr->saddr, &fl6->saddr);
1646 ipv6_addr_copy(&hdr->daddr, final_dst);
1648 skb->priority = sk->sk_priority;
1649 skb->mark = sk->sk_mark;
1651 skb_dst_set(skb, dst_clone(&rt->dst));
1652 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1653 if (proto == IPPROTO_ICMPV6) {
1654 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1656 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1657 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1660 err = ip6_local_out(skb);
1663 err = net_xmit_errno(err);
1669 ip6_cork_release(inet, np);
1672 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1676 void ip6_flush_pending_frames(struct sock *sk)
1678 struct sk_buff *skb;
1680 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1682 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1683 IPSTATS_MIB_OUTDISCARDS);
1687 ip6_cork_release(inet_sk(sk), inet6_sk(sk));