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/bpf-cgroup.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>
60 #include <net/lwtunnel.h>
62 static int ip6_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
64 struct dst_entry *dst = skb_dst(skb);
65 struct net_device *dev = dst->dev;
66 struct neighbour *neigh;
67 struct in6_addr *nexthop;
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)) {
106 if (lwtunnel_xmit_redirect(dst->lwtstate)) {
107 int res = lwtunnel_xmit(skb);
109 if (res < 0 || res == LWTUNNEL_XMIT_DONE)
114 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
115 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
116 if (unlikely(!neigh))
117 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
118 if (!IS_ERR(neigh)) {
119 sock_confirm_neigh(skb, neigh);
120 ret = neigh_output(neigh, skb);
121 rcu_read_unlock_bh();
124 rcu_read_unlock_bh();
126 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
132 ip6_finish_output_gso_slowpath_drop(struct net *net, struct sock *sk,
133 struct sk_buff *skb, unsigned int mtu)
135 struct sk_buff *segs, *nskb;
136 netdev_features_t features;
139 /* Please see corresponding comment in ip_finish_output_gso
140 * describing the cases where GSO segment length exceeds the
143 features = netif_skb_features(skb);
144 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
145 if (IS_ERR_OR_NULL(segs)) {
152 skb_list_walk_safe(segs, segs, nskb) {
155 skb_mark_not_on_list(segs);
156 err = ip6_fragment(net, sk, segs, ip6_finish_output2);
164 static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
169 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
175 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
176 /* Policy lookup after SNAT yielded a new policy */
177 if (skb_dst(skb)->xfrm) {
178 IP6CB(skb)->flags |= IP6SKB_REROUTED;
179 return dst_output(net, sk, skb);
183 mtu = ip6_skb_dst_mtu(skb);
184 if (skb_is_gso(skb) && !skb_gso_validate_mtu(skb, mtu))
185 return ip6_finish_output_gso_slowpath_drop(net, sk, skb, mtu);
187 if ((skb->len > mtu && !skb_is_gso(skb)) ||
188 dst_allfrag(skb_dst(skb)) ||
189 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
190 return ip6_fragment(net, sk, skb, ip6_finish_output2);
192 return ip6_finish_output2(net, sk, skb);
195 int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
197 struct net_device *dev = skb_dst(skb)->dev;
198 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
200 skb->protocol = htons(ETH_P_IPV6);
203 if (unlikely(idev->cnf.disable_ipv6)) {
204 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
209 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
210 net, sk, skb, NULL, dev,
212 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
215 bool ip6_autoflowlabel(struct net *net, const struct ipv6_pinfo *np)
217 if (!np->autoflowlabel_set)
218 return ip6_default_np_autolabel(net);
220 return np->autoflowlabel;
224 * xmit an sk_buff (used by TCP, SCTP and DCCP)
225 * Note : socket lock is not held for SYNACK packets, but might be modified
226 * by calls to skb_set_owner_w() and ipv6_local_error(),
227 * which are using proper atomic operations or spinlocks.
229 int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
230 __u32 mark, struct ipv6_txoptions *opt, int tclass)
232 struct net *net = sock_net(sk);
233 const struct ipv6_pinfo *np = inet6_sk(sk);
234 struct in6_addr *first_hop = &fl6->daddr;
235 struct dst_entry *dst = skb_dst(skb);
236 unsigned int head_room;
238 u8 proto = fl6->flowi6_proto;
239 int seg_len = skb->len;
243 head_room = sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
245 head_room += opt->opt_nflen + opt->opt_flen;
247 if (unlikely(skb_headroom(skb) < head_room)) {
248 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
250 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
251 IPSTATS_MIB_OUTDISCARDS);
256 skb_set_owner_w(skb2, skb->sk);
262 seg_len += opt->opt_nflen + opt->opt_flen;
265 ipv6_push_frag_opts(skb, opt, &proto);
268 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop,
272 skb_push(skb, sizeof(struct ipv6hdr));
273 skb_reset_network_header(skb);
277 * Fill in the IPv6 header
280 hlimit = np->hop_limit;
282 hlimit = ip6_dst_hoplimit(dst);
284 ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
285 ip6_autoflowlabel(net, np), fl6));
287 hdr->payload_len = htons(seg_len);
288 hdr->nexthdr = proto;
289 hdr->hop_limit = hlimit;
291 hdr->saddr = fl6->saddr;
292 hdr->daddr = *first_hop;
294 skb->protocol = htons(ETH_P_IPV6);
295 skb->priority = sk->sk_priority;
299 if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
300 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
301 IPSTATS_MIB_OUT, skb->len);
303 /* if egress device is enslaved to an L3 master device pass the
304 * skb to its handler for processing
306 skb = l3mdev_ip6_out((struct sock *)sk, skb);
310 /* hooks should never assume socket lock is held.
311 * we promote our socket to non const
313 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
314 net, (struct sock *)sk, skb, NULL, dst->dev,
319 /* ipv6_local_error() does not require socket lock,
320 * we promote our socket to non const
322 ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
324 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
328 EXPORT_SYMBOL(ip6_xmit);
330 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
332 struct ip6_ra_chain *ra;
333 struct sock *last = NULL;
335 read_lock(&ip6_ra_lock);
336 for (ra = ip6_ra_chain; ra; ra = ra->next) {
337 struct sock *sk = ra->sk;
338 if (sk && ra->sel == sel &&
339 (!sk->sk_bound_dev_if ||
340 sk->sk_bound_dev_if == skb->dev->ifindex)) {
342 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
344 rawv6_rcv(last, skb2);
351 rawv6_rcv(last, skb);
352 read_unlock(&ip6_ra_lock);
355 read_unlock(&ip6_ra_lock);
359 static int ip6_forward_proxy_check(struct sk_buff *skb)
361 struct ipv6hdr *hdr = ipv6_hdr(skb);
362 u8 nexthdr = hdr->nexthdr;
366 if (ipv6_ext_hdr(nexthdr)) {
367 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
371 offset = sizeof(struct ipv6hdr);
373 if (nexthdr == IPPROTO_ICMPV6) {
374 struct icmp6hdr *icmp6;
376 if (!pskb_may_pull(skb, (skb_network_header(skb) +
377 offset + 1 - skb->data)))
380 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
382 switch (icmp6->icmp6_type) {
383 case NDISC_ROUTER_SOLICITATION:
384 case NDISC_ROUTER_ADVERTISEMENT:
385 case NDISC_NEIGHBOUR_SOLICITATION:
386 case NDISC_NEIGHBOUR_ADVERTISEMENT:
388 /* For reaction involving unicast neighbor discovery
389 * message destined to the proxied address, pass it to
399 * The proxying router can't forward traffic sent to a link-local
400 * address, so signal the sender and discard the packet. This
401 * behavior is clarified by the MIPv6 specification.
403 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
404 dst_link_failure(skb);
411 static inline int ip6_forward_finish(struct net *net, struct sock *sk,
414 struct dst_entry *dst = skb_dst(skb);
416 __IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
417 __IP6_ADD_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
419 return dst_output(net, sk, skb);
422 static unsigned int ip6_dst_mtu_forward(const struct dst_entry *dst)
425 struct inet6_dev *idev;
427 if (dst_metric_locked(dst, RTAX_MTU)) {
428 mtu = dst_metric_raw(dst, RTAX_MTU);
435 idev = __in6_dev_get(dst->dev);
437 mtu = idev->cnf.mtu6;
443 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
448 /* ipv6 conntrack defrag sets max_frag_size + ignore_df */
449 if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
455 if (skb_is_gso(skb) && skb_gso_validate_mtu(skb, mtu))
461 int ip6_forward(struct sk_buff *skb)
463 struct dst_entry *dst = skb_dst(skb);
464 struct ipv6hdr *hdr = ipv6_hdr(skb);
465 struct inet6_skb_parm *opt = IP6CB(skb);
466 struct net *net = dev_net(dst->dev);
469 if (net->ipv6.devconf_all->forwarding == 0)
472 if (skb->pkt_type != PACKET_HOST)
475 if (unlikely(skb->sk))
478 if (skb_warn_if_lro(skb))
481 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
482 __IP6_INC_STATS(net, ip6_dst_idev(dst),
483 IPSTATS_MIB_INDISCARDS);
487 skb_forward_csum(skb);
490 * We DO NOT make any processing on
491 * RA packets, pushing them to user level AS IS
492 * without ane WARRANTY that application will be able
493 * to interpret them. The reason is that we
494 * cannot make anything clever here.
496 * We are not end-node, so that if packet contains
497 * AH/ESP, we cannot make anything.
498 * Defragmentation also would be mistake, RA packets
499 * cannot be fragmented, because there is no warranty
500 * that different fragments will go along one path. --ANK
502 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
503 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
508 * check and decrement ttl
510 if (hdr->hop_limit <= 1) {
511 /* Force OUTPUT device used as source address */
513 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
514 __IP6_INC_STATS(net, ip6_dst_idev(dst),
515 IPSTATS_MIB_INHDRERRORS);
521 /* XXX: idev->cnf.proxy_ndp? */
522 if (net->ipv6.devconf_all->proxy_ndp &&
523 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
524 int proxied = ip6_forward_proxy_check(skb);
526 return ip6_input(skb);
527 else if (proxied < 0) {
528 __IP6_INC_STATS(net, ip6_dst_idev(dst),
529 IPSTATS_MIB_INDISCARDS);
534 if (!xfrm6_route_forward(skb)) {
535 __IP6_INC_STATS(net, ip6_dst_idev(dst),
536 IPSTATS_MIB_INDISCARDS);
541 /* IPv6 specs say nothing about it, but it is clear that we cannot
542 send redirects to source routed frames.
543 We don't send redirects to frames decapsulated from IPsec.
545 if (IP6CB(skb)->iif == dst->dev->ifindex &&
546 opt->srcrt == 0 && !skb_sec_path(skb)) {
547 struct in6_addr *target = NULL;
548 struct inet_peer *peer;
552 * incoming and outgoing devices are the same
556 rt = (struct rt6_info *) dst;
557 if (rt->rt6i_flags & RTF_GATEWAY)
558 target = &rt->rt6i_gateway;
560 target = &hdr->daddr;
562 peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
564 /* Limit redirects both by destination (here)
565 and by source (inside ndisc_send_redirect)
567 if (inet_peer_xrlim_allow(peer, 1*HZ))
568 ndisc_send_redirect(skb, target);
572 int addrtype = ipv6_addr_type(&hdr->saddr);
574 /* This check is security critical. */
575 if (addrtype == IPV6_ADDR_ANY ||
576 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
578 if (addrtype & IPV6_ADDR_LINKLOCAL) {
579 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
580 ICMPV6_NOT_NEIGHBOUR, 0);
585 mtu = ip6_dst_mtu_forward(dst);
586 if (mtu < IPV6_MIN_MTU)
589 if (ip6_pkt_too_big(skb, mtu)) {
590 /* Again, force OUTPUT device used as source address */
592 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
593 __IP6_INC_STATS(net, ip6_dst_idev(dst),
594 IPSTATS_MIB_INTOOBIGERRORS);
595 __IP6_INC_STATS(net, ip6_dst_idev(dst),
596 IPSTATS_MIB_FRAGFAILS);
601 if (skb_cow(skb, dst->dev->hard_header_len)) {
602 __IP6_INC_STATS(net, ip6_dst_idev(dst),
603 IPSTATS_MIB_OUTDISCARDS);
609 /* Mangling hops number delayed to point after skb COW */
613 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
614 net, NULL, skb, skb->dev, dst->dev,
618 __IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
624 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
626 to->pkt_type = from->pkt_type;
627 to->priority = from->priority;
628 to->protocol = from->protocol;
630 skb_dst_set(to, dst_clone(skb_dst(from)));
632 to->mark = from->mark;
634 skb_copy_hash(to, from);
636 #ifdef CONFIG_NET_SCHED
637 to->tc_index = from->tc_index;
640 skb_copy_secmark(to, from);
643 int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
644 int (*output)(struct net *, struct sock *, struct sk_buff *))
646 struct sk_buff *frag;
647 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
648 struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
649 inet6_sk(skb->sk) : NULL;
650 struct ipv6hdr *tmp_hdr;
652 unsigned int mtu, hlen, left, len, nexthdr_offset;
655 int ptr, offset = 0, err = 0;
656 u8 *prevhdr, nexthdr = 0;
658 err = ip6_find_1stfragopt(skb, &prevhdr);
663 nexthdr_offset = prevhdr - skb_network_header(skb);
665 mtu = ip6_skb_dst_mtu(skb);
667 /* We must not fragment if the socket is set to force MTU discovery
668 * or if the skb it not generated by a local socket.
670 if (unlikely(!skb->ignore_df && skb->len > mtu))
673 if (IP6CB(skb)->frag_max_size) {
674 if (IP6CB(skb)->frag_max_size > mtu)
677 /* don't send fragments larger than what we received */
678 mtu = IP6CB(skb)->frag_max_size;
679 if (mtu < IPV6_MIN_MTU)
683 if (np && np->frag_size < mtu) {
687 if (mtu < hlen + sizeof(struct frag_hdr) + 8)
689 mtu -= hlen + sizeof(struct frag_hdr);
691 frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
692 &ipv6_hdr(skb)->saddr);
694 if (skb->ip_summed == CHECKSUM_PARTIAL &&
695 (err = skb_checksum_help(skb)))
698 prevhdr = skb_network_header(skb) + nexthdr_offset;
699 hroom = LL_RESERVED_SPACE(rt->dst.dev);
700 if (skb_has_frag_list(skb)) {
701 unsigned int first_len = skb_pagelen(skb);
702 struct sk_buff *frag2;
704 if (first_len - hlen > mtu ||
705 ((first_len - hlen) & 7) ||
707 skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
710 skb_walk_frags(skb, frag) {
711 /* Correct geometry. */
712 if (frag->len > mtu ||
713 ((frag->len & 7) && frag->next) ||
714 skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
715 goto slow_path_clean;
717 /* Partially cloned skb? */
718 if (skb_shared(frag))
719 goto slow_path_clean;
724 frag->destructor = sock_wfree;
726 skb->truesize -= frag->truesize;
733 *prevhdr = NEXTHDR_FRAGMENT;
734 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
739 frag = skb_shinfo(skb)->frag_list;
740 skb_frag_list_init(skb);
742 __skb_pull(skb, hlen);
743 fh = __skb_push(skb, sizeof(struct frag_hdr));
744 __skb_push(skb, hlen);
745 skb_reset_network_header(skb);
746 memcpy(skb_network_header(skb), tmp_hdr, hlen);
748 fh->nexthdr = nexthdr;
750 fh->frag_off = htons(IP6_MF);
751 fh->identification = frag_id;
753 first_len = skb_pagelen(skb);
754 skb->data_len = first_len - skb_headlen(skb);
755 skb->len = first_len;
756 ipv6_hdr(skb)->payload_len = htons(first_len -
757 sizeof(struct ipv6hdr));
759 /* We prevent @rt from being freed. */
763 /* Prepare header of the next frame,
764 * before previous one went down. */
766 frag->ip_summed = CHECKSUM_NONE;
767 skb_reset_transport_header(frag);
768 fh = __skb_push(frag, sizeof(struct frag_hdr));
769 __skb_push(frag, hlen);
770 skb_reset_network_header(frag);
771 memcpy(skb_network_header(frag), tmp_hdr,
773 offset += skb->len - hlen - sizeof(struct frag_hdr);
774 fh->nexthdr = nexthdr;
776 fh->frag_off = htons(offset);
778 fh->frag_off |= htons(IP6_MF);
779 fh->identification = frag_id;
780 ipv6_hdr(frag)->payload_len =
782 sizeof(struct ipv6hdr));
783 ip6_copy_metadata(frag, skb);
786 err = output(net, sk, skb);
788 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
789 IPSTATS_MIB_FRAGCREATES);
802 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
803 IPSTATS_MIB_FRAGOKS);
808 kfree_skb_list(frag);
810 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
811 IPSTATS_MIB_FRAGFAILS);
816 skb_walk_frags(skb, frag2) {
820 frag2->destructor = NULL;
821 skb->truesize += frag2->truesize;
826 left = skb->len - hlen; /* Space per frame */
827 ptr = hlen; /* Where to start from */
830 * Fragment the datagram.
833 troom = rt->dst.dev->needed_tailroom;
836 * Keep copying data until we run out.
839 u8 *fragnexthdr_offset;
842 /* IF: it doesn't fit, use 'mtu' - the data space left */
845 /* IF: we are not sending up to and including the packet end
846 then align the next start on an eight byte boundary */
851 /* Allocate buffer */
852 frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
853 hroom + troom, GFP_ATOMIC);
860 * Set up data on packet
863 ip6_copy_metadata(frag, skb);
864 skb_reserve(frag, hroom);
865 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
866 skb_reset_network_header(frag);
867 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
868 frag->transport_header = (frag->network_header + hlen +
869 sizeof(struct frag_hdr));
872 * Charge the memory for the fragment to any owner
876 skb_set_owner_w(frag, skb->sk);
879 * Copy the packet header into the new buffer.
881 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
883 fragnexthdr_offset = skb_network_header(frag);
884 fragnexthdr_offset += prevhdr - skb_network_header(skb);
885 *fragnexthdr_offset = NEXTHDR_FRAGMENT;
888 * Build fragment header.
890 fh->nexthdr = nexthdr;
892 fh->identification = frag_id;
895 * Copy a block of the IP datagram.
897 BUG_ON(skb_copy_bits(skb, ptr, skb_transport_header(frag),
901 fh->frag_off = htons(offset);
903 fh->frag_off |= htons(IP6_MF);
904 ipv6_hdr(frag)->payload_len = htons(frag->len -
905 sizeof(struct ipv6hdr));
911 * Put this fragment into the sending queue.
913 err = output(net, sk, frag);
917 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
918 IPSTATS_MIB_FRAGCREATES);
920 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
921 IPSTATS_MIB_FRAGOKS);
926 if (skb->sk && dst_allfrag(skb_dst(skb)))
927 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
929 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
933 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
934 IPSTATS_MIB_FRAGFAILS);
939 static inline int ip6_rt_check(const struct rt6key *rt_key,
940 const struct in6_addr *fl_addr,
941 const struct in6_addr *addr_cache)
943 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
944 (!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
947 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
948 struct dst_entry *dst,
949 const struct flowi6 *fl6)
951 struct ipv6_pinfo *np = inet6_sk(sk);
957 if (dst->ops->family != AF_INET6) {
962 rt = (struct rt6_info *)dst;
963 /* Yes, checking route validity in not connected
964 * case is not very simple. Take into account,
965 * that we do not support routing by source, TOS,
966 * and MSG_DONTROUTE --ANK (980726)
968 * 1. ip6_rt_check(): If route was host route,
969 * check that cached destination is current.
970 * If it is network route, we still may
971 * check its validity using saved pointer
972 * to the last used address: daddr_cache.
973 * We do not want to save whole address now,
974 * (because main consumer of this service
975 * is tcp, which has not this problem),
976 * so that the last trick works only on connected
978 * 2. oif also should be the same.
980 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
981 #ifdef CONFIG_IPV6_SUBTREES
982 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
984 (!(fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF) &&
985 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex))) {
994 static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
995 struct dst_entry **dst, struct flowi6 *fl6)
997 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1004 /* The correct way to handle this would be to do
1005 * ip6_route_get_saddr, and then ip6_route_output; however,
1006 * the route-specific preferred source forces the
1007 * ip6_route_output call _before_ ip6_route_get_saddr.
1009 * In source specific routing (no src=any default route),
1010 * ip6_route_output will fail given src=any saddr, though, so
1011 * that's why we try it again later.
1013 if (ipv6_addr_any(&fl6->saddr) && (!*dst || !(*dst)->error)) {
1014 struct rt6_info *rt;
1015 bool had_dst = *dst != NULL;
1018 *dst = ip6_route_output(net, sk, fl6);
1019 rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
1020 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
1021 sk ? inet6_sk(sk)->srcprefs : 0,
1024 goto out_err_release;
1026 /* If we had an erroneous initial result, pretend it
1027 * never existed and let the SA-enabled version take
1030 if (!had_dst && (*dst)->error) {
1035 if (fl6->flowi6_oif)
1036 flags |= RT6_LOOKUP_F_IFACE;
1040 *dst = ip6_route_output_flags(net, sk, fl6, flags);
1042 err = (*dst)->error;
1044 goto out_err_release;
1046 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1048 * Here if the dst entry we've looked up
1049 * has a neighbour entry that is in the INCOMPLETE
1050 * state and the src address from the flow is
1051 * marked as OPTIMISTIC, we release the found
1052 * dst entry and replace it instead with the
1053 * dst entry of the nexthop router
1055 rt = (struct rt6_info *) *dst;
1057 n = __ipv6_neigh_lookup_noref(rt->dst.dev,
1058 rt6_nexthop(rt, &fl6->daddr));
1059 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
1060 rcu_read_unlock_bh();
1063 struct inet6_ifaddr *ifp;
1064 struct flowi6 fl_gw6;
1067 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1070 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1076 * We need to get the dst entry for the
1077 * default router instead
1080 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1081 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1082 *dst = ip6_route_output(net, sk, &fl_gw6);
1083 err = (*dst)->error;
1085 goto out_err_release;
1089 if (ipv6_addr_v4mapped(&fl6->saddr) &&
1090 !(ipv6_addr_v4mapped(&fl6->daddr) || ipv6_addr_any(&fl6->daddr))) {
1091 err = -EAFNOSUPPORT;
1092 goto out_err_release;
1101 if (err == -ENETUNREACH)
1102 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1107 * ip6_dst_lookup - perform route lookup on flow
1108 * @sk: socket which provides route info
1109 * @dst: pointer to dst_entry * for result
1110 * @fl6: flow to lookup
1112 * This function performs a route lookup on the given flow.
1114 * It returns zero on success, or a standard errno code on error.
1116 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1120 return ip6_dst_lookup_tail(net, sk, dst, fl6);
1122 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1125 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1126 * @sk: socket which provides route info
1127 * @fl6: flow to lookup
1128 * @final_dst: final destination address for ipsec lookup
1130 * This function performs a route lookup on the given flow.
1132 * It returns a valid dst pointer on success, or a pointer encoded
1135 struct dst_entry *ip6_dst_lookup_flow(struct net *net, const struct sock *sk, struct flowi6 *fl6,
1136 const struct in6_addr *final_dst)
1138 struct dst_entry *dst = NULL;
1141 err = ip6_dst_lookup_tail(net, sk, &dst, fl6);
1143 return ERR_PTR(err);
1145 fl6->daddr = *final_dst;
1147 return xfrm_lookup_route(net, dst, flowi6_to_flowi(fl6), sk, 0);
1149 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1152 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1153 * @sk: socket which provides the dst cache and route info
1154 * @fl6: flow to lookup
1155 * @final_dst: final destination address for ipsec lookup
1157 * This function performs a route lookup on the given flow with the
1158 * possibility of using the cached route in the socket if it is valid.
1159 * It will take the socket dst lock when operating on the dst cache.
1160 * As a result, this function can only be used in process context.
1162 * It returns a valid dst pointer on success, or a pointer encoded
1165 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1166 const struct in6_addr *final_dst)
1168 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1170 dst = ip6_sk_dst_check(sk, dst, fl6);
1172 dst = ip6_dst_lookup_flow(sock_net(sk), sk, fl6, final_dst);
1176 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1178 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1181 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1184 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1187 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1190 static void ip6_append_data_mtu(unsigned int *mtu,
1192 unsigned int fragheaderlen,
1193 struct sk_buff *skb,
1194 struct rt6_info *rt,
1195 unsigned int orig_mtu)
1197 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1199 /* first fragment, reserve header_len */
1200 *mtu = orig_mtu - rt->dst.header_len;
1204 * this fragment is not first, the headers
1205 * space is regarded as data space.
1209 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1210 + fragheaderlen - sizeof(struct frag_hdr);
1214 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1215 struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6,
1216 struct rt6_info *rt, struct flowi6 *fl6)
1218 struct ipv6_pinfo *np = inet6_sk(sk);
1220 struct ipv6_txoptions *opt = ipc6->opt;
1226 if (WARN_ON(v6_cork->opt))
1229 v6_cork->opt = kzalloc(sizeof(*opt), sk->sk_allocation);
1230 if (unlikely(!v6_cork->opt))
1233 v6_cork->opt->tot_len = sizeof(*opt);
1234 v6_cork->opt->opt_flen = opt->opt_flen;
1235 v6_cork->opt->opt_nflen = opt->opt_nflen;
1237 v6_cork->opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1239 if (opt->dst0opt && !v6_cork->opt->dst0opt)
1242 v6_cork->opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1244 if (opt->dst1opt && !v6_cork->opt->dst1opt)
1247 v6_cork->opt->hopopt = ip6_opt_dup(opt->hopopt,
1249 if (opt->hopopt && !v6_cork->opt->hopopt)
1252 v6_cork->opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1254 if (opt->srcrt && !v6_cork->opt->srcrt)
1257 /* need source address above miyazawa*/
1260 cork->base.dst = &rt->dst;
1261 cork->fl.u.ip6 = *fl6;
1262 v6_cork->hop_limit = ipc6->hlimit;
1263 v6_cork->tclass = ipc6->tclass;
1264 if (rt->dst.flags & DST_XFRM_TUNNEL)
1265 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1266 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(&rt->dst);
1268 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1269 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(rt->dst.path);
1270 if (np->frag_size < mtu) {
1272 mtu = np->frag_size;
1274 cork->base.fragsize = mtu;
1275 if (dst_allfrag(rt->dst.path))
1276 cork->base.flags |= IPCORK_ALLFRAG;
1277 cork->base.length = 0;
1282 static int __ip6_append_data(struct sock *sk,
1284 struct sk_buff_head *queue,
1285 struct inet_cork *cork,
1286 struct inet6_cork *v6_cork,
1287 struct page_frag *pfrag,
1288 int getfrag(void *from, char *to, int offset,
1289 int len, int odd, struct sk_buff *skb),
1290 void *from, int length, int transhdrlen,
1291 unsigned int flags, struct ipcm6_cookie *ipc6,
1292 const struct sockcm_cookie *sockc)
1294 struct sk_buff *skb, *skb_prev = NULL;
1295 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu, pmtu;
1297 int dst_exthdrlen = 0;
1304 struct rt6_info *rt = (struct rt6_info *)cork->dst;
1305 struct ipv6_txoptions *opt = v6_cork->opt;
1306 int csummode = CHECKSUM_NONE;
1307 unsigned int maxnonfragsize, headersize;
1309 skb = skb_peek_tail(queue);
1311 exthdrlen = opt ? opt->opt_flen : 0;
1312 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1315 mtu = cork->fragsize;
1318 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1320 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1321 (opt ? opt->opt_nflen : 0);
1323 headersize = sizeof(struct ipv6hdr) +
1324 (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1325 (dst_allfrag(&rt->dst) ?
1326 sizeof(struct frag_hdr) : 0) +
1327 rt->rt6i_nfheader_len;
1329 if (mtu <= fragheaderlen ||
1330 ((mtu - fragheaderlen) & ~7) + fragheaderlen <= sizeof(struct frag_hdr))
1333 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1334 sizeof(struct frag_hdr);
1336 /* as per RFC 7112 section 5, the entire IPv6 Header Chain must fit
1337 * the first fragment
1339 if (headersize + transhdrlen > mtu)
1342 if (cork->length + length > mtu - headersize && ipc6->dontfrag &&
1343 (sk->sk_protocol == IPPROTO_UDP ||
1344 sk->sk_protocol == IPPROTO_RAW)) {
1345 ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1346 sizeof(struct ipv6hdr));
1350 if (ip6_sk_ignore_df(sk))
1351 maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1353 maxnonfragsize = mtu;
1355 if (cork->length + length > maxnonfragsize - headersize) {
1357 pmtu = max_t(int, mtu - headersize + sizeof(struct ipv6hdr), 0);
1358 ipv6_local_error(sk, EMSGSIZE, fl6, pmtu);
1362 /* CHECKSUM_PARTIAL only with no extension headers and when
1363 * we are not going to fragment
1365 if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1366 headersize == sizeof(struct ipv6hdr) &&
1367 length <= mtu - headersize &&
1368 !(flags & MSG_MORE) &&
1369 rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1370 csummode = CHECKSUM_PARTIAL;
1372 if (sk->sk_type == SOCK_DGRAM || sk->sk_type == SOCK_RAW) {
1373 sock_tx_timestamp(sk, sockc->tsflags, &tx_flags);
1374 if (tx_flags & SKBTX_ANY_SW_TSTAMP &&
1375 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1376 tskey = sk->sk_tskey++;
1380 * Let's try using as much space as possible.
1381 * Use MTU if total length of the message fits into the MTU.
1382 * Otherwise, we need to reserve fragment header and
1383 * fragment alignment (= 8-15 octects, in total).
1385 * Note that we may need to "move" the data from the tail of
1386 * of the buffer to the new fragment when we split
1389 * FIXME: It may be fragmented into multiple chunks
1390 * at once if non-fragmentable extension headers
1395 cork->length += length;
1399 while (length > 0) {
1400 /* Check if the remaining data fits into current packet. */
1401 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1403 copy = maxfraglen - skb->len;
1407 unsigned int datalen;
1408 unsigned int fraglen;
1409 unsigned int fraggap;
1410 unsigned int alloclen;
1412 /* There's no room in the current skb */
1414 fraggap = skb->len - maxfraglen;
1417 /* update mtu and maxfraglen if necessary */
1418 if (!skb || !skb_prev)
1419 ip6_append_data_mtu(&mtu, &maxfraglen,
1420 fragheaderlen, skb, rt,
1426 * If remaining data exceeds the mtu,
1427 * we know we need more fragment(s).
1429 datalen = length + fraggap;
1431 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1432 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1433 if ((flags & MSG_MORE) &&
1434 !(rt->dst.dev->features&NETIF_F_SG))
1437 alloclen = datalen + fragheaderlen;
1439 alloclen += dst_exthdrlen;
1441 if (datalen != length + fraggap) {
1443 * this is not the last fragment, the trailer
1444 * space is regarded as data space.
1446 datalen += rt->dst.trailer_len;
1449 alloclen += rt->dst.trailer_len;
1450 fraglen = datalen + fragheaderlen;
1453 * We just reserve space for fragment header.
1454 * Note: this may be overallocation if the message
1455 * (without MSG_MORE) fits into the MTU.
1457 alloclen += sizeof(struct frag_hdr);
1459 copy = datalen - transhdrlen - fraggap;
1465 skb = sock_alloc_send_skb(sk,
1467 (flags & MSG_DONTWAIT), &err);
1470 if (refcount_read(&sk->sk_wmem_alloc) <=
1472 skb = sock_wmalloc(sk,
1473 alloclen + hh_len, 1,
1481 * Fill in the control structures
1483 skb->protocol = htons(ETH_P_IPV6);
1484 skb->ip_summed = csummode;
1486 /* reserve for fragmentation and ipsec header */
1487 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1490 /* Only the initial fragment is time stamped */
1491 skb_shinfo(skb)->tx_flags = tx_flags;
1493 skb_shinfo(skb)->tskey = tskey;
1497 * Find where to start putting bytes
1499 data = skb_put(skb, fraglen);
1500 skb_set_network_header(skb, exthdrlen);
1501 data += fragheaderlen;
1502 skb->transport_header = (skb->network_header +
1505 skb->csum = skb_copy_and_csum_bits(
1506 skb_prev, maxfraglen,
1507 data + transhdrlen, fraggap, 0);
1508 skb_prev->csum = csum_sub(skb_prev->csum,
1511 pskb_trim_unique(skb_prev, maxfraglen);
1514 getfrag(from, data + transhdrlen, offset,
1515 copy, fraggap, skb) < 0) {
1522 length -= datalen - fraggap;
1527 if ((flags & MSG_CONFIRM) && !skb_prev)
1528 skb_set_dst_pending_confirm(skb, 1);
1531 * Put the packet on the pending queue
1533 __skb_queue_tail(queue, skb);
1540 if (!(rt->dst.dev->features&NETIF_F_SG) &&
1541 skb_tailroom(skb) >= copy) {
1545 if (getfrag(from, skb_put(skb, copy),
1546 offset, copy, off, skb) < 0) {
1547 __skb_trim(skb, off);
1552 int i = skb_shinfo(skb)->nr_frags;
1555 if (!sk_page_frag_refill(sk, pfrag))
1558 if (!skb_can_coalesce(skb, i, pfrag->page,
1561 if (i == MAX_SKB_FRAGS)
1564 __skb_fill_page_desc(skb, i, pfrag->page,
1566 skb_shinfo(skb)->nr_frags = ++i;
1567 get_page(pfrag->page);
1569 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1571 page_address(pfrag->page) + pfrag->offset,
1572 offset, copy, skb->len, skb) < 0)
1575 pfrag->offset += copy;
1576 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1578 skb->data_len += copy;
1579 skb->truesize += copy;
1580 refcount_add(copy, &sk->sk_wmem_alloc);
1591 cork->length -= length;
1592 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1596 int ip6_append_data(struct sock *sk,
1597 int getfrag(void *from, char *to, int offset, int len,
1598 int odd, struct sk_buff *skb),
1599 void *from, int length, int transhdrlen,
1600 struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1601 struct rt6_info *rt, unsigned int flags,
1602 const struct sockcm_cookie *sockc)
1604 struct inet_sock *inet = inet_sk(sk);
1605 struct ipv6_pinfo *np = inet6_sk(sk);
1609 if (flags&MSG_PROBE)
1611 if (skb_queue_empty(&sk->sk_write_queue)) {
1615 err = ip6_setup_cork(sk, &inet->cork, &np->cork,
1620 exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1621 length += exthdrlen;
1622 transhdrlen += exthdrlen;
1624 fl6 = &inet->cork.fl.u.ip6;
1628 return __ip6_append_data(sk, fl6, &sk->sk_write_queue, &inet->cork.base,
1629 &np->cork, sk_page_frag(sk), getfrag,
1630 from, length, transhdrlen, flags, ipc6, sockc);
1632 EXPORT_SYMBOL_GPL(ip6_append_data);
1634 static void ip6_cork_release(struct inet_cork_full *cork,
1635 struct inet6_cork *v6_cork)
1638 kfree(v6_cork->opt->dst0opt);
1639 kfree(v6_cork->opt->dst1opt);
1640 kfree(v6_cork->opt->hopopt);
1641 kfree(v6_cork->opt->srcrt);
1642 kfree(v6_cork->opt);
1643 v6_cork->opt = NULL;
1646 if (cork->base.dst) {
1647 dst_release(cork->base.dst);
1648 cork->base.dst = NULL;
1649 cork->base.flags &= ~IPCORK_ALLFRAG;
1651 memset(&cork->fl, 0, sizeof(cork->fl));
1654 struct sk_buff *__ip6_make_skb(struct sock *sk,
1655 struct sk_buff_head *queue,
1656 struct inet_cork_full *cork,
1657 struct inet6_cork *v6_cork)
1659 struct sk_buff *skb, *tmp_skb;
1660 struct sk_buff **tail_skb;
1661 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1662 struct ipv6_pinfo *np = inet6_sk(sk);
1663 struct net *net = sock_net(sk);
1664 struct ipv6hdr *hdr;
1665 struct ipv6_txoptions *opt = v6_cork->opt;
1666 struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1667 struct flowi6 *fl6 = &cork->fl.u.ip6;
1668 unsigned char proto = fl6->flowi6_proto;
1670 skb = __skb_dequeue(queue);
1673 tail_skb = &(skb_shinfo(skb)->frag_list);
1675 /* move skb->data to ip header from ext header */
1676 if (skb->data < skb_network_header(skb))
1677 __skb_pull(skb, skb_network_offset(skb));
1678 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1679 __skb_pull(tmp_skb, skb_network_header_len(skb));
1680 *tail_skb = tmp_skb;
1681 tail_skb = &(tmp_skb->next);
1682 skb->len += tmp_skb->len;
1683 skb->data_len += tmp_skb->len;
1684 skb->truesize += tmp_skb->truesize;
1685 tmp_skb->destructor = NULL;
1689 /* Allow local fragmentation. */
1690 skb->ignore_df = ip6_sk_ignore_df(sk);
1692 *final_dst = fl6->daddr;
1693 __skb_pull(skb, skb_network_header_len(skb));
1694 if (opt && opt->opt_flen)
1695 ipv6_push_frag_opts(skb, opt, &proto);
1696 if (opt && opt->opt_nflen)
1697 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr);
1699 skb_push(skb, sizeof(struct ipv6hdr));
1700 skb_reset_network_header(skb);
1701 hdr = ipv6_hdr(skb);
1703 ip6_flow_hdr(hdr, v6_cork->tclass,
1704 ip6_make_flowlabel(net, skb, fl6->flowlabel,
1705 ip6_autoflowlabel(net, np), fl6));
1706 hdr->hop_limit = v6_cork->hop_limit;
1707 hdr->nexthdr = proto;
1708 hdr->saddr = fl6->saddr;
1709 hdr->daddr = *final_dst;
1711 skb->priority = sk->sk_priority;
1712 skb->mark = sk->sk_mark;
1714 skb_dst_set(skb, dst_clone(&rt->dst));
1715 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1716 if (proto == IPPROTO_ICMPV6) {
1717 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1719 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1720 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1723 ip6_cork_release(cork, v6_cork);
1728 int ip6_send_skb(struct sk_buff *skb)
1730 struct net *net = sock_net(skb->sk);
1731 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1734 err = ip6_local_out(net, skb->sk, skb);
1737 err = net_xmit_errno(err);
1739 IP6_INC_STATS(net, rt->rt6i_idev,
1740 IPSTATS_MIB_OUTDISCARDS);
1746 int ip6_push_pending_frames(struct sock *sk)
1748 struct sk_buff *skb;
1750 skb = ip6_finish_skb(sk);
1754 return ip6_send_skb(skb);
1756 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1758 static void __ip6_flush_pending_frames(struct sock *sk,
1759 struct sk_buff_head *queue,
1760 struct inet_cork_full *cork,
1761 struct inet6_cork *v6_cork)
1763 struct sk_buff *skb;
1765 while ((skb = __skb_dequeue_tail(queue)) != NULL) {
1767 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1768 IPSTATS_MIB_OUTDISCARDS);
1772 ip6_cork_release(cork, v6_cork);
1775 void ip6_flush_pending_frames(struct sock *sk)
1777 __ip6_flush_pending_frames(sk, &sk->sk_write_queue,
1778 &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
1780 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1782 struct sk_buff *ip6_make_skb(struct sock *sk,
1783 int getfrag(void *from, char *to, int offset,
1784 int len, int odd, struct sk_buff *skb),
1785 void *from, int length, int transhdrlen,
1786 struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1787 struct rt6_info *rt, unsigned int flags,
1788 const struct sockcm_cookie *sockc)
1790 struct inet_cork_full cork;
1791 struct inet6_cork v6_cork;
1792 struct sk_buff_head queue;
1793 int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1796 if (flags & MSG_PROBE)
1799 __skb_queue_head_init(&queue);
1801 cork.base.flags = 0;
1803 cork.base.opt = NULL;
1804 cork.base.dst = NULL;
1806 err = ip6_setup_cork(sk, &cork, &v6_cork, ipc6, rt, fl6);
1808 ip6_cork_release(&cork, &v6_cork);
1809 return ERR_PTR(err);
1811 if (ipc6->dontfrag < 0)
1812 ipc6->dontfrag = inet6_sk(sk)->dontfrag;
1814 err = __ip6_append_data(sk, fl6, &queue, &cork.base, &v6_cork,
1815 ¤t->task_frag, getfrag, from,
1816 length + exthdrlen, transhdrlen + exthdrlen,
1817 flags, ipc6, sockc);
1819 __ip6_flush_pending_frames(sk, &queue, &cork, &v6_cork);
1820 return ERR_PTR(err);
1823 return __ip6_make_skb(sk, &queue, &cork, &v6_cork);