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_is_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);
131 static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
135 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
141 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
142 /* Policy lookup after SNAT yielded a new policy */
143 if (skb_dst(skb)->xfrm) {
144 IPCB(skb)->flags |= IPSKB_REROUTED;
145 return dst_output(net, sk, skb);
149 if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
150 dst_allfrag(skb_dst(skb)) ||
151 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
152 return ip6_fragment(net, sk, skb, ip6_finish_output2);
154 return ip6_finish_output2(net, sk, skb);
157 int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
159 struct net_device *dev = skb_dst(skb)->dev;
160 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
162 skb->protocol = htons(ETH_P_IPV6);
165 if (unlikely(idev->cnf.disable_ipv6)) {
166 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
171 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
172 net, sk, skb, NULL, dev,
174 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
177 bool ip6_autoflowlabel(struct net *net, const struct ipv6_pinfo *np)
179 if (!np->autoflowlabel_set)
180 return ip6_default_np_autolabel(net);
182 return np->autoflowlabel;
186 * xmit an sk_buff (used by TCP, SCTP and DCCP)
187 * Note : socket lock is not held for SYNACK packets, but might be modified
188 * by calls to skb_set_owner_w() and ipv6_local_error(),
189 * which are using proper atomic operations or spinlocks.
191 int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
192 __u32 mark, struct ipv6_txoptions *opt, int tclass)
194 struct net *net = sock_net(sk);
195 const struct ipv6_pinfo *np = inet6_sk(sk);
196 struct in6_addr *first_hop = &fl6->daddr;
197 struct dst_entry *dst = skb_dst(skb);
199 u8 proto = fl6->flowi6_proto;
200 int seg_len = skb->len;
205 unsigned int head_room;
207 /* First: exthdrs may take lots of space (~8K for now)
208 MAX_HEADER is not enough.
210 head_room = opt->opt_nflen + opt->opt_flen;
211 seg_len += head_room;
212 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
214 if (skb_headroom(skb) < head_room) {
215 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
217 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
218 IPSTATS_MIB_OUTDISCARDS);
224 /* skb_set_owner_w() changes sk->sk_wmem_alloc atomically,
225 * it is safe to call in our context (socket lock not held)
227 skb_set_owner_w(skb, (struct sock *)sk);
230 ipv6_push_frag_opts(skb, opt, &proto);
232 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop,
236 skb_push(skb, sizeof(struct ipv6hdr));
237 skb_reset_network_header(skb);
241 * Fill in the IPv6 header
244 hlimit = np->hop_limit;
246 hlimit = ip6_dst_hoplimit(dst);
248 ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
249 ip6_autoflowlabel(net, np), fl6));
251 hdr->payload_len = htons(seg_len);
252 hdr->nexthdr = proto;
253 hdr->hop_limit = hlimit;
255 hdr->saddr = fl6->saddr;
256 hdr->daddr = *first_hop;
258 skb->protocol = htons(ETH_P_IPV6);
259 skb->priority = sk->sk_priority;
263 if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
264 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
265 IPSTATS_MIB_OUT, skb->len);
267 /* if egress device is enslaved to an L3 master device pass the
268 * skb to its handler for processing
270 skb = l3mdev_ip6_out((struct sock *)sk, skb);
274 /* hooks should never assume socket lock is held.
275 * we promote our socket to non const
277 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
278 net, (struct sock *)sk, skb, NULL, dst->dev,
283 /* ipv6_local_error() does not require socket lock,
284 * we promote our socket to non const
286 ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
288 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
292 EXPORT_SYMBOL(ip6_xmit);
294 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
296 struct ip6_ra_chain *ra;
297 struct sock *last = NULL;
299 read_lock(&ip6_ra_lock);
300 for (ra = ip6_ra_chain; ra; ra = ra->next) {
301 struct sock *sk = ra->sk;
302 if (sk && ra->sel == sel &&
303 (!sk->sk_bound_dev_if ||
304 sk->sk_bound_dev_if == skb->dev->ifindex)) {
306 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
308 rawv6_rcv(last, skb2);
315 rawv6_rcv(last, skb);
316 read_unlock(&ip6_ra_lock);
319 read_unlock(&ip6_ra_lock);
323 static int ip6_forward_proxy_check(struct sk_buff *skb)
325 struct ipv6hdr *hdr = ipv6_hdr(skb);
326 u8 nexthdr = hdr->nexthdr;
330 if (ipv6_ext_hdr(nexthdr)) {
331 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
335 offset = sizeof(struct ipv6hdr);
337 if (nexthdr == IPPROTO_ICMPV6) {
338 struct icmp6hdr *icmp6;
340 if (!pskb_may_pull(skb, (skb_network_header(skb) +
341 offset + 1 - skb->data)))
344 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
346 switch (icmp6->icmp6_type) {
347 case NDISC_ROUTER_SOLICITATION:
348 case NDISC_ROUTER_ADVERTISEMENT:
349 case NDISC_NEIGHBOUR_SOLICITATION:
350 case NDISC_NEIGHBOUR_ADVERTISEMENT:
352 /* For reaction involving unicast neighbor discovery
353 * message destined to the proxied address, pass it to
363 * The proxying router can't forward traffic sent to a link-local
364 * address, so signal the sender and discard the packet. This
365 * behavior is clarified by the MIPv6 specification.
367 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
368 dst_link_failure(skb);
375 static inline int ip6_forward_finish(struct net *net, struct sock *sk,
378 struct dst_entry *dst = skb_dst(skb);
380 __IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
381 __IP6_ADD_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
383 return dst_output(net, sk, skb);
386 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
391 /* ipv6 conntrack defrag sets max_frag_size + ignore_df */
392 if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
398 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
404 int ip6_forward(struct sk_buff *skb)
406 struct inet6_dev *idev = __in6_dev_get_safely(skb->dev);
407 struct dst_entry *dst = skb_dst(skb);
408 struct ipv6hdr *hdr = ipv6_hdr(skb);
409 struct inet6_skb_parm *opt = IP6CB(skb);
410 struct net *net = dev_net(dst->dev);
413 if (net->ipv6.devconf_all->forwarding == 0)
416 if (skb->pkt_type != PACKET_HOST)
419 if (unlikely(skb->sk))
422 if (skb_warn_if_lro(skb))
425 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
426 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
430 skb_forward_csum(skb);
433 * We DO NOT make any processing on
434 * RA packets, pushing them to user level AS IS
435 * without ane WARRANTY that application will be able
436 * to interpret them. The reason is that we
437 * cannot make anything clever here.
439 * We are not end-node, so that if packet contains
440 * AH/ESP, we cannot make anything.
441 * Defragmentation also would be mistake, RA packets
442 * cannot be fragmented, because there is no warranty
443 * that different fragments will go along one path. --ANK
445 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
446 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
451 * check and decrement ttl
453 if (hdr->hop_limit <= 1) {
454 /* Force OUTPUT device used as source address */
456 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
457 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
463 /* XXX: idev->cnf.proxy_ndp? */
464 if (net->ipv6.devconf_all->proxy_ndp &&
465 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
466 int proxied = ip6_forward_proxy_check(skb);
468 return ip6_input(skb);
469 else if (proxied < 0) {
470 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
475 if (!xfrm6_route_forward(skb)) {
476 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
481 /* IPv6 specs say nothing about it, but it is clear that we cannot
482 send redirects to source routed frames.
483 We don't send redirects to frames decapsulated from IPsec.
485 if (IP6CB(skb)->iif == dst->dev->ifindex &&
486 opt->srcrt == 0 && !skb_sec_path(skb)) {
487 struct in6_addr *target = NULL;
488 struct inet_peer *peer;
492 * incoming and outgoing devices are the same
496 rt = (struct rt6_info *) dst;
497 if (rt->rt6i_flags & RTF_GATEWAY)
498 target = &rt->rt6i_gateway;
500 target = &hdr->daddr;
502 peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
504 /* Limit redirects both by destination (here)
505 and by source (inside ndisc_send_redirect)
507 if (inet_peer_xrlim_allow(peer, 1*HZ))
508 ndisc_send_redirect(skb, target);
512 int addrtype = ipv6_addr_type(&hdr->saddr);
514 /* This check is security critical. */
515 if (addrtype == IPV6_ADDR_ANY ||
516 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
518 if (addrtype & IPV6_ADDR_LINKLOCAL) {
519 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
520 ICMPV6_NOT_NEIGHBOUR, 0);
525 mtu = ip6_dst_mtu_forward(dst);
526 if (mtu < IPV6_MIN_MTU)
529 if (ip6_pkt_too_big(skb, mtu)) {
530 /* Again, force OUTPUT device used as source address */
532 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
533 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INTOOBIGERRORS);
534 __IP6_INC_STATS(net, ip6_dst_idev(dst),
535 IPSTATS_MIB_FRAGFAILS);
540 if (skb_cow(skb, dst->dev->hard_header_len)) {
541 __IP6_INC_STATS(net, ip6_dst_idev(dst),
542 IPSTATS_MIB_OUTDISCARDS);
548 /* Mangling hops number delayed to point after skb COW */
552 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
553 net, NULL, skb, skb->dev, dst->dev,
557 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
563 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
565 to->pkt_type = from->pkt_type;
566 to->priority = from->priority;
567 to->protocol = from->protocol;
569 skb_dst_set(to, dst_clone(skb_dst(from)));
571 to->mark = from->mark;
573 skb_copy_hash(to, from);
575 #ifdef CONFIG_NET_SCHED
576 to->tc_index = from->tc_index;
579 skb_copy_secmark(to, from);
582 int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
583 int (*output)(struct net *, struct sock *, struct sk_buff *))
585 struct sk_buff *frag;
586 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
587 struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
588 inet6_sk(skb->sk) : NULL;
589 struct ipv6hdr *tmp_hdr;
591 unsigned int mtu, hlen, left, len;
594 int ptr, offset = 0, err = 0;
595 u8 *prevhdr, nexthdr = 0;
597 err = ip6_find_1stfragopt(skb, &prevhdr);
603 mtu = ip6_skb_dst_mtu(skb);
605 /* We must not fragment if the socket is set to force MTU discovery
606 * or if the skb it not generated by a local socket.
608 if (unlikely(!skb->ignore_df && skb->len > mtu))
611 if (IP6CB(skb)->frag_max_size) {
612 if (IP6CB(skb)->frag_max_size > mtu)
615 /* don't send fragments larger than what we received */
616 mtu = IP6CB(skb)->frag_max_size;
617 if (mtu < IPV6_MIN_MTU)
621 if (np && np->frag_size < mtu) {
625 if (mtu < hlen + sizeof(struct frag_hdr) + 8)
627 mtu -= hlen + sizeof(struct frag_hdr);
629 frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
630 &ipv6_hdr(skb)->saddr);
632 if (skb->ip_summed == CHECKSUM_PARTIAL &&
633 (err = skb_checksum_help(skb)))
636 hroom = LL_RESERVED_SPACE(rt->dst.dev);
637 if (skb_has_frag_list(skb)) {
638 unsigned int first_len = skb_pagelen(skb);
639 struct sk_buff *frag2;
641 if (first_len - hlen > mtu ||
642 ((first_len - hlen) & 7) ||
644 skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
647 skb_walk_frags(skb, frag) {
648 /* Correct geometry. */
649 if (frag->len > mtu ||
650 ((frag->len & 7) && frag->next) ||
651 skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
652 goto slow_path_clean;
654 /* Partially cloned skb? */
655 if (skb_shared(frag))
656 goto slow_path_clean;
661 frag->destructor = sock_wfree;
663 skb->truesize -= frag->truesize;
670 *prevhdr = NEXTHDR_FRAGMENT;
671 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
676 frag = skb_shinfo(skb)->frag_list;
677 skb_frag_list_init(skb);
679 __skb_pull(skb, hlen);
680 fh = __skb_push(skb, sizeof(struct frag_hdr));
681 __skb_push(skb, hlen);
682 skb_reset_network_header(skb);
683 memcpy(skb_network_header(skb), tmp_hdr, hlen);
685 fh->nexthdr = nexthdr;
687 fh->frag_off = htons(IP6_MF);
688 fh->identification = frag_id;
690 first_len = skb_pagelen(skb);
691 skb->data_len = first_len - skb_headlen(skb);
692 skb->len = first_len;
693 ipv6_hdr(skb)->payload_len = htons(first_len -
694 sizeof(struct ipv6hdr));
697 /* Prepare header of the next frame,
698 * before previous one went down. */
700 frag->ip_summed = CHECKSUM_NONE;
701 skb_reset_transport_header(frag);
702 fh = __skb_push(frag, sizeof(struct frag_hdr));
703 __skb_push(frag, hlen);
704 skb_reset_network_header(frag);
705 memcpy(skb_network_header(frag), tmp_hdr,
707 offset += skb->len - hlen - sizeof(struct frag_hdr);
708 fh->nexthdr = nexthdr;
710 fh->frag_off = htons(offset);
712 fh->frag_off |= htons(IP6_MF);
713 fh->identification = frag_id;
714 ipv6_hdr(frag)->payload_len =
716 sizeof(struct ipv6hdr));
717 ip6_copy_metadata(frag, skb);
720 err = output(net, sk, skb);
722 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
723 IPSTATS_MIB_FRAGCREATES);
736 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
737 IPSTATS_MIB_FRAGOKS);
741 kfree_skb_list(frag);
743 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
744 IPSTATS_MIB_FRAGFAILS);
748 skb_walk_frags(skb, frag2) {
752 frag2->destructor = NULL;
753 skb->truesize += frag2->truesize;
758 left = skb->len - hlen; /* Space per frame */
759 ptr = hlen; /* Where to start from */
762 * Fragment the datagram.
765 troom = rt->dst.dev->needed_tailroom;
768 * Keep copying data until we run out.
771 u8 *fragnexthdr_offset;
774 /* IF: it doesn't fit, use 'mtu' - the data space left */
777 /* IF: we are not sending up to and including the packet end
778 then align the next start on an eight byte boundary */
783 /* Allocate buffer */
784 frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
785 hroom + troom, GFP_ATOMIC);
792 * Set up data on packet
795 ip6_copy_metadata(frag, skb);
796 skb_reserve(frag, hroom);
797 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
798 skb_reset_network_header(frag);
799 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
800 frag->transport_header = (frag->network_header + hlen +
801 sizeof(struct frag_hdr));
804 * Charge the memory for the fragment to any owner
808 skb_set_owner_w(frag, skb->sk);
811 * Copy the packet header into the new buffer.
813 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
815 fragnexthdr_offset = skb_network_header(frag);
816 fragnexthdr_offset += prevhdr - skb_network_header(skb);
817 *fragnexthdr_offset = NEXTHDR_FRAGMENT;
820 * Build fragment header.
822 fh->nexthdr = nexthdr;
824 fh->identification = frag_id;
827 * Copy a block of the IP datagram.
829 BUG_ON(skb_copy_bits(skb, ptr, skb_transport_header(frag),
833 fh->frag_off = htons(offset);
835 fh->frag_off |= htons(IP6_MF);
836 ipv6_hdr(frag)->payload_len = htons(frag->len -
837 sizeof(struct ipv6hdr));
843 * Put this fragment into the sending queue.
845 err = output(net, sk, frag);
849 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
850 IPSTATS_MIB_FRAGCREATES);
852 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
853 IPSTATS_MIB_FRAGOKS);
858 if (skb->sk && dst_allfrag(skb_dst(skb)))
859 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
861 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
865 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
866 IPSTATS_MIB_FRAGFAILS);
871 static inline int ip6_rt_check(const struct rt6key *rt_key,
872 const struct in6_addr *fl_addr,
873 const struct in6_addr *addr_cache)
875 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
876 (!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
879 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
880 struct dst_entry *dst,
881 const struct flowi6 *fl6)
883 struct ipv6_pinfo *np = inet6_sk(sk);
889 if (dst->ops->family != AF_INET6) {
894 rt = (struct rt6_info *)dst;
895 /* Yes, checking route validity in not connected
896 * case is not very simple. Take into account,
897 * that we do not support routing by source, TOS,
898 * and MSG_DONTROUTE --ANK (980726)
900 * 1. ip6_rt_check(): If route was host route,
901 * check that cached destination is current.
902 * If it is network route, we still may
903 * check its validity using saved pointer
904 * to the last used address: daddr_cache.
905 * We do not want to save whole address now,
906 * (because main consumer of this service
907 * is tcp, which has not this problem),
908 * so that the last trick works only on connected
910 * 2. oif also should be the same.
912 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
913 #ifdef CONFIG_IPV6_SUBTREES
914 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
916 (!(fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF) &&
917 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex))) {
926 static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
927 struct dst_entry **dst, struct flowi6 *fl6)
929 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
936 /* The correct way to handle this would be to do
937 * ip6_route_get_saddr, and then ip6_route_output; however,
938 * the route-specific preferred source forces the
939 * ip6_route_output call _before_ ip6_route_get_saddr.
941 * In source specific routing (no src=any default route),
942 * ip6_route_output will fail given src=any saddr, though, so
943 * that's why we try it again later.
945 if (ipv6_addr_any(&fl6->saddr) && (!*dst || !(*dst)->error)) {
946 struct fib6_info *from;
948 bool had_dst = *dst != NULL;
951 *dst = ip6_route_output(net, sk, fl6);
952 rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
955 from = rt ? rcu_dereference(rt->from) : NULL;
956 err = ip6_route_get_saddr(net, from, &fl6->daddr,
957 sk ? inet6_sk(sk)->srcprefs : 0,
962 goto out_err_release;
964 /* If we had an erroneous initial result, pretend it
965 * never existed and let the SA-enabled version take
968 if (!had_dst && (*dst)->error) {
974 flags |= RT6_LOOKUP_F_IFACE;
978 *dst = ip6_route_output_flags(net, sk, fl6, flags);
982 goto out_err_release;
984 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
986 * Here if the dst entry we've looked up
987 * has a neighbour entry that is in the INCOMPLETE
988 * state and the src address from the flow is
989 * marked as OPTIMISTIC, we release the found
990 * dst entry and replace it instead with the
991 * dst entry of the nexthop router
993 rt = (struct rt6_info *) *dst;
995 n = __ipv6_neigh_lookup_noref(rt->dst.dev,
996 rt6_nexthop(rt, &fl6->daddr));
997 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
998 rcu_read_unlock_bh();
1001 struct inet6_ifaddr *ifp;
1002 struct flowi6 fl_gw6;
1005 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1008 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1014 * We need to get the dst entry for the
1015 * default router instead
1018 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1019 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1020 *dst = ip6_route_output(net, sk, &fl_gw6);
1021 err = (*dst)->error;
1023 goto out_err_release;
1027 if (ipv6_addr_v4mapped(&fl6->saddr) &&
1028 !(ipv6_addr_v4mapped(&fl6->daddr) || ipv6_addr_any(&fl6->daddr))) {
1029 err = -EAFNOSUPPORT;
1030 goto out_err_release;
1039 if (err == -ENETUNREACH)
1040 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1045 * ip6_dst_lookup - perform route lookup on flow
1046 * @sk: socket which provides route info
1047 * @dst: pointer to dst_entry * for result
1048 * @fl6: flow to lookup
1050 * This function performs a route lookup on the given flow.
1052 * It returns zero on success, or a standard errno code on error.
1054 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1058 return ip6_dst_lookup_tail(net, sk, dst, fl6);
1060 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1063 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1064 * @sk: socket which provides route info
1065 * @fl6: flow to lookup
1066 * @final_dst: final destination address for ipsec lookup
1068 * This function performs a route lookup on the given flow.
1070 * It returns a valid dst pointer on success, or a pointer encoded
1073 struct dst_entry *ip6_dst_lookup_flow(const struct sock *sk, struct flowi6 *fl6,
1074 const struct in6_addr *final_dst)
1076 struct dst_entry *dst = NULL;
1079 err = ip6_dst_lookup_tail(sock_net(sk), sk, &dst, fl6);
1081 return ERR_PTR(err);
1083 fl6->daddr = *final_dst;
1085 return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1087 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1090 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1091 * @sk: socket which provides the dst cache and route info
1092 * @fl6: flow to lookup
1093 * @final_dst: final destination address for ipsec lookup
1094 * @connected: whether @sk is connected or not
1096 * This function performs a route lookup on the given flow with the
1097 * possibility of using the cached route in the socket if it is valid.
1098 * It will take the socket dst lock when operating on the dst cache.
1099 * As a result, this function can only be used in process context.
1101 * In addition, for a connected socket, cache the dst in the socket
1102 * if the current cache is not valid.
1104 * It returns a valid dst pointer on success, or a pointer encoded
1107 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1108 const struct in6_addr *final_dst,
1111 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1113 dst = ip6_sk_dst_check(sk, dst, fl6);
1117 dst = ip6_dst_lookup_flow(sk, fl6, final_dst);
1118 if (connected && !IS_ERR(dst))
1119 ip6_sk_dst_store_flow(sk, dst_clone(dst), fl6);
1123 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1125 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1128 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1131 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1134 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1137 static void ip6_append_data_mtu(unsigned int *mtu,
1139 unsigned int fragheaderlen,
1140 struct sk_buff *skb,
1141 struct rt6_info *rt,
1142 unsigned int orig_mtu)
1144 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1146 /* first fragment, reserve header_len */
1147 *mtu = orig_mtu - rt->dst.header_len;
1151 * this fragment is not first, the headers
1152 * space is regarded as data space.
1156 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1157 + fragheaderlen - sizeof(struct frag_hdr);
1161 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1162 struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6,
1163 struct rt6_info *rt, struct flowi6 *fl6)
1165 struct ipv6_pinfo *np = inet6_sk(sk);
1167 struct ipv6_txoptions *opt = ipc6->opt;
1173 if (WARN_ON(v6_cork->opt))
1176 v6_cork->opt = kzalloc(sizeof(*opt), sk->sk_allocation);
1177 if (unlikely(!v6_cork->opt))
1180 v6_cork->opt->tot_len = sizeof(*opt);
1181 v6_cork->opt->opt_flen = opt->opt_flen;
1182 v6_cork->opt->opt_nflen = opt->opt_nflen;
1184 v6_cork->opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1186 if (opt->dst0opt && !v6_cork->opt->dst0opt)
1189 v6_cork->opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1191 if (opt->dst1opt && !v6_cork->opt->dst1opt)
1194 v6_cork->opt->hopopt = ip6_opt_dup(opt->hopopt,
1196 if (opt->hopopt && !v6_cork->opt->hopopt)
1199 v6_cork->opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1201 if (opt->srcrt && !v6_cork->opt->srcrt)
1204 /* need source address above miyazawa*/
1207 cork->base.dst = &rt->dst;
1208 cork->fl.u.ip6 = *fl6;
1209 v6_cork->hop_limit = ipc6->hlimit;
1210 v6_cork->tclass = ipc6->tclass;
1211 if (rt->dst.flags & DST_XFRM_TUNNEL)
1212 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1213 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(&rt->dst);
1215 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1216 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(xfrm_dst_path(&rt->dst));
1217 if (np->frag_size < mtu) {
1219 mtu = np->frag_size;
1221 if (mtu < IPV6_MIN_MTU)
1223 cork->base.fragsize = mtu;
1224 cork->base.gso_size = sk->sk_type == SOCK_DGRAM &&
1225 sk->sk_protocol == IPPROTO_UDP ? ipc6->gso_size : 0;
1227 if (dst_allfrag(xfrm_dst_path(&rt->dst)))
1228 cork->base.flags |= IPCORK_ALLFRAG;
1229 cork->base.length = 0;
1234 static int __ip6_append_data(struct sock *sk,
1236 struct sk_buff_head *queue,
1237 struct inet_cork *cork,
1238 struct inet6_cork *v6_cork,
1239 struct page_frag *pfrag,
1240 int getfrag(void *from, char *to, int offset,
1241 int len, int odd, struct sk_buff *skb),
1242 void *from, int length, int transhdrlen,
1243 unsigned int flags, struct ipcm6_cookie *ipc6,
1244 const struct sockcm_cookie *sockc)
1246 struct sk_buff *skb, *skb_prev = NULL;
1247 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu, pmtu;
1249 int dst_exthdrlen = 0;
1256 struct rt6_info *rt = (struct rt6_info *)cork->dst;
1257 struct ipv6_txoptions *opt = v6_cork->opt;
1258 int csummode = CHECKSUM_NONE;
1259 unsigned int maxnonfragsize, headersize;
1260 unsigned int wmem_alloc_delta = 0;
1263 skb = skb_peek_tail(queue);
1265 exthdrlen = opt ? opt->opt_flen : 0;
1266 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1269 paged = !!cork->gso_size;
1270 mtu = cork->gso_size ? IP6_MAX_MTU : cork->fragsize;
1273 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1275 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1276 (opt ? opt->opt_nflen : 0);
1277 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1278 sizeof(struct frag_hdr);
1280 headersize = sizeof(struct ipv6hdr) +
1281 (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1282 (dst_allfrag(&rt->dst) ?
1283 sizeof(struct frag_hdr) : 0) +
1284 rt->rt6i_nfheader_len;
1286 /* as per RFC 7112 section 5, the entire IPv6 Header Chain must fit
1287 * the first fragment
1289 if (headersize + transhdrlen > mtu)
1292 if (cork->length + length > mtu - headersize && ipc6->dontfrag &&
1293 (sk->sk_protocol == IPPROTO_UDP ||
1294 sk->sk_protocol == IPPROTO_RAW)) {
1295 ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1296 sizeof(struct ipv6hdr));
1300 if (ip6_sk_ignore_df(sk))
1301 maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1303 maxnonfragsize = mtu;
1305 if (cork->length + length > maxnonfragsize - headersize) {
1307 pmtu = max_t(int, mtu - headersize + sizeof(struct ipv6hdr), 0);
1308 ipv6_local_error(sk, EMSGSIZE, fl6, pmtu);
1312 /* CHECKSUM_PARTIAL only with no extension headers and when
1313 * we are not going to fragment
1315 if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1316 headersize == sizeof(struct ipv6hdr) &&
1317 length <= mtu - headersize &&
1318 (!(flags & MSG_MORE) || cork->gso_size) &&
1319 rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1320 csummode = CHECKSUM_PARTIAL;
1322 if (sk->sk_type == SOCK_DGRAM || sk->sk_type == SOCK_RAW) {
1323 sock_tx_timestamp(sk, sockc->tsflags, &tx_flags);
1324 if (tx_flags & SKBTX_ANY_SW_TSTAMP &&
1325 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1326 tskey = sk->sk_tskey++;
1330 * Let's try using as much space as possible.
1331 * Use MTU if total length of the message fits into the MTU.
1332 * Otherwise, we need to reserve fragment header and
1333 * fragment alignment (= 8-15 octects, in total).
1335 * Note that we may need to "move" the data from the tail of
1336 * of the buffer to the new fragment when we split
1339 * FIXME: It may be fragmented into multiple chunks
1340 * at once if non-fragmentable extension headers
1345 cork->length += length;
1349 while (length > 0) {
1350 /* Check if the remaining data fits into current packet. */
1351 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1353 copy = maxfraglen - skb->len;
1357 unsigned int datalen;
1358 unsigned int fraglen;
1359 unsigned int fraggap;
1360 unsigned int alloclen;
1361 unsigned int pagedlen = 0;
1363 /* There's no room in the current skb */
1365 fraggap = skb->len - maxfraglen;
1368 /* update mtu and maxfraglen if necessary */
1369 if (!skb || !skb_prev)
1370 ip6_append_data_mtu(&mtu, &maxfraglen,
1371 fragheaderlen, skb, rt,
1377 * If remaining data exceeds the mtu,
1378 * we know we need more fragment(s).
1380 datalen = length + fraggap;
1382 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1383 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1384 fraglen = datalen + fragheaderlen;
1386 if ((flags & MSG_MORE) &&
1387 !(rt->dst.dev->features&NETIF_F_SG))
1392 alloclen = min_t(int, fraglen, MAX_HEADER);
1393 pagedlen = fraglen - alloclen;
1396 alloclen += dst_exthdrlen;
1398 if (datalen != length + fraggap) {
1400 * this is not the last fragment, the trailer
1401 * space is regarded as data space.
1403 datalen += rt->dst.trailer_len;
1406 alloclen += rt->dst.trailer_len;
1407 fraglen = datalen + fragheaderlen;
1410 * We just reserve space for fragment header.
1411 * Note: this may be overallocation if the message
1412 * (without MSG_MORE) fits into the MTU.
1414 alloclen += sizeof(struct frag_hdr);
1416 copy = datalen - transhdrlen - fraggap - pagedlen;
1422 skb = sock_alloc_send_skb(sk,
1424 (flags & MSG_DONTWAIT), &err);
1427 if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1429 skb = alloc_skb(alloclen + hh_len,
1437 * Fill in the control structures
1439 skb->protocol = htons(ETH_P_IPV6);
1440 skb->ip_summed = csummode;
1442 /* reserve for fragmentation and ipsec header */
1443 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1446 /* Only the initial fragment is time stamped */
1447 skb_shinfo(skb)->tx_flags = tx_flags;
1449 skb_shinfo(skb)->tskey = tskey;
1453 * Find where to start putting bytes
1455 data = skb_put(skb, fraglen - pagedlen);
1456 skb_set_network_header(skb, exthdrlen);
1457 data += fragheaderlen;
1458 skb->transport_header = (skb->network_header +
1461 skb->csum = skb_copy_and_csum_bits(
1462 skb_prev, maxfraglen,
1463 data + transhdrlen, fraggap, 0);
1464 skb_prev->csum = csum_sub(skb_prev->csum,
1467 pskb_trim_unique(skb_prev, maxfraglen);
1470 getfrag(from, data + transhdrlen, offset,
1471 copy, fraggap, skb) < 0) {
1478 length -= copy + transhdrlen;
1483 if ((flags & MSG_CONFIRM) && !skb_prev)
1484 skb_set_dst_pending_confirm(skb, 1);
1487 * Put the packet on the pending queue
1489 if (!skb->destructor) {
1490 skb->destructor = sock_wfree;
1492 wmem_alloc_delta += skb->truesize;
1494 __skb_queue_tail(queue, skb);
1501 if (!(rt->dst.dev->features&NETIF_F_SG) &&
1502 skb_tailroom(skb) >= copy) {
1506 if (getfrag(from, skb_put(skb, copy),
1507 offset, copy, off, skb) < 0) {
1508 __skb_trim(skb, off);
1513 int i = skb_shinfo(skb)->nr_frags;
1516 if (!sk_page_frag_refill(sk, pfrag))
1519 if (!skb_can_coalesce(skb, i, pfrag->page,
1522 if (i == MAX_SKB_FRAGS)
1525 __skb_fill_page_desc(skb, i, pfrag->page,
1527 skb_shinfo(skb)->nr_frags = ++i;
1528 get_page(pfrag->page);
1530 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1532 page_address(pfrag->page) + pfrag->offset,
1533 offset, copy, skb->len, skb) < 0)
1536 pfrag->offset += copy;
1537 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1539 skb->data_len += copy;
1540 skb->truesize += copy;
1541 wmem_alloc_delta += copy;
1547 if (wmem_alloc_delta)
1548 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1554 cork->length -= length;
1555 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1556 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1560 int ip6_append_data(struct sock *sk,
1561 int getfrag(void *from, char *to, int offset, int len,
1562 int odd, struct sk_buff *skb),
1563 void *from, int length, int transhdrlen,
1564 struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1565 struct rt6_info *rt, unsigned int flags,
1566 const struct sockcm_cookie *sockc)
1568 struct inet_sock *inet = inet_sk(sk);
1569 struct ipv6_pinfo *np = inet6_sk(sk);
1573 if (flags&MSG_PROBE)
1575 if (skb_queue_empty(&sk->sk_write_queue)) {
1579 err = ip6_setup_cork(sk, &inet->cork, &np->cork,
1584 exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1585 length += exthdrlen;
1586 transhdrlen += exthdrlen;
1588 fl6 = &inet->cork.fl.u.ip6;
1592 return __ip6_append_data(sk, fl6, &sk->sk_write_queue, &inet->cork.base,
1593 &np->cork, sk_page_frag(sk), getfrag,
1594 from, length, transhdrlen, flags, ipc6, sockc);
1596 EXPORT_SYMBOL_GPL(ip6_append_data);
1598 static void ip6_cork_release(struct inet_cork_full *cork,
1599 struct inet6_cork *v6_cork)
1602 kfree(v6_cork->opt->dst0opt);
1603 kfree(v6_cork->opt->dst1opt);
1604 kfree(v6_cork->opt->hopopt);
1605 kfree(v6_cork->opt->srcrt);
1606 kfree(v6_cork->opt);
1607 v6_cork->opt = NULL;
1610 if (cork->base.dst) {
1611 dst_release(cork->base.dst);
1612 cork->base.dst = NULL;
1613 cork->base.flags &= ~IPCORK_ALLFRAG;
1615 memset(&cork->fl, 0, sizeof(cork->fl));
1618 struct sk_buff *__ip6_make_skb(struct sock *sk,
1619 struct sk_buff_head *queue,
1620 struct inet_cork_full *cork,
1621 struct inet6_cork *v6_cork)
1623 struct sk_buff *skb, *tmp_skb;
1624 struct sk_buff **tail_skb;
1625 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1626 struct ipv6_pinfo *np = inet6_sk(sk);
1627 struct net *net = sock_net(sk);
1628 struct ipv6hdr *hdr;
1629 struct ipv6_txoptions *opt = v6_cork->opt;
1630 struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1631 struct flowi6 *fl6 = &cork->fl.u.ip6;
1632 unsigned char proto = fl6->flowi6_proto;
1634 skb = __skb_dequeue(queue);
1637 tail_skb = &(skb_shinfo(skb)->frag_list);
1639 /* move skb->data to ip header from ext header */
1640 if (skb->data < skb_network_header(skb))
1641 __skb_pull(skb, skb_network_offset(skb));
1642 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1643 __skb_pull(tmp_skb, skb_network_header_len(skb));
1644 *tail_skb = tmp_skb;
1645 tail_skb = &(tmp_skb->next);
1646 skb->len += tmp_skb->len;
1647 skb->data_len += tmp_skb->len;
1648 skb->truesize += tmp_skb->truesize;
1649 tmp_skb->destructor = NULL;
1653 /* Allow local fragmentation. */
1654 skb->ignore_df = ip6_sk_ignore_df(sk);
1656 *final_dst = fl6->daddr;
1657 __skb_pull(skb, skb_network_header_len(skb));
1658 if (opt && opt->opt_flen)
1659 ipv6_push_frag_opts(skb, opt, &proto);
1660 if (opt && opt->opt_nflen)
1661 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr);
1663 skb_push(skb, sizeof(struct ipv6hdr));
1664 skb_reset_network_header(skb);
1665 hdr = ipv6_hdr(skb);
1667 ip6_flow_hdr(hdr, v6_cork->tclass,
1668 ip6_make_flowlabel(net, skb, fl6->flowlabel,
1669 ip6_autoflowlabel(net, np), fl6));
1670 hdr->hop_limit = v6_cork->hop_limit;
1671 hdr->nexthdr = proto;
1672 hdr->saddr = fl6->saddr;
1673 hdr->daddr = *final_dst;
1675 skb->priority = sk->sk_priority;
1676 skb->mark = sk->sk_mark;
1678 skb_dst_set(skb, dst_clone(&rt->dst));
1679 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1680 if (proto == IPPROTO_ICMPV6) {
1681 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1683 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1684 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1687 ip6_cork_release(cork, v6_cork);
1692 int ip6_send_skb(struct sk_buff *skb)
1694 struct net *net = sock_net(skb->sk);
1695 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1698 err = ip6_local_out(net, skb->sk, skb);
1701 err = net_xmit_errno(err);
1703 IP6_INC_STATS(net, rt->rt6i_idev,
1704 IPSTATS_MIB_OUTDISCARDS);
1710 int ip6_push_pending_frames(struct sock *sk)
1712 struct sk_buff *skb;
1714 skb = ip6_finish_skb(sk);
1718 return ip6_send_skb(skb);
1720 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1722 static void __ip6_flush_pending_frames(struct sock *sk,
1723 struct sk_buff_head *queue,
1724 struct inet_cork_full *cork,
1725 struct inet6_cork *v6_cork)
1727 struct sk_buff *skb;
1729 while ((skb = __skb_dequeue_tail(queue)) != NULL) {
1731 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1732 IPSTATS_MIB_OUTDISCARDS);
1736 ip6_cork_release(cork, v6_cork);
1739 void ip6_flush_pending_frames(struct sock *sk)
1741 __ip6_flush_pending_frames(sk, &sk->sk_write_queue,
1742 &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
1744 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1746 struct sk_buff *ip6_make_skb(struct sock *sk,
1747 int getfrag(void *from, char *to, int offset,
1748 int len, int odd, struct sk_buff *skb),
1749 void *from, int length, int transhdrlen,
1750 struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1751 struct rt6_info *rt, unsigned int flags,
1752 struct inet_cork_full *cork,
1753 const struct sockcm_cookie *sockc)
1755 struct inet6_cork v6_cork;
1756 struct sk_buff_head queue;
1757 int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1760 if (flags & MSG_PROBE)
1763 __skb_queue_head_init(&queue);
1765 cork->base.flags = 0;
1766 cork->base.addr = 0;
1767 cork->base.opt = NULL;
1768 cork->base.dst = NULL;
1770 err = ip6_setup_cork(sk, cork, &v6_cork, ipc6, rt, fl6);
1772 ip6_cork_release(cork, &v6_cork);
1773 return ERR_PTR(err);
1775 if (ipc6->dontfrag < 0)
1776 ipc6->dontfrag = inet6_sk(sk)->dontfrag;
1778 err = __ip6_append_data(sk, fl6, &queue, &cork->base, &v6_cork,
1779 ¤t->task_frag, getfrag, from,
1780 length + exthdrlen, transhdrlen + exthdrlen,
1781 flags, ipc6, sockc);
1783 __ip6_flush_pending_frames(sk, &queue, cork, &v6_cork);
1784 return ERR_PTR(err);
1787 return __ip6_make_skb(sk, &queue, cork, &v6_cork);