2 * IPv6 output functions
3 * Linux INET6 implementation
6 * Pedro Roque <roque@di.fc.ul.pt>
8 * Based on linux/net/ipv4/ip_output.c
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
16 * A.N.Kuznetsov : airthmetics in fragmentation.
17 * extension headers are implemented.
18 * route changes now work.
19 * ip6_forward does not confuse sniffers.
22 * H. von Brand : Added missing #include <linux/string.h>
23 * Imran Patel : frag id should be in NBO
24 * Kazunori MIYAZAWA @USAGI
25 * : add ip6_append_data and related functions
29 #include <linux/errno.h>
30 #include <linux/kernel.h>
31 #include <linux/string.h>
32 #include <linux/socket.h>
33 #include <linux/net.h>
34 #include <linux/netdevice.h>
35 #include <linux/if_arp.h>
36 #include <linux/in6.h>
37 #include <linux/tcp.h>
38 #include <linux/route.h>
39 #include <linux/module.h>
40 #include <linux/slab.h>
42 #include <linux/netfilter.h>
43 #include <linux/netfilter_ipv6.h>
49 #include <net/ndisc.h>
50 #include <net/protocol.h>
51 #include <net/ip6_route.h>
52 #include <net/addrconf.h>
53 #include <net/rawv6.h>
56 #include <net/checksum.h>
57 #include <linux/mroute6.h>
59 static int ip6_finish_output2(struct sk_buff *skb)
61 struct dst_entry *dst = skb_dst(skb);
62 struct net_device *dev = dst->dev;
63 struct neighbour *neigh;
64 struct in6_addr *nexthop;
67 skb->protocol = htons(ETH_P_IPV6);
70 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
71 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
73 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(skb->sk) &&
74 ((mroute6_socket(dev_net(dev), 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 newskb, NULL, newskb->dev,
88 if (ipv6_hdr(skb)->hop_limit == 0) {
89 IP6_INC_STATS(dev_net(dev), idev,
90 IPSTATS_MIB_OUTDISCARDS);
96 IP6_UPD_PO_STATS(dev_net(dev), idev, IPSTATS_MIB_OUTMCAST,
99 if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb)->daddr) <=
100 IPV6_ADDR_SCOPE_NODELOCAL &&
101 !(dev->flags & IFF_LOOPBACK)) {
108 nexthop = rt6_nexthop((struct rt6_info *)dst);
109 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
110 if (unlikely(!neigh))
111 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
112 if (!IS_ERR(neigh)) {
113 ret = dst_neigh_output(dst, neigh, skb);
114 rcu_read_unlock_bh();
117 rcu_read_unlock_bh();
119 IP6_INC_STATS(dev_net(dst->dev),
120 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
125 static int ip6_finish_output(struct sk_buff *skb)
127 if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
128 dst_allfrag(skb_dst(skb)) ||
129 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
130 return ip6_fragment(skb, ip6_finish_output2);
132 return ip6_finish_output2(skb);
135 int ip6_output(struct sock *sk, struct sk_buff *skb)
137 struct net_device *dev = skb_dst(skb)->dev;
138 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
139 if (unlikely(idev->cnf.disable_ipv6)) {
140 IP6_INC_STATS(dev_net(dev), idev,
141 IPSTATS_MIB_OUTDISCARDS);
146 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, skb, NULL, dev,
148 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
152 * xmit an sk_buff (used by TCP, SCTP and DCCP)
155 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
156 struct ipv6_txoptions *opt, int tclass)
158 struct net *net = sock_net(sk);
159 struct ipv6_pinfo *np = inet6_sk(sk);
160 struct in6_addr *first_hop = &fl6->daddr;
161 struct dst_entry *dst = skb_dst(skb);
163 u8 proto = fl6->flowi6_proto;
164 int seg_len = skb->len;
169 unsigned int head_room;
171 /* First: exthdrs may take lots of space (~8K for now)
172 MAX_HEADER is not enough.
174 head_room = opt->opt_nflen + opt->opt_flen;
175 seg_len += head_room;
176 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
178 if (skb_headroom(skb) < head_room) {
179 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
181 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
182 IPSTATS_MIB_OUTDISCARDS);
188 skb_set_owner_w(skb, sk);
191 ipv6_push_frag_opts(skb, opt, &proto);
193 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
196 skb_push(skb, sizeof(struct ipv6hdr));
197 skb_reset_network_header(skb);
201 * Fill in the IPv6 header
204 hlimit = np->hop_limit;
206 hlimit = ip6_dst_hoplimit(dst);
208 ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
211 hdr->payload_len = htons(seg_len);
212 hdr->nexthdr = proto;
213 hdr->hop_limit = hlimit;
215 hdr->saddr = fl6->saddr;
216 hdr->daddr = *first_hop;
218 skb->protocol = htons(ETH_P_IPV6);
219 skb->priority = sk->sk_priority;
220 skb->mark = sk->sk_mark;
223 if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
224 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
225 IPSTATS_MIB_OUT, skb->len);
226 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, skb, NULL,
227 dst->dev, dst_output);
231 ipv6_local_error(sk, EMSGSIZE, fl6, mtu);
232 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
236 EXPORT_SYMBOL(ip6_xmit);
238 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
240 struct ip6_ra_chain *ra;
241 struct sock *last = NULL;
243 read_lock(&ip6_ra_lock);
244 for (ra = ip6_ra_chain; ra; ra = ra->next) {
245 struct sock *sk = ra->sk;
246 if (sk && ra->sel == sel &&
247 (!sk->sk_bound_dev_if ||
248 sk->sk_bound_dev_if == skb->dev->ifindex)) {
250 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
252 rawv6_rcv(last, skb2);
259 rawv6_rcv(last, skb);
260 read_unlock(&ip6_ra_lock);
263 read_unlock(&ip6_ra_lock);
267 static int ip6_forward_proxy_check(struct sk_buff *skb)
269 struct ipv6hdr *hdr = ipv6_hdr(skb);
270 u8 nexthdr = hdr->nexthdr;
274 if (ipv6_ext_hdr(nexthdr)) {
275 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
279 offset = sizeof(struct ipv6hdr);
281 if (nexthdr == IPPROTO_ICMPV6) {
282 struct icmp6hdr *icmp6;
284 if (!pskb_may_pull(skb, (skb_network_header(skb) +
285 offset + 1 - skb->data)))
288 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
290 switch (icmp6->icmp6_type) {
291 case NDISC_ROUTER_SOLICITATION:
292 case NDISC_ROUTER_ADVERTISEMENT:
293 case NDISC_NEIGHBOUR_SOLICITATION:
294 case NDISC_NEIGHBOUR_ADVERTISEMENT:
296 /* For reaction involving unicast neighbor discovery
297 * message destined to the proxied address, pass it to
307 * The proxying router can't forward traffic sent to a link-local
308 * address, so signal the sender and discard the packet. This
309 * behavior is clarified by the MIPv6 specification.
311 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
312 dst_link_failure(skb);
319 static inline int ip6_forward_finish(struct sk_buff *skb)
321 return dst_output(skb);
324 static unsigned int ip6_dst_mtu_forward(const struct dst_entry *dst)
327 struct inet6_dev *idev;
329 if (dst_metric_locked(dst, RTAX_MTU)) {
330 mtu = dst_metric_raw(dst, RTAX_MTU);
337 idev = __in6_dev_get(dst->dev);
339 mtu = idev->cnf.mtu6;
345 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
350 /* ipv6 conntrack defrag sets max_frag_size + ignore_df */
351 if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
357 if (skb_is_gso(skb) && skb_gso_network_seglen(skb) <= mtu)
363 int ip6_forward(struct sk_buff *skb)
365 struct dst_entry *dst = skb_dst(skb);
366 struct ipv6hdr *hdr = ipv6_hdr(skb);
367 struct inet6_skb_parm *opt = IP6CB(skb);
368 struct net *net = dev_net(dst->dev);
371 if (net->ipv6.devconf_all->forwarding == 0)
374 if (skb->pkt_type != PACKET_HOST)
377 if (skb_warn_if_lro(skb))
380 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
381 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
382 IPSTATS_MIB_INDISCARDS);
386 skb_forward_csum(skb);
389 * We DO NOT make any processing on
390 * RA packets, pushing them to user level AS IS
391 * without ane WARRANTY that application will be able
392 * to interpret them. The reason is that we
393 * cannot make anything clever here.
395 * We are not end-node, so that if packet contains
396 * AH/ESP, we cannot make anything.
397 * Defragmentation also would be mistake, RA packets
398 * cannot be fragmented, because there is no warranty
399 * that different fragments will go along one path. --ANK
401 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
402 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
407 * check and decrement ttl
409 if (hdr->hop_limit <= 1) {
410 /* Force OUTPUT device used as source address */
412 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
413 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
414 IPSTATS_MIB_INHDRERRORS);
420 /* XXX: idev->cnf.proxy_ndp? */
421 if (net->ipv6.devconf_all->proxy_ndp &&
422 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
423 int proxied = ip6_forward_proxy_check(skb);
425 return ip6_input(skb);
426 else if (proxied < 0) {
427 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
428 IPSTATS_MIB_INDISCARDS);
433 if (!xfrm6_route_forward(skb)) {
434 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
435 IPSTATS_MIB_INDISCARDS);
440 /* IPv6 specs say nothing about it, but it is clear that we cannot
441 send redirects to source routed frames.
442 We don't send redirects to frames decapsulated from IPsec.
444 if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
445 struct in6_addr *target = NULL;
446 struct inet_peer *peer;
450 * incoming and outgoing devices are the same
454 rt = (struct rt6_info *) dst;
455 if (rt->rt6i_flags & RTF_GATEWAY)
456 target = &rt->rt6i_gateway;
458 target = &hdr->daddr;
460 peer = inet_getpeer_v6(net->ipv6.peers, &rt->rt6i_dst.addr, 1);
462 /* Limit redirects both by destination (here)
463 and by source (inside ndisc_send_redirect)
465 if (inet_peer_xrlim_allow(peer, 1*HZ))
466 ndisc_send_redirect(skb, target);
470 int addrtype = ipv6_addr_type(&hdr->saddr);
472 /* This check is security critical. */
473 if (addrtype == IPV6_ADDR_ANY ||
474 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
476 if (addrtype & IPV6_ADDR_LINKLOCAL) {
477 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
478 ICMPV6_NOT_NEIGHBOUR, 0);
483 mtu = ip6_dst_mtu_forward(dst);
484 if (mtu < IPV6_MIN_MTU)
487 if (ip6_pkt_too_big(skb, mtu)) {
488 /* Again, force OUTPUT device used as source address */
490 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
491 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
492 IPSTATS_MIB_INTOOBIGERRORS);
493 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
494 IPSTATS_MIB_FRAGFAILS);
499 if (skb_cow(skb, dst->dev->hard_header_len)) {
500 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
501 IPSTATS_MIB_OUTDISCARDS);
507 /* Mangling hops number delayed to point after skb COW */
511 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
512 IP6_ADD_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
513 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD, skb, skb->dev, dst->dev,
517 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
523 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
525 to->pkt_type = from->pkt_type;
526 to->priority = from->priority;
527 to->protocol = from->protocol;
529 skb_dst_set(to, dst_clone(skb_dst(from)));
531 to->mark = from->mark;
533 #ifdef CONFIG_NET_SCHED
534 to->tc_index = from->tc_index;
537 skb_copy_secmark(to, from);
540 int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
542 struct sk_buff *frag;
543 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
544 struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
545 struct ipv6hdr *tmp_hdr;
547 unsigned int mtu, hlen, left, len;
550 int ptr, offset = 0, err = 0;
551 u8 *prevhdr, nexthdr = 0;
552 struct net *net = dev_net(skb_dst(skb)->dev);
554 hlen = ip6_find_1stfragopt(skb, &prevhdr);
557 mtu = ip6_skb_dst_mtu(skb);
559 /* We must not fragment if the socket is set to force MTU discovery
560 * or if the skb it not generated by a local socket.
562 if (unlikely(!skb->ignore_df && skb->len > mtu) ||
563 (IP6CB(skb)->frag_max_size &&
564 IP6CB(skb)->frag_max_size > mtu)) {
565 if (skb->sk && dst_allfrag(skb_dst(skb)))
566 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
568 skb->dev = skb_dst(skb)->dev;
569 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
570 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
571 IPSTATS_MIB_FRAGFAILS);
576 if (np && np->frag_size < mtu) {
580 mtu -= hlen + sizeof(struct frag_hdr);
582 if (skb_has_frag_list(skb)) {
583 int first_len = skb_pagelen(skb);
584 struct sk_buff *frag2;
586 if (first_len - hlen > mtu ||
587 ((first_len - hlen) & 7) ||
591 skb_walk_frags(skb, frag) {
592 /* Correct geometry. */
593 if (frag->len > mtu ||
594 ((frag->len & 7) && frag->next) ||
595 skb_headroom(frag) < hlen)
596 goto slow_path_clean;
598 /* Partially cloned skb? */
599 if (skb_shared(frag))
600 goto slow_path_clean;
605 frag->destructor = sock_wfree;
607 skb->truesize -= frag->truesize;
612 frag = skb_shinfo(skb)->frag_list;
613 skb_frag_list_init(skb);
616 *prevhdr = NEXTHDR_FRAGMENT;
617 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
619 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
620 IPSTATS_MIB_FRAGFAILS);
624 __skb_pull(skb, hlen);
625 fh = (struct frag_hdr *)__skb_push(skb, sizeof(struct frag_hdr));
626 __skb_push(skb, hlen);
627 skb_reset_network_header(skb);
628 memcpy(skb_network_header(skb), tmp_hdr, hlen);
630 ipv6_select_ident(fh, rt);
631 fh->nexthdr = nexthdr;
633 fh->frag_off = htons(IP6_MF);
634 frag_id = fh->identification;
636 first_len = skb_pagelen(skb);
637 skb->data_len = first_len - skb_headlen(skb);
638 skb->len = first_len;
639 ipv6_hdr(skb)->payload_len = htons(first_len -
640 sizeof(struct ipv6hdr));
645 /* Prepare header of the next frame,
646 * before previous one went down. */
648 frag->ip_summed = CHECKSUM_NONE;
649 skb_reset_transport_header(frag);
650 fh = (struct frag_hdr *)__skb_push(frag, sizeof(struct frag_hdr));
651 __skb_push(frag, hlen);
652 skb_reset_network_header(frag);
653 memcpy(skb_network_header(frag), tmp_hdr,
655 offset += skb->len - hlen - sizeof(struct frag_hdr);
656 fh->nexthdr = nexthdr;
658 fh->frag_off = htons(offset);
659 if (frag->next != NULL)
660 fh->frag_off |= htons(IP6_MF);
661 fh->identification = frag_id;
662 ipv6_hdr(frag)->payload_len =
664 sizeof(struct ipv6hdr));
665 ip6_copy_metadata(frag, skb);
670 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
671 IPSTATS_MIB_FRAGCREATES);
684 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
685 IPSTATS_MIB_FRAGOKS);
690 kfree_skb_list(frag);
692 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
693 IPSTATS_MIB_FRAGFAILS);
698 skb_walk_frags(skb, frag2) {
702 frag2->destructor = NULL;
703 skb->truesize += frag2->truesize;
708 if ((skb->ip_summed == CHECKSUM_PARTIAL) &&
709 skb_checksum_help(skb))
712 left = skb->len - hlen; /* Space per frame */
713 ptr = hlen; /* Where to start from */
716 * Fragment the datagram.
719 *prevhdr = NEXTHDR_FRAGMENT;
720 hroom = LL_RESERVED_SPACE(rt->dst.dev);
721 troom = rt->dst.dev->needed_tailroom;
724 * Keep copying data until we run out.
728 /* IF: it doesn't fit, use 'mtu' - the data space left */
731 /* IF: we are not sending up to and including the packet end
732 then align the next start on an eight byte boundary */
737 /* Allocate buffer */
738 frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
739 hroom + troom, GFP_ATOMIC);
741 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
742 IPSTATS_MIB_FRAGFAILS);
748 * Set up data on packet
751 ip6_copy_metadata(frag, skb);
752 skb_reserve(frag, hroom);
753 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
754 skb_reset_network_header(frag);
755 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
756 frag->transport_header = (frag->network_header + hlen +
757 sizeof(struct frag_hdr));
760 * Charge the memory for the fragment to any owner
764 skb_set_owner_w(frag, skb->sk);
767 * Copy the packet header into the new buffer.
769 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
772 * Build fragment header.
774 fh->nexthdr = nexthdr;
777 ipv6_select_ident(fh, rt);
778 frag_id = fh->identification;
780 fh->identification = frag_id;
783 * Copy a block of the IP datagram.
785 BUG_ON(skb_copy_bits(skb, ptr, skb_transport_header(frag),
789 fh->frag_off = htons(offset);
791 fh->frag_off |= htons(IP6_MF);
792 ipv6_hdr(frag)->payload_len = htons(frag->len -
793 sizeof(struct ipv6hdr));
799 * Put this fragment into the sending queue.
805 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
806 IPSTATS_MIB_FRAGCREATES);
808 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
809 IPSTATS_MIB_FRAGOKS);
814 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
815 IPSTATS_MIB_FRAGFAILS);
820 static inline int ip6_rt_check(const struct rt6key *rt_key,
821 const struct in6_addr *fl_addr,
822 const struct in6_addr *addr_cache)
824 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
825 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache));
828 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
829 struct dst_entry *dst,
830 const struct flowi6 *fl6)
832 struct ipv6_pinfo *np = inet6_sk(sk);
838 if (dst->ops->family != AF_INET6) {
843 rt = (struct rt6_info *)dst;
844 /* Yes, checking route validity in not connected
845 * case is not very simple. Take into account,
846 * that we do not support routing by source, TOS,
847 * and MSG_DONTROUTE --ANK (980726)
849 * 1. ip6_rt_check(): If route was host route,
850 * check that cached destination is current.
851 * If it is network route, we still may
852 * check its validity using saved pointer
853 * to the last used address: daddr_cache.
854 * We do not want to save whole address now,
855 * (because main consumer of this service
856 * is tcp, which has not this problem),
857 * so that the last trick works only on connected
859 * 2. oif also should be the same.
861 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
862 #ifdef CONFIG_IPV6_SUBTREES
863 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
865 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
874 static int ip6_dst_lookup_tail(struct sock *sk,
875 struct dst_entry **dst, struct flowi6 *fl6)
877 struct net *net = sock_net(sk);
878 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
885 *dst = ip6_route_output(net, sk, fl6);
889 goto out_err_release;
891 if (ipv6_addr_any(&fl6->saddr)) {
892 struct rt6_info *rt = (struct rt6_info *) *dst;
893 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
894 sk ? inet6_sk(sk)->srcprefs : 0,
897 goto out_err_release;
900 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
902 * Here if the dst entry we've looked up
903 * has a neighbour entry that is in the INCOMPLETE
904 * state and the src address from the flow is
905 * marked as OPTIMISTIC, we release the found
906 * dst entry and replace it instead with the
907 * dst entry of the nexthop router
909 rt = (struct rt6_info *) *dst;
911 n = __ipv6_neigh_lookup_noref(rt->dst.dev, rt6_nexthop(rt));
912 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
913 rcu_read_unlock_bh();
916 struct inet6_ifaddr *ifp;
917 struct flowi6 fl_gw6;
920 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
923 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
929 * We need to get the dst entry for the
930 * default router instead
933 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
934 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
935 *dst = ip6_route_output(net, sk, &fl_gw6);
938 goto out_err_release;
946 if (err == -ENETUNREACH)
947 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
954 * ip6_dst_lookup - perform route lookup on flow
955 * @sk: socket which provides route info
956 * @dst: pointer to dst_entry * for result
957 * @fl6: flow to lookup
959 * This function performs a route lookup on the given flow.
961 * It returns zero on success, or a standard errno code on error.
963 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi6 *fl6)
966 return ip6_dst_lookup_tail(sk, dst, fl6);
968 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
971 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
972 * @sk: socket which provides route info
973 * @fl6: flow to lookup
974 * @final_dst: final destination address for ipsec lookup
976 * This function performs a route lookup on the given flow.
978 * It returns a valid dst pointer on success, or a pointer encoded
981 struct dst_entry *ip6_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
982 const struct in6_addr *final_dst)
984 struct dst_entry *dst = NULL;
987 err = ip6_dst_lookup_tail(sk, &dst, fl6);
991 fl6->daddr = *final_dst;
993 return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
995 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
998 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
999 * @sk: socket which provides the dst cache and route info
1000 * @fl6: flow to lookup
1001 * @final_dst: final destination address for ipsec lookup
1003 * This function performs a route lookup on the given flow with the
1004 * possibility of using the cached route in the socket if it is valid.
1005 * It will take the socket dst lock when operating on the dst cache.
1006 * As a result, this function can only be used in process context.
1008 * It returns a valid dst pointer on success, or a pointer encoded
1011 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1012 const struct in6_addr *final_dst)
1014 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1017 dst = ip6_sk_dst_check(sk, dst, fl6);
1019 err = ip6_dst_lookup_tail(sk, &dst, fl6);
1021 return ERR_PTR(err);
1023 fl6->daddr = *final_dst;
1025 return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1027 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1029 static inline int ip6_ufo_append_data(struct sock *sk,
1030 int getfrag(void *from, char *to, int offset, int len,
1031 int odd, struct sk_buff *skb),
1032 void *from, int length, int hh_len, int fragheaderlen,
1033 int transhdrlen, int mtu, unsigned int flags,
1034 struct rt6_info *rt)
1037 struct sk_buff *skb;
1038 struct frag_hdr fhdr;
1041 /* There is support for UDP large send offload by network
1042 * device, so create one single skb packet containing complete
1045 skb = skb_peek_tail(&sk->sk_write_queue);
1047 skb = sock_alloc_send_skb(sk,
1048 hh_len + fragheaderlen + transhdrlen + 20,
1049 (flags & MSG_DONTWAIT), &err);
1053 /* reserve space for Hardware header */
1054 skb_reserve(skb, hh_len);
1056 /* create space for UDP/IP header */
1057 skb_put(skb, fragheaderlen + transhdrlen);
1059 /* initialize network header pointer */
1060 skb_reset_network_header(skb);
1062 /* initialize protocol header pointer */
1063 skb->transport_header = skb->network_header + fragheaderlen;
1065 skb->protocol = htons(ETH_P_IPV6);
1068 __skb_queue_tail(&sk->sk_write_queue, skb);
1069 } else if (skb_is_gso(skb)) {
1073 skb->ip_summed = CHECKSUM_PARTIAL;
1074 /* Specify the length of each IPv6 datagram fragment.
1075 * It has to be a multiple of 8.
1077 skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1078 sizeof(struct frag_hdr)) & ~7;
1079 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1080 ipv6_select_ident(&fhdr, rt);
1081 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1084 return skb_append_datato_frags(sk, skb, getfrag, from,
1085 (length - transhdrlen));
1088 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1091 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1094 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1097 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1100 static void ip6_append_data_mtu(unsigned int *mtu,
1102 unsigned int fragheaderlen,
1103 struct sk_buff *skb,
1104 struct rt6_info *rt,
1105 unsigned int orig_mtu)
1107 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1109 /* first fragment, reserve header_len */
1110 *mtu = orig_mtu - rt->dst.header_len;
1114 * this fragment is not first, the headers
1115 * space is regarded as data space.
1119 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1120 + fragheaderlen - sizeof(struct frag_hdr);
1124 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1125 int offset, int len, int odd, struct sk_buff *skb),
1126 void *from, int length, int transhdrlen,
1127 int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1128 struct rt6_info *rt, unsigned int flags, int dontfrag)
1130 struct inet_sock *inet = inet_sk(sk);
1131 struct ipv6_pinfo *np = inet6_sk(sk);
1132 struct inet_cork *cork;
1133 struct sk_buff *skb, *skb_prev = NULL;
1134 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu;
1144 if (flags&MSG_PROBE)
1146 cork = &inet->cork.base;
1147 if (skb_queue_empty(&sk->sk_write_queue)) {
1152 if (WARN_ON(np->cork.opt))
1155 np->cork.opt = kzalloc(opt->tot_len, sk->sk_allocation);
1156 if (unlikely(np->cork.opt == NULL))
1159 np->cork.opt->tot_len = opt->tot_len;
1160 np->cork.opt->opt_flen = opt->opt_flen;
1161 np->cork.opt->opt_nflen = opt->opt_nflen;
1163 np->cork.opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1165 if (opt->dst0opt && !np->cork.opt->dst0opt)
1168 np->cork.opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1170 if (opt->dst1opt && !np->cork.opt->dst1opt)
1173 np->cork.opt->hopopt = ip6_opt_dup(opt->hopopt,
1175 if (opt->hopopt && !np->cork.opt->hopopt)
1178 np->cork.opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1180 if (opt->srcrt && !np->cork.opt->srcrt)
1183 /* need source address above miyazawa*/
1186 cork->dst = &rt->dst;
1187 inet->cork.fl.u.ip6 = *fl6;
1188 np->cork.hop_limit = hlimit;
1189 np->cork.tclass = tclass;
1190 if (rt->dst.flags & DST_XFRM_TUNNEL)
1191 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1192 rt->dst.dev->mtu : dst_mtu(&rt->dst);
1194 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1195 rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1196 if (np->frag_size < mtu) {
1198 mtu = np->frag_size;
1200 cork->fragsize = mtu;
1201 if (dst_allfrag(rt->dst.path))
1202 cork->flags |= IPCORK_ALLFRAG;
1204 exthdrlen = (opt ? opt->opt_flen : 0);
1205 length += exthdrlen;
1206 transhdrlen += exthdrlen;
1207 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1209 rt = (struct rt6_info *)cork->dst;
1210 fl6 = &inet->cork.fl.u.ip6;
1215 mtu = cork->fragsize;
1219 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1221 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1222 (opt ? opt->opt_nflen : 0);
1223 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1224 sizeof(struct frag_hdr);
1226 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1227 unsigned int maxnonfragsize, headersize;
1229 headersize = sizeof(struct ipv6hdr) +
1230 (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1231 (dst_allfrag(&rt->dst) ?
1232 sizeof(struct frag_hdr) : 0) +
1233 rt->rt6i_nfheader_len;
1235 if (ip6_sk_ignore_df(sk))
1236 maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1238 maxnonfragsize = mtu;
1240 /* dontfrag active */
1241 if ((cork->length + length > mtu - headersize) && dontfrag &&
1242 (sk->sk_protocol == IPPROTO_UDP ||
1243 sk->sk_protocol == IPPROTO_RAW)) {
1244 ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1245 sizeof(struct ipv6hdr));
1249 if (cork->length + length > maxnonfragsize - headersize) {
1251 ipv6_local_error(sk, EMSGSIZE, fl6,
1253 sizeof(struct ipv6hdr));
1258 if (sk->sk_type == SOCK_DGRAM || sk->sk_type == SOCK_RAW) {
1259 sock_tx_timestamp(sk, &tx_flags);
1260 if (tx_flags & SKBTX_ANY_SW_TSTAMP &&
1261 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1262 tskey = sk->sk_tskey++;
1266 * Let's try using as much space as possible.
1267 * Use MTU if total length of the message fits into the MTU.
1268 * Otherwise, we need to reserve fragment header and
1269 * fragment alignment (= 8-15 octects, in total).
1271 * Note that we may need to "move" the data from the tail of
1272 * of the buffer to the new fragment when we split
1275 * FIXME: It may be fragmented into multiple chunks
1276 * at once if non-fragmentable extension headers
1281 skb = skb_peek_tail(&sk->sk_write_queue);
1282 cork->length += length;
1283 if (((length > mtu) ||
1284 (skb && skb_is_gso(skb))) &&
1285 (sk->sk_protocol == IPPROTO_UDP) &&
1286 (rt->dst.dev->features & NETIF_F_UFO)) {
1287 err = ip6_ufo_append_data(sk, getfrag, from, length,
1288 hh_len, fragheaderlen,
1289 transhdrlen, mtu, flags, rt);
1298 while (length > 0) {
1299 /* Check if the remaining data fits into current packet. */
1300 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1302 copy = maxfraglen - skb->len;
1306 unsigned int datalen;
1307 unsigned int fraglen;
1308 unsigned int fraggap;
1309 unsigned int alloclen;
1311 /* There's no room in the current skb */
1313 fraggap = skb->len - maxfraglen;
1316 /* update mtu and maxfraglen if necessary */
1317 if (skb == NULL || skb_prev == NULL)
1318 ip6_append_data_mtu(&mtu, &maxfraglen,
1319 fragheaderlen, skb, rt,
1325 * If remaining data exceeds the mtu,
1326 * we know we need more fragment(s).
1328 datalen = length + fraggap;
1330 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1331 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1332 if ((flags & MSG_MORE) &&
1333 !(rt->dst.dev->features&NETIF_F_SG))
1336 alloclen = datalen + fragheaderlen;
1338 alloclen += dst_exthdrlen;
1340 if (datalen != length + fraggap) {
1342 * this is not the last fragment, the trailer
1343 * space is regarded as data space.
1345 datalen += rt->dst.trailer_len;
1348 alloclen += rt->dst.trailer_len;
1349 fraglen = datalen + fragheaderlen;
1352 * We just reserve space for fragment header.
1353 * Note: this may be overallocation if the message
1354 * (without MSG_MORE) fits into the MTU.
1356 alloclen += sizeof(struct frag_hdr);
1359 skb = sock_alloc_send_skb(sk,
1361 (flags & MSG_DONTWAIT), &err);
1364 if (atomic_read(&sk->sk_wmem_alloc) <=
1366 skb = sock_wmalloc(sk,
1367 alloclen + hh_len, 1,
1369 if (unlikely(skb == NULL))
1375 * Fill in the control structures
1377 skb->protocol = htons(ETH_P_IPV6);
1378 skb->ip_summed = CHECKSUM_NONE;
1380 /* reserve for fragmentation and ipsec header */
1381 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1384 /* Only the initial fragment is time stamped */
1385 skb_shinfo(skb)->tx_flags = tx_flags;
1387 skb_shinfo(skb)->tskey = tskey;
1391 * Find where to start putting bytes
1393 data = skb_put(skb, fraglen);
1394 skb_set_network_header(skb, exthdrlen);
1395 data += fragheaderlen;
1396 skb->transport_header = (skb->network_header +
1399 skb->csum = skb_copy_and_csum_bits(
1400 skb_prev, maxfraglen,
1401 data + transhdrlen, fraggap, 0);
1402 skb_prev->csum = csum_sub(skb_prev->csum,
1405 pskb_trim_unique(skb_prev, maxfraglen);
1407 copy = datalen - transhdrlen - fraggap;
1413 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1420 length -= datalen - fraggap;
1426 * Put the packet on the pending queue
1428 __skb_queue_tail(&sk->sk_write_queue, skb);
1435 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1439 if (getfrag(from, skb_put(skb, copy),
1440 offset, copy, off, skb) < 0) {
1441 __skb_trim(skb, off);
1446 int i = skb_shinfo(skb)->nr_frags;
1447 struct page_frag *pfrag = sk_page_frag(sk);
1450 if (!sk_page_frag_refill(sk, pfrag))
1453 if (!skb_can_coalesce(skb, i, pfrag->page,
1456 if (i == MAX_SKB_FRAGS)
1459 __skb_fill_page_desc(skb, i, pfrag->page,
1461 skb_shinfo(skb)->nr_frags = ++i;
1462 get_page(pfrag->page);
1464 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1466 page_address(pfrag->page) + pfrag->offset,
1467 offset, copy, skb->len, skb) < 0)
1470 pfrag->offset += copy;
1471 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1473 skb->data_len += copy;
1474 skb->truesize += copy;
1475 atomic_add(copy, &sk->sk_wmem_alloc);
1486 cork->length -= length;
1487 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1490 EXPORT_SYMBOL_GPL(ip6_append_data);
1492 static void ip6_cork_release(struct inet_sock *inet, struct ipv6_pinfo *np)
1495 kfree(np->cork.opt->dst0opt);
1496 kfree(np->cork.opt->dst1opt);
1497 kfree(np->cork.opt->hopopt);
1498 kfree(np->cork.opt->srcrt);
1499 kfree(np->cork.opt);
1500 np->cork.opt = NULL;
1503 if (inet->cork.base.dst) {
1504 dst_release(inet->cork.base.dst);
1505 inet->cork.base.dst = NULL;
1506 inet->cork.base.flags &= ~IPCORK_ALLFRAG;
1508 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1511 int ip6_push_pending_frames(struct sock *sk)
1513 struct sk_buff *skb, *tmp_skb;
1514 struct sk_buff **tail_skb;
1515 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1516 struct inet_sock *inet = inet_sk(sk);
1517 struct ipv6_pinfo *np = inet6_sk(sk);
1518 struct net *net = sock_net(sk);
1519 struct ipv6hdr *hdr;
1520 struct ipv6_txoptions *opt = np->cork.opt;
1521 struct rt6_info *rt = (struct rt6_info *)inet->cork.base.dst;
1522 struct flowi6 *fl6 = &inet->cork.fl.u.ip6;
1523 unsigned char proto = fl6->flowi6_proto;
1526 skb = __skb_dequeue(&sk->sk_write_queue);
1529 tail_skb = &(skb_shinfo(skb)->frag_list);
1531 /* move skb->data to ip header from ext header */
1532 if (skb->data < skb_network_header(skb))
1533 __skb_pull(skb, skb_network_offset(skb));
1534 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1535 __skb_pull(tmp_skb, skb_network_header_len(skb));
1536 *tail_skb = tmp_skb;
1537 tail_skb = &(tmp_skb->next);
1538 skb->len += tmp_skb->len;
1539 skb->data_len += tmp_skb->len;
1540 skb->truesize += tmp_skb->truesize;
1541 tmp_skb->destructor = NULL;
1545 /* Allow local fragmentation. */
1546 skb->ignore_df = ip6_sk_ignore_df(sk);
1548 *final_dst = fl6->daddr;
1549 __skb_pull(skb, skb_network_header_len(skb));
1550 if (opt && opt->opt_flen)
1551 ipv6_push_frag_opts(skb, opt, &proto);
1552 if (opt && opt->opt_nflen)
1553 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1555 skb_push(skb, sizeof(struct ipv6hdr));
1556 skb_reset_network_header(skb);
1557 hdr = ipv6_hdr(skb);
1559 ip6_flow_hdr(hdr, np->cork.tclass,
1560 ip6_make_flowlabel(net, skb, fl6->flowlabel,
1561 np->autoflowlabel));
1562 hdr->hop_limit = np->cork.hop_limit;
1563 hdr->nexthdr = proto;
1564 hdr->saddr = fl6->saddr;
1565 hdr->daddr = *final_dst;
1567 skb->priority = sk->sk_priority;
1568 skb->mark = sk->sk_mark;
1570 skb_dst_set(skb, dst_clone(&rt->dst));
1571 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1572 if (proto == IPPROTO_ICMPV6) {
1573 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1575 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1576 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1579 err = ip6_local_out(skb);
1582 err = net_xmit_errno(err);
1588 ip6_cork_release(inet, np);
1591 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1594 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1596 void ip6_flush_pending_frames(struct sock *sk)
1598 struct sk_buff *skb;
1600 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1602 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1603 IPSTATS_MIB_OUTDISCARDS);
1607 ip6_cork_release(inet_sk(sk), inet6_sk(sk));
1609 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);