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 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, fl6->flowlabel);
210 hdr->payload_len = htons(seg_len);
211 hdr->nexthdr = proto;
212 hdr->hop_limit = hlimit;
214 hdr->saddr = fl6->saddr;
215 hdr->daddr = *first_hop;
217 skb->protocol = htons(ETH_P_IPV6);
218 skb->priority = sk->sk_priority;
219 skb->mark = sk->sk_mark;
222 if ((skb->len <= mtu) || skb->local_df || skb_is_gso(skb)) {
223 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
224 IPSTATS_MIB_OUT, skb->len);
225 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, skb, NULL,
226 dst->dev, dst_output);
230 ipv6_local_error(sk, EMSGSIZE, fl6, mtu);
231 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 int ip6_forward(struct sk_buff *skb)
326 struct dst_entry *dst = skb_dst(skb);
327 struct ipv6hdr *hdr = ipv6_hdr(skb);
328 struct inet6_skb_parm *opt = IP6CB(skb);
329 struct net *net = dev_net(dst->dev);
332 if (net->ipv6.devconf_all->forwarding == 0)
335 if (skb_warn_if_lro(skb))
338 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
339 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
340 IPSTATS_MIB_INDISCARDS);
344 if (skb->pkt_type != PACKET_HOST)
347 skb_forward_csum(skb);
350 * We DO NOT make any processing on
351 * RA packets, pushing them to user level AS IS
352 * without ane WARRANTY that application will be able
353 * to interpret them. The reason is that we
354 * cannot make anything clever here.
356 * We are not end-node, so that if packet contains
357 * AH/ESP, we cannot make anything.
358 * Defragmentation also would be mistake, RA packets
359 * cannot be fragmented, because there is no warranty
360 * that different fragments will go along one path. --ANK
362 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
363 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
368 * check and decrement ttl
370 if (hdr->hop_limit <= 1) {
371 /* Force OUTPUT device used as source address */
373 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
374 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
375 IPSTATS_MIB_INHDRERRORS);
381 /* XXX: idev->cnf.proxy_ndp? */
382 if (net->ipv6.devconf_all->proxy_ndp &&
383 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
384 int proxied = ip6_forward_proxy_check(skb);
386 return ip6_input(skb);
387 else if (proxied < 0) {
388 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
389 IPSTATS_MIB_INDISCARDS);
394 if (!xfrm6_route_forward(skb)) {
395 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
396 IPSTATS_MIB_INDISCARDS);
401 /* IPv6 specs say nothing about it, but it is clear that we cannot
402 send redirects to source routed frames.
403 We don't send redirects to frames decapsulated from IPsec.
405 if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
406 struct in6_addr *target = NULL;
407 struct inet_peer *peer;
411 * incoming and outgoing devices are the same
415 rt = (struct rt6_info *) dst;
416 if (rt->rt6i_flags & RTF_GATEWAY)
417 target = &rt->rt6i_gateway;
419 target = &hdr->daddr;
421 peer = inet_getpeer_v6(net->ipv6.peers, &rt->rt6i_dst.addr, 1);
423 /* Limit redirects both by destination (here)
424 and by source (inside ndisc_send_redirect)
426 if (inet_peer_xrlim_allow(peer, 1*HZ))
427 ndisc_send_redirect(skb, target);
431 int addrtype = ipv6_addr_type(&hdr->saddr);
433 /* This check is security critical. */
434 if (addrtype == IPV6_ADDR_ANY ||
435 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
437 if (addrtype & IPV6_ADDR_LINKLOCAL) {
438 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
439 ICMPV6_NOT_NEIGHBOUR, 0);
445 if (mtu < IPV6_MIN_MTU)
448 if ((!skb->local_df && skb->len > mtu && !skb_is_gso(skb)) ||
449 (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)) {
450 /* Again, force OUTPUT device used as source address */
452 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
453 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
454 IPSTATS_MIB_INTOOBIGERRORS);
455 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
456 IPSTATS_MIB_FRAGFAILS);
461 if (skb_cow(skb, dst->dev->hard_header_len)) {
462 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
463 IPSTATS_MIB_OUTDISCARDS);
469 /* Mangling hops number delayed to point after skb COW */
473 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
474 IP6_ADD_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
475 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD, skb, skb->dev, dst->dev,
479 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
485 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
487 to->pkt_type = from->pkt_type;
488 to->priority = from->priority;
489 to->protocol = from->protocol;
491 skb_dst_set(to, dst_clone(skb_dst(from)));
493 to->mark = from->mark;
495 #ifdef CONFIG_NET_SCHED
496 to->tc_index = from->tc_index;
499 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
500 to->nf_trace = from->nf_trace;
502 skb_copy_secmark(to, from);
505 int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
507 struct sk_buff *frag;
508 struct rt6_info *rt = (struct rt6_info*)skb_dst(skb);
509 struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
510 struct ipv6hdr *tmp_hdr;
512 unsigned int mtu, hlen, left, len;
515 int ptr, offset = 0, err=0;
516 u8 *prevhdr, nexthdr = 0;
517 struct net *net = dev_net(skb_dst(skb)->dev);
519 hlen = ip6_find_1stfragopt(skb, &prevhdr);
522 mtu = ip6_skb_dst_mtu(skb);
524 /* We must not fragment if the socket is set to force MTU discovery
525 * or if the skb it not generated by a local socket.
527 if (unlikely(!skb->local_df && skb->len > mtu) ||
528 (IP6CB(skb)->frag_max_size &&
529 IP6CB(skb)->frag_max_size > mtu)) {
530 if (skb->sk && dst_allfrag(skb_dst(skb)))
531 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
533 skb->dev = skb_dst(skb)->dev;
534 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
535 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
536 IPSTATS_MIB_FRAGFAILS);
541 if (np && np->frag_size < mtu) {
545 mtu -= hlen + sizeof(struct frag_hdr);
547 if (skb_has_frag_list(skb)) {
548 int first_len = skb_pagelen(skb);
549 struct sk_buff *frag2;
551 if (first_len - hlen > mtu ||
552 ((first_len - hlen) & 7) ||
556 skb_walk_frags(skb, frag) {
557 /* Correct geometry. */
558 if (frag->len > mtu ||
559 ((frag->len & 7) && frag->next) ||
560 skb_headroom(frag) < hlen)
561 goto slow_path_clean;
563 /* Partially cloned skb? */
564 if (skb_shared(frag))
565 goto slow_path_clean;
570 frag->destructor = sock_wfree;
572 skb->truesize -= frag->truesize;
577 frag = skb_shinfo(skb)->frag_list;
578 skb_frag_list_init(skb);
581 *prevhdr = NEXTHDR_FRAGMENT;
582 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
584 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
585 IPSTATS_MIB_FRAGFAILS);
589 __skb_pull(skb, hlen);
590 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
591 __skb_push(skb, hlen);
592 skb_reset_network_header(skb);
593 memcpy(skb_network_header(skb), tmp_hdr, hlen);
595 ipv6_select_ident(fh, rt);
596 fh->nexthdr = nexthdr;
598 fh->frag_off = htons(IP6_MF);
599 frag_id = fh->identification;
601 first_len = skb_pagelen(skb);
602 skb->data_len = first_len - skb_headlen(skb);
603 skb->len = first_len;
604 ipv6_hdr(skb)->payload_len = htons(first_len -
605 sizeof(struct ipv6hdr));
610 /* Prepare header of the next frame,
611 * before previous one went down. */
613 frag->ip_summed = CHECKSUM_NONE;
614 skb_reset_transport_header(frag);
615 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
616 __skb_push(frag, hlen);
617 skb_reset_network_header(frag);
618 memcpy(skb_network_header(frag), tmp_hdr,
620 offset += skb->len - hlen - sizeof(struct frag_hdr);
621 fh->nexthdr = nexthdr;
623 fh->frag_off = htons(offset);
624 if (frag->next != NULL)
625 fh->frag_off |= htons(IP6_MF);
626 fh->identification = frag_id;
627 ipv6_hdr(frag)->payload_len =
629 sizeof(struct ipv6hdr));
630 ip6_copy_metadata(frag, skb);
635 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
636 IPSTATS_MIB_FRAGCREATES);
649 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
650 IPSTATS_MIB_FRAGOKS);
661 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
662 IPSTATS_MIB_FRAGFAILS);
667 skb_walk_frags(skb, frag2) {
671 frag2->destructor = NULL;
672 skb->truesize += frag2->truesize;
677 if ((skb->ip_summed == CHECKSUM_PARTIAL) &&
678 skb_checksum_help(skb))
681 left = skb->len - hlen; /* Space per frame */
682 ptr = hlen; /* Where to start from */
685 * Fragment the datagram.
688 *prevhdr = NEXTHDR_FRAGMENT;
689 hroom = LL_RESERVED_SPACE(rt->dst.dev);
690 troom = rt->dst.dev->needed_tailroom;
693 * Keep copying data until we run out.
697 /* IF: it doesn't fit, use 'mtu' - the data space left */
700 /* IF: we are not sending up to and including the packet end
701 then align the next start on an eight byte boundary */
709 if ((frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
710 hroom + troom, GFP_ATOMIC)) == NULL) {
711 NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
712 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
713 IPSTATS_MIB_FRAGFAILS);
719 * Set up data on packet
722 ip6_copy_metadata(frag, skb);
723 skb_reserve(frag, hroom);
724 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
725 skb_reset_network_header(frag);
726 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
727 frag->transport_header = (frag->network_header + hlen +
728 sizeof(struct frag_hdr));
731 * Charge the memory for the fragment to any owner
735 skb_set_owner_w(frag, skb->sk);
738 * Copy the packet header into the new buffer.
740 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
743 * Build fragment header.
745 fh->nexthdr = nexthdr;
748 ipv6_select_ident(fh, rt);
749 frag_id = fh->identification;
751 fh->identification = frag_id;
754 * Copy a block of the IP datagram.
756 if (skb_copy_bits(skb, ptr, skb_transport_header(frag), len))
760 fh->frag_off = htons(offset);
762 fh->frag_off |= htons(IP6_MF);
763 ipv6_hdr(frag)->payload_len = htons(frag->len -
764 sizeof(struct ipv6hdr));
770 * Put this fragment into the sending queue.
776 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
777 IPSTATS_MIB_FRAGCREATES);
779 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
780 IPSTATS_MIB_FRAGOKS);
785 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
786 IPSTATS_MIB_FRAGFAILS);
791 static inline int ip6_rt_check(const struct rt6key *rt_key,
792 const struct in6_addr *fl_addr,
793 const struct in6_addr *addr_cache)
795 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
796 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache));
799 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
800 struct dst_entry *dst,
801 const struct flowi6 *fl6)
803 struct ipv6_pinfo *np = inet6_sk(sk);
809 if (dst->ops->family != AF_INET6) {
814 rt = (struct rt6_info *)dst;
815 /* Yes, checking route validity in not connected
816 * case is not very simple. Take into account,
817 * that we do not support routing by source, TOS,
818 * and MSG_DONTROUTE --ANK (980726)
820 * 1. ip6_rt_check(): If route was host route,
821 * check that cached destination is current.
822 * If it is network route, we still may
823 * check its validity using saved pointer
824 * to the last used address: daddr_cache.
825 * We do not want to save whole address now,
826 * (because main consumer of this service
827 * is tcp, which has not this problem),
828 * so that the last trick works only on connected
830 * 2. oif also should be the same.
832 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
833 #ifdef CONFIG_IPV6_SUBTREES
834 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
836 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
845 static int ip6_dst_lookup_tail(struct sock *sk,
846 struct dst_entry **dst, struct flowi6 *fl6)
848 struct net *net = sock_net(sk);
849 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
856 *dst = ip6_route_output(net, sk, fl6);
858 if ((err = (*dst)->error))
859 goto out_err_release;
861 if (ipv6_addr_any(&fl6->saddr)) {
862 struct rt6_info *rt = (struct rt6_info *) *dst;
863 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
864 sk ? inet6_sk(sk)->srcprefs : 0,
867 goto out_err_release;
870 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
872 * Here if the dst entry we've looked up
873 * has a neighbour entry that is in the INCOMPLETE
874 * state and the src address from the flow is
875 * marked as OPTIMISTIC, we release the found
876 * dst entry and replace it instead with the
877 * dst entry of the nexthop router
879 rt = (struct rt6_info *) *dst;
881 n = __ipv6_neigh_lookup_noref(rt->dst.dev, rt6_nexthop(rt));
882 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
883 rcu_read_unlock_bh();
886 struct inet6_ifaddr *ifp;
887 struct flowi6 fl_gw6;
890 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
893 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
899 * We need to get the dst entry for the
900 * default router instead
903 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
904 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
905 *dst = ip6_route_output(net, sk, &fl_gw6);
906 if ((err = (*dst)->error))
907 goto out_err_release;
915 if (err == -ENETUNREACH)
916 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
923 * ip6_dst_lookup - perform route lookup on flow
924 * @sk: socket which provides route info
925 * @dst: pointer to dst_entry * for result
926 * @fl6: flow to lookup
928 * This function performs a route lookup on the given flow.
930 * It returns zero on success, or a standard errno code on error.
932 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi6 *fl6)
935 return ip6_dst_lookup_tail(sk, dst, fl6);
937 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
940 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
941 * @sk: socket which provides route info
942 * @fl6: flow to lookup
943 * @final_dst: final destination address for ipsec lookup
945 * This function performs a route lookup on the given flow.
947 * It returns a valid dst pointer on success, or a pointer encoded
950 struct dst_entry *ip6_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
951 const struct in6_addr *final_dst)
953 struct dst_entry *dst = NULL;
956 err = ip6_dst_lookup_tail(sk, &dst, fl6);
960 fl6->daddr = *final_dst;
962 return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
964 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
967 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
968 * @sk: socket which provides the dst cache and route info
969 * @fl6: flow to lookup
970 * @final_dst: final destination address for ipsec lookup
972 * This function performs a route lookup on the given flow with the
973 * possibility of using the cached route in the socket if it is valid.
974 * It will take the socket dst lock when operating on the dst cache.
975 * As a result, this function can only be used in process context.
977 * It returns a valid dst pointer on success, or a pointer encoded
980 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
981 const struct in6_addr *final_dst)
983 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
986 dst = ip6_sk_dst_check(sk, dst, fl6);
988 err = ip6_dst_lookup_tail(sk, &dst, fl6);
992 fl6->daddr = *final_dst;
994 return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
996 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
998 static inline int ip6_ufo_append_data(struct sock *sk,
999 int getfrag(void *from, char *to, int offset, int len,
1000 int odd, struct sk_buff *skb),
1001 void *from, int length, int hh_len, int fragheaderlen,
1002 int transhdrlen, int mtu,unsigned int flags,
1003 struct rt6_info *rt)
1006 struct sk_buff *skb;
1007 struct frag_hdr fhdr;
1010 /* There is support for UDP large send offload by network
1011 * device, so create one single skb packet containing complete
1014 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
1015 skb = sock_alloc_send_skb(sk,
1016 hh_len + fragheaderlen + transhdrlen + 20,
1017 (flags & MSG_DONTWAIT), &err);
1021 /* reserve space for Hardware header */
1022 skb_reserve(skb, hh_len);
1024 /* create space for UDP/IP header */
1025 skb_put(skb,fragheaderlen + transhdrlen);
1027 /* initialize network header pointer */
1028 skb_reset_network_header(skb);
1030 /* initialize protocol header pointer */
1031 skb->transport_header = skb->network_header + fragheaderlen;
1033 skb->protocol = htons(ETH_P_IPV6);
1036 __skb_queue_tail(&sk->sk_write_queue, skb);
1037 } else if (skb_is_gso(skb)) {
1041 skb->ip_summed = CHECKSUM_PARTIAL;
1042 /* Specify the length of each IPv6 datagram fragment.
1043 * It has to be a multiple of 8.
1045 skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1046 sizeof(struct frag_hdr)) & ~7;
1047 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1048 ipv6_select_ident(&fhdr, rt);
1049 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1052 return skb_append_datato_frags(sk, skb, getfrag, from,
1053 (length - transhdrlen));
1056 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1059 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1062 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1065 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1068 static void ip6_append_data_mtu(unsigned int *mtu,
1070 unsigned int fragheaderlen,
1071 struct sk_buff *skb,
1072 struct rt6_info *rt,
1075 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1077 /* first fragment, reserve header_len */
1078 *mtu = *mtu - rt->dst.header_len;
1082 * this fragment is not first, the headers
1083 * space is regarded as data space.
1085 *mtu = min(*mtu, pmtuprobe ?
1087 dst_mtu(rt->dst.path));
1089 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1090 + fragheaderlen - sizeof(struct frag_hdr);
1094 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1095 int offset, int len, int odd, struct sk_buff *skb),
1096 void *from, int length, int transhdrlen,
1097 int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1098 struct rt6_info *rt, unsigned int flags, int dontfrag)
1100 struct inet_sock *inet = inet_sk(sk);
1101 struct ipv6_pinfo *np = inet6_sk(sk);
1102 struct inet_cork *cork;
1103 struct sk_buff *skb, *skb_prev = NULL;
1104 unsigned int maxfraglen, fragheaderlen, mtu;
1113 if (flags&MSG_PROBE)
1115 cork = &inet->cork.base;
1116 if (skb_queue_empty(&sk->sk_write_queue)) {
1121 if (WARN_ON(np->cork.opt))
1124 np->cork.opt = kzalloc(opt->tot_len, sk->sk_allocation);
1125 if (unlikely(np->cork.opt == NULL))
1128 np->cork.opt->tot_len = opt->tot_len;
1129 np->cork.opt->opt_flen = opt->opt_flen;
1130 np->cork.opt->opt_nflen = opt->opt_nflen;
1132 np->cork.opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1134 if (opt->dst0opt && !np->cork.opt->dst0opt)
1137 np->cork.opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1139 if (opt->dst1opt && !np->cork.opt->dst1opt)
1142 np->cork.opt->hopopt = ip6_opt_dup(opt->hopopt,
1144 if (opt->hopopt && !np->cork.opt->hopopt)
1147 np->cork.opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1149 if (opt->srcrt && !np->cork.opt->srcrt)
1152 /* need source address above miyazawa*/
1155 cork->dst = &rt->dst;
1156 inet->cork.fl.u.ip6 = *fl6;
1157 np->cork.hop_limit = hlimit;
1158 np->cork.tclass = tclass;
1159 if (rt->dst.flags & DST_XFRM_TUNNEL)
1160 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1161 rt->dst.dev->mtu : dst_mtu(&rt->dst);
1163 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1164 rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1165 if (np->frag_size < mtu) {
1167 mtu = np->frag_size;
1169 cork->fragsize = mtu;
1170 if (dst_allfrag(rt->dst.path))
1171 cork->flags |= IPCORK_ALLFRAG;
1173 exthdrlen = (opt ? opt->opt_flen : 0);
1174 length += exthdrlen;
1175 transhdrlen += exthdrlen;
1176 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1178 rt = (struct rt6_info *)cork->dst;
1179 fl6 = &inet->cork.fl.u.ip6;
1184 mtu = cork->fragsize;
1187 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1189 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1190 (opt ? opt->opt_nflen : 0);
1191 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
1193 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1194 if (cork->length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
1195 ipv6_local_error(sk, EMSGSIZE, fl6, mtu-exthdrlen);
1200 /* For UDP, check if TX timestamp is enabled */
1201 if (sk->sk_type == SOCK_DGRAM)
1202 sock_tx_timestamp(sk, &tx_flags);
1205 * Let's try using as much space as possible.
1206 * Use MTU if total length of the message fits into the MTU.
1207 * Otherwise, we need to reserve fragment header and
1208 * fragment alignment (= 8-15 octects, in total).
1210 * Note that we may need to "move" the data from the tail of
1211 * of the buffer to the new fragment when we split
1214 * FIXME: It may be fragmented into multiple chunks
1215 * at once if non-fragmentable extension headers
1220 if ((length > mtu) && dontfrag && (sk->sk_protocol == IPPROTO_UDP ||
1221 sk->sk_protocol == IPPROTO_RAW)) {
1222 ipv6_local_rxpmtu(sk, fl6, mtu-exthdrlen);
1226 skb = skb_peek_tail(&sk->sk_write_queue);
1227 cork->length += length;
1228 if (((length > mtu) ||
1229 (skb && skb_is_gso(skb))) &&
1230 (sk->sk_protocol == IPPROTO_UDP) &&
1231 (rt->dst.dev->features & NETIF_F_UFO)) {
1232 err = ip6_ufo_append_data(sk, getfrag, from, length,
1233 hh_len, fragheaderlen,
1234 transhdrlen, mtu, flags, rt);
1243 while (length > 0) {
1244 /* Check if the remaining data fits into current packet. */
1245 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1247 copy = maxfraglen - skb->len;
1251 unsigned int datalen;
1252 unsigned int fraglen;
1253 unsigned int fraggap;
1254 unsigned int alloclen;
1256 /* There's no room in the current skb */
1258 fraggap = skb->len - maxfraglen;
1261 /* update mtu and maxfraglen if necessary */
1262 if (skb == NULL || skb_prev == NULL)
1263 ip6_append_data_mtu(&mtu, &maxfraglen,
1264 fragheaderlen, skb, rt,
1266 IPV6_PMTUDISC_PROBE);
1271 * If remaining data exceeds the mtu,
1272 * we know we need more fragment(s).
1274 datalen = length + fraggap;
1276 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1277 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1278 if ((flags & MSG_MORE) &&
1279 !(rt->dst.dev->features&NETIF_F_SG))
1282 alloclen = datalen + fragheaderlen;
1284 alloclen += dst_exthdrlen;
1286 if (datalen != length + fraggap) {
1288 * this is not the last fragment, the trailer
1289 * space is regarded as data space.
1291 datalen += rt->dst.trailer_len;
1294 alloclen += rt->dst.trailer_len;
1295 fraglen = datalen + fragheaderlen;
1298 * We just reserve space for fragment header.
1299 * Note: this may be overallocation if the message
1300 * (without MSG_MORE) fits into the MTU.
1302 alloclen += sizeof(struct frag_hdr);
1305 skb = sock_alloc_send_skb(sk,
1307 (flags & MSG_DONTWAIT), &err);
1310 if (atomic_read(&sk->sk_wmem_alloc) <=
1312 skb = sock_wmalloc(sk,
1313 alloclen + hh_len, 1,
1315 if (unlikely(skb == NULL))
1318 /* Only the initial fragment
1327 * Fill in the control structures
1329 skb->protocol = htons(ETH_P_IPV6);
1330 skb->ip_summed = CHECKSUM_NONE;
1332 /* reserve for fragmentation and ipsec header */
1333 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1336 if (sk->sk_type == SOCK_DGRAM)
1337 skb_shinfo(skb)->tx_flags = tx_flags;
1340 * Find where to start putting bytes
1342 data = skb_put(skb, fraglen);
1343 skb_set_network_header(skb, exthdrlen);
1344 data += fragheaderlen;
1345 skb->transport_header = (skb->network_header +
1348 skb->csum = skb_copy_and_csum_bits(
1349 skb_prev, maxfraglen,
1350 data + transhdrlen, fraggap, 0);
1351 skb_prev->csum = csum_sub(skb_prev->csum,
1354 pskb_trim_unique(skb_prev, maxfraglen);
1356 copy = datalen - transhdrlen - fraggap;
1362 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1369 length -= datalen - fraggap;
1375 * Put the packet on the pending queue
1377 __skb_queue_tail(&sk->sk_write_queue, skb);
1384 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1388 if (getfrag(from, skb_put(skb, copy),
1389 offset, copy, off, skb) < 0) {
1390 __skb_trim(skb, off);
1395 int i = skb_shinfo(skb)->nr_frags;
1396 struct page_frag *pfrag = sk_page_frag(sk);
1399 if (!sk_page_frag_refill(sk, pfrag))
1402 if (!skb_can_coalesce(skb, i, pfrag->page,
1405 if (i == MAX_SKB_FRAGS)
1408 __skb_fill_page_desc(skb, i, pfrag->page,
1410 skb_shinfo(skb)->nr_frags = ++i;
1411 get_page(pfrag->page);
1413 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1415 page_address(pfrag->page) + pfrag->offset,
1416 offset, copy, skb->len, skb) < 0)
1419 pfrag->offset += copy;
1420 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1422 skb->data_len += copy;
1423 skb->truesize += copy;
1424 atomic_add(copy, &sk->sk_wmem_alloc);
1435 cork->length -= length;
1436 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1439 EXPORT_SYMBOL_GPL(ip6_append_data);
1441 static void ip6_cork_release(struct inet_sock *inet, struct ipv6_pinfo *np)
1444 kfree(np->cork.opt->dst0opt);
1445 kfree(np->cork.opt->dst1opt);
1446 kfree(np->cork.opt->hopopt);
1447 kfree(np->cork.opt->srcrt);
1448 kfree(np->cork.opt);
1449 np->cork.opt = NULL;
1452 if (inet->cork.base.dst) {
1453 dst_release(inet->cork.base.dst);
1454 inet->cork.base.dst = NULL;
1455 inet->cork.base.flags &= ~IPCORK_ALLFRAG;
1457 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1460 int ip6_push_pending_frames(struct sock *sk)
1462 struct sk_buff *skb, *tmp_skb;
1463 struct sk_buff **tail_skb;
1464 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1465 struct inet_sock *inet = inet_sk(sk);
1466 struct ipv6_pinfo *np = inet6_sk(sk);
1467 struct net *net = sock_net(sk);
1468 struct ipv6hdr *hdr;
1469 struct ipv6_txoptions *opt = np->cork.opt;
1470 struct rt6_info *rt = (struct rt6_info *)inet->cork.base.dst;
1471 struct flowi6 *fl6 = &inet->cork.fl.u.ip6;
1472 unsigned char proto = fl6->flowi6_proto;
1475 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1477 tail_skb = &(skb_shinfo(skb)->frag_list);
1479 /* move skb->data to ip header from ext header */
1480 if (skb->data < skb_network_header(skb))
1481 __skb_pull(skb, skb_network_offset(skb));
1482 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1483 __skb_pull(tmp_skb, skb_network_header_len(skb));
1484 *tail_skb = tmp_skb;
1485 tail_skb = &(tmp_skb->next);
1486 skb->len += tmp_skb->len;
1487 skb->data_len += tmp_skb->len;
1488 skb->truesize += tmp_skb->truesize;
1489 tmp_skb->destructor = NULL;
1493 /* Allow local fragmentation. */
1494 if (np->pmtudisc < IPV6_PMTUDISC_DO)
1497 *final_dst = fl6->daddr;
1498 __skb_pull(skb, skb_network_header_len(skb));
1499 if (opt && opt->opt_flen)
1500 ipv6_push_frag_opts(skb, opt, &proto);
1501 if (opt && opt->opt_nflen)
1502 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1504 skb_push(skb, sizeof(struct ipv6hdr));
1505 skb_reset_network_header(skb);
1506 hdr = ipv6_hdr(skb);
1508 ip6_flow_hdr(hdr, np->cork.tclass, fl6->flowlabel);
1509 hdr->hop_limit = np->cork.hop_limit;
1510 hdr->nexthdr = proto;
1511 hdr->saddr = fl6->saddr;
1512 hdr->daddr = *final_dst;
1514 skb->priority = sk->sk_priority;
1515 skb->mark = sk->sk_mark;
1517 skb_dst_set(skb, dst_clone(&rt->dst));
1518 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1519 if (proto == IPPROTO_ICMPV6) {
1520 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1522 ICMP6MSGOUT_INC_STATS_BH(net, idev, icmp6_hdr(skb)->icmp6_type);
1523 ICMP6_INC_STATS_BH(net, idev, ICMP6_MIB_OUTMSGS);
1526 err = ip6_local_out(skb);
1529 err = net_xmit_errno(err);
1535 ip6_cork_release(inet, np);
1538 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1541 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1543 void ip6_flush_pending_frames(struct sock *sk)
1545 struct sk_buff *skb;
1547 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1549 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1550 IPSTATS_MIB_OUTDISCARDS);
1554 ip6_cork_release(inet_sk(sk), inet6_sk(sk));
1556 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);