2 * IPv6 output functions
3 * Linux INET6 implementation
6 * Pedro Roque <roque@di.fc.ul.pt>
8 * Based on linux/net/ipv4/ip_output.c
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
16 * A.N.Kuznetsov : airthmetics in fragmentation.
17 * extension headers are implemented.
18 * route changes now work.
19 * ip6_forward does not confuse sniffers.
22 * H. von Brand : Added missing #include <linux/string.h>
23 * Imran Patel : frag id should be in NBO
24 * Kazunori MIYAZAWA @USAGI
25 * : add ip6_append_data and related functions
29 #include <linux/errno.h>
30 #include <linux/kernel.h>
31 #include <linux/string.h>
32 #include <linux/socket.h>
33 #include <linux/net.h>
34 #include <linux/netdevice.h>
35 #include <linux/if_arp.h>
36 #include <linux/in6.h>
37 #include <linux/tcp.h>
38 #include <linux/route.h>
39 #include <linux/module.h>
40 #include <linux/slab.h>
42 #include <linux/netfilter.h>
43 #include <linux/netfilter_ipv6.h>
49 #include <net/ndisc.h>
50 #include <net/protocol.h>
51 #include <net/ip6_route.h>
52 #include <net/addrconf.h>
53 #include <net/rawv6.h>
56 #include <net/checksum.h>
57 #include <linux/mroute6.h>
59 int __ip6_local_out(struct sk_buff *skb)
63 len = skb->len - sizeof(struct ipv6hdr);
64 if (len > IPV6_MAXPLEN)
66 ipv6_hdr(skb)->payload_len = htons(len);
68 return nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, skb, NULL,
69 skb_dst(skb)->dev, dst_output);
72 int ip6_local_out(struct sk_buff *skb)
76 err = __ip6_local_out(skb);
78 err = dst_output(skb);
82 EXPORT_SYMBOL_GPL(ip6_local_out);
84 static int ip6_finish_output2(struct sk_buff *skb)
86 struct dst_entry *dst = skb_dst(skb);
87 struct net_device *dev = dst->dev;
88 struct neighbour *neigh;
89 struct in6_addr *nexthop;
92 skb->protocol = htons(ETH_P_IPV6);
95 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
96 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
98 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(skb->sk) &&
99 ((mroute6_socket(dev_net(dev), skb) &&
100 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
101 ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
102 &ipv6_hdr(skb)->saddr))) {
103 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
105 /* Do not check for IFF_ALLMULTI; multicast routing
106 is not supported in any case.
109 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
110 newskb, NULL, newskb->dev,
113 if (ipv6_hdr(skb)->hop_limit == 0) {
114 IP6_INC_STATS(dev_net(dev), idev,
115 IPSTATS_MIB_OUTDISCARDS);
121 IP6_UPD_PO_STATS(dev_net(dev), idev, IPSTATS_MIB_OUTMCAST,
124 if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb)->daddr) <=
125 IPV6_ADDR_SCOPE_NODELOCAL &&
126 !(dev->flags & IFF_LOOPBACK)) {
133 nexthop = rt6_nexthop((struct rt6_info *)dst);
134 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
135 if (unlikely(!neigh))
136 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
137 if (!IS_ERR(neigh)) {
138 ret = dst_neigh_output(dst, neigh, skb);
139 rcu_read_unlock_bh();
142 rcu_read_unlock_bh();
144 IP6_INC_STATS_BH(dev_net(dst->dev),
145 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
150 static int ip6_finish_output(struct sk_buff *skb)
152 if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
153 dst_allfrag(skb_dst(skb)) ||
154 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
155 return ip6_fragment(skb, ip6_finish_output2);
157 return ip6_finish_output2(skb);
160 int ip6_output(struct sk_buff *skb)
162 struct net_device *dev = skb_dst(skb)->dev;
163 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
164 if (unlikely(idev->cnf.disable_ipv6)) {
165 IP6_INC_STATS(dev_net(dev), idev,
166 IPSTATS_MIB_OUTDISCARDS);
171 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, skb, NULL, dev,
173 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
177 * xmit an sk_buff (used by TCP, SCTP and DCCP)
180 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
181 struct ipv6_txoptions *opt, int tclass)
183 struct net *net = sock_net(sk);
184 struct ipv6_pinfo *np = inet6_sk(sk);
185 struct in6_addr *first_hop = &fl6->daddr;
186 struct dst_entry *dst = skb_dst(skb);
188 u8 proto = fl6->flowi6_proto;
189 int seg_len = skb->len;
194 unsigned int head_room;
196 /* First: exthdrs may take lots of space (~8K for now)
197 MAX_HEADER is not enough.
199 head_room = opt->opt_nflen + opt->opt_flen;
200 seg_len += head_room;
201 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
203 if (skb_headroom(skb) < head_room) {
204 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
206 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
207 IPSTATS_MIB_OUTDISCARDS);
213 skb_set_owner_w(skb, sk);
216 ipv6_push_frag_opts(skb, opt, &proto);
218 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
221 skb_push(skb, sizeof(struct ipv6hdr));
222 skb_reset_network_header(skb);
226 * Fill in the IPv6 header
229 hlimit = np->hop_limit;
231 hlimit = ip6_dst_hoplimit(dst);
233 ip6_flow_hdr(hdr, tclass, fl6->flowlabel);
235 hdr->payload_len = htons(seg_len);
236 hdr->nexthdr = proto;
237 hdr->hop_limit = hlimit;
239 hdr->saddr = fl6->saddr;
240 hdr->daddr = *first_hop;
242 skb->priority = sk->sk_priority;
243 skb->mark = sk->sk_mark;
246 if ((skb->len <= mtu) || skb->local_df || skb_is_gso(skb)) {
247 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
248 IPSTATS_MIB_OUT, skb->len);
249 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, skb, NULL,
250 dst->dev, dst_output);
254 ipv6_local_error(sk, EMSGSIZE, fl6, mtu);
255 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
260 EXPORT_SYMBOL(ip6_xmit);
262 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
264 struct ip6_ra_chain *ra;
265 struct sock *last = NULL;
267 read_lock(&ip6_ra_lock);
268 for (ra = ip6_ra_chain; ra; ra = ra->next) {
269 struct sock *sk = ra->sk;
270 if (sk && ra->sel == sel &&
271 (!sk->sk_bound_dev_if ||
272 sk->sk_bound_dev_if == skb->dev->ifindex)) {
274 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
276 rawv6_rcv(last, skb2);
283 rawv6_rcv(last, skb);
284 read_unlock(&ip6_ra_lock);
287 read_unlock(&ip6_ra_lock);
291 static int ip6_forward_proxy_check(struct sk_buff *skb)
293 struct ipv6hdr *hdr = ipv6_hdr(skb);
294 u8 nexthdr = hdr->nexthdr;
298 if (ipv6_ext_hdr(nexthdr)) {
299 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
303 offset = sizeof(struct ipv6hdr);
305 if (nexthdr == IPPROTO_ICMPV6) {
306 struct icmp6hdr *icmp6;
308 if (!pskb_may_pull(skb, (skb_network_header(skb) +
309 offset + 1 - skb->data)))
312 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
314 switch (icmp6->icmp6_type) {
315 case NDISC_ROUTER_SOLICITATION:
316 case NDISC_ROUTER_ADVERTISEMENT:
317 case NDISC_NEIGHBOUR_SOLICITATION:
318 case NDISC_NEIGHBOUR_ADVERTISEMENT:
320 /* For reaction involving unicast neighbor discovery
321 * message destined to the proxied address, pass it to
331 * The proxying router can't forward traffic sent to a link-local
332 * address, so signal the sender and discard the packet. This
333 * behavior is clarified by the MIPv6 specification.
335 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
336 dst_link_failure(skb);
343 static inline int ip6_forward_finish(struct sk_buff *skb)
345 return dst_output(skb);
348 int ip6_forward(struct sk_buff *skb)
350 struct dst_entry *dst = skb_dst(skb);
351 struct ipv6hdr *hdr = ipv6_hdr(skb);
352 struct inet6_skb_parm *opt = IP6CB(skb);
353 struct net *net = dev_net(dst->dev);
356 if (net->ipv6.devconf_all->forwarding == 0)
359 if (skb_warn_if_lro(skb))
362 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
363 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
367 if (skb->pkt_type != PACKET_HOST)
370 skb_forward_csum(skb);
373 * We DO NOT make any processing on
374 * RA packets, pushing them to user level AS IS
375 * without ane WARRANTY that application will be able
376 * to interpret them. The reason is that we
377 * cannot make anything clever here.
379 * We are not end-node, so that if packet contains
380 * AH/ESP, we cannot make anything.
381 * Defragmentation also would be mistake, RA packets
382 * cannot be fragmented, because there is no warranty
383 * that different fragments will go along one path. --ANK
385 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
386 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
391 * check and decrement ttl
393 if (hdr->hop_limit <= 1) {
394 /* Force OUTPUT device used as source address */
396 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
397 IP6_INC_STATS_BH(net,
398 ip6_dst_idev(dst), IPSTATS_MIB_INHDRERRORS);
404 /* XXX: idev->cnf.proxy_ndp? */
405 if (net->ipv6.devconf_all->proxy_ndp &&
406 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
407 int proxied = ip6_forward_proxy_check(skb);
409 return ip6_input(skb);
410 else if (proxied < 0) {
411 IP6_INC_STATS(net, ip6_dst_idev(dst),
412 IPSTATS_MIB_INDISCARDS);
417 if (!xfrm6_route_forward(skb)) {
418 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
423 /* IPv6 specs say nothing about it, but it is clear that we cannot
424 send redirects to source routed frames.
425 We don't send redirects to frames decapsulated from IPsec.
427 if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
428 struct in6_addr *target = NULL;
429 struct inet_peer *peer;
433 * incoming and outgoing devices are the same
437 rt = (struct rt6_info *) dst;
438 if (rt->rt6i_flags & RTF_GATEWAY)
439 target = &rt->rt6i_gateway;
441 target = &hdr->daddr;
443 peer = inet_getpeer_v6(net->ipv6.peers, &rt->rt6i_dst.addr, 1);
445 /* Limit redirects both by destination (here)
446 and by source (inside ndisc_send_redirect)
448 if (inet_peer_xrlim_allow(peer, 1*HZ))
449 ndisc_send_redirect(skb, target);
453 int addrtype = ipv6_addr_type(&hdr->saddr);
455 /* This check is security critical. */
456 if (addrtype == IPV6_ADDR_ANY ||
457 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
459 if (addrtype & IPV6_ADDR_LINKLOCAL) {
460 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
461 ICMPV6_NOT_NEIGHBOUR, 0);
467 if (mtu < IPV6_MIN_MTU)
470 if ((!skb->local_df && skb->len > mtu && !skb_is_gso(skb)) ||
471 (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)) {
472 /* Again, force OUTPUT device used as source address */
474 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
475 IP6_INC_STATS_BH(net,
476 ip6_dst_idev(dst), IPSTATS_MIB_INTOOBIGERRORS);
477 IP6_INC_STATS_BH(net,
478 ip6_dst_idev(dst), IPSTATS_MIB_FRAGFAILS);
483 if (skb_cow(skb, dst->dev->hard_header_len)) {
484 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTDISCARDS);
490 /* Mangling hops number delayed to point after skb COW */
494 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
495 IP6_ADD_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
496 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD, skb, skb->dev, dst->dev,
500 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
506 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
508 to->pkt_type = from->pkt_type;
509 to->priority = from->priority;
510 to->protocol = from->protocol;
512 skb_dst_set(to, dst_clone(skb_dst(from)));
514 to->mark = from->mark;
516 #ifdef CONFIG_NET_SCHED
517 to->tc_index = from->tc_index;
520 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
521 to->nf_trace = from->nf_trace;
523 skb_copy_secmark(to, from);
526 int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
528 struct sk_buff *frag;
529 struct rt6_info *rt = (struct rt6_info*)skb_dst(skb);
530 struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
531 struct ipv6hdr *tmp_hdr;
533 unsigned int mtu, hlen, left, len;
536 int ptr, offset = 0, err=0;
537 u8 *prevhdr, nexthdr = 0;
538 struct net *net = dev_net(skb_dst(skb)->dev);
540 hlen = ip6_find_1stfragopt(skb, &prevhdr);
543 mtu = ip6_skb_dst_mtu(skb);
545 /* We must not fragment if the socket is set to force MTU discovery
546 * or if the skb it not generated by a local socket.
548 if (unlikely(!skb->local_df && skb->len > mtu) ||
549 (IP6CB(skb)->frag_max_size &&
550 IP6CB(skb)->frag_max_size > mtu)) {
551 if (skb->sk && dst_allfrag(skb_dst(skb)))
552 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
554 skb->dev = skb_dst(skb)->dev;
555 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
556 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
557 IPSTATS_MIB_FRAGFAILS);
562 if (np && np->frag_size < mtu) {
566 mtu -= hlen + sizeof(struct frag_hdr);
568 if (skb_has_frag_list(skb)) {
569 int first_len = skb_pagelen(skb);
570 struct sk_buff *frag2;
572 if (first_len - hlen > mtu ||
573 ((first_len - hlen) & 7) ||
577 skb_walk_frags(skb, frag) {
578 /* Correct geometry. */
579 if (frag->len > mtu ||
580 ((frag->len & 7) && frag->next) ||
581 skb_headroom(frag) < hlen)
582 goto slow_path_clean;
584 /* Partially cloned skb? */
585 if (skb_shared(frag))
586 goto slow_path_clean;
591 frag->destructor = sock_wfree;
593 skb->truesize -= frag->truesize;
598 frag = skb_shinfo(skb)->frag_list;
599 skb_frag_list_init(skb);
602 *prevhdr = NEXTHDR_FRAGMENT;
603 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
605 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
606 IPSTATS_MIB_FRAGFAILS);
610 __skb_pull(skb, hlen);
611 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
612 __skb_push(skb, hlen);
613 skb_reset_network_header(skb);
614 memcpy(skb_network_header(skb), tmp_hdr, hlen);
616 ipv6_select_ident(fh, rt);
617 fh->nexthdr = nexthdr;
619 fh->frag_off = htons(IP6_MF);
620 frag_id = fh->identification;
622 first_len = skb_pagelen(skb);
623 skb->data_len = first_len - skb_headlen(skb);
624 skb->len = first_len;
625 ipv6_hdr(skb)->payload_len = htons(first_len -
626 sizeof(struct ipv6hdr));
631 /* Prepare header of the next frame,
632 * before previous one went down. */
634 frag->ip_summed = CHECKSUM_NONE;
635 skb_reset_transport_header(frag);
636 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
637 __skb_push(frag, hlen);
638 skb_reset_network_header(frag);
639 memcpy(skb_network_header(frag), tmp_hdr,
641 offset += skb->len - hlen - sizeof(struct frag_hdr);
642 fh->nexthdr = nexthdr;
644 fh->frag_off = htons(offset);
645 if (frag->next != NULL)
646 fh->frag_off |= htons(IP6_MF);
647 fh->identification = frag_id;
648 ipv6_hdr(frag)->payload_len =
650 sizeof(struct ipv6hdr));
651 ip6_copy_metadata(frag, skb);
656 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
657 IPSTATS_MIB_FRAGCREATES);
670 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
671 IPSTATS_MIB_FRAGOKS);
682 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
683 IPSTATS_MIB_FRAGFAILS);
688 skb_walk_frags(skb, frag2) {
692 frag2->destructor = NULL;
693 skb->truesize += frag2->truesize;
698 if ((skb->ip_summed == CHECKSUM_PARTIAL) &&
699 skb_checksum_help(skb))
702 left = skb->len - hlen; /* Space per frame */
703 ptr = hlen; /* Where to start from */
706 * Fragment the datagram.
709 *prevhdr = NEXTHDR_FRAGMENT;
710 hroom = LL_RESERVED_SPACE(rt->dst.dev);
711 troom = rt->dst.dev->needed_tailroom;
714 * Keep copying data until we run out.
718 /* IF: it doesn't fit, use 'mtu' - the data space left */
721 /* IF: we are not sending up to and including the packet end
722 then align the next start on an eight byte boundary */
730 if ((frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
731 hroom + troom, GFP_ATOMIC)) == NULL) {
732 NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
733 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
734 IPSTATS_MIB_FRAGFAILS);
740 * Set up data on packet
743 ip6_copy_metadata(frag, skb);
744 skb_reserve(frag, hroom);
745 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
746 skb_reset_network_header(frag);
747 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
748 frag->transport_header = (frag->network_header + hlen +
749 sizeof(struct frag_hdr));
752 * Charge the memory for the fragment to any owner
756 skb_set_owner_w(frag, skb->sk);
759 * Copy the packet header into the new buffer.
761 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
764 * Build fragment header.
766 fh->nexthdr = nexthdr;
769 ipv6_select_ident(fh, rt);
770 frag_id = fh->identification;
772 fh->identification = frag_id;
775 * Copy a block of the IP datagram.
777 if (skb_copy_bits(skb, ptr, skb_transport_header(frag), len))
781 fh->frag_off = htons(offset);
783 fh->frag_off |= htons(IP6_MF);
784 ipv6_hdr(frag)->payload_len = htons(frag->len -
785 sizeof(struct ipv6hdr));
791 * Put this fragment into the sending queue.
797 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
798 IPSTATS_MIB_FRAGCREATES);
800 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
801 IPSTATS_MIB_FRAGOKS);
806 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
807 IPSTATS_MIB_FRAGFAILS);
812 static inline int ip6_rt_check(const struct rt6key *rt_key,
813 const struct in6_addr *fl_addr,
814 const struct in6_addr *addr_cache)
816 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
817 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache));
820 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
821 struct dst_entry *dst,
822 const struct flowi6 *fl6)
824 struct ipv6_pinfo *np = inet6_sk(sk);
830 if (dst->ops->family != AF_INET6) {
835 rt = (struct rt6_info *)dst;
836 /* Yes, checking route validity in not connected
837 * case is not very simple. Take into account,
838 * that we do not support routing by source, TOS,
839 * and MSG_DONTROUTE --ANK (980726)
841 * 1. ip6_rt_check(): If route was host route,
842 * check that cached destination is current.
843 * If it is network route, we still may
844 * check its validity using saved pointer
845 * to the last used address: daddr_cache.
846 * We do not want to save whole address now,
847 * (because main consumer of this service
848 * is tcp, which has not this problem),
849 * so that the last trick works only on connected
851 * 2. oif also should be the same.
853 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
854 #ifdef CONFIG_IPV6_SUBTREES
855 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
857 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
866 static int ip6_dst_lookup_tail(struct sock *sk,
867 struct dst_entry **dst, struct flowi6 *fl6)
869 struct net *net = sock_net(sk);
870 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
877 *dst = ip6_route_output(net, sk, fl6);
879 if ((err = (*dst)->error))
880 goto out_err_release;
882 if (ipv6_addr_any(&fl6->saddr)) {
883 struct rt6_info *rt = (struct rt6_info *) *dst;
884 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
885 sk ? inet6_sk(sk)->srcprefs : 0,
888 goto out_err_release;
891 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
893 * Here if the dst entry we've looked up
894 * has a neighbour entry that is in the INCOMPLETE
895 * state and the src address from the flow is
896 * marked as OPTIMISTIC, we release the found
897 * dst entry and replace it instead with the
898 * dst entry of the nexthop router
900 rt = (struct rt6_info *) *dst;
902 n = __ipv6_neigh_lookup_noref(rt->dst.dev, rt6_nexthop(rt));
903 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
904 rcu_read_unlock_bh();
907 struct inet6_ifaddr *ifp;
908 struct flowi6 fl_gw6;
911 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
914 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
920 * We need to get the dst entry for the
921 * default router instead
924 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
925 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
926 *dst = ip6_route_output(net, sk, &fl_gw6);
927 if ((err = (*dst)->error))
928 goto out_err_release;
936 if (err == -ENETUNREACH)
937 IP6_INC_STATS_BH(net, NULL, IPSTATS_MIB_OUTNOROUTES);
944 * ip6_dst_lookup - perform route lookup on flow
945 * @sk: socket which provides route info
946 * @dst: pointer to dst_entry * for result
947 * @fl6: flow to lookup
949 * This function performs a route lookup on the given flow.
951 * It returns zero on success, or a standard errno code on error.
953 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi6 *fl6)
956 return ip6_dst_lookup_tail(sk, dst, fl6);
958 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
961 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
962 * @sk: socket which provides route info
963 * @fl6: flow to lookup
964 * @final_dst: final destination address for ipsec lookup
965 * @can_sleep: we are in a sleepable context
967 * This function performs a route lookup on the given flow.
969 * It returns a valid dst pointer on success, or a pointer encoded
972 struct dst_entry *ip6_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
973 const struct in6_addr *final_dst,
976 struct dst_entry *dst = NULL;
979 err = ip6_dst_lookup_tail(sk, &dst, fl6);
983 fl6->daddr = *final_dst;
985 fl6->flowi6_flags |= FLOWI_FLAG_CAN_SLEEP;
987 return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
989 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
992 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
993 * @sk: socket which provides the dst cache and route info
994 * @fl6: flow to lookup
995 * @final_dst: final destination address for ipsec lookup
996 * @can_sleep: we are in a sleepable context
998 * This function performs a route lookup on the given flow with the
999 * possibility of using the cached route in the socket if it is valid.
1000 * It will take the socket dst lock when operating on the dst cache.
1001 * As a result, this function can only be used in process context.
1003 * It returns a valid dst pointer on success, or a pointer encoded
1006 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1007 const struct in6_addr *final_dst,
1010 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1013 dst = ip6_sk_dst_check(sk, dst, fl6);
1015 err = ip6_dst_lookup_tail(sk, &dst, fl6);
1017 return ERR_PTR(err);
1019 fl6->daddr = *final_dst;
1021 fl6->flowi6_flags |= FLOWI_FLAG_CAN_SLEEP;
1023 return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1025 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1027 static inline int ip6_ufo_append_data(struct sock *sk,
1028 int getfrag(void *from, char *to, int offset, int len,
1029 int odd, struct sk_buff *skb),
1030 void *from, int length, int hh_len, int fragheaderlen,
1031 int transhdrlen, int mtu,unsigned int flags,
1032 struct rt6_info *rt)
1035 struct sk_buff *skb;
1038 /* There is support for UDP large send offload by network
1039 * device, so create one single skb packet containing complete
1042 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
1043 struct frag_hdr fhdr;
1045 skb = sock_alloc_send_skb(sk,
1046 hh_len + fragheaderlen + transhdrlen + 20,
1047 (flags & MSG_DONTWAIT), &err);
1051 /* reserve space for Hardware header */
1052 skb_reserve(skb, hh_len);
1054 /* create space for UDP/IP header */
1055 skb_put(skb,fragheaderlen + transhdrlen);
1057 /* initialize network header pointer */
1058 skb_reset_network_header(skb);
1060 /* initialize protocol header pointer */
1061 skb->transport_header = skb->network_header + fragheaderlen;
1063 skb->ip_summed = CHECKSUM_PARTIAL;
1066 /* Specify the length of each IPv6 datagram fragment.
1067 * It has to be a multiple of 8.
1069 skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1070 sizeof(struct frag_hdr)) & ~7;
1071 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1072 ipv6_select_ident(&fhdr, rt);
1073 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1074 __skb_queue_tail(&sk->sk_write_queue, skb);
1077 return skb_append_datato_frags(sk, skb, getfrag, from,
1078 (length - transhdrlen));
1081 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1084 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1087 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1090 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1093 static void ip6_append_data_mtu(unsigned int *mtu,
1095 unsigned int fragheaderlen,
1096 struct sk_buff *skb,
1097 struct rt6_info *rt,
1100 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1102 /* first fragment, reserve header_len */
1103 *mtu = *mtu - rt->dst.header_len;
1107 * this fragment is not first, the headers
1108 * space is regarded as data space.
1110 *mtu = min(*mtu, pmtuprobe ?
1112 dst_mtu(rt->dst.path));
1114 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1115 + fragheaderlen - sizeof(struct frag_hdr);
1119 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1120 int offset, int len, int odd, struct sk_buff *skb),
1121 void *from, int length, int transhdrlen,
1122 int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1123 struct rt6_info *rt, unsigned int flags, int dontfrag)
1125 struct inet_sock *inet = inet_sk(sk);
1126 struct ipv6_pinfo *np = inet6_sk(sk);
1127 struct inet_cork *cork;
1128 struct sk_buff *skb, *skb_prev = NULL;
1129 unsigned int maxfraglen, fragheaderlen, mtu;
1138 if (flags&MSG_PROBE)
1140 cork = &inet->cork.base;
1141 if (skb_queue_empty(&sk->sk_write_queue)) {
1146 if (WARN_ON(np->cork.opt))
1149 np->cork.opt = kzalloc(opt->tot_len, sk->sk_allocation);
1150 if (unlikely(np->cork.opt == NULL))
1153 np->cork.opt->tot_len = opt->tot_len;
1154 np->cork.opt->opt_flen = opt->opt_flen;
1155 np->cork.opt->opt_nflen = opt->opt_nflen;
1157 np->cork.opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1159 if (opt->dst0opt && !np->cork.opt->dst0opt)
1162 np->cork.opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1164 if (opt->dst1opt && !np->cork.opt->dst1opt)
1167 np->cork.opt->hopopt = ip6_opt_dup(opt->hopopt,
1169 if (opt->hopopt && !np->cork.opt->hopopt)
1172 np->cork.opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1174 if (opt->srcrt && !np->cork.opt->srcrt)
1177 /* need source address above miyazawa*/
1180 cork->dst = &rt->dst;
1181 inet->cork.fl.u.ip6 = *fl6;
1182 np->cork.hop_limit = hlimit;
1183 np->cork.tclass = tclass;
1184 if (rt->dst.flags & DST_XFRM_TUNNEL)
1185 mtu = np->pmtudisc == IPV6_PMTUDISC_PROBE ?
1186 rt->dst.dev->mtu : dst_mtu(&rt->dst);
1188 mtu = np->pmtudisc == IPV6_PMTUDISC_PROBE ?
1189 rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1190 if (np->frag_size < mtu) {
1192 mtu = np->frag_size;
1194 cork->fragsize = mtu;
1195 if (dst_allfrag(rt->dst.path))
1196 cork->flags |= IPCORK_ALLFRAG;
1198 exthdrlen = (opt ? opt->opt_flen : 0);
1199 length += exthdrlen;
1200 transhdrlen += exthdrlen;
1201 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1203 rt = (struct rt6_info *)cork->dst;
1204 fl6 = &inet->cork.fl.u.ip6;
1209 mtu = cork->fragsize;
1212 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1214 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1215 (opt ? opt->opt_nflen : 0);
1216 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
1218 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1219 if (cork->length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
1220 ipv6_local_error(sk, EMSGSIZE, fl6, mtu-exthdrlen);
1225 /* For UDP, check if TX timestamp is enabled */
1226 if (sk->sk_type == SOCK_DGRAM)
1227 sock_tx_timestamp(sk, &tx_flags);
1230 * Let's try using as much space as possible.
1231 * Use MTU if total length of the message fits into the MTU.
1232 * Otherwise, we need to reserve fragment header and
1233 * fragment alignment (= 8-15 octects, in total).
1235 * Note that we may need to "move" the data from the tail of
1236 * of the buffer to the new fragment when we split
1239 * FIXME: It may be fragmented into multiple chunks
1240 * at once if non-fragmentable extension headers
1245 if ((length > mtu) && dontfrag && (sk->sk_protocol == IPPROTO_UDP ||
1246 sk->sk_protocol == IPPROTO_RAW)) {
1247 ipv6_local_rxpmtu(sk, fl6, mtu-exthdrlen);
1251 skb = skb_peek_tail(&sk->sk_write_queue);
1252 cork->length += length;
1253 if (((length > mtu) ||
1254 (skb && skb_has_frags(skb))) &&
1255 (sk->sk_protocol == IPPROTO_UDP) &&
1256 (rt->dst.dev->features & NETIF_F_UFO)) {
1257 err = ip6_ufo_append_data(sk, getfrag, from, length,
1258 hh_len, fragheaderlen,
1259 transhdrlen, mtu, flags, rt);
1268 while (length > 0) {
1269 /* Check if the remaining data fits into current packet. */
1270 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1272 copy = maxfraglen - skb->len;
1276 unsigned int datalen;
1277 unsigned int fraglen;
1278 unsigned int fraggap;
1279 unsigned int alloclen;
1281 /* There's no room in the current skb */
1283 fraggap = skb->len - maxfraglen;
1286 /* update mtu and maxfraglen if necessary */
1287 if (skb == NULL || skb_prev == NULL)
1288 ip6_append_data_mtu(&mtu, &maxfraglen,
1289 fragheaderlen, skb, rt,
1291 IPV6_PMTUDISC_PROBE);
1296 * If remaining data exceeds the mtu,
1297 * we know we need more fragment(s).
1299 datalen = length + fraggap;
1301 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1302 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1303 if ((flags & MSG_MORE) &&
1304 !(rt->dst.dev->features&NETIF_F_SG))
1307 alloclen = datalen + fragheaderlen;
1309 alloclen += dst_exthdrlen;
1311 if (datalen != length + fraggap) {
1313 * this is not the last fragment, the trailer
1314 * space is regarded as data space.
1316 datalen += rt->dst.trailer_len;
1319 alloclen += rt->dst.trailer_len;
1320 fraglen = datalen + fragheaderlen;
1323 * We just reserve space for fragment header.
1324 * Note: this may be overallocation if the message
1325 * (without MSG_MORE) fits into the MTU.
1327 alloclen += sizeof(struct frag_hdr);
1330 skb = sock_alloc_send_skb(sk,
1332 (flags & MSG_DONTWAIT), &err);
1335 if (atomic_read(&sk->sk_wmem_alloc) <=
1337 skb = sock_wmalloc(sk,
1338 alloclen + hh_len, 1,
1340 if (unlikely(skb == NULL))
1343 /* Only the initial fragment
1352 * Fill in the control structures
1354 skb->ip_summed = CHECKSUM_NONE;
1356 /* reserve for fragmentation and ipsec header */
1357 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1360 if (sk->sk_type == SOCK_DGRAM)
1361 skb_shinfo(skb)->tx_flags = tx_flags;
1364 * Find where to start putting bytes
1366 data = skb_put(skb, fraglen);
1367 skb_set_network_header(skb, exthdrlen);
1368 data += fragheaderlen;
1369 skb->transport_header = (skb->network_header +
1372 skb->csum = skb_copy_and_csum_bits(
1373 skb_prev, maxfraglen,
1374 data + transhdrlen, fraggap, 0);
1375 skb_prev->csum = csum_sub(skb_prev->csum,
1378 pskb_trim_unique(skb_prev, maxfraglen);
1380 copy = datalen - transhdrlen - fraggap;
1386 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1393 length -= datalen - fraggap;
1399 * Put the packet on the pending queue
1401 __skb_queue_tail(&sk->sk_write_queue, skb);
1408 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1412 if (getfrag(from, skb_put(skb, copy),
1413 offset, copy, off, skb) < 0) {
1414 __skb_trim(skb, off);
1419 int i = skb_shinfo(skb)->nr_frags;
1420 struct page_frag *pfrag = sk_page_frag(sk);
1423 if (!sk_page_frag_refill(sk, pfrag))
1426 if (!skb_can_coalesce(skb, i, pfrag->page,
1429 if (i == MAX_SKB_FRAGS)
1432 __skb_fill_page_desc(skb, i, pfrag->page,
1434 skb_shinfo(skb)->nr_frags = ++i;
1435 get_page(pfrag->page);
1437 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1439 page_address(pfrag->page) + pfrag->offset,
1440 offset, copy, skb->len, skb) < 0)
1443 pfrag->offset += copy;
1444 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1446 skb->data_len += copy;
1447 skb->truesize += copy;
1448 atomic_add(copy, &sk->sk_wmem_alloc);
1459 cork->length -= length;
1460 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1463 EXPORT_SYMBOL_GPL(ip6_append_data);
1465 static void ip6_cork_release(struct inet_sock *inet, struct ipv6_pinfo *np)
1468 kfree(np->cork.opt->dst0opt);
1469 kfree(np->cork.opt->dst1opt);
1470 kfree(np->cork.opt->hopopt);
1471 kfree(np->cork.opt->srcrt);
1472 kfree(np->cork.opt);
1473 np->cork.opt = NULL;
1476 if (inet->cork.base.dst) {
1477 dst_release(inet->cork.base.dst);
1478 inet->cork.base.dst = NULL;
1479 inet->cork.base.flags &= ~IPCORK_ALLFRAG;
1481 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1484 int ip6_push_pending_frames(struct sock *sk)
1486 struct sk_buff *skb, *tmp_skb;
1487 struct sk_buff **tail_skb;
1488 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1489 struct inet_sock *inet = inet_sk(sk);
1490 struct ipv6_pinfo *np = inet6_sk(sk);
1491 struct net *net = sock_net(sk);
1492 struct ipv6hdr *hdr;
1493 struct ipv6_txoptions *opt = np->cork.opt;
1494 struct rt6_info *rt = (struct rt6_info *)inet->cork.base.dst;
1495 struct flowi6 *fl6 = &inet->cork.fl.u.ip6;
1496 unsigned char proto = fl6->flowi6_proto;
1499 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1501 tail_skb = &(skb_shinfo(skb)->frag_list);
1503 /* move skb->data to ip header from ext header */
1504 if (skb->data < skb_network_header(skb))
1505 __skb_pull(skb, skb_network_offset(skb));
1506 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1507 __skb_pull(tmp_skb, skb_network_header_len(skb));
1508 *tail_skb = tmp_skb;
1509 tail_skb = &(tmp_skb->next);
1510 skb->len += tmp_skb->len;
1511 skb->data_len += tmp_skb->len;
1512 skb->truesize += tmp_skb->truesize;
1513 tmp_skb->destructor = NULL;
1517 /* Allow local fragmentation. */
1518 if (np->pmtudisc < IPV6_PMTUDISC_DO)
1521 *final_dst = fl6->daddr;
1522 __skb_pull(skb, skb_network_header_len(skb));
1523 if (opt && opt->opt_flen)
1524 ipv6_push_frag_opts(skb, opt, &proto);
1525 if (opt && opt->opt_nflen)
1526 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1528 skb_push(skb, sizeof(struct ipv6hdr));
1529 skb_reset_network_header(skb);
1530 hdr = ipv6_hdr(skb);
1532 ip6_flow_hdr(hdr, np->cork.tclass, fl6->flowlabel);
1533 hdr->hop_limit = np->cork.hop_limit;
1534 hdr->nexthdr = proto;
1535 hdr->saddr = fl6->saddr;
1536 hdr->daddr = *final_dst;
1538 skb->priority = sk->sk_priority;
1539 skb->mark = sk->sk_mark;
1541 skb_dst_set(skb, dst_clone(&rt->dst));
1542 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1543 if (proto == IPPROTO_ICMPV6) {
1544 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1546 ICMP6MSGOUT_INC_STATS_BH(net, idev, icmp6_hdr(skb)->icmp6_type);
1547 ICMP6_INC_STATS_BH(net, idev, ICMP6_MIB_OUTMSGS);
1550 err = ip6_local_out(skb);
1553 err = net_xmit_errno(err);
1559 ip6_cork_release(inet, np);
1562 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1565 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1567 void ip6_flush_pending_frames(struct sock *sk)
1569 struct sk_buff *skb;
1571 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1573 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1574 IPSTATS_MIB_OUTDISCARDS);
1578 ip6_cork_release(inet_sk(sk), inet6_sk(sk));
1580 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);