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, &ipv6_hdr(skb)->daddr);
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 return ip6_fragment(skb, ip6_finish_output2);
156 return ip6_finish_output2(skb);
159 int ip6_output(struct sk_buff *skb)
161 struct net_device *dev = skb_dst(skb)->dev;
162 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
163 if (unlikely(idev->cnf.disable_ipv6)) {
164 IP6_INC_STATS(dev_net(dev), idev,
165 IPSTATS_MIB_OUTDISCARDS);
170 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, skb, NULL, dev,
172 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
176 * xmit an sk_buff (used by TCP, SCTP and DCCP)
179 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
180 struct ipv6_txoptions *opt, int tclass)
182 struct net *net = sock_net(sk);
183 struct ipv6_pinfo *np = inet6_sk(sk);
184 struct in6_addr *first_hop = &fl6->daddr;
185 struct dst_entry *dst = skb_dst(skb);
187 u8 proto = fl6->flowi6_proto;
188 int seg_len = skb->len;
193 unsigned int head_room;
195 /* First: exthdrs may take lots of space (~8K for now)
196 MAX_HEADER is not enough.
198 head_room = opt->opt_nflen + opt->opt_flen;
199 seg_len += head_room;
200 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
202 if (skb_headroom(skb) < head_room) {
203 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
205 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
206 IPSTATS_MIB_OUTDISCARDS);
212 skb_set_owner_w(skb, sk);
215 ipv6_push_frag_opts(skb, opt, &proto);
217 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
220 skb_push(skb, sizeof(struct ipv6hdr));
221 skb_reset_network_header(skb);
225 * Fill in the IPv6 header
228 hlimit = np->hop_limit;
230 hlimit = ip6_dst_hoplimit(dst);
232 ip6_flow_hdr(hdr, tclass, fl6->flowlabel);
234 hdr->payload_len = htons(seg_len);
235 hdr->nexthdr = proto;
236 hdr->hop_limit = hlimit;
238 hdr->saddr = fl6->saddr;
239 hdr->daddr = *first_hop;
241 skb->priority = sk->sk_priority;
242 skb->mark = sk->sk_mark;
245 if ((skb->len <= mtu) || skb->local_df || skb_is_gso(skb)) {
246 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
247 IPSTATS_MIB_OUT, skb->len);
248 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, skb, NULL,
249 dst->dev, dst_output);
253 ipv6_local_error(sk, EMSGSIZE, fl6, mtu);
254 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
259 EXPORT_SYMBOL(ip6_xmit);
261 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
263 struct ip6_ra_chain *ra;
264 struct sock *last = NULL;
266 read_lock(&ip6_ra_lock);
267 for (ra = ip6_ra_chain; ra; ra = ra->next) {
268 struct sock *sk = ra->sk;
269 if (sk && ra->sel == sel &&
270 (!sk->sk_bound_dev_if ||
271 sk->sk_bound_dev_if == skb->dev->ifindex)) {
273 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
275 rawv6_rcv(last, skb2);
282 rawv6_rcv(last, skb);
283 read_unlock(&ip6_ra_lock);
286 read_unlock(&ip6_ra_lock);
290 static int ip6_forward_proxy_check(struct sk_buff *skb)
292 struct ipv6hdr *hdr = ipv6_hdr(skb);
293 u8 nexthdr = hdr->nexthdr;
297 if (ipv6_ext_hdr(nexthdr)) {
298 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
302 offset = sizeof(struct ipv6hdr);
304 if (nexthdr == IPPROTO_ICMPV6) {
305 struct icmp6hdr *icmp6;
307 if (!pskb_may_pull(skb, (skb_network_header(skb) +
308 offset + 1 - skb->data)))
311 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
313 switch (icmp6->icmp6_type) {
314 case NDISC_ROUTER_SOLICITATION:
315 case NDISC_ROUTER_ADVERTISEMENT:
316 case NDISC_NEIGHBOUR_SOLICITATION:
317 case NDISC_NEIGHBOUR_ADVERTISEMENT:
319 /* For reaction involving unicast neighbor discovery
320 * message destined to the proxied address, pass it to
330 * The proxying router can't forward traffic sent to a link-local
331 * address, so signal the sender and discard the packet. This
332 * behavior is clarified by the MIPv6 specification.
334 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
335 dst_link_failure(skb);
342 static inline int ip6_forward_finish(struct sk_buff *skb)
344 return dst_output(skb);
347 int ip6_forward(struct sk_buff *skb)
349 struct dst_entry *dst = skb_dst(skb);
350 struct ipv6hdr *hdr = ipv6_hdr(skb);
351 struct inet6_skb_parm *opt = IP6CB(skb);
352 struct net *net = dev_net(dst->dev);
355 if (net->ipv6.devconf_all->forwarding == 0)
358 if (skb_warn_if_lro(skb))
361 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
362 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
366 if (skb->pkt_type != PACKET_HOST)
369 skb_forward_csum(skb);
372 * We DO NOT make any processing on
373 * RA packets, pushing them to user level AS IS
374 * without ane WARRANTY that application will be able
375 * to interpret them. The reason is that we
376 * cannot make anything clever here.
378 * We are not end-node, so that if packet contains
379 * AH/ESP, we cannot make anything.
380 * Defragmentation also would be mistake, RA packets
381 * cannot be fragmented, because there is no warranty
382 * that different fragments will go along one path. --ANK
384 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
385 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
390 * check and decrement ttl
392 if (hdr->hop_limit <= 1) {
393 /* Force OUTPUT device used as source address */
395 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
396 IP6_INC_STATS_BH(net,
397 ip6_dst_idev(dst), IPSTATS_MIB_INHDRERRORS);
403 /* XXX: idev->cnf.proxy_ndp? */
404 if (net->ipv6.devconf_all->proxy_ndp &&
405 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
406 int proxied = ip6_forward_proxy_check(skb);
408 return ip6_input(skb);
409 else if (proxied < 0) {
410 IP6_INC_STATS(net, ip6_dst_idev(dst),
411 IPSTATS_MIB_INDISCARDS);
416 if (!xfrm6_route_forward(skb)) {
417 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
422 /* IPv6 specs say nothing about it, but it is clear that we cannot
423 send redirects to source routed frames.
424 We don't send redirects to frames decapsulated from IPsec.
426 if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
427 struct in6_addr *target = NULL;
428 struct inet_peer *peer;
432 * incoming and outgoing devices are the same
436 rt = (struct rt6_info *) dst;
437 if (rt->rt6i_flags & RTF_GATEWAY)
438 target = &rt->rt6i_gateway;
440 target = &hdr->daddr;
442 peer = inet_getpeer_v6(net->ipv6.peers, &rt->rt6i_dst.addr, 1);
444 /* Limit redirects both by destination (here)
445 and by source (inside ndisc_send_redirect)
447 if (inet_peer_xrlim_allow(peer, 1*HZ))
448 ndisc_send_redirect(skb, target);
452 int addrtype = ipv6_addr_type(&hdr->saddr);
454 /* This check is security critical. */
455 if (addrtype == IPV6_ADDR_ANY ||
456 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
458 if (addrtype & IPV6_ADDR_LINKLOCAL) {
459 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
460 ICMPV6_NOT_NEIGHBOUR, 0);
466 if (mtu < IPV6_MIN_MTU)
469 if ((!skb->local_df && skb->len > mtu && !skb_is_gso(skb)) ||
470 (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)) {
471 /* Again, force OUTPUT device used as source address */
473 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
474 IP6_INC_STATS_BH(net,
475 ip6_dst_idev(dst), IPSTATS_MIB_INTOOBIGERRORS);
476 IP6_INC_STATS_BH(net,
477 ip6_dst_idev(dst), IPSTATS_MIB_FRAGFAILS);
482 if (skb_cow(skb, dst->dev->hard_header_len)) {
483 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTDISCARDS);
489 /* Mangling hops number delayed to point after skb COW */
493 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
494 IP6_ADD_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
495 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD, skb, skb->dev, dst->dev,
499 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
505 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
507 to->pkt_type = from->pkt_type;
508 to->priority = from->priority;
509 to->protocol = from->protocol;
511 skb_dst_set(to, dst_clone(skb_dst(from)));
513 to->mark = from->mark;
515 #ifdef CONFIG_NET_SCHED
516 to->tc_index = from->tc_index;
519 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
520 to->nf_trace = from->nf_trace;
522 skb_copy_secmark(to, from);
525 int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
527 struct sk_buff *frag;
528 struct rt6_info *rt = (struct rt6_info*)skb_dst(skb);
529 struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
530 struct ipv6hdr *tmp_hdr;
532 unsigned int mtu, hlen, left, len;
535 int ptr, offset = 0, err=0;
536 u8 *prevhdr, nexthdr = 0;
537 struct net *net = dev_net(skb_dst(skb)->dev);
539 hlen = ip6_find_1stfragopt(skb, &prevhdr);
542 mtu = ip6_skb_dst_mtu(skb);
544 /* We must not fragment if the socket is set to force MTU discovery
545 * or if the skb it not generated by a local socket.
547 if (unlikely(!skb->local_df && skb->len > mtu) ||
548 (IP6CB(skb)->frag_max_size &&
549 IP6CB(skb)->frag_max_size > mtu)) {
550 if (skb->sk && dst_allfrag(skb_dst(skb)))
551 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
553 skb->dev = skb_dst(skb)->dev;
554 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
555 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
556 IPSTATS_MIB_FRAGFAILS);
561 if (np && np->frag_size < mtu) {
565 mtu -= hlen + sizeof(struct frag_hdr);
567 if (skb_has_frag_list(skb)) {
568 int first_len = skb_pagelen(skb);
569 struct sk_buff *frag2;
571 if (first_len - hlen > mtu ||
572 ((first_len - hlen) & 7) ||
576 skb_walk_frags(skb, frag) {
577 /* Correct geometry. */
578 if (frag->len > mtu ||
579 ((frag->len & 7) && frag->next) ||
580 skb_headroom(frag) < hlen)
581 goto slow_path_clean;
583 /* Partially cloned skb? */
584 if (skb_shared(frag))
585 goto slow_path_clean;
590 frag->destructor = sock_wfree;
592 skb->truesize -= frag->truesize;
597 frag = skb_shinfo(skb)->frag_list;
598 skb_frag_list_init(skb);
601 *prevhdr = NEXTHDR_FRAGMENT;
602 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
604 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
605 IPSTATS_MIB_FRAGFAILS);
609 __skb_pull(skb, hlen);
610 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
611 __skb_push(skb, hlen);
612 skb_reset_network_header(skb);
613 memcpy(skb_network_header(skb), tmp_hdr, hlen);
615 ipv6_select_ident(fh, rt);
616 fh->nexthdr = nexthdr;
618 fh->frag_off = htons(IP6_MF);
619 frag_id = fh->identification;
621 first_len = skb_pagelen(skb);
622 skb->data_len = first_len - skb_headlen(skb);
623 skb->len = first_len;
624 ipv6_hdr(skb)->payload_len = htons(first_len -
625 sizeof(struct ipv6hdr));
630 /* Prepare header of the next frame,
631 * before previous one went down. */
633 frag->ip_summed = CHECKSUM_NONE;
634 skb_reset_transport_header(frag);
635 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
636 __skb_push(frag, hlen);
637 skb_reset_network_header(frag);
638 memcpy(skb_network_header(frag), tmp_hdr,
640 offset += skb->len - hlen - sizeof(struct frag_hdr);
641 fh->nexthdr = nexthdr;
643 fh->frag_off = htons(offset);
644 if (frag->next != NULL)
645 fh->frag_off |= htons(IP6_MF);
646 fh->identification = frag_id;
647 ipv6_hdr(frag)->payload_len =
649 sizeof(struct ipv6hdr));
650 ip6_copy_metadata(frag, skb);
655 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
656 IPSTATS_MIB_FRAGCREATES);
669 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
670 IPSTATS_MIB_FRAGOKS);
681 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
682 IPSTATS_MIB_FRAGFAILS);
687 skb_walk_frags(skb, frag2) {
691 frag2->destructor = NULL;
692 skb->truesize += frag2->truesize;
697 if ((skb->ip_summed == CHECKSUM_PARTIAL) &&
698 skb_checksum_help(skb))
701 left = skb->len - hlen; /* Space per frame */
702 ptr = hlen; /* Where to start from */
705 * Fragment the datagram.
708 *prevhdr = NEXTHDR_FRAGMENT;
709 hroom = LL_RESERVED_SPACE(rt->dst.dev);
710 troom = rt->dst.dev->needed_tailroom;
713 * Keep copying data until we run out.
717 /* IF: it doesn't fit, use 'mtu' - the data space left */
720 /* IF: we are not sending up to and including the packet end
721 then align the next start on an eight byte boundary */
729 if ((frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
730 hroom + troom, GFP_ATOMIC)) == NULL) {
731 NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
732 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
733 IPSTATS_MIB_FRAGFAILS);
739 * Set up data on packet
742 ip6_copy_metadata(frag, skb);
743 skb_reserve(frag, hroom);
744 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
745 skb_reset_network_header(frag);
746 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
747 frag->transport_header = (frag->network_header + hlen +
748 sizeof(struct frag_hdr));
751 * Charge the memory for the fragment to any owner
755 skb_set_owner_w(frag, skb->sk);
758 * Copy the packet header into the new buffer.
760 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
763 * Build fragment header.
765 fh->nexthdr = nexthdr;
768 ipv6_select_ident(fh, rt);
769 frag_id = fh->identification;
771 fh->identification = frag_id;
774 * Copy a block of the IP datagram.
776 if (skb_copy_bits(skb, ptr, skb_transport_header(frag), len))
780 fh->frag_off = htons(offset);
782 fh->frag_off |= htons(IP6_MF);
783 ipv6_hdr(frag)->payload_len = htons(frag->len -
784 sizeof(struct ipv6hdr));
790 * Put this fragment into the sending queue.
796 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
797 IPSTATS_MIB_FRAGCREATES);
799 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
800 IPSTATS_MIB_FRAGOKS);
805 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
806 IPSTATS_MIB_FRAGFAILS);
811 static inline int ip6_rt_check(const struct rt6key *rt_key,
812 const struct in6_addr *fl_addr,
813 const struct in6_addr *addr_cache)
815 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
816 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache));
819 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
820 struct dst_entry *dst,
821 const struct flowi6 *fl6)
823 struct ipv6_pinfo *np = inet6_sk(sk);
824 struct rt6_info *rt = (struct rt6_info *)dst;
829 /* Yes, checking route validity in not connected
830 * case is not very simple. Take into account,
831 * that we do not support routing by source, TOS,
832 * and MSG_DONTROUTE --ANK (980726)
834 * 1. ip6_rt_check(): If route was host route,
835 * check that cached destination is current.
836 * If it is network route, we still may
837 * check its validity using saved pointer
838 * to the last used address: daddr_cache.
839 * We do not want to save whole address now,
840 * (because main consumer of this service
841 * is tcp, which has not this problem),
842 * so that the last trick works only on connected
844 * 2. oif also should be the same.
846 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
847 #ifdef CONFIG_IPV6_SUBTREES
848 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
850 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
859 static int ip6_dst_lookup_tail(struct sock *sk,
860 struct dst_entry **dst, struct flowi6 *fl6)
862 struct net *net = sock_net(sk);
863 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
870 *dst = ip6_route_output(net, sk, fl6);
872 if ((err = (*dst)->error))
873 goto out_err_release;
875 if (ipv6_addr_any(&fl6->saddr)) {
876 struct rt6_info *rt = (struct rt6_info *) *dst;
877 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
878 sk ? inet6_sk(sk)->srcprefs : 0,
881 goto out_err_release;
884 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
886 * Here if the dst entry we've looked up
887 * has a neighbour entry that is in the INCOMPLETE
888 * state and the src address from the flow is
889 * marked as OPTIMISTIC, we release the found
890 * dst entry and replace it instead with the
891 * dst entry of the nexthop router
893 rt = (struct rt6_info *) *dst;
895 n = __ipv6_neigh_lookup_noref(rt->dst.dev, rt6_nexthop(rt, &fl6->daddr));
896 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
897 rcu_read_unlock_bh();
900 struct inet6_ifaddr *ifp;
901 struct flowi6 fl_gw6;
904 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
907 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
913 * We need to get the dst entry for the
914 * default router instead
917 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
918 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
919 *dst = ip6_route_output(net, sk, &fl_gw6);
920 if ((err = (*dst)->error))
921 goto out_err_release;
929 if (err == -ENETUNREACH)
930 IP6_INC_STATS_BH(net, NULL, IPSTATS_MIB_OUTNOROUTES);
937 * ip6_dst_lookup - perform route lookup on flow
938 * @sk: socket which provides route info
939 * @dst: pointer to dst_entry * for result
940 * @fl6: flow to lookup
942 * This function performs a route lookup on the given flow.
944 * It returns zero on success, or a standard errno code on error.
946 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi6 *fl6)
949 return ip6_dst_lookup_tail(sk, dst, fl6);
951 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
954 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
955 * @sk: socket which provides route info
956 * @fl6: flow to lookup
957 * @final_dst: final destination address for ipsec lookup
958 * @can_sleep: we are in a sleepable context
960 * This function performs a route lookup on the given flow.
962 * It returns a valid dst pointer on success, or a pointer encoded
965 struct dst_entry *ip6_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
966 const struct in6_addr *final_dst,
969 struct dst_entry *dst = NULL;
972 err = ip6_dst_lookup_tail(sk, &dst, fl6);
976 fl6->daddr = *final_dst;
978 fl6->flowi6_flags |= FLOWI_FLAG_CAN_SLEEP;
980 return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
982 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
985 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
986 * @sk: socket which provides the dst cache and route info
987 * @fl6: flow to lookup
988 * @final_dst: final destination address for ipsec lookup
989 * @can_sleep: we are in a sleepable context
991 * This function performs a route lookup on the given flow with the
992 * possibility of using the cached route in the socket if it is valid.
993 * It will take the socket dst lock when operating on the dst cache.
994 * As a result, this function can only be used in process context.
996 * It returns a valid dst pointer on success, or a pointer encoded
999 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1000 const struct in6_addr *final_dst,
1003 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1006 dst = ip6_sk_dst_check(sk, dst, fl6);
1008 err = ip6_dst_lookup_tail(sk, &dst, fl6);
1010 return ERR_PTR(err);
1012 fl6->daddr = *final_dst;
1014 fl6->flowi6_flags |= FLOWI_FLAG_CAN_SLEEP;
1016 return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1018 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1020 static inline int ip6_ufo_append_data(struct sock *sk,
1021 int getfrag(void *from, char *to, int offset, int len,
1022 int odd, struct sk_buff *skb),
1023 void *from, int length, int hh_len, int fragheaderlen,
1024 int transhdrlen, int mtu,unsigned int flags,
1025 struct rt6_info *rt)
1028 struct sk_buff *skb;
1031 /* There is support for UDP large send offload by network
1032 * device, so create one single skb packet containing complete
1035 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
1036 skb = sock_alloc_send_skb(sk,
1037 hh_len + fragheaderlen + transhdrlen + 20,
1038 (flags & MSG_DONTWAIT), &err);
1042 /* reserve space for Hardware header */
1043 skb_reserve(skb, hh_len);
1045 /* create space for UDP/IP header */
1046 skb_put(skb,fragheaderlen + transhdrlen);
1048 /* initialize network header pointer */
1049 skb_reset_network_header(skb);
1051 /* initialize protocol header pointer */
1052 skb->transport_header = skb->network_header + fragheaderlen;
1054 skb->ip_summed = CHECKSUM_PARTIAL;
1058 err = skb_append_datato_frags(sk,skb, getfrag, from,
1059 (length - transhdrlen));
1061 struct frag_hdr fhdr;
1063 /* Specify the length of each IPv6 datagram fragment.
1064 * It has to be a multiple of 8.
1066 skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1067 sizeof(struct frag_hdr)) & ~7;
1068 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1069 ipv6_select_ident(&fhdr, rt);
1070 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1071 __skb_queue_tail(&sk->sk_write_queue, skb);
1075 /* There is not enough support do UPD LSO,
1076 * so follow normal path
1083 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1086 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1089 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1092 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1095 static void ip6_append_data_mtu(int *mtu,
1097 unsigned int fragheaderlen,
1098 struct sk_buff *skb,
1099 struct rt6_info *rt)
1101 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1103 /* first fragment, reserve header_len */
1104 *mtu = *mtu - rt->dst.header_len;
1108 * this fragment is not first, the headers
1109 * space is regarded as data space.
1111 *mtu = dst_mtu(rt->dst.path);
1113 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1114 + fragheaderlen - sizeof(struct frag_hdr);
1118 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1119 int offset, int len, int odd, struct sk_buff *skb),
1120 void *from, int length, int transhdrlen,
1121 int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1122 struct rt6_info *rt, unsigned int flags, int dontfrag)
1124 struct inet_sock *inet = inet_sk(sk);
1125 struct ipv6_pinfo *np = inet6_sk(sk);
1126 struct inet_cork *cork;
1127 struct sk_buff *skb, *skb_prev = NULL;
1128 unsigned int maxfraglen, fragheaderlen;
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 cork->length += length;
1247 int proto = sk->sk_protocol;
1248 if (dontfrag && (proto == IPPROTO_UDP || proto == IPPROTO_RAW)){
1249 ipv6_local_rxpmtu(sk, fl6, mtu-exthdrlen);
1253 if (proto == IPPROTO_UDP &&
1254 (rt->dst.dev->features & NETIF_F_UFO)) {
1256 err = ip6_ufo_append_data(sk, getfrag, from, length,
1257 hh_len, fragheaderlen,
1258 transhdrlen, mtu, flags, rt);
1265 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
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);
1294 * If remaining data exceeds the mtu,
1295 * we know we need more fragment(s).
1297 datalen = length + fraggap;
1299 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1300 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1301 if ((flags & MSG_MORE) &&
1302 !(rt->dst.dev->features&NETIF_F_SG))
1305 alloclen = datalen + fragheaderlen;
1307 alloclen += dst_exthdrlen;
1309 if (datalen != length + fraggap) {
1311 * this is not the last fragment, the trailer
1312 * space is regarded as data space.
1314 datalen += rt->dst.trailer_len;
1317 alloclen += rt->dst.trailer_len;
1318 fraglen = datalen + fragheaderlen;
1321 * We just reserve space for fragment header.
1322 * Note: this may be overallocation if the message
1323 * (without MSG_MORE) fits into the MTU.
1325 alloclen += sizeof(struct frag_hdr);
1328 skb = sock_alloc_send_skb(sk,
1330 (flags & MSG_DONTWAIT), &err);
1333 if (atomic_read(&sk->sk_wmem_alloc) <=
1335 skb = sock_wmalloc(sk,
1336 alloclen + hh_len, 1,
1338 if (unlikely(skb == NULL))
1341 /* Only the initial fragment
1350 * Fill in the control structures
1352 skb->ip_summed = CHECKSUM_NONE;
1354 /* reserve for fragmentation and ipsec header */
1355 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1358 if (sk->sk_type == SOCK_DGRAM)
1359 skb_shinfo(skb)->tx_flags = tx_flags;
1362 * Find where to start putting bytes
1364 data = skb_put(skb, fraglen);
1365 skb_set_network_header(skb, exthdrlen);
1366 data += fragheaderlen;
1367 skb->transport_header = (skb->network_header +
1370 skb->csum = skb_copy_and_csum_bits(
1371 skb_prev, maxfraglen,
1372 data + transhdrlen, fraggap, 0);
1373 skb_prev->csum = csum_sub(skb_prev->csum,
1376 pskb_trim_unique(skb_prev, maxfraglen);
1378 copy = datalen - transhdrlen - fraggap;
1384 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1391 length -= datalen - fraggap;
1397 * Put the packet on the pending queue
1399 __skb_queue_tail(&sk->sk_write_queue, skb);
1406 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1410 if (getfrag(from, skb_put(skb, copy),
1411 offset, copy, off, skb) < 0) {
1412 __skb_trim(skb, off);
1417 int i = skb_shinfo(skb)->nr_frags;
1418 struct page_frag *pfrag = sk_page_frag(sk);
1421 if (!sk_page_frag_refill(sk, pfrag))
1424 if (!skb_can_coalesce(skb, i, pfrag->page,
1427 if (i == MAX_SKB_FRAGS)
1430 __skb_fill_page_desc(skb, i, pfrag->page,
1432 skb_shinfo(skb)->nr_frags = ++i;
1433 get_page(pfrag->page);
1435 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1437 page_address(pfrag->page) + pfrag->offset,
1438 offset, copy, skb->len, skb) < 0)
1441 pfrag->offset += copy;
1442 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1444 skb->data_len += copy;
1445 skb->truesize += copy;
1446 atomic_add(copy, &sk->sk_wmem_alloc);
1457 cork->length -= length;
1458 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1461 EXPORT_SYMBOL_GPL(ip6_append_data);
1463 static void ip6_cork_release(struct inet_sock *inet, struct ipv6_pinfo *np)
1466 kfree(np->cork.opt->dst0opt);
1467 kfree(np->cork.opt->dst1opt);
1468 kfree(np->cork.opt->hopopt);
1469 kfree(np->cork.opt->srcrt);
1470 kfree(np->cork.opt);
1471 np->cork.opt = NULL;
1474 if (inet->cork.base.dst) {
1475 dst_release(inet->cork.base.dst);
1476 inet->cork.base.dst = NULL;
1477 inet->cork.base.flags &= ~IPCORK_ALLFRAG;
1479 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1482 int ip6_push_pending_frames(struct sock *sk)
1484 struct sk_buff *skb, *tmp_skb;
1485 struct sk_buff **tail_skb;
1486 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1487 struct inet_sock *inet = inet_sk(sk);
1488 struct ipv6_pinfo *np = inet6_sk(sk);
1489 struct net *net = sock_net(sk);
1490 struct ipv6hdr *hdr;
1491 struct ipv6_txoptions *opt = np->cork.opt;
1492 struct rt6_info *rt = (struct rt6_info *)inet->cork.base.dst;
1493 struct flowi6 *fl6 = &inet->cork.fl.u.ip6;
1494 unsigned char proto = fl6->flowi6_proto;
1497 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1499 tail_skb = &(skb_shinfo(skb)->frag_list);
1501 /* move skb->data to ip header from ext header */
1502 if (skb->data < skb_network_header(skb))
1503 __skb_pull(skb, skb_network_offset(skb));
1504 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1505 __skb_pull(tmp_skb, skb_network_header_len(skb));
1506 *tail_skb = tmp_skb;
1507 tail_skb = &(tmp_skb->next);
1508 skb->len += tmp_skb->len;
1509 skb->data_len += tmp_skb->len;
1510 skb->truesize += tmp_skb->truesize;
1511 tmp_skb->destructor = NULL;
1515 /* Allow local fragmentation. */
1516 if (np->pmtudisc < IPV6_PMTUDISC_DO)
1519 *final_dst = fl6->daddr;
1520 __skb_pull(skb, skb_network_header_len(skb));
1521 if (opt && opt->opt_flen)
1522 ipv6_push_frag_opts(skb, opt, &proto);
1523 if (opt && opt->opt_nflen)
1524 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1526 skb_push(skb, sizeof(struct ipv6hdr));
1527 skb_reset_network_header(skb);
1528 hdr = ipv6_hdr(skb);
1530 ip6_flow_hdr(hdr, np->cork.tclass, fl6->flowlabel);
1531 hdr->hop_limit = np->cork.hop_limit;
1532 hdr->nexthdr = proto;
1533 hdr->saddr = fl6->saddr;
1534 hdr->daddr = *final_dst;
1536 skb->priority = sk->sk_priority;
1537 skb->mark = sk->sk_mark;
1539 skb_dst_set(skb, dst_clone(&rt->dst));
1540 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1541 if (proto == IPPROTO_ICMPV6) {
1542 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1544 ICMP6MSGOUT_INC_STATS_BH(net, idev, icmp6_hdr(skb)->icmp6_type);
1545 ICMP6_INC_STATS_BH(net, idev, ICMP6_MIB_OUTMSGS);
1548 err = ip6_local_out(skb);
1551 err = net_xmit_errno(err);
1557 ip6_cork_release(inet, np);
1560 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1563 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1565 void ip6_flush_pending_frames(struct sock *sk)
1567 struct sk_buff *skb;
1569 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1571 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1572 IPSTATS_MIB_OUTDISCARDS);
1576 ip6_cork_release(inet_sk(sk), inet6_sk(sk));
1578 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);