2 * IPv6 output functions
3 * Linux INET6 implementation
6 * Pedro Roque <roque@di.fc.ul.pt>
8 * Based on linux/net/ipv4/ip_output.c
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
16 * A.N.Kuznetsov : airthmetics in fragmentation.
17 * extension headers are implemented.
18 * route changes now work.
19 * ip6_forward does not confuse sniffers.
22 * H. von Brand : Added missing #include <linux/string.h>
23 * Imran Patel : frag id should be in NBO
24 * Kazunori MIYAZAWA @USAGI
25 * : add ip6_append_data and related functions
29 #include <linux/errno.h>
30 #include <linux/kernel.h>
31 #include <linux/string.h>
32 #include <linux/socket.h>
33 #include <linux/net.h>
34 #include <linux/netdevice.h>
35 #include <linux/if_arp.h>
36 #include <linux/in6.h>
37 #include <linux/tcp.h>
38 #include <linux/route.h>
39 #include <linux/module.h>
40 #include <linux/slab.h>
42 #include <linux/netfilter.h>
43 #include <linux/netfilter_ipv6.h>
49 #include <net/ndisc.h>
50 #include <net/protocol.h>
51 #include <net/ip6_route.h>
52 #include <net/addrconf.h>
53 #include <net/rawv6.h>
56 #include <net/checksum.h>
57 #include <linux/mroute6.h>
59 static int ip6_finish_output2(struct sk_buff *skb)
61 struct dst_entry *dst = skb_dst(skb);
62 struct net_device *dev = dst->dev;
63 struct neighbour *neigh;
64 struct in6_addr *nexthop;
67 skb->protocol = htons(ETH_P_IPV6);
70 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
71 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
73 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(skb->sk) &&
74 ((mroute6_socket(dev_net(dev), skb) &&
75 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
76 ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
77 &ipv6_hdr(skb)->saddr))) {
78 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
80 /* Do not check for IFF_ALLMULTI; multicast routing
81 is not supported in any case.
84 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
85 newskb, NULL, newskb->dev,
88 if (ipv6_hdr(skb)->hop_limit == 0) {
89 IP6_INC_STATS(dev_net(dev), idev,
90 IPSTATS_MIB_OUTDISCARDS);
96 IP6_UPD_PO_STATS(dev_net(dev), idev, IPSTATS_MIB_OUTMCAST,
99 if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb)->daddr) <=
100 IPV6_ADDR_SCOPE_NODELOCAL &&
101 !(dev->flags & IFF_LOOPBACK)) {
108 nexthop = rt6_nexthop((struct rt6_info *)dst);
109 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
110 if (unlikely(!neigh))
111 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
112 if (!IS_ERR(neigh)) {
113 ret = dst_neigh_output(dst, neigh, skb);
114 rcu_read_unlock_bh();
117 rcu_read_unlock_bh();
119 IP6_INC_STATS(dev_net(dst->dev),
120 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
125 static int ip6_finish_output(struct sk_buff *skb)
127 if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
128 dst_allfrag(skb_dst(skb)) ||
129 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
130 return ip6_fragment(skb, ip6_finish_output2);
132 return ip6_finish_output2(skb);
135 int ip6_output(struct sock *sk, struct sk_buff *skb)
137 struct net_device *dev = skb_dst(skb)->dev;
138 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
139 if (unlikely(idev->cnf.disable_ipv6)) {
140 IP6_INC_STATS(dev_net(dev), idev,
141 IPSTATS_MIB_OUTDISCARDS);
146 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, skb, NULL, dev,
148 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
152 * xmit an sk_buff (used by TCP, SCTP and DCCP)
155 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
156 struct ipv6_txoptions *opt, int tclass)
158 struct net *net = sock_net(sk);
159 struct ipv6_pinfo *np = inet6_sk(sk);
160 struct in6_addr *first_hop = &fl6->daddr;
161 struct dst_entry *dst = skb_dst(skb);
163 u8 proto = fl6->flowi6_proto;
164 int seg_len = skb->len;
169 unsigned int head_room;
171 /* First: exthdrs may take lots of space (~8K for now)
172 MAX_HEADER is not enough.
174 head_room = opt->opt_nflen + opt->opt_flen;
175 seg_len += head_room;
176 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
178 if (skb_headroom(skb) < head_room) {
179 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
181 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
182 IPSTATS_MIB_OUTDISCARDS);
188 skb_set_owner_w(skb, sk);
191 ipv6_push_frag_opts(skb, opt, &proto);
193 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
196 skb_push(skb, sizeof(struct ipv6hdr));
197 skb_reset_network_header(skb);
201 * Fill in the IPv6 header
204 hlimit = np->hop_limit;
206 hlimit = ip6_dst_hoplimit(dst);
208 ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
211 hdr->payload_len = htons(seg_len);
212 hdr->nexthdr = proto;
213 hdr->hop_limit = hlimit;
215 hdr->saddr = fl6->saddr;
216 hdr->daddr = *first_hop;
218 skb->protocol = htons(ETH_P_IPV6);
219 skb->priority = sk->sk_priority;
220 skb->mark = sk->sk_mark;
223 if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
224 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
225 IPSTATS_MIB_OUT, skb->len);
226 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, skb, NULL,
227 dst->dev, dst_output);
231 ipv6_local_error(sk, EMSGSIZE, fl6, mtu);
232 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
236 EXPORT_SYMBOL(ip6_xmit);
238 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
240 struct ip6_ra_chain *ra;
241 struct sock *last = NULL;
243 read_lock(&ip6_ra_lock);
244 for (ra = ip6_ra_chain; ra; ra = ra->next) {
245 struct sock *sk = ra->sk;
246 if (sk && ra->sel == sel &&
247 (!sk->sk_bound_dev_if ||
248 sk->sk_bound_dev_if == skb->dev->ifindex)) {
250 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
252 rawv6_rcv(last, skb2);
259 rawv6_rcv(last, skb);
260 read_unlock(&ip6_ra_lock);
263 read_unlock(&ip6_ra_lock);
267 static int ip6_forward_proxy_check(struct sk_buff *skb)
269 struct ipv6hdr *hdr = ipv6_hdr(skb);
270 u8 nexthdr = hdr->nexthdr;
274 if (ipv6_ext_hdr(nexthdr)) {
275 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
279 offset = sizeof(struct ipv6hdr);
281 if (nexthdr == IPPROTO_ICMPV6) {
282 struct icmp6hdr *icmp6;
284 if (!pskb_may_pull(skb, (skb_network_header(skb) +
285 offset + 1 - skb->data)))
288 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
290 switch (icmp6->icmp6_type) {
291 case NDISC_ROUTER_SOLICITATION:
292 case NDISC_ROUTER_ADVERTISEMENT:
293 case NDISC_NEIGHBOUR_SOLICITATION:
294 case NDISC_NEIGHBOUR_ADVERTISEMENT:
296 /* For reaction involving unicast neighbor discovery
297 * message destined to the proxied address, pass it to
307 * The proxying router can't forward traffic sent to a link-local
308 * address, so signal the sender and discard the packet. This
309 * behavior is clarified by the MIPv6 specification.
311 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
312 dst_link_failure(skb);
319 static inline int ip6_forward_finish(struct sk_buff *skb)
321 return dst_output(skb);
324 static unsigned int ip6_dst_mtu_forward(const struct dst_entry *dst)
327 struct inet6_dev *idev;
329 if (dst_metric_locked(dst, RTAX_MTU)) {
330 mtu = dst_metric_raw(dst, RTAX_MTU);
337 idev = __in6_dev_get(dst->dev);
339 mtu = idev->cnf.mtu6;
345 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
350 /* ipv6 conntrack defrag sets max_frag_size + ignore_df */
351 if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
357 if (skb_is_gso(skb) && skb_gso_network_seglen(skb) <= mtu)
363 int ip6_forward(struct sk_buff *skb)
365 struct dst_entry *dst = skb_dst(skb);
366 struct ipv6hdr *hdr = ipv6_hdr(skb);
367 struct inet6_skb_parm *opt = IP6CB(skb);
368 struct net *net = dev_net(dst->dev);
371 if (net->ipv6.devconf_all->forwarding == 0)
374 if (skb->pkt_type != PACKET_HOST)
377 if (skb_warn_if_lro(skb))
380 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
381 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
382 IPSTATS_MIB_INDISCARDS);
386 skb_forward_csum(skb);
389 * We DO NOT make any processing on
390 * RA packets, pushing them to user level AS IS
391 * without ane WARRANTY that application will be able
392 * to interpret them. The reason is that we
393 * cannot make anything clever here.
395 * We are not end-node, so that if packet contains
396 * AH/ESP, we cannot make anything.
397 * Defragmentation also would be mistake, RA packets
398 * cannot be fragmented, because there is no warranty
399 * that different fragments will go along one path. --ANK
401 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
402 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
407 * check and decrement ttl
409 if (hdr->hop_limit <= 1) {
410 /* Force OUTPUT device used as source address */
412 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
413 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
414 IPSTATS_MIB_INHDRERRORS);
420 /* XXX: idev->cnf.proxy_ndp? */
421 if (net->ipv6.devconf_all->proxy_ndp &&
422 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
423 int proxied = ip6_forward_proxy_check(skb);
425 return ip6_input(skb);
426 else if (proxied < 0) {
427 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
428 IPSTATS_MIB_INDISCARDS);
433 if (!xfrm6_route_forward(skb)) {
434 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
435 IPSTATS_MIB_INDISCARDS);
440 /* IPv6 specs say nothing about it, but it is clear that we cannot
441 send redirects to source routed frames.
442 We don't send redirects to frames decapsulated from IPsec.
444 if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
445 struct in6_addr *target = NULL;
446 struct inet_peer *peer;
450 * incoming and outgoing devices are the same
454 rt = (struct rt6_info *) dst;
455 if (rt->rt6i_flags & RTF_GATEWAY)
456 target = &rt->rt6i_gateway;
458 target = &hdr->daddr;
460 peer = inet_getpeer_v6(net->ipv6.peers, &rt->rt6i_dst.addr, 1);
462 /* Limit redirects both by destination (here)
463 and by source (inside ndisc_send_redirect)
465 if (inet_peer_xrlim_allow(peer, 1*HZ))
466 ndisc_send_redirect(skb, target);
470 int addrtype = ipv6_addr_type(&hdr->saddr);
472 /* This check is security critical. */
473 if (addrtype == IPV6_ADDR_ANY ||
474 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
476 if (addrtype & IPV6_ADDR_LINKLOCAL) {
477 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
478 ICMPV6_NOT_NEIGHBOUR, 0);
483 mtu = ip6_dst_mtu_forward(dst);
484 if (mtu < IPV6_MIN_MTU)
487 if (ip6_pkt_too_big(skb, mtu)) {
488 /* Again, force OUTPUT device used as source address */
490 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
491 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
492 IPSTATS_MIB_INTOOBIGERRORS);
493 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
494 IPSTATS_MIB_FRAGFAILS);
499 if (skb_cow(skb, dst->dev->hard_header_len)) {
500 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
501 IPSTATS_MIB_OUTDISCARDS);
507 /* Mangling hops number delayed to point after skb COW */
511 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
512 IP6_ADD_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
513 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD, skb, skb->dev, dst->dev,
517 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
523 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
525 to->pkt_type = from->pkt_type;
526 to->priority = from->priority;
527 to->protocol = from->protocol;
529 skb_dst_set(to, dst_clone(skb_dst(from)));
531 to->mark = from->mark;
533 #ifdef CONFIG_NET_SCHED
534 to->tc_index = from->tc_index;
537 skb_copy_secmark(to, from);
540 static void ipv6_select_ident(struct frag_hdr *fhdr, struct rt6_info *rt)
542 static u32 ip6_idents_hashrnd __read_mostly;
545 net_get_random_once(&ip6_idents_hashrnd, sizeof(ip6_idents_hashrnd));
547 hash = __ipv6_addr_jhash(&rt->rt6i_dst.addr, ip6_idents_hashrnd);
548 hash = __ipv6_addr_jhash(&rt->rt6i_src.addr, hash);
550 id = ip_idents_reserve(hash, 1);
551 fhdr->identification = htonl(id);
554 int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
556 struct sk_buff *frag;
557 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
558 struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
559 struct ipv6hdr *tmp_hdr;
561 unsigned int mtu, hlen, left, len;
564 int ptr, offset = 0, err = 0;
565 u8 *prevhdr, nexthdr = 0;
566 struct net *net = dev_net(skb_dst(skb)->dev);
568 hlen = ip6_find_1stfragopt(skb, &prevhdr);
571 mtu = ip6_skb_dst_mtu(skb);
573 /* We must not fragment if the socket is set to force MTU discovery
574 * or if the skb it not generated by a local socket.
576 if (unlikely(!skb->ignore_df && skb->len > mtu) ||
577 (IP6CB(skb)->frag_max_size &&
578 IP6CB(skb)->frag_max_size > mtu)) {
579 if (skb->sk && dst_allfrag(skb_dst(skb)))
580 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
582 skb->dev = skb_dst(skb)->dev;
583 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
584 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
585 IPSTATS_MIB_FRAGFAILS);
590 if (np && np->frag_size < mtu) {
594 mtu -= hlen + sizeof(struct frag_hdr);
596 if (skb_has_frag_list(skb)) {
597 int first_len = skb_pagelen(skb);
598 struct sk_buff *frag2;
600 if (first_len - hlen > mtu ||
601 ((first_len - hlen) & 7) ||
605 skb_walk_frags(skb, frag) {
606 /* Correct geometry. */
607 if (frag->len > mtu ||
608 ((frag->len & 7) && frag->next) ||
609 skb_headroom(frag) < hlen)
610 goto slow_path_clean;
612 /* Partially cloned skb? */
613 if (skb_shared(frag))
614 goto slow_path_clean;
619 frag->destructor = sock_wfree;
621 skb->truesize -= frag->truesize;
626 frag = skb_shinfo(skb)->frag_list;
627 skb_frag_list_init(skb);
630 *prevhdr = NEXTHDR_FRAGMENT;
631 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
633 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
634 IPSTATS_MIB_FRAGFAILS);
638 __skb_pull(skb, hlen);
639 fh = (struct frag_hdr *)__skb_push(skb, sizeof(struct frag_hdr));
640 __skb_push(skb, hlen);
641 skb_reset_network_header(skb);
642 memcpy(skb_network_header(skb), tmp_hdr, hlen);
644 ipv6_select_ident(fh, rt);
645 fh->nexthdr = nexthdr;
647 fh->frag_off = htons(IP6_MF);
648 frag_id = fh->identification;
650 first_len = skb_pagelen(skb);
651 skb->data_len = first_len - skb_headlen(skb);
652 skb->len = first_len;
653 ipv6_hdr(skb)->payload_len = htons(first_len -
654 sizeof(struct ipv6hdr));
659 /* Prepare header of the next frame,
660 * before previous one went down. */
662 frag->ip_summed = CHECKSUM_NONE;
663 skb_reset_transport_header(frag);
664 fh = (struct frag_hdr *)__skb_push(frag, sizeof(struct frag_hdr));
665 __skb_push(frag, hlen);
666 skb_reset_network_header(frag);
667 memcpy(skb_network_header(frag), tmp_hdr,
669 offset += skb->len - hlen - sizeof(struct frag_hdr);
670 fh->nexthdr = nexthdr;
672 fh->frag_off = htons(offset);
673 if (frag->next != NULL)
674 fh->frag_off |= htons(IP6_MF);
675 fh->identification = frag_id;
676 ipv6_hdr(frag)->payload_len =
678 sizeof(struct ipv6hdr));
679 ip6_copy_metadata(frag, skb);
684 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
685 IPSTATS_MIB_FRAGCREATES);
698 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
699 IPSTATS_MIB_FRAGOKS);
704 kfree_skb_list(frag);
706 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
707 IPSTATS_MIB_FRAGFAILS);
712 skb_walk_frags(skb, frag2) {
716 frag2->destructor = NULL;
717 skb->truesize += frag2->truesize;
722 if ((skb->ip_summed == CHECKSUM_PARTIAL) &&
723 skb_checksum_help(skb))
726 left = skb->len - hlen; /* Space per frame */
727 ptr = hlen; /* Where to start from */
730 * Fragment the datagram.
733 *prevhdr = NEXTHDR_FRAGMENT;
734 hroom = LL_RESERVED_SPACE(rt->dst.dev);
735 troom = rt->dst.dev->needed_tailroom;
738 * Keep copying data until we run out.
742 /* IF: it doesn't fit, use 'mtu' - the data space left */
745 /* IF: we are not sending up to and including the packet end
746 then align the next start on an eight byte boundary */
754 if ((frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
755 hroom + troom, GFP_ATOMIC)) == NULL) {
756 NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
757 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
758 IPSTATS_MIB_FRAGFAILS);
764 * Set up data on packet
767 ip6_copy_metadata(frag, skb);
768 skb_reserve(frag, hroom);
769 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
770 skb_reset_network_header(frag);
771 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
772 frag->transport_header = (frag->network_header + hlen +
773 sizeof(struct frag_hdr));
776 * Charge the memory for the fragment to any owner
780 skb_set_owner_w(frag, skb->sk);
783 * Copy the packet header into the new buffer.
785 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
788 * Build fragment header.
790 fh->nexthdr = nexthdr;
793 ipv6_select_ident(fh, rt);
794 frag_id = fh->identification;
796 fh->identification = frag_id;
799 * Copy a block of the IP datagram.
801 BUG_ON(skb_copy_bits(skb, ptr, skb_transport_header(frag),
805 fh->frag_off = htons(offset);
807 fh->frag_off |= htons(IP6_MF);
808 ipv6_hdr(frag)->payload_len = htons(frag->len -
809 sizeof(struct ipv6hdr));
815 * Put this fragment into the sending queue.
821 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
822 IPSTATS_MIB_FRAGCREATES);
824 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
825 IPSTATS_MIB_FRAGOKS);
830 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
831 IPSTATS_MIB_FRAGFAILS);
836 static inline int ip6_rt_check(const struct rt6key *rt_key,
837 const struct in6_addr *fl_addr,
838 const struct in6_addr *addr_cache)
840 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
841 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache));
844 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
845 struct dst_entry *dst,
846 const struct flowi6 *fl6)
848 struct ipv6_pinfo *np = inet6_sk(sk);
854 if (dst->ops->family != AF_INET6) {
859 rt = (struct rt6_info *)dst;
860 /* Yes, checking route validity in not connected
861 * case is not very simple. Take into account,
862 * that we do not support routing by source, TOS,
863 * and MSG_DONTROUTE --ANK (980726)
865 * 1. ip6_rt_check(): If route was host route,
866 * check that cached destination is current.
867 * If it is network route, we still may
868 * check its validity using saved pointer
869 * to the last used address: daddr_cache.
870 * We do not want to save whole address now,
871 * (because main consumer of this service
872 * is tcp, which has not this problem),
873 * so that the last trick works only on connected
875 * 2. oif also should be the same.
877 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
878 #ifdef CONFIG_IPV6_SUBTREES
879 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
881 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
890 static int ip6_dst_lookup_tail(struct sock *sk,
891 struct dst_entry **dst, struct flowi6 *fl6)
893 struct net *net = sock_net(sk);
894 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
901 *dst = ip6_route_output(net, sk, fl6);
903 if ((err = (*dst)->error))
904 goto out_err_release;
906 if (ipv6_addr_any(&fl6->saddr)) {
907 struct rt6_info *rt = (struct rt6_info *) *dst;
908 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
909 sk ? inet6_sk(sk)->srcprefs : 0,
912 goto out_err_release;
915 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
917 * Here if the dst entry we've looked up
918 * has a neighbour entry that is in the INCOMPLETE
919 * state and the src address from the flow is
920 * marked as OPTIMISTIC, we release the found
921 * dst entry and replace it instead with the
922 * dst entry of the nexthop router
924 rt = (struct rt6_info *) *dst;
926 n = __ipv6_neigh_lookup_noref(rt->dst.dev, rt6_nexthop(rt));
927 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
928 rcu_read_unlock_bh();
931 struct inet6_ifaddr *ifp;
932 struct flowi6 fl_gw6;
935 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
938 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
944 * We need to get the dst entry for the
945 * default router instead
948 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
949 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
950 *dst = ip6_route_output(net, sk, &fl_gw6);
951 if ((err = (*dst)->error))
952 goto out_err_release;
960 if (err == -ENETUNREACH)
961 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
968 * ip6_dst_lookup - perform route lookup on flow
969 * @sk: socket which provides route info
970 * @dst: pointer to dst_entry * for result
971 * @fl6: flow to lookup
973 * This function performs a route lookup on the given flow.
975 * It returns zero on success, or a standard errno code on error.
977 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi6 *fl6)
980 return ip6_dst_lookup_tail(sk, dst, fl6);
982 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
985 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
986 * @sk: socket which provides route info
987 * @fl6: flow to lookup
988 * @final_dst: final destination address for ipsec lookup
990 * This function performs a route lookup on the given flow.
992 * It returns a valid dst pointer on success, or a pointer encoded
995 struct dst_entry *ip6_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
996 const struct in6_addr *final_dst)
998 struct dst_entry *dst = NULL;
1001 err = ip6_dst_lookup_tail(sk, &dst, fl6);
1003 return ERR_PTR(err);
1005 fl6->daddr = *final_dst;
1007 return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1009 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1012 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1013 * @sk: socket which provides the dst cache and route info
1014 * @fl6: flow to lookup
1015 * @final_dst: final destination address for ipsec lookup
1017 * This function performs a route lookup on the given flow with the
1018 * possibility of using the cached route in the socket if it is valid.
1019 * It will take the socket dst lock when operating on the dst cache.
1020 * As a result, this function can only be used in process context.
1022 * It returns a valid dst pointer on success, or a pointer encoded
1025 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1026 const struct in6_addr *final_dst)
1028 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1031 dst = ip6_sk_dst_check(sk, dst, fl6);
1033 err = ip6_dst_lookup_tail(sk, &dst, fl6);
1035 return ERR_PTR(err);
1037 fl6->daddr = *final_dst;
1039 return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1041 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1043 static inline int ip6_ufo_append_data(struct sock *sk,
1044 int getfrag(void *from, char *to, int offset, int len,
1045 int odd, struct sk_buff *skb),
1046 void *from, int length, int hh_len, int fragheaderlen,
1047 int transhdrlen, int mtu, unsigned int flags,
1048 struct rt6_info *rt)
1051 struct sk_buff *skb;
1052 struct frag_hdr fhdr;
1055 /* There is support for UDP large send offload by network
1056 * device, so create one single skb packet containing complete
1059 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
1060 skb = sock_alloc_send_skb(sk,
1061 hh_len + fragheaderlen + transhdrlen + 20,
1062 (flags & MSG_DONTWAIT), &err);
1066 /* reserve space for Hardware header */
1067 skb_reserve(skb, hh_len);
1069 /* create space for UDP/IP header */
1070 skb_put(skb, fragheaderlen + transhdrlen);
1072 /* initialize network header pointer */
1073 skb_reset_network_header(skb);
1075 /* initialize protocol header pointer */
1076 skb->transport_header = skb->network_header + fragheaderlen;
1078 skb->protocol = htons(ETH_P_IPV6);
1081 __skb_queue_tail(&sk->sk_write_queue, skb);
1082 } else if (skb_is_gso(skb)) {
1086 skb->ip_summed = CHECKSUM_PARTIAL;
1087 /* Specify the length of each IPv6 datagram fragment.
1088 * It has to be a multiple of 8.
1090 skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1091 sizeof(struct frag_hdr)) & ~7;
1092 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1093 ipv6_select_ident(&fhdr, rt);
1094 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1097 return skb_append_datato_frags(sk, skb, getfrag, from,
1098 (length - transhdrlen));
1101 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1104 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1107 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1110 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1113 static void ip6_append_data_mtu(unsigned int *mtu,
1115 unsigned int fragheaderlen,
1116 struct sk_buff *skb,
1117 struct rt6_info *rt,
1118 unsigned int orig_mtu)
1120 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1122 /* first fragment, reserve header_len */
1123 *mtu = orig_mtu - rt->dst.header_len;
1127 * this fragment is not first, the headers
1128 * space is regarded as data space.
1132 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1133 + fragheaderlen - sizeof(struct frag_hdr);
1137 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1138 int offset, int len, int odd, struct sk_buff *skb),
1139 void *from, int length, int transhdrlen,
1140 int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1141 struct rt6_info *rt, unsigned int flags, int dontfrag)
1143 struct inet_sock *inet = inet_sk(sk);
1144 struct ipv6_pinfo *np = inet6_sk(sk);
1145 struct inet_cork *cork;
1146 struct sk_buff *skb, *skb_prev = NULL;
1147 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu;
1157 if (flags&MSG_PROBE)
1159 cork = &inet->cork.base;
1160 if (skb_queue_empty(&sk->sk_write_queue)) {
1165 if (WARN_ON(np->cork.opt))
1168 np->cork.opt = kzalloc(opt->tot_len, sk->sk_allocation);
1169 if (unlikely(np->cork.opt == NULL))
1172 np->cork.opt->tot_len = opt->tot_len;
1173 np->cork.opt->opt_flen = opt->opt_flen;
1174 np->cork.opt->opt_nflen = opt->opt_nflen;
1176 np->cork.opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1178 if (opt->dst0opt && !np->cork.opt->dst0opt)
1181 np->cork.opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1183 if (opt->dst1opt && !np->cork.opt->dst1opt)
1186 np->cork.opt->hopopt = ip6_opt_dup(opt->hopopt,
1188 if (opt->hopopt && !np->cork.opt->hopopt)
1191 np->cork.opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1193 if (opt->srcrt && !np->cork.opt->srcrt)
1196 /* need source address above miyazawa*/
1199 cork->dst = &rt->dst;
1200 inet->cork.fl.u.ip6 = *fl6;
1201 np->cork.hop_limit = hlimit;
1202 np->cork.tclass = tclass;
1203 if (rt->dst.flags & DST_XFRM_TUNNEL)
1204 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1205 rt->dst.dev->mtu : dst_mtu(&rt->dst);
1207 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1208 rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1209 if (np->frag_size < mtu) {
1211 mtu = np->frag_size;
1213 cork->fragsize = mtu;
1214 if (dst_allfrag(rt->dst.path))
1215 cork->flags |= IPCORK_ALLFRAG;
1217 exthdrlen = (opt ? opt->opt_flen : 0);
1218 length += exthdrlen;
1219 transhdrlen += exthdrlen;
1220 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1222 rt = (struct rt6_info *)cork->dst;
1223 fl6 = &inet->cork.fl.u.ip6;
1228 mtu = cork->fragsize;
1232 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1234 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1235 (opt ? opt->opt_nflen : 0);
1236 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1237 sizeof(struct frag_hdr);
1239 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1240 unsigned int maxnonfragsize, headersize;
1242 headersize = sizeof(struct ipv6hdr) +
1243 (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1244 (dst_allfrag(&rt->dst) ?
1245 sizeof(struct frag_hdr) : 0) +
1246 rt->rt6i_nfheader_len;
1248 if (ip6_sk_ignore_df(sk))
1249 maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1251 maxnonfragsize = mtu;
1253 /* dontfrag active */
1254 if ((cork->length + length > mtu - headersize) && dontfrag &&
1255 (sk->sk_protocol == IPPROTO_UDP ||
1256 sk->sk_protocol == IPPROTO_RAW)) {
1257 ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1258 sizeof(struct ipv6hdr));
1262 if (cork->length + length > maxnonfragsize - headersize) {
1264 ipv6_local_error(sk, EMSGSIZE, fl6,
1266 sizeof(struct ipv6hdr));
1271 if (sk->sk_type == SOCK_DGRAM || sk->sk_type == SOCK_RAW) {
1272 sock_tx_timestamp(sk, &tx_flags);
1273 if (tx_flags & SKBTX_ANY_SW_TSTAMP &&
1274 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1275 tskey = sk->sk_tskey++;
1279 * Let's try using as much space as possible.
1280 * Use MTU if total length of the message fits into the MTU.
1281 * Otherwise, we need to reserve fragment header and
1282 * fragment alignment (= 8-15 octects, in total).
1284 * Note that we may need to "move" the data from the tail of
1285 * of the buffer to the new fragment when we split
1288 * FIXME: It may be fragmented into multiple chunks
1289 * at once if non-fragmentable extension headers
1294 skb = skb_peek_tail(&sk->sk_write_queue);
1295 cork->length += length;
1296 if (((length > mtu) ||
1297 (skb && skb_is_gso(skb))) &&
1298 (sk->sk_protocol == IPPROTO_UDP) &&
1299 (rt->dst.dev->features & NETIF_F_UFO)) {
1300 err = ip6_ufo_append_data(sk, getfrag, from, length,
1301 hh_len, fragheaderlen,
1302 transhdrlen, mtu, flags, rt);
1311 while (length > 0) {
1312 /* Check if the remaining data fits into current packet. */
1313 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1315 copy = maxfraglen - skb->len;
1319 unsigned int datalen;
1320 unsigned int fraglen;
1321 unsigned int fraggap;
1322 unsigned int alloclen;
1324 /* There's no room in the current skb */
1326 fraggap = skb->len - maxfraglen;
1329 /* update mtu and maxfraglen if necessary */
1330 if (skb == NULL || skb_prev == NULL)
1331 ip6_append_data_mtu(&mtu, &maxfraglen,
1332 fragheaderlen, skb, rt,
1338 * If remaining data exceeds the mtu,
1339 * we know we need more fragment(s).
1341 datalen = length + fraggap;
1343 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1344 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1345 if ((flags & MSG_MORE) &&
1346 !(rt->dst.dev->features&NETIF_F_SG))
1349 alloclen = datalen + fragheaderlen;
1351 alloclen += dst_exthdrlen;
1353 if (datalen != length + fraggap) {
1355 * this is not the last fragment, the trailer
1356 * space is regarded as data space.
1358 datalen += rt->dst.trailer_len;
1361 alloclen += rt->dst.trailer_len;
1362 fraglen = datalen + fragheaderlen;
1365 * We just reserve space for fragment header.
1366 * Note: this may be overallocation if the message
1367 * (without MSG_MORE) fits into the MTU.
1369 alloclen += sizeof(struct frag_hdr);
1372 skb = sock_alloc_send_skb(sk,
1374 (flags & MSG_DONTWAIT), &err);
1377 if (atomic_read(&sk->sk_wmem_alloc) <=
1379 skb = sock_wmalloc(sk,
1380 alloclen + hh_len, 1,
1382 if (unlikely(skb == NULL))
1388 * Fill in the control structures
1390 skb->protocol = htons(ETH_P_IPV6);
1391 skb->ip_summed = CHECKSUM_NONE;
1393 /* reserve for fragmentation and ipsec header */
1394 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1397 /* Only the initial fragment is time stamped */
1398 skb_shinfo(skb)->tx_flags = tx_flags;
1400 skb_shinfo(skb)->tskey = tskey;
1404 * Find where to start putting bytes
1406 data = skb_put(skb, fraglen);
1407 skb_set_network_header(skb, exthdrlen);
1408 data += fragheaderlen;
1409 skb->transport_header = (skb->network_header +
1412 skb->csum = skb_copy_and_csum_bits(
1413 skb_prev, maxfraglen,
1414 data + transhdrlen, fraggap, 0);
1415 skb_prev->csum = csum_sub(skb_prev->csum,
1418 pskb_trim_unique(skb_prev, maxfraglen);
1420 copy = datalen - transhdrlen - fraggap;
1426 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1433 length -= datalen - fraggap;
1439 * Put the packet on the pending queue
1441 __skb_queue_tail(&sk->sk_write_queue, skb);
1448 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1452 if (getfrag(from, skb_put(skb, copy),
1453 offset, copy, off, skb) < 0) {
1454 __skb_trim(skb, off);
1459 int i = skb_shinfo(skb)->nr_frags;
1460 struct page_frag *pfrag = sk_page_frag(sk);
1463 if (!sk_page_frag_refill(sk, pfrag))
1466 if (!skb_can_coalesce(skb, i, pfrag->page,
1469 if (i == MAX_SKB_FRAGS)
1472 __skb_fill_page_desc(skb, i, pfrag->page,
1474 skb_shinfo(skb)->nr_frags = ++i;
1475 get_page(pfrag->page);
1477 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1479 page_address(pfrag->page) + pfrag->offset,
1480 offset, copy, skb->len, skb) < 0)
1483 pfrag->offset += copy;
1484 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1486 skb->data_len += copy;
1487 skb->truesize += copy;
1488 atomic_add(copy, &sk->sk_wmem_alloc);
1499 cork->length -= length;
1500 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1503 EXPORT_SYMBOL_GPL(ip6_append_data);
1505 static void ip6_cork_release(struct inet_sock *inet, struct ipv6_pinfo *np)
1508 kfree(np->cork.opt->dst0opt);
1509 kfree(np->cork.opt->dst1opt);
1510 kfree(np->cork.opt->hopopt);
1511 kfree(np->cork.opt->srcrt);
1512 kfree(np->cork.opt);
1513 np->cork.opt = NULL;
1516 if (inet->cork.base.dst) {
1517 dst_release(inet->cork.base.dst);
1518 inet->cork.base.dst = NULL;
1519 inet->cork.base.flags &= ~IPCORK_ALLFRAG;
1521 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1524 int ip6_push_pending_frames(struct sock *sk)
1526 struct sk_buff *skb, *tmp_skb;
1527 struct sk_buff **tail_skb;
1528 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1529 struct inet_sock *inet = inet_sk(sk);
1530 struct ipv6_pinfo *np = inet6_sk(sk);
1531 struct net *net = sock_net(sk);
1532 struct ipv6hdr *hdr;
1533 struct ipv6_txoptions *opt = np->cork.opt;
1534 struct rt6_info *rt = (struct rt6_info *)inet->cork.base.dst;
1535 struct flowi6 *fl6 = &inet->cork.fl.u.ip6;
1536 unsigned char proto = fl6->flowi6_proto;
1539 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1541 tail_skb = &(skb_shinfo(skb)->frag_list);
1543 /* move skb->data to ip header from ext header */
1544 if (skb->data < skb_network_header(skb))
1545 __skb_pull(skb, skb_network_offset(skb));
1546 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1547 __skb_pull(tmp_skb, skb_network_header_len(skb));
1548 *tail_skb = tmp_skb;
1549 tail_skb = &(tmp_skb->next);
1550 skb->len += tmp_skb->len;
1551 skb->data_len += tmp_skb->len;
1552 skb->truesize += tmp_skb->truesize;
1553 tmp_skb->destructor = NULL;
1557 /* Allow local fragmentation. */
1558 skb->ignore_df = ip6_sk_ignore_df(sk);
1560 *final_dst = fl6->daddr;
1561 __skb_pull(skb, skb_network_header_len(skb));
1562 if (opt && opt->opt_flen)
1563 ipv6_push_frag_opts(skb, opt, &proto);
1564 if (opt && opt->opt_nflen)
1565 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1567 skb_push(skb, sizeof(struct ipv6hdr));
1568 skb_reset_network_header(skb);
1569 hdr = ipv6_hdr(skb);
1571 ip6_flow_hdr(hdr, np->cork.tclass,
1572 ip6_make_flowlabel(net, skb, fl6->flowlabel,
1573 np->autoflowlabel));
1574 hdr->hop_limit = np->cork.hop_limit;
1575 hdr->nexthdr = proto;
1576 hdr->saddr = fl6->saddr;
1577 hdr->daddr = *final_dst;
1579 skb->priority = sk->sk_priority;
1580 skb->mark = sk->sk_mark;
1582 skb_dst_set(skb, dst_clone(&rt->dst));
1583 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1584 if (proto == IPPROTO_ICMPV6) {
1585 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1587 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1588 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1591 err = ip6_local_out(skb);
1594 err = net_xmit_errno(err);
1600 ip6_cork_release(inet, np);
1603 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1606 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1608 void ip6_flush_pending_frames(struct sock *sk)
1610 struct sk_buff *skb;
1612 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1614 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1615 IPSTATS_MIB_OUTDISCARDS);
1619 ip6_cork_release(inet_sk(sk), inet6_sk(sk));
1621 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);