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_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *));
61 int __ip6_local_out(struct sk_buff *skb)
65 len = skb->len - sizeof(struct ipv6hdr);
66 if (len > IPV6_MAXPLEN)
68 ipv6_hdr(skb)->payload_len = htons(len);
70 return nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, skb, NULL,
71 skb_dst(skb)->dev, dst_output);
74 int ip6_local_out(struct sk_buff *skb)
78 err = __ip6_local_out(skb);
80 err = dst_output(skb);
84 EXPORT_SYMBOL_GPL(ip6_local_out);
86 /* dev_loopback_xmit for use with netfilter. */
87 static int ip6_dev_loopback_xmit(struct sk_buff *newskb)
89 skb_reset_mac_header(newskb);
90 __skb_pull(newskb, skb_network_offset(newskb));
91 newskb->pkt_type = PACKET_LOOPBACK;
92 newskb->ip_summed = CHECKSUM_UNNECESSARY;
93 WARN_ON(!skb_dst(newskb));
99 static int ip6_finish_output2(struct sk_buff *skb)
101 struct dst_entry *dst = skb_dst(skb);
102 struct net_device *dev = dst->dev;
103 struct neighbour *neigh;
105 skb->protocol = htons(ETH_P_IPV6);
108 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
109 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
111 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(skb->sk) &&
112 ((mroute6_socket(dev_net(dev), skb) &&
113 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
114 ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
115 &ipv6_hdr(skb)->saddr))) {
116 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
118 /* Do not check for IFF_ALLMULTI; multicast routing
119 is not supported in any case.
122 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
123 newskb, NULL, newskb->dev,
124 ip6_dev_loopback_xmit);
126 if (ipv6_hdr(skb)->hop_limit == 0) {
127 IP6_INC_STATS(dev_net(dev), idev,
128 IPSTATS_MIB_OUTDISCARDS);
134 IP6_UPD_PO_STATS(dev_net(dev), idev, IPSTATS_MIB_OUTMCAST,
139 neigh = dst_get_neighbour_noref(dst);
141 int res = neigh_output(neigh, skb);
147 IP6_INC_STATS_BH(dev_net(dst->dev),
148 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
153 static int ip6_finish_output(struct sk_buff *skb)
155 if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
156 dst_allfrag(skb_dst(skb)))
157 return ip6_fragment(skb, ip6_finish_output2);
159 return ip6_finish_output2(skb);
162 int ip6_output(struct sk_buff *skb)
164 struct net_device *dev = skb_dst(skb)->dev;
165 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
166 if (unlikely(idev->cnf.disable_ipv6)) {
167 IP6_INC_STATS(dev_net(dev), idev,
168 IPSTATS_MIB_OUTDISCARDS);
173 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, skb, NULL, dev,
175 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
179 * xmit an sk_buff (used by TCP, SCTP and DCCP)
182 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
183 struct ipv6_txoptions *opt, int tclass)
185 struct net *net = sock_net(sk);
186 struct ipv6_pinfo *np = inet6_sk(sk);
187 struct in6_addr *first_hop = &fl6->daddr;
188 struct dst_entry *dst = skb_dst(skb);
190 u8 proto = fl6->flowi6_proto;
191 int seg_len = skb->len;
196 unsigned int head_room;
198 /* First: exthdrs may take lots of space (~8K for now)
199 MAX_HEADER is not enough.
201 head_room = opt->opt_nflen + opt->opt_flen;
202 seg_len += head_room;
203 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
205 if (skb_headroom(skb) < head_room) {
206 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
208 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
209 IPSTATS_MIB_OUTDISCARDS);
215 skb_set_owner_w(skb, sk);
218 ipv6_push_frag_opts(skb, opt, &proto);
220 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
223 skb_push(skb, sizeof(struct ipv6hdr));
224 skb_reset_network_header(skb);
228 * Fill in the IPv6 header
231 hlimit = np->hop_limit;
233 hlimit = ip6_dst_hoplimit(dst);
235 *(__be32 *)hdr = htonl(0x60000000 | (tclass << 20)) | fl6->flowlabel;
237 hdr->payload_len = htons(seg_len);
238 hdr->nexthdr = proto;
239 hdr->hop_limit = hlimit;
241 hdr->saddr = fl6->saddr;
242 hdr->daddr = *first_hop;
244 skb->priority = sk->sk_priority;
245 skb->mark = sk->sk_mark;
248 if ((skb->len <= mtu) || skb->local_df || skb_is_gso(skb)) {
249 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
250 IPSTATS_MIB_OUT, skb->len);
251 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, skb, NULL,
252 dst->dev, dst_output);
255 net_dbg_ratelimited("IPv6: sending pkt_too_big to self\n");
257 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
258 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
263 EXPORT_SYMBOL(ip6_xmit);
266 * To avoid extra problems ND packets are send through this
267 * routine. It's code duplication but I really want to avoid
268 * extra checks since ipv6_build_header is used by TCP (which
269 * is for us performance critical)
272 int ip6_nd_hdr(struct sock *sk, struct sk_buff *skb, struct net_device *dev,
273 const struct in6_addr *saddr, const struct in6_addr *daddr,
276 struct ipv6_pinfo *np = inet6_sk(sk);
279 skb->protocol = htons(ETH_P_IPV6);
282 skb_reset_network_header(skb);
283 skb_put(skb, sizeof(struct ipv6hdr));
286 *(__be32*)hdr = htonl(0x60000000);
288 hdr->payload_len = htons(len);
289 hdr->nexthdr = proto;
290 hdr->hop_limit = np->hop_limit;
298 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
300 struct ip6_ra_chain *ra;
301 struct sock *last = NULL;
303 read_lock(&ip6_ra_lock);
304 for (ra = ip6_ra_chain; ra; ra = ra->next) {
305 struct sock *sk = ra->sk;
306 if (sk && ra->sel == sel &&
307 (!sk->sk_bound_dev_if ||
308 sk->sk_bound_dev_if == skb->dev->ifindex)) {
310 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
312 rawv6_rcv(last, skb2);
319 rawv6_rcv(last, skb);
320 read_unlock(&ip6_ra_lock);
323 read_unlock(&ip6_ra_lock);
327 static int ip6_forward_proxy_check(struct sk_buff *skb)
329 struct ipv6hdr *hdr = ipv6_hdr(skb);
330 u8 nexthdr = hdr->nexthdr;
334 if (ipv6_ext_hdr(nexthdr)) {
335 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
339 offset = sizeof(struct ipv6hdr);
341 if (nexthdr == IPPROTO_ICMPV6) {
342 struct icmp6hdr *icmp6;
344 if (!pskb_may_pull(skb, (skb_network_header(skb) +
345 offset + 1 - skb->data)))
348 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
350 switch (icmp6->icmp6_type) {
351 case NDISC_ROUTER_SOLICITATION:
352 case NDISC_ROUTER_ADVERTISEMENT:
353 case NDISC_NEIGHBOUR_SOLICITATION:
354 case NDISC_NEIGHBOUR_ADVERTISEMENT:
356 /* For reaction involving unicast neighbor discovery
357 * message destined to the proxied address, pass it to
367 * The proxying router can't forward traffic sent to a link-local
368 * address, so signal the sender and discard the packet. This
369 * behavior is clarified by the MIPv6 specification.
371 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
372 dst_link_failure(skb);
379 static inline int ip6_forward_finish(struct sk_buff *skb)
381 return dst_output(skb);
384 int ip6_forward(struct sk_buff *skb)
386 struct dst_entry *dst = skb_dst(skb);
387 struct ipv6hdr *hdr = ipv6_hdr(skb);
388 struct inet6_skb_parm *opt = IP6CB(skb);
389 struct net *net = dev_net(dst->dev);
392 if (net->ipv6.devconf_all->forwarding == 0)
395 if (skb_warn_if_lro(skb))
398 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
399 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
403 if (skb->pkt_type != PACKET_HOST)
406 skb_forward_csum(skb);
409 * We DO NOT make any processing on
410 * RA packets, pushing them to user level AS IS
411 * without ane WARRANTY that application will be able
412 * to interpret them. The reason is that we
413 * cannot make anything clever here.
415 * We are not end-node, so that if packet contains
416 * AH/ESP, we cannot make anything.
417 * Defragmentation also would be mistake, RA packets
418 * cannot be fragmented, because there is no warranty
419 * that different fragments will go along one path. --ANK
422 u8 *ptr = skb_network_header(skb) + opt->ra;
423 if (ip6_call_ra_chain(skb, (ptr[2]<<8) + ptr[3]))
428 * check and decrement ttl
430 if (hdr->hop_limit <= 1) {
431 /* Force OUTPUT device used as source address */
433 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
434 IP6_INC_STATS_BH(net,
435 ip6_dst_idev(dst), IPSTATS_MIB_INHDRERRORS);
441 /* XXX: idev->cnf.proxy_ndp? */
442 if (net->ipv6.devconf_all->proxy_ndp &&
443 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
444 int proxied = ip6_forward_proxy_check(skb);
446 return ip6_input(skb);
447 else if (proxied < 0) {
448 IP6_INC_STATS(net, ip6_dst_idev(dst),
449 IPSTATS_MIB_INDISCARDS);
454 if (!xfrm6_route_forward(skb)) {
455 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
460 /* IPv6 specs say nothing about it, but it is clear that we cannot
461 send redirects to source routed frames.
462 We don't send redirects to frames decapsulated from IPsec.
464 if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
465 struct in6_addr *target = NULL;
469 * incoming and outgoing devices are the same
473 rt = (struct rt6_info *) dst;
474 if (rt->rt6i_flags & RTF_GATEWAY)
475 target = &rt->rt6i_gateway;
477 target = &hdr->daddr;
480 rt6_bind_peer(rt, 1);
482 /* Limit redirects both by destination (here)
483 and by source (inside ndisc_send_redirect)
485 if (inet_peer_xrlim_allow(rt->rt6i_peer, 1*HZ))
486 ndisc_send_redirect(skb, target);
488 int addrtype = ipv6_addr_type(&hdr->saddr);
490 /* This check is security critical. */
491 if (addrtype == IPV6_ADDR_ANY ||
492 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
494 if (addrtype & IPV6_ADDR_LINKLOCAL) {
495 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
496 ICMPV6_NOT_NEIGHBOUR, 0);
502 if (mtu < IPV6_MIN_MTU)
505 if (skb->len > mtu && !skb_is_gso(skb)) {
506 /* Again, force OUTPUT device used as source address */
508 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
509 IP6_INC_STATS_BH(net,
510 ip6_dst_idev(dst), IPSTATS_MIB_INTOOBIGERRORS);
511 IP6_INC_STATS_BH(net,
512 ip6_dst_idev(dst), IPSTATS_MIB_FRAGFAILS);
517 if (skb_cow(skb, dst->dev->hard_header_len)) {
518 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTDISCARDS);
524 /* Mangling hops number delayed to point after skb COW */
528 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
529 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD, skb, skb->dev, dst->dev,
533 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
539 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
541 to->pkt_type = from->pkt_type;
542 to->priority = from->priority;
543 to->protocol = from->protocol;
545 skb_dst_set(to, dst_clone(skb_dst(from)));
547 to->mark = from->mark;
549 #ifdef CONFIG_NET_SCHED
550 to->tc_index = from->tc_index;
553 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
554 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
555 to->nf_trace = from->nf_trace;
557 skb_copy_secmark(to, from);
560 int ip6_find_1stfragopt(struct sk_buff *skb, u8 **nexthdr)
562 u16 offset = sizeof(struct ipv6hdr);
563 struct ipv6_opt_hdr *exthdr =
564 (struct ipv6_opt_hdr *)(ipv6_hdr(skb) + 1);
565 unsigned int packet_len = skb->tail - skb->network_header;
567 *nexthdr = &ipv6_hdr(skb)->nexthdr;
569 while (offset + 1 <= packet_len) {
575 case NEXTHDR_ROUTING:
579 #if defined(CONFIG_IPV6_MIP6) || defined(CONFIG_IPV6_MIP6_MODULE)
580 if (ipv6_find_tlv(skb, offset, IPV6_TLV_HAO) >= 0)
590 offset += ipv6_optlen(exthdr);
591 *nexthdr = &exthdr->nexthdr;
592 exthdr = (struct ipv6_opt_hdr *)(skb_network_header(skb) +
599 void ipv6_select_ident(struct frag_hdr *fhdr, struct rt6_info *rt)
601 static atomic_t ipv6_fragmentation_id;
604 if (rt && !(rt->dst.flags & DST_NOPEER)) {
605 struct inet_peer *peer;
608 rt6_bind_peer(rt, 1);
609 peer = rt->rt6i_peer;
611 fhdr->identification = htonl(inet_getid(peer, 0));
616 old = atomic_read(&ipv6_fragmentation_id);
620 } while (atomic_cmpxchg(&ipv6_fragmentation_id, old, new) != old);
621 fhdr->identification = htonl(new);
624 int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
626 struct sk_buff *frag;
627 struct rt6_info *rt = (struct rt6_info*)skb_dst(skb);
628 struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
629 struct ipv6hdr *tmp_hdr;
631 unsigned int mtu, hlen, left, len;
634 int ptr, offset = 0, err=0;
635 u8 *prevhdr, nexthdr = 0;
636 struct net *net = dev_net(skb_dst(skb)->dev);
638 hlen = ip6_find_1stfragopt(skb, &prevhdr);
641 mtu = ip6_skb_dst_mtu(skb);
643 /* We must not fragment if the socket is set to force MTU discovery
644 * or if the skb it not generated by a local socket.
646 if (unlikely(!skb->local_df && skb->len > mtu)) {
647 if (skb->sk && dst_allfrag(skb_dst(skb)))
648 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
650 skb->dev = skb_dst(skb)->dev;
651 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
652 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
653 IPSTATS_MIB_FRAGFAILS);
658 if (np && np->frag_size < mtu) {
662 mtu -= hlen + sizeof(struct frag_hdr);
664 if (skb_has_frag_list(skb)) {
665 int first_len = skb_pagelen(skb);
666 struct sk_buff *frag2;
668 if (first_len - hlen > mtu ||
669 ((first_len - hlen) & 7) ||
673 skb_walk_frags(skb, frag) {
674 /* Correct geometry. */
675 if (frag->len > mtu ||
676 ((frag->len & 7) && frag->next) ||
677 skb_headroom(frag) < hlen)
678 goto slow_path_clean;
680 /* Partially cloned skb? */
681 if (skb_shared(frag))
682 goto slow_path_clean;
687 frag->destructor = sock_wfree;
689 skb->truesize -= frag->truesize;
694 frag = skb_shinfo(skb)->frag_list;
695 skb_frag_list_init(skb);
698 *prevhdr = NEXTHDR_FRAGMENT;
699 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
701 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
702 IPSTATS_MIB_FRAGFAILS);
706 __skb_pull(skb, hlen);
707 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
708 __skb_push(skb, hlen);
709 skb_reset_network_header(skb);
710 memcpy(skb_network_header(skb), tmp_hdr, hlen);
712 ipv6_select_ident(fh, rt);
713 fh->nexthdr = nexthdr;
715 fh->frag_off = htons(IP6_MF);
716 frag_id = fh->identification;
718 first_len = skb_pagelen(skb);
719 skb->data_len = first_len - skb_headlen(skb);
720 skb->len = first_len;
721 ipv6_hdr(skb)->payload_len = htons(first_len -
722 sizeof(struct ipv6hdr));
727 /* Prepare header of the next frame,
728 * before previous one went down. */
730 frag->ip_summed = CHECKSUM_NONE;
731 skb_reset_transport_header(frag);
732 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
733 __skb_push(frag, hlen);
734 skb_reset_network_header(frag);
735 memcpy(skb_network_header(frag), tmp_hdr,
737 offset += skb->len - hlen - sizeof(struct frag_hdr);
738 fh->nexthdr = nexthdr;
740 fh->frag_off = htons(offset);
741 if (frag->next != NULL)
742 fh->frag_off |= htons(IP6_MF);
743 fh->identification = frag_id;
744 ipv6_hdr(frag)->payload_len =
746 sizeof(struct ipv6hdr));
747 ip6_copy_metadata(frag, skb);
752 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
753 IPSTATS_MIB_FRAGCREATES);
766 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
767 IPSTATS_MIB_FRAGOKS);
768 dst_release(&rt->dst);
778 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
779 IPSTATS_MIB_FRAGFAILS);
780 dst_release(&rt->dst);
784 skb_walk_frags(skb, frag2) {
788 frag2->destructor = NULL;
789 skb->truesize += frag2->truesize;
794 if ((skb->ip_summed == CHECKSUM_PARTIAL) &&
795 skb_checksum_help(skb))
798 left = skb->len - hlen; /* Space per frame */
799 ptr = hlen; /* Where to start from */
802 * Fragment the datagram.
805 *prevhdr = NEXTHDR_FRAGMENT;
806 hroom = LL_RESERVED_SPACE(rt->dst.dev);
807 troom = rt->dst.dev->needed_tailroom;
810 * Keep copying data until we run out.
814 /* IF: it doesn't fit, use 'mtu' - the data space left */
817 /* IF: we are not sending up to and including the packet end
818 then align the next start on an eight byte boundary */
826 if ((frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
827 hroom + troom, GFP_ATOMIC)) == NULL) {
828 NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
829 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
830 IPSTATS_MIB_FRAGFAILS);
836 * Set up data on packet
839 ip6_copy_metadata(frag, skb);
840 skb_reserve(frag, hroom);
841 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
842 skb_reset_network_header(frag);
843 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
844 frag->transport_header = (frag->network_header + hlen +
845 sizeof(struct frag_hdr));
848 * Charge the memory for the fragment to any owner
852 skb_set_owner_w(frag, skb->sk);
855 * Copy the packet header into the new buffer.
857 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
860 * Build fragment header.
862 fh->nexthdr = nexthdr;
865 ipv6_select_ident(fh, rt);
866 frag_id = fh->identification;
868 fh->identification = frag_id;
871 * Copy a block of the IP datagram.
873 if (skb_copy_bits(skb, ptr, skb_transport_header(frag), len))
877 fh->frag_off = htons(offset);
879 fh->frag_off |= htons(IP6_MF);
880 ipv6_hdr(frag)->payload_len = htons(frag->len -
881 sizeof(struct ipv6hdr));
887 * Put this fragment into the sending queue.
893 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
894 IPSTATS_MIB_FRAGCREATES);
896 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
897 IPSTATS_MIB_FRAGOKS);
902 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
903 IPSTATS_MIB_FRAGFAILS);
908 static inline int ip6_rt_check(const struct rt6key *rt_key,
909 const struct in6_addr *fl_addr,
910 const struct in6_addr *addr_cache)
912 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
913 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache));
916 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
917 struct dst_entry *dst,
918 const struct flowi6 *fl6)
920 struct ipv6_pinfo *np = inet6_sk(sk);
921 struct rt6_info *rt = (struct rt6_info *)dst;
926 /* Yes, checking route validity in not connected
927 * case is not very simple. Take into account,
928 * that we do not support routing by source, TOS,
929 * and MSG_DONTROUTE --ANK (980726)
931 * 1. ip6_rt_check(): If route was host route,
932 * check that cached destination is current.
933 * If it is network route, we still may
934 * check its validity using saved pointer
935 * to the last used address: daddr_cache.
936 * We do not want to save whole address now,
937 * (because main consumer of this service
938 * is tcp, which has not this problem),
939 * so that the last trick works only on connected
941 * 2. oif also should be the same.
943 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
944 #ifdef CONFIG_IPV6_SUBTREES
945 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
947 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
956 static int ip6_dst_lookup_tail(struct sock *sk,
957 struct dst_entry **dst, struct flowi6 *fl6)
959 struct net *net = sock_net(sk);
960 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
966 *dst = ip6_route_output(net, sk, fl6);
968 if ((err = (*dst)->error))
969 goto out_err_release;
971 if (ipv6_addr_any(&fl6->saddr)) {
972 struct rt6_info *rt = (struct rt6_info *) *dst;
973 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
974 sk ? inet6_sk(sk)->srcprefs : 0,
977 goto out_err_release;
980 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
982 * Here if the dst entry we've looked up
983 * has a neighbour entry that is in the INCOMPLETE
984 * state and the src address from the flow is
985 * marked as OPTIMISTIC, we release the found
986 * dst entry and replace it instead with the
987 * dst entry of the nexthop router
990 n = dst_get_neighbour_noref(*dst);
991 if (n && !(n->nud_state & NUD_VALID)) {
992 struct inet6_ifaddr *ifp;
993 struct flowi6 fl_gw6;
997 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1000 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1006 * We need to get the dst entry for the
1007 * default router instead
1010 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1011 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1012 *dst = ip6_route_output(net, sk, &fl_gw6);
1013 if ((err = (*dst)->error))
1014 goto out_err_release;
1024 if (err == -ENETUNREACH)
1025 IP6_INC_STATS_BH(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1032 * ip6_dst_lookup - perform route lookup on flow
1033 * @sk: socket which provides route info
1034 * @dst: pointer to dst_entry * for result
1035 * @fl6: flow to lookup
1037 * This function performs a route lookup on the given flow.
1039 * It returns zero on success, or a standard errno code on error.
1041 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi6 *fl6)
1044 return ip6_dst_lookup_tail(sk, dst, fl6);
1046 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1049 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1050 * @sk: socket which provides route info
1051 * @fl6: flow to lookup
1052 * @final_dst: final destination address for ipsec lookup
1053 * @can_sleep: we are in a sleepable context
1055 * This function performs a route lookup on the given flow.
1057 * It returns a valid dst pointer on success, or a pointer encoded
1060 struct dst_entry *ip6_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1061 const struct in6_addr *final_dst,
1064 struct dst_entry *dst = NULL;
1067 err = ip6_dst_lookup_tail(sk, &dst, fl6);
1069 return ERR_PTR(err);
1071 fl6->daddr = *final_dst;
1073 fl6->flowi6_flags |= FLOWI_FLAG_CAN_SLEEP;
1075 return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1077 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1080 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1081 * @sk: socket which provides the dst cache and route info
1082 * @fl6: flow to lookup
1083 * @final_dst: final destination address for ipsec lookup
1084 * @can_sleep: we are in a sleepable context
1086 * This function performs a route lookup on the given flow with the
1087 * possibility of using the cached route in the socket if it is valid.
1088 * It will take the socket dst lock when operating on the dst cache.
1089 * As a result, this function can only be used in process context.
1091 * It returns a valid dst pointer on success, or a pointer encoded
1094 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1095 const struct in6_addr *final_dst,
1098 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1101 dst = ip6_sk_dst_check(sk, dst, fl6);
1103 err = ip6_dst_lookup_tail(sk, &dst, fl6);
1105 return ERR_PTR(err);
1107 fl6->daddr = *final_dst;
1109 fl6->flowi6_flags |= FLOWI_FLAG_CAN_SLEEP;
1111 return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1113 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1115 static inline int ip6_ufo_append_data(struct sock *sk,
1116 int getfrag(void *from, char *to, int offset, int len,
1117 int odd, struct sk_buff *skb),
1118 void *from, int length, int hh_len, int fragheaderlen,
1119 int transhdrlen, int mtu,unsigned int flags,
1120 struct rt6_info *rt)
1123 struct sk_buff *skb;
1126 /* There is support for UDP large send offload by network
1127 * device, so create one single skb packet containing complete
1130 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
1131 skb = sock_alloc_send_skb(sk,
1132 hh_len + fragheaderlen + transhdrlen + 20,
1133 (flags & MSG_DONTWAIT), &err);
1137 /* reserve space for Hardware header */
1138 skb_reserve(skb, hh_len);
1140 /* create space for UDP/IP header */
1141 skb_put(skb,fragheaderlen + transhdrlen);
1143 /* initialize network header pointer */
1144 skb_reset_network_header(skb);
1146 /* initialize protocol header pointer */
1147 skb->transport_header = skb->network_header + fragheaderlen;
1149 skb->ip_summed = CHECKSUM_PARTIAL;
1153 err = skb_append_datato_frags(sk,skb, getfrag, from,
1154 (length - transhdrlen));
1156 struct frag_hdr fhdr;
1158 /* Specify the length of each IPv6 datagram fragment.
1159 * It has to be a multiple of 8.
1161 skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1162 sizeof(struct frag_hdr)) & ~7;
1163 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1164 ipv6_select_ident(&fhdr, rt);
1165 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1166 __skb_queue_tail(&sk->sk_write_queue, skb);
1170 /* There is not enough support do UPD LSO,
1171 * so follow normal path
1178 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1181 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1184 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1187 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1190 static void ip6_append_data_mtu(int *mtu,
1192 unsigned int fragheaderlen,
1193 struct sk_buff *skb,
1194 struct rt6_info *rt)
1196 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1198 /* first fragment, reserve header_len */
1199 *mtu = *mtu - rt->dst.header_len;
1203 * this fragment is not first, the headers
1204 * space is regarded as data space.
1206 *mtu = dst_mtu(rt->dst.path);
1208 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1209 + fragheaderlen - sizeof(struct frag_hdr);
1213 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1214 int offset, int len, int odd, struct sk_buff *skb),
1215 void *from, int length, int transhdrlen,
1216 int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1217 struct rt6_info *rt, unsigned int flags, int dontfrag)
1219 struct inet_sock *inet = inet_sk(sk);
1220 struct ipv6_pinfo *np = inet6_sk(sk);
1221 struct inet_cork *cork;
1222 struct sk_buff *skb, *skb_prev = NULL;
1223 unsigned int maxfraglen, fragheaderlen;
1233 if (flags&MSG_PROBE)
1235 cork = &inet->cork.base;
1236 if (skb_queue_empty(&sk->sk_write_queue)) {
1241 if (WARN_ON(np->cork.opt))
1244 np->cork.opt = kmalloc(opt->tot_len, sk->sk_allocation);
1245 if (unlikely(np->cork.opt == NULL))
1248 np->cork.opt->tot_len = opt->tot_len;
1249 np->cork.opt->opt_flen = opt->opt_flen;
1250 np->cork.opt->opt_nflen = opt->opt_nflen;
1252 np->cork.opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1254 if (opt->dst0opt && !np->cork.opt->dst0opt)
1257 np->cork.opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1259 if (opt->dst1opt && !np->cork.opt->dst1opt)
1262 np->cork.opt->hopopt = ip6_opt_dup(opt->hopopt,
1264 if (opt->hopopt && !np->cork.opt->hopopt)
1267 np->cork.opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1269 if (opt->srcrt && !np->cork.opt->srcrt)
1272 /* need source address above miyazawa*/
1275 cork->dst = &rt->dst;
1276 inet->cork.fl.u.ip6 = *fl6;
1277 np->cork.hop_limit = hlimit;
1278 np->cork.tclass = tclass;
1279 if (rt->dst.flags & DST_XFRM_TUNNEL)
1280 mtu = np->pmtudisc == IPV6_PMTUDISC_PROBE ?
1281 rt->dst.dev->mtu : dst_mtu(&rt->dst);
1283 mtu = np->pmtudisc == IPV6_PMTUDISC_PROBE ?
1284 rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1285 if (np->frag_size < mtu) {
1287 mtu = np->frag_size;
1289 cork->fragsize = mtu;
1290 if (dst_allfrag(rt->dst.path))
1291 cork->flags |= IPCORK_ALLFRAG;
1293 sk->sk_sndmsg_page = NULL;
1294 sk->sk_sndmsg_off = 0;
1295 exthdrlen = (opt ? opt->opt_flen : 0) - rt->rt6i_nfheader_len;
1296 length += exthdrlen;
1297 transhdrlen += exthdrlen;
1298 dst_exthdrlen = rt->dst.header_len;
1300 rt = (struct rt6_info *)cork->dst;
1301 fl6 = &inet->cork.fl.u.ip6;
1306 mtu = cork->fragsize;
1309 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1311 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1312 (opt ? opt->opt_nflen : 0);
1313 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
1315 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1316 if (cork->length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
1317 ipv6_local_error(sk, EMSGSIZE, fl6, mtu-exthdrlen);
1322 /* For UDP, check if TX timestamp is enabled */
1323 if (sk->sk_type == SOCK_DGRAM) {
1324 err = sock_tx_timestamp(sk, &tx_flags);
1330 * Let's try using as much space as possible.
1331 * Use MTU if total length of the message fits into the MTU.
1332 * Otherwise, we need to reserve fragment header and
1333 * fragment alignment (= 8-15 octects, in total).
1335 * Note that we may need to "move" the data from the tail of
1336 * of the buffer to the new fragment when we split
1339 * FIXME: It may be fragmented into multiple chunks
1340 * at once if non-fragmentable extension headers
1345 cork->length += length;
1347 int proto = sk->sk_protocol;
1348 if (dontfrag && (proto == IPPROTO_UDP || proto == IPPROTO_RAW)){
1349 ipv6_local_rxpmtu(sk, fl6, mtu-exthdrlen);
1353 if (proto == IPPROTO_UDP &&
1354 (rt->dst.dev->features & NETIF_F_UFO)) {
1356 err = ip6_ufo_append_data(sk, getfrag, from, length,
1357 hh_len, fragheaderlen,
1358 transhdrlen, mtu, flags, rt);
1365 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1368 while (length > 0) {
1369 /* Check if the remaining data fits into current packet. */
1370 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1372 copy = maxfraglen - skb->len;
1376 unsigned int datalen;
1377 unsigned int fraglen;
1378 unsigned int fraggap;
1379 unsigned int alloclen;
1381 /* There's no room in the current skb */
1383 fraggap = skb->len - maxfraglen;
1386 /* update mtu and maxfraglen if necessary */
1387 if (skb == NULL || skb_prev == NULL)
1388 ip6_append_data_mtu(&mtu, &maxfraglen,
1389 fragheaderlen, skb, rt);
1394 * If remaining data exceeds the mtu,
1395 * we know we need more fragment(s).
1397 datalen = length + fraggap;
1399 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1400 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1401 if ((flags & MSG_MORE) &&
1402 !(rt->dst.dev->features&NETIF_F_SG))
1405 alloclen = datalen + fragheaderlen;
1407 alloclen += dst_exthdrlen;
1409 if (datalen != length + fraggap) {
1411 * this is not the last fragment, the trailer
1412 * space is regarded as data space.
1414 datalen += rt->dst.trailer_len;
1417 alloclen += rt->dst.trailer_len;
1418 fraglen = datalen + fragheaderlen;
1421 * We just reserve space for fragment header.
1422 * Note: this may be overallocation if the message
1423 * (without MSG_MORE) fits into the MTU.
1425 alloclen += sizeof(struct frag_hdr);
1428 skb = sock_alloc_send_skb(sk,
1430 (flags & MSG_DONTWAIT), &err);
1433 if (atomic_read(&sk->sk_wmem_alloc) <=
1435 skb = sock_wmalloc(sk,
1436 alloclen + hh_len, 1,
1438 if (unlikely(skb == NULL))
1441 /* Only the initial fragment
1450 * Fill in the control structures
1452 skb->ip_summed = CHECKSUM_NONE;
1454 /* reserve for fragmentation and ipsec header */
1455 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1458 if (sk->sk_type == SOCK_DGRAM)
1459 skb_shinfo(skb)->tx_flags = tx_flags;
1462 * Find where to start putting bytes
1464 data = skb_put(skb, fraglen);
1465 skb_set_network_header(skb, exthdrlen);
1466 data += fragheaderlen;
1467 skb->transport_header = (skb->network_header +
1470 skb->csum = skb_copy_and_csum_bits(
1471 skb_prev, maxfraglen,
1472 data + transhdrlen, fraggap, 0);
1473 skb_prev->csum = csum_sub(skb_prev->csum,
1476 pskb_trim_unique(skb_prev, maxfraglen);
1478 copy = datalen - transhdrlen - fraggap;
1484 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1491 length -= datalen - fraggap;
1497 * Put the packet on the pending queue
1499 __skb_queue_tail(&sk->sk_write_queue, skb);
1506 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1510 if (getfrag(from, skb_put(skb, copy),
1511 offset, copy, off, skb) < 0) {
1512 __skb_trim(skb, off);
1517 int i = skb_shinfo(skb)->nr_frags;
1518 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1519 struct page *page = sk->sk_sndmsg_page;
1520 int off = sk->sk_sndmsg_off;
1523 if (page && (left = PAGE_SIZE - off) > 0) {
1526 if (page != skb_frag_page(frag)) {
1527 if (i == MAX_SKB_FRAGS) {
1531 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1532 skb_frag_ref(skb, i);
1533 frag = &skb_shinfo(skb)->frags[i];
1535 } else if(i < MAX_SKB_FRAGS) {
1536 if (copy > PAGE_SIZE)
1538 page = alloc_pages(sk->sk_allocation, 0);
1543 sk->sk_sndmsg_page = page;
1544 sk->sk_sndmsg_off = 0;
1546 skb_fill_page_desc(skb, i, page, 0, 0);
1547 frag = &skb_shinfo(skb)->frags[i];
1553 skb_frag_address(frag) + skb_frag_size(frag),
1554 offset, copy, skb->len, skb) < 0) {
1558 sk->sk_sndmsg_off += copy;
1559 skb_frag_size_add(frag, copy);
1561 skb->data_len += copy;
1562 skb->truesize += copy;
1563 atomic_add(copy, &sk->sk_wmem_alloc);
1570 cork->length -= length;
1571 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1574 EXPORT_SYMBOL_GPL(ip6_append_data);
1576 static void ip6_cork_release(struct inet_sock *inet, struct ipv6_pinfo *np)
1579 kfree(np->cork.opt->dst0opt);
1580 kfree(np->cork.opt->dst1opt);
1581 kfree(np->cork.opt->hopopt);
1582 kfree(np->cork.opt->srcrt);
1583 kfree(np->cork.opt);
1584 np->cork.opt = NULL;
1587 if (inet->cork.base.dst) {
1588 dst_release(inet->cork.base.dst);
1589 inet->cork.base.dst = NULL;
1590 inet->cork.base.flags &= ~IPCORK_ALLFRAG;
1592 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1595 int ip6_push_pending_frames(struct sock *sk)
1597 struct sk_buff *skb, *tmp_skb;
1598 struct sk_buff **tail_skb;
1599 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1600 struct inet_sock *inet = inet_sk(sk);
1601 struct ipv6_pinfo *np = inet6_sk(sk);
1602 struct net *net = sock_net(sk);
1603 struct ipv6hdr *hdr;
1604 struct ipv6_txoptions *opt = np->cork.opt;
1605 struct rt6_info *rt = (struct rt6_info *)inet->cork.base.dst;
1606 struct flowi6 *fl6 = &inet->cork.fl.u.ip6;
1607 unsigned char proto = fl6->flowi6_proto;
1610 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1612 tail_skb = &(skb_shinfo(skb)->frag_list);
1614 /* move skb->data to ip header from ext header */
1615 if (skb->data < skb_network_header(skb))
1616 __skb_pull(skb, skb_network_offset(skb));
1617 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1618 __skb_pull(tmp_skb, skb_network_header_len(skb));
1619 *tail_skb = tmp_skb;
1620 tail_skb = &(tmp_skb->next);
1621 skb->len += tmp_skb->len;
1622 skb->data_len += tmp_skb->len;
1623 skb->truesize += tmp_skb->truesize;
1624 tmp_skb->destructor = NULL;
1628 /* Allow local fragmentation. */
1629 if (np->pmtudisc < IPV6_PMTUDISC_DO)
1632 *final_dst = fl6->daddr;
1633 __skb_pull(skb, skb_network_header_len(skb));
1634 if (opt && opt->opt_flen)
1635 ipv6_push_frag_opts(skb, opt, &proto);
1636 if (opt && opt->opt_nflen)
1637 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1639 skb_push(skb, sizeof(struct ipv6hdr));
1640 skb_reset_network_header(skb);
1641 hdr = ipv6_hdr(skb);
1643 *(__be32*)hdr = fl6->flowlabel |
1644 htonl(0x60000000 | ((int)np->cork.tclass << 20));
1646 hdr->hop_limit = np->cork.hop_limit;
1647 hdr->nexthdr = proto;
1648 hdr->saddr = fl6->saddr;
1649 hdr->daddr = *final_dst;
1651 skb->priority = sk->sk_priority;
1652 skb->mark = sk->sk_mark;
1654 skb_dst_set(skb, dst_clone(&rt->dst));
1655 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1656 if (proto == IPPROTO_ICMPV6) {
1657 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1659 ICMP6MSGOUT_INC_STATS_BH(net, idev, icmp6_hdr(skb)->icmp6_type);
1660 ICMP6_INC_STATS_BH(net, idev, ICMP6_MIB_OUTMSGS);
1663 err = ip6_local_out(skb);
1666 err = net_xmit_errno(err);
1672 ip6_cork_release(inet, np);
1675 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1678 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1680 void ip6_flush_pending_frames(struct sock *sk)
1682 struct sk_buff *skb;
1684 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1686 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1687 IPSTATS_MIB_OUTDISCARDS);
1691 ip6_cork_release(inet_sk(sk), inet6_sk(sk));
1693 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);