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(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);
256 printk(KERN_DEBUG "IPv6: sending pkt_too_big to self\n");
258 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
259 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
264 EXPORT_SYMBOL(ip6_xmit);
267 * To avoid extra problems ND packets are send through this
268 * routine. It's code duplication but I really want to avoid
269 * extra checks since ipv6_build_header is used by TCP (which
270 * is for us performance critical)
273 int ip6_nd_hdr(struct sock *sk, struct sk_buff *skb, struct net_device *dev,
274 const struct in6_addr *saddr, const struct in6_addr *daddr,
277 struct ipv6_pinfo *np = inet6_sk(sk);
280 skb->protocol = htons(ETH_P_IPV6);
283 skb_reset_network_header(skb);
284 skb_put(skb, sizeof(struct ipv6hdr));
287 *(__be32*)hdr = htonl(0x60000000);
289 hdr->payload_len = htons(len);
290 hdr->nexthdr = proto;
291 hdr->hop_limit = np->hop_limit;
299 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
301 struct ip6_ra_chain *ra;
302 struct sock *last = NULL;
304 read_lock(&ip6_ra_lock);
305 for (ra = ip6_ra_chain; ra; ra = ra->next) {
306 struct sock *sk = ra->sk;
307 if (sk && ra->sel == sel &&
308 (!sk->sk_bound_dev_if ||
309 sk->sk_bound_dev_if == skb->dev->ifindex)) {
311 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
313 rawv6_rcv(last, skb2);
320 rawv6_rcv(last, skb);
321 read_unlock(&ip6_ra_lock);
324 read_unlock(&ip6_ra_lock);
328 static int ip6_forward_proxy_check(struct sk_buff *skb)
330 struct ipv6hdr *hdr = ipv6_hdr(skb);
331 u8 nexthdr = hdr->nexthdr;
334 if (ipv6_ext_hdr(nexthdr)) {
335 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr);
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);
393 if (net->ipv6.devconf_all->forwarding == 0)
396 if (skb_warn_if_lro(skb))
399 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
400 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
404 if (skb->pkt_type != PACKET_HOST)
407 skb_forward_csum(skb);
410 * We DO NOT make any processing on
411 * RA packets, pushing them to user level AS IS
412 * without ane WARRANTY that application will be able
413 * to interpret them. The reason is that we
414 * cannot make anything clever here.
416 * We are not end-node, so that if packet contains
417 * AH/ESP, we cannot make anything.
418 * Defragmentation also would be mistake, RA packets
419 * cannot be fragmented, because there is no warranty
420 * that different fragments will go along one path. --ANK
423 u8 *ptr = skb_network_header(skb) + opt->ra;
424 if (ip6_call_ra_chain(skb, (ptr[2]<<8) + ptr[3]))
429 * check and decrement ttl
431 if (hdr->hop_limit <= 1) {
432 /* Force OUTPUT device used as source address */
434 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
435 IP6_INC_STATS_BH(net,
436 ip6_dst_idev(dst), IPSTATS_MIB_INHDRERRORS);
442 /* XXX: idev->cnf.proxy_ndp? */
443 if (net->ipv6.devconf_all->proxy_ndp &&
444 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
445 int proxied = ip6_forward_proxy_check(skb);
447 return ip6_input(skb);
448 else if (proxied < 0) {
449 IP6_INC_STATS(net, ip6_dst_idev(dst),
450 IPSTATS_MIB_INDISCARDS);
455 if (!xfrm6_route_forward(skb)) {
456 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
461 /* IPv6 specs say nothing about it, but it is clear that we cannot
462 send redirects to source routed frames.
463 We don't send redirects to frames decapsulated from IPsec.
465 n = dst_get_neighbour(dst);
466 if (skb->dev == dst->dev && n && opt->srcrt == 0 && !skb_sec_path(skb)) {
467 struct in6_addr *target = NULL;
471 * incoming and outgoing devices are the same
475 rt = (struct rt6_info *) dst;
476 if ((rt->rt6i_flags & RTF_GATEWAY))
477 target = (struct in6_addr*)&n->primary_key;
479 target = &hdr->daddr;
482 rt6_bind_peer(rt, 1);
484 /* Limit redirects both by destination (here)
485 and by source (inside ndisc_send_redirect)
487 if (inet_peer_xrlim_allow(rt->rt6i_peer, 1*HZ))
488 ndisc_send_redirect(skb, n, target);
490 int addrtype = ipv6_addr_type(&hdr->saddr);
492 /* This check is security critical. */
493 if (addrtype == IPV6_ADDR_ANY ||
494 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
496 if (addrtype & IPV6_ADDR_LINKLOCAL) {
497 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
498 ICMPV6_NOT_NEIGHBOUR, 0);
504 if (mtu < IPV6_MIN_MTU)
507 if (skb->len > mtu && !skb_is_gso(skb)) {
508 /* Again, force OUTPUT device used as source address */
510 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
511 IP6_INC_STATS_BH(net,
512 ip6_dst_idev(dst), IPSTATS_MIB_INTOOBIGERRORS);
513 IP6_INC_STATS_BH(net,
514 ip6_dst_idev(dst), IPSTATS_MIB_FRAGFAILS);
519 if (skb_cow(skb, dst->dev->hard_header_len)) {
520 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTDISCARDS);
526 /* Mangling hops number delayed to point after skb COW */
530 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
531 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD, skb, skb->dev, dst->dev,
535 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
541 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
543 to->pkt_type = from->pkt_type;
544 to->priority = from->priority;
545 to->protocol = from->protocol;
547 skb_dst_set(to, dst_clone(skb_dst(from)));
549 to->mark = from->mark;
551 #ifdef CONFIG_NET_SCHED
552 to->tc_index = from->tc_index;
555 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
556 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
557 to->nf_trace = from->nf_trace;
559 skb_copy_secmark(to, from);
562 int ip6_find_1stfragopt(struct sk_buff *skb, u8 **nexthdr)
564 u16 offset = sizeof(struct ipv6hdr);
565 struct ipv6_opt_hdr *exthdr =
566 (struct ipv6_opt_hdr *)(ipv6_hdr(skb) + 1);
567 unsigned int packet_len = skb->tail - skb->network_header;
569 *nexthdr = &ipv6_hdr(skb)->nexthdr;
571 while (offset + 1 <= packet_len) {
577 case NEXTHDR_ROUTING:
581 #if defined(CONFIG_IPV6_MIP6) || defined(CONFIG_IPV6_MIP6_MODULE)
582 if (ipv6_find_tlv(skb, offset, IPV6_TLV_HAO) >= 0)
592 offset += ipv6_optlen(exthdr);
593 *nexthdr = &exthdr->nexthdr;
594 exthdr = (struct ipv6_opt_hdr *)(skb_network_header(skb) +
601 void ipv6_select_ident(struct frag_hdr *fhdr, struct rt6_info *rt)
603 static atomic_t ipv6_fragmentation_id;
607 struct inet_peer *peer;
610 rt6_bind_peer(rt, 1);
611 peer = rt->rt6i_peer;
613 fhdr->identification = htonl(inet_getid(peer, 0));
618 old = atomic_read(&ipv6_fragmentation_id);
622 } while (atomic_cmpxchg(&ipv6_fragmentation_id, old, new) != old);
623 fhdr->identification = htonl(new);
626 int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
628 struct sk_buff *frag;
629 struct rt6_info *rt = (struct rt6_info*)skb_dst(skb);
630 struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
631 struct ipv6hdr *tmp_hdr;
633 unsigned int mtu, hlen, left, len;
636 int ptr, offset = 0, err=0;
637 u8 *prevhdr, nexthdr = 0;
638 struct net *net = dev_net(skb_dst(skb)->dev);
640 hlen = ip6_find_1stfragopt(skb, &prevhdr);
643 mtu = ip6_skb_dst_mtu(skb);
645 /* We must not fragment if the socket is set to force MTU discovery
646 * or if the skb it not generated by a local socket.
648 if (!skb->local_df && skb->len > mtu) {
649 skb->dev = skb_dst(skb)->dev;
650 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
651 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
652 IPSTATS_MIB_FRAGFAILS);
657 if (np && np->frag_size < mtu) {
661 mtu -= hlen + sizeof(struct frag_hdr);
663 if (skb_has_frag_list(skb)) {
664 int first_len = skb_pagelen(skb);
665 struct sk_buff *frag2;
667 if (first_len - hlen > mtu ||
668 ((first_len - hlen) & 7) ||
672 skb_walk_frags(skb, frag) {
673 /* Correct geometry. */
674 if (frag->len > mtu ||
675 ((frag->len & 7) && frag->next) ||
676 skb_headroom(frag) < hlen)
677 goto slow_path_clean;
679 /* Partially cloned skb? */
680 if (skb_shared(frag))
681 goto slow_path_clean;
686 frag->destructor = sock_wfree;
688 skb->truesize -= frag->truesize;
693 frag = skb_shinfo(skb)->frag_list;
694 skb_frag_list_init(skb);
697 *prevhdr = NEXTHDR_FRAGMENT;
698 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
700 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
701 IPSTATS_MIB_FRAGFAILS);
705 __skb_pull(skb, hlen);
706 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
707 __skb_push(skb, hlen);
708 skb_reset_network_header(skb);
709 memcpy(skb_network_header(skb), tmp_hdr, hlen);
711 ipv6_select_ident(fh, rt);
712 fh->nexthdr = nexthdr;
714 fh->frag_off = htons(IP6_MF);
715 frag_id = fh->identification;
717 first_len = skb_pagelen(skb);
718 skb->data_len = first_len - skb_headlen(skb);
719 skb->len = first_len;
720 ipv6_hdr(skb)->payload_len = htons(first_len -
721 sizeof(struct ipv6hdr));
726 /* Prepare header of the next frame,
727 * before previous one went down. */
729 frag->ip_summed = CHECKSUM_NONE;
730 skb_reset_transport_header(frag);
731 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
732 __skb_push(frag, hlen);
733 skb_reset_network_header(frag);
734 memcpy(skb_network_header(frag), tmp_hdr,
736 offset += skb->len - hlen - sizeof(struct frag_hdr);
737 fh->nexthdr = nexthdr;
739 fh->frag_off = htons(offset);
740 if (frag->next != NULL)
741 fh->frag_off |= htons(IP6_MF);
742 fh->identification = frag_id;
743 ipv6_hdr(frag)->payload_len =
745 sizeof(struct ipv6hdr));
746 ip6_copy_metadata(frag, skb);
751 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
752 IPSTATS_MIB_FRAGCREATES);
765 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
766 IPSTATS_MIB_FRAGOKS);
767 dst_release(&rt->dst);
777 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
778 IPSTATS_MIB_FRAGFAILS);
779 dst_release(&rt->dst);
783 skb_walk_frags(skb, frag2) {
787 frag2->destructor = NULL;
788 skb->truesize += frag2->truesize;
793 left = skb->len - hlen; /* Space per frame */
794 ptr = hlen; /* Where to start from */
797 * Fragment the datagram.
800 *prevhdr = NEXTHDR_FRAGMENT;
801 hroom = LL_RESERVED_SPACE(rt->dst.dev);
802 troom = rt->dst.dev->needed_tailroom;
805 * Keep copying data until we run out.
809 /* IF: it doesn't fit, use 'mtu' - the data space left */
812 /* IF: we are not sending up to and including the packet end
813 then align the next start on an eight byte boundary */
821 if ((frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
822 hroom + troom, GFP_ATOMIC)) == NULL) {
823 NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
824 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
825 IPSTATS_MIB_FRAGFAILS);
831 * Set up data on packet
834 ip6_copy_metadata(frag, skb);
835 skb_reserve(frag, hroom);
836 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
837 skb_reset_network_header(frag);
838 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
839 frag->transport_header = (frag->network_header + hlen +
840 sizeof(struct frag_hdr));
843 * Charge the memory for the fragment to any owner
847 skb_set_owner_w(frag, skb->sk);
850 * Copy the packet header into the new buffer.
852 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
855 * Build fragment header.
857 fh->nexthdr = nexthdr;
860 ipv6_select_ident(fh, rt);
861 frag_id = fh->identification;
863 fh->identification = frag_id;
866 * Copy a block of the IP datagram.
868 if (skb_copy_bits(skb, ptr, skb_transport_header(frag), len))
872 fh->frag_off = htons(offset);
874 fh->frag_off |= htons(IP6_MF);
875 ipv6_hdr(frag)->payload_len = htons(frag->len -
876 sizeof(struct ipv6hdr));
882 * Put this fragment into the sending queue.
888 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
889 IPSTATS_MIB_FRAGCREATES);
891 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
892 IPSTATS_MIB_FRAGOKS);
897 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
898 IPSTATS_MIB_FRAGFAILS);
903 static inline int ip6_rt_check(const struct rt6key *rt_key,
904 const struct in6_addr *fl_addr,
905 const struct in6_addr *addr_cache)
907 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
908 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache));
911 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
912 struct dst_entry *dst,
913 const struct flowi6 *fl6)
915 struct ipv6_pinfo *np = inet6_sk(sk);
916 struct rt6_info *rt = (struct rt6_info *)dst;
921 /* Yes, checking route validity in not connected
922 * case is not very simple. Take into account,
923 * that we do not support routing by source, TOS,
924 * and MSG_DONTROUTE --ANK (980726)
926 * 1. ip6_rt_check(): If route was host route,
927 * check that cached destination is current.
928 * If it is network route, we still may
929 * check its validity using saved pointer
930 * to the last used address: daddr_cache.
931 * We do not want to save whole address now,
932 * (because main consumer of this service
933 * is tcp, which has not this problem),
934 * so that the last trick works only on connected
936 * 2. oif also should be the same.
938 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
939 #ifdef CONFIG_IPV6_SUBTREES
940 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
942 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
951 static int ip6_dst_lookup_tail(struct sock *sk,
952 struct dst_entry **dst, struct flowi6 *fl6)
954 struct net *net = sock_net(sk);
955 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
961 *dst = ip6_route_output(net, sk, fl6);
963 if ((err = (*dst)->error))
964 goto out_err_release;
966 if (ipv6_addr_any(&fl6->saddr)) {
967 struct rt6_info *rt = (struct rt6_info *) *dst;
968 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
969 sk ? inet6_sk(sk)->srcprefs : 0,
972 goto out_err_release;
975 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
977 * Here if the dst entry we've looked up
978 * has a neighbour entry that is in the INCOMPLETE
979 * state and the src address from the flow is
980 * marked as OPTIMISTIC, we release the found
981 * dst entry and replace it instead with the
982 * dst entry of the nexthop router
985 n = dst_get_neighbour(*dst);
986 if (n && !(n->nud_state & NUD_VALID)) {
987 struct inet6_ifaddr *ifp;
988 struct flowi6 fl_gw6;
992 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
995 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1001 * We need to get the dst entry for the
1002 * default router instead
1005 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1006 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1007 *dst = ip6_route_output(net, sk, &fl_gw6);
1008 if ((err = (*dst)->error))
1009 goto out_err_release;
1019 if (err == -ENETUNREACH)
1020 IP6_INC_STATS_BH(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1027 * ip6_dst_lookup - perform route lookup on flow
1028 * @sk: socket which provides route info
1029 * @dst: pointer to dst_entry * for result
1030 * @fl6: flow to lookup
1032 * This function performs a route lookup on the given flow.
1034 * It returns zero on success, or a standard errno code on error.
1036 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi6 *fl6)
1039 return ip6_dst_lookup_tail(sk, dst, fl6);
1041 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1044 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1045 * @sk: socket which provides route info
1046 * @fl6: flow to lookup
1047 * @final_dst: final destination address for ipsec lookup
1048 * @can_sleep: we are in a sleepable context
1050 * This function performs a route lookup on the given flow.
1052 * It returns a valid dst pointer on success, or a pointer encoded
1055 struct dst_entry *ip6_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1056 const struct in6_addr *final_dst,
1059 struct dst_entry *dst = NULL;
1062 err = ip6_dst_lookup_tail(sk, &dst, fl6);
1064 return ERR_PTR(err);
1066 fl6->daddr = *final_dst;
1068 fl6->flowi6_flags |= FLOWI_FLAG_CAN_SLEEP;
1070 return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1072 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1075 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1076 * @sk: socket which provides the dst cache and route info
1077 * @fl6: flow to lookup
1078 * @final_dst: final destination address for ipsec lookup
1079 * @can_sleep: we are in a sleepable context
1081 * This function performs a route lookup on the given flow with the
1082 * possibility of using the cached route in the socket if it is valid.
1083 * It will take the socket dst lock when operating on the dst cache.
1084 * As a result, this function can only be used in process context.
1086 * It returns a valid dst pointer on success, or a pointer encoded
1089 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1090 const struct in6_addr *final_dst,
1093 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1096 dst = ip6_sk_dst_check(sk, dst, fl6);
1098 err = ip6_dst_lookup_tail(sk, &dst, fl6);
1100 return ERR_PTR(err);
1102 fl6->daddr = *final_dst;
1104 fl6->flowi6_flags |= FLOWI_FLAG_CAN_SLEEP;
1106 return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1108 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1110 static inline int ip6_ufo_append_data(struct sock *sk,
1111 int getfrag(void *from, char *to, int offset, int len,
1112 int odd, struct sk_buff *skb),
1113 void *from, int length, int hh_len, int fragheaderlen,
1114 int transhdrlen, int mtu,unsigned int flags,
1115 struct rt6_info *rt)
1118 struct sk_buff *skb;
1121 /* There is support for UDP large send offload by network
1122 * device, so create one single skb packet containing complete
1125 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
1126 skb = sock_alloc_send_skb(sk,
1127 hh_len + fragheaderlen + transhdrlen + 20,
1128 (flags & MSG_DONTWAIT), &err);
1132 /* reserve space for Hardware header */
1133 skb_reserve(skb, hh_len);
1135 /* create space for UDP/IP header */
1136 skb_put(skb,fragheaderlen + transhdrlen);
1138 /* initialize network header pointer */
1139 skb_reset_network_header(skb);
1141 /* initialize protocol header pointer */
1142 skb->transport_header = skb->network_header + fragheaderlen;
1144 skb->ip_summed = CHECKSUM_PARTIAL;
1148 err = skb_append_datato_frags(sk,skb, getfrag, from,
1149 (length - transhdrlen));
1151 struct frag_hdr fhdr;
1153 /* Specify the length of each IPv6 datagram fragment.
1154 * It has to be a multiple of 8.
1156 skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1157 sizeof(struct frag_hdr)) & ~7;
1158 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1159 ipv6_select_ident(&fhdr, rt);
1160 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1161 __skb_queue_tail(&sk->sk_write_queue, skb);
1165 /* There is not enough support do UPD LSO,
1166 * so follow normal path
1173 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1176 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1179 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1182 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1185 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1186 int offset, int len, int odd, struct sk_buff *skb),
1187 void *from, int length, int transhdrlen,
1188 int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1189 struct rt6_info *rt, unsigned int flags, int dontfrag)
1191 struct inet_sock *inet = inet_sk(sk);
1192 struct ipv6_pinfo *np = inet6_sk(sk);
1193 struct inet_cork *cork;
1194 struct sk_buff *skb;
1195 unsigned int maxfraglen, fragheaderlen;
1203 int csummode = CHECKSUM_NONE;
1206 if (flags&MSG_PROBE)
1208 cork = &inet->cork.base;
1209 if (skb_queue_empty(&sk->sk_write_queue)) {
1214 if (WARN_ON(np->cork.opt))
1217 np->cork.opt = kmalloc(opt->tot_len, sk->sk_allocation);
1218 if (unlikely(np->cork.opt == NULL))
1221 np->cork.opt->tot_len = opt->tot_len;
1222 np->cork.opt->opt_flen = opt->opt_flen;
1223 np->cork.opt->opt_nflen = opt->opt_nflen;
1225 np->cork.opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1227 if (opt->dst0opt && !np->cork.opt->dst0opt)
1230 np->cork.opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1232 if (opt->dst1opt && !np->cork.opt->dst1opt)
1235 np->cork.opt->hopopt = ip6_opt_dup(opt->hopopt,
1237 if (opt->hopopt && !np->cork.opt->hopopt)
1240 np->cork.opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1242 if (opt->srcrt && !np->cork.opt->srcrt)
1245 /* need source address above miyazawa*/
1248 cork->dst = &rt->dst;
1249 inet->cork.fl.u.ip6 = *fl6;
1250 np->cork.hop_limit = hlimit;
1251 np->cork.tclass = tclass;
1252 mtu = np->pmtudisc == IPV6_PMTUDISC_PROBE ?
1253 rt->dst.dev->mtu : dst_mtu(&rt->dst);
1254 if (np->frag_size < mtu) {
1256 mtu = np->frag_size;
1258 cork->fragsize = mtu;
1259 if (dst_allfrag(rt->dst.path))
1260 cork->flags |= IPCORK_ALLFRAG;
1262 sk->sk_sndmsg_page = NULL;
1263 sk->sk_sndmsg_off = 0;
1264 exthdrlen = (opt ? opt->opt_flen : 0) - rt->rt6i_nfheader_len;
1265 length += exthdrlen;
1266 transhdrlen += exthdrlen;
1267 dst_exthdrlen = rt->dst.header_len;
1269 rt = (struct rt6_info *)cork->dst;
1270 fl6 = &inet->cork.fl.u.ip6;
1275 mtu = cork->fragsize;
1278 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1280 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1281 (opt ? opt->opt_nflen : 0);
1282 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
1284 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1285 if (cork->length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
1286 ipv6_local_error(sk, EMSGSIZE, fl6, mtu-exthdrlen);
1291 /* For UDP, check if TX timestamp is enabled */
1292 if (sk->sk_type == SOCK_DGRAM) {
1293 err = sock_tx_timestamp(sk, &tx_flags);
1299 * Let's try using as much space as possible.
1300 * Use MTU if total length of the message fits into the MTU.
1301 * Otherwise, we need to reserve fragment header and
1302 * fragment alignment (= 8-15 octects, in total).
1304 * Note that we may need to "move" the data from the tail of
1305 * of the buffer to the new fragment when we split
1308 * FIXME: It may be fragmented into multiple chunks
1309 * at once if non-fragmentable extension headers
1314 cork->length += length;
1316 int proto = sk->sk_protocol;
1317 if (dontfrag && (proto == IPPROTO_UDP || proto == IPPROTO_RAW)){
1318 ipv6_local_rxpmtu(sk, fl6, mtu-exthdrlen);
1322 if (proto == IPPROTO_UDP &&
1323 (rt->dst.dev->features & NETIF_F_UFO)) {
1325 err = ip6_ufo_append_data(sk, getfrag, from, length,
1326 hh_len, fragheaderlen,
1327 transhdrlen, mtu, flags, rt);
1334 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1337 while (length > 0) {
1338 /* Check if the remaining data fits into current packet. */
1339 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1341 copy = maxfraglen - skb->len;
1345 unsigned int datalen;
1346 unsigned int fraglen;
1347 unsigned int fraggap;
1348 unsigned int alloclen;
1349 struct sk_buff *skb_prev;
1353 /* There's no room in the current skb */
1355 fraggap = skb_prev->len - maxfraglen;
1360 * If remaining data exceeds the mtu,
1361 * we know we need more fragment(s).
1363 datalen = length + fraggap;
1364 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1365 datalen = maxfraglen - fragheaderlen;
1367 fraglen = datalen + fragheaderlen;
1368 if ((flags & MSG_MORE) &&
1369 !(rt->dst.dev->features&NETIF_F_SG))
1372 alloclen = datalen + fragheaderlen;
1374 alloclen += dst_exthdrlen;
1377 * The last fragment gets additional space at tail.
1378 * Note: we overallocate on fragments with MSG_MODE
1379 * because we have no idea if we're the last one.
1381 if (datalen == length + fraggap)
1382 alloclen += rt->dst.trailer_len;
1385 * We just reserve space for fragment header.
1386 * Note: this may be overallocation if the message
1387 * (without MSG_MORE) fits into the MTU.
1389 alloclen += sizeof(struct frag_hdr);
1392 skb = sock_alloc_send_skb(sk,
1394 (flags & MSG_DONTWAIT), &err);
1397 if (atomic_read(&sk->sk_wmem_alloc) <=
1399 skb = sock_wmalloc(sk,
1400 alloclen + hh_len, 1,
1402 if (unlikely(skb == NULL))
1405 /* Only the initial fragment
1414 * Fill in the control structures
1416 skb->ip_summed = csummode;
1418 /* reserve for fragmentation */
1419 skb_reserve(skb, hh_len+sizeof(struct frag_hdr));
1421 if (sk->sk_type == SOCK_DGRAM)
1422 skb_shinfo(skb)->tx_flags = tx_flags;
1425 * Find where to start putting bytes
1427 data = skb_put(skb, fraglen + dst_exthdrlen);
1428 skb_set_network_header(skb, exthdrlen + dst_exthdrlen);
1429 data += fragheaderlen + dst_exthdrlen;
1430 skb->transport_header = (skb->network_header +
1433 skb->csum = skb_copy_and_csum_bits(
1434 skb_prev, maxfraglen,
1435 data + transhdrlen, fraggap, 0);
1436 skb_prev->csum = csum_sub(skb_prev->csum,
1439 pskb_trim_unique(skb_prev, maxfraglen);
1441 copy = datalen - transhdrlen - fraggap;
1447 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1454 length -= datalen - fraggap;
1458 csummode = CHECKSUM_NONE;
1461 * Put the packet on the pending queue
1463 __skb_queue_tail(&sk->sk_write_queue, skb);
1470 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1474 if (getfrag(from, skb_put(skb, copy),
1475 offset, copy, off, skb) < 0) {
1476 __skb_trim(skb, off);
1481 int i = skb_shinfo(skb)->nr_frags;
1482 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1483 struct page *page = sk->sk_sndmsg_page;
1484 int off = sk->sk_sndmsg_off;
1487 if (page && (left = PAGE_SIZE - off) > 0) {
1490 if (page != skb_frag_page(frag)) {
1491 if (i == MAX_SKB_FRAGS) {
1495 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1496 skb_frag_ref(skb, i);
1497 frag = &skb_shinfo(skb)->frags[i];
1499 } else if(i < MAX_SKB_FRAGS) {
1500 if (copy > PAGE_SIZE)
1502 page = alloc_pages(sk->sk_allocation, 0);
1507 sk->sk_sndmsg_page = page;
1508 sk->sk_sndmsg_off = 0;
1510 skb_fill_page_desc(skb, i, page, 0, 0);
1511 frag = &skb_shinfo(skb)->frags[i];
1517 skb_frag_address(frag) + skb_frag_size(frag),
1518 offset, copy, skb->len, skb) < 0) {
1522 sk->sk_sndmsg_off += copy;
1523 skb_frag_size_add(frag, copy);
1525 skb->data_len += copy;
1526 skb->truesize += copy;
1527 atomic_add(copy, &sk->sk_wmem_alloc);
1534 cork->length -= length;
1535 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1539 static void ip6_cork_release(struct inet_sock *inet, struct ipv6_pinfo *np)
1542 kfree(np->cork.opt->dst0opt);
1543 kfree(np->cork.opt->dst1opt);
1544 kfree(np->cork.opt->hopopt);
1545 kfree(np->cork.opt->srcrt);
1546 kfree(np->cork.opt);
1547 np->cork.opt = NULL;
1550 if (inet->cork.base.dst) {
1551 dst_release(inet->cork.base.dst);
1552 inet->cork.base.dst = NULL;
1553 inet->cork.base.flags &= ~IPCORK_ALLFRAG;
1555 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1558 int ip6_push_pending_frames(struct sock *sk)
1560 struct sk_buff *skb, *tmp_skb;
1561 struct sk_buff **tail_skb;
1562 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1563 struct inet_sock *inet = inet_sk(sk);
1564 struct ipv6_pinfo *np = inet6_sk(sk);
1565 struct net *net = sock_net(sk);
1566 struct ipv6hdr *hdr;
1567 struct ipv6_txoptions *opt = np->cork.opt;
1568 struct rt6_info *rt = (struct rt6_info *)inet->cork.base.dst;
1569 struct flowi6 *fl6 = &inet->cork.fl.u.ip6;
1570 unsigned char proto = fl6->flowi6_proto;
1573 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1575 tail_skb = &(skb_shinfo(skb)->frag_list);
1577 /* move skb->data to ip header from ext header */
1578 if (skb->data < skb_network_header(skb))
1579 __skb_pull(skb, skb_network_offset(skb));
1580 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1581 __skb_pull(tmp_skb, skb_network_header_len(skb));
1582 *tail_skb = tmp_skb;
1583 tail_skb = &(tmp_skb->next);
1584 skb->len += tmp_skb->len;
1585 skb->data_len += tmp_skb->len;
1586 skb->truesize += tmp_skb->truesize;
1587 tmp_skb->destructor = NULL;
1591 /* Allow local fragmentation. */
1592 if (np->pmtudisc < IPV6_PMTUDISC_DO)
1595 *final_dst = fl6->daddr;
1596 __skb_pull(skb, skb_network_header_len(skb));
1597 if (opt && opt->opt_flen)
1598 ipv6_push_frag_opts(skb, opt, &proto);
1599 if (opt && opt->opt_nflen)
1600 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1602 skb_push(skb, sizeof(struct ipv6hdr));
1603 skb_reset_network_header(skb);
1604 hdr = ipv6_hdr(skb);
1606 *(__be32*)hdr = fl6->flowlabel |
1607 htonl(0x60000000 | ((int)np->cork.tclass << 20));
1609 hdr->hop_limit = np->cork.hop_limit;
1610 hdr->nexthdr = proto;
1611 hdr->saddr = fl6->saddr;
1612 hdr->daddr = *final_dst;
1614 skb->priority = sk->sk_priority;
1615 skb->mark = sk->sk_mark;
1617 skb_dst_set(skb, dst_clone(&rt->dst));
1618 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1619 if (proto == IPPROTO_ICMPV6) {
1620 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1622 ICMP6MSGOUT_INC_STATS_BH(net, idev, icmp6_hdr(skb)->icmp6_type);
1623 ICMP6_INC_STATS_BH(net, idev, ICMP6_MIB_OUTMSGS);
1626 err = ip6_local_out(skb);
1629 err = net_xmit_errno(err);
1635 ip6_cork_release(inet, np);
1638 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1642 void ip6_flush_pending_frames(struct sock *sk)
1644 struct sk_buff *skb;
1646 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1648 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1649 IPSTATS_MIB_OUTDISCARDS);
1653 ip6_cork_release(inet_sk(sk), inet6_sk(sk));