2 * IPv6 output functions
3 * Linux INET6 implementation
6 * Pedro Roque <roque@di.fc.ul.pt>
8 * $Id: ip6_output.c,v 1.34 2002/02/01 22:01:04 davem Exp $
10 * Based on linux/net/ipv4/ip_output.c
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation; either version
15 * 2 of the License, or (at your option) any later version.
18 * A.N.Kuznetsov : airthmetics in fragmentation.
19 * extension headers are implemented.
20 * route changes now work.
21 * ip6_forward does not confuse sniffers.
24 * H. von Brand : Added missing #include <linux/string.h>
25 * Imran Patel : frag id should be in NBO
26 * Kazunori MIYAZAWA @USAGI
27 * : add ip6_append_data and related functions
31 #include <linux/errno.h>
32 #include <linux/types.h>
33 #include <linux/string.h>
34 #include <linux/socket.h>
35 #include <linux/net.h>
36 #include <linux/netdevice.h>
37 #include <linux/if_arp.h>
38 #include <linux/in6.h>
39 #include <linux/tcp.h>
40 #include <linux/route.h>
41 #include <linux/module.h>
43 #include <linux/netfilter.h>
44 #include <linux/netfilter_ipv6.h>
50 #include <net/ndisc.h>
51 #include <net/protocol.h>
52 #include <net/ip6_route.h>
53 #include <net/addrconf.h>
54 #include <net/rawv6.h>
57 #include <net/checksum.h>
59 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *));
61 static __inline__ void ipv6_select_ident(struct sk_buff *skb, struct frag_hdr *fhdr)
63 static u32 ipv6_fragmentation_id = 1;
64 static DEFINE_SPINLOCK(ip6_id_lock);
66 spin_lock_bh(&ip6_id_lock);
67 fhdr->identification = htonl(ipv6_fragmentation_id);
68 if (++ipv6_fragmentation_id == 0)
69 ipv6_fragmentation_id = 1;
70 spin_unlock_bh(&ip6_id_lock);
73 static inline int ip6_output_finish(struct sk_buff *skb)
75 struct dst_entry *dst = skb->dst;
78 return neigh_hh_output(dst->hh, skb);
79 else if (dst->neighbour)
80 return dst->neighbour->output(skb);
82 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
88 /* dev_loopback_xmit for use with netfilter. */
89 static int ip6_dev_loopback_xmit(struct sk_buff *newskb)
91 skb_reset_mac_header(newskb);
92 __skb_pull(newskb, skb_network_offset(newskb));
93 newskb->pkt_type = PACKET_LOOPBACK;
94 newskb->ip_summed = CHECKSUM_UNNECESSARY;
95 BUG_TRAP(newskb->dst);
102 static int ip6_output2(struct sk_buff *skb)
104 struct dst_entry *dst = skb->dst;
105 struct net_device *dev = dst->dev;
107 skb->protocol = htons(ETH_P_IPV6);
110 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
111 struct ipv6_pinfo* np = skb->sk ? inet6_sk(skb->sk) : NULL;
112 struct inet6_dev *idev = ip6_dst_idev(skb->dst);
114 if (!(dev->flags & IFF_LOOPBACK) && (!np || np->mc_loop) &&
115 ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
116 &ipv6_hdr(skb)->saddr)) {
117 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
119 /* Do not check for IFF_ALLMULTI; multicast routing
120 is not supported in any case.
123 NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, newskb, NULL,
125 ip6_dev_loopback_xmit);
127 if (ipv6_hdr(skb)->hop_limit == 0) {
128 IP6_INC_STATS(idev, IPSTATS_MIB_OUTDISCARDS);
134 IP6_INC_STATS(idev, IPSTATS_MIB_OUTMCASTPKTS);
137 return NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, skb,NULL, skb->dev,ip6_output_finish);
140 int ip6_output(struct sk_buff *skb)
142 if ((skb->len > dst_mtu(skb->dst) && !skb_is_gso(skb)) ||
143 dst_allfrag(skb->dst))
144 return ip6_fragment(skb, ip6_output2);
146 return ip6_output2(skb);
150 * xmit an sk_buff (used by TCP)
153 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
154 struct ipv6_txoptions *opt, int ipfragok)
156 struct ipv6_pinfo *np = inet6_sk(sk);
157 struct in6_addr *first_hop = &fl->fl6_dst;
158 struct dst_entry *dst = skb->dst;
160 u8 proto = fl->proto;
161 int seg_len = skb->len;
168 /* First: exthdrs may take lots of space (~8K for now)
169 MAX_HEADER is not enough.
171 head_room = opt->opt_nflen + opt->opt_flen;
172 seg_len += head_room;
173 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
175 if (skb_headroom(skb) < head_room) {
176 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
178 IP6_INC_STATS(ip6_dst_idev(skb->dst),
179 IPSTATS_MIB_OUTDISCARDS);
186 skb_set_owner_w(skb, sk);
189 ipv6_push_frag_opts(skb, opt, &proto);
191 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
194 skb_push(skb, sizeof(struct ipv6hdr));
195 skb_reset_network_header(skb);
199 * Fill in the IPv6 header
204 hlimit = np->hop_limit;
206 hlimit = dst_metric(dst, RTAX_HOPLIMIT);
208 hlimit = ipv6_get_hoplimit(dst->dev);
216 *(__be32 *)hdr = htonl(0x60000000 | (tclass << 20)) | fl->fl6_flowlabel;
218 hdr->payload_len = htons(seg_len);
219 hdr->nexthdr = proto;
220 hdr->hop_limit = hlimit;
222 ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
223 ipv6_addr_copy(&hdr->daddr, first_hop);
225 skb->priority = sk->sk_priority;
228 if ((skb->len <= mtu) || ipfragok || skb_is_gso(skb)) {
229 IP6_INC_STATS(ip6_dst_idev(skb->dst),
230 IPSTATS_MIB_OUTREQUESTS);
231 return NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, dst->dev,
236 printk(KERN_DEBUG "IPv6: sending pkt_too_big to self\n");
238 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev);
239 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGFAILS);
244 EXPORT_SYMBOL(ip6_xmit);
247 * To avoid extra problems ND packets are send through this
248 * routine. It's code duplication but I really want to avoid
249 * extra checks since ipv6_build_header is used by TCP (which
250 * is for us performance critical)
253 int ip6_nd_hdr(struct sock *sk, struct sk_buff *skb, struct net_device *dev,
254 struct in6_addr *saddr, struct in6_addr *daddr,
257 struct ipv6_pinfo *np = inet6_sk(sk);
261 skb->protocol = htons(ETH_P_IPV6);
264 totlen = len + sizeof(struct ipv6hdr);
266 skb->nh.raw = skb_put(skb, sizeof(struct ipv6hdr));
269 *(__be32*)hdr = htonl(0x60000000);
271 hdr->payload_len = htons(len);
272 hdr->nexthdr = proto;
273 hdr->hop_limit = np->hop_limit;
275 ipv6_addr_copy(&hdr->saddr, saddr);
276 ipv6_addr_copy(&hdr->daddr, daddr);
281 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
283 struct ip6_ra_chain *ra;
284 struct sock *last = NULL;
286 read_lock(&ip6_ra_lock);
287 for (ra = ip6_ra_chain; ra; ra = ra->next) {
288 struct sock *sk = ra->sk;
289 if (sk && ra->sel == sel &&
290 (!sk->sk_bound_dev_if ||
291 sk->sk_bound_dev_if == skb->dev->ifindex)) {
293 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
295 rawv6_rcv(last, skb2);
302 rawv6_rcv(last, skb);
303 read_unlock(&ip6_ra_lock);
306 read_unlock(&ip6_ra_lock);
310 static int ip6_forward_proxy_check(struct sk_buff *skb)
312 struct ipv6hdr *hdr = ipv6_hdr(skb);
313 u8 nexthdr = hdr->nexthdr;
316 if (ipv6_ext_hdr(nexthdr)) {
317 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr);
321 offset = sizeof(struct ipv6hdr);
323 if (nexthdr == IPPROTO_ICMPV6) {
324 struct icmp6hdr *icmp6;
326 if (!pskb_may_pull(skb, (skb_network_header(skb) +
327 offset + 1 - skb->data)))
330 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
332 switch (icmp6->icmp6_type) {
333 case NDISC_ROUTER_SOLICITATION:
334 case NDISC_ROUTER_ADVERTISEMENT:
335 case NDISC_NEIGHBOUR_SOLICITATION:
336 case NDISC_NEIGHBOUR_ADVERTISEMENT:
338 /* For reaction involving unicast neighbor discovery
339 * message destined to the proxied address, pass it to
349 * The proxying router can't forward traffic sent to a link-local
350 * address, so signal the sender and discard the packet. This
351 * behavior is clarified by the MIPv6 specification.
353 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
354 dst_link_failure(skb);
361 static inline int ip6_forward_finish(struct sk_buff *skb)
363 return dst_output(skb);
366 int ip6_forward(struct sk_buff *skb)
368 struct dst_entry *dst = skb->dst;
369 struct ipv6hdr *hdr = ipv6_hdr(skb);
370 struct inet6_skb_parm *opt = IP6CB(skb);
372 if (ipv6_devconf.forwarding == 0)
375 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
376 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
380 skb->ip_summed = CHECKSUM_NONE;
383 * We DO NOT make any processing on
384 * RA packets, pushing them to user level AS IS
385 * without ane WARRANTY that application will be able
386 * to interpret them. The reason is that we
387 * cannot make anything clever here.
389 * We are not end-node, so that if packet contains
390 * AH/ESP, we cannot make anything.
391 * Defragmentation also would be mistake, RA packets
392 * cannot be fragmented, because there is no warranty
393 * that different fragments will go along one path. --ANK
396 u8 *ptr = skb_network_header(skb) + opt->ra;
397 if (ip6_call_ra_chain(skb, (ptr[2]<<8) + ptr[3]))
402 * check and decrement ttl
404 if (hdr->hop_limit <= 1) {
405 /* Force OUTPUT device used as source address */
407 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT,
409 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INHDRERRORS);
415 /* XXX: idev->cnf.proxy_ndp? */
416 if (ipv6_devconf.proxy_ndp &&
417 pneigh_lookup(&nd_tbl, &hdr->daddr, skb->dev, 0)) {
418 int proxied = ip6_forward_proxy_check(skb);
420 return ip6_input(skb);
421 else if (proxied < 0) {
422 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
427 if (!xfrm6_route_forward(skb)) {
428 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
433 /* IPv6 specs say nothing about it, but it is clear that we cannot
434 send redirects to source routed frames.
436 if (skb->dev == dst->dev && dst->neighbour && opt->srcrt == 0) {
437 struct in6_addr *target = NULL;
439 struct neighbour *n = dst->neighbour;
442 * incoming and outgoing devices are the same
446 rt = (struct rt6_info *) dst;
447 if ((rt->rt6i_flags & RTF_GATEWAY))
448 target = (struct in6_addr*)&n->primary_key;
450 target = &hdr->daddr;
452 /* Limit redirects both by destination (here)
453 and by source (inside ndisc_send_redirect)
455 if (xrlim_allow(dst, 1*HZ))
456 ndisc_send_redirect(skb, n, target);
457 } else if (ipv6_addr_type(&hdr->saddr)&(IPV6_ADDR_MULTICAST|IPV6_ADDR_LOOPBACK
458 |IPV6_ADDR_LINKLOCAL)) {
459 /* This check is security critical. */
463 if (skb->len > dst_mtu(dst)) {
464 /* Again, force OUTPUT device used as source address */
466 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, dst_mtu(dst), skb->dev);
467 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INTOOBIGERRORS);
468 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_FRAGFAILS);
473 if (skb_cow(skb, dst->dev->hard_header_len)) {
474 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_OUTDISCARDS);
480 /* Mangling hops number delayed to point after skb COW */
484 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
485 return NF_HOOK(PF_INET6,NF_IP6_FORWARD, skb, skb->dev, dst->dev, ip6_forward_finish);
488 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
494 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
496 to->pkt_type = from->pkt_type;
497 to->priority = from->priority;
498 to->protocol = from->protocol;
499 dst_release(to->dst);
500 to->dst = dst_clone(from->dst);
502 to->mark = from->mark;
504 #ifdef CONFIG_NET_SCHED
505 to->tc_index = from->tc_index;
508 skb_copy_secmark(to, from);
511 int ip6_find_1stfragopt(struct sk_buff *skb, u8 **nexthdr)
513 u16 offset = sizeof(struct ipv6hdr);
514 struct ipv6_opt_hdr *exthdr =
515 (struct ipv6_opt_hdr *)(ipv6_hdr(skb) + 1);
516 unsigned int packet_len = skb->tail - skb_network_header(skb);
518 *nexthdr = &ipv6_hdr(skb)->nexthdr;
520 while (offset + 1 <= packet_len) {
526 case NEXTHDR_ROUTING:
530 #ifdef CONFIG_IPV6_MIP6
531 if (ipv6_find_tlv(skb, offset, IPV6_TLV_HAO) >= 0)
541 offset += ipv6_optlen(exthdr);
542 *nexthdr = &exthdr->nexthdr;
543 exthdr = (struct ipv6_opt_hdr *)(skb_network_header(skb) +
549 EXPORT_SYMBOL_GPL(ip6_find_1stfragopt);
551 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
553 struct net_device *dev;
554 struct sk_buff *frag;
555 struct rt6_info *rt = (struct rt6_info*)skb->dst;
556 struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
557 struct ipv6hdr *tmp_hdr;
559 unsigned int mtu, hlen, left, len;
561 int ptr, offset = 0, err=0;
562 u8 *prevhdr, nexthdr = 0;
565 hlen = ip6_find_1stfragopt(skb, &prevhdr);
568 mtu = dst_mtu(&rt->u.dst);
569 if (np && np->frag_size < mtu) {
573 mtu -= hlen + sizeof(struct frag_hdr);
575 if (skb_shinfo(skb)->frag_list) {
576 int first_len = skb_pagelen(skb);
578 if (first_len - hlen > mtu ||
579 ((first_len - hlen) & 7) ||
583 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
584 /* Correct geometry. */
585 if (frag->len > mtu ||
586 ((frag->len & 7) && frag->next) ||
587 skb_headroom(frag) < hlen)
590 /* Partially cloned skb? */
591 if (skb_shared(frag))
598 frag->destructor = sock_wfree;
599 skb->truesize -= frag->truesize;
605 frag = skb_shinfo(skb)->frag_list;
606 skb_shinfo(skb)->frag_list = NULL;
609 *prevhdr = NEXTHDR_FRAGMENT;
610 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
612 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGFAILS);
616 __skb_pull(skb, hlen);
617 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
618 __skb_push(skb, hlen);
619 skb_reset_network_header(skb);
620 memcpy(skb_network_header(skb), tmp_hdr, hlen);
622 ipv6_select_ident(skb, fh);
623 fh->nexthdr = nexthdr;
625 fh->frag_off = htons(IP6_MF);
626 frag_id = fh->identification;
628 first_len = skb_pagelen(skb);
629 skb->data_len = first_len - skb_headlen(skb);
630 skb->len = first_len;
631 ipv6_hdr(skb)->payload_len = htons(first_len -
632 sizeof(struct ipv6hdr));
634 dst_hold(&rt->u.dst);
637 /* Prepare header of the next frame,
638 * before previous one went down. */
640 frag->ip_summed = CHECKSUM_NONE;
641 skb_reset_transport_header(frag);
642 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
643 __skb_push(frag, hlen);
644 skb_reset_network_header(frag);
645 memcpy(skb_network_header(frag), tmp_hdr,
647 offset += skb->len - hlen - sizeof(struct frag_hdr);
648 fh->nexthdr = nexthdr;
650 fh->frag_off = htons(offset);
651 if (frag->next != NULL)
652 fh->frag_off |= htons(IP6_MF);
653 fh->identification = frag_id;
654 ipv6_hdr(frag)->payload_len =
656 sizeof(struct ipv6hdr));
657 ip6_copy_metadata(frag, skb);
662 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGCREATES);
675 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGOKS);
676 dst_release(&rt->u.dst);
686 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGFAILS);
687 dst_release(&rt->u.dst);
692 left = skb->len - hlen; /* Space per frame */
693 ptr = hlen; /* Where to start from */
696 * Fragment the datagram.
699 *prevhdr = NEXTHDR_FRAGMENT;
702 * Keep copying data until we run out.
706 /* IF: it doesn't fit, use 'mtu' - the data space left */
709 /* IF: we are not sending upto and including the packet end
710 then align the next start on an eight byte boundary */
718 if ((frag = alloc_skb(len+hlen+sizeof(struct frag_hdr)+LL_RESERVED_SPACE(rt->u.dst.dev), GFP_ATOMIC)) == NULL) {
719 NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
720 IP6_INC_STATS(ip6_dst_idev(skb->dst),
721 IPSTATS_MIB_FRAGFAILS);
727 * Set up data on packet
730 ip6_copy_metadata(frag, skb);
731 skb_reserve(frag, LL_RESERVED_SPACE(rt->u.dst.dev));
732 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
733 skb_reset_network_header(frag);
734 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
735 frag->h.raw = frag->nh.raw + hlen + sizeof(struct frag_hdr);
738 * Charge the memory for the fragment to any owner
742 skb_set_owner_w(frag, skb->sk);
745 * Copy the packet header into the new buffer.
747 memcpy(skb_network_header(frag), skb->data, hlen);
750 * Build fragment header.
752 fh->nexthdr = nexthdr;
755 ipv6_select_ident(skb, fh);
756 frag_id = fh->identification;
758 fh->identification = frag_id;
761 * Copy a block of the IP datagram.
763 if (skb_copy_bits(skb, ptr, frag->h.raw, len))
767 fh->frag_off = htons(offset);
769 fh->frag_off |= htons(IP6_MF);
770 ipv6_hdr(frag)->payload_len = htons(frag->len -
771 sizeof(struct ipv6hdr));
777 * Put this fragment into the sending queue.
783 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGCREATES);
785 IP6_INC_STATS(ip6_dst_idev(skb->dst),
786 IPSTATS_MIB_FRAGOKS);
791 IP6_INC_STATS(ip6_dst_idev(skb->dst),
792 IPSTATS_MIB_FRAGFAILS);
797 static inline int ip6_rt_check(struct rt6key *rt_key,
798 struct in6_addr *fl_addr,
799 struct in6_addr *addr_cache)
801 return ((rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
802 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache)));
805 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
806 struct dst_entry *dst,
809 struct ipv6_pinfo *np = inet6_sk(sk);
810 struct rt6_info *rt = (struct rt6_info *)dst;
815 /* Yes, checking route validity in not connected
816 * case is not very simple. Take into account,
817 * that we do not support routing by source, TOS,
818 * and MSG_DONTROUTE --ANK (980726)
820 * 1. ip6_rt_check(): If route was host route,
821 * check that cached destination is current.
822 * If it is network route, we still may
823 * check its validity using saved pointer
824 * to the last used address: daddr_cache.
825 * We do not want to save whole address now,
826 * (because main consumer of this service
827 * is tcp, which has not this problem),
828 * so that the last trick works only on connected
830 * 2. oif also should be the same.
832 if (ip6_rt_check(&rt->rt6i_dst, &fl->fl6_dst, np->daddr_cache) ||
833 #ifdef CONFIG_IPV6_SUBTREES
834 ip6_rt_check(&rt->rt6i_src, &fl->fl6_src, np->saddr_cache) ||
836 (fl->oif && fl->oif != dst->dev->ifindex)) {
845 static int ip6_dst_lookup_tail(struct sock *sk,
846 struct dst_entry **dst, struct flowi *fl)
851 *dst = ip6_route_output(sk, fl);
853 if ((err = (*dst)->error))
854 goto out_err_release;
856 if (ipv6_addr_any(&fl->fl6_src)) {
857 err = ipv6_get_saddr(*dst, &fl->fl6_dst, &fl->fl6_src);
859 goto out_err_release;
862 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
864 * Here if the dst entry we've looked up
865 * has a neighbour entry that is in the INCOMPLETE
866 * state and the src address from the flow is
867 * marked as OPTIMISTIC, we release the found
868 * dst entry and replace it instead with the
869 * dst entry of the nexthop router
871 if (!((*dst)->neighbour->nud_state & NUD_VALID)) {
872 struct inet6_ifaddr *ifp;
876 ifp = ipv6_get_ifaddr(&fl->fl6_src, (*dst)->dev, 1);
878 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
884 * We need to get the dst entry for the
885 * default router instead
888 memcpy(&fl_gw, fl, sizeof(struct flowi));
889 memset(&fl_gw.fl6_dst, 0, sizeof(struct in6_addr));
890 *dst = ip6_route_output(sk, &fl_gw);
891 if ((err = (*dst)->error))
892 goto out_err_release;
906 * ip6_dst_lookup - perform route lookup on flow
907 * @sk: socket which provides route info
908 * @dst: pointer to dst_entry * for result
909 * @fl: flow to lookup
911 * This function performs a route lookup on the given flow.
913 * It returns zero on success, or a standard errno code on error.
915 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
918 return ip6_dst_lookup_tail(sk, dst, fl);
920 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
923 * ip6_sk_dst_lookup - perform socket cached route lookup on flow
924 * @sk: socket which provides the dst cache and route info
925 * @dst: pointer to dst_entry * for result
926 * @fl: flow to lookup
928 * This function performs a route lookup on the given flow with the
929 * possibility of using the cached route in the socket if it is valid.
930 * It will take the socket dst lock when operating on the dst cache.
931 * As a result, this function can only be used in process context.
933 * It returns zero on success, or a standard errno code on error.
935 int ip6_sk_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
939 *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
940 *dst = ip6_sk_dst_check(sk, *dst, fl);
943 return ip6_dst_lookup_tail(sk, dst, fl);
945 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup);
947 static inline int ip6_ufo_append_data(struct sock *sk,
948 int getfrag(void *from, char *to, int offset, int len,
949 int odd, struct sk_buff *skb),
950 void *from, int length, int hh_len, int fragheaderlen,
951 int transhdrlen, int mtu,unsigned int flags)
957 /* There is support for UDP large send offload by network
958 * device, so create one single skb packet containing complete
961 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
962 skb = sock_alloc_send_skb(sk,
963 hh_len + fragheaderlen + transhdrlen + 20,
964 (flags & MSG_DONTWAIT), &err);
968 /* reserve space for Hardware header */
969 skb_reserve(skb, hh_len);
971 /* create space for UDP/IP header */
972 skb_put(skb,fragheaderlen + transhdrlen);
974 /* initialize network header pointer */
975 skb_reset_network_header(skb);
977 /* initialize protocol header pointer */
978 skb->h.raw = skb->nh.raw + fragheaderlen;
980 skb->ip_summed = CHECKSUM_PARTIAL;
982 sk->sk_sndmsg_off = 0;
985 err = skb_append_datato_frags(sk,skb, getfrag, from,
986 (length - transhdrlen));
988 struct frag_hdr fhdr;
990 /* specify the length of each IP datagram fragment*/
991 skb_shinfo(skb)->gso_size = mtu - fragheaderlen -
992 sizeof(struct frag_hdr);
993 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
994 ipv6_select_ident(skb, &fhdr);
995 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
996 __skb_queue_tail(&sk->sk_write_queue, skb);
1000 /* There is not enough support do UPD LSO,
1001 * so follow normal path
1008 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1009 int offset, int len, int odd, struct sk_buff *skb),
1010 void *from, int length, int transhdrlen,
1011 int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi *fl,
1012 struct rt6_info *rt, unsigned int flags)
1014 struct inet_sock *inet = inet_sk(sk);
1015 struct ipv6_pinfo *np = inet6_sk(sk);
1016 struct sk_buff *skb;
1017 unsigned int maxfraglen, fragheaderlen;
1024 int csummode = CHECKSUM_NONE;
1026 if (flags&MSG_PROBE)
1028 if (skb_queue_empty(&sk->sk_write_queue)) {
1033 if (np->cork.opt == NULL) {
1034 np->cork.opt = kmalloc(opt->tot_len,
1036 if (unlikely(np->cork.opt == NULL))
1038 } else if (np->cork.opt->tot_len < opt->tot_len) {
1039 printk(KERN_DEBUG "ip6_append_data: invalid option length\n");
1042 memcpy(np->cork.opt, opt, opt->tot_len);
1043 inet->cork.flags |= IPCORK_OPT;
1044 /* need source address above miyazawa*/
1046 dst_hold(&rt->u.dst);
1048 inet->cork.fl = *fl;
1049 np->cork.hop_limit = hlimit;
1050 np->cork.tclass = tclass;
1051 mtu = dst_mtu(rt->u.dst.path);
1052 if (np->frag_size < mtu) {
1054 mtu = np->frag_size;
1056 inet->cork.fragsize = mtu;
1057 if (dst_allfrag(rt->u.dst.path))
1058 inet->cork.flags |= IPCORK_ALLFRAG;
1059 inet->cork.length = 0;
1060 sk->sk_sndmsg_page = NULL;
1061 sk->sk_sndmsg_off = 0;
1062 exthdrlen = rt->u.dst.header_len + (opt ? opt->opt_flen : 0);
1063 length += exthdrlen;
1064 transhdrlen += exthdrlen;
1067 fl = &inet->cork.fl;
1068 if (inet->cork.flags & IPCORK_OPT)
1072 mtu = inet->cork.fragsize;
1075 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1077 fragheaderlen = sizeof(struct ipv6hdr) + rt->u.dst.nfheader_len + (opt ? opt->opt_nflen : 0);
1078 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
1080 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1081 if (inet->cork.length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
1082 ipv6_local_error(sk, EMSGSIZE, fl, mtu-exthdrlen);
1088 * Let's try using as much space as possible.
1089 * Use MTU if total length of the message fits into the MTU.
1090 * Otherwise, we need to reserve fragment header and
1091 * fragment alignment (= 8-15 octects, in total).
1093 * Note that we may need to "move" the data from the tail of
1094 * of the buffer to the new fragment when we split
1097 * FIXME: It may be fragmented into multiple chunks
1098 * at once if non-fragmentable extension headers
1103 inet->cork.length += length;
1104 if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
1105 (rt->u.dst.dev->features & NETIF_F_UFO)) {
1107 err = ip6_ufo_append_data(sk, getfrag, from, length, hh_len,
1108 fragheaderlen, transhdrlen, mtu,
1115 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1118 while (length > 0) {
1119 /* Check if the remaining data fits into current packet. */
1120 copy = (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1122 copy = maxfraglen - skb->len;
1126 unsigned int datalen;
1127 unsigned int fraglen;
1128 unsigned int fraggap;
1129 unsigned int alloclen;
1130 struct sk_buff *skb_prev;
1134 /* There's no room in the current skb */
1136 fraggap = skb_prev->len - maxfraglen;
1141 * If remaining data exceeds the mtu,
1142 * we know we need more fragment(s).
1144 datalen = length + fraggap;
1145 if (datalen > (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1146 datalen = maxfraglen - fragheaderlen;
1148 fraglen = datalen + fragheaderlen;
1149 if ((flags & MSG_MORE) &&
1150 !(rt->u.dst.dev->features&NETIF_F_SG))
1153 alloclen = datalen + fragheaderlen;
1156 * The last fragment gets additional space at tail.
1157 * Note: we overallocate on fragments with MSG_MODE
1158 * because we have no idea if we're the last one.
1160 if (datalen == length + fraggap)
1161 alloclen += rt->u.dst.trailer_len;
1164 * We just reserve space for fragment header.
1165 * Note: this may be overallocation if the message
1166 * (without MSG_MORE) fits into the MTU.
1168 alloclen += sizeof(struct frag_hdr);
1171 skb = sock_alloc_send_skb(sk,
1173 (flags & MSG_DONTWAIT), &err);
1176 if (atomic_read(&sk->sk_wmem_alloc) <=
1178 skb = sock_wmalloc(sk,
1179 alloclen + hh_len, 1,
1181 if (unlikely(skb == NULL))
1187 * Fill in the control structures
1189 skb->ip_summed = csummode;
1191 /* reserve for fragmentation */
1192 skb_reserve(skb, hh_len+sizeof(struct frag_hdr));
1195 * Find where to start putting bytes
1197 data = skb_put(skb, fraglen);
1198 skb_set_network_header(skb, exthdrlen);
1199 data += fragheaderlen;
1200 skb->h.raw = skb->nh.raw + fragheaderlen;
1203 skb->csum = skb_copy_and_csum_bits(
1204 skb_prev, maxfraglen,
1205 data + transhdrlen, fraggap, 0);
1206 skb_prev->csum = csum_sub(skb_prev->csum,
1209 pskb_trim_unique(skb_prev, maxfraglen);
1211 copy = datalen - transhdrlen - fraggap;
1216 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1223 length -= datalen - fraggap;
1226 csummode = CHECKSUM_NONE;
1229 * Put the packet on the pending queue
1231 __skb_queue_tail(&sk->sk_write_queue, skb);
1238 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
1242 if (getfrag(from, skb_put(skb, copy),
1243 offset, copy, off, skb) < 0) {
1244 __skb_trim(skb, off);
1249 int i = skb_shinfo(skb)->nr_frags;
1250 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1251 struct page *page = sk->sk_sndmsg_page;
1252 int off = sk->sk_sndmsg_off;
1255 if (page && (left = PAGE_SIZE - off) > 0) {
1258 if (page != frag->page) {
1259 if (i == MAX_SKB_FRAGS) {
1264 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1265 frag = &skb_shinfo(skb)->frags[i];
1267 } else if(i < MAX_SKB_FRAGS) {
1268 if (copy > PAGE_SIZE)
1270 page = alloc_pages(sk->sk_allocation, 0);
1275 sk->sk_sndmsg_page = page;
1276 sk->sk_sndmsg_off = 0;
1278 skb_fill_page_desc(skb, i, page, 0, 0);
1279 frag = &skb_shinfo(skb)->frags[i];
1280 skb->truesize += PAGE_SIZE;
1281 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1286 if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1290 sk->sk_sndmsg_off += copy;
1293 skb->data_len += copy;
1300 inet->cork.length -= length;
1301 IP6_INC_STATS(rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1305 int ip6_push_pending_frames(struct sock *sk)
1307 struct sk_buff *skb, *tmp_skb;
1308 struct sk_buff **tail_skb;
1309 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1310 struct inet_sock *inet = inet_sk(sk);
1311 struct ipv6_pinfo *np = inet6_sk(sk);
1312 struct ipv6hdr *hdr;
1313 struct ipv6_txoptions *opt = np->cork.opt;
1314 struct rt6_info *rt = np->cork.rt;
1315 struct flowi *fl = &inet->cork.fl;
1316 unsigned char proto = fl->proto;
1319 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1321 tail_skb = &(skb_shinfo(skb)->frag_list);
1323 /* move skb->data to ip header from ext header */
1324 if (skb->data < skb_network_header(skb))
1325 __skb_pull(skb, skb_network_offset(skb));
1326 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1327 __skb_pull(tmp_skb, skb->h.raw - skb->nh.raw);
1328 *tail_skb = tmp_skb;
1329 tail_skb = &(tmp_skb->next);
1330 skb->len += tmp_skb->len;
1331 skb->data_len += tmp_skb->len;
1332 skb->truesize += tmp_skb->truesize;
1333 __sock_put(tmp_skb->sk);
1334 tmp_skb->destructor = NULL;
1338 ipv6_addr_copy(final_dst, &fl->fl6_dst);
1339 __skb_pull(skb, skb->h.raw - skb->nh.raw);
1340 if (opt && opt->opt_flen)
1341 ipv6_push_frag_opts(skb, opt, &proto);
1342 if (opt && opt->opt_nflen)
1343 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1345 skb_push(skb, sizeof(struct ipv6hdr));
1346 skb_reset_network_header(skb);
1347 hdr = ipv6_hdr(skb);
1349 *(__be32*)hdr = fl->fl6_flowlabel |
1350 htonl(0x60000000 | ((int)np->cork.tclass << 20));
1352 if (skb->len <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN)
1353 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
1355 hdr->payload_len = 0;
1356 hdr->hop_limit = np->cork.hop_limit;
1357 hdr->nexthdr = proto;
1358 ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
1359 ipv6_addr_copy(&hdr->daddr, final_dst);
1361 skb->priority = sk->sk_priority;
1363 skb->dst = dst_clone(&rt->u.dst);
1364 IP6_INC_STATS(rt->rt6i_idev, IPSTATS_MIB_OUTREQUESTS);
1365 err = NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, skb->dst->dev, dst_output);
1368 err = np->recverr ? net_xmit_errno(err) : 0;
1374 inet->cork.flags &= ~IPCORK_OPT;
1375 kfree(np->cork.opt);
1376 np->cork.opt = NULL;
1378 dst_release(&np->cork.rt->u.dst);
1380 inet->cork.flags &= ~IPCORK_ALLFRAG;
1382 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1388 void ip6_flush_pending_frames(struct sock *sk)
1390 struct inet_sock *inet = inet_sk(sk);
1391 struct ipv6_pinfo *np = inet6_sk(sk);
1392 struct sk_buff *skb;
1394 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1395 IP6_INC_STATS(ip6_dst_idev(skb->dst),
1396 IPSTATS_MIB_OUTDISCARDS);
1400 inet->cork.flags &= ~IPCORK_OPT;
1402 kfree(np->cork.opt);
1403 np->cork.opt = NULL;
1405 dst_release(&np->cork.rt->u.dst);
1407 inet->cork.flags &= ~IPCORK_ALLFRAG;
1409 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));