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)
76 struct dst_entry *dst = skb->dst;
77 struct hh_cache *hh = dst->hh;
82 read_lock_bh(&hh->hh_lock);
83 hh_alen = HH_DATA_ALIGN(hh->hh_len);
84 memcpy(skb->data - hh_alen, hh->hh_data, hh_alen);
85 read_unlock_bh(&hh->hh_lock);
86 skb_push(skb, hh->hh_len);
87 return hh->hh_output(skb);
88 } else if (dst->neighbour)
89 return dst->neighbour->output(skb);
91 IP6_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
97 /* dev_loopback_xmit for use with netfilter. */
98 static int ip6_dev_loopback_xmit(struct sk_buff *newskb)
100 newskb->mac.raw = newskb->data;
101 __skb_pull(newskb, newskb->nh.raw - newskb->data);
102 newskb->pkt_type = PACKET_LOOPBACK;
103 newskb->ip_summed = CHECKSUM_UNNECESSARY;
104 BUG_TRAP(newskb->dst);
111 static int ip6_output2(struct sk_buff *skb)
113 struct dst_entry *dst = skb->dst;
114 struct net_device *dev = dst->dev;
116 skb->protocol = htons(ETH_P_IPV6);
119 if (ipv6_addr_is_multicast(&skb->nh.ipv6h->daddr)) {
120 struct ipv6_pinfo* np = skb->sk ? inet6_sk(skb->sk) : NULL;
122 if (!(dev->flags & IFF_LOOPBACK) && (!np || np->mc_loop) &&
123 ipv6_chk_mcast_addr(dev, &skb->nh.ipv6h->daddr,
124 &skb->nh.ipv6h->saddr)) {
125 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
127 /* Do not check for IFF_ALLMULTI; multicast routing
128 is not supported in any case.
131 NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, newskb, NULL,
133 ip6_dev_loopback_xmit);
135 if (skb->nh.ipv6h->hop_limit == 0) {
136 IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
142 IP6_INC_STATS(IPSTATS_MIB_OUTMCASTPKTS);
145 return NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, skb,NULL, skb->dev,ip6_output_finish);
148 int ip6_output(struct sk_buff *skb)
150 if ((skb->len > dst_mtu(skb->dst) && !skb_is_gso(skb)) ||
151 dst_allfrag(skb->dst))
152 return ip6_fragment(skb, ip6_output2);
154 return ip6_output2(skb);
158 * xmit an sk_buff (used by TCP)
161 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
162 struct ipv6_txoptions *opt, int ipfragok)
164 struct ipv6_pinfo *np = inet6_sk(sk);
165 struct in6_addr *first_hop = &fl->fl6_dst;
166 struct dst_entry *dst = skb->dst;
168 u8 proto = fl->proto;
169 int seg_len = skb->len;
176 /* First: exthdrs may take lots of space (~8K for now)
177 MAX_HEADER is not enough.
179 head_room = opt->opt_nflen + opt->opt_flen;
180 seg_len += head_room;
181 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
183 if (skb_headroom(skb) < head_room) {
184 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
188 IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
192 skb_set_owner_w(skb, sk);
195 ipv6_push_frag_opts(skb, opt, &proto);
197 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
200 hdr = skb->nh.ipv6h = (struct ipv6hdr*)skb_push(skb, sizeof(struct ipv6hdr));
203 * Fill in the IPv6 header
208 hlimit = np->hop_limit;
210 hlimit = dst_metric(dst, RTAX_HOPLIMIT);
212 hlimit = ipv6_get_hoplimit(dst->dev);
220 *(u32 *)hdr = htonl(0x60000000 | (tclass << 20)) | fl->fl6_flowlabel;
222 hdr->payload_len = htons(seg_len);
223 hdr->nexthdr = proto;
224 hdr->hop_limit = hlimit;
226 ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
227 ipv6_addr_copy(&hdr->daddr, first_hop);
229 skb->priority = sk->sk_priority;
232 if ((skb->len <= mtu) || ipfragok || skb_is_gso(skb)) {
233 IP6_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
234 return NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, dst->dev,
239 printk(KERN_DEBUG "IPv6: sending pkt_too_big to self\n");
241 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev);
242 IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
248 * To avoid extra problems ND packets are send through this
249 * routine. It's code duplication but I really want to avoid
250 * extra checks since ipv6_build_header is used by TCP (which
251 * is for us performance critical)
254 int ip6_nd_hdr(struct sock *sk, struct sk_buff *skb, struct net_device *dev,
255 struct in6_addr *saddr, struct in6_addr *daddr,
258 struct ipv6_pinfo *np = inet6_sk(sk);
262 skb->protocol = htons(ETH_P_IPV6);
265 totlen = len + sizeof(struct ipv6hdr);
267 hdr = (struct ipv6hdr *) skb_put(skb, sizeof(struct ipv6hdr));
270 *(u32*)hdr = htonl(0x60000000);
272 hdr->payload_len = htons(len);
273 hdr->nexthdr = proto;
274 hdr->hop_limit = np->hop_limit;
276 ipv6_addr_copy(&hdr->saddr, saddr);
277 ipv6_addr_copy(&hdr->daddr, daddr);
282 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
284 struct ip6_ra_chain *ra;
285 struct sock *last = NULL;
287 read_lock(&ip6_ra_lock);
288 for (ra = ip6_ra_chain; ra; ra = ra->next) {
289 struct sock *sk = ra->sk;
290 if (sk && ra->sel == sel &&
291 (!sk->sk_bound_dev_if ||
292 sk->sk_bound_dev_if == skb->dev->ifindex)) {
294 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
296 rawv6_rcv(last, skb2);
303 rawv6_rcv(last, skb);
304 read_unlock(&ip6_ra_lock);
307 read_unlock(&ip6_ra_lock);
311 static int ip6_forward_proxy_check(struct sk_buff *skb)
313 struct ipv6hdr *hdr = skb->nh.ipv6h;
314 u8 nexthdr = hdr->nexthdr;
317 if (ipv6_ext_hdr(nexthdr)) {
318 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr);
322 offset = sizeof(struct ipv6hdr);
324 if (nexthdr == IPPROTO_ICMPV6) {
325 struct icmp6hdr *icmp6;
327 if (!pskb_may_pull(skb, skb->nh.raw + offset + 1 - skb->data))
330 icmp6 = (struct icmp6hdr *)(skb->nh.raw + 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 = skb->nh.ipv6h;
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(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->nh.raw + 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(IPSTATS_MIB_INHDRERRORS);
415 if (pneigh_lookup(&nd_tbl, &hdr->daddr, skb->dev, 0)) {
416 int proxied = ip6_forward_proxy_check(skb);
418 return ip6_input(skb);
419 else if (proxied < 0) {
420 IP6_INC_STATS(IPSTATS_MIB_INDISCARDS);
425 if (!xfrm6_route_forward(skb)) {
426 IP6_INC_STATS(IPSTATS_MIB_INDISCARDS);
431 /* IPv6 specs say nothing about it, but it is clear that we cannot
432 send redirects to source routed frames.
434 if (skb->dev == dst->dev && dst->neighbour && opt->srcrt == 0) {
435 struct in6_addr *target = NULL;
437 struct neighbour *n = dst->neighbour;
440 * incoming and outgoing devices are the same
444 rt = (struct rt6_info *) dst;
445 if ((rt->rt6i_flags & RTF_GATEWAY))
446 target = (struct in6_addr*)&n->primary_key;
448 target = &hdr->daddr;
450 /* Limit redirects both by destination (here)
451 and by source (inside ndisc_send_redirect)
453 if (xrlim_allow(dst, 1*HZ))
454 ndisc_send_redirect(skb, n, target);
455 } else if (ipv6_addr_type(&hdr->saddr)&(IPV6_ADDR_MULTICAST|IPV6_ADDR_LOOPBACK
456 |IPV6_ADDR_LINKLOCAL)) {
457 /* This check is security critical. */
461 if (skb->len > dst_mtu(dst)) {
462 /* Again, force OUTPUT device used as source address */
464 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, dst_mtu(dst), skb->dev);
465 IP6_INC_STATS_BH(IPSTATS_MIB_INTOOBIGERRORS);
466 IP6_INC_STATS_BH(IPSTATS_MIB_FRAGFAILS);
471 if (skb_cow(skb, dst->dev->hard_header_len)) {
472 IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
478 /* Mangling hops number delayed to point after skb COW */
482 IP6_INC_STATS_BH(IPSTATS_MIB_OUTFORWDATAGRAMS);
483 return NF_HOOK(PF_INET6,NF_IP6_FORWARD, skb, skb->dev, dst->dev, ip6_forward_finish);
486 IP6_INC_STATS_BH(IPSTATS_MIB_INADDRERRORS);
492 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
494 to->pkt_type = from->pkt_type;
495 to->priority = from->priority;
496 to->protocol = from->protocol;
497 dst_release(to->dst);
498 to->dst = dst_clone(from->dst);
501 #ifdef CONFIG_NET_SCHED
502 to->tc_index = from->tc_index;
504 #ifdef CONFIG_NETFILTER
505 to->nfmark = from->nfmark;
506 /* Connection association is same as pre-frag packet */
507 nf_conntrack_put(to->nfct);
508 to->nfct = from->nfct;
509 nf_conntrack_get(to->nfct);
510 to->nfctinfo = from->nfctinfo;
511 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
512 nf_conntrack_put_reasm(to->nfct_reasm);
513 to->nfct_reasm = from->nfct_reasm;
514 nf_conntrack_get_reasm(to->nfct_reasm);
516 #ifdef CONFIG_BRIDGE_NETFILTER
517 nf_bridge_put(to->nf_bridge);
518 to->nf_bridge = from->nf_bridge;
519 nf_bridge_get(to->nf_bridge);
522 skb_copy_secmark(to, from);
525 int ip6_find_1stfragopt(struct sk_buff *skb, u8 **nexthdr)
527 u16 offset = sizeof(struct ipv6hdr);
528 struct ipv6_opt_hdr *exthdr = (struct ipv6_opt_hdr*)(skb->nh.ipv6h + 1);
529 unsigned int packet_len = skb->tail - skb->nh.raw;
531 *nexthdr = &skb->nh.ipv6h->nexthdr;
533 while (offset + 1 <= packet_len) {
539 case NEXTHDR_ROUTING:
543 #ifdef CONFIG_IPV6_MIP6
544 if (ipv6_find_tlv(skb, offset, IPV6_TLV_HAO) >= 0)
554 offset += ipv6_optlen(exthdr);
555 *nexthdr = &exthdr->nexthdr;
556 exthdr = (struct ipv6_opt_hdr*)(skb->nh.raw + offset);
561 EXPORT_SYMBOL_GPL(ip6_find_1stfragopt);
563 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
565 struct net_device *dev;
566 struct sk_buff *frag;
567 struct rt6_info *rt = (struct rt6_info*)skb->dst;
568 struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
569 struct ipv6hdr *tmp_hdr;
571 unsigned int mtu, hlen, left, len;
573 int ptr, offset = 0, err=0;
574 u8 *prevhdr, nexthdr = 0;
577 hlen = ip6_find_1stfragopt(skb, &prevhdr);
580 mtu = dst_mtu(&rt->u.dst);
581 if (np && np->frag_size < mtu) {
585 mtu -= hlen + sizeof(struct frag_hdr);
587 if (skb_shinfo(skb)->frag_list) {
588 int first_len = skb_pagelen(skb);
590 if (first_len - hlen > mtu ||
591 ((first_len - hlen) & 7) ||
595 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
596 /* Correct geometry. */
597 if (frag->len > mtu ||
598 ((frag->len & 7) && frag->next) ||
599 skb_headroom(frag) < hlen)
602 /* Partially cloned skb? */
603 if (skb_shared(frag))
610 frag->destructor = sock_wfree;
611 skb->truesize -= frag->truesize;
617 frag = skb_shinfo(skb)->frag_list;
618 skb_shinfo(skb)->frag_list = NULL;
621 tmp_hdr = kmalloc(hlen, GFP_ATOMIC);
623 IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
627 *prevhdr = NEXTHDR_FRAGMENT;
628 memcpy(tmp_hdr, skb->nh.raw, hlen);
629 __skb_pull(skb, hlen);
630 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
631 skb->nh.raw = __skb_push(skb, hlen);
632 memcpy(skb->nh.raw, tmp_hdr, hlen);
634 ipv6_select_ident(skb, fh);
635 fh->nexthdr = nexthdr;
637 fh->frag_off = htons(IP6_MF);
638 frag_id = fh->identification;
640 first_len = skb_pagelen(skb);
641 skb->data_len = first_len - skb_headlen(skb);
642 skb->len = first_len;
643 skb->nh.ipv6h->payload_len = htons(first_len - sizeof(struct ipv6hdr));
647 /* Prepare header of the next frame,
648 * before previous one went down. */
650 frag->ip_summed = CHECKSUM_NONE;
651 frag->h.raw = frag->data;
652 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
653 frag->nh.raw = __skb_push(frag, hlen);
654 memcpy(frag->nh.raw, tmp_hdr, hlen);
655 offset += skb->len - hlen - sizeof(struct frag_hdr);
656 fh->nexthdr = nexthdr;
658 fh->frag_off = htons(offset);
659 if (frag->next != NULL)
660 fh->frag_off |= htons(IP6_MF);
661 fh->identification = frag_id;
662 frag->nh.ipv6h->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
663 ip6_copy_metadata(frag, skb);
668 IP6_INC_STATS(IPSTATS_MIB_FRAGCREATES);
681 IP6_INC_STATS(IPSTATS_MIB_FRAGOKS);
691 IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
696 left = skb->len - hlen; /* Space per frame */
697 ptr = hlen; /* Where to start from */
700 * Fragment the datagram.
703 *prevhdr = NEXTHDR_FRAGMENT;
706 * Keep copying data until we run out.
710 /* IF: it doesn't fit, use 'mtu' - the data space left */
713 /* IF: we are not sending upto and including the packet end
714 then align the next start on an eight byte boundary */
722 if ((frag = alloc_skb(len+hlen+sizeof(struct frag_hdr)+LL_RESERVED_SPACE(rt->u.dst.dev), GFP_ATOMIC)) == NULL) {
723 NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
724 IP6_INC_STATS(IPSTATS_MIB_FRAGFAILS);
730 * Set up data on packet
733 ip6_copy_metadata(frag, skb);
734 skb_reserve(frag, LL_RESERVED_SPACE(rt->u.dst.dev));
735 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
736 frag->nh.raw = frag->data;
737 fh = (struct frag_hdr*)(frag->data + hlen);
738 frag->h.raw = frag->data + hlen + sizeof(struct frag_hdr);
741 * Charge the memory for the fragment to any owner
745 skb_set_owner_w(frag, skb->sk);
748 * Copy the packet header into the new buffer.
750 memcpy(frag->nh.raw, skb->data, hlen);
753 * Build fragment header.
755 fh->nexthdr = nexthdr;
758 ipv6_select_ident(skb, fh);
759 frag_id = fh->identification;
761 fh->identification = frag_id;
764 * Copy a block of the IP datagram.
766 if (skb_copy_bits(skb, ptr, frag->h.raw, len))
770 fh->frag_off = htons(offset);
772 fh->frag_off |= htons(IP6_MF);
773 frag->nh.ipv6h->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
779 * Put this fragment into the sending queue.
785 IP6_INC_STATS(IPSTATS_MIB_FRAGCREATES);
788 IP6_INC_STATS(IPSTATS_MIB_FRAGOKS);
793 IP6_INC_STATS(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;
871 * ip6_dst_lookup - perform route lookup on flow
872 * @sk: socket which provides route info
873 * @dst: pointer to dst_entry * for result
874 * @fl: flow to lookup
876 * This function performs a route lookup on the given flow.
878 * It returns zero on success, or a standard errno code on error.
880 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
883 return ip6_dst_lookup_tail(sk, dst, fl);
885 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
888 * ip6_sk_dst_lookup - perform socket cached route lookup on flow
889 * @sk: socket which provides the dst cache and route info
890 * @dst: pointer to dst_entry * for result
891 * @fl: flow to lookup
893 * This function performs a route lookup on the given flow with the
894 * possibility of using the cached route in the socket if it is valid.
895 * It will take the socket dst lock when operating on the dst cache.
896 * As a result, this function can only be used in process context.
898 * It returns zero on success, or a standard errno code on error.
900 int ip6_sk_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
904 *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
905 *dst = ip6_sk_dst_check(sk, *dst, fl);
908 return ip6_dst_lookup_tail(sk, dst, fl);
910 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup);
912 static inline int ip6_ufo_append_data(struct sock *sk,
913 int getfrag(void *from, char *to, int offset, int len,
914 int odd, struct sk_buff *skb),
915 void *from, int length, int hh_len, int fragheaderlen,
916 int transhdrlen, int mtu,unsigned int flags)
922 /* There is support for UDP large send offload by network
923 * device, so create one single skb packet containing complete
926 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
927 skb = sock_alloc_send_skb(sk,
928 hh_len + fragheaderlen + transhdrlen + 20,
929 (flags & MSG_DONTWAIT), &err);
933 /* reserve space for Hardware header */
934 skb_reserve(skb, hh_len);
936 /* create space for UDP/IP header */
937 skb_put(skb,fragheaderlen + transhdrlen);
939 /* initialize network header pointer */
940 skb->nh.raw = skb->data;
942 /* initialize protocol header pointer */
943 skb->h.raw = skb->data + fragheaderlen;
945 skb->ip_summed = CHECKSUM_PARTIAL;
947 sk->sk_sndmsg_off = 0;
950 err = skb_append_datato_frags(sk,skb, getfrag, from,
951 (length - transhdrlen));
953 struct frag_hdr fhdr;
955 /* specify the length of each IP datagram fragment*/
956 skb_shinfo(skb)->gso_size = mtu - fragheaderlen -
957 sizeof(struct frag_hdr);
958 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
959 ipv6_select_ident(skb, &fhdr);
960 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
961 __skb_queue_tail(&sk->sk_write_queue, skb);
965 /* There is not enough support do UPD LSO,
966 * so follow normal path
973 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
974 int offset, int len, int odd, struct sk_buff *skb),
975 void *from, int length, int transhdrlen,
976 int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi *fl,
977 struct rt6_info *rt, unsigned int flags)
979 struct inet_sock *inet = inet_sk(sk);
980 struct ipv6_pinfo *np = inet6_sk(sk);
982 unsigned int maxfraglen, fragheaderlen;
989 int csummode = CHECKSUM_NONE;
993 if (skb_queue_empty(&sk->sk_write_queue)) {
998 if (np->cork.opt == NULL) {
999 np->cork.opt = kmalloc(opt->tot_len,
1001 if (unlikely(np->cork.opt == NULL))
1003 } else if (np->cork.opt->tot_len < opt->tot_len) {
1004 printk(KERN_DEBUG "ip6_append_data: invalid option length\n");
1007 memcpy(np->cork.opt, opt, opt->tot_len);
1008 inet->cork.flags |= IPCORK_OPT;
1009 /* need source address above miyazawa*/
1011 dst_hold(&rt->u.dst);
1013 inet->cork.fl = *fl;
1014 np->cork.hop_limit = hlimit;
1015 np->cork.tclass = tclass;
1016 mtu = dst_mtu(rt->u.dst.path);
1017 if (np->frag_size < mtu) {
1019 mtu = np->frag_size;
1021 inet->cork.fragsize = mtu;
1022 if (dst_allfrag(rt->u.dst.path))
1023 inet->cork.flags |= IPCORK_ALLFRAG;
1024 inet->cork.length = 0;
1025 sk->sk_sndmsg_page = NULL;
1026 sk->sk_sndmsg_off = 0;
1027 exthdrlen = rt->u.dst.header_len + (opt ? opt->opt_flen : 0);
1028 length += exthdrlen;
1029 transhdrlen += exthdrlen;
1032 fl = &inet->cork.fl;
1033 if (inet->cork.flags & IPCORK_OPT)
1037 mtu = inet->cork.fragsize;
1040 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1042 fragheaderlen = sizeof(struct ipv6hdr) + rt->u.dst.nfheader_len + (opt ? opt->opt_nflen : 0);
1043 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
1045 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1046 if (inet->cork.length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
1047 ipv6_local_error(sk, EMSGSIZE, fl, mtu-exthdrlen);
1053 * Let's try using as much space as possible.
1054 * Use MTU if total length of the message fits into the MTU.
1055 * Otherwise, we need to reserve fragment header and
1056 * fragment alignment (= 8-15 octects, in total).
1058 * Note that we may need to "move" the data from the tail of
1059 * of the buffer to the new fragment when we split
1062 * FIXME: It may be fragmented into multiple chunks
1063 * at once if non-fragmentable extension headers
1068 inet->cork.length += length;
1069 if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
1070 (rt->u.dst.dev->features & NETIF_F_UFO)) {
1072 err = ip6_ufo_append_data(sk, getfrag, from, length, hh_len,
1073 fragheaderlen, transhdrlen, mtu,
1080 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1083 while (length > 0) {
1084 /* Check if the remaining data fits into current packet. */
1085 copy = (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1087 copy = maxfraglen - skb->len;
1091 unsigned int datalen;
1092 unsigned int fraglen;
1093 unsigned int fraggap;
1094 unsigned int alloclen;
1095 struct sk_buff *skb_prev;
1099 /* There's no room in the current skb */
1101 fraggap = skb_prev->len - maxfraglen;
1106 * If remaining data exceeds the mtu,
1107 * we know we need more fragment(s).
1109 datalen = length + fraggap;
1110 if (datalen > (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1111 datalen = maxfraglen - fragheaderlen;
1113 fraglen = datalen + fragheaderlen;
1114 if ((flags & MSG_MORE) &&
1115 !(rt->u.dst.dev->features&NETIF_F_SG))
1118 alloclen = datalen + fragheaderlen;
1121 * The last fragment gets additional space at tail.
1122 * Note: we overallocate on fragments with MSG_MODE
1123 * because we have no idea if we're the last one.
1125 if (datalen == length + fraggap)
1126 alloclen += rt->u.dst.trailer_len;
1129 * We just reserve space for fragment header.
1130 * Note: this may be overallocation if the message
1131 * (without MSG_MORE) fits into the MTU.
1133 alloclen += sizeof(struct frag_hdr);
1136 skb = sock_alloc_send_skb(sk,
1138 (flags & MSG_DONTWAIT), &err);
1141 if (atomic_read(&sk->sk_wmem_alloc) <=
1143 skb = sock_wmalloc(sk,
1144 alloclen + hh_len, 1,
1146 if (unlikely(skb == NULL))
1152 * Fill in the control structures
1154 skb->ip_summed = csummode;
1156 /* reserve for fragmentation */
1157 skb_reserve(skb, hh_len+sizeof(struct frag_hdr));
1160 * Find where to start putting bytes
1162 data = skb_put(skb, fraglen);
1163 skb->nh.raw = data + exthdrlen;
1164 data += fragheaderlen;
1165 skb->h.raw = data + exthdrlen;
1168 skb->csum = skb_copy_and_csum_bits(
1169 skb_prev, maxfraglen,
1170 data + transhdrlen, fraggap, 0);
1171 skb_prev->csum = csum_sub(skb_prev->csum,
1174 pskb_trim_unique(skb_prev, maxfraglen);
1176 copy = datalen - transhdrlen - fraggap;
1181 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1188 length -= datalen - fraggap;
1191 csummode = CHECKSUM_NONE;
1194 * Put the packet on the pending queue
1196 __skb_queue_tail(&sk->sk_write_queue, skb);
1203 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
1207 if (getfrag(from, skb_put(skb, copy),
1208 offset, copy, off, skb) < 0) {
1209 __skb_trim(skb, off);
1214 int i = skb_shinfo(skb)->nr_frags;
1215 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1216 struct page *page = sk->sk_sndmsg_page;
1217 int off = sk->sk_sndmsg_off;
1220 if (page && (left = PAGE_SIZE - off) > 0) {
1223 if (page != frag->page) {
1224 if (i == MAX_SKB_FRAGS) {
1229 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1230 frag = &skb_shinfo(skb)->frags[i];
1232 } else if(i < MAX_SKB_FRAGS) {
1233 if (copy > PAGE_SIZE)
1235 page = alloc_pages(sk->sk_allocation, 0);
1240 sk->sk_sndmsg_page = page;
1241 sk->sk_sndmsg_off = 0;
1243 skb_fill_page_desc(skb, i, page, 0, 0);
1244 frag = &skb_shinfo(skb)->frags[i];
1245 skb->truesize += PAGE_SIZE;
1246 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1251 if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1255 sk->sk_sndmsg_off += copy;
1258 skb->data_len += copy;
1265 inet->cork.length -= length;
1266 IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1270 int ip6_push_pending_frames(struct sock *sk)
1272 struct sk_buff *skb, *tmp_skb;
1273 struct sk_buff **tail_skb;
1274 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1275 struct inet_sock *inet = inet_sk(sk);
1276 struct ipv6_pinfo *np = inet6_sk(sk);
1277 struct ipv6hdr *hdr;
1278 struct ipv6_txoptions *opt = np->cork.opt;
1279 struct rt6_info *rt = np->cork.rt;
1280 struct flowi *fl = &inet->cork.fl;
1281 unsigned char proto = fl->proto;
1284 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1286 tail_skb = &(skb_shinfo(skb)->frag_list);
1288 /* move skb->data to ip header from ext header */
1289 if (skb->data < skb->nh.raw)
1290 __skb_pull(skb, skb->nh.raw - skb->data);
1291 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1292 __skb_pull(tmp_skb, skb->h.raw - skb->nh.raw);
1293 *tail_skb = tmp_skb;
1294 tail_skb = &(tmp_skb->next);
1295 skb->len += tmp_skb->len;
1296 skb->data_len += tmp_skb->len;
1297 skb->truesize += tmp_skb->truesize;
1298 __sock_put(tmp_skb->sk);
1299 tmp_skb->destructor = NULL;
1303 ipv6_addr_copy(final_dst, &fl->fl6_dst);
1304 __skb_pull(skb, skb->h.raw - skb->nh.raw);
1305 if (opt && opt->opt_flen)
1306 ipv6_push_frag_opts(skb, opt, &proto);
1307 if (opt && opt->opt_nflen)
1308 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1310 skb->nh.ipv6h = hdr = (struct ipv6hdr*) skb_push(skb, sizeof(struct ipv6hdr));
1312 *(u32*)hdr = fl->fl6_flowlabel |
1313 htonl(0x60000000 | ((int)np->cork.tclass << 20));
1315 if (skb->len <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN)
1316 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
1318 hdr->payload_len = 0;
1319 hdr->hop_limit = np->cork.hop_limit;
1320 hdr->nexthdr = proto;
1321 ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
1322 ipv6_addr_copy(&hdr->daddr, final_dst);
1324 skb->priority = sk->sk_priority;
1326 skb->dst = dst_clone(&rt->u.dst);
1327 IP6_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
1328 err = NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, skb->dst->dev, dst_output);
1331 err = np->recverr ? net_xmit_errno(err) : 0;
1337 inet->cork.flags &= ~IPCORK_OPT;
1338 kfree(np->cork.opt);
1339 np->cork.opt = NULL;
1341 dst_release(&np->cork.rt->u.dst);
1343 inet->cork.flags &= ~IPCORK_ALLFRAG;
1345 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1351 void ip6_flush_pending_frames(struct sock *sk)
1353 struct inet_sock *inet = inet_sk(sk);
1354 struct ipv6_pinfo *np = inet6_sk(sk);
1355 struct sk_buff *skb;
1357 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1358 IP6_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1362 inet->cork.flags &= ~IPCORK_OPT;
1364 kfree(np->cork.opt);
1365 np->cork.opt = NULL;
1367 dst_release(&np->cork.rt->u.dst);
1369 inet->cork.flags &= ~IPCORK_ALLFRAG;
1371 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));