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_local_out(struct sk_buff *skb)
63 len = skb->len - sizeof(struct ipv6hdr);
64 if (len > IPV6_MAXPLEN)
66 ipv6_hdr(skb)->payload_len = htons(len);
68 return nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, skb, NULL,
69 skb_dst(skb)->dev, dst_output);
72 int ip6_local_out(struct sk_buff *skb)
76 err = __ip6_local_out(skb);
78 err = dst_output(skb);
82 EXPORT_SYMBOL_GPL(ip6_local_out);
84 static int ip6_finish_output2(struct sk_buff *skb)
86 struct dst_entry *dst = skb_dst(skb);
87 struct net_device *dev = dst->dev;
88 struct neighbour *neigh;
89 struct in6_addr *nexthop;
92 skb->protocol = htons(ETH_P_IPV6);
95 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
96 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
98 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(skb->sk) &&
99 ((mroute6_socket(dev_net(dev), skb) &&
100 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
101 ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
102 &ipv6_hdr(skb)->saddr))) {
103 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
105 /* Do not check for IFF_ALLMULTI; multicast routing
106 is not supported in any case.
109 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
110 newskb, NULL, newskb->dev,
113 if (ipv6_hdr(skb)->hop_limit == 0) {
114 IP6_INC_STATS(dev_net(dev), idev,
115 IPSTATS_MIB_OUTDISCARDS);
121 IP6_UPD_PO_STATS(dev_net(dev), idev, IPSTATS_MIB_OUTMCAST,
124 if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb)->daddr) <=
125 IPV6_ADDR_SCOPE_NODELOCAL &&
126 !(dev->flags & IFF_LOOPBACK)) {
133 nexthop = rt6_nexthop((struct rt6_info *)dst);
134 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
135 if (unlikely(!neigh))
136 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
137 if (!IS_ERR(neigh)) {
138 ret = dst_neigh_output(dst, neigh, skb);
139 rcu_read_unlock_bh();
142 rcu_read_unlock_bh();
144 IP6_INC_STATS(dev_net(dst->dev),
145 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
150 static int ip6_finish_output(struct sk_buff *skb)
152 if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
153 dst_allfrag(skb_dst(skb)) ||
154 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
155 return ip6_fragment(skb, ip6_finish_output2);
157 return ip6_finish_output2(skb);
160 int ip6_output(struct sk_buff *skb)
162 struct net_device *dev = skb_dst(skb)->dev;
163 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
164 if (unlikely(idev->cnf.disable_ipv6)) {
165 IP6_INC_STATS(dev_net(dev), idev,
166 IPSTATS_MIB_OUTDISCARDS);
171 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, skb, NULL, dev,
173 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
177 * xmit an sk_buff (used by TCP, SCTP and DCCP)
180 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
181 struct ipv6_txoptions *opt, int tclass)
183 struct net *net = sock_net(sk);
184 struct ipv6_pinfo *np = inet6_sk(sk);
185 struct in6_addr *first_hop = &fl6->daddr;
186 struct dst_entry *dst = skb_dst(skb);
188 u8 proto = fl6->flowi6_proto;
189 int seg_len = skb->len;
194 unsigned int head_room;
196 /* First: exthdrs may take lots of space (~8K for now)
197 MAX_HEADER is not enough.
199 head_room = opt->opt_nflen + opt->opt_flen;
200 seg_len += head_room;
201 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
203 if (skb_headroom(skb) < head_room) {
204 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
206 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
207 IPSTATS_MIB_OUTDISCARDS);
213 skb_set_owner_w(skb, sk);
216 ipv6_push_frag_opts(skb, opt, &proto);
218 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
221 skb_push(skb, sizeof(struct ipv6hdr));
222 skb_reset_network_header(skb);
226 * Fill in the IPv6 header
229 hlimit = np->hop_limit;
231 hlimit = ip6_dst_hoplimit(dst);
233 ip6_flow_hdr(hdr, tclass, fl6->flowlabel);
235 hdr->payload_len = htons(seg_len);
236 hdr->nexthdr = proto;
237 hdr->hop_limit = hlimit;
239 hdr->saddr = fl6->saddr;
240 hdr->daddr = *first_hop;
242 skb->priority = sk->sk_priority;
243 skb->mark = sk->sk_mark;
246 if ((skb->len <= mtu) || skb->local_df || skb_is_gso(skb)) {
247 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
248 IPSTATS_MIB_OUT, skb->len);
249 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, skb, NULL,
250 dst->dev, dst_output);
254 ipv6_local_error(sk, EMSGSIZE, fl6, mtu);
255 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
260 EXPORT_SYMBOL(ip6_xmit);
262 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
264 struct ip6_ra_chain *ra;
265 struct sock *last = NULL;
267 read_lock(&ip6_ra_lock);
268 for (ra = ip6_ra_chain; ra; ra = ra->next) {
269 struct sock *sk = ra->sk;
270 if (sk && ra->sel == sel &&
271 (!sk->sk_bound_dev_if ||
272 sk->sk_bound_dev_if == skb->dev->ifindex)) {
274 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
276 rawv6_rcv(last, skb2);
283 rawv6_rcv(last, skb);
284 read_unlock(&ip6_ra_lock);
287 read_unlock(&ip6_ra_lock);
291 static int ip6_forward_proxy_check(struct sk_buff *skb)
293 struct ipv6hdr *hdr = ipv6_hdr(skb);
294 u8 nexthdr = hdr->nexthdr;
298 if (ipv6_ext_hdr(nexthdr)) {
299 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
303 offset = sizeof(struct ipv6hdr);
305 if (nexthdr == IPPROTO_ICMPV6) {
306 struct icmp6hdr *icmp6;
308 if (!pskb_may_pull(skb, (skb_network_header(skb) +
309 offset + 1 - skb->data)))
312 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
314 switch (icmp6->icmp6_type) {
315 case NDISC_ROUTER_SOLICITATION:
316 case NDISC_ROUTER_ADVERTISEMENT:
317 case NDISC_NEIGHBOUR_SOLICITATION:
318 case NDISC_NEIGHBOUR_ADVERTISEMENT:
320 /* For reaction involving unicast neighbor discovery
321 * message destined to the proxied address, pass it to
331 * The proxying router can't forward traffic sent to a link-local
332 * address, so signal the sender and discard the packet. This
333 * behavior is clarified by the MIPv6 specification.
335 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
336 dst_link_failure(skb);
343 static inline int ip6_forward_finish(struct sk_buff *skb)
345 return dst_output(skb);
348 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
353 /* ipv6 conntrack defrag sets max_frag_size + local_df */
354 if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
360 if (skb_is_gso(skb) && skb_gso_network_seglen(skb) <= mtu)
366 int ip6_forward(struct sk_buff *skb)
368 struct dst_entry *dst = skb_dst(skb);
369 struct ipv6hdr *hdr = ipv6_hdr(skb);
370 struct inet6_skb_parm *opt = IP6CB(skb);
371 struct net *net = dev_net(dst->dev);
374 if (net->ipv6.devconf_all->forwarding == 0)
377 if (skb_warn_if_lro(skb))
380 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
381 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
385 if (skb->pkt_type != PACKET_HOST)
388 skb_forward_csum(skb);
391 * We DO NOT make any processing on
392 * RA packets, pushing them to user level AS IS
393 * without ane WARRANTY that application will be able
394 * to interpret them. The reason is that we
395 * cannot make anything clever here.
397 * We are not end-node, so that if packet contains
398 * AH/ESP, we cannot make anything.
399 * Defragmentation also would be mistake, RA packets
400 * cannot be fragmented, because there is no warranty
401 * that different fragments will go along one path. --ANK
403 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
404 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
409 * check and decrement ttl
411 if (hdr->hop_limit <= 1) {
412 /* Force OUTPUT device used as source address */
414 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
415 IP6_INC_STATS_BH(net,
416 ip6_dst_idev(dst), IPSTATS_MIB_INHDRERRORS);
422 /* XXX: idev->cnf.proxy_ndp? */
423 if (net->ipv6.devconf_all->proxy_ndp &&
424 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
425 int proxied = ip6_forward_proxy_check(skb);
427 return ip6_input(skb);
428 else if (proxied < 0) {
429 IP6_INC_STATS(net, ip6_dst_idev(dst),
430 IPSTATS_MIB_INDISCARDS);
435 if (!xfrm6_route_forward(skb)) {
436 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
441 /* IPv6 specs say nothing about it, but it is clear that we cannot
442 send redirects to source routed frames.
443 We don't send redirects to frames decapsulated from IPsec.
445 if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
446 struct in6_addr *target = NULL;
447 struct inet_peer *peer;
451 * incoming and outgoing devices are the same
455 rt = (struct rt6_info *) dst;
456 if (rt->rt6i_flags & RTF_GATEWAY)
457 target = &rt->rt6i_gateway;
459 target = &hdr->daddr;
461 peer = inet_getpeer_v6(net->ipv6.peers, &rt->rt6i_dst.addr, 1);
463 /* Limit redirects both by destination (here)
464 and by source (inside ndisc_send_redirect)
466 if (inet_peer_xrlim_allow(peer, 1*HZ))
467 ndisc_send_redirect(skb, target);
471 int addrtype = ipv6_addr_type(&hdr->saddr);
473 /* This check is security critical. */
474 if (addrtype == IPV6_ADDR_ANY ||
475 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
477 if (addrtype & IPV6_ADDR_LINKLOCAL) {
478 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
479 ICMPV6_NOT_NEIGHBOUR, 0);
485 if (mtu < IPV6_MIN_MTU)
488 if (ip6_pkt_too_big(skb, mtu)) {
489 /* Again, force OUTPUT device used as source address */
491 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
492 IP6_INC_STATS_BH(net,
493 ip6_dst_idev(dst), IPSTATS_MIB_INTOOBIGERRORS);
494 IP6_INC_STATS_BH(net,
495 ip6_dst_idev(dst), IPSTATS_MIB_FRAGFAILS);
500 if (skb_cow(skb, dst->dev->hard_header_len)) {
501 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTDISCARDS);
507 /* Mangling hops number delayed to point after skb COW */
511 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
512 IP6_ADD_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
513 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD, skb, skb->dev, dst->dev,
517 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
523 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
525 to->pkt_type = from->pkt_type;
526 to->priority = from->priority;
527 to->protocol = from->protocol;
529 skb_dst_set(to, dst_clone(skb_dst(from)));
531 to->mark = from->mark;
533 #ifdef CONFIG_NET_SCHED
534 to->tc_index = from->tc_index;
537 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
538 to->nf_trace = from->nf_trace;
540 skb_copy_secmark(to, from);
543 static void ipv6_select_ident(struct frag_hdr *fhdr, struct rt6_info *rt)
545 static u32 ip6_idents_hashrnd __read_mostly;
546 static bool hashrnd_initialized = false;
549 if (unlikely(!hashrnd_initialized)) {
550 hashrnd_initialized = true;
551 get_random_bytes(&ip6_idents_hashrnd, sizeof(ip6_idents_hashrnd));
553 hash = __ipv6_addr_jhash(&rt->rt6i_dst.addr, ip6_idents_hashrnd);
554 hash = __ipv6_addr_jhash(&rt->rt6i_src.addr, hash);
556 id = ip_idents_reserve(hash, 1);
557 fhdr->identification = htonl(id);
560 int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
562 struct sk_buff *frag;
563 struct rt6_info *rt = (struct rt6_info*)skb_dst(skb);
564 struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
565 struct ipv6hdr *tmp_hdr;
567 unsigned int mtu, hlen, left, len;
570 int ptr, offset = 0, err=0;
571 u8 *prevhdr, nexthdr = 0;
572 struct net *net = dev_net(skb_dst(skb)->dev);
574 hlen = ip6_find_1stfragopt(skb, &prevhdr);
577 mtu = ip6_skb_dst_mtu(skb);
579 /* We must not fragment if the socket is set to force MTU discovery
580 * or if the skb it not generated by a local socket.
582 if (unlikely(!skb->local_df && skb->len > mtu) ||
583 (IP6CB(skb)->frag_max_size &&
584 IP6CB(skb)->frag_max_size > mtu)) {
585 if (skb->sk && dst_allfrag(skb_dst(skb)))
586 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
588 skb->dev = skb_dst(skb)->dev;
589 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
590 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
591 IPSTATS_MIB_FRAGFAILS);
596 if (np && np->frag_size < mtu) {
600 mtu -= hlen + sizeof(struct frag_hdr);
602 if (skb_has_frag_list(skb)) {
603 int first_len = skb_pagelen(skb);
604 struct sk_buff *frag2;
606 if (first_len - hlen > mtu ||
607 ((first_len - hlen) & 7) ||
611 skb_walk_frags(skb, frag) {
612 /* Correct geometry. */
613 if (frag->len > mtu ||
614 ((frag->len & 7) && frag->next) ||
615 skb_headroom(frag) < hlen)
616 goto slow_path_clean;
618 /* Partially cloned skb? */
619 if (skb_shared(frag))
620 goto slow_path_clean;
625 frag->destructor = sock_wfree;
627 skb->truesize -= frag->truesize;
632 frag = skb_shinfo(skb)->frag_list;
633 skb_frag_list_init(skb);
636 *prevhdr = NEXTHDR_FRAGMENT;
637 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
639 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
640 IPSTATS_MIB_FRAGFAILS);
644 __skb_pull(skb, hlen);
645 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
646 __skb_push(skb, hlen);
647 skb_reset_network_header(skb);
648 memcpy(skb_network_header(skb), tmp_hdr, hlen);
650 ipv6_select_ident(fh, rt);
651 fh->nexthdr = nexthdr;
653 fh->frag_off = htons(IP6_MF);
654 frag_id = fh->identification;
656 first_len = skb_pagelen(skb);
657 skb->data_len = first_len - skb_headlen(skb);
658 skb->len = first_len;
659 ipv6_hdr(skb)->payload_len = htons(first_len -
660 sizeof(struct ipv6hdr));
665 /* Prepare header of the next frame,
666 * before previous one went down. */
668 frag->ip_summed = CHECKSUM_NONE;
669 skb_reset_transport_header(frag);
670 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
671 __skb_push(frag, hlen);
672 skb_reset_network_header(frag);
673 memcpy(skb_network_header(frag), tmp_hdr,
675 offset += skb->len - hlen - sizeof(struct frag_hdr);
676 fh->nexthdr = nexthdr;
678 fh->frag_off = htons(offset);
679 if (frag->next != NULL)
680 fh->frag_off |= htons(IP6_MF);
681 fh->identification = frag_id;
682 ipv6_hdr(frag)->payload_len =
684 sizeof(struct ipv6hdr));
685 ip6_copy_metadata(frag, skb);
690 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
691 IPSTATS_MIB_FRAGCREATES);
704 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
705 IPSTATS_MIB_FRAGOKS);
716 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
717 IPSTATS_MIB_FRAGFAILS);
722 skb_walk_frags(skb, frag2) {
726 frag2->destructor = NULL;
727 skb->truesize += frag2->truesize;
732 if ((skb->ip_summed == CHECKSUM_PARTIAL) &&
733 skb_checksum_help(skb))
736 left = skb->len - hlen; /* Space per frame */
737 ptr = hlen; /* Where to start from */
740 * Fragment the datagram.
743 *prevhdr = NEXTHDR_FRAGMENT;
744 hroom = LL_RESERVED_SPACE(rt->dst.dev);
745 troom = rt->dst.dev->needed_tailroom;
748 * Keep copying data until we run out.
752 /* IF: it doesn't fit, use 'mtu' - the data space left */
755 /* IF: we are not sending up to and including the packet end
756 then align the next start on an eight byte boundary */
764 if ((frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
765 hroom + troom, GFP_ATOMIC)) == NULL) {
766 NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
767 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
768 IPSTATS_MIB_FRAGFAILS);
774 * Set up data on packet
777 ip6_copy_metadata(frag, skb);
778 skb_reserve(frag, hroom);
779 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
780 skb_reset_network_header(frag);
781 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
782 frag->transport_header = (frag->network_header + hlen +
783 sizeof(struct frag_hdr));
786 * Charge the memory for the fragment to any owner
790 skb_set_owner_w(frag, skb->sk);
793 * Copy the packet header into the new buffer.
795 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
798 * Build fragment header.
800 fh->nexthdr = nexthdr;
803 ipv6_select_ident(fh, rt);
804 frag_id = fh->identification;
806 fh->identification = frag_id;
809 * Copy a block of the IP datagram.
811 if (skb_copy_bits(skb, ptr, skb_transport_header(frag), len))
815 fh->frag_off = htons(offset);
817 fh->frag_off |= htons(IP6_MF);
818 ipv6_hdr(frag)->payload_len = htons(frag->len -
819 sizeof(struct ipv6hdr));
825 * Put this fragment into the sending queue.
831 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
832 IPSTATS_MIB_FRAGCREATES);
834 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
835 IPSTATS_MIB_FRAGOKS);
840 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
841 IPSTATS_MIB_FRAGFAILS);
846 static inline int ip6_rt_check(const struct rt6key *rt_key,
847 const struct in6_addr *fl_addr,
848 const struct in6_addr *addr_cache)
850 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
851 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache));
854 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
855 struct dst_entry *dst,
856 const struct flowi6 *fl6)
858 struct ipv6_pinfo *np = inet6_sk(sk);
864 if (dst->ops->family != AF_INET6) {
869 rt = (struct rt6_info *)dst;
870 /* Yes, checking route validity in not connected
871 * case is not very simple. Take into account,
872 * that we do not support routing by source, TOS,
873 * and MSG_DONTROUTE --ANK (980726)
875 * 1. ip6_rt_check(): If route was host route,
876 * check that cached destination is current.
877 * If it is network route, we still may
878 * check its validity using saved pointer
879 * to the last used address: daddr_cache.
880 * We do not want to save whole address now,
881 * (because main consumer of this service
882 * is tcp, which has not this problem),
883 * so that the last trick works only on connected
885 * 2. oif also should be the same.
887 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
888 #ifdef CONFIG_IPV6_SUBTREES
889 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
891 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
900 static int ip6_dst_lookup_tail(struct sock *sk,
901 struct dst_entry **dst, struct flowi6 *fl6)
903 struct net *net = sock_net(sk);
904 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
911 *dst = ip6_route_output(net, sk, fl6);
913 if ((err = (*dst)->error))
914 goto out_err_release;
916 if (ipv6_addr_any(&fl6->saddr)) {
917 struct rt6_info *rt = (struct rt6_info *) *dst;
918 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
919 sk ? inet6_sk(sk)->srcprefs : 0,
922 goto out_err_release;
925 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
927 * Here if the dst entry we've looked up
928 * has a neighbour entry that is in the INCOMPLETE
929 * state and the src address from the flow is
930 * marked as OPTIMISTIC, we release the found
931 * dst entry and replace it instead with the
932 * dst entry of the nexthop router
934 rt = (struct rt6_info *) *dst;
936 n = __ipv6_neigh_lookup_noref(rt->dst.dev, rt6_nexthop(rt));
937 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
938 rcu_read_unlock_bh();
941 struct inet6_ifaddr *ifp;
942 struct flowi6 fl_gw6;
945 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
948 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
954 * We need to get the dst entry for the
955 * default router instead
958 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
959 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
960 *dst = ip6_route_output(net, sk, &fl_gw6);
961 if ((err = (*dst)->error))
962 goto out_err_release;
970 if (err == -ENETUNREACH)
971 IP6_INC_STATS_BH(net, NULL, IPSTATS_MIB_OUTNOROUTES);
978 * ip6_dst_lookup - perform route lookup on flow
979 * @sk: socket which provides route info
980 * @dst: pointer to dst_entry * for result
981 * @fl6: flow to lookup
983 * This function performs a route lookup on the given flow.
985 * It returns zero on success, or a standard errno code on error.
987 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi6 *fl6)
990 return ip6_dst_lookup_tail(sk, dst, fl6);
992 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
995 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
996 * @sk: socket which provides route info
997 * @fl6: flow to lookup
998 * @final_dst: final destination address for ipsec lookup
999 * @can_sleep: we are in a sleepable context
1001 * This function performs a route lookup on the given flow.
1003 * It returns a valid dst pointer on success, or a pointer encoded
1006 struct dst_entry *ip6_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1007 const struct in6_addr *final_dst,
1010 struct dst_entry *dst = NULL;
1013 err = ip6_dst_lookup_tail(sk, &dst, fl6);
1015 return ERR_PTR(err);
1017 fl6->daddr = *final_dst;
1019 fl6->flowi6_flags |= FLOWI_FLAG_CAN_SLEEP;
1021 return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1023 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1026 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1027 * @sk: socket which provides the dst cache and route info
1028 * @fl6: flow to lookup
1029 * @final_dst: final destination address for ipsec lookup
1030 * @can_sleep: we are in a sleepable context
1032 * This function performs a route lookup on the given flow with the
1033 * possibility of using the cached route in the socket if it is valid.
1034 * It will take the socket dst lock when operating on the dst cache.
1035 * As a result, this function can only be used in process context.
1037 * It returns a valid dst pointer on success, or a pointer encoded
1040 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1041 const struct in6_addr *final_dst,
1044 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1047 dst = ip6_sk_dst_check(sk, dst, fl6);
1049 err = ip6_dst_lookup_tail(sk, &dst, fl6);
1051 return ERR_PTR(err);
1053 fl6->daddr = *final_dst;
1055 fl6->flowi6_flags |= FLOWI_FLAG_CAN_SLEEP;
1057 return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1059 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1061 static inline int ip6_ufo_append_data(struct sock *sk,
1062 int getfrag(void *from, char *to, int offset, int len,
1063 int odd, struct sk_buff *skb),
1064 void *from, int length, int hh_len, int fragheaderlen,
1065 int transhdrlen, int mtu,unsigned int flags,
1066 struct rt6_info *rt)
1069 struct sk_buff *skb;
1072 /* There is support for UDP large send offload by network
1073 * device, so create one single skb packet containing complete
1076 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
1077 struct frag_hdr fhdr;
1079 skb = sock_alloc_send_skb(sk,
1080 hh_len + fragheaderlen + transhdrlen + 20,
1081 (flags & MSG_DONTWAIT), &err);
1085 /* reserve space for Hardware header */
1086 skb_reserve(skb, hh_len);
1088 /* create space for UDP/IP header */
1089 skb_put(skb,fragheaderlen + transhdrlen);
1091 /* initialize network header pointer */
1092 skb_reset_network_header(skb);
1094 /* initialize protocol header pointer */
1095 skb->transport_header = skb->network_header + fragheaderlen;
1097 skb->ip_summed = CHECKSUM_PARTIAL;
1100 /* Specify the length of each IPv6 datagram fragment.
1101 * It has to be a multiple of 8.
1103 skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1104 sizeof(struct frag_hdr)) & ~7;
1105 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1106 ipv6_select_ident(&fhdr, rt);
1107 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1108 __skb_queue_tail(&sk->sk_write_queue, skb);
1111 return skb_append_datato_frags(sk, skb, getfrag, from,
1112 (length - transhdrlen));
1115 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1118 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1121 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1124 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1127 static void ip6_append_data_mtu(unsigned int *mtu,
1129 unsigned int fragheaderlen,
1130 struct sk_buff *skb,
1131 struct rt6_info *rt,
1132 unsigned int orig_mtu)
1134 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1136 /* first fragment, reserve header_len */
1137 *mtu = orig_mtu - rt->dst.header_len;
1141 * this fragment is not first, the headers
1142 * space is regarded as data space.
1146 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1147 + fragheaderlen - sizeof(struct frag_hdr);
1151 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1152 int offset, int len, int odd, struct sk_buff *skb),
1153 void *from, int length, int transhdrlen,
1154 int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1155 struct rt6_info *rt, unsigned int flags, int dontfrag)
1157 struct inet_sock *inet = inet_sk(sk);
1158 struct ipv6_pinfo *np = inet6_sk(sk);
1159 struct inet_cork *cork;
1160 struct sk_buff *skb, *skb_prev = NULL;
1161 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu;
1170 if (flags&MSG_PROBE)
1172 cork = &inet->cork.base;
1173 if (skb_queue_empty(&sk->sk_write_queue)) {
1178 if (WARN_ON(np->cork.opt))
1181 np->cork.opt = kzalloc(opt->tot_len, sk->sk_allocation);
1182 if (unlikely(np->cork.opt == NULL))
1185 np->cork.opt->tot_len = opt->tot_len;
1186 np->cork.opt->opt_flen = opt->opt_flen;
1187 np->cork.opt->opt_nflen = opt->opt_nflen;
1189 np->cork.opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1191 if (opt->dst0opt && !np->cork.opt->dst0opt)
1194 np->cork.opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1196 if (opt->dst1opt && !np->cork.opt->dst1opt)
1199 np->cork.opt->hopopt = ip6_opt_dup(opt->hopopt,
1201 if (opt->hopopt && !np->cork.opt->hopopt)
1204 np->cork.opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1206 if (opt->srcrt && !np->cork.opt->srcrt)
1209 /* need source address above miyazawa*/
1212 cork->dst = &rt->dst;
1213 inet->cork.fl.u.ip6 = *fl6;
1214 np->cork.hop_limit = hlimit;
1215 np->cork.tclass = tclass;
1216 if (rt->dst.flags & DST_XFRM_TUNNEL)
1217 mtu = np->pmtudisc == IPV6_PMTUDISC_PROBE ?
1218 rt->dst.dev->mtu : dst_mtu(&rt->dst);
1220 mtu = np->pmtudisc == IPV6_PMTUDISC_PROBE ?
1221 rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1222 if (np->frag_size < mtu) {
1224 mtu = np->frag_size;
1226 cork->fragsize = mtu;
1227 if (dst_allfrag(rt->dst.path))
1228 cork->flags |= IPCORK_ALLFRAG;
1230 exthdrlen = (opt ? opt->opt_flen : 0);
1231 length += exthdrlen;
1232 transhdrlen += exthdrlen;
1233 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1235 rt = (struct rt6_info *)cork->dst;
1236 fl6 = &inet->cork.fl.u.ip6;
1241 mtu = cork->fragsize;
1245 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1247 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1248 (opt ? opt->opt_nflen : 0);
1249 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
1251 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1252 if (cork->length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
1253 ipv6_local_error(sk, EMSGSIZE, fl6, mtu-exthdrlen);
1258 /* For UDP, check if TX timestamp is enabled */
1259 if (sk->sk_type == SOCK_DGRAM)
1260 sock_tx_timestamp(sk, &tx_flags);
1263 * Let's try using as much space as possible.
1264 * Use MTU if total length of the message fits into the MTU.
1265 * Otherwise, we need to reserve fragment header and
1266 * fragment alignment (= 8-15 octects, in total).
1268 * Note that we may need to "move" the data from the tail of
1269 * of the buffer to the new fragment when we split
1272 * FIXME: It may be fragmented into multiple chunks
1273 * at once if non-fragmentable extension headers
1278 if ((length > mtu) && dontfrag && (sk->sk_protocol == IPPROTO_UDP ||
1279 sk->sk_protocol == IPPROTO_RAW)) {
1280 ipv6_local_rxpmtu(sk, fl6, mtu-exthdrlen);
1284 skb = skb_peek_tail(&sk->sk_write_queue);
1285 cork->length += length;
1286 if (((length > mtu) ||
1287 (skb && skb_has_frags(skb))) &&
1288 (sk->sk_protocol == IPPROTO_UDP) &&
1289 (rt->dst.dev->features & NETIF_F_UFO) &&
1290 (sk->sk_type == SOCK_DGRAM)) {
1291 err = ip6_ufo_append_data(sk, getfrag, from, length,
1292 hh_len, fragheaderlen,
1293 transhdrlen, mtu, flags, rt);
1302 while (length > 0) {
1303 /* Check if the remaining data fits into current packet. */
1304 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1306 copy = maxfraglen - skb->len;
1310 unsigned int datalen;
1311 unsigned int fraglen;
1312 unsigned int fraggap;
1313 unsigned int alloclen;
1315 /* There's no room in the current skb */
1317 fraggap = skb->len - maxfraglen;
1320 /* update mtu and maxfraglen if necessary */
1321 if (skb == NULL || skb_prev == NULL)
1322 ip6_append_data_mtu(&mtu, &maxfraglen,
1323 fragheaderlen, skb, rt,
1329 * If remaining data exceeds the mtu,
1330 * we know we need more fragment(s).
1332 datalen = length + fraggap;
1334 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1335 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1336 if ((flags & MSG_MORE) &&
1337 !(rt->dst.dev->features&NETIF_F_SG))
1340 alloclen = datalen + fragheaderlen;
1342 alloclen += dst_exthdrlen;
1344 if (datalen != length + fraggap) {
1346 * this is not the last fragment, the trailer
1347 * space is regarded as data space.
1349 datalen += rt->dst.trailer_len;
1352 alloclen += rt->dst.trailer_len;
1353 fraglen = datalen + fragheaderlen;
1356 * We just reserve space for fragment header.
1357 * Note: this may be overallocation if the message
1358 * (without MSG_MORE) fits into the MTU.
1360 alloclen += sizeof(struct frag_hdr);
1363 skb = sock_alloc_send_skb(sk,
1365 (flags & MSG_DONTWAIT), &err);
1368 if (atomic_read(&sk->sk_wmem_alloc) <=
1370 skb = sock_wmalloc(sk,
1371 alloclen + hh_len, 1,
1373 if (unlikely(skb == NULL))
1376 /* Only the initial fragment
1385 * Fill in the control structures
1387 skb->ip_summed = CHECKSUM_NONE;
1389 /* reserve for fragmentation and ipsec header */
1390 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1393 if (sk->sk_type == SOCK_DGRAM)
1394 skb_shinfo(skb)->tx_flags = tx_flags;
1397 * Find where to start putting bytes
1399 data = skb_put(skb, fraglen);
1400 skb_set_network_header(skb, exthdrlen);
1401 data += fragheaderlen;
1402 skb->transport_header = (skb->network_header +
1405 skb->csum = skb_copy_and_csum_bits(
1406 skb_prev, maxfraglen,
1407 data + transhdrlen, fraggap, 0);
1408 skb_prev->csum = csum_sub(skb_prev->csum,
1411 pskb_trim_unique(skb_prev, maxfraglen);
1413 copy = datalen - transhdrlen - fraggap;
1419 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1426 length -= datalen - fraggap;
1432 * Put the packet on the pending queue
1434 __skb_queue_tail(&sk->sk_write_queue, skb);
1441 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1445 if (getfrag(from, skb_put(skb, copy),
1446 offset, copy, off, skb) < 0) {
1447 __skb_trim(skb, off);
1452 int i = skb_shinfo(skb)->nr_frags;
1453 struct page_frag *pfrag = sk_page_frag(sk);
1456 if (!sk_page_frag_refill(sk, pfrag))
1459 if (!skb_can_coalesce(skb, i, pfrag->page,
1462 if (i == MAX_SKB_FRAGS)
1465 __skb_fill_page_desc(skb, i, pfrag->page,
1467 skb_shinfo(skb)->nr_frags = ++i;
1468 get_page(pfrag->page);
1470 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1472 page_address(pfrag->page) + pfrag->offset,
1473 offset, copy, skb->len, skb) < 0)
1476 pfrag->offset += copy;
1477 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1479 skb->data_len += copy;
1480 skb->truesize += copy;
1481 atomic_add(copy, &sk->sk_wmem_alloc);
1492 cork->length -= length;
1493 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1496 EXPORT_SYMBOL_GPL(ip6_append_data);
1498 static void ip6_cork_release(struct inet_sock *inet, struct ipv6_pinfo *np)
1501 kfree(np->cork.opt->dst0opt);
1502 kfree(np->cork.opt->dst1opt);
1503 kfree(np->cork.opt->hopopt);
1504 kfree(np->cork.opt->srcrt);
1505 kfree(np->cork.opt);
1506 np->cork.opt = NULL;
1509 if (inet->cork.base.dst) {
1510 dst_release(inet->cork.base.dst);
1511 inet->cork.base.dst = NULL;
1512 inet->cork.base.flags &= ~IPCORK_ALLFRAG;
1514 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1517 int ip6_push_pending_frames(struct sock *sk)
1519 struct sk_buff *skb, *tmp_skb;
1520 struct sk_buff **tail_skb;
1521 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1522 struct inet_sock *inet = inet_sk(sk);
1523 struct ipv6_pinfo *np = inet6_sk(sk);
1524 struct net *net = sock_net(sk);
1525 struct ipv6hdr *hdr;
1526 struct ipv6_txoptions *opt = np->cork.opt;
1527 struct rt6_info *rt = (struct rt6_info *)inet->cork.base.dst;
1528 struct flowi6 *fl6 = &inet->cork.fl.u.ip6;
1529 unsigned char proto = fl6->flowi6_proto;
1532 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1534 tail_skb = &(skb_shinfo(skb)->frag_list);
1536 /* move skb->data to ip header from ext header */
1537 if (skb->data < skb_network_header(skb))
1538 __skb_pull(skb, skb_network_offset(skb));
1539 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1540 __skb_pull(tmp_skb, skb_network_header_len(skb));
1541 *tail_skb = tmp_skb;
1542 tail_skb = &(tmp_skb->next);
1543 skb->len += tmp_skb->len;
1544 skb->data_len += tmp_skb->len;
1545 skb->truesize += tmp_skb->truesize;
1546 tmp_skb->destructor = NULL;
1550 /* Allow local fragmentation. */
1551 if (np->pmtudisc < IPV6_PMTUDISC_DO)
1554 *final_dst = fl6->daddr;
1555 __skb_pull(skb, skb_network_header_len(skb));
1556 if (opt && opt->opt_flen)
1557 ipv6_push_frag_opts(skb, opt, &proto);
1558 if (opt && opt->opt_nflen)
1559 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1561 skb_push(skb, sizeof(struct ipv6hdr));
1562 skb_reset_network_header(skb);
1563 hdr = ipv6_hdr(skb);
1565 ip6_flow_hdr(hdr, np->cork.tclass, fl6->flowlabel);
1566 hdr->hop_limit = np->cork.hop_limit;
1567 hdr->nexthdr = proto;
1568 hdr->saddr = fl6->saddr;
1569 hdr->daddr = *final_dst;
1571 skb->priority = sk->sk_priority;
1572 skb->mark = sk->sk_mark;
1574 skb_dst_set(skb, dst_clone(&rt->dst));
1575 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1576 if (proto == IPPROTO_ICMPV6) {
1577 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1579 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1580 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1583 err = ip6_local_out(skb);
1586 err = net_xmit_errno(err);
1592 ip6_cork_release(inet, np);
1595 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1598 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1600 void ip6_flush_pending_frames(struct sock *sk)
1602 struct sk_buff *skb;
1604 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1606 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1607 IPSTATS_MIB_OUTDISCARDS);
1611 ip6_cork_release(inet_sk(sk), inet6_sk(sk));
1613 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);