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_reset_network_header(skb
);
267 skb_put(skb
, sizeof(struct ipv6hdr
));
270 *(__be32
*)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
= ipv6_hdr(skb
);
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_network_header(skb
) +
328 offset
+ 1 - skb
->data
)))
331 icmp6
= (struct icmp6hdr
*)(skb_network_header(skb
) + offset
);
333 switch (icmp6
->icmp6_type
) {
334 case NDISC_ROUTER_SOLICITATION
:
335 case NDISC_ROUTER_ADVERTISEMENT
:
336 case NDISC_NEIGHBOUR_SOLICITATION
:
337 case NDISC_NEIGHBOUR_ADVERTISEMENT
:
339 /* For reaction involving unicast neighbor discovery
340 * message destined to the proxied address, pass it to
350 * The proxying router can't forward traffic sent to a link-local
351 * address, so signal the sender and discard the packet. This
352 * behavior is clarified by the MIPv6 specification.
354 if (ipv6_addr_type(&hdr
->daddr
) & IPV6_ADDR_LINKLOCAL
) {
355 dst_link_failure(skb
);
362 static inline int ip6_forward_finish(struct sk_buff
*skb
)
364 return dst_output(skb
);
367 int ip6_forward(struct sk_buff
*skb
)
369 struct dst_entry
*dst
= skb
->dst
;
370 struct ipv6hdr
*hdr
= ipv6_hdr(skb
);
371 struct inet6_skb_parm
*opt
= IP6CB(skb
);
373 if (ipv6_devconf
.forwarding
== 0)
376 if (!xfrm6_policy_check(NULL
, XFRM_POLICY_FWD
, skb
)) {
377 IP6_INC_STATS(ip6_dst_idev(dst
), IPSTATS_MIB_INDISCARDS
);
381 skb_forward_csum(skb
);
384 * We DO NOT make any processing on
385 * RA packets, pushing them to user level AS IS
386 * without ane WARRANTY that application will be able
387 * to interpret them. The reason is that we
388 * cannot make anything clever here.
390 * We are not end-node, so that if packet contains
391 * AH/ESP, we cannot make anything.
392 * Defragmentation also would be mistake, RA packets
393 * cannot be fragmented, because there is no warranty
394 * that different fragments will go along one path. --ANK
397 u8
*ptr
= skb_network_header(skb
) + opt
->ra
;
398 if (ip6_call_ra_chain(skb
, (ptr
[2]<<8) + ptr
[3]))
403 * check and decrement ttl
405 if (hdr
->hop_limit
<= 1) {
406 /* Force OUTPUT device used as source address */
408 icmpv6_send(skb
, ICMPV6_TIME_EXCEED
, ICMPV6_EXC_HOPLIMIT
,
410 IP6_INC_STATS_BH(ip6_dst_idev(dst
), IPSTATS_MIB_INHDRERRORS
);
416 /* XXX: idev->cnf.proxy_ndp? */
417 if (ipv6_devconf
.proxy_ndp
&&
418 pneigh_lookup(&nd_tbl
, &hdr
->daddr
, skb
->dev
, 0)) {
419 int proxied
= ip6_forward_proxy_check(skb
);
421 return ip6_input(skb
);
422 else if (proxied
< 0) {
423 IP6_INC_STATS(ip6_dst_idev(dst
), IPSTATS_MIB_INDISCARDS
);
428 if (!xfrm6_route_forward(skb
)) {
429 IP6_INC_STATS(ip6_dst_idev(dst
), IPSTATS_MIB_INDISCARDS
);
434 /* IPv6 specs say nothing about it, but it is clear that we cannot
435 send redirects to source routed frames.
437 if (skb
->dev
== dst
->dev
&& dst
->neighbour
&& opt
->srcrt
== 0) {
438 struct in6_addr
*target
= NULL
;
440 struct neighbour
*n
= dst
->neighbour
;
443 * incoming and outgoing devices are the same
447 rt
= (struct rt6_info
*) dst
;
448 if ((rt
->rt6i_flags
& RTF_GATEWAY
))
449 target
= (struct in6_addr
*)&n
->primary_key
;
451 target
= &hdr
->daddr
;
453 /* Limit redirects both by destination (here)
454 and by source (inside ndisc_send_redirect)
456 if (xrlim_allow(dst
, 1*HZ
))
457 ndisc_send_redirect(skb
, n
, target
);
458 } else if (ipv6_addr_type(&hdr
->saddr
)&(IPV6_ADDR_MULTICAST
|IPV6_ADDR_LOOPBACK
459 |IPV6_ADDR_LINKLOCAL
)) {
460 /* This check is security critical. */
464 if (skb
->len
> dst_mtu(dst
)) {
465 /* Again, force OUTPUT device used as source address */
467 icmpv6_send(skb
, ICMPV6_PKT_TOOBIG
, 0, dst_mtu(dst
), skb
->dev
);
468 IP6_INC_STATS_BH(ip6_dst_idev(dst
), IPSTATS_MIB_INTOOBIGERRORS
);
469 IP6_INC_STATS_BH(ip6_dst_idev(dst
), IPSTATS_MIB_FRAGFAILS
);
474 if (skb_cow(skb
, dst
->dev
->hard_header_len
)) {
475 IP6_INC_STATS(ip6_dst_idev(dst
), IPSTATS_MIB_OUTDISCARDS
);
481 /* Mangling hops number delayed to point after skb COW */
485 IP6_INC_STATS_BH(ip6_dst_idev(dst
), IPSTATS_MIB_OUTFORWDATAGRAMS
);
486 return NF_HOOK(PF_INET6
,NF_IP6_FORWARD
, skb
, skb
->dev
, dst
->dev
, ip6_forward_finish
);
489 IP6_INC_STATS_BH(ip6_dst_idev(dst
), IPSTATS_MIB_INADDRERRORS
);
495 static void ip6_copy_metadata(struct sk_buff
*to
, struct sk_buff
*from
)
497 to
->pkt_type
= from
->pkt_type
;
498 to
->priority
= from
->priority
;
499 to
->protocol
= from
->protocol
;
500 dst_release(to
->dst
);
501 to
->dst
= dst_clone(from
->dst
);
503 to
->mark
= from
->mark
;
505 #ifdef CONFIG_NET_SCHED
506 to
->tc_index
= from
->tc_index
;
509 skb_copy_secmark(to
, from
);
512 int ip6_find_1stfragopt(struct sk_buff
*skb
, u8
**nexthdr
)
514 u16 offset
= sizeof(struct ipv6hdr
);
515 struct ipv6_opt_hdr
*exthdr
=
516 (struct ipv6_opt_hdr
*)(ipv6_hdr(skb
) + 1);
517 unsigned int packet_len
= skb
->tail
- skb
->network_header
;
519 *nexthdr
= &ipv6_hdr(skb
)->nexthdr
;
521 while (offset
+ 1 <= packet_len
) {
527 case NEXTHDR_ROUTING
:
531 #ifdef CONFIG_IPV6_MIP6
532 if (ipv6_find_tlv(skb
, offset
, IPV6_TLV_HAO
) >= 0)
542 offset
+= ipv6_optlen(exthdr
);
543 *nexthdr
= &exthdr
->nexthdr
;
544 exthdr
= (struct ipv6_opt_hdr
*)(skb_network_header(skb
) +
550 EXPORT_SYMBOL_GPL(ip6_find_1stfragopt
);
552 static int ip6_fragment(struct sk_buff
*skb
, int (*output
)(struct sk_buff
*))
554 struct net_device
*dev
;
555 struct sk_buff
*frag
;
556 struct rt6_info
*rt
= (struct rt6_info
*)skb
->dst
;
557 struct ipv6_pinfo
*np
= skb
->sk
? inet6_sk(skb
->sk
) : NULL
;
558 struct ipv6hdr
*tmp_hdr
;
560 unsigned int mtu
, hlen
, left
, len
;
562 int ptr
, offset
= 0, err
=0;
563 u8
*prevhdr
, nexthdr
= 0;
566 hlen
= ip6_find_1stfragopt(skb
, &prevhdr
);
569 mtu
= dst_mtu(&rt
->u
.dst
);
571 /* We must not fragment if the socket is set to force MTU discovery
572 * or if the skb it not generated by a local socket. (This last
573 * check should be redundant, but it's free.)
575 if (!np
|| np
->pmtudisc
>= IPV6_PMTUDISC_DO
) {
576 skb
->dev
= skb
->dst
->dev
;
577 icmpv6_send(skb
, ICMPV6_PKT_TOOBIG
, 0, mtu
, skb
->dev
);
578 IP6_INC_STATS(ip6_dst_idev(skb
->dst
), IPSTATS_MIB_FRAGFAILS
);
583 if (np
&& np
->frag_size
< mtu
) {
587 mtu
-= hlen
+ sizeof(struct frag_hdr
);
589 if (skb_shinfo(skb
)->frag_list
) {
590 int first_len
= skb_pagelen(skb
);
592 if (first_len
- hlen
> mtu
||
593 ((first_len
- hlen
) & 7) ||
597 for (frag
= skb_shinfo(skb
)->frag_list
; frag
; frag
= frag
->next
) {
598 /* Correct geometry. */
599 if (frag
->len
> mtu
||
600 ((frag
->len
& 7) && frag
->next
) ||
601 skb_headroom(frag
) < hlen
)
604 /* Partially cloned skb? */
605 if (skb_shared(frag
))
612 frag
->destructor
= sock_wfree
;
613 skb
->truesize
-= frag
->truesize
;
619 frag
= skb_shinfo(skb
)->frag_list
;
620 skb_shinfo(skb
)->frag_list
= NULL
;
623 *prevhdr
= NEXTHDR_FRAGMENT
;
624 tmp_hdr
= kmemdup(skb_network_header(skb
), hlen
, GFP_ATOMIC
);
626 IP6_INC_STATS(ip6_dst_idev(skb
->dst
), IPSTATS_MIB_FRAGFAILS
);
630 __skb_pull(skb
, hlen
);
631 fh
= (struct frag_hdr
*)__skb_push(skb
, sizeof(struct frag_hdr
));
632 __skb_push(skb
, hlen
);
633 skb_reset_network_header(skb
);
634 memcpy(skb_network_header(skb
), tmp_hdr
, hlen
);
636 ipv6_select_ident(skb
, fh
);
637 fh
->nexthdr
= nexthdr
;
639 fh
->frag_off
= htons(IP6_MF
);
640 frag_id
= fh
->identification
;
642 first_len
= skb_pagelen(skb
);
643 skb
->data_len
= first_len
- skb_headlen(skb
);
644 skb
->len
= first_len
;
645 ipv6_hdr(skb
)->payload_len
= htons(first_len
-
646 sizeof(struct ipv6hdr
));
648 dst_hold(&rt
->u
.dst
);
651 /* Prepare header of the next frame,
652 * before previous one went down. */
654 frag
->ip_summed
= CHECKSUM_NONE
;
655 skb_reset_transport_header(frag
);
656 fh
= (struct frag_hdr
*)__skb_push(frag
, sizeof(struct frag_hdr
));
657 __skb_push(frag
, hlen
);
658 skb_reset_network_header(frag
);
659 memcpy(skb_network_header(frag
), tmp_hdr
,
661 offset
+= skb
->len
- hlen
- sizeof(struct frag_hdr
);
662 fh
->nexthdr
= nexthdr
;
664 fh
->frag_off
= htons(offset
);
665 if (frag
->next
!= NULL
)
666 fh
->frag_off
|= htons(IP6_MF
);
667 fh
->identification
= frag_id
;
668 ipv6_hdr(frag
)->payload_len
=
670 sizeof(struct ipv6hdr
));
671 ip6_copy_metadata(frag
, skb
);
676 IP6_INC_STATS(ip6_dst_idev(&rt
->u
.dst
), IPSTATS_MIB_FRAGCREATES
);
689 IP6_INC_STATS(ip6_dst_idev(&rt
->u
.dst
), IPSTATS_MIB_FRAGOKS
);
690 dst_release(&rt
->u
.dst
);
700 IP6_INC_STATS(ip6_dst_idev(&rt
->u
.dst
), IPSTATS_MIB_FRAGFAILS
);
701 dst_release(&rt
->u
.dst
);
706 left
= skb
->len
- hlen
; /* Space per frame */
707 ptr
= hlen
; /* Where to start from */
710 * Fragment the datagram.
713 *prevhdr
= NEXTHDR_FRAGMENT
;
716 * Keep copying data until we run out.
720 /* IF: it doesn't fit, use 'mtu' - the data space left */
723 /* IF: we are not sending upto and including the packet end
724 then align the next start on an eight byte boundary */
732 if ((frag
= alloc_skb(len
+hlen
+sizeof(struct frag_hdr
)+LL_RESERVED_SPACE(rt
->u
.dst
.dev
), GFP_ATOMIC
)) == NULL
) {
733 NETDEBUG(KERN_INFO
"IPv6: frag: no memory for new fragment!\n");
734 IP6_INC_STATS(ip6_dst_idev(skb
->dst
),
735 IPSTATS_MIB_FRAGFAILS
);
741 * Set up data on packet
744 ip6_copy_metadata(frag
, skb
);
745 skb_reserve(frag
, LL_RESERVED_SPACE(rt
->u
.dst
.dev
));
746 skb_put(frag
, len
+ hlen
+ sizeof(struct frag_hdr
));
747 skb_reset_network_header(frag
);
748 fh
= (struct frag_hdr
*)(skb_network_header(frag
) + hlen
);
749 frag
->transport_header
= (frag
->network_header
+ hlen
+
750 sizeof(struct frag_hdr
));
753 * Charge the memory for the fragment to any owner
757 skb_set_owner_w(frag
, skb
->sk
);
760 * Copy the packet header into the new buffer.
762 skb_copy_from_linear_data(skb
, skb_network_header(frag
), hlen
);
765 * Build fragment header.
767 fh
->nexthdr
= nexthdr
;
770 ipv6_select_ident(skb
, fh
);
771 frag_id
= fh
->identification
;
773 fh
->identification
= frag_id
;
776 * Copy a block of the IP datagram.
778 if (skb_copy_bits(skb
, ptr
, skb_transport_header(skb
), len
))
782 fh
->frag_off
= htons(offset
);
784 fh
->frag_off
|= htons(IP6_MF
);
785 ipv6_hdr(frag
)->payload_len
= htons(frag
->len
-
786 sizeof(struct ipv6hdr
));
792 * Put this fragment into the sending queue.
798 IP6_INC_STATS(ip6_dst_idev(skb
->dst
), IPSTATS_MIB_FRAGCREATES
);
800 IP6_INC_STATS(ip6_dst_idev(skb
->dst
),
801 IPSTATS_MIB_FRAGOKS
);
806 IP6_INC_STATS(ip6_dst_idev(skb
->dst
),
807 IPSTATS_MIB_FRAGFAILS
);
812 static inline int ip6_rt_check(struct rt6key
*rt_key
,
813 struct in6_addr
*fl_addr
,
814 struct in6_addr
*addr_cache
)
816 return ((rt_key
->plen
!= 128 || !ipv6_addr_equal(fl_addr
, &rt_key
->addr
)) &&
817 (addr_cache
== NULL
|| !ipv6_addr_equal(fl_addr
, addr_cache
)));
820 static struct dst_entry
*ip6_sk_dst_check(struct sock
*sk
,
821 struct dst_entry
*dst
,
824 struct ipv6_pinfo
*np
= inet6_sk(sk
);
825 struct rt6_info
*rt
= (struct rt6_info
*)dst
;
830 /* Yes, checking route validity in not connected
831 * case is not very simple. Take into account,
832 * that we do not support routing by source, TOS,
833 * and MSG_DONTROUTE --ANK (980726)
835 * 1. ip6_rt_check(): If route was host route,
836 * check that cached destination is current.
837 * If it is network route, we still may
838 * check its validity using saved pointer
839 * to the last used address: daddr_cache.
840 * We do not want to save whole address now,
841 * (because main consumer of this service
842 * is tcp, which has not this problem),
843 * so that the last trick works only on connected
845 * 2. oif also should be the same.
847 if (ip6_rt_check(&rt
->rt6i_dst
, &fl
->fl6_dst
, np
->daddr_cache
) ||
848 #ifdef CONFIG_IPV6_SUBTREES
849 ip6_rt_check(&rt
->rt6i_src
, &fl
->fl6_src
, np
->saddr_cache
) ||
851 (fl
->oif
&& fl
->oif
!= dst
->dev
->ifindex
)) {
860 static int ip6_dst_lookup_tail(struct sock
*sk
,
861 struct dst_entry
**dst
, struct flowi
*fl
)
866 *dst
= ip6_route_output(sk
, fl
);
868 if ((err
= (*dst
)->error
))
869 goto out_err_release
;
871 if (ipv6_addr_any(&fl
->fl6_src
)) {
872 err
= ipv6_get_saddr(*dst
, &fl
->fl6_dst
, &fl
->fl6_src
);
874 goto out_err_release
;
877 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
879 * Here if the dst entry we've looked up
880 * has a neighbour entry that is in the INCOMPLETE
881 * state and the src address from the flow is
882 * marked as OPTIMISTIC, we release the found
883 * dst entry and replace it instead with the
884 * dst entry of the nexthop router
886 if (!((*dst
)->neighbour
->nud_state
& NUD_VALID
)) {
887 struct inet6_ifaddr
*ifp
;
891 ifp
= ipv6_get_ifaddr(&fl
->fl6_src
, (*dst
)->dev
, 1);
893 redirect
= (ifp
&& ifp
->flags
& IFA_F_OPTIMISTIC
);
899 * We need to get the dst entry for the
900 * default router instead
903 memcpy(&fl_gw
, fl
, sizeof(struct flowi
));
904 memset(&fl_gw
.fl6_dst
, 0, sizeof(struct in6_addr
));
905 *dst
= ip6_route_output(sk
, &fl_gw
);
906 if ((err
= (*dst
)->error
))
907 goto out_err_release
;
921 * ip6_dst_lookup - perform route lookup on flow
922 * @sk: socket which provides route info
923 * @dst: pointer to dst_entry * for result
924 * @fl: flow to lookup
926 * This function performs a route lookup on the given flow.
928 * It returns zero on success, or a standard errno code on error.
930 int ip6_dst_lookup(struct sock
*sk
, struct dst_entry
**dst
, struct flowi
*fl
)
933 return ip6_dst_lookup_tail(sk
, dst
, fl
);
935 EXPORT_SYMBOL_GPL(ip6_dst_lookup
);
938 * ip6_sk_dst_lookup - perform socket cached route lookup on flow
939 * @sk: socket which provides the dst cache and route info
940 * @dst: pointer to dst_entry * for result
941 * @fl: flow to lookup
943 * This function performs a route lookup on the given flow with the
944 * possibility of using the cached route in the socket if it is valid.
945 * It will take the socket dst lock when operating on the dst cache.
946 * As a result, this function can only be used in process context.
948 * It returns zero on success, or a standard errno code on error.
950 int ip6_sk_dst_lookup(struct sock
*sk
, struct dst_entry
**dst
, struct flowi
*fl
)
954 *dst
= sk_dst_check(sk
, inet6_sk(sk
)->dst_cookie
);
955 *dst
= ip6_sk_dst_check(sk
, *dst
, fl
);
958 return ip6_dst_lookup_tail(sk
, dst
, fl
);
960 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup
);
962 static inline int ip6_ufo_append_data(struct sock
*sk
,
963 int getfrag(void *from
, char *to
, int offset
, int len
,
964 int odd
, struct sk_buff
*skb
),
965 void *from
, int length
, int hh_len
, int fragheaderlen
,
966 int transhdrlen
, int mtu
,unsigned int flags
)
972 /* There is support for UDP large send offload by network
973 * device, so create one single skb packet containing complete
976 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
) {
977 skb
= sock_alloc_send_skb(sk
,
978 hh_len
+ fragheaderlen
+ transhdrlen
+ 20,
979 (flags
& MSG_DONTWAIT
), &err
);
983 /* reserve space for Hardware header */
984 skb_reserve(skb
, hh_len
);
986 /* create space for UDP/IP header */
987 skb_put(skb
,fragheaderlen
+ transhdrlen
);
989 /* initialize network header pointer */
990 skb_reset_network_header(skb
);
992 /* initialize protocol header pointer */
993 skb
->transport_header
= skb
->network_header
+ fragheaderlen
;
995 skb
->ip_summed
= CHECKSUM_PARTIAL
;
997 sk
->sk_sndmsg_off
= 0;
1000 err
= skb_append_datato_frags(sk
,skb
, getfrag
, from
,
1001 (length
- transhdrlen
));
1003 struct frag_hdr fhdr
;
1005 /* specify the length of each IP datagram fragment*/
1006 skb_shinfo(skb
)->gso_size
= mtu
- fragheaderlen
-
1007 sizeof(struct frag_hdr
);
1008 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
1009 ipv6_select_ident(skb
, &fhdr
);
1010 skb_shinfo(skb
)->ip6_frag_id
= fhdr
.identification
;
1011 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1015 /* There is not enough support do UPD LSO,
1016 * so follow normal path
1023 int ip6_append_data(struct sock
*sk
, int getfrag(void *from
, char *to
,
1024 int offset
, int len
, int odd
, struct sk_buff
*skb
),
1025 void *from
, int length
, int transhdrlen
,
1026 int hlimit
, int tclass
, struct ipv6_txoptions
*opt
, struct flowi
*fl
,
1027 struct rt6_info
*rt
, unsigned int flags
)
1029 struct inet_sock
*inet
= inet_sk(sk
);
1030 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1031 struct sk_buff
*skb
;
1032 unsigned int maxfraglen
, fragheaderlen
;
1039 int csummode
= CHECKSUM_NONE
;
1041 if (flags
&MSG_PROBE
)
1043 if (skb_queue_empty(&sk
->sk_write_queue
)) {
1048 if (np
->cork
.opt
== NULL
) {
1049 np
->cork
.opt
= kmalloc(opt
->tot_len
,
1051 if (unlikely(np
->cork
.opt
== NULL
))
1053 } else if (np
->cork
.opt
->tot_len
< opt
->tot_len
) {
1054 printk(KERN_DEBUG
"ip6_append_data: invalid option length\n");
1057 memcpy(np
->cork
.opt
, opt
, opt
->tot_len
);
1058 inet
->cork
.flags
|= IPCORK_OPT
;
1059 /* need source address above miyazawa*/
1061 dst_hold(&rt
->u
.dst
);
1063 inet
->cork
.fl
= *fl
;
1064 np
->cork
.hop_limit
= hlimit
;
1065 np
->cork
.tclass
= tclass
;
1066 mtu
= dst_mtu(rt
->u
.dst
.path
);
1067 if (np
->frag_size
< mtu
) {
1069 mtu
= np
->frag_size
;
1071 inet
->cork
.fragsize
= mtu
;
1072 if (dst_allfrag(rt
->u
.dst
.path
))
1073 inet
->cork
.flags
|= IPCORK_ALLFRAG
;
1074 inet
->cork
.length
= 0;
1075 sk
->sk_sndmsg_page
= NULL
;
1076 sk
->sk_sndmsg_off
= 0;
1077 exthdrlen
= rt
->u
.dst
.header_len
+ (opt
? opt
->opt_flen
: 0);
1078 length
+= exthdrlen
;
1079 transhdrlen
+= exthdrlen
;
1082 fl
= &inet
->cork
.fl
;
1083 if (inet
->cork
.flags
& IPCORK_OPT
)
1087 mtu
= inet
->cork
.fragsize
;
1090 hh_len
= LL_RESERVED_SPACE(rt
->u
.dst
.dev
);
1092 fragheaderlen
= sizeof(struct ipv6hdr
) + rt
->u
.dst
.nfheader_len
+ (opt
? opt
->opt_nflen
: 0);
1093 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
- sizeof(struct frag_hdr
);
1095 if (mtu
<= sizeof(struct ipv6hdr
) + IPV6_MAXPLEN
) {
1096 if (inet
->cork
.length
+ length
> sizeof(struct ipv6hdr
) + IPV6_MAXPLEN
- fragheaderlen
) {
1097 ipv6_local_error(sk
, EMSGSIZE
, fl
, mtu
-exthdrlen
);
1103 * Let's try using as much space as possible.
1104 * Use MTU if total length of the message fits into the MTU.
1105 * Otherwise, we need to reserve fragment header and
1106 * fragment alignment (= 8-15 octects, in total).
1108 * Note that we may need to "move" the data from the tail of
1109 * of the buffer to the new fragment when we split
1112 * FIXME: It may be fragmented into multiple chunks
1113 * at once if non-fragmentable extension headers
1118 inet
->cork
.length
+= length
;
1119 if (((length
> mtu
) && (sk
->sk_protocol
== IPPROTO_UDP
)) &&
1120 (rt
->u
.dst
.dev
->features
& NETIF_F_UFO
)) {
1122 err
= ip6_ufo_append_data(sk
, getfrag
, from
, length
, hh_len
,
1123 fragheaderlen
, transhdrlen
, mtu
,
1130 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
)
1133 while (length
> 0) {
1134 /* Check if the remaining data fits into current packet. */
1135 copy
= (inet
->cork
.length
<= mtu
&& !(inet
->cork
.flags
& IPCORK_ALLFRAG
) ? mtu
: maxfraglen
) - skb
->len
;
1137 copy
= maxfraglen
- skb
->len
;
1141 unsigned int datalen
;
1142 unsigned int fraglen
;
1143 unsigned int fraggap
;
1144 unsigned int alloclen
;
1145 struct sk_buff
*skb_prev
;
1149 /* There's no room in the current skb */
1151 fraggap
= skb_prev
->len
- maxfraglen
;
1156 * If remaining data exceeds the mtu,
1157 * we know we need more fragment(s).
1159 datalen
= length
+ fraggap
;
1160 if (datalen
> (inet
->cork
.length
<= mtu
&& !(inet
->cork
.flags
& IPCORK_ALLFRAG
) ? mtu
: maxfraglen
) - fragheaderlen
)
1161 datalen
= maxfraglen
- fragheaderlen
;
1163 fraglen
= datalen
+ fragheaderlen
;
1164 if ((flags
& MSG_MORE
) &&
1165 !(rt
->u
.dst
.dev
->features
&NETIF_F_SG
))
1168 alloclen
= datalen
+ fragheaderlen
;
1171 * The last fragment gets additional space at tail.
1172 * Note: we overallocate on fragments with MSG_MODE
1173 * because we have no idea if we're the last one.
1175 if (datalen
== length
+ fraggap
)
1176 alloclen
+= rt
->u
.dst
.trailer_len
;
1179 * We just reserve space for fragment header.
1180 * Note: this may be overallocation if the message
1181 * (without MSG_MORE) fits into the MTU.
1183 alloclen
+= sizeof(struct frag_hdr
);
1186 skb
= sock_alloc_send_skb(sk
,
1188 (flags
& MSG_DONTWAIT
), &err
);
1191 if (atomic_read(&sk
->sk_wmem_alloc
) <=
1193 skb
= sock_wmalloc(sk
,
1194 alloclen
+ hh_len
, 1,
1196 if (unlikely(skb
== NULL
))
1202 * Fill in the control structures
1204 skb
->ip_summed
= csummode
;
1206 /* reserve for fragmentation */
1207 skb_reserve(skb
, hh_len
+sizeof(struct frag_hdr
));
1210 * Find where to start putting bytes
1212 data
= skb_put(skb
, fraglen
);
1213 skb_set_network_header(skb
, exthdrlen
);
1214 data
+= fragheaderlen
;
1215 skb
->transport_header
= (skb
->network_header
+
1218 skb
->csum
= skb_copy_and_csum_bits(
1219 skb_prev
, maxfraglen
,
1220 data
+ transhdrlen
, fraggap
, 0);
1221 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
1224 pskb_trim_unique(skb_prev
, maxfraglen
);
1226 copy
= datalen
- transhdrlen
- fraggap
;
1231 } else if (copy
> 0 && getfrag(from
, data
+ transhdrlen
, offset
, copy
, fraggap
, skb
) < 0) {
1238 length
-= datalen
- fraggap
;
1241 csummode
= CHECKSUM_NONE
;
1244 * Put the packet on the pending queue
1246 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1253 if (!(rt
->u
.dst
.dev
->features
&NETIF_F_SG
)) {
1257 if (getfrag(from
, skb_put(skb
, copy
),
1258 offset
, copy
, off
, skb
) < 0) {
1259 __skb_trim(skb
, off
);
1264 int i
= skb_shinfo(skb
)->nr_frags
;
1265 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
-1];
1266 struct page
*page
= sk
->sk_sndmsg_page
;
1267 int off
= sk
->sk_sndmsg_off
;
1270 if (page
&& (left
= PAGE_SIZE
- off
) > 0) {
1273 if (page
!= frag
->page
) {
1274 if (i
== MAX_SKB_FRAGS
) {
1279 skb_fill_page_desc(skb
, i
, page
, sk
->sk_sndmsg_off
, 0);
1280 frag
= &skb_shinfo(skb
)->frags
[i
];
1282 } else if(i
< MAX_SKB_FRAGS
) {
1283 if (copy
> PAGE_SIZE
)
1285 page
= alloc_pages(sk
->sk_allocation
, 0);
1290 sk
->sk_sndmsg_page
= page
;
1291 sk
->sk_sndmsg_off
= 0;
1293 skb_fill_page_desc(skb
, i
, page
, 0, 0);
1294 frag
= &skb_shinfo(skb
)->frags
[i
];
1295 skb
->truesize
+= PAGE_SIZE
;
1296 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
1301 if (getfrag(from
, page_address(frag
->page
)+frag
->page_offset
+frag
->size
, offset
, copy
, skb
->len
, skb
) < 0) {
1305 sk
->sk_sndmsg_off
+= copy
;
1308 skb
->data_len
+= copy
;
1315 inet
->cork
.length
-= length
;
1316 IP6_INC_STATS(rt
->rt6i_idev
, IPSTATS_MIB_OUTDISCARDS
);
1320 int ip6_push_pending_frames(struct sock
*sk
)
1322 struct sk_buff
*skb
, *tmp_skb
;
1323 struct sk_buff
**tail_skb
;
1324 struct in6_addr final_dst_buf
, *final_dst
= &final_dst_buf
;
1325 struct inet_sock
*inet
= inet_sk(sk
);
1326 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1327 struct ipv6hdr
*hdr
;
1328 struct ipv6_txoptions
*opt
= np
->cork
.opt
;
1329 struct rt6_info
*rt
= np
->cork
.rt
;
1330 struct flowi
*fl
= &inet
->cork
.fl
;
1331 unsigned char proto
= fl
->proto
;
1334 if ((skb
= __skb_dequeue(&sk
->sk_write_queue
)) == NULL
)
1336 tail_skb
= &(skb_shinfo(skb
)->frag_list
);
1338 /* move skb->data to ip header from ext header */
1339 if (skb
->data
< skb_network_header(skb
))
1340 __skb_pull(skb
, skb_network_offset(skb
));
1341 while ((tmp_skb
= __skb_dequeue(&sk
->sk_write_queue
)) != NULL
) {
1342 __skb_pull(tmp_skb
, skb_network_header_len(skb
));
1343 *tail_skb
= tmp_skb
;
1344 tail_skb
= &(tmp_skb
->next
);
1345 skb
->len
+= tmp_skb
->len
;
1346 skb
->data_len
+= tmp_skb
->len
;
1347 skb
->truesize
+= tmp_skb
->truesize
;
1348 __sock_put(tmp_skb
->sk
);
1349 tmp_skb
->destructor
= NULL
;
1353 ipv6_addr_copy(final_dst
, &fl
->fl6_dst
);
1354 __skb_pull(skb
, skb_network_header_len(skb
));
1355 if (opt
&& opt
->opt_flen
)
1356 ipv6_push_frag_opts(skb
, opt
, &proto
);
1357 if (opt
&& opt
->opt_nflen
)
1358 ipv6_push_nfrag_opts(skb
, opt
, &proto
, &final_dst
);
1360 skb_push(skb
, sizeof(struct ipv6hdr
));
1361 skb_reset_network_header(skb
);
1362 hdr
= ipv6_hdr(skb
);
1364 *(__be32
*)hdr
= fl
->fl6_flowlabel
|
1365 htonl(0x60000000 | ((int)np
->cork
.tclass
<< 20));
1367 if (skb
->len
<= sizeof(struct ipv6hdr
) + IPV6_MAXPLEN
)
1368 hdr
->payload_len
= htons(skb
->len
- sizeof(struct ipv6hdr
));
1370 hdr
->payload_len
= 0;
1371 hdr
->hop_limit
= np
->cork
.hop_limit
;
1372 hdr
->nexthdr
= proto
;
1373 ipv6_addr_copy(&hdr
->saddr
, &fl
->fl6_src
);
1374 ipv6_addr_copy(&hdr
->daddr
, final_dst
);
1376 skb
->priority
= sk
->sk_priority
;
1378 skb
->dst
= dst_clone(&rt
->u
.dst
);
1379 IP6_INC_STATS(rt
->rt6i_idev
, IPSTATS_MIB_OUTREQUESTS
);
1380 err
= NF_HOOK(PF_INET6
, NF_IP6_LOCAL_OUT
, skb
, NULL
, skb
->dst
->dev
, dst_output
);
1383 err
= np
->recverr
? net_xmit_errno(err
) : 0;
1389 inet
->cork
.flags
&= ~IPCORK_OPT
;
1390 kfree(np
->cork
.opt
);
1391 np
->cork
.opt
= NULL
;
1393 dst_release(&np
->cork
.rt
->u
.dst
);
1395 inet
->cork
.flags
&= ~IPCORK_ALLFRAG
;
1397 memset(&inet
->cork
.fl
, 0, sizeof(inet
->cork
.fl
));
1403 void ip6_flush_pending_frames(struct sock
*sk
)
1405 struct inet_sock
*inet
= inet_sk(sk
);
1406 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1407 struct sk_buff
*skb
;
1409 while ((skb
= __skb_dequeue_tail(&sk
->sk_write_queue
)) != NULL
) {
1410 IP6_INC_STATS(ip6_dst_idev(skb
->dst
),
1411 IPSTATS_MIB_OUTDISCARDS
);
1415 inet
->cork
.flags
&= ~IPCORK_OPT
;
1417 kfree(np
->cork
.opt
);
1418 np
->cork
.opt
= NULL
;
1420 dst_release(&np
->cork
.rt
->u
.dst
);
1422 inet
->cork
.flags
&= ~IPCORK_ALLFRAG
;
1424 memset(&inet
->cork
.fl
, 0, sizeof(inet
->cork
.fl
));