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 static inline int ip6_skb_dst_mtu(struct sk_buff
*skb
)
142 struct ipv6_pinfo
*np
= skb
->sk
? inet6_sk(skb
->sk
) : NULL
;
144 return (np
&& np
->pmtudisc
== IPV6_PMTUDISC_PROBE
) ?
145 skb
->dst
->dev
->mtu
: dst_mtu(skb
->dst
);
148 int ip6_output(struct sk_buff
*skb
)
150 if ((skb
->len
> ip6_skb_dst_mtu(skb
) && !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
);
186 IP6_INC_STATS(ip6_dst_idev(skb
->dst
),
187 IPSTATS_MIB_OUTDISCARDS
);
194 skb_set_owner_w(skb
, sk
);
197 ipv6_push_frag_opts(skb
, opt
, &proto
);
199 ipv6_push_nfrag_opts(skb
, opt
, &proto
, &first_hop
);
202 skb_push(skb
, sizeof(struct ipv6hdr
));
203 skb_reset_network_header(skb
);
207 * Fill in the IPv6 header
212 hlimit
= np
->hop_limit
;
214 hlimit
= dst_metric(dst
, RTAX_HOPLIMIT
);
216 hlimit
= ipv6_get_hoplimit(dst
->dev
);
224 *(__be32
*)hdr
= htonl(0x60000000 | (tclass
<< 20)) | fl
->fl6_flowlabel
;
226 hdr
->payload_len
= htons(seg_len
);
227 hdr
->nexthdr
= proto
;
228 hdr
->hop_limit
= hlimit
;
230 ipv6_addr_copy(&hdr
->saddr
, &fl
->fl6_src
);
231 ipv6_addr_copy(&hdr
->daddr
, first_hop
);
233 skb
->priority
= sk
->sk_priority
;
236 if ((skb
->len
<= mtu
) || ipfragok
|| skb_is_gso(skb
)) {
237 IP6_INC_STATS(ip6_dst_idev(skb
->dst
),
238 IPSTATS_MIB_OUTREQUESTS
);
239 return NF_HOOK(PF_INET6
, NF_IP6_LOCAL_OUT
, skb
, NULL
, dst
->dev
,
244 printk(KERN_DEBUG
"IPv6: sending pkt_too_big to self\n");
246 icmpv6_send(skb
, ICMPV6_PKT_TOOBIG
, 0, mtu
, skb
->dev
);
247 IP6_INC_STATS(ip6_dst_idev(skb
->dst
), IPSTATS_MIB_FRAGFAILS
);
252 EXPORT_SYMBOL(ip6_xmit
);
255 * To avoid extra problems ND packets are send through this
256 * routine. It's code duplication but I really want to avoid
257 * extra checks since ipv6_build_header is used by TCP (which
258 * is for us performance critical)
261 int ip6_nd_hdr(struct sock
*sk
, struct sk_buff
*skb
, struct net_device
*dev
,
262 struct in6_addr
*saddr
, struct in6_addr
*daddr
,
265 struct ipv6_pinfo
*np
= inet6_sk(sk
);
269 skb
->protocol
= htons(ETH_P_IPV6
);
272 totlen
= len
+ sizeof(struct ipv6hdr
);
274 skb_reset_network_header(skb
);
275 skb_put(skb
, sizeof(struct ipv6hdr
));
278 *(__be32
*)hdr
= htonl(0x60000000);
280 hdr
->payload_len
= htons(len
);
281 hdr
->nexthdr
= proto
;
282 hdr
->hop_limit
= np
->hop_limit
;
284 ipv6_addr_copy(&hdr
->saddr
, saddr
);
285 ipv6_addr_copy(&hdr
->daddr
, daddr
);
290 static int ip6_call_ra_chain(struct sk_buff
*skb
, int sel
)
292 struct ip6_ra_chain
*ra
;
293 struct sock
*last
= NULL
;
295 read_lock(&ip6_ra_lock
);
296 for (ra
= ip6_ra_chain
; ra
; ra
= ra
->next
) {
297 struct sock
*sk
= ra
->sk
;
298 if (sk
&& ra
->sel
== sel
&&
299 (!sk
->sk_bound_dev_if
||
300 sk
->sk_bound_dev_if
== skb
->dev
->ifindex
)) {
302 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
304 rawv6_rcv(last
, skb2
);
311 rawv6_rcv(last
, skb
);
312 read_unlock(&ip6_ra_lock
);
315 read_unlock(&ip6_ra_lock
);
319 static int ip6_forward_proxy_check(struct sk_buff
*skb
)
321 struct ipv6hdr
*hdr
= ipv6_hdr(skb
);
322 u8 nexthdr
= hdr
->nexthdr
;
325 if (ipv6_ext_hdr(nexthdr
)) {
326 offset
= ipv6_skip_exthdr(skb
, sizeof(*hdr
), &nexthdr
);
330 offset
= sizeof(struct ipv6hdr
);
332 if (nexthdr
== IPPROTO_ICMPV6
) {
333 struct icmp6hdr
*icmp6
;
335 if (!pskb_may_pull(skb
, (skb_network_header(skb
) +
336 offset
+ 1 - skb
->data
)))
339 icmp6
= (struct icmp6hdr
*)(skb_network_header(skb
) + offset
);
341 switch (icmp6
->icmp6_type
) {
342 case NDISC_ROUTER_SOLICITATION
:
343 case NDISC_ROUTER_ADVERTISEMENT
:
344 case NDISC_NEIGHBOUR_SOLICITATION
:
345 case NDISC_NEIGHBOUR_ADVERTISEMENT
:
347 /* For reaction involving unicast neighbor discovery
348 * message destined to the proxied address, pass it to
358 * The proxying router can't forward traffic sent to a link-local
359 * address, so signal the sender and discard the packet. This
360 * behavior is clarified by the MIPv6 specification.
362 if (ipv6_addr_type(&hdr
->daddr
) & IPV6_ADDR_LINKLOCAL
) {
363 dst_link_failure(skb
);
370 static inline int ip6_forward_finish(struct sk_buff
*skb
)
372 return dst_output(skb
);
375 int ip6_forward(struct sk_buff
*skb
)
377 struct dst_entry
*dst
= skb
->dst
;
378 struct ipv6hdr
*hdr
= ipv6_hdr(skb
);
379 struct inet6_skb_parm
*opt
= IP6CB(skb
);
381 if (ipv6_devconf
.forwarding
== 0)
384 if (!xfrm6_policy_check(NULL
, XFRM_POLICY_FWD
, skb
)) {
385 IP6_INC_STATS(ip6_dst_idev(dst
), IPSTATS_MIB_INDISCARDS
);
389 skb_forward_csum(skb
);
392 * We DO NOT make any processing on
393 * RA packets, pushing them to user level AS IS
394 * without ane WARRANTY that application will be able
395 * to interpret them. The reason is that we
396 * cannot make anything clever here.
398 * We are not end-node, so that if packet contains
399 * AH/ESP, we cannot make anything.
400 * Defragmentation also would be mistake, RA packets
401 * cannot be fragmented, because there is no warranty
402 * that different fragments will go along one path. --ANK
405 u8
*ptr
= skb_network_header(skb
) + opt
->ra
;
406 if (ip6_call_ra_chain(skb
, (ptr
[2]<<8) + ptr
[3]))
411 * check and decrement ttl
413 if (hdr
->hop_limit
<= 1) {
414 /* Force OUTPUT device used as source address */
416 icmpv6_send(skb
, ICMPV6_TIME_EXCEED
, ICMPV6_EXC_HOPLIMIT
,
418 IP6_INC_STATS_BH(ip6_dst_idev(dst
), IPSTATS_MIB_INHDRERRORS
);
424 /* XXX: idev->cnf.proxy_ndp? */
425 if (ipv6_devconf
.proxy_ndp
&&
426 pneigh_lookup(&nd_tbl
, &hdr
->daddr
, skb
->dev
, 0)) {
427 int proxied
= ip6_forward_proxy_check(skb
);
429 return ip6_input(skb
);
430 else if (proxied
< 0) {
431 IP6_INC_STATS(ip6_dst_idev(dst
), IPSTATS_MIB_INDISCARDS
);
436 if (!xfrm6_route_forward(skb
)) {
437 IP6_INC_STATS(ip6_dst_idev(dst
), IPSTATS_MIB_INDISCARDS
);
442 /* IPv6 specs say nothing about it, but it is clear that we cannot
443 send redirects to source routed frames.
445 if (skb
->dev
== dst
->dev
&& dst
->neighbour
&& opt
->srcrt
== 0) {
446 struct in6_addr
*target
= NULL
;
448 struct neighbour
*n
= dst
->neighbour
;
451 * incoming and outgoing devices are the same
455 rt
= (struct rt6_info
*) dst
;
456 if ((rt
->rt6i_flags
& RTF_GATEWAY
))
457 target
= (struct in6_addr
*)&n
->primary_key
;
459 target
= &hdr
->daddr
;
461 /* Limit redirects both by destination (here)
462 and by source (inside ndisc_send_redirect)
464 if (xrlim_allow(dst
, 1*HZ
))
465 ndisc_send_redirect(skb
, n
, target
);
467 int addrtype
= ipv6_addr_type(&hdr
->saddr
);
469 /* This check is security critical. */
470 if (addrtype
& (IPV6_ADDR_MULTICAST
|IPV6_ADDR_LOOPBACK
))
472 if (addrtype
& IPV6_ADDR_LINKLOCAL
) {
473 icmpv6_send(skb
, ICMPV6_DEST_UNREACH
,
474 ICMPV6_NOT_NEIGHBOUR
, 0, skb
->dev
);
479 if (skb
->len
> dst_mtu(dst
)) {
480 /* Again, force OUTPUT device used as source address */
482 icmpv6_send(skb
, ICMPV6_PKT_TOOBIG
, 0, dst_mtu(dst
), skb
->dev
);
483 IP6_INC_STATS_BH(ip6_dst_idev(dst
), IPSTATS_MIB_INTOOBIGERRORS
);
484 IP6_INC_STATS_BH(ip6_dst_idev(dst
), IPSTATS_MIB_FRAGFAILS
);
489 if (skb_cow(skb
, dst
->dev
->hard_header_len
)) {
490 IP6_INC_STATS(ip6_dst_idev(dst
), IPSTATS_MIB_OUTDISCARDS
);
496 /* Mangling hops number delayed to point after skb COW */
500 IP6_INC_STATS_BH(ip6_dst_idev(dst
), IPSTATS_MIB_OUTFORWDATAGRAMS
);
501 return NF_HOOK(PF_INET6
,NF_IP6_FORWARD
, skb
, skb
->dev
, dst
->dev
, ip6_forward_finish
);
504 IP6_INC_STATS_BH(ip6_dst_idev(dst
), IPSTATS_MIB_INADDRERRORS
);
510 static void ip6_copy_metadata(struct sk_buff
*to
, struct sk_buff
*from
)
512 to
->pkt_type
= from
->pkt_type
;
513 to
->priority
= from
->priority
;
514 to
->protocol
= from
->protocol
;
515 dst_release(to
->dst
);
516 to
->dst
= dst_clone(from
->dst
);
518 to
->mark
= from
->mark
;
520 #ifdef CONFIG_NET_SCHED
521 to
->tc_index
= from
->tc_index
;
524 skb_copy_secmark(to
, from
);
527 int ip6_find_1stfragopt(struct sk_buff
*skb
, u8
**nexthdr
)
529 u16 offset
= sizeof(struct ipv6hdr
);
530 struct ipv6_opt_hdr
*exthdr
=
531 (struct ipv6_opt_hdr
*)(ipv6_hdr(skb
) + 1);
532 unsigned int packet_len
= skb
->tail
- skb
->network_header
;
534 *nexthdr
= &ipv6_hdr(skb
)->nexthdr
;
536 while (offset
+ 1 <= packet_len
) {
542 case NEXTHDR_ROUTING
:
546 #ifdef CONFIG_IPV6_MIP6
547 if (ipv6_find_tlv(skb
, offset
, IPV6_TLV_HAO
) >= 0)
557 offset
+= ipv6_optlen(exthdr
);
558 *nexthdr
= &exthdr
->nexthdr
;
559 exthdr
= (struct ipv6_opt_hdr
*)(skb_network_header(skb
) +
565 EXPORT_SYMBOL_GPL(ip6_find_1stfragopt
);
567 static int ip6_fragment(struct sk_buff
*skb
, int (*output
)(struct sk_buff
*))
569 struct net_device
*dev
;
570 struct sk_buff
*frag
;
571 struct rt6_info
*rt
= (struct rt6_info
*)skb
->dst
;
572 struct ipv6_pinfo
*np
= skb
->sk
? inet6_sk(skb
->sk
) : NULL
;
573 struct ipv6hdr
*tmp_hdr
;
575 unsigned int mtu
, hlen
, left
, len
;
577 int ptr
, offset
= 0, err
=0;
578 u8
*prevhdr
, nexthdr
= 0;
581 hlen
= ip6_find_1stfragopt(skb
, &prevhdr
);
584 mtu
= ip6_skb_dst_mtu(skb
);
586 /* We must not fragment if the socket is set to force MTU discovery
587 * or if the skb it not generated by a local socket. (This last
588 * check should be redundant, but it's free.)
590 if (!np
|| np
->pmtudisc
>= IPV6_PMTUDISC_DO
) {
591 skb
->dev
= skb
->dst
->dev
;
592 icmpv6_send(skb
, ICMPV6_PKT_TOOBIG
, 0, mtu
, skb
->dev
);
593 IP6_INC_STATS(ip6_dst_idev(skb
->dst
), IPSTATS_MIB_FRAGFAILS
);
598 if (np
&& np
->frag_size
< mtu
) {
602 mtu
-= hlen
+ sizeof(struct frag_hdr
);
604 if (skb_shinfo(skb
)->frag_list
) {
605 int first_len
= skb_pagelen(skb
);
607 if (first_len
- hlen
> mtu
||
608 ((first_len
- hlen
) & 7) ||
612 for (frag
= skb_shinfo(skb
)->frag_list
; frag
; frag
= frag
->next
) {
613 /* Correct geometry. */
614 if (frag
->len
> mtu
||
615 ((frag
->len
& 7) && frag
->next
) ||
616 skb_headroom(frag
) < hlen
)
619 /* Partially cloned skb? */
620 if (skb_shared(frag
))
627 frag
->destructor
= sock_wfree
;
628 skb
->truesize
-= frag
->truesize
;
634 frag
= skb_shinfo(skb
)->frag_list
;
635 skb_shinfo(skb
)->frag_list
= NULL
;
638 *prevhdr
= NEXTHDR_FRAGMENT
;
639 tmp_hdr
= kmemdup(skb_network_header(skb
), hlen
, GFP_ATOMIC
);
641 IP6_INC_STATS(ip6_dst_idev(skb
->dst
), IPSTATS_MIB_FRAGFAILS
);
645 __skb_pull(skb
, hlen
);
646 fh
= (struct frag_hdr
*)__skb_push(skb
, sizeof(struct frag_hdr
));
647 __skb_push(skb
, hlen
);
648 skb_reset_network_header(skb
);
649 memcpy(skb_network_header(skb
), tmp_hdr
, hlen
);
651 ipv6_select_ident(skb
, fh
);
652 fh
->nexthdr
= nexthdr
;
654 fh
->frag_off
= htons(IP6_MF
);
655 frag_id
= fh
->identification
;
657 first_len
= skb_pagelen(skb
);
658 skb
->data_len
= first_len
- skb_headlen(skb
);
659 skb
->len
= first_len
;
660 ipv6_hdr(skb
)->payload_len
= htons(first_len
-
661 sizeof(struct ipv6hdr
));
663 dst_hold(&rt
->u
.dst
);
666 /* Prepare header of the next frame,
667 * before previous one went down. */
669 frag
->ip_summed
= CHECKSUM_NONE
;
670 skb_reset_transport_header(frag
);
671 fh
= (struct frag_hdr
*)__skb_push(frag
, sizeof(struct frag_hdr
));
672 __skb_push(frag
, hlen
);
673 skb_reset_network_header(frag
);
674 memcpy(skb_network_header(frag
), tmp_hdr
,
676 offset
+= skb
->len
- hlen
- sizeof(struct frag_hdr
);
677 fh
->nexthdr
= nexthdr
;
679 fh
->frag_off
= htons(offset
);
680 if (frag
->next
!= NULL
)
681 fh
->frag_off
|= htons(IP6_MF
);
682 fh
->identification
= frag_id
;
683 ipv6_hdr(frag
)->payload_len
=
685 sizeof(struct ipv6hdr
));
686 ip6_copy_metadata(frag
, skb
);
691 IP6_INC_STATS(ip6_dst_idev(&rt
->u
.dst
), IPSTATS_MIB_FRAGCREATES
);
704 IP6_INC_STATS(ip6_dst_idev(&rt
->u
.dst
), IPSTATS_MIB_FRAGOKS
);
705 dst_release(&rt
->u
.dst
);
715 IP6_INC_STATS(ip6_dst_idev(&rt
->u
.dst
), IPSTATS_MIB_FRAGFAILS
);
716 dst_release(&rt
->u
.dst
);
721 left
= skb
->len
- hlen
; /* Space per frame */
722 ptr
= hlen
; /* Where to start from */
725 * Fragment the datagram.
728 *prevhdr
= NEXTHDR_FRAGMENT
;
731 * Keep copying data until we run out.
735 /* IF: it doesn't fit, use 'mtu' - the data space left */
738 /* IF: we are not sending upto and including the packet end
739 then align the next start on an eight byte boundary */
747 if ((frag
= alloc_skb(len
+hlen
+sizeof(struct frag_hdr
)+LL_RESERVED_SPACE(rt
->u
.dst
.dev
), GFP_ATOMIC
)) == NULL
) {
748 NETDEBUG(KERN_INFO
"IPv6: frag: no memory for new fragment!\n");
749 IP6_INC_STATS(ip6_dst_idev(skb
->dst
),
750 IPSTATS_MIB_FRAGFAILS
);
756 * Set up data on packet
759 ip6_copy_metadata(frag
, skb
);
760 skb_reserve(frag
, LL_RESERVED_SPACE(rt
->u
.dst
.dev
));
761 skb_put(frag
, len
+ hlen
+ sizeof(struct frag_hdr
));
762 skb_reset_network_header(frag
);
763 fh
= (struct frag_hdr
*)(skb_network_header(frag
) + hlen
);
764 frag
->transport_header
= (frag
->network_header
+ hlen
+
765 sizeof(struct frag_hdr
));
768 * Charge the memory for the fragment to any owner
772 skb_set_owner_w(frag
, skb
->sk
);
775 * Copy the packet header into the new buffer.
777 skb_copy_from_linear_data(skb
, skb_network_header(frag
), hlen
);
780 * Build fragment header.
782 fh
->nexthdr
= nexthdr
;
785 ipv6_select_ident(skb
, fh
);
786 frag_id
= fh
->identification
;
788 fh
->identification
= frag_id
;
791 * Copy a block of the IP datagram.
793 if (skb_copy_bits(skb
, ptr
, skb_transport_header(skb
), len
))
797 fh
->frag_off
= htons(offset
);
799 fh
->frag_off
|= htons(IP6_MF
);
800 ipv6_hdr(frag
)->payload_len
= htons(frag
->len
-
801 sizeof(struct ipv6hdr
));
807 * Put this fragment into the sending queue.
813 IP6_INC_STATS(ip6_dst_idev(skb
->dst
), IPSTATS_MIB_FRAGCREATES
);
815 IP6_INC_STATS(ip6_dst_idev(skb
->dst
),
816 IPSTATS_MIB_FRAGOKS
);
821 IP6_INC_STATS(ip6_dst_idev(skb
->dst
),
822 IPSTATS_MIB_FRAGFAILS
);
827 static inline int ip6_rt_check(struct rt6key
*rt_key
,
828 struct in6_addr
*fl_addr
,
829 struct in6_addr
*addr_cache
)
831 return ((rt_key
->plen
!= 128 || !ipv6_addr_equal(fl_addr
, &rt_key
->addr
)) &&
832 (addr_cache
== NULL
|| !ipv6_addr_equal(fl_addr
, addr_cache
)));
835 static struct dst_entry
*ip6_sk_dst_check(struct sock
*sk
,
836 struct dst_entry
*dst
,
839 struct ipv6_pinfo
*np
= inet6_sk(sk
);
840 struct rt6_info
*rt
= (struct rt6_info
*)dst
;
845 /* Yes, checking route validity in not connected
846 * case is not very simple. Take into account,
847 * that we do not support routing by source, TOS,
848 * and MSG_DONTROUTE --ANK (980726)
850 * 1. ip6_rt_check(): If route was host route,
851 * check that cached destination is current.
852 * If it is network route, we still may
853 * check its validity using saved pointer
854 * to the last used address: daddr_cache.
855 * We do not want to save whole address now,
856 * (because main consumer of this service
857 * is tcp, which has not this problem),
858 * so that the last trick works only on connected
860 * 2. oif also should be the same.
862 if (ip6_rt_check(&rt
->rt6i_dst
, &fl
->fl6_dst
, np
->daddr_cache
) ||
863 #ifdef CONFIG_IPV6_SUBTREES
864 ip6_rt_check(&rt
->rt6i_src
, &fl
->fl6_src
, np
->saddr_cache
) ||
866 (fl
->oif
&& fl
->oif
!= dst
->dev
->ifindex
)) {
875 static int ip6_dst_lookup_tail(struct sock
*sk
,
876 struct dst_entry
**dst
, struct flowi
*fl
)
881 *dst
= ip6_route_output(sk
, fl
);
883 if ((err
= (*dst
)->error
))
884 goto out_err_release
;
886 if (ipv6_addr_any(&fl
->fl6_src
)) {
887 err
= ipv6_get_saddr(*dst
, &fl
->fl6_dst
, &fl
->fl6_src
);
889 goto out_err_release
;
892 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
894 * Here if the dst entry we've looked up
895 * has a neighbour entry that is in the INCOMPLETE
896 * state and the src address from the flow is
897 * marked as OPTIMISTIC, we release the found
898 * dst entry and replace it instead with the
899 * dst entry of the nexthop router
901 if (!((*dst
)->neighbour
->nud_state
& NUD_VALID
)) {
902 struct inet6_ifaddr
*ifp
;
906 ifp
= ipv6_get_ifaddr(&fl
->fl6_src
, (*dst
)->dev
, 1);
908 redirect
= (ifp
&& ifp
->flags
& IFA_F_OPTIMISTIC
);
914 * We need to get the dst entry for the
915 * default router instead
918 memcpy(&fl_gw
, fl
, sizeof(struct flowi
));
919 memset(&fl_gw
.fl6_dst
, 0, sizeof(struct in6_addr
));
920 *dst
= ip6_route_output(sk
, &fl_gw
);
921 if ((err
= (*dst
)->error
))
922 goto out_err_release
;
936 * ip6_dst_lookup - perform route lookup on flow
937 * @sk: socket which provides route info
938 * @dst: pointer to dst_entry * for result
939 * @fl: flow to lookup
941 * This function performs a route lookup on the given flow.
943 * It returns zero on success, or a standard errno code on error.
945 int ip6_dst_lookup(struct sock
*sk
, struct dst_entry
**dst
, struct flowi
*fl
)
948 return ip6_dst_lookup_tail(sk
, dst
, fl
);
950 EXPORT_SYMBOL_GPL(ip6_dst_lookup
);
953 * ip6_sk_dst_lookup - perform socket cached route lookup on flow
954 * @sk: socket which provides the dst cache and route info
955 * @dst: pointer to dst_entry * for result
956 * @fl: flow to lookup
958 * This function performs a route lookup on the given flow with the
959 * possibility of using the cached route in the socket if it is valid.
960 * It will take the socket dst lock when operating on the dst cache.
961 * As a result, this function can only be used in process context.
963 * It returns zero on success, or a standard errno code on error.
965 int ip6_sk_dst_lookup(struct sock
*sk
, struct dst_entry
**dst
, struct flowi
*fl
)
969 *dst
= sk_dst_check(sk
, inet6_sk(sk
)->dst_cookie
);
970 *dst
= ip6_sk_dst_check(sk
, *dst
, fl
);
973 return ip6_dst_lookup_tail(sk
, dst
, fl
);
975 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup
);
977 static inline int ip6_ufo_append_data(struct sock
*sk
,
978 int getfrag(void *from
, char *to
, int offset
, int len
,
979 int odd
, struct sk_buff
*skb
),
980 void *from
, int length
, int hh_len
, int fragheaderlen
,
981 int transhdrlen
, int mtu
,unsigned int flags
)
987 /* There is support for UDP large send offload by network
988 * device, so create one single skb packet containing complete
991 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
) {
992 skb
= sock_alloc_send_skb(sk
,
993 hh_len
+ fragheaderlen
+ transhdrlen
+ 20,
994 (flags
& MSG_DONTWAIT
), &err
);
998 /* reserve space for Hardware header */
999 skb_reserve(skb
, hh_len
);
1001 /* create space for UDP/IP header */
1002 skb_put(skb
,fragheaderlen
+ transhdrlen
);
1004 /* initialize network header pointer */
1005 skb_reset_network_header(skb
);
1007 /* initialize protocol header pointer */
1008 skb
->transport_header
= skb
->network_header
+ fragheaderlen
;
1010 skb
->ip_summed
= CHECKSUM_PARTIAL
;
1012 sk
->sk_sndmsg_off
= 0;
1015 err
= skb_append_datato_frags(sk
,skb
, getfrag
, from
,
1016 (length
- transhdrlen
));
1018 struct frag_hdr fhdr
;
1020 /* specify the length of each IP datagram fragment*/
1021 skb_shinfo(skb
)->gso_size
= mtu
- fragheaderlen
-
1022 sizeof(struct frag_hdr
);
1023 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
1024 ipv6_select_ident(skb
, &fhdr
);
1025 skb_shinfo(skb
)->ip6_frag_id
= fhdr
.identification
;
1026 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1030 /* There is not enough support do UPD LSO,
1031 * so follow normal path
1038 int ip6_append_data(struct sock
*sk
, int getfrag(void *from
, char *to
,
1039 int offset
, int len
, int odd
, struct sk_buff
*skb
),
1040 void *from
, int length
, int transhdrlen
,
1041 int hlimit
, int tclass
, struct ipv6_txoptions
*opt
, struct flowi
*fl
,
1042 struct rt6_info
*rt
, unsigned int flags
)
1044 struct inet_sock
*inet
= inet_sk(sk
);
1045 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1046 struct sk_buff
*skb
;
1047 unsigned int maxfraglen
, fragheaderlen
;
1054 int csummode
= CHECKSUM_NONE
;
1056 if (flags
&MSG_PROBE
)
1058 if (skb_queue_empty(&sk
->sk_write_queue
)) {
1063 if (np
->cork
.opt
== NULL
) {
1064 np
->cork
.opt
= kmalloc(opt
->tot_len
,
1066 if (unlikely(np
->cork
.opt
== NULL
))
1068 } else if (np
->cork
.opt
->tot_len
< opt
->tot_len
) {
1069 printk(KERN_DEBUG
"ip6_append_data: invalid option length\n");
1072 memcpy(np
->cork
.opt
, opt
, opt
->tot_len
);
1073 inet
->cork
.flags
|= IPCORK_OPT
;
1074 /* need source address above miyazawa*/
1076 dst_hold(&rt
->u
.dst
);
1078 inet
->cork
.fl
= *fl
;
1079 np
->cork
.hop_limit
= hlimit
;
1080 np
->cork
.tclass
= tclass
;
1081 mtu
= np
->pmtudisc
== IPV6_PMTUDISC_PROBE
?
1082 rt
->u
.dst
.dev
->mtu
: dst_mtu(rt
->u
.dst
.path
);
1083 if (np
->frag_size
< mtu
) {
1085 mtu
= np
->frag_size
;
1087 inet
->cork
.fragsize
= mtu
;
1088 if (dst_allfrag(rt
->u
.dst
.path
))
1089 inet
->cork
.flags
|= IPCORK_ALLFRAG
;
1090 inet
->cork
.length
= 0;
1091 sk
->sk_sndmsg_page
= NULL
;
1092 sk
->sk_sndmsg_off
= 0;
1093 exthdrlen
= rt
->u
.dst
.header_len
+ (opt
? opt
->opt_flen
: 0);
1094 length
+= exthdrlen
;
1095 transhdrlen
+= exthdrlen
;
1098 fl
= &inet
->cork
.fl
;
1099 if (inet
->cork
.flags
& IPCORK_OPT
)
1103 mtu
= inet
->cork
.fragsize
;
1106 hh_len
= LL_RESERVED_SPACE(rt
->u
.dst
.dev
);
1108 fragheaderlen
= sizeof(struct ipv6hdr
) + rt
->u
.dst
.nfheader_len
+ (opt
? opt
->opt_nflen
: 0);
1109 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
- sizeof(struct frag_hdr
);
1111 if (mtu
<= sizeof(struct ipv6hdr
) + IPV6_MAXPLEN
) {
1112 if (inet
->cork
.length
+ length
> sizeof(struct ipv6hdr
) + IPV6_MAXPLEN
- fragheaderlen
) {
1113 ipv6_local_error(sk
, EMSGSIZE
, fl
, mtu
-exthdrlen
);
1119 * Let's try using as much space as possible.
1120 * Use MTU if total length of the message fits into the MTU.
1121 * Otherwise, we need to reserve fragment header and
1122 * fragment alignment (= 8-15 octects, in total).
1124 * Note that we may need to "move" the data from the tail of
1125 * of the buffer to the new fragment when we split
1128 * FIXME: It may be fragmented into multiple chunks
1129 * at once if non-fragmentable extension headers
1134 inet
->cork
.length
+= length
;
1135 if (((length
> mtu
) && (sk
->sk_protocol
== IPPROTO_UDP
)) &&
1136 (rt
->u
.dst
.dev
->features
& NETIF_F_UFO
)) {
1138 err
= ip6_ufo_append_data(sk
, getfrag
, from
, length
, hh_len
,
1139 fragheaderlen
, transhdrlen
, mtu
,
1146 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
)
1149 while (length
> 0) {
1150 /* Check if the remaining data fits into current packet. */
1151 copy
= (inet
->cork
.length
<= mtu
&& !(inet
->cork
.flags
& IPCORK_ALLFRAG
) ? mtu
: maxfraglen
) - skb
->len
;
1153 copy
= maxfraglen
- skb
->len
;
1157 unsigned int datalen
;
1158 unsigned int fraglen
;
1159 unsigned int fraggap
;
1160 unsigned int alloclen
;
1161 struct sk_buff
*skb_prev
;
1165 /* There's no room in the current skb */
1167 fraggap
= skb_prev
->len
- maxfraglen
;
1172 * If remaining data exceeds the mtu,
1173 * we know we need more fragment(s).
1175 datalen
= length
+ fraggap
;
1176 if (datalen
> (inet
->cork
.length
<= mtu
&& !(inet
->cork
.flags
& IPCORK_ALLFRAG
) ? mtu
: maxfraglen
) - fragheaderlen
)
1177 datalen
= maxfraglen
- fragheaderlen
;
1179 fraglen
= datalen
+ fragheaderlen
;
1180 if ((flags
& MSG_MORE
) &&
1181 !(rt
->u
.dst
.dev
->features
&NETIF_F_SG
))
1184 alloclen
= datalen
+ fragheaderlen
;
1187 * The last fragment gets additional space at tail.
1188 * Note: we overallocate on fragments with MSG_MODE
1189 * because we have no idea if we're the last one.
1191 if (datalen
== length
+ fraggap
)
1192 alloclen
+= rt
->u
.dst
.trailer_len
;
1195 * We just reserve space for fragment header.
1196 * Note: this may be overallocation if the message
1197 * (without MSG_MORE) fits into the MTU.
1199 alloclen
+= sizeof(struct frag_hdr
);
1202 skb
= sock_alloc_send_skb(sk
,
1204 (flags
& MSG_DONTWAIT
), &err
);
1207 if (atomic_read(&sk
->sk_wmem_alloc
) <=
1209 skb
= sock_wmalloc(sk
,
1210 alloclen
+ hh_len
, 1,
1212 if (unlikely(skb
== NULL
))
1218 * Fill in the control structures
1220 skb
->ip_summed
= csummode
;
1222 /* reserve for fragmentation */
1223 skb_reserve(skb
, hh_len
+sizeof(struct frag_hdr
));
1226 * Find where to start putting bytes
1228 data
= skb_put(skb
, fraglen
);
1229 skb_set_network_header(skb
, exthdrlen
);
1230 data
+= fragheaderlen
;
1231 skb
->transport_header
= (skb
->network_header
+
1234 skb
->csum
= skb_copy_and_csum_bits(
1235 skb_prev
, maxfraglen
,
1236 data
+ transhdrlen
, fraggap
, 0);
1237 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
1240 pskb_trim_unique(skb_prev
, maxfraglen
);
1242 copy
= datalen
- transhdrlen
- fraggap
;
1247 } else if (copy
> 0 && getfrag(from
, data
+ transhdrlen
, offset
, copy
, fraggap
, skb
) < 0) {
1254 length
-= datalen
- fraggap
;
1257 csummode
= CHECKSUM_NONE
;
1260 * Put the packet on the pending queue
1262 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1269 if (!(rt
->u
.dst
.dev
->features
&NETIF_F_SG
)) {
1273 if (getfrag(from
, skb_put(skb
, copy
),
1274 offset
, copy
, off
, skb
) < 0) {
1275 __skb_trim(skb
, off
);
1280 int i
= skb_shinfo(skb
)->nr_frags
;
1281 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
-1];
1282 struct page
*page
= sk
->sk_sndmsg_page
;
1283 int off
= sk
->sk_sndmsg_off
;
1286 if (page
&& (left
= PAGE_SIZE
- off
) > 0) {
1289 if (page
!= frag
->page
) {
1290 if (i
== MAX_SKB_FRAGS
) {
1295 skb_fill_page_desc(skb
, i
, page
, sk
->sk_sndmsg_off
, 0);
1296 frag
= &skb_shinfo(skb
)->frags
[i
];
1298 } else if(i
< MAX_SKB_FRAGS
) {
1299 if (copy
> PAGE_SIZE
)
1301 page
= alloc_pages(sk
->sk_allocation
, 0);
1306 sk
->sk_sndmsg_page
= page
;
1307 sk
->sk_sndmsg_off
= 0;
1309 skb_fill_page_desc(skb
, i
, page
, 0, 0);
1310 frag
= &skb_shinfo(skb
)->frags
[i
];
1311 skb
->truesize
+= PAGE_SIZE
;
1312 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
1317 if (getfrag(from
, page_address(frag
->page
)+frag
->page_offset
+frag
->size
, offset
, copy
, skb
->len
, skb
) < 0) {
1321 sk
->sk_sndmsg_off
+= copy
;
1324 skb
->data_len
+= copy
;
1331 inet
->cork
.length
-= length
;
1332 IP6_INC_STATS(rt
->rt6i_idev
, IPSTATS_MIB_OUTDISCARDS
);
1336 int ip6_push_pending_frames(struct sock
*sk
)
1338 struct sk_buff
*skb
, *tmp_skb
;
1339 struct sk_buff
**tail_skb
;
1340 struct in6_addr final_dst_buf
, *final_dst
= &final_dst_buf
;
1341 struct inet_sock
*inet
= inet_sk(sk
);
1342 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1343 struct ipv6hdr
*hdr
;
1344 struct ipv6_txoptions
*opt
= np
->cork
.opt
;
1345 struct rt6_info
*rt
= np
->cork
.rt
;
1346 struct flowi
*fl
= &inet
->cork
.fl
;
1347 unsigned char proto
= fl
->proto
;
1350 if ((skb
= __skb_dequeue(&sk
->sk_write_queue
)) == NULL
)
1352 tail_skb
= &(skb_shinfo(skb
)->frag_list
);
1354 /* move skb->data to ip header from ext header */
1355 if (skb
->data
< skb_network_header(skb
))
1356 __skb_pull(skb
, skb_network_offset(skb
));
1357 while ((tmp_skb
= __skb_dequeue(&sk
->sk_write_queue
)) != NULL
) {
1358 __skb_pull(tmp_skb
, skb_network_header_len(skb
));
1359 *tail_skb
= tmp_skb
;
1360 tail_skb
= &(tmp_skb
->next
);
1361 skb
->len
+= tmp_skb
->len
;
1362 skb
->data_len
+= tmp_skb
->len
;
1363 skb
->truesize
+= tmp_skb
->truesize
;
1364 __sock_put(tmp_skb
->sk
);
1365 tmp_skb
->destructor
= NULL
;
1369 ipv6_addr_copy(final_dst
, &fl
->fl6_dst
);
1370 __skb_pull(skb
, skb_network_header_len(skb
));
1371 if (opt
&& opt
->opt_flen
)
1372 ipv6_push_frag_opts(skb
, opt
, &proto
);
1373 if (opt
&& opt
->opt_nflen
)
1374 ipv6_push_nfrag_opts(skb
, opt
, &proto
, &final_dst
);
1376 skb_push(skb
, sizeof(struct ipv6hdr
));
1377 skb_reset_network_header(skb
);
1378 hdr
= ipv6_hdr(skb
);
1380 *(__be32
*)hdr
= fl
->fl6_flowlabel
|
1381 htonl(0x60000000 | ((int)np
->cork
.tclass
<< 20));
1383 if (skb
->len
<= sizeof(struct ipv6hdr
) + IPV6_MAXPLEN
)
1384 hdr
->payload_len
= htons(skb
->len
- sizeof(struct ipv6hdr
));
1386 hdr
->payload_len
= 0;
1387 hdr
->hop_limit
= np
->cork
.hop_limit
;
1388 hdr
->nexthdr
= proto
;
1389 ipv6_addr_copy(&hdr
->saddr
, &fl
->fl6_src
);
1390 ipv6_addr_copy(&hdr
->daddr
, final_dst
);
1392 skb
->priority
= sk
->sk_priority
;
1394 skb
->dst
= dst_clone(&rt
->u
.dst
);
1395 IP6_INC_STATS(rt
->rt6i_idev
, IPSTATS_MIB_OUTREQUESTS
);
1396 err
= NF_HOOK(PF_INET6
, NF_IP6_LOCAL_OUT
, skb
, NULL
, skb
->dst
->dev
, dst_output
);
1399 err
= np
->recverr
? net_xmit_errno(err
) : 0;
1405 inet
->cork
.flags
&= ~IPCORK_OPT
;
1406 kfree(np
->cork
.opt
);
1407 np
->cork
.opt
= NULL
;
1409 dst_release(&np
->cork
.rt
->u
.dst
);
1411 inet
->cork
.flags
&= ~IPCORK_ALLFRAG
;
1413 memset(&inet
->cork
.fl
, 0, sizeof(inet
->cork
.fl
));
1419 void ip6_flush_pending_frames(struct sock
*sk
)
1421 struct inet_sock
*inet
= inet_sk(sk
);
1422 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1423 struct sk_buff
*skb
;
1425 while ((skb
= __skb_dequeue_tail(&sk
->sk_write_queue
)) != NULL
) {
1426 IP6_INC_STATS(ip6_dst_idev(skb
->dst
),
1427 IPSTATS_MIB_OUTDISCARDS
);
1431 inet
->cork
.flags
&= ~IPCORK_OPT
;
1433 kfree(np
->cork
.opt
);
1434 np
->cork
.opt
= NULL
;
1436 dst_release(&np
->cork
.rt
->u
.dst
);
1438 inet
->cork
.flags
&= ~IPCORK_ALLFRAG
;
1440 memset(&inet
->cork
.fl
, 0, sizeof(inet
->cork
.fl
));