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
, newskb
->nh
.raw
- newskb
->data
);
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(&skb
->nh
.ipv6h
->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
, &skb
->nh
.ipv6h
->daddr
,
116 &skb
->nh
.ipv6h
->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 (skb
->nh
.ipv6h
->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 hdr
= skb
->nh
.ipv6h
= (struct ipv6hdr
*)skb_push(skb
, sizeof(struct ipv6hdr
));
197 * Fill in the IPv6 header
202 hlimit
= np
->hop_limit
;
204 hlimit
= dst_metric(dst
, RTAX_HOPLIMIT
);
206 hlimit
= ipv6_get_hoplimit(dst
->dev
);
214 *(__be32
*)hdr
= htonl(0x60000000 | (tclass
<< 20)) | fl
->fl6_flowlabel
;
216 hdr
->payload_len
= htons(seg_len
);
217 hdr
->nexthdr
= proto
;
218 hdr
->hop_limit
= hlimit
;
220 ipv6_addr_copy(&hdr
->saddr
, &fl
->fl6_src
);
221 ipv6_addr_copy(&hdr
->daddr
, first_hop
);
223 skb
->priority
= sk
->sk_priority
;
226 if ((skb
->len
<= mtu
) || ipfragok
|| skb_is_gso(skb
)) {
227 IP6_INC_STATS(ip6_dst_idev(skb
->dst
),
228 IPSTATS_MIB_OUTREQUESTS
);
229 return NF_HOOK(PF_INET6
, NF_IP6_LOCAL_OUT
, skb
, NULL
, dst
->dev
,
234 printk(KERN_DEBUG
"IPv6: sending pkt_too_big to self\n");
236 icmpv6_send(skb
, ICMPV6_PKT_TOOBIG
, 0, mtu
, skb
->dev
);
237 IP6_INC_STATS(ip6_dst_idev(skb
->dst
), IPSTATS_MIB_FRAGFAILS
);
242 EXPORT_SYMBOL(ip6_xmit
);
245 * To avoid extra problems ND packets are send through this
246 * routine. It's code duplication but I really want to avoid
247 * extra checks since ipv6_build_header is used by TCP (which
248 * is for us performance critical)
251 int ip6_nd_hdr(struct sock
*sk
, struct sk_buff
*skb
, struct net_device
*dev
,
252 struct in6_addr
*saddr
, struct in6_addr
*daddr
,
255 struct ipv6_pinfo
*np
= inet6_sk(sk
);
259 skb
->protocol
= htons(ETH_P_IPV6
);
262 totlen
= len
+ sizeof(struct ipv6hdr
);
264 hdr
= (struct ipv6hdr
*) skb_put(skb
, sizeof(struct ipv6hdr
));
267 *(__be32
*)hdr
= htonl(0x60000000);
269 hdr
->payload_len
= htons(len
);
270 hdr
->nexthdr
= proto
;
271 hdr
->hop_limit
= np
->hop_limit
;
273 ipv6_addr_copy(&hdr
->saddr
, saddr
);
274 ipv6_addr_copy(&hdr
->daddr
, daddr
);
279 static int ip6_call_ra_chain(struct sk_buff
*skb
, int sel
)
281 struct ip6_ra_chain
*ra
;
282 struct sock
*last
= NULL
;
284 read_lock(&ip6_ra_lock
);
285 for (ra
= ip6_ra_chain
; ra
; ra
= ra
->next
) {
286 struct sock
*sk
= ra
->sk
;
287 if (sk
&& ra
->sel
== sel
&&
288 (!sk
->sk_bound_dev_if
||
289 sk
->sk_bound_dev_if
== skb
->dev
->ifindex
)) {
291 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
293 rawv6_rcv(last
, skb2
);
300 rawv6_rcv(last
, skb
);
301 read_unlock(&ip6_ra_lock
);
304 read_unlock(&ip6_ra_lock
);
308 static int ip6_forward_proxy_check(struct sk_buff
*skb
)
310 struct ipv6hdr
*hdr
= skb
->nh
.ipv6h
;
311 u8 nexthdr
= hdr
->nexthdr
;
314 if (ipv6_ext_hdr(nexthdr
)) {
315 offset
= ipv6_skip_exthdr(skb
, sizeof(*hdr
), &nexthdr
);
319 offset
= sizeof(struct ipv6hdr
);
321 if (nexthdr
== IPPROTO_ICMPV6
) {
322 struct icmp6hdr
*icmp6
;
324 if (!pskb_may_pull(skb
, skb
->nh
.raw
+ offset
+ 1 - skb
->data
))
327 icmp6
= (struct icmp6hdr
*)(skb
->nh
.raw
+ offset
);
329 switch (icmp6
->icmp6_type
) {
330 case NDISC_ROUTER_SOLICITATION
:
331 case NDISC_ROUTER_ADVERTISEMENT
:
332 case NDISC_NEIGHBOUR_SOLICITATION
:
333 case NDISC_NEIGHBOUR_ADVERTISEMENT
:
335 /* For reaction involving unicast neighbor discovery
336 * message destined to the proxied address, pass it to
346 * The proxying router can't forward traffic sent to a link-local
347 * address, so signal the sender and discard the packet. This
348 * behavior is clarified by the MIPv6 specification.
350 if (ipv6_addr_type(&hdr
->daddr
) & IPV6_ADDR_LINKLOCAL
) {
351 dst_link_failure(skb
);
358 static inline int ip6_forward_finish(struct sk_buff
*skb
)
360 return dst_output(skb
);
363 int ip6_forward(struct sk_buff
*skb
)
365 struct dst_entry
*dst
= skb
->dst
;
366 struct ipv6hdr
*hdr
= skb
->nh
.ipv6h
;
367 struct inet6_skb_parm
*opt
= IP6CB(skb
);
369 if (ipv6_devconf
.forwarding
== 0)
372 if (!xfrm6_policy_check(NULL
, XFRM_POLICY_FWD
, skb
)) {
373 IP6_INC_STATS(ip6_dst_idev(dst
), IPSTATS_MIB_INDISCARDS
);
377 skb
->ip_summed
= CHECKSUM_NONE
;
380 * We DO NOT make any processing on
381 * RA packets, pushing them to user level AS IS
382 * without ane WARRANTY that application will be able
383 * to interpret them. The reason is that we
384 * cannot make anything clever here.
386 * We are not end-node, so that if packet contains
387 * AH/ESP, we cannot make anything.
388 * Defragmentation also would be mistake, RA packets
389 * cannot be fragmented, because there is no warranty
390 * that different fragments will go along one path. --ANK
393 u8
*ptr
= skb
->nh
.raw
+ opt
->ra
;
394 if (ip6_call_ra_chain(skb
, (ptr
[2]<<8) + ptr
[3]))
399 * check and decrement ttl
401 if (hdr
->hop_limit
<= 1) {
402 /* Force OUTPUT device used as source address */
404 icmpv6_send(skb
, ICMPV6_TIME_EXCEED
, ICMPV6_EXC_HOPLIMIT
,
406 IP6_INC_STATS_BH(ip6_dst_idev(dst
), IPSTATS_MIB_INHDRERRORS
);
412 /* XXX: idev->cnf.proxy_ndp? */
413 if (ipv6_devconf
.proxy_ndp
&&
414 pneigh_lookup(&nd_tbl
, &hdr
->daddr
, skb
->dev
, 0)) {
415 int proxied
= ip6_forward_proxy_check(skb
);
417 return ip6_input(skb
);
418 else if (proxied
< 0) {
419 IP6_INC_STATS(ip6_dst_idev(dst
), IPSTATS_MIB_INDISCARDS
);
424 if (!xfrm6_route_forward(skb
)) {
425 IP6_INC_STATS(ip6_dst_idev(dst
), IPSTATS_MIB_INDISCARDS
);
430 /* IPv6 specs say nothing about it, but it is clear that we cannot
431 send redirects to source routed frames.
433 if (skb
->dev
== dst
->dev
&& dst
->neighbour
&& opt
->srcrt
== 0) {
434 struct in6_addr
*target
= NULL
;
436 struct neighbour
*n
= dst
->neighbour
;
439 * incoming and outgoing devices are the same
443 rt
= (struct rt6_info
*) dst
;
444 if ((rt
->rt6i_flags
& RTF_GATEWAY
))
445 target
= (struct in6_addr
*)&n
->primary_key
;
447 target
= &hdr
->daddr
;
449 /* Limit redirects both by destination (here)
450 and by source (inside ndisc_send_redirect)
452 if (xrlim_allow(dst
, 1*HZ
))
453 ndisc_send_redirect(skb
, n
, target
);
454 } else if (ipv6_addr_type(&hdr
->saddr
)&(IPV6_ADDR_MULTICAST
|IPV6_ADDR_LOOPBACK
455 |IPV6_ADDR_LINKLOCAL
)) {
456 /* This check is security critical. */
460 if (skb
->len
> dst_mtu(dst
)) {
461 /* Again, force OUTPUT device used as source address */
463 icmpv6_send(skb
, ICMPV6_PKT_TOOBIG
, 0, dst_mtu(dst
), skb
->dev
);
464 IP6_INC_STATS_BH(ip6_dst_idev(dst
), IPSTATS_MIB_INTOOBIGERRORS
);
465 IP6_INC_STATS_BH(ip6_dst_idev(dst
), IPSTATS_MIB_FRAGFAILS
);
470 if (skb_cow(skb
, dst
->dev
->hard_header_len
)) {
471 IP6_INC_STATS(ip6_dst_idev(dst
), IPSTATS_MIB_OUTDISCARDS
);
477 /* Mangling hops number delayed to point after skb COW */
481 IP6_INC_STATS_BH(ip6_dst_idev(dst
), IPSTATS_MIB_OUTFORWDATAGRAMS
);
482 return NF_HOOK(PF_INET6
,NF_IP6_FORWARD
, skb
, skb
->dev
, dst
->dev
, ip6_forward_finish
);
485 IP6_INC_STATS_BH(ip6_dst_idev(dst
), IPSTATS_MIB_INADDRERRORS
);
491 static void ip6_copy_metadata(struct sk_buff
*to
, struct sk_buff
*from
)
493 to
->pkt_type
= from
->pkt_type
;
494 to
->priority
= from
->priority
;
495 to
->protocol
= from
->protocol
;
496 dst_release(to
->dst
);
497 to
->dst
= dst_clone(from
->dst
);
499 to
->mark
= from
->mark
;
501 #ifdef CONFIG_NET_SCHED
502 to
->tc_index
= from
->tc_index
;
504 #ifdef CONFIG_NETFILTER
505 /* Connection association is same as pre-frag packet */
506 nf_conntrack_put(to
->nfct
);
507 to
->nfct
= from
->nfct
;
508 nf_conntrack_get(to
->nfct
);
509 to
->nfctinfo
= from
->nfctinfo
;
510 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
511 nf_conntrack_put_reasm(to
->nfct_reasm
);
512 to
->nfct_reasm
= from
->nfct_reasm
;
513 nf_conntrack_get_reasm(to
->nfct_reasm
);
515 #ifdef CONFIG_BRIDGE_NETFILTER
516 nf_bridge_put(to
->nf_bridge
);
517 to
->nf_bridge
= from
->nf_bridge
;
518 nf_bridge_get(to
->nf_bridge
);
521 skb_copy_secmark(to
, from
);
524 int ip6_find_1stfragopt(struct sk_buff
*skb
, u8
**nexthdr
)
526 u16 offset
= sizeof(struct ipv6hdr
);
527 struct ipv6_opt_hdr
*exthdr
= (struct ipv6_opt_hdr
*)(skb
->nh
.ipv6h
+ 1);
528 unsigned int packet_len
= skb
->tail
- skb
->nh
.raw
;
530 *nexthdr
= &skb
->nh
.ipv6h
->nexthdr
;
532 while (offset
+ 1 <= packet_len
) {
538 case NEXTHDR_ROUTING
:
542 #ifdef CONFIG_IPV6_MIP6
543 if (ipv6_find_tlv(skb
, offset
, IPV6_TLV_HAO
) >= 0)
553 offset
+= ipv6_optlen(exthdr
);
554 *nexthdr
= &exthdr
->nexthdr
;
555 exthdr
= (struct ipv6_opt_hdr
*)(skb
->nh
.raw
+ offset
);
560 EXPORT_SYMBOL_GPL(ip6_find_1stfragopt
);
562 static int ip6_fragment(struct sk_buff
*skb
, int (*output
)(struct sk_buff
*))
564 struct net_device
*dev
;
565 struct sk_buff
*frag
;
566 struct rt6_info
*rt
= (struct rt6_info
*)skb
->dst
;
567 struct ipv6_pinfo
*np
= skb
->sk
? inet6_sk(skb
->sk
) : NULL
;
568 struct ipv6hdr
*tmp_hdr
;
570 unsigned int mtu
, hlen
, left
, len
;
572 int ptr
, offset
= 0, err
=0;
573 u8
*prevhdr
, nexthdr
= 0;
576 hlen
= ip6_find_1stfragopt(skb
, &prevhdr
);
579 mtu
= dst_mtu(&rt
->u
.dst
);
580 if (np
&& np
->frag_size
< mtu
) {
584 mtu
-= hlen
+ sizeof(struct frag_hdr
);
586 if (skb_shinfo(skb
)->frag_list
) {
587 int first_len
= skb_pagelen(skb
);
589 if (first_len
- hlen
> mtu
||
590 ((first_len
- hlen
) & 7) ||
594 for (frag
= skb_shinfo(skb
)->frag_list
; frag
; frag
= frag
->next
) {
595 /* Correct geometry. */
596 if (frag
->len
> mtu
||
597 ((frag
->len
& 7) && frag
->next
) ||
598 skb_headroom(frag
) < hlen
)
601 /* Partially cloned skb? */
602 if (skb_shared(frag
))
609 frag
->destructor
= sock_wfree
;
610 skb
->truesize
-= frag
->truesize
;
616 frag
= skb_shinfo(skb
)->frag_list
;
617 skb_shinfo(skb
)->frag_list
= NULL
;
620 *prevhdr
= NEXTHDR_FRAGMENT
;
621 tmp_hdr
= kmemdup(skb
->nh
.raw
, hlen
, GFP_ATOMIC
);
623 IP6_INC_STATS(ip6_dst_idev(skb
->dst
), IPSTATS_MIB_FRAGFAILS
);
627 __skb_pull(skb
, hlen
);
628 fh
= (struct frag_hdr
*)__skb_push(skb
, sizeof(struct frag_hdr
));
629 skb
->nh
.raw
= __skb_push(skb
, hlen
);
630 memcpy(skb
->nh
.raw
, tmp_hdr
, hlen
);
632 ipv6_select_ident(skb
, fh
);
633 fh
->nexthdr
= nexthdr
;
635 fh
->frag_off
= htons(IP6_MF
);
636 frag_id
= fh
->identification
;
638 first_len
= skb_pagelen(skb
);
639 skb
->data_len
= first_len
- skb_headlen(skb
);
640 skb
->len
= first_len
;
641 skb
->nh
.ipv6h
->payload_len
= htons(first_len
- sizeof(struct ipv6hdr
));
643 dst_hold(&rt
->u
.dst
);
646 /* Prepare header of the next frame,
647 * before previous one went down. */
649 frag
->ip_summed
= CHECKSUM_NONE
;
650 frag
->h
.raw
= frag
->data
;
651 fh
= (struct frag_hdr
*)__skb_push(frag
, sizeof(struct frag_hdr
));
652 frag
->nh
.raw
= __skb_push(frag
, hlen
);
653 memcpy(frag
->nh
.raw
, tmp_hdr
, hlen
);
654 offset
+= skb
->len
- hlen
- sizeof(struct frag_hdr
);
655 fh
->nexthdr
= nexthdr
;
657 fh
->frag_off
= htons(offset
);
658 if (frag
->next
!= NULL
)
659 fh
->frag_off
|= htons(IP6_MF
);
660 fh
->identification
= frag_id
;
661 frag
->nh
.ipv6h
->payload_len
= htons(frag
->len
- sizeof(struct ipv6hdr
));
662 ip6_copy_metadata(frag
, skb
);
667 IP6_INC_STATS(ip6_dst_idev(&rt
->u
.dst
), IPSTATS_MIB_FRAGCREATES
);
680 IP6_INC_STATS(ip6_dst_idev(&rt
->u
.dst
), IPSTATS_MIB_FRAGOKS
);
681 dst_release(&rt
->u
.dst
);
691 IP6_INC_STATS(ip6_dst_idev(&rt
->u
.dst
), IPSTATS_MIB_FRAGFAILS
);
692 dst_release(&rt
->u
.dst
);
697 left
= skb
->len
- hlen
; /* Space per frame */
698 ptr
= hlen
; /* Where to start from */
701 * Fragment the datagram.
704 *prevhdr
= NEXTHDR_FRAGMENT
;
707 * Keep copying data until we run out.
711 /* IF: it doesn't fit, use 'mtu' - the data space left */
714 /* IF: we are not sending upto and including the packet end
715 then align the next start on an eight byte boundary */
723 if ((frag
= alloc_skb(len
+hlen
+sizeof(struct frag_hdr
)+LL_RESERVED_SPACE(rt
->u
.dst
.dev
), GFP_ATOMIC
)) == NULL
) {
724 NETDEBUG(KERN_INFO
"IPv6: frag: no memory for new fragment!\n");
725 IP6_INC_STATS(ip6_dst_idev(skb
->dst
),
726 IPSTATS_MIB_FRAGFAILS
);
732 * Set up data on packet
735 ip6_copy_metadata(frag
, skb
);
736 skb_reserve(frag
, LL_RESERVED_SPACE(rt
->u
.dst
.dev
));
737 skb_put(frag
, len
+ hlen
+ sizeof(struct frag_hdr
));
738 skb_reset_network_header(frag
);
739 fh
= (struct frag_hdr
*)(frag
->data
+ hlen
);
740 frag
->h
.raw
= frag
->data
+ hlen
+ sizeof(struct frag_hdr
);
743 * Charge the memory for the fragment to any owner
747 skb_set_owner_w(frag
, skb
->sk
);
750 * Copy the packet header into the new buffer.
752 memcpy(frag
->nh
.raw
, skb
->data
, hlen
);
755 * Build fragment header.
757 fh
->nexthdr
= nexthdr
;
760 ipv6_select_ident(skb
, fh
);
761 frag_id
= fh
->identification
;
763 fh
->identification
= frag_id
;
766 * Copy a block of the IP datagram.
768 if (skb_copy_bits(skb
, ptr
, frag
->h
.raw
, len
))
772 fh
->frag_off
= htons(offset
);
774 fh
->frag_off
|= htons(IP6_MF
);
775 frag
->nh
.ipv6h
->payload_len
= htons(frag
->len
- sizeof(struct ipv6hdr
));
781 * Put this fragment into the sending queue.
787 IP6_INC_STATS(ip6_dst_idev(skb
->dst
), IPSTATS_MIB_FRAGCREATES
);
789 IP6_INC_STATS(ip6_dst_idev(skb
->dst
),
790 IPSTATS_MIB_FRAGOKS
);
795 IP6_INC_STATS(ip6_dst_idev(skb
->dst
),
796 IPSTATS_MIB_FRAGFAILS
);
801 static inline int ip6_rt_check(struct rt6key
*rt_key
,
802 struct in6_addr
*fl_addr
,
803 struct in6_addr
*addr_cache
)
805 return ((rt_key
->plen
!= 128 || !ipv6_addr_equal(fl_addr
, &rt_key
->addr
)) &&
806 (addr_cache
== NULL
|| !ipv6_addr_equal(fl_addr
, addr_cache
)));
809 static struct dst_entry
*ip6_sk_dst_check(struct sock
*sk
,
810 struct dst_entry
*dst
,
813 struct ipv6_pinfo
*np
= inet6_sk(sk
);
814 struct rt6_info
*rt
= (struct rt6_info
*)dst
;
819 /* Yes, checking route validity in not connected
820 * case is not very simple. Take into account,
821 * that we do not support routing by source, TOS,
822 * and MSG_DONTROUTE --ANK (980726)
824 * 1. ip6_rt_check(): If route was host route,
825 * check that cached destination is current.
826 * If it is network route, we still may
827 * check its validity using saved pointer
828 * to the last used address: daddr_cache.
829 * We do not want to save whole address now,
830 * (because main consumer of this service
831 * is tcp, which has not this problem),
832 * so that the last trick works only on connected
834 * 2. oif also should be the same.
836 if (ip6_rt_check(&rt
->rt6i_dst
, &fl
->fl6_dst
, np
->daddr_cache
) ||
837 #ifdef CONFIG_IPV6_SUBTREES
838 ip6_rt_check(&rt
->rt6i_src
, &fl
->fl6_src
, np
->saddr_cache
) ||
840 (fl
->oif
&& fl
->oif
!= dst
->dev
->ifindex
)) {
849 static int ip6_dst_lookup_tail(struct sock
*sk
,
850 struct dst_entry
**dst
, struct flowi
*fl
)
855 *dst
= ip6_route_output(sk
, fl
);
857 if ((err
= (*dst
)->error
))
858 goto out_err_release
;
860 if (ipv6_addr_any(&fl
->fl6_src
)) {
861 err
= ipv6_get_saddr(*dst
, &fl
->fl6_dst
, &fl
->fl6_src
);
863 goto out_err_release
;
866 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
868 * Here if the dst entry we've looked up
869 * has a neighbour entry that is in the INCOMPLETE
870 * state and the src address from the flow is
871 * marked as OPTIMISTIC, we release the found
872 * dst entry and replace it instead with the
873 * dst entry of the nexthop router
875 if (!((*dst
)->neighbour
->nud_state
& NUD_VALID
)) {
876 struct inet6_ifaddr
*ifp
;
880 ifp
= ipv6_get_ifaddr(&fl
->fl6_src
, (*dst
)->dev
, 1);
882 redirect
= (ifp
&& ifp
->flags
& IFA_F_OPTIMISTIC
);
888 * We need to get the dst entry for the
889 * default router instead
892 memcpy(&fl_gw
, fl
, sizeof(struct flowi
));
893 memset(&fl_gw
.fl6_dst
, 0, sizeof(struct in6_addr
));
894 *dst
= ip6_route_output(sk
, &fl_gw
);
895 if ((err
= (*dst
)->error
))
896 goto out_err_release
;
910 * ip6_dst_lookup - perform route lookup on flow
911 * @sk: socket which provides route info
912 * @dst: pointer to dst_entry * for result
913 * @fl: flow to lookup
915 * This function performs a route lookup on the given flow.
917 * It returns zero on success, or a standard errno code on error.
919 int ip6_dst_lookup(struct sock
*sk
, struct dst_entry
**dst
, struct flowi
*fl
)
922 return ip6_dst_lookup_tail(sk
, dst
, fl
);
924 EXPORT_SYMBOL_GPL(ip6_dst_lookup
);
927 * ip6_sk_dst_lookup - perform socket cached route lookup on flow
928 * @sk: socket which provides the dst cache and route info
929 * @dst: pointer to dst_entry * for result
930 * @fl: flow to lookup
932 * This function performs a route lookup on the given flow with the
933 * possibility of using the cached route in the socket if it is valid.
934 * It will take the socket dst lock when operating on the dst cache.
935 * As a result, this function can only be used in process context.
937 * It returns zero on success, or a standard errno code on error.
939 int ip6_sk_dst_lookup(struct sock
*sk
, struct dst_entry
**dst
, struct flowi
*fl
)
943 *dst
= sk_dst_check(sk
, inet6_sk(sk
)->dst_cookie
);
944 *dst
= ip6_sk_dst_check(sk
, *dst
, fl
);
947 return ip6_dst_lookup_tail(sk
, dst
, fl
);
949 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup
);
951 static inline int ip6_ufo_append_data(struct sock
*sk
,
952 int getfrag(void *from
, char *to
, int offset
, int len
,
953 int odd
, struct sk_buff
*skb
),
954 void *from
, int length
, int hh_len
, int fragheaderlen
,
955 int transhdrlen
, int mtu
,unsigned int flags
)
961 /* There is support for UDP large send offload by network
962 * device, so create one single skb packet containing complete
965 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
) {
966 skb
= sock_alloc_send_skb(sk
,
967 hh_len
+ fragheaderlen
+ transhdrlen
+ 20,
968 (flags
& MSG_DONTWAIT
), &err
);
972 /* reserve space for Hardware header */
973 skb_reserve(skb
, hh_len
);
975 /* create space for UDP/IP header */
976 skb_put(skb
,fragheaderlen
+ transhdrlen
);
978 /* initialize network header pointer */
979 skb_reset_network_header(skb
);
981 /* initialize protocol header pointer */
982 skb
->h
.raw
= skb
->data
+ fragheaderlen
;
984 skb
->ip_summed
= CHECKSUM_PARTIAL
;
986 sk
->sk_sndmsg_off
= 0;
989 err
= skb_append_datato_frags(sk
,skb
, getfrag
, from
,
990 (length
- transhdrlen
));
992 struct frag_hdr fhdr
;
994 /* specify the length of each IP datagram fragment*/
995 skb_shinfo(skb
)->gso_size
= mtu
- fragheaderlen
-
996 sizeof(struct frag_hdr
);
997 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
998 ipv6_select_ident(skb
, &fhdr
);
999 skb_shinfo(skb
)->ip6_frag_id
= fhdr
.identification
;
1000 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1004 /* There is not enough support do UPD LSO,
1005 * so follow normal path
1012 int ip6_append_data(struct sock
*sk
, int getfrag(void *from
, char *to
,
1013 int offset
, int len
, int odd
, struct sk_buff
*skb
),
1014 void *from
, int length
, int transhdrlen
,
1015 int hlimit
, int tclass
, struct ipv6_txoptions
*opt
, struct flowi
*fl
,
1016 struct rt6_info
*rt
, unsigned int flags
)
1018 struct inet_sock
*inet
= inet_sk(sk
);
1019 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1020 struct sk_buff
*skb
;
1021 unsigned int maxfraglen
, fragheaderlen
;
1028 int csummode
= CHECKSUM_NONE
;
1030 if (flags
&MSG_PROBE
)
1032 if (skb_queue_empty(&sk
->sk_write_queue
)) {
1037 if (np
->cork
.opt
== NULL
) {
1038 np
->cork
.opt
= kmalloc(opt
->tot_len
,
1040 if (unlikely(np
->cork
.opt
== NULL
))
1042 } else if (np
->cork
.opt
->tot_len
< opt
->tot_len
) {
1043 printk(KERN_DEBUG
"ip6_append_data: invalid option length\n");
1046 memcpy(np
->cork
.opt
, opt
, opt
->tot_len
);
1047 inet
->cork
.flags
|= IPCORK_OPT
;
1048 /* need source address above miyazawa*/
1050 dst_hold(&rt
->u
.dst
);
1052 inet
->cork
.fl
= *fl
;
1053 np
->cork
.hop_limit
= hlimit
;
1054 np
->cork
.tclass
= tclass
;
1055 mtu
= dst_mtu(rt
->u
.dst
.path
);
1056 if (np
->frag_size
< mtu
) {
1058 mtu
= np
->frag_size
;
1060 inet
->cork
.fragsize
= mtu
;
1061 if (dst_allfrag(rt
->u
.dst
.path
))
1062 inet
->cork
.flags
|= IPCORK_ALLFRAG
;
1063 inet
->cork
.length
= 0;
1064 sk
->sk_sndmsg_page
= NULL
;
1065 sk
->sk_sndmsg_off
= 0;
1066 exthdrlen
= rt
->u
.dst
.header_len
+ (opt
? opt
->opt_flen
: 0);
1067 length
+= exthdrlen
;
1068 transhdrlen
+= exthdrlen
;
1071 fl
= &inet
->cork
.fl
;
1072 if (inet
->cork
.flags
& IPCORK_OPT
)
1076 mtu
= inet
->cork
.fragsize
;
1079 hh_len
= LL_RESERVED_SPACE(rt
->u
.dst
.dev
);
1081 fragheaderlen
= sizeof(struct ipv6hdr
) + rt
->u
.dst
.nfheader_len
+ (opt
? opt
->opt_nflen
: 0);
1082 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
- sizeof(struct frag_hdr
);
1084 if (mtu
<= sizeof(struct ipv6hdr
) + IPV6_MAXPLEN
) {
1085 if (inet
->cork
.length
+ length
> sizeof(struct ipv6hdr
) + IPV6_MAXPLEN
- fragheaderlen
) {
1086 ipv6_local_error(sk
, EMSGSIZE
, fl
, mtu
-exthdrlen
);
1092 * Let's try using as much space as possible.
1093 * Use MTU if total length of the message fits into the MTU.
1094 * Otherwise, we need to reserve fragment header and
1095 * fragment alignment (= 8-15 octects, in total).
1097 * Note that we may need to "move" the data from the tail of
1098 * of the buffer to the new fragment when we split
1101 * FIXME: It may be fragmented into multiple chunks
1102 * at once if non-fragmentable extension headers
1107 inet
->cork
.length
+= length
;
1108 if (((length
> mtu
) && (sk
->sk_protocol
== IPPROTO_UDP
)) &&
1109 (rt
->u
.dst
.dev
->features
& NETIF_F_UFO
)) {
1111 err
= ip6_ufo_append_data(sk
, getfrag
, from
, length
, hh_len
,
1112 fragheaderlen
, transhdrlen
, mtu
,
1119 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
)
1122 while (length
> 0) {
1123 /* Check if the remaining data fits into current packet. */
1124 copy
= (inet
->cork
.length
<= mtu
&& !(inet
->cork
.flags
& IPCORK_ALLFRAG
) ? mtu
: maxfraglen
) - skb
->len
;
1126 copy
= maxfraglen
- skb
->len
;
1130 unsigned int datalen
;
1131 unsigned int fraglen
;
1132 unsigned int fraggap
;
1133 unsigned int alloclen
;
1134 struct sk_buff
*skb_prev
;
1138 /* There's no room in the current skb */
1140 fraggap
= skb_prev
->len
- maxfraglen
;
1145 * If remaining data exceeds the mtu,
1146 * we know we need more fragment(s).
1148 datalen
= length
+ fraggap
;
1149 if (datalen
> (inet
->cork
.length
<= mtu
&& !(inet
->cork
.flags
& IPCORK_ALLFRAG
) ? mtu
: maxfraglen
) - fragheaderlen
)
1150 datalen
= maxfraglen
- fragheaderlen
;
1152 fraglen
= datalen
+ fragheaderlen
;
1153 if ((flags
& MSG_MORE
) &&
1154 !(rt
->u
.dst
.dev
->features
&NETIF_F_SG
))
1157 alloclen
= datalen
+ fragheaderlen
;
1160 * The last fragment gets additional space at tail.
1161 * Note: we overallocate on fragments with MSG_MODE
1162 * because we have no idea if we're the last one.
1164 if (datalen
== length
+ fraggap
)
1165 alloclen
+= rt
->u
.dst
.trailer_len
;
1168 * We just reserve space for fragment header.
1169 * Note: this may be overallocation if the message
1170 * (without MSG_MORE) fits into the MTU.
1172 alloclen
+= sizeof(struct frag_hdr
);
1175 skb
= sock_alloc_send_skb(sk
,
1177 (flags
& MSG_DONTWAIT
), &err
);
1180 if (atomic_read(&sk
->sk_wmem_alloc
) <=
1182 skb
= sock_wmalloc(sk
,
1183 alloclen
+ hh_len
, 1,
1185 if (unlikely(skb
== NULL
))
1191 * Fill in the control structures
1193 skb
->ip_summed
= csummode
;
1195 /* reserve for fragmentation */
1196 skb_reserve(skb
, hh_len
+sizeof(struct frag_hdr
));
1199 * Find where to start putting bytes
1201 data
= skb_put(skb
, fraglen
);
1202 skb
->nh
.raw
= data
+ exthdrlen
;
1203 data
+= fragheaderlen
;
1204 skb
->h
.raw
= data
+ exthdrlen
;
1207 skb
->csum
= skb_copy_and_csum_bits(
1208 skb_prev
, maxfraglen
,
1209 data
+ transhdrlen
, fraggap
, 0);
1210 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
1213 pskb_trim_unique(skb_prev
, maxfraglen
);
1215 copy
= datalen
- transhdrlen
- fraggap
;
1220 } else if (copy
> 0 && getfrag(from
, data
+ transhdrlen
, offset
, copy
, fraggap
, skb
) < 0) {
1227 length
-= datalen
- fraggap
;
1230 csummode
= CHECKSUM_NONE
;
1233 * Put the packet on the pending queue
1235 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1242 if (!(rt
->u
.dst
.dev
->features
&NETIF_F_SG
)) {
1246 if (getfrag(from
, skb_put(skb
, copy
),
1247 offset
, copy
, off
, skb
) < 0) {
1248 __skb_trim(skb
, off
);
1253 int i
= skb_shinfo(skb
)->nr_frags
;
1254 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
-1];
1255 struct page
*page
= sk
->sk_sndmsg_page
;
1256 int off
= sk
->sk_sndmsg_off
;
1259 if (page
&& (left
= PAGE_SIZE
- off
) > 0) {
1262 if (page
!= frag
->page
) {
1263 if (i
== MAX_SKB_FRAGS
) {
1268 skb_fill_page_desc(skb
, i
, page
, sk
->sk_sndmsg_off
, 0);
1269 frag
= &skb_shinfo(skb
)->frags
[i
];
1271 } else if(i
< MAX_SKB_FRAGS
) {
1272 if (copy
> PAGE_SIZE
)
1274 page
= alloc_pages(sk
->sk_allocation
, 0);
1279 sk
->sk_sndmsg_page
= page
;
1280 sk
->sk_sndmsg_off
= 0;
1282 skb_fill_page_desc(skb
, i
, page
, 0, 0);
1283 frag
= &skb_shinfo(skb
)->frags
[i
];
1284 skb
->truesize
+= PAGE_SIZE
;
1285 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
1290 if (getfrag(from
, page_address(frag
->page
)+frag
->page_offset
+frag
->size
, offset
, copy
, skb
->len
, skb
) < 0) {
1294 sk
->sk_sndmsg_off
+= copy
;
1297 skb
->data_len
+= copy
;
1304 inet
->cork
.length
-= length
;
1305 IP6_INC_STATS(rt
->rt6i_idev
, IPSTATS_MIB_OUTDISCARDS
);
1309 int ip6_push_pending_frames(struct sock
*sk
)
1311 struct sk_buff
*skb
, *tmp_skb
;
1312 struct sk_buff
**tail_skb
;
1313 struct in6_addr final_dst_buf
, *final_dst
= &final_dst_buf
;
1314 struct inet_sock
*inet
= inet_sk(sk
);
1315 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1316 struct ipv6hdr
*hdr
;
1317 struct ipv6_txoptions
*opt
= np
->cork
.opt
;
1318 struct rt6_info
*rt
= np
->cork
.rt
;
1319 struct flowi
*fl
= &inet
->cork
.fl
;
1320 unsigned char proto
= fl
->proto
;
1323 if ((skb
= __skb_dequeue(&sk
->sk_write_queue
)) == NULL
)
1325 tail_skb
= &(skb_shinfo(skb
)->frag_list
);
1327 /* move skb->data to ip header from ext header */
1328 if (skb
->data
< skb
->nh
.raw
)
1329 __skb_pull(skb
, skb
->nh
.raw
- skb
->data
);
1330 while ((tmp_skb
= __skb_dequeue(&sk
->sk_write_queue
)) != NULL
) {
1331 __skb_pull(tmp_skb
, skb
->h
.raw
- skb
->nh
.raw
);
1332 *tail_skb
= tmp_skb
;
1333 tail_skb
= &(tmp_skb
->next
);
1334 skb
->len
+= tmp_skb
->len
;
1335 skb
->data_len
+= tmp_skb
->len
;
1336 skb
->truesize
+= tmp_skb
->truesize
;
1337 __sock_put(tmp_skb
->sk
);
1338 tmp_skb
->destructor
= NULL
;
1342 ipv6_addr_copy(final_dst
, &fl
->fl6_dst
);
1343 __skb_pull(skb
, skb
->h
.raw
- skb
->nh
.raw
);
1344 if (opt
&& opt
->opt_flen
)
1345 ipv6_push_frag_opts(skb
, opt
, &proto
);
1346 if (opt
&& opt
->opt_nflen
)
1347 ipv6_push_nfrag_opts(skb
, opt
, &proto
, &final_dst
);
1349 skb
->nh
.ipv6h
= hdr
= (struct ipv6hdr
*) skb_push(skb
, sizeof(struct ipv6hdr
));
1351 *(__be32
*)hdr
= fl
->fl6_flowlabel
|
1352 htonl(0x60000000 | ((int)np
->cork
.tclass
<< 20));
1354 if (skb
->len
<= sizeof(struct ipv6hdr
) + IPV6_MAXPLEN
)
1355 hdr
->payload_len
= htons(skb
->len
- sizeof(struct ipv6hdr
));
1357 hdr
->payload_len
= 0;
1358 hdr
->hop_limit
= np
->cork
.hop_limit
;
1359 hdr
->nexthdr
= proto
;
1360 ipv6_addr_copy(&hdr
->saddr
, &fl
->fl6_src
);
1361 ipv6_addr_copy(&hdr
->daddr
, final_dst
);
1363 skb
->priority
= sk
->sk_priority
;
1365 skb
->dst
= dst_clone(&rt
->u
.dst
);
1366 IP6_INC_STATS(rt
->rt6i_idev
, IPSTATS_MIB_OUTREQUESTS
);
1367 err
= NF_HOOK(PF_INET6
, NF_IP6_LOCAL_OUT
, skb
, NULL
, skb
->dst
->dev
, dst_output
);
1370 err
= np
->recverr
? net_xmit_errno(err
) : 0;
1376 inet
->cork
.flags
&= ~IPCORK_OPT
;
1377 kfree(np
->cork
.opt
);
1378 np
->cork
.opt
= NULL
;
1380 dst_release(&np
->cork
.rt
->u
.dst
);
1382 inet
->cork
.flags
&= ~IPCORK_ALLFRAG
;
1384 memset(&inet
->cork
.fl
, 0, sizeof(inet
->cork
.fl
));
1390 void ip6_flush_pending_frames(struct sock
*sk
)
1392 struct inet_sock
*inet
= inet_sk(sk
);
1393 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1394 struct sk_buff
*skb
;
1396 while ((skb
= __skb_dequeue_tail(&sk
->sk_write_queue
)) != NULL
) {
1397 IP6_INC_STATS(ip6_dst_idev(skb
->dst
),
1398 IPSTATS_MIB_OUTDISCARDS
);
1402 inet
->cork
.flags
&= ~IPCORK_OPT
;
1404 kfree(np
->cork
.opt
);
1405 np
->cork
.opt
= NULL
;
1407 dst_release(&np
->cork
.rt
->u
.dst
);
1409 inet
->cork
.flags
&= ~IPCORK_ALLFRAG
;
1411 memset(&inet
->cork
.fl
, 0, sizeof(inet
->cork
.fl
));