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 newskb
->mac
.raw
= newskb
->data
;
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
);
243 * To avoid extra problems ND packets are send through this
244 * routine. It's code duplication but I really want to avoid
245 * extra checks since ipv6_build_header is used by TCP (which
246 * is for us performance critical)
249 int ip6_nd_hdr(struct sock
*sk
, struct sk_buff
*skb
, struct net_device
*dev
,
250 struct in6_addr
*saddr
, struct in6_addr
*daddr
,
253 struct ipv6_pinfo
*np
= inet6_sk(sk
);
257 skb
->protocol
= htons(ETH_P_IPV6
);
260 totlen
= len
+ sizeof(struct ipv6hdr
);
262 hdr
= (struct ipv6hdr
*) skb_put(skb
, sizeof(struct ipv6hdr
));
265 *(__be32
*)hdr
= htonl(0x60000000);
267 hdr
->payload_len
= htons(len
);
268 hdr
->nexthdr
= proto
;
269 hdr
->hop_limit
= np
->hop_limit
;
271 ipv6_addr_copy(&hdr
->saddr
, saddr
);
272 ipv6_addr_copy(&hdr
->daddr
, daddr
);
277 static int ip6_call_ra_chain(struct sk_buff
*skb
, int sel
)
279 struct ip6_ra_chain
*ra
;
280 struct sock
*last
= NULL
;
282 read_lock(&ip6_ra_lock
);
283 for (ra
= ip6_ra_chain
; ra
; ra
= ra
->next
) {
284 struct sock
*sk
= ra
->sk
;
285 if (sk
&& ra
->sel
== sel
&&
286 (!sk
->sk_bound_dev_if
||
287 sk
->sk_bound_dev_if
== skb
->dev
->ifindex
)) {
289 struct sk_buff
*skb2
= skb_clone(skb
, GFP_ATOMIC
);
291 rawv6_rcv(last
, skb2
);
298 rawv6_rcv(last
, skb
);
299 read_unlock(&ip6_ra_lock
);
302 read_unlock(&ip6_ra_lock
);
306 static int ip6_forward_proxy_check(struct sk_buff
*skb
)
308 struct ipv6hdr
*hdr
= skb
->nh
.ipv6h
;
309 u8 nexthdr
= hdr
->nexthdr
;
312 if (ipv6_ext_hdr(nexthdr
)) {
313 offset
= ipv6_skip_exthdr(skb
, sizeof(*hdr
), &nexthdr
);
317 offset
= sizeof(struct ipv6hdr
);
319 if (nexthdr
== IPPROTO_ICMPV6
) {
320 struct icmp6hdr
*icmp6
;
322 if (!pskb_may_pull(skb
, skb
->nh
.raw
+ offset
+ 1 - skb
->data
))
325 icmp6
= (struct icmp6hdr
*)(skb
->nh
.raw
+ offset
);
327 switch (icmp6
->icmp6_type
) {
328 case NDISC_ROUTER_SOLICITATION
:
329 case NDISC_ROUTER_ADVERTISEMENT
:
330 case NDISC_NEIGHBOUR_SOLICITATION
:
331 case NDISC_NEIGHBOUR_ADVERTISEMENT
:
333 /* For reaction involving unicast neighbor discovery
334 * message destined to the proxied address, pass it to
344 * The proxying router can't forward traffic sent to a link-local
345 * address, so signal the sender and discard the packet. This
346 * behavior is clarified by the MIPv6 specification.
348 if (ipv6_addr_type(&hdr
->daddr
) & IPV6_ADDR_LINKLOCAL
) {
349 dst_link_failure(skb
);
356 static inline int ip6_forward_finish(struct sk_buff
*skb
)
358 return dst_output(skb
);
361 int ip6_forward(struct sk_buff
*skb
)
363 struct dst_entry
*dst
= skb
->dst
;
364 struct ipv6hdr
*hdr
= skb
->nh
.ipv6h
;
365 struct inet6_skb_parm
*opt
= IP6CB(skb
);
367 if (ipv6_devconf
.forwarding
== 0)
370 if (!xfrm6_policy_check(NULL
, XFRM_POLICY_FWD
, skb
)) {
371 IP6_INC_STATS(ip6_dst_idev(dst
), IPSTATS_MIB_INDISCARDS
);
375 skb
->ip_summed
= CHECKSUM_NONE
;
378 * We DO NOT make any processing on
379 * RA packets, pushing them to user level AS IS
380 * without ane WARRANTY that application will be able
381 * to interpret them. The reason is that we
382 * cannot make anything clever here.
384 * We are not end-node, so that if packet contains
385 * AH/ESP, we cannot make anything.
386 * Defragmentation also would be mistake, RA packets
387 * cannot be fragmented, because there is no warranty
388 * that different fragments will go along one path. --ANK
391 u8
*ptr
= skb
->nh
.raw
+ opt
->ra
;
392 if (ip6_call_ra_chain(skb
, (ptr
[2]<<8) + ptr
[3]))
397 * check and decrement ttl
399 if (hdr
->hop_limit
<= 1) {
400 /* Force OUTPUT device used as source address */
402 icmpv6_send(skb
, ICMPV6_TIME_EXCEED
, ICMPV6_EXC_HOPLIMIT
,
404 IP6_INC_STATS_BH(ip6_dst_idev(dst
), IPSTATS_MIB_INHDRERRORS
);
410 /* XXX: idev->cnf.proxy_ndp? */
411 if (ipv6_devconf
.proxy_ndp
&&
412 pneigh_lookup(&nd_tbl
, &hdr
->daddr
, skb
->dev
, 0)) {
413 int proxied
= ip6_forward_proxy_check(skb
);
415 return ip6_input(skb
);
416 else if (proxied
< 0) {
417 IP6_INC_STATS(ip6_dst_idev(dst
), IPSTATS_MIB_INDISCARDS
);
422 if (!xfrm6_route_forward(skb
)) {
423 IP6_INC_STATS(ip6_dst_idev(dst
), IPSTATS_MIB_INDISCARDS
);
428 /* IPv6 specs say nothing about it, but it is clear that we cannot
429 send redirects to source routed frames.
431 if (skb
->dev
== dst
->dev
&& dst
->neighbour
&& opt
->srcrt
== 0) {
432 struct in6_addr
*target
= NULL
;
434 struct neighbour
*n
= dst
->neighbour
;
437 * incoming and outgoing devices are the same
441 rt
= (struct rt6_info
*) dst
;
442 if ((rt
->rt6i_flags
& RTF_GATEWAY
))
443 target
= (struct in6_addr
*)&n
->primary_key
;
445 target
= &hdr
->daddr
;
447 /* Limit redirects both by destination (here)
448 and by source (inside ndisc_send_redirect)
450 if (xrlim_allow(dst
, 1*HZ
))
451 ndisc_send_redirect(skb
, n
, target
);
452 } else if (ipv6_addr_type(&hdr
->saddr
)&(IPV6_ADDR_MULTICAST
|IPV6_ADDR_LOOPBACK
453 |IPV6_ADDR_LINKLOCAL
)) {
454 /* This check is security critical. */
458 if (skb
->len
> dst_mtu(dst
)) {
459 /* Again, force OUTPUT device used as source address */
461 icmpv6_send(skb
, ICMPV6_PKT_TOOBIG
, 0, dst_mtu(dst
), skb
->dev
);
462 IP6_INC_STATS_BH(ip6_dst_idev(dst
), IPSTATS_MIB_INTOOBIGERRORS
);
463 IP6_INC_STATS_BH(ip6_dst_idev(dst
), IPSTATS_MIB_FRAGFAILS
);
468 if (skb_cow(skb
, dst
->dev
->hard_header_len
)) {
469 IP6_INC_STATS(ip6_dst_idev(dst
), IPSTATS_MIB_OUTDISCARDS
);
475 /* Mangling hops number delayed to point after skb COW */
479 IP6_INC_STATS_BH(ip6_dst_idev(dst
), IPSTATS_MIB_OUTFORWDATAGRAMS
);
480 return NF_HOOK(PF_INET6
,NF_IP6_FORWARD
, skb
, skb
->dev
, dst
->dev
, ip6_forward_finish
);
483 IP6_INC_STATS_BH(ip6_dst_idev(dst
), IPSTATS_MIB_INADDRERRORS
);
489 static void ip6_copy_metadata(struct sk_buff
*to
, struct sk_buff
*from
)
491 to
->pkt_type
= from
->pkt_type
;
492 to
->priority
= from
->priority
;
493 to
->protocol
= from
->protocol
;
494 dst_release(to
->dst
);
495 to
->dst
= dst_clone(from
->dst
);
497 to
->mark
= from
->mark
;
499 #ifdef CONFIG_NET_SCHED
500 to
->tc_index
= from
->tc_index
;
502 #ifdef CONFIG_NETFILTER
503 /* Connection association is same as pre-frag packet */
504 nf_conntrack_put(to
->nfct
);
505 to
->nfct
= from
->nfct
;
506 nf_conntrack_get(to
->nfct
);
507 to
->nfctinfo
= from
->nfctinfo
;
508 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
509 nf_conntrack_put_reasm(to
->nfct_reasm
);
510 to
->nfct_reasm
= from
->nfct_reasm
;
511 nf_conntrack_get_reasm(to
->nfct_reasm
);
513 #ifdef CONFIG_BRIDGE_NETFILTER
514 nf_bridge_put(to
->nf_bridge
);
515 to
->nf_bridge
= from
->nf_bridge
;
516 nf_bridge_get(to
->nf_bridge
);
519 skb_copy_secmark(to
, from
);
522 int ip6_find_1stfragopt(struct sk_buff
*skb
, u8
**nexthdr
)
524 u16 offset
= sizeof(struct ipv6hdr
);
525 struct ipv6_opt_hdr
*exthdr
= (struct ipv6_opt_hdr
*)(skb
->nh
.ipv6h
+ 1);
526 unsigned int packet_len
= skb
->tail
- skb
->nh
.raw
;
528 *nexthdr
= &skb
->nh
.ipv6h
->nexthdr
;
530 while (offset
+ 1 <= packet_len
) {
536 case NEXTHDR_ROUTING
:
540 #ifdef CONFIG_IPV6_MIP6
541 if (ipv6_find_tlv(skb
, offset
, IPV6_TLV_HAO
) >= 0)
551 offset
+= ipv6_optlen(exthdr
);
552 *nexthdr
= &exthdr
->nexthdr
;
553 exthdr
= (struct ipv6_opt_hdr
*)(skb
->nh
.raw
+ offset
);
558 EXPORT_SYMBOL_GPL(ip6_find_1stfragopt
);
560 static int ip6_fragment(struct sk_buff
*skb
, int (*output
)(struct sk_buff
*))
562 struct net_device
*dev
;
563 struct sk_buff
*frag
;
564 struct rt6_info
*rt
= (struct rt6_info
*)skb
->dst
;
565 struct ipv6_pinfo
*np
= skb
->sk
? inet6_sk(skb
->sk
) : NULL
;
566 struct ipv6hdr
*tmp_hdr
;
568 unsigned int mtu
, hlen
, left
, len
;
570 int ptr
, offset
= 0, err
=0;
571 u8
*prevhdr
, nexthdr
= 0;
574 hlen
= ip6_find_1stfragopt(skb
, &prevhdr
);
577 mtu
= dst_mtu(&rt
->u
.dst
);
578 if (np
&& np
->frag_size
< mtu
) {
582 mtu
-= hlen
+ sizeof(struct frag_hdr
);
584 if (skb_shinfo(skb
)->frag_list
) {
585 int first_len
= skb_pagelen(skb
);
587 if (first_len
- hlen
> mtu
||
588 ((first_len
- hlen
) & 7) ||
592 for (frag
= skb_shinfo(skb
)->frag_list
; frag
; frag
= frag
->next
) {
593 /* Correct geometry. */
594 if (frag
->len
> mtu
||
595 ((frag
->len
& 7) && frag
->next
) ||
596 skb_headroom(frag
) < hlen
)
599 /* Partially cloned skb? */
600 if (skb_shared(frag
))
607 frag
->destructor
= sock_wfree
;
608 skb
->truesize
-= frag
->truesize
;
614 frag
= skb_shinfo(skb
)->frag_list
;
615 skb_shinfo(skb
)->frag_list
= NULL
;
618 *prevhdr
= NEXTHDR_FRAGMENT
;
619 tmp_hdr
= kmemdup(skb
->nh
.raw
, hlen
, GFP_ATOMIC
);
621 IP6_INC_STATS(ip6_dst_idev(skb
->dst
), IPSTATS_MIB_FRAGFAILS
);
625 __skb_pull(skb
, hlen
);
626 fh
= (struct frag_hdr
*)__skb_push(skb
, sizeof(struct frag_hdr
));
627 skb
->nh
.raw
= __skb_push(skb
, hlen
);
628 memcpy(skb
->nh
.raw
, tmp_hdr
, hlen
);
630 ipv6_select_ident(skb
, fh
);
631 fh
->nexthdr
= nexthdr
;
633 fh
->frag_off
= htons(IP6_MF
);
634 frag_id
= fh
->identification
;
636 first_len
= skb_pagelen(skb
);
637 skb
->data_len
= first_len
- skb_headlen(skb
);
638 skb
->len
= first_len
;
639 skb
->nh
.ipv6h
->payload_len
= htons(first_len
- sizeof(struct ipv6hdr
));
641 dst_hold(&rt
->u
.dst
);
644 /* Prepare header of the next frame,
645 * before previous one went down. */
647 frag
->ip_summed
= CHECKSUM_NONE
;
648 frag
->h
.raw
= frag
->data
;
649 fh
= (struct frag_hdr
*)__skb_push(frag
, sizeof(struct frag_hdr
));
650 frag
->nh
.raw
= __skb_push(frag
, hlen
);
651 memcpy(frag
->nh
.raw
, tmp_hdr
, hlen
);
652 offset
+= skb
->len
- hlen
- sizeof(struct frag_hdr
);
653 fh
->nexthdr
= nexthdr
;
655 fh
->frag_off
= htons(offset
);
656 if (frag
->next
!= NULL
)
657 fh
->frag_off
|= htons(IP6_MF
);
658 fh
->identification
= frag_id
;
659 frag
->nh
.ipv6h
->payload_len
= htons(frag
->len
- sizeof(struct ipv6hdr
));
660 ip6_copy_metadata(frag
, skb
);
665 IP6_INC_STATS(ip6_dst_idev(&rt
->u
.dst
), IPSTATS_MIB_FRAGCREATES
);
678 IP6_INC_STATS(ip6_dst_idev(&rt
->u
.dst
), IPSTATS_MIB_FRAGOKS
);
679 dst_release(&rt
->u
.dst
);
689 IP6_INC_STATS(ip6_dst_idev(&rt
->u
.dst
), IPSTATS_MIB_FRAGFAILS
);
690 dst_release(&rt
->u
.dst
);
695 left
= skb
->len
- hlen
; /* Space per frame */
696 ptr
= hlen
; /* Where to start from */
699 * Fragment the datagram.
702 *prevhdr
= NEXTHDR_FRAGMENT
;
705 * Keep copying data until we run out.
709 /* IF: it doesn't fit, use 'mtu' - the data space left */
712 /* IF: we are not sending upto and including the packet end
713 then align the next start on an eight byte boundary */
721 if ((frag
= alloc_skb(len
+hlen
+sizeof(struct frag_hdr
)+LL_RESERVED_SPACE(rt
->u
.dst
.dev
), GFP_ATOMIC
)) == NULL
) {
722 NETDEBUG(KERN_INFO
"IPv6: frag: no memory for new fragment!\n");
723 IP6_INC_STATS(ip6_dst_idev(skb
->dst
),
724 IPSTATS_MIB_FRAGFAILS
);
730 * Set up data on packet
733 ip6_copy_metadata(frag
, skb
);
734 skb_reserve(frag
, LL_RESERVED_SPACE(rt
->u
.dst
.dev
));
735 skb_put(frag
, len
+ hlen
+ sizeof(struct frag_hdr
));
736 frag
->nh
.raw
= frag
->data
;
737 fh
= (struct frag_hdr
*)(frag
->data
+ hlen
);
738 frag
->h
.raw
= frag
->data
+ hlen
+ sizeof(struct frag_hdr
);
741 * Charge the memory for the fragment to any owner
745 skb_set_owner_w(frag
, skb
->sk
);
748 * Copy the packet header into the new buffer.
750 memcpy(frag
->nh
.raw
, skb
->data
, hlen
);
753 * Build fragment header.
755 fh
->nexthdr
= nexthdr
;
758 ipv6_select_ident(skb
, fh
);
759 frag_id
= fh
->identification
;
761 fh
->identification
= frag_id
;
764 * Copy a block of the IP datagram.
766 if (skb_copy_bits(skb
, ptr
, frag
->h
.raw
, len
))
770 fh
->frag_off
= htons(offset
);
772 fh
->frag_off
|= htons(IP6_MF
);
773 frag
->nh
.ipv6h
->payload_len
= htons(frag
->len
- sizeof(struct ipv6hdr
));
779 * Put this fragment into the sending queue.
785 IP6_INC_STATS(ip6_dst_idev(skb
->dst
), IPSTATS_MIB_FRAGCREATES
);
787 IP6_INC_STATS(ip6_dst_idev(skb
->dst
),
788 IPSTATS_MIB_FRAGOKS
);
793 IP6_INC_STATS(ip6_dst_idev(skb
->dst
),
794 IPSTATS_MIB_FRAGFAILS
);
799 static inline int ip6_rt_check(struct rt6key
*rt_key
,
800 struct in6_addr
*fl_addr
,
801 struct in6_addr
*addr_cache
)
803 return ((rt_key
->plen
!= 128 || !ipv6_addr_equal(fl_addr
, &rt_key
->addr
)) &&
804 (addr_cache
== NULL
|| !ipv6_addr_equal(fl_addr
, addr_cache
)));
807 static struct dst_entry
*ip6_sk_dst_check(struct sock
*sk
,
808 struct dst_entry
*dst
,
811 struct ipv6_pinfo
*np
= inet6_sk(sk
);
812 struct rt6_info
*rt
= (struct rt6_info
*)dst
;
817 /* Yes, checking route validity in not connected
818 * case is not very simple. Take into account,
819 * that we do not support routing by source, TOS,
820 * and MSG_DONTROUTE --ANK (980726)
822 * 1. ip6_rt_check(): If route was host route,
823 * check that cached destination is current.
824 * If it is network route, we still may
825 * check its validity using saved pointer
826 * to the last used address: daddr_cache.
827 * We do not want to save whole address now,
828 * (because main consumer of this service
829 * is tcp, which has not this problem),
830 * so that the last trick works only on connected
832 * 2. oif also should be the same.
834 if (ip6_rt_check(&rt
->rt6i_dst
, &fl
->fl6_dst
, np
->daddr_cache
) ||
835 #ifdef CONFIG_IPV6_SUBTREES
836 ip6_rt_check(&rt
->rt6i_src
, &fl
->fl6_src
, np
->saddr_cache
) ||
838 (fl
->oif
&& fl
->oif
!= dst
->dev
->ifindex
)) {
847 static int ip6_dst_lookup_tail(struct sock
*sk
,
848 struct dst_entry
**dst
, struct flowi
*fl
)
853 *dst
= ip6_route_output(sk
, fl
);
855 if ((err
= (*dst
)->error
))
856 goto out_err_release
;
858 if (ipv6_addr_any(&fl
->fl6_src
)) {
859 err
= ipv6_get_saddr(*dst
, &fl
->fl6_dst
, &fl
->fl6_src
);
861 goto out_err_release
;
873 * ip6_dst_lookup - perform route lookup on flow
874 * @sk: socket which provides route info
875 * @dst: pointer to dst_entry * for result
876 * @fl: flow to lookup
878 * This function performs a route lookup on the given flow.
880 * It returns zero on success, or a standard errno code on error.
882 int ip6_dst_lookup(struct sock
*sk
, struct dst_entry
**dst
, struct flowi
*fl
)
885 return ip6_dst_lookup_tail(sk
, dst
, fl
);
887 EXPORT_SYMBOL_GPL(ip6_dst_lookup
);
890 * ip6_sk_dst_lookup - perform socket cached route lookup on flow
891 * @sk: socket which provides the dst cache and route info
892 * @dst: pointer to dst_entry * for result
893 * @fl: flow to lookup
895 * This function performs a route lookup on the given flow with the
896 * possibility of using the cached route in the socket if it is valid.
897 * It will take the socket dst lock when operating on the dst cache.
898 * As a result, this function can only be used in process context.
900 * It returns zero on success, or a standard errno code on error.
902 int ip6_sk_dst_lookup(struct sock
*sk
, struct dst_entry
**dst
, struct flowi
*fl
)
906 *dst
= sk_dst_check(sk
, inet6_sk(sk
)->dst_cookie
);
907 *dst
= ip6_sk_dst_check(sk
, *dst
, fl
);
910 return ip6_dst_lookup_tail(sk
, dst
, fl
);
912 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup
);
914 static inline int ip6_ufo_append_data(struct sock
*sk
,
915 int getfrag(void *from
, char *to
, int offset
, int len
,
916 int odd
, struct sk_buff
*skb
),
917 void *from
, int length
, int hh_len
, int fragheaderlen
,
918 int transhdrlen
, int mtu
,unsigned int flags
)
924 /* There is support for UDP large send offload by network
925 * device, so create one single skb packet containing complete
928 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
) {
929 skb
= sock_alloc_send_skb(sk
,
930 hh_len
+ fragheaderlen
+ transhdrlen
+ 20,
931 (flags
& MSG_DONTWAIT
), &err
);
935 /* reserve space for Hardware header */
936 skb_reserve(skb
, hh_len
);
938 /* create space for UDP/IP header */
939 skb_put(skb
,fragheaderlen
+ transhdrlen
);
941 /* initialize network header pointer */
942 skb
->nh
.raw
= skb
->data
;
944 /* initialize protocol header pointer */
945 skb
->h
.raw
= skb
->data
+ fragheaderlen
;
947 skb
->ip_summed
= CHECKSUM_PARTIAL
;
949 sk
->sk_sndmsg_off
= 0;
952 err
= skb_append_datato_frags(sk
,skb
, getfrag
, from
,
953 (length
- transhdrlen
));
955 struct frag_hdr fhdr
;
957 /* specify the length of each IP datagram fragment*/
958 skb_shinfo(skb
)->gso_size
= mtu
- fragheaderlen
-
959 sizeof(struct frag_hdr
);
960 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
961 ipv6_select_ident(skb
, &fhdr
);
962 skb_shinfo(skb
)->ip6_frag_id
= fhdr
.identification
;
963 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
967 /* There is not enough support do UPD LSO,
968 * so follow normal path
975 int ip6_append_data(struct sock
*sk
, int getfrag(void *from
, char *to
,
976 int offset
, int len
, int odd
, struct sk_buff
*skb
),
977 void *from
, int length
, int transhdrlen
,
978 int hlimit
, int tclass
, struct ipv6_txoptions
*opt
, struct flowi
*fl
,
979 struct rt6_info
*rt
, unsigned int flags
)
981 struct inet_sock
*inet
= inet_sk(sk
);
982 struct ipv6_pinfo
*np
= inet6_sk(sk
);
984 unsigned int maxfraglen
, fragheaderlen
;
991 int csummode
= CHECKSUM_NONE
;
995 if (skb_queue_empty(&sk
->sk_write_queue
)) {
1000 if (np
->cork
.opt
== NULL
) {
1001 np
->cork
.opt
= kmalloc(opt
->tot_len
,
1003 if (unlikely(np
->cork
.opt
== NULL
))
1005 } else if (np
->cork
.opt
->tot_len
< opt
->tot_len
) {
1006 printk(KERN_DEBUG
"ip6_append_data: invalid option length\n");
1009 memcpy(np
->cork
.opt
, opt
, opt
->tot_len
);
1010 inet
->cork
.flags
|= IPCORK_OPT
;
1011 /* need source address above miyazawa*/
1013 dst_hold(&rt
->u
.dst
);
1015 inet
->cork
.fl
= *fl
;
1016 np
->cork
.hop_limit
= hlimit
;
1017 np
->cork
.tclass
= tclass
;
1018 mtu
= dst_mtu(rt
->u
.dst
.path
);
1019 if (np
->frag_size
< mtu
) {
1021 mtu
= np
->frag_size
;
1023 inet
->cork
.fragsize
= mtu
;
1024 if (dst_allfrag(rt
->u
.dst
.path
))
1025 inet
->cork
.flags
|= IPCORK_ALLFRAG
;
1026 inet
->cork
.length
= 0;
1027 sk
->sk_sndmsg_page
= NULL
;
1028 sk
->sk_sndmsg_off
= 0;
1029 exthdrlen
= rt
->u
.dst
.header_len
+ (opt
? opt
->opt_flen
: 0);
1030 length
+= exthdrlen
;
1031 transhdrlen
+= exthdrlen
;
1034 fl
= &inet
->cork
.fl
;
1035 if (inet
->cork
.flags
& IPCORK_OPT
)
1039 mtu
= inet
->cork
.fragsize
;
1042 hh_len
= LL_RESERVED_SPACE(rt
->u
.dst
.dev
);
1044 fragheaderlen
= sizeof(struct ipv6hdr
) + rt
->u
.dst
.nfheader_len
+ (opt
? opt
->opt_nflen
: 0);
1045 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
- sizeof(struct frag_hdr
);
1047 if (mtu
<= sizeof(struct ipv6hdr
) + IPV6_MAXPLEN
) {
1048 if (inet
->cork
.length
+ length
> sizeof(struct ipv6hdr
) + IPV6_MAXPLEN
- fragheaderlen
) {
1049 ipv6_local_error(sk
, EMSGSIZE
, fl
, mtu
-exthdrlen
);
1055 * Let's try using as much space as possible.
1056 * Use MTU if total length of the message fits into the MTU.
1057 * Otherwise, we need to reserve fragment header and
1058 * fragment alignment (= 8-15 octects, in total).
1060 * Note that we may need to "move" the data from the tail of
1061 * of the buffer to the new fragment when we split
1064 * FIXME: It may be fragmented into multiple chunks
1065 * at once if non-fragmentable extension headers
1070 inet
->cork
.length
+= length
;
1071 if (((length
> mtu
) && (sk
->sk_protocol
== IPPROTO_UDP
)) &&
1072 (rt
->u
.dst
.dev
->features
& NETIF_F_UFO
)) {
1074 err
= ip6_ufo_append_data(sk
, getfrag
, from
, length
, hh_len
,
1075 fragheaderlen
, transhdrlen
, mtu
,
1082 if ((skb
= skb_peek_tail(&sk
->sk_write_queue
)) == NULL
)
1085 while (length
> 0) {
1086 /* Check if the remaining data fits into current packet. */
1087 copy
= (inet
->cork
.length
<= mtu
&& !(inet
->cork
.flags
& IPCORK_ALLFRAG
) ? mtu
: maxfraglen
) - skb
->len
;
1089 copy
= maxfraglen
- skb
->len
;
1093 unsigned int datalen
;
1094 unsigned int fraglen
;
1095 unsigned int fraggap
;
1096 unsigned int alloclen
;
1097 struct sk_buff
*skb_prev
;
1101 /* There's no room in the current skb */
1103 fraggap
= skb_prev
->len
- maxfraglen
;
1108 * If remaining data exceeds the mtu,
1109 * we know we need more fragment(s).
1111 datalen
= length
+ fraggap
;
1112 if (datalen
> (inet
->cork
.length
<= mtu
&& !(inet
->cork
.flags
& IPCORK_ALLFRAG
) ? mtu
: maxfraglen
) - fragheaderlen
)
1113 datalen
= maxfraglen
- fragheaderlen
;
1115 fraglen
= datalen
+ fragheaderlen
;
1116 if ((flags
& MSG_MORE
) &&
1117 !(rt
->u
.dst
.dev
->features
&NETIF_F_SG
))
1120 alloclen
= datalen
+ fragheaderlen
;
1123 * The last fragment gets additional space at tail.
1124 * Note: we overallocate on fragments with MSG_MODE
1125 * because we have no idea if we're the last one.
1127 if (datalen
== length
+ fraggap
)
1128 alloclen
+= rt
->u
.dst
.trailer_len
;
1131 * We just reserve space for fragment header.
1132 * Note: this may be overallocation if the message
1133 * (without MSG_MORE) fits into the MTU.
1135 alloclen
+= sizeof(struct frag_hdr
);
1138 skb
= sock_alloc_send_skb(sk
,
1140 (flags
& MSG_DONTWAIT
), &err
);
1143 if (atomic_read(&sk
->sk_wmem_alloc
) <=
1145 skb
= sock_wmalloc(sk
,
1146 alloclen
+ hh_len
, 1,
1148 if (unlikely(skb
== NULL
))
1154 * Fill in the control structures
1156 skb
->ip_summed
= csummode
;
1158 /* reserve for fragmentation */
1159 skb_reserve(skb
, hh_len
+sizeof(struct frag_hdr
));
1162 * Find where to start putting bytes
1164 data
= skb_put(skb
, fraglen
);
1165 skb
->nh
.raw
= data
+ exthdrlen
;
1166 data
+= fragheaderlen
;
1167 skb
->h
.raw
= data
+ exthdrlen
;
1170 skb
->csum
= skb_copy_and_csum_bits(
1171 skb_prev
, maxfraglen
,
1172 data
+ transhdrlen
, fraggap
, 0);
1173 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
1176 pskb_trim_unique(skb_prev
, maxfraglen
);
1178 copy
= datalen
- transhdrlen
- fraggap
;
1183 } else if (copy
> 0 && getfrag(from
, data
+ transhdrlen
, offset
, copy
, fraggap
, skb
) < 0) {
1190 length
-= datalen
- fraggap
;
1193 csummode
= CHECKSUM_NONE
;
1196 * Put the packet on the pending queue
1198 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1205 if (!(rt
->u
.dst
.dev
->features
&NETIF_F_SG
)) {
1209 if (getfrag(from
, skb_put(skb
, copy
),
1210 offset
, copy
, off
, skb
) < 0) {
1211 __skb_trim(skb
, off
);
1216 int i
= skb_shinfo(skb
)->nr_frags
;
1217 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
-1];
1218 struct page
*page
= sk
->sk_sndmsg_page
;
1219 int off
= sk
->sk_sndmsg_off
;
1222 if (page
&& (left
= PAGE_SIZE
- off
) > 0) {
1225 if (page
!= frag
->page
) {
1226 if (i
== MAX_SKB_FRAGS
) {
1231 skb_fill_page_desc(skb
, i
, page
, sk
->sk_sndmsg_off
, 0);
1232 frag
= &skb_shinfo(skb
)->frags
[i
];
1234 } else if(i
< MAX_SKB_FRAGS
) {
1235 if (copy
> PAGE_SIZE
)
1237 page
= alloc_pages(sk
->sk_allocation
, 0);
1242 sk
->sk_sndmsg_page
= page
;
1243 sk
->sk_sndmsg_off
= 0;
1245 skb_fill_page_desc(skb
, i
, page
, 0, 0);
1246 frag
= &skb_shinfo(skb
)->frags
[i
];
1247 skb
->truesize
+= PAGE_SIZE
;
1248 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
1253 if (getfrag(from
, page_address(frag
->page
)+frag
->page_offset
+frag
->size
, offset
, copy
, skb
->len
, skb
) < 0) {
1257 sk
->sk_sndmsg_off
+= copy
;
1260 skb
->data_len
+= copy
;
1267 inet
->cork
.length
-= length
;
1268 IP6_INC_STATS(rt
->rt6i_idev
, IPSTATS_MIB_OUTDISCARDS
);
1272 int ip6_push_pending_frames(struct sock
*sk
)
1274 struct sk_buff
*skb
, *tmp_skb
;
1275 struct sk_buff
**tail_skb
;
1276 struct in6_addr final_dst_buf
, *final_dst
= &final_dst_buf
;
1277 struct inet_sock
*inet
= inet_sk(sk
);
1278 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1279 struct ipv6hdr
*hdr
;
1280 struct ipv6_txoptions
*opt
= np
->cork
.opt
;
1281 struct rt6_info
*rt
= np
->cork
.rt
;
1282 struct flowi
*fl
= &inet
->cork
.fl
;
1283 unsigned char proto
= fl
->proto
;
1286 if ((skb
= __skb_dequeue(&sk
->sk_write_queue
)) == NULL
)
1288 tail_skb
= &(skb_shinfo(skb
)->frag_list
);
1290 /* move skb->data to ip header from ext header */
1291 if (skb
->data
< skb
->nh
.raw
)
1292 __skb_pull(skb
, skb
->nh
.raw
- skb
->data
);
1293 while ((tmp_skb
= __skb_dequeue(&sk
->sk_write_queue
)) != NULL
) {
1294 __skb_pull(tmp_skb
, skb
->h
.raw
- skb
->nh
.raw
);
1295 *tail_skb
= tmp_skb
;
1296 tail_skb
= &(tmp_skb
->next
);
1297 skb
->len
+= tmp_skb
->len
;
1298 skb
->data_len
+= tmp_skb
->len
;
1299 skb
->truesize
+= tmp_skb
->truesize
;
1300 __sock_put(tmp_skb
->sk
);
1301 tmp_skb
->destructor
= NULL
;
1305 ipv6_addr_copy(final_dst
, &fl
->fl6_dst
);
1306 __skb_pull(skb
, skb
->h
.raw
- skb
->nh
.raw
);
1307 if (opt
&& opt
->opt_flen
)
1308 ipv6_push_frag_opts(skb
, opt
, &proto
);
1309 if (opt
&& opt
->opt_nflen
)
1310 ipv6_push_nfrag_opts(skb
, opt
, &proto
, &final_dst
);
1312 skb
->nh
.ipv6h
= hdr
= (struct ipv6hdr
*) skb_push(skb
, sizeof(struct ipv6hdr
));
1314 *(__be32
*)hdr
= fl
->fl6_flowlabel
|
1315 htonl(0x60000000 | ((int)np
->cork
.tclass
<< 20));
1317 if (skb
->len
<= sizeof(struct ipv6hdr
) + IPV6_MAXPLEN
)
1318 hdr
->payload_len
= htons(skb
->len
- sizeof(struct ipv6hdr
));
1320 hdr
->payload_len
= 0;
1321 hdr
->hop_limit
= np
->cork
.hop_limit
;
1322 hdr
->nexthdr
= proto
;
1323 ipv6_addr_copy(&hdr
->saddr
, &fl
->fl6_src
);
1324 ipv6_addr_copy(&hdr
->daddr
, final_dst
);
1326 skb
->priority
= sk
->sk_priority
;
1328 skb
->dst
= dst_clone(&rt
->u
.dst
);
1329 IP6_INC_STATS(rt
->rt6i_idev
, IPSTATS_MIB_OUTREQUESTS
);
1330 err
= NF_HOOK(PF_INET6
, NF_IP6_LOCAL_OUT
, skb
, NULL
, skb
->dst
->dev
, dst_output
);
1333 err
= np
->recverr
? net_xmit_errno(err
) : 0;
1339 inet
->cork
.flags
&= ~IPCORK_OPT
;
1340 kfree(np
->cork
.opt
);
1341 np
->cork
.opt
= NULL
;
1343 dst_release(&np
->cork
.rt
->u
.dst
);
1345 inet
->cork
.flags
&= ~IPCORK_ALLFRAG
;
1347 memset(&inet
->cork
.fl
, 0, sizeof(inet
->cork
.fl
));
1353 void ip6_flush_pending_frames(struct sock
*sk
)
1355 struct inet_sock
*inet
= inet_sk(sk
);
1356 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1357 struct sk_buff
*skb
;
1359 while ((skb
= __skb_dequeue_tail(&sk
->sk_write_queue
)) != NULL
) {
1360 IP6_INC_STATS(ip6_dst_idev(skb
->dst
),
1361 IPSTATS_MIB_OUTDISCARDS
);
1365 inet
->cork
.flags
&= ~IPCORK_OPT
;
1367 kfree(np
->cork
.opt
);
1368 np
->cork
.opt
= NULL
;
1370 dst_release(&np
->cork
.rt
->u
.dst
);
1372 inet
->cork
.flags
&= ~IPCORK_ALLFRAG
;
1374 memset(&inet
->cork
.fl
, 0, sizeof(inet
->cork
.fl
));