2 * IPv6 output functions
3 * Linux INET6 implementation
6 * Pedro Roque <roque@di.fc.ul.pt>
8 * Based on linux/net/ipv4/ip_output.c
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
16 * A.N.Kuznetsov : airthmetics in fragmentation.
17 * extension headers are implemented.
18 * route changes now work.
19 * ip6_forward does not confuse sniffers.
22 * H. von Brand : Added missing #include <linux/string.h>
23 * Imran Patel : frag id should be in NBO
24 * Kazunori MIYAZAWA @USAGI
25 * : add ip6_append_data and related functions
29 #include <linux/errno.h>
30 #include <linux/kernel.h>
31 #include <linux/string.h>
32 #include <linux/socket.h>
33 #include <linux/net.h>
34 #include <linux/netdevice.h>
35 #include <linux/if_arp.h>
36 #include <linux/in6.h>
37 #include <linux/tcp.h>
38 #include <linux/route.h>
39 #include <linux/module.h>
40 #include <linux/slab.h>
42 #include <linux/bpf-cgroup.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>
58 #include <linux/mroute6.h>
59 #include <net/l3mdev.h>
60 #include <net/lwtunnel.h>
62 static int ip6_finish_output2(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
64 struct dst_entry
*dst
= skb_dst(skb
);
65 struct net_device
*dev
= dst
->dev
;
66 struct neighbour
*neigh
;
67 struct in6_addr
*nexthop
;
70 skb
->protocol
= htons(ETH_P_IPV6
);
73 if (ipv6_addr_is_multicast(&ipv6_hdr(skb
)->daddr
)) {
74 struct inet6_dev
*idev
= ip6_dst_idev(skb_dst(skb
));
76 if (!(dev
->flags
& IFF_LOOPBACK
) && sk_mc_loop(sk
) &&
77 ((mroute6_socket(net
, skb
) &&
78 !(IP6CB(skb
)->flags
& IP6SKB_FORWARDED
)) ||
79 ipv6_chk_mcast_addr(dev
, &ipv6_hdr(skb
)->daddr
,
80 &ipv6_hdr(skb
)->saddr
))) {
81 struct sk_buff
*newskb
= skb_clone(skb
, GFP_ATOMIC
);
83 /* Do not check for IFF_ALLMULTI; multicast routing
84 is not supported in any case.
87 NF_HOOK(NFPROTO_IPV6
, NF_INET_POST_ROUTING
,
88 net
, sk
, newskb
, NULL
, newskb
->dev
,
91 if (ipv6_hdr(skb
)->hop_limit
== 0) {
92 IP6_INC_STATS(net
, idev
,
93 IPSTATS_MIB_OUTDISCARDS
);
99 IP6_UPD_PO_STATS(net
, idev
, IPSTATS_MIB_OUTMCAST
, skb
->len
);
101 if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb
)->daddr
) <=
102 IPV6_ADDR_SCOPE_NODELOCAL
&&
103 !(dev
->flags
& IFF_LOOPBACK
)) {
109 if (lwtunnel_xmit_redirect(dst
->lwtstate
)) {
110 int res
= lwtunnel_xmit(skb
);
112 if (res
< 0 || res
== LWTUNNEL_XMIT_DONE
)
117 nexthop
= rt6_nexthop((struct rt6_info
*)dst
, &ipv6_hdr(skb
)->daddr
);
118 neigh
= __ipv6_neigh_lookup_noref(dst
->dev
, nexthop
);
119 if (unlikely(!neigh
))
120 neigh
= __neigh_create(&nd_tbl
, nexthop
, dst
->dev
, false);
121 if (!IS_ERR(neigh
)) {
122 ret
= dst_neigh_output(dst
, neigh
, skb
);
123 rcu_read_unlock_bh();
126 rcu_read_unlock_bh();
128 IP6_INC_STATS(net
, ip6_dst_idev(dst
), IPSTATS_MIB_OUTNOROUTES
);
133 static int ip6_finish_output(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
137 ret
= BPF_CGROUP_RUN_PROG_INET_EGRESS(sk
, skb
);
143 if ((skb
->len
> ip6_skb_dst_mtu(skb
) && !skb_is_gso(skb
)) ||
144 dst_allfrag(skb_dst(skb
)) ||
145 (IP6CB(skb
)->frag_max_size
&& skb
->len
> IP6CB(skb
)->frag_max_size
))
146 return ip6_fragment(net
, sk
, skb
, ip6_finish_output2
);
148 return ip6_finish_output2(net
, sk
, skb
);
151 int ip6_output(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
153 struct net_device
*dev
= skb_dst(skb
)->dev
;
154 struct inet6_dev
*idev
= ip6_dst_idev(skb_dst(skb
));
156 if (unlikely(idev
->cnf
.disable_ipv6
)) {
157 IP6_INC_STATS(net
, idev
, IPSTATS_MIB_OUTDISCARDS
);
162 return NF_HOOK_COND(NFPROTO_IPV6
, NF_INET_POST_ROUTING
,
163 net
, sk
, skb
, NULL
, dev
,
165 !(IP6CB(skb
)->flags
& IP6SKB_REROUTED
));
169 * xmit an sk_buff (used by TCP, SCTP and DCCP)
170 * Note : socket lock is not held for SYNACK packets, but might be modified
171 * by calls to skb_set_owner_w() and ipv6_local_error(),
172 * which are using proper atomic operations or spinlocks.
174 int ip6_xmit(const struct sock
*sk
, struct sk_buff
*skb
, struct flowi6
*fl6
,
175 __u32 mark
, struct ipv6_txoptions
*opt
, int tclass
)
177 struct net
*net
= sock_net(sk
);
178 const struct ipv6_pinfo
*np
= inet6_sk(sk
);
179 struct in6_addr
*first_hop
= &fl6
->daddr
;
180 struct dst_entry
*dst
= skb_dst(skb
);
182 u8 proto
= fl6
->flowi6_proto
;
183 int seg_len
= skb
->len
;
188 unsigned int head_room
;
190 /* First: exthdrs may take lots of space (~8K for now)
191 MAX_HEADER is not enough.
193 head_room
= opt
->opt_nflen
+ opt
->opt_flen
;
194 seg_len
+= head_room
;
195 head_room
+= sizeof(struct ipv6hdr
) + LL_RESERVED_SPACE(dst
->dev
);
197 if (skb_headroom(skb
) < head_room
) {
198 struct sk_buff
*skb2
= skb_realloc_headroom(skb
, head_room
);
200 IP6_INC_STATS(net
, ip6_dst_idev(skb_dst(skb
)),
201 IPSTATS_MIB_OUTDISCARDS
);
207 /* skb_set_owner_w() changes sk->sk_wmem_alloc atomically,
208 * it is safe to call in our context (socket lock not held)
210 skb_set_owner_w(skb
, (struct sock
*)sk
);
213 ipv6_push_frag_opts(skb
, opt
, &proto
);
215 ipv6_push_nfrag_opts(skb
, opt
, &proto
, &first_hop
,
219 skb_push(skb
, sizeof(struct ipv6hdr
));
220 skb_reset_network_header(skb
);
224 * Fill in the IPv6 header
227 hlimit
= np
->hop_limit
;
229 hlimit
= ip6_dst_hoplimit(dst
);
231 ip6_flow_hdr(hdr
, tclass
, ip6_make_flowlabel(net
, skb
, fl6
->flowlabel
,
232 np
->autoflowlabel
, fl6
));
234 hdr
->payload_len
= htons(seg_len
);
235 hdr
->nexthdr
= proto
;
236 hdr
->hop_limit
= hlimit
;
238 hdr
->saddr
= fl6
->saddr
;
239 hdr
->daddr
= *first_hop
;
241 skb
->protocol
= htons(ETH_P_IPV6
);
242 skb
->priority
= sk
->sk_priority
;
246 if ((skb
->len
<= mtu
) || skb
->ignore_df
|| skb_is_gso(skb
)) {
247 IP6_UPD_PO_STATS(net
, ip6_dst_idev(skb_dst(skb
)),
248 IPSTATS_MIB_OUT
, skb
->len
);
250 /* if egress device is enslaved to an L3 master device pass the
251 * skb to its handler for processing
253 skb
= l3mdev_ip6_out((struct sock
*)sk
, skb
);
257 /* hooks should never assume socket lock is held.
258 * we promote our socket to non const
260 return NF_HOOK(NFPROTO_IPV6
, NF_INET_LOCAL_OUT
,
261 net
, (struct sock
*)sk
, skb
, NULL
, dst
->dev
,
266 /* ipv6_local_error() does not require socket lock,
267 * we promote our socket to non const
269 ipv6_local_error((struct sock
*)sk
, EMSGSIZE
, fl6
, mtu
);
271 IP6_INC_STATS(net
, ip6_dst_idev(skb_dst(skb
)), IPSTATS_MIB_FRAGFAILS
);
275 EXPORT_SYMBOL(ip6_xmit
);
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
= ipv6_hdr(skb
);
309 u8 nexthdr
= hdr
->nexthdr
;
313 if (ipv6_ext_hdr(nexthdr
)) {
314 offset
= ipv6_skip_exthdr(skb
, sizeof(*hdr
), &nexthdr
, &frag_off
);
318 offset
= sizeof(struct ipv6hdr
);
320 if (nexthdr
== IPPROTO_ICMPV6
) {
321 struct icmp6hdr
*icmp6
;
323 if (!pskb_may_pull(skb
, (skb_network_header(skb
) +
324 offset
+ 1 - skb
->data
)))
327 icmp6
= (struct icmp6hdr
*)(skb_network_header(skb
) + 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 net
*net
, struct sock
*sk
,
361 return dst_output(net
, sk
, skb
);
364 static unsigned int ip6_dst_mtu_forward(const struct dst_entry
*dst
)
367 struct inet6_dev
*idev
;
369 if (dst_metric_locked(dst
, RTAX_MTU
)) {
370 mtu
= dst_metric_raw(dst
, RTAX_MTU
);
377 idev
= __in6_dev_get(dst
->dev
);
379 mtu
= idev
->cnf
.mtu6
;
385 static bool ip6_pkt_too_big(const struct sk_buff
*skb
, unsigned int mtu
)
390 /* ipv6 conntrack defrag sets max_frag_size + ignore_df */
391 if (IP6CB(skb
)->frag_max_size
&& IP6CB(skb
)->frag_max_size
> mtu
)
397 if (skb_is_gso(skb
) && skb_gso_validate_mtu(skb
, mtu
))
403 int ip6_forward(struct sk_buff
*skb
)
405 struct dst_entry
*dst
= skb_dst(skb
);
406 struct ipv6hdr
*hdr
= ipv6_hdr(skb
);
407 struct inet6_skb_parm
*opt
= IP6CB(skb
);
408 struct net
*net
= dev_net(dst
->dev
);
411 if (net
->ipv6
.devconf_all
->forwarding
== 0)
414 if (skb
->pkt_type
!= PACKET_HOST
)
417 if (unlikely(skb
->sk
))
420 if (skb_warn_if_lro(skb
))
423 if (!xfrm6_policy_check(NULL
, XFRM_POLICY_FWD
, skb
)) {
424 __IP6_INC_STATS(net
, ip6_dst_idev(dst
),
425 IPSTATS_MIB_INDISCARDS
);
429 skb_forward_csum(skb
);
432 * We DO NOT make any processing on
433 * RA packets, pushing them to user level AS IS
434 * without ane WARRANTY that application will be able
435 * to interpret them. The reason is that we
436 * cannot make anything clever here.
438 * We are not end-node, so that if packet contains
439 * AH/ESP, we cannot make anything.
440 * Defragmentation also would be mistake, RA packets
441 * cannot be fragmented, because there is no warranty
442 * that different fragments will go along one path. --ANK
444 if (unlikely(opt
->flags
& IP6SKB_ROUTERALERT
)) {
445 if (ip6_call_ra_chain(skb
, ntohs(opt
->ra
)))
450 * check and decrement ttl
452 if (hdr
->hop_limit
<= 1) {
453 /* Force OUTPUT device used as source address */
455 icmpv6_send(skb
, ICMPV6_TIME_EXCEED
, ICMPV6_EXC_HOPLIMIT
, 0);
456 __IP6_INC_STATS(net
, ip6_dst_idev(dst
),
457 IPSTATS_MIB_INHDRERRORS
);
463 /* XXX: idev->cnf.proxy_ndp? */
464 if (net
->ipv6
.devconf_all
->proxy_ndp
&&
465 pneigh_lookup(&nd_tbl
, net
, &hdr
->daddr
, skb
->dev
, 0)) {
466 int proxied
= ip6_forward_proxy_check(skb
);
468 return ip6_input(skb
);
469 else if (proxied
< 0) {
470 __IP6_INC_STATS(net
, ip6_dst_idev(dst
),
471 IPSTATS_MIB_INDISCARDS
);
476 if (!xfrm6_route_forward(skb
)) {
477 __IP6_INC_STATS(net
, ip6_dst_idev(dst
),
478 IPSTATS_MIB_INDISCARDS
);
483 /* IPv6 specs say nothing about it, but it is clear that we cannot
484 send redirects to source routed frames.
485 We don't send redirects to frames decapsulated from IPsec.
487 if (skb
->dev
== dst
->dev
&& opt
->srcrt
== 0 && !skb_sec_path(skb
)) {
488 struct in6_addr
*target
= NULL
;
489 struct inet_peer
*peer
;
493 * incoming and outgoing devices are the same
497 rt
= (struct rt6_info
*) dst
;
498 if (rt
->rt6i_flags
& RTF_GATEWAY
)
499 target
= &rt
->rt6i_gateway
;
501 target
= &hdr
->daddr
;
503 peer
= inet_getpeer_v6(net
->ipv6
.peers
, &hdr
->daddr
, 1);
505 /* Limit redirects both by destination (here)
506 and by source (inside ndisc_send_redirect)
508 if (inet_peer_xrlim_allow(peer
, 1*HZ
))
509 ndisc_send_redirect(skb
, target
);
513 int addrtype
= ipv6_addr_type(&hdr
->saddr
);
515 /* This check is security critical. */
516 if (addrtype
== IPV6_ADDR_ANY
||
517 addrtype
& (IPV6_ADDR_MULTICAST
| IPV6_ADDR_LOOPBACK
))
519 if (addrtype
& IPV6_ADDR_LINKLOCAL
) {
520 icmpv6_send(skb
, ICMPV6_DEST_UNREACH
,
521 ICMPV6_NOT_NEIGHBOUR
, 0);
526 mtu
= ip6_dst_mtu_forward(dst
);
527 if (mtu
< IPV6_MIN_MTU
)
530 if (ip6_pkt_too_big(skb
, mtu
)) {
531 /* Again, force OUTPUT device used as source address */
533 icmpv6_send(skb
, ICMPV6_PKT_TOOBIG
, 0, mtu
);
534 __IP6_INC_STATS(net
, ip6_dst_idev(dst
),
535 IPSTATS_MIB_INTOOBIGERRORS
);
536 __IP6_INC_STATS(net
, ip6_dst_idev(dst
),
537 IPSTATS_MIB_FRAGFAILS
);
542 if (skb_cow(skb
, dst
->dev
->hard_header_len
)) {
543 __IP6_INC_STATS(net
, ip6_dst_idev(dst
),
544 IPSTATS_MIB_OUTDISCARDS
);
550 /* Mangling hops number delayed to point after skb COW */
554 __IP6_INC_STATS(net
, ip6_dst_idev(dst
), IPSTATS_MIB_OUTFORWDATAGRAMS
);
555 __IP6_ADD_STATS(net
, ip6_dst_idev(dst
), IPSTATS_MIB_OUTOCTETS
, skb
->len
);
556 return NF_HOOK(NFPROTO_IPV6
, NF_INET_FORWARD
,
557 net
, NULL
, skb
, skb
->dev
, dst
->dev
,
561 __IP6_INC_STATS(net
, ip6_dst_idev(dst
), IPSTATS_MIB_INADDRERRORS
);
567 static void ip6_copy_metadata(struct sk_buff
*to
, struct sk_buff
*from
)
569 to
->pkt_type
= from
->pkt_type
;
570 to
->priority
= from
->priority
;
571 to
->protocol
= from
->protocol
;
573 skb_dst_set(to
, dst_clone(skb_dst(from
)));
575 to
->mark
= from
->mark
;
577 #ifdef CONFIG_NET_SCHED
578 to
->tc_index
= from
->tc_index
;
581 skb_copy_secmark(to
, from
);
584 int ip6_fragment(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
,
585 int (*output
)(struct net
*, struct sock
*, struct sk_buff
*))
587 struct sk_buff
*frag
;
588 struct rt6_info
*rt
= (struct rt6_info
*)skb_dst(skb
);
589 struct ipv6_pinfo
*np
= skb
->sk
&& !dev_recursion_level() ?
590 inet6_sk(skb
->sk
) : NULL
;
591 struct ipv6hdr
*tmp_hdr
;
593 unsigned int mtu
, hlen
, left
, len
;
596 int ptr
, offset
= 0, err
= 0;
597 u8
*prevhdr
, nexthdr
= 0;
599 hlen
= ip6_find_1stfragopt(skb
, &prevhdr
);
602 mtu
= ip6_skb_dst_mtu(skb
);
604 /* We must not fragment if the socket is set to force MTU discovery
605 * or if the skb it not generated by a local socket.
607 if (unlikely(!skb
->ignore_df
&& skb
->len
> mtu
))
610 if (IP6CB(skb
)->frag_max_size
) {
611 if (IP6CB(skb
)->frag_max_size
> mtu
)
614 /* don't send fragments larger than what we received */
615 mtu
= IP6CB(skb
)->frag_max_size
;
616 if (mtu
< IPV6_MIN_MTU
)
620 if (np
&& np
->frag_size
< mtu
) {
624 if (mtu
< hlen
+ sizeof(struct frag_hdr
) + 8)
626 mtu
-= hlen
+ sizeof(struct frag_hdr
);
628 frag_id
= ipv6_select_ident(net
, &ipv6_hdr(skb
)->daddr
,
629 &ipv6_hdr(skb
)->saddr
);
631 if (skb
->ip_summed
== CHECKSUM_PARTIAL
&&
632 (err
= skb_checksum_help(skb
)))
635 hroom
= LL_RESERVED_SPACE(rt
->dst
.dev
);
636 if (skb_has_frag_list(skb
)) {
637 unsigned int first_len
= skb_pagelen(skb
);
638 struct sk_buff
*frag2
;
640 if (first_len
- hlen
> mtu
||
641 ((first_len
- hlen
) & 7) ||
643 skb_headroom(skb
) < (hroom
+ sizeof(struct frag_hdr
)))
646 skb_walk_frags(skb
, frag
) {
647 /* Correct geometry. */
648 if (frag
->len
> mtu
||
649 ((frag
->len
& 7) && frag
->next
) ||
650 skb_headroom(frag
) < (hlen
+ hroom
+ sizeof(struct frag_hdr
)))
651 goto slow_path_clean
;
653 /* Partially cloned skb? */
654 if (skb_shared(frag
))
655 goto slow_path_clean
;
660 frag
->destructor
= sock_wfree
;
662 skb
->truesize
-= frag
->truesize
;
669 *prevhdr
= NEXTHDR_FRAGMENT
;
670 tmp_hdr
= kmemdup(skb_network_header(skb
), hlen
, GFP_ATOMIC
);
672 IP6_INC_STATS(net
, ip6_dst_idev(skb_dst(skb
)),
673 IPSTATS_MIB_FRAGFAILS
);
677 frag
= skb_shinfo(skb
)->frag_list
;
678 skb_frag_list_init(skb
);
680 __skb_pull(skb
, hlen
);
681 fh
= (struct frag_hdr
*)__skb_push(skb
, sizeof(struct frag_hdr
));
682 __skb_push(skb
, hlen
);
683 skb_reset_network_header(skb
);
684 memcpy(skb_network_header(skb
), tmp_hdr
, hlen
);
686 fh
->nexthdr
= nexthdr
;
688 fh
->frag_off
= htons(IP6_MF
);
689 fh
->identification
= frag_id
;
691 first_len
= skb_pagelen(skb
);
692 skb
->data_len
= first_len
- skb_headlen(skb
);
693 skb
->len
= first_len
;
694 ipv6_hdr(skb
)->payload_len
= htons(first_len
-
695 sizeof(struct ipv6hdr
));
700 /* Prepare header of the next frame,
701 * before previous one went down. */
703 frag
->ip_summed
= CHECKSUM_NONE
;
704 skb_reset_transport_header(frag
);
705 fh
= (struct frag_hdr
*)__skb_push(frag
, sizeof(struct frag_hdr
));
706 __skb_push(frag
, hlen
);
707 skb_reset_network_header(frag
);
708 memcpy(skb_network_header(frag
), tmp_hdr
,
710 offset
+= skb
->len
- hlen
- sizeof(struct frag_hdr
);
711 fh
->nexthdr
= nexthdr
;
713 fh
->frag_off
= htons(offset
);
715 fh
->frag_off
|= htons(IP6_MF
);
716 fh
->identification
= frag_id
;
717 ipv6_hdr(frag
)->payload_len
=
719 sizeof(struct ipv6hdr
));
720 ip6_copy_metadata(frag
, skb
);
723 err
= output(net
, sk
, skb
);
725 IP6_INC_STATS(net
, ip6_dst_idev(&rt
->dst
),
726 IPSTATS_MIB_FRAGCREATES
);
739 IP6_INC_STATS(net
, ip6_dst_idev(&rt
->dst
),
740 IPSTATS_MIB_FRAGOKS
);
745 kfree_skb_list(frag
);
747 IP6_INC_STATS(net
, ip6_dst_idev(&rt
->dst
),
748 IPSTATS_MIB_FRAGFAILS
);
753 skb_walk_frags(skb
, frag2
) {
757 frag2
->destructor
= NULL
;
758 skb
->truesize
+= frag2
->truesize
;
763 left
= skb
->len
- hlen
; /* Space per frame */
764 ptr
= hlen
; /* Where to start from */
767 * Fragment the datagram.
770 *prevhdr
= NEXTHDR_FRAGMENT
;
771 troom
= rt
->dst
.dev
->needed_tailroom
;
774 * Keep copying data until we run out.
778 /* IF: it doesn't fit, use 'mtu' - the data space left */
781 /* IF: we are not sending up to and including the packet end
782 then align the next start on an eight byte boundary */
787 /* Allocate buffer */
788 frag
= alloc_skb(len
+ hlen
+ sizeof(struct frag_hdr
) +
789 hroom
+ troom
, GFP_ATOMIC
);
791 IP6_INC_STATS(net
, ip6_dst_idev(skb_dst(skb
)),
792 IPSTATS_MIB_FRAGFAILS
);
798 * Set up data on packet
801 ip6_copy_metadata(frag
, skb
);
802 skb_reserve(frag
, hroom
);
803 skb_put(frag
, len
+ hlen
+ sizeof(struct frag_hdr
));
804 skb_reset_network_header(frag
);
805 fh
= (struct frag_hdr
*)(skb_network_header(frag
) + hlen
);
806 frag
->transport_header
= (frag
->network_header
+ hlen
+
807 sizeof(struct frag_hdr
));
810 * Charge the memory for the fragment to any owner
814 skb_set_owner_w(frag
, skb
->sk
);
817 * Copy the packet header into the new buffer.
819 skb_copy_from_linear_data(skb
, skb_network_header(frag
), hlen
);
822 * Build fragment header.
824 fh
->nexthdr
= nexthdr
;
826 fh
->identification
= frag_id
;
829 * Copy a block of the IP datagram.
831 BUG_ON(skb_copy_bits(skb
, ptr
, skb_transport_header(frag
),
835 fh
->frag_off
= htons(offset
);
837 fh
->frag_off
|= htons(IP6_MF
);
838 ipv6_hdr(frag
)->payload_len
= htons(frag
->len
-
839 sizeof(struct ipv6hdr
));
845 * Put this fragment into the sending queue.
847 err
= output(net
, sk
, frag
);
851 IP6_INC_STATS(net
, ip6_dst_idev(skb_dst(skb
)),
852 IPSTATS_MIB_FRAGCREATES
);
854 IP6_INC_STATS(net
, ip6_dst_idev(skb_dst(skb
)),
855 IPSTATS_MIB_FRAGOKS
);
860 if (skb
->sk
&& dst_allfrag(skb_dst(skb
)))
861 sk_nocaps_add(skb
->sk
, NETIF_F_GSO_MASK
);
863 skb
->dev
= skb_dst(skb
)->dev
;
864 icmpv6_send(skb
, ICMPV6_PKT_TOOBIG
, 0, mtu
);
868 IP6_INC_STATS(net
, ip6_dst_idev(skb_dst(skb
)),
869 IPSTATS_MIB_FRAGFAILS
);
874 static inline int ip6_rt_check(const struct rt6key
*rt_key
,
875 const struct in6_addr
*fl_addr
,
876 const struct in6_addr
*addr_cache
)
878 return (rt_key
->plen
!= 128 || !ipv6_addr_equal(fl_addr
, &rt_key
->addr
)) &&
879 (!addr_cache
|| !ipv6_addr_equal(fl_addr
, addr_cache
));
882 static struct dst_entry
*ip6_sk_dst_check(struct sock
*sk
,
883 struct dst_entry
*dst
,
884 const struct flowi6
*fl6
)
886 struct ipv6_pinfo
*np
= inet6_sk(sk
);
892 if (dst
->ops
->family
!= AF_INET6
) {
897 rt
= (struct rt6_info
*)dst
;
898 /* Yes, checking route validity in not connected
899 * case is not very simple. Take into account,
900 * that we do not support routing by source, TOS,
901 * and MSG_DONTROUTE --ANK (980726)
903 * 1. ip6_rt_check(): If route was host route,
904 * check that cached destination is current.
905 * If it is network route, we still may
906 * check its validity using saved pointer
907 * to the last used address: daddr_cache.
908 * We do not want to save whole address now,
909 * (because main consumer of this service
910 * is tcp, which has not this problem),
911 * so that the last trick works only on connected
913 * 2. oif also should be the same.
915 if (ip6_rt_check(&rt
->rt6i_dst
, &fl6
->daddr
, np
->daddr_cache
) ||
916 #ifdef CONFIG_IPV6_SUBTREES
917 ip6_rt_check(&rt
->rt6i_src
, &fl6
->saddr
, np
->saddr_cache
) ||
919 (!(fl6
->flowi6_flags
& FLOWI_FLAG_SKIP_NH_OIF
) &&
920 (fl6
->flowi6_oif
&& fl6
->flowi6_oif
!= dst
->dev
->ifindex
))) {
929 static int ip6_dst_lookup_tail(struct net
*net
, const struct sock
*sk
,
930 struct dst_entry
**dst
, struct flowi6
*fl6
)
932 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
939 /* The correct way to handle this would be to do
940 * ip6_route_get_saddr, and then ip6_route_output; however,
941 * the route-specific preferred source forces the
942 * ip6_route_output call _before_ ip6_route_get_saddr.
944 * In source specific routing (no src=any default route),
945 * ip6_route_output will fail given src=any saddr, though, so
946 * that's why we try it again later.
948 if (ipv6_addr_any(&fl6
->saddr
) && (!*dst
|| !(*dst
)->error
)) {
950 bool had_dst
= *dst
!= NULL
;
953 *dst
= ip6_route_output(net
, sk
, fl6
);
954 rt
= (*dst
)->error
? NULL
: (struct rt6_info
*)*dst
;
955 err
= ip6_route_get_saddr(net
, rt
, &fl6
->daddr
,
956 sk
? inet6_sk(sk
)->srcprefs
: 0,
959 goto out_err_release
;
961 /* If we had an erroneous initial result, pretend it
962 * never existed and let the SA-enabled version take
965 if (!had_dst
&& (*dst
)->error
) {
971 flags
|= RT6_LOOKUP_F_IFACE
;
975 *dst
= ip6_route_output_flags(net
, sk
, fl6
, flags
);
979 goto out_err_release
;
981 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
983 * Here if the dst entry we've looked up
984 * has a neighbour entry that is in the INCOMPLETE
985 * state and the src address from the flow is
986 * marked as OPTIMISTIC, we release the found
987 * dst entry and replace it instead with the
988 * dst entry of the nexthop router
990 rt
= (struct rt6_info
*) *dst
;
992 n
= __ipv6_neigh_lookup_noref(rt
->dst
.dev
,
993 rt6_nexthop(rt
, &fl6
->daddr
));
994 err
= n
&& !(n
->nud_state
& NUD_VALID
) ? -EINVAL
: 0;
995 rcu_read_unlock_bh();
998 struct inet6_ifaddr
*ifp
;
999 struct flowi6 fl_gw6
;
1002 ifp
= ipv6_get_ifaddr(net
, &fl6
->saddr
,
1005 redirect
= (ifp
&& ifp
->flags
& IFA_F_OPTIMISTIC
);
1011 * We need to get the dst entry for the
1012 * default router instead
1015 memcpy(&fl_gw6
, fl6
, sizeof(struct flowi6
));
1016 memset(&fl_gw6
.daddr
, 0, sizeof(struct in6_addr
));
1017 *dst
= ip6_route_output(net
, sk
, &fl_gw6
);
1018 err
= (*dst
)->error
;
1020 goto out_err_release
;
1024 if (ipv6_addr_v4mapped(&fl6
->saddr
) &&
1025 !(ipv6_addr_v4mapped(&fl6
->daddr
) || ipv6_addr_any(&fl6
->daddr
)))
1026 return -EAFNOSUPPORT
;
1034 if (err
== -ENETUNREACH
)
1035 IP6_INC_STATS(net
, NULL
, IPSTATS_MIB_OUTNOROUTES
);
1040 * ip6_dst_lookup - perform route lookup on flow
1041 * @sk: socket which provides route info
1042 * @dst: pointer to dst_entry * for result
1043 * @fl6: flow to lookup
1045 * This function performs a route lookup on the given flow.
1047 * It returns zero on success, or a standard errno code on error.
1049 int ip6_dst_lookup(struct net
*net
, struct sock
*sk
, struct dst_entry
**dst
,
1053 return ip6_dst_lookup_tail(net
, sk
, dst
, fl6
);
1055 EXPORT_SYMBOL_GPL(ip6_dst_lookup
);
1058 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1059 * @sk: socket which provides route info
1060 * @fl6: flow to lookup
1061 * @final_dst: final destination address for ipsec lookup
1063 * This function performs a route lookup on the given flow.
1065 * It returns a valid dst pointer on success, or a pointer encoded
1068 struct dst_entry
*ip6_dst_lookup_flow(const struct sock
*sk
, struct flowi6
*fl6
,
1069 const struct in6_addr
*final_dst
)
1071 struct dst_entry
*dst
= NULL
;
1074 err
= ip6_dst_lookup_tail(sock_net(sk
), sk
, &dst
, fl6
);
1076 return ERR_PTR(err
);
1078 fl6
->daddr
= *final_dst
;
1080 return xfrm_lookup_route(sock_net(sk
), dst
, flowi6_to_flowi(fl6
), sk
, 0);
1082 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow
);
1085 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1086 * @sk: socket which provides the dst cache and route info
1087 * @fl6: flow to lookup
1088 * @final_dst: final destination address for ipsec lookup
1090 * This function performs a route lookup on the given flow with the
1091 * possibility of using the cached route in the socket if it is valid.
1092 * It will take the socket dst lock when operating on the dst cache.
1093 * As a result, this function can only be used in process context.
1095 * It returns a valid dst pointer on success, or a pointer encoded
1098 struct dst_entry
*ip6_sk_dst_lookup_flow(struct sock
*sk
, struct flowi6
*fl6
,
1099 const struct in6_addr
*final_dst
)
1101 struct dst_entry
*dst
= sk_dst_check(sk
, inet6_sk(sk
)->dst_cookie
);
1103 dst
= ip6_sk_dst_check(sk
, dst
, fl6
);
1105 dst
= ip6_dst_lookup_flow(sk
, fl6
, final_dst
);
1109 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow
);
1111 static inline int ip6_ufo_append_data(struct sock
*sk
,
1112 struct sk_buff_head
*queue
,
1113 int getfrag(void *from
, char *to
, int offset
, int len
,
1114 int odd
, struct sk_buff
*skb
),
1115 void *from
, int length
, int hh_len
, int fragheaderlen
,
1116 int exthdrlen
, int transhdrlen
, int mtu
,
1117 unsigned int flags
, const struct flowi6
*fl6
)
1120 struct sk_buff
*skb
;
1123 /* There is support for UDP large send offload by network
1124 * device, so create one single skb packet containing complete
1127 skb
= skb_peek_tail(queue
);
1129 skb
= sock_alloc_send_skb(sk
,
1130 hh_len
+ fragheaderlen
+ transhdrlen
+ 20,
1131 (flags
& MSG_DONTWAIT
), &err
);
1135 /* reserve space for Hardware header */
1136 skb_reserve(skb
, hh_len
);
1138 /* create space for UDP/IP header */
1139 skb_put(skb
, fragheaderlen
+ transhdrlen
);
1141 /* initialize network header pointer */
1142 skb_set_network_header(skb
, exthdrlen
);
1144 /* initialize protocol header pointer */
1145 skb
->transport_header
= skb
->network_header
+ fragheaderlen
;
1147 skb
->protocol
= htons(ETH_P_IPV6
);
1150 __skb_queue_tail(queue
, skb
);
1151 } else if (skb_is_gso(skb
)) {
1155 skb
->ip_summed
= CHECKSUM_PARTIAL
;
1156 /* Specify the length of each IPv6 datagram fragment.
1157 * It has to be a multiple of 8.
1159 skb_shinfo(skb
)->gso_size
= (mtu
- fragheaderlen
-
1160 sizeof(struct frag_hdr
)) & ~7;
1161 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
1162 skb_shinfo(skb
)->ip6_frag_id
= ipv6_select_ident(sock_net(sk
),
1167 return skb_append_datato_frags(sk
, skb
, getfrag
, from
,
1168 (length
- transhdrlen
));
1171 static inline struct ipv6_opt_hdr
*ip6_opt_dup(struct ipv6_opt_hdr
*src
,
1174 return src
? kmemdup(src
, (src
->hdrlen
+ 1) * 8, gfp
) : NULL
;
1177 static inline struct ipv6_rt_hdr
*ip6_rthdr_dup(struct ipv6_rt_hdr
*src
,
1180 return src
? kmemdup(src
, (src
->hdrlen
+ 1) * 8, gfp
) : NULL
;
1183 static void ip6_append_data_mtu(unsigned int *mtu
,
1185 unsigned int fragheaderlen
,
1186 struct sk_buff
*skb
,
1187 struct rt6_info
*rt
,
1188 unsigned int orig_mtu
)
1190 if (!(rt
->dst
.flags
& DST_XFRM_TUNNEL
)) {
1192 /* first fragment, reserve header_len */
1193 *mtu
= orig_mtu
- rt
->dst
.header_len
;
1197 * this fragment is not first, the headers
1198 * space is regarded as data space.
1202 *maxfraglen
= ((*mtu
- fragheaderlen
) & ~7)
1203 + fragheaderlen
- sizeof(struct frag_hdr
);
1207 static int ip6_setup_cork(struct sock
*sk
, struct inet_cork_full
*cork
,
1208 struct inet6_cork
*v6_cork
, struct ipcm6_cookie
*ipc6
,
1209 struct rt6_info
*rt
, struct flowi6
*fl6
)
1211 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1213 struct ipv6_txoptions
*opt
= ipc6
->opt
;
1219 if (WARN_ON(v6_cork
->opt
))
1222 v6_cork
->opt
= kzalloc(opt
->tot_len
, sk
->sk_allocation
);
1223 if (unlikely(!v6_cork
->opt
))
1226 v6_cork
->opt
->tot_len
= opt
->tot_len
;
1227 v6_cork
->opt
->opt_flen
= opt
->opt_flen
;
1228 v6_cork
->opt
->opt_nflen
= opt
->opt_nflen
;
1230 v6_cork
->opt
->dst0opt
= ip6_opt_dup(opt
->dst0opt
,
1232 if (opt
->dst0opt
&& !v6_cork
->opt
->dst0opt
)
1235 v6_cork
->opt
->dst1opt
= ip6_opt_dup(opt
->dst1opt
,
1237 if (opt
->dst1opt
&& !v6_cork
->opt
->dst1opt
)
1240 v6_cork
->opt
->hopopt
= ip6_opt_dup(opt
->hopopt
,
1242 if (opt
->hopopt
&& !v6_cork
->opt
->hopopt
)
1245 v6_cork
->opt
->srcrt
= ip6_rthdr_dup(opt
->srcrt
,
1247 if (opt
->srcrt
&& !v6_cork
->opt
->srcrt
)
1250 /* need source address above miyazawa*/
1253 cork
->base
.dst
= &rt
->dst
;
1254 cork
->fl
.u
.ip6
= *fl6
;
1255 v6_cork
->hop_limit
= ipc6
->hlimit
;
1256 v6_cork
->tclass
= ipc6
->tclass
;
1257 if (rt
->dst
.flags
& DST_XFRM_TUNNEL
)
1258 mtu
= np
->pmtudisc
>= IPV6_PMTUDISC_PROBE
?
1259 rt
->dst
.dev
->mtu
: dst_mtu(&rt
->dst
);
1261 mtu
= np
->pmtudisc
>= IPV6_PMTUDISC_PROBE
?
1262 rt
->dst
.dev
->mtu
: dst_mtu(rt
->dst
.path
);
1263 if (np
->frag_size
< mtu
) {
1265 mtu
= np
->frag_size
;
1267 cork
->base
.fragsize
= mtu
;
1268 if (dst_allfrag(rt
->dst
.path
))
1269 cork
->base
.flags
|= IPCORK_ALLFRAG
;
1270 cork
->base
.length
= 0;
1275 static int __ip6_append_data(struct sock
*sk
,
1277 struct sk_buff_head
*queue
,
1278 struct inet_cork
*cork
,
1279 struct inet6_cork
*v6_cork
,
1280 struct page_frag
*pfrag
,
1281 int getfrag(void *from
, char *to
, int offset
,
1282 int len
, int odd
, struct sk_buff
*skb
),
1283 void *from
, int length
, int transhdrlen
,
1284 unsigned int flags
, struct ipcm6_cookie
*ipc6
,
1285 const struct sockcm_cookie
*sockc
)
1287 struct sk_buff
*skb
, *skb_prev
= NULL
;
1288 unsigned int maxfraglen
, fragheaderlen
, mtu
, orig_mtu
;
1290 int dst_exthdrlen
= 0;
1297 struct rt6_info
*rt
= (struct rt6_info
*)cork
->dst
;
1298 struct ipv6_txoptions
*opt
= v6_cork
->opt
;
1299 int csummode
= CHECKSUM_NONE
;
1300 unsigned int maxnonfragsize
, headersize
;
1302 skb
= skb_peek_tail(queue
);
1304 exthdrlen
= opt
? opt
->opt_flen
: 0;
1305 dst_exthdrlen
= rt
->dst
.header_len
- rt
->rt6i_nfheader_len
;
1308 mtu
= cork
->fragsize
;
1311 hh_len
= LL_RESERVED_SPACE(rt
->dst
.dev
);
1313 fragheaderlen
= sizeof(struct ipv6hdr
) + rt
->rt6i_nfheader_len
+
1314 (opt
? opt
->opt_nflen
: 0);
1315 maxfraglen
= ((mtu
- fragheaderlen
) & ~7) + fragheaderlen
-
1316 sizeof(struct frag_hdr
);
1318 headersize
= sizeof(struct ipv6hdr
) +
1319 (opt
? opt
->opt_flen
+ opt
->opt_nflen
: 0) +
1320 (dst_allfrag(&rt
->dst
) ?
1321 sizeof(struct frag_hdr
) : 0) +
1322 rt
->rt6i_nfheader_len
;
1324 if (cork
->length
+ length
> mtu
- headersize
&& ipc6
->dontfrag
&&
1325 (sk
->sk_protocol
== IPPROTO_UDP
||
1326 sk
->sk_protocol
== IPPROTO_RAW
)) {
1327 ipv6_local_rxpmtu(sk
, fl6
, mtu
- headersize
+
1328 sizeof(struct ipv6hdr
));
1332 if (ip6_sk_ignore_df(sk
))
1333 maxnonfragsize
= sizeof(struct ipv6hdr
) + IPV6_MAXPLEN
;
1335 maxnonfragsize
= mtu
;
1337 if (cork
->length
+ length
> maxnonfragsize
- headersize
) {
1339 ipv6_local_error(sk
, EMSGSIZE
, fl6
,
1341 sizeof(struct ipv6hdr
));
1345 /* CHECKSUM_PARTIAL only with no extension headers and when
1346 * we are not going to fragment
1348 if (transhdrlen
&& sk
->sk_protocol
== IPPROTO_UDP
&&
1349 headersize
== sizeof(struct ipv6hdr
) &&
1350 length
<= mtu
- headersize
&&
1351 !(flags
& MSG_MORE
) &&
1352 rt
->dst
.dev
->features
& (NETIF_F_IPV6_CSUM
| NETIF_F_HW_CSUM
))
1353 csummode
= CHECKSUM_PARTIAL
;
1355 if (sk
->sk_type
== SOCK_DGRAM
|| sk
->sk_type
== SOCK_RAW
) {
1356 sock_tx_timestamp(sk
, sockc
->tsflags
, &tx_flags
);
1357 if (tx_flags
& SKBTX_ANY_SW_TSTAMP
&&
1358 sk
->sk_tsflags
& SOF_TIMESTAMPING_OPT_ID
)
1359 tskey
= sk
->sk_tskey
++;
1363 * Let's try using as much space as possible.
1364 * Use MTU if total length of the message fits into the MTU.
1365 * Otherwise, we need to reserve fragment header and
1366 * fragment alignment (= 8-15 octects, in total).
1368 * Note that we may need to "move" the data from the tail of
1369 * of the buffer to the new fragment when we split
1372 * FIXME: It may be fragmented into multiple chunks
1373 * at once if non-fragmentable extension headers
1378 cork
->length
+= length
;
1379 if ((((length
+ fragheaderlen
) > mtu
) ||
1380 (skb
&& skb_is_gso(skb
))) &&
1381 (sk
->sk_protocol
== IPPROTO_UDP
) &&
1382 (rt
->dst
.dev
->features
& NETIF_F_UFO
) && !rt
->dst
.header_len
&&
1383 (sk
->sk_type
== SOCK_DGRAM
) && !udp_get_no_check6_tx(sk
)) {
1384 err
= ip6_ufo_append_data(sk
, queue
, getfrag
, from
, length
,
1385 hh_len
, fragheaderlen
, exthdrlen
,
1386 transhdrlen
, mtu
, flags
, fl6
);
1395 while (length
> 0) {
1396 /* Check if the remaining data fits into current packet. */
1397 copy
= (cork
->length
<= mtu
&& !(cork
->flags
& IPCORK_ALLFRAG
) ? mtu
: maxfraglen
) - skb
->len
;
1399 copy
= maxfraglen
- skb
->len
;
1403 unsigned int datalen
;
1404 unsigned int fraglen
;
1405 unsigned int fraggap
;
1406 unsigned int alloclen
;
1408 /* There's no room in the current skb */
1410 fraggap
= skb
->len
- maxfraglen
;
1413 /* update mtu and maxfraglen if necessary */
1414 if (!skb
|| !skb_prev
)
1415 ip6_append_data_mtu(&mtu
, &maxfraglen
,
1416 fragheaderlen
, skb
, rt
,
1422 * If remaining data exceeds the mtu,
1423 * we know we need more fragment(s).
1425 datalen
= length
+ fraggap
;
1427 if (datalen
> (cork
->length
<= mtu
&& !(cork
->flags
& IPCORK_ALLFRAG
) ? mtu
: maxfraglen
) - fragheaderlen
)
1428 datalen
= maxfraglen
- fragheaderlen
- rt
->dst
.trailer_len
;
1429 if ((flags
& MSG_MORE
) &&
1430 !(rt
->dst
.dev
->features
&NETIF_F_SG
))
1433 alloclen
= datalen
+ fragheaderlen
;
1435 alloclen
+= dst_exthdrlen
;
1437 if (datalen
!= length
+ fraggap
) {
1439 * this is not the last fragment, the trailer
1440 * space is regarded as data space.
1442 datalen
+= rt
->dst
.trailer_len
;
1445 alloclen
+= rt
->dst
.trailer_len
;
1446 fraglen
= datalen
+ fragheaderlen
;
1449 * We just reserve space for fragment header.
1450 * Note: this may be overallocation if the message
1451 * (without MSG_MORE) fits into the MTU.
1453 alloclen
+= sizeof(struct frag_hdr
);
1456 skb
= sock_alloc_send_skb(sk
,
1458 (flags
& MSG_DONTWAIT
), &err
);
1461 if (atomic_read(&sk
->sk_wmem_alloc
) <=
1463 skb
= sock_wmalloc(sk
,
1464 alloclen
+ hh_len
, 1,
1472 * Fill in the control structures
1474 skb
->protocol
= htons(ETH_P_IPV6
);
1475 skb
->ip_summed
= csummode
;
1477 /* reserve for fragmentation and ipsec header */
1478 skb_reserve(skb
, hh_len
+ sizeof(struct frag_hdr
) +
1481 /* Only the initial fragment is time stamped */
1482 skb_shinfo(skb
)->tx_flags
= tx_flags
;
1484 skb_shinfo(skb
)->tskey
= tskey
;
1488 * Find where to start putting bytes
1490 data
= skb_put(skb
, fraglen
);
1491 skb_set_network_header(skb
, exthdrlen
);
1492 data
+= fragheaderlen
;
1493 skb
->transport_header
= (skb
->network_header
+
1496 skb
->csum
= skb_copy_and_csum_bits(
1497 skb_prev
, maxfraglen
,
1498 data
+ transhdrlen
, fraggap
, 0);
1499 skb_prev
->csum
= csum_sub(skb_prev
->csum
,
1502 pskb_trim_unique(skb_prev
, maxfraglen
);
1504 copy
= datalen
- transhdrlen
- fraggap
;
1510 } else if (copy
> 0 && getfrag(from
, data
+ transhdrlen
, offset
, copy
, fraggap
, skb
) < 0) {
1517 length
-= datalen
- fraggap
;
1523 * Put the packet on the pending queue
1525 __skb_queue_tail(queue
, skb
);
1532 if (!(rt
->dst
.dev
->features
&NETIF_F_SG
)) {
1536 if (getfrag(from
, skb_put(skb
, copy
),
1537 offset
, copy
, off
, skb
) < 0) {
1538 __skb_trim(skb
, off
);
1543 int i
= skb_shinfo(skb
)->nr_frags
;
1546 if (!sk_page_frag_refill(sk
, pfrag
))
1549 if (!skb_can_coalesce(skb
, i
, pfrag
->page
,
1552 if (i
== MAX_SKB_FRAGS
)
1555 __skb_fill_page_desc(skb
, i
, pfrag
->page
,
1557 skb_shinfo(skb
)->nr_frags
= ++i
;
1558 get_page(pfrag
->page
);
1560 copy
= min_t(int, copy
, pfrag
->size
- pfrag
->offset
);
1562 page_address(pfrag
->page
) + pfrag
->offset
,
1563 offset
, copy
, skb
->len
, skb
) < 0)
1566 pfrag
->offset
+= copy
;
1567 skb_frag_size_add(&skb_shinfo(skb
)->frags
[i
- 1], copy
);
1569 skb
->data_len
+= copy
;
1570 skb
->truesize
+= copy
;
1571 atomic_add(copy
, &sk
->sk_wmem_alloc
);
1582 cork
->length
-= length
;
1583 IP6_INC_STATS(sock_net(sk
), rt
->rt6i_idev
, IPSTATS_MIB_OUTDISCARDS
);
1587 int ip6_append_data(struct sock
*sk
,
1588 int getfrag(void *from
, char *to
, int offset
, int len
,
1589 int odd
, struct sk_buff
*skb
),
1590 void *from
, int length
, int transhdrlen
,
1591 struct ipcm6_cookie
*ipc6
, struct flowi6
*fl6
,
1592 struct rt6_info
*rt
, unsigned int flags
,
1593 const struct sockcm_cookie
*sockc
)
1595 struct inet_sock
*inet
= inet_sk(sk
);
1596 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1600 if (flags
&MSG_PROBE
)
1602 if (skb_queue_empty(&sk
->sk_write_queue
)) {
1606 err
= ip6_setup_cork(sk
, &inet
->cork
, &np
->cork
,
1611 exthdrlen
= (ipc6
->opt
? ipc6
->opt
->opt_flen
: 0);
1612 length
+= exthdrlen
;
1613 transhdrlen
+= exthdrlen
;
1615 fl6
= &inet
->cork
.fl
.u
.ip6
;
1619 return __ip6_append_data(sk
, fl6
, &sk
->sk_write_queue
, &inet
->cork
.base
,
1620 &np
->cork
, sk_page_frag(sk
), getfrag
,
1621 from
, length
, transhdrlen
, flags
, ipc6
, sockc
);
1623 EXPORT_SYMBOL_GPL(ip6_append_data
);
1625 static void ip6_cork_release(struct inet_cork_full
*cork
,
1626 struct inet6_cork
*v6_cork
)
1629 kfree(v6_cork
->opt
->dst0opt
);
1630 kfree(v6_cork
->opt
->dst1opt
);
1631 kfree(v6_cork
->opt
->hopopt
);
1632 kfree(v6_cork
->opt
->srcrt
);
1633 kfree(v6_cork
->opt
);
1634 v6_cork
->opt
= NULL
;
1637 if (cork
->base
.dst
) {
1638 dst_release(cork
->base
.dst
);
1639 cork
->base
.dst
= NULL
;
1640 cork
->base
.flags
&= ~IPCORK_ALLFRAG
;
1642 memset(&cork
->fl
, 0, sizeof(cork
->fl
));
1645 struct sk_buff
*__ip6_make_skb(struct sock
*sk
,
1646 struct sk_buff_head
*queue
,
1647 struct inet_cork_full
*cork
,
1648 struct inet6_cork
*v6_cork
)
1650 struct sk_buff
*skb
, *tmp_skb
;
1651 struct sk_buff
**tail_skb
;
1652 struct in6_addr final_dst_buf
, *final_dst
= &final_dst_buf
;
1653 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1654 struct net
*net
= sock_net(sk
);
1655 struct ipv6hdr
*hdr
;
1656 struct ipv6_txoptions
*opt
= v6_cork
->opt
;
1657 struct rt6_info
*rt
= (struct rt6_info
*)cork
->base
.dst
;
1658 struct flowi6
*fl6
= &cork
->fl
.u
.ip6
;
1659 unsigned char proto
= fl6
->flowi6_proto
;
1661 skb
= __skb_dequeue(queue
);
1664 tail_skb
= &(skb_shinfo(skb
)->frag_list
);
1666 /* move skb->data to ip header from ext header */
1667 if (skb
->data
< skb_network_header(skb
))
1668 __skb_pull(skb
, skb_network_offset(skb
));
1669 while ((tmp_skb
= __skb_dequeue(queue
)) != NULL
) {
1670 __skb_pull(tmp_skb
, skb_network_header_len(skb
));
1671 *tail_skb
= tmp_skb
;
1672 tail_skb
= &(tmp_skb
->next
);
1673 skb
->len
+= tmp_skb
->len
;
1674 skb
->data_len
+= tmp_skb
->len
;
1675 skb
->truesize
+= tmp_skb
->truesize
;
1676 tmp_skb
->destructor
= NULL
;
1680 /* Allow local fragmentation. */
1681 skb
->ignore_df
= ip6_sk_ignore_df(sk
);
1683 *final_dst
= fl6
->daddr
;
1684 __skb_pull(skb
, skb_network_header_len(skb
));
1685 if (opt
&& opt
->opt_flen
)
1686 ipv6_push_frag_opts(skb
, opt
, &proto
);
1687 if (opt
&& opt
->opt_nflen
)
1688 ipv6_push_nfrag_opts(skb
, opt
, &proto
, &final_dst
, &fl6
->saddr
);
1690 skb_push(skb
, sizeof(struct ipv6hdr
));
1691 skb_reset_network_header(skb
);
1692 hdr
= ipv6_hdr(skb
);
1694 ip6_flow_hdr(hdr
, v6_cork
->tclass
,
1695 ip6_make_flowlabel(net
, skb
, fl6
->flowlabel
,
1696 np
->autoflowlabel
, fl6
));
1697 hdr
->hop_limit
= v6_cork
->hop_limit
;
1698 hdr
->nexthdr
= proto
;
1699 hdr
->saddr
= fl6
->saddr
;
1700 hdr
->daddr
= *final_dst
;
1702 skb
->priority
= sk
->sk_priority
;
1703 skb
->mark
= sk
->sk_mark
;
1705 skb_dst_set(skb
, dst_clone(&rt
->dst
));
1706 IP6_UPD_PO_STATS(net
, rt
->rt6i_idev
, IPSTATS_MIB_OUT
, skb
->len
);
1707 if (proto
== IPPROTO_ICMPV6
) {
1708 struct inet6_dev
*idev
= ip6_dst_idev(skb_dst(skb
));
1710 ICMP6MSGOUT_INC_STATS(net
, idev
, icmp6_hdr(skb
)->icmp6_type
);
1711 ICMP6_INC_STATS(net
, idev
, ICMP6_MIB_OUTMSGS
);
1714 ip6_cork_release(cork
, v6_cork
);
1719 int ip6_send_skb(struct sk_buff
*skb
)
1721 struct net
*net
= sock_net(skb
->sk
);
1722 struct rt6_info
*rt
= (struct rt6_info
*)skb_dst(skb
);
1725 err
= ip6_local_out(net
, skb
->sk
, skb
);
1728 err
= net_xmit_errno(err
);
1730 IP6_INC_STATS(net
, rt
->rt6i_idev
,
1731 IPSTATS_MIB_OUTDISCARDS
);
1737 int ip6_push_pending_frames(struct sock
*sk
)
1739 struct sk_buff
*skb
;
1741 skb
= ip6_finish_skb(sk
);
1745 return ip6_send_skb(skb
);
1747 EXPORT_SYMBOL_GPL(ip6_push_pending_frames
);
1749 static void __ip6_flush_pending_frames(struct sock
*sk
,
1750 struct sk_buff_head
*queue
,
1751 struct inet_cork_full
*cork
,
1752 struct inet6_cork
*v6_cork
)
1754 struct sk_buff
*skb
;
1756 while ((skb
= __skb_dequeue_tail(queue
)) != NULL
) {
1758 IP6_INC_STATS(sock_net(sk
), ip6_dst_idev(skb_dst(skb
)),
1759 IPSTATS_MIB_OUTDISCARDS
);
1763 ip6_cork_release(cork
, v6_cork
);
1766 void ip6_flush_pending_frames(struct sock
*sk
)
1768 __ip6_flush_pending_frames(sk
, &sk
->sk_write_queue
,
1769 &inet_sk(sk
)->cork
, &inet6_sk(sk
)->cork
);
1771 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames
);
1773 struct sk_buff
*ip6_make_skb(struct sock
*sk
,
1774 int getfrag(void *from
, char *to
, int offset
,
1775 int len
, int odd
, struct sk_buff
*skb
),
1776 void *from
, int length
, int transhdrlen
,
1777 struct ipcm6_cookie
*ipc6
, struct flowi6
*fl6
,
1778 struct rt6_info
*rt
, unsigned int flags
,
1779 const struct sockcm_cookie
*sockc
)
1781 struct inet_cork_full cork
;
1782 struct inet6_cork v6_cork
;
1783 struct sk_buff_head queue
;
1784 int exthdrlen
= (ipc6
->opt
? ipc6
->opt
->opt_flen
: 0);
1787 if (flags
& MSG_PROBE
)
1790 __skb_queue_head_init(&queue
);
1792 cork
.base
.flags
= 0;
1794 cork
.base
.opt
= NULL
;
1796 err
= ip6_setup_cork(sk
, &cork
, &v6_cork
, ipc6
, rt
, fl6
);
1798 return ERR_PTR(err
);
1800 if (ipc6
->dontfrag
< 0)
1801 ipc6
->dontfrag
= inet6_sk(sk
)->dontfrag
;
1803 err
= __ip6_append_data(sk
, fl6
, &queue
, &cork
.base
, &v6_cork
,
1804 ¤t
->task_frag
, getfrag
, from
,
1805 length
+ exthdrlen
, transhdrlen
+ exthdrlen
,
1806 flags
, ipc6
, sockc
);
1808 __ip6_flush_pending_frames(sk
, &queue
, &cork
, &v6_cork
);
1809 return ERR_PTR(err
);
1812 return __ip6_make_skb(sk
, &queue
, &cork
, &v6_cork
);