3 * Linux INET6 implementation
6 * Pedro Roque <roque@di.fc.ul.pt>
8 * Based on linux/ipv4/udp.c
11 * Hideaki YOSHIFUJI : sin6_scope_id support
12 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
13 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
14 * a single port at the same time.
15 * Kazunori MIYAZAWA @USAGI: change process style to use ip6_append_data
16 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/udp6 to seq_file.
18 * This program is free software; you can redistribute it and/or
19 * modify it under the terms of the GNU General Public License
20 * as published by the Free Software Foundation; either version
21 * 2 of the License, or (at your option) any later version.
24 #include <linux/errno.h>
25 #include <linux/types.h>
26 #include <linux/socket.h>
27 #include <linux/sockios.h>
28 #include <linux/net.h>
29 #include <linux/in6.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_arp.h>
32 #include <linux/ipv6.h>
33 #include <linux/icmpv6.h>
34 #include <linux/init.h>
35 #include <linux/module.h>
36 #include <linux/skbuff.h>
37 #include <asm/uaccess.h>
39 #include <net/ndisc.h>
40 #include <net/protocol.h>
41 #include <net/transp_v6.h>
42 #include <net/ip6_route.h>
44 #include <net/tcp_states.h>
45 #include <net/ip6_checksum.h>
48 #include <linux/proc_fs.h>
49 #include <linux/seq_file.h>
52 int ipv6_rcv_saddr_equal(const struct sock
*sk
, const struct sock
*sk2
)
54 const struct in6_addr
*sk_rcv_saddr6
= &inet6_sk(sk
)->rcv_saddr
;
55 const struct in6_addr
*sk2_rcv_saddr6
= inet6_rcv_saddr(sk2
);
56 __be32 sk1_rcv_saddr
= inet_sk(sk
)->inet_rcv_saddr
;
57 __be32 sk2_rcv_saddr
= inet_rcv_saddr(sk2
);
58 int sk_ipv6only
= ipv6_only_sock(sk
);
59 int sk2_ipv6only
= inet_v6_ipv6only(sk2
);
60 int addr_type
= ipv6_addr_type(sk_rcv_saddr6
);
61 int addr_type2
= sk2_rcv_saddr6
? ipv6_addr_type(sk2_rcv_saddr6
) : IPV6_ADDR_MAPPED
;
63 /* if both are mapped, treat as IPv4 */
64 if (addr_type
== IPV6_ADDR_MAPPED
&& addr_type2
== IPV6_ADDR_MAPPED
)
65 return (!sk2_ipv6only
&&
66 (!sk1_rcv_saddr
|| !sk2_rcv_saddr
||
67 sk1_rcv_saddr
== sk2_rcv_saddr
));
69 if (addr_type2
== IPV6_ADDR_ANY
&&
70 !(sk2_ipv6only
&& addr_type
== IPV6_ADDR_MAPPED
))
73 if (addr_type
== IPV6_ADDR_ANY
&&
74 !(sk_ipv6only
&& addr_type2
== IPV6_ADDR_MAPPED
))
78 ipv6_addr_equal(sk_rcv_saddr6
, sk2_rcv_saddr6
))
84 static unsigned int udp6_portaddr_hash(struct net
*net
,
85 const struct in6_addr
*addr6
,
88 unsigned int hash
, mix
= net_hash_mix(net
);
90 if (ipv6_addr_any(addr6
))
91 hash
= jhash_1word(0, mix
);
92 else if (ipv6_addr_v4mapped(addr6
))
93 hash
= jhash_1word(addr6
->s6_addr32
[3], mix
);
95 hash
= jhash2(addr6
->s6_addr32
, 4, mix
);
101 int udp_v6_get_port(struct sock
*sk
, unsigned short snum
)
103 unsigned int hash2_nulladdr
=
104 udp6_portaddr_hash(sock_net(sk
), &in6addr_any
, snum
);
105 unsigned int hash2_partial
=
106 udp6_portaddr_hash(sock_net(sk
), &inet6_sk(sk
)->rcv_saddr
, 0);
108 /* precompute partial secondary hash */
109 udp_sk(sk
)->udp_portaddr_hash
= hash2_partial
;
110 return udp_lib_get_port(sk
, snum
, ipv6_rcv_saddr_equal
, hash2_nulladdr
);
113 static inline int compute_score(struct sock
*sk
, struct net
*net
,
115 struct in6_addr
*saddr
, __be16 sport
,
116 struct in6_addr
*daddr
, __be16 dport
,
121 if (net_eq(sock_net(sk
), net
) && udp_sk(sk
)->udp_port_hash
== hnum
&&
122 sk
->sk_family
== PF_INET6
) {
123 struct ipv6_pinfo
*np
= inet6_sk(sk
);
124 struct inet_sock
*inet
= inet_sk(sk
);
127 if (inet
->inet_dport
) {
128 if (inet
->inet_dport
!= sport
)
132 if (!ipv6_addr_any(&np
->rcv_saddr
)) {
133 if (!ipv6_addr_equal(&np
->rcv_saddr
, daddr
))
137 if (!ipv6_addr_any(&np
->daddr
)) {
138 if (!ipv6_addr_equal(&np
->daddr
, saddr
))
142 if (sk
->sk_bound_dev_if
) {
143 if (sk
->sk_bound_dev_if
!= dif
)
151 #define SCORE2_MAX (1 + 1 + 1)
152 static inline int compute_score2(struct sock
*sk
, struct net
*net
,
153 const struct in6_addr
*saddr
, __be16 sport
,
154 const struct in6_addr
*daddr
, unsigned short hnum
,
159 if (net_eq(sock_net(sk
), net
) && udp_sk(sk
)->udp_port_hash
== hnum
&&
160 sk
->sk_family
== PF_INET6
) {
161 struct ipv6_pinfo
*np
= inet6_sk(sk
);
162 struct inet_sock
*inet
= inet_sk(sk
);
164 if (!ipv6_addr_equal(&np
->rcv_saddr
, daddr
))
167 if (inet
->inet_dport
) {
168 if (inet
->inet_dport
!= sport
)
172 if (!ipv6_addr_any(&np
->daddr
)) {
173 if (!ipv6_addr_equal(&np
->daddr
, saddr
))
177 if (sk
->sk_bound_dev_if
) {
178 if (sk
->sk_bound_dev_if
!= dif
)
187 /* called with read_rcu_lock() */
188 static struct sock
*udp6_lib_lookup2(struct net
*net
,
189 const struct in6_addr
*saddr
, __be16 sport
,
190 const struct in6_addr
*daddr
, unsigned int hnum
, int dif
,
191 struct udp_hslot
*hslot2
, unsigned int slot2
)
193 struct sock
*sk
, *result
;
194 struct hlist_nulls_node
*node
;
200 udp_portaddr_for_each_entry_rcu(sk
, node
, &hslot2
->head
) {
201 score
= compute_score2(sk
, net
, saddr
, sport
,
203 if (score
> badness
) {
206 if (score
== SCORE2_MAX
)
211 * if the nulls value we got at the end of this lookup is
212 * not the expected one, we must restart lookup.
213 * We probably met an item that was moved to another chain.
215 if (get_nulls_value(node
) != slot2
)
220 if (unlikely(!atomic_inc_not_zero(&result
->sk_refcnt
)))
222 else if (unlikely(compute_score2(result
, net
, saddr
, sport
,
223 daddr
, hnum
, dif
) < badness
)) {
231 static struct sock
*__udp6_lib_lookup(struct net
*net
,
232 struct in6_addr
*saddr
, __be16 sport
,
233 struct in6_addr
*daddr
, __be16 dport
,
234 int dif
, struct udp_table
*udptable
)
236 struct sock
*sk
, *result
;
237 struct hlist_nulls_node
*node
;
238 unsigned short hnum
= ntohs(dport
);
239 unsigned int hash2
, slot2
, slot
= udp_hashfn(net
, hnum
, udptable
->mask
);
240 struct udp_hslot
*hslot2
, *hslot
= &udptable
->hash
[slot
];
244 if (hslot
->count
> 10) {
245 hash2
= udp6_portaddr_hash(net
, daddr
, hnum
);
246 slot2
= hash2
& udptable
->mask
;
247 hslot2
= &udptable
->hash2
[slot2
];
248 if (hslot
->count
< hslot2
->count
)
251 result
= udp6_lib_lookup2(net
, saddr
, sport
,
255 hash2
= udp6_portaddr_hash(net
, &in6addr_any
, hnum
);
256 slot2
= hash2
& udptable
->mask
;
257 hslot2
= &udptable
->hash2
[slot2
];
258 if (hslot
->count
< hslot2
->count
)
261 result
= udp6_lib_lookup2(net
, &in6addr_any
, sport
,
271 sk_nulls_for_each_rcu(sk
, node
, &hslot
->head
) {
272 score
= compute_score(sk
, net
, hnum
, saddr
, sport
, daddr
, dport
, dif
);
273 if (score
> badness
) {
279 * if the nulls value we got at the end of this lookup is
280 * not the expected one, we must restart lookup.
281 * We probably met an item that was moved to another chain.
283 if (get_nulls_value(node
) != slot
)
287 if (unlikely(!atomic_inc_not_zero(&result
->sk_refcnt
)))
289 else if (unlikely(compute_score(result
, net
, hnum
, saddr
, sport
,
290 daddr
, dport
, dif
) < badness
)) {
299 static struct sock
*__udp6_lib_lookup_skb(struct sk_buff
*skb
,
300 __be16 sport
, __be16 dport
,
301 struct udp_table
*udptable
)
304 struct ipv6hdr
*iph
= ipv6_hdr(skb
);
306 if (unlikely(sk
= skb_steal_sock(skb
)))
308 return __udp6_lib_lookup(dev_net(skb_dst(skb
)->dev
), &iph
->saddr
, sport
,
309 &iph
->daddr
, dport
, inet6_iif(skb
),
314 * This should be easy, if there is something there we
315 * return it, otherwise we block.
318 int udpv6_recvmsg(struct kiocb
*iocb
, struct sock
*sk
,
319 struct msghdr
*msg
, size_t len
,
320 int noblock
, int flags
, int *addr_len
)
322 struct ipv6_pinfo
*np
= inet6_sk(sk
);
323 struct inet_sock
*inet
= inet_sk(sk
);
325 unsigned int ulen
, copied
;
328 int is_udplite
= IS_UDPLITE(sk
);
332 *addr_len
=sizeof(struct sockaddr_in6
);
334 if (flags
& MSG_ERRQUEUE
)
335 return ipv6_recv_error(sk
, msg
, len
);
338 skb
= __skb_recv_datagram(sk
, flags
| (noblock
? MSG_DONTWAIT
: 0),
343 ulen
= skb
->len
- sizeof(struct udphdr
);
347 else if (copied
< ulen
)
348 msg
->msg_flags
|= MSG_TRUNC
;
350 is_udp4
= (skb
->protocol
== htons(ETH_P_IP
));
353 * If checksum is needed at all, try to do it while copying the
354 * data. If the data is truncated, or if we only want a partial
355 * coverage checksum (UDP-Lite), do it before the copy.
358 if (copied
< ulen
|| UDP_SKB_CB(skb
)->partial_cov
) {
359 if (udp_lib_checksum_complete(skb
))
363 if (skb_csum_unnecessary(skb
))
364 err
= skb_copy_datagram_iovec(skb
, sizeof(struct udphdr
),
365 msg
->msg_iov
, copied
);
367 err
= skb_copy_and_csum_datagram_iovec(skb
, sizeof(struct udphdr
), msg
->msg_iov
);
376 UDP_INC_STATS_USER(sock_net(sk
),
377 UDP_MIB_INDATAGRAMS
, is_udplite
);
379 UDP6_INC_STATS_USER(sock_net(sk
),
380 UDP_MIB_INDATAGRAMS
, is_udplite
);
383 sock_recv_ts_and_drops(msg
, sk
, skb
);
385 /* Copy the address. */
387 struct sockaddr_in6
*sin6
;
389 sin6
= (struct sockaddr_in6
*) msg
->msg_name
;
390 sin6
->sin6_family
= AF_INET6
;
391 sin6
->sin6_port
= udp_hdr(skb
)->source
;
392 sin6
->sin6_flowinfo
= 0;
393 sin6
->sin6_scope_id
= 0;
396 ipv6_addr_set_v4mapped(ip_hdr(skb
)->saddr
,
399 ipv6_addr_copy(&sin6
->sin6_addr
,
400 &ipv6_hdr(skb
)->saddr
);
401 if (ipv6_addr_type(&sin6
->sin6_addr
) & IPV6_ADDR_LINKLOCAL
)
402 sin6
->sin6_scope_id
= IP6CB(skb
)->iif
;
407 if (inet
->cmsg_flags
)
408 ip_cmsg_recv(msg
, skb
);
411 datagram_recv_ctl(sk
, msg
, skb
);
415 if (flags
& MSG_TRUNC
)
419 skb_free_datagram_locked(sk
, skb
);
425 if (!skb_kill_datagram(sk
, skb
, flags
)) {
427 UDP_INC_STATS_USER(sock_net(sk
),
428 UDP_MIB_INERRORS
, is_udplite
);
430 UDP6_INC_STATS_USER(sock_net(sk
),
431 UDP_MIB_INERRORS
, is_udplite
);
435 if (flags
& MSG_DONTWAIT
)
440 void __udp6_lib_err(struct sk_buff
*skb
, struct inet6_skb_parm
*opt
,
441 u8 type
, u8 code
, int offset
, __be32 info
,
442 struct udp_table
*udptable
)
444 struct ipv6_pinfo
*np
;
445 struct ipv6hdr
*hdr
= (struct ipv6hdr
*)skb
->data
;
446 struct in6_addr
*saddr
= &hdr
->saddr
;
447 struct in6_addr
*daddr
= &hdr
->daddr
;
448 struct udphdr
*uh
= (struct udphdr
*)(skb
->data
+offset
);
452 sk
= __udp6_lib_lookup(dev_net(skb
->dev
), daddr
, uh
->dest
,
453 saddr
, uh
->source
, inet6_iif(skb
), udptable
);
459 if (!icmpv6_err_convert(type
, code
, &err
) && !np
->recverr
)
462 if (sk
->sk_state
!= TCP_ESTABLISHED
&& !np
->recverr
)
466 ipv6_icmp_error(sk
, skb
, err
, uh
->dest
, ntohl(info
), (u8
*)(uh
+1));
469 sk
->sk_error_report(sk
);
474 static __inline__
void udpv6_err(struct sk_buff
*skb
,
475 struct inet6_skb_parm
*opt
, u8 type
,
476 u8 code
, int offset
, __be32 info
)
478 __udp6_lib_err(skb
, opt
, type
, code
, offset
, info
, &udp_table
);
481 int udpv6_queue_rcv_skb(struct sock
* sk
, struct sk_buff
*skb
)
483 struct udp_sock
*up
= udp_sk(sk
);
485 int is_udplite
= IS_UDPLITE(sk
);
487 if (!xfrm6_policy_check(sk
, XFRM_POLICY_IN
, skb
))
491 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
493 if ((is_udplite
& UDPLITE_RECV_CC
) && UDP_SKB_CB(skb
)->partial_cov
) {
495 if (up
->pcrlen
== 0) { /* full coverage was set */
496 LIMIT_NETDEBUG(KERN_WARNING
"UDPLITE6: partial coverage"
497 " %d while full coverage %d requested\n",
498 UDP_SKB_CB(skb
)->cscov
, skb
->len
);
501 if (UDP_SKB_CB(skb
)->cscov
< up
->pcrlen
) {
502 LIMIT_NETDEBUG(KERN_WARNING
"UDPLITE6: coverage %d "
503 "too small, need min %d\n",
504 UDP_SKB_CB(skb
)->cscov
, up
->pcrlen
);
510 if (udp_lib_checksum_complete(skb
))
514 if ((rc
= sock_queue_rcv_skb(sk
, skb
)) < 0) {
515 /* Note that an ENOMEM error is charged twice */
517 UDP6_INC_STATS_BH(sock_net(sk
),
518 UDP_MIB_RCVBUFERRORS
, is_udplite
);
519 goto drop_no_sk_drops_inc
;
524 atomic_inc(&sk
->sk_drops
);
525 drop_no_sk_drops_inc
:
526 UDP6_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
531 static struct sock
*udp_v6_mcast_next(struct net
*net
, struct sock
*sk
,
532 __be16 loc_port
, struct in6_addr
*loc_addr
,
533 __be16 rmt_port
, struct in6_addr
*rmt_addr
,
536 struct hlist_nulls_node
*node
;
538 unsigned short num
= ntohs(loc_port
);
540 sk_nulls_for_each_from(s
, node
) {
541 struct inet_sock
*inet
= inet_sk(s
);
543 if (!net_eq(sock_net(s
), net
))
546 if (udp_sk(s
)->udp_port_hash
== num
&&
547 s
->sk_family
== PF_INET6
) {
548 struct ipv6_pinfo
*np
= inet6_sk(s
);
549 if (inet
->inet_dport
) {
550 if (inet
->inet_dport
!= rmt_port
)
553 if (!ipv6_addr_any(&np
->daddr
) &&
554 !ipv6_addr_equal(&np
->daddr
, rmt_addr
))
557 if (s
->sk_bound_dev_if
&& s
->sk_bound_dev_if
!= dif
)
560 if (!ipv6_addr_any(&np
->rcv_saddr
)) {
561 if (!ipv6_addr_equal(&np
->rcv_saddr
, loc_addr
))
564 if (!inet6_mc_check(s
, loc_addr
, rmt_addr
))
572 static void flush_stack(struct sock
**stack
, unsigned int count
,
573 struct sk_buff
*skb
, unsigned int final
)
577 struct sk_buff
*skb1
;
579 for (i
= 0; i
< count
; i
++) {
580 skb1
= (i
== final
) ? skb
: skb_clone(skb
, GFP_ATOMIC
);
585 if (!sock_owned_by_user(sk
))
586 udpv6_queue_rcv_skb(sk
, skb1
);
588 sk_add_backlog(sk
, skb1
);
591 atomic_inc(&sk
->sk_drops
);
592 UDP6_INC_STATS_BH(sock_net(sk
),
593 UDP_MIB_RCVBUFERRORS
, IS_UDPLITE(sk
));
594 UDP6_INC_STATS_BH(sock_net(sk
),
595 UDP_MIB_INERRORS
, IS_UDPLITE(sk
));
600 * Note: called only from the BH handler context,
601 * so we don't need to lock the hashes.
603 static int __udp6_lib_mcast_deliver(struct net
*net
, struct sk_buff
*skb
,
604 struct in6_addr
*saddr
, struct in6_addr
*daddr
,
605 struct udp_table
*udptable
)
607 struct sock
*sk
, *stack
[256 / sizeof(struct sock
*)];
608 const struct udphdr
*uh
= udp_hdr(skb
);
609 struct udp_hslot
*hslot
= udp_hashslot(udptable
, net
, ntohs(uh
->dest
));
611 unsigned int i
, count
= 0;
613 spin_lock(&hslot
->lock
);
614 sk
= sk_nulls_head(&hslot
->head
);
615 dif
= inet6_iif(skb
);
616 sk
= udp_v6_mcast_next(net
, sk
, uh
->dest
, daddr
, uh
->source
, saddr
, dif
);
619 sk
= udp_v6_mcast_next(net
, sk_nulls_next(sk
), uh
->dest
, daddr
,
620 uh
->source
, saddr
, dif
);
621 if (unlikely(count
== ARRAY_SIZE(stack
))) {
624 flush_stack(stack
, count
, skb
, ~0);
629 * before releasing the lock, we must take reference on sockets
631 for (i
= 0; i
< count
; i
++)
634 spin_unlock(&hslot
->lock
);
637 flush_stack(stack
, count
, skb
, count
- 1);
639 for (i
= 0; i
< count
; i
++)
647 static inline int udp6_csum_init(struct sk_buff
*skb
, struct udphdr
*uh
,
652 UDP_SKB_CB(skb
)->partial_cov
= 0;
653 UDP_SKB_CB(skb
)->cscov
= skb
->len
;
655 if (proto
== IPPROTO_UDPLITE
) {
656 err
= udplite_checksum_init(skb
, uh
);
661 if (uh
->check
== 0) {
662 /* RFC 2460 section 8.1 says that we SHOULD log
663 this error. Well, it is reasonable.
665 LIMIT_NETDEBUG(KERN_INFO
"IPv6: udp checksum is 0\n");
668 if (skb
->ip_summed
== CHECKSUM_COMPLETE
&&
669 !csum_ipv6_magic(&ipv6_hdr(skb
)->saddr
, &ipv6_hdr(skb
)->daddr
,
670 skb
->len
, proto
, skb
->csum
))
671 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
673 if (!skb_csum_unnecessary(skb
))
674 skb
->csum
= ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb
)->saddr
,
675 &ipv6_hdr(skb
)->daddr
,
676 skb
->len
, proto
, 0));
681 int __udp6_lib_rcv(struct sk_buff
*skb
, struct udp_table
*udptable
,
686 struct net_device
*dev
= skb
->dev
;
687 struct in6_addr
*saddr
, *daddr
;
689 struct net
*net
= dev_net(skb
->dev
);
691 if (!pskb_may_pull(skb
, sizeof(struct udphdr
)))
694 saddr
= &ipv6_hdr(skb
)->saddr
;
695 daddr
= &ipv6_hdr(skb
)->daddr
;
698 ulen
= ntohs(uh
->len
);
702 if (proto
== IPPROTO_UDP
) {
703 /* UDP validates ulen. */
705 /* Check for jumbo payload */
709 if (ulen
< sizeof(*uh
))
712 if (ulen
< skb
->len
) {
713 if (pskb_trim_rcsum(skb
, ulen
))
715 saddr
= &ipv6_hdr(skb
)->saddr
;
716 daddr
= &ipv6_hdr(skb
)->daddr
;
721 if (udp6_csum_init(skb
, uh
, proto
))
725 * Multicast receive code
727 if (ipv6_addr_is_multicast(daddr
))
728 return __udp6_lib_mcast_deliver(net
, skb
,
729 saddr
, daddr
, udptable
);
734 * check socket cache ... must talk to Alan about his plans
735 * for sock caches... i'll skip this for now.
737 sk
= __udp6_lib_lookup_skb(skb
, uh
->source
, uh
->dest
, udptable
);
740 if (!xfrm6_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
743 if (udp_lib_checksum_complete(skb
))
745 UDP6_INC_STATS_BH(net
, UDP_MIB_NOPORTS
,
746 proto
== IPPROTO_UDPLITE
);
748 icmpv6_send(skb
, ICMPV6_DEST_UNREACH
, ICMPV6_PORT_UNREACH
, 0, dev
);
757 if (!sock_owned_by_user(sk
))
758 udpv6_queue_rcv_skb(sk
, skb
);
760 sk_add_backlog(sk
, skb
);
766 LIMIT_NETDEBUG(KERN_DEBUG
"UDP%sv6: short packet: %d/%u\n",
767 proto
== IPPROTO_UDPLITE
? "-Lite" : "",
771 UDP6_INC_STATS_BH(net
, UDP_MIB_INERRORS
, proto
== IPPROTO_UDPLITE
);
776 static __inline__
int udpv6_rcv(struct sk_buff
*skb
)
778 return __udp6_lib_rcv(skb
, &udp_table
, IPPROTO_UDP
);
782 * Throw away all pending data and cancel the corking. Socket is locked.
784 static void udp_v6_flush_pending_frames(struct sock
*sk
)
786 struct udp_sock
*up
= udp_sk(sk
);
788 if (up
->pending
== AF_INET
)
789 udp_flush_pending_frames(sk
);
790 else if (up
->pending
) {
793 ip6_flush_pending_frames(sk
);
798 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
799 * @sk: socket we are sending on
800 * @skb: sk_buff containing the filled-in UDP header
801 * (checksum field must be zeroed out)
803 static void udp6_hwcsum_outgoing(struct sock
*sk
, struct sk_buff
*skb
,
804 const struct in6_addr
*saddr
,
805 const struct in6_addr
*daddr
, int len
)
808 struct udphdr
*uh
= udp_hdr(skb
);
811 if (skb_queue_len(&sk
->sk_write_queue
) == 1) {
812 /* Only one fragment on the socket. */
813 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
814 skb
->csum_offset
= offsetof(struct udphdr
, check
);
815 uh
->check
= ~csum_ipv6_magic(saddr
, daddr
, len
, IPPROTO_UDP
, 0);
818 * HW-checksum won't work as there are two or more
819 * fragments on the socket so that all csums of sk_buffs
822 offset
= skb_transport_offset(skb
);
823 skb
->csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
825 skb
->ip_summed
= CHECKSUM_NONE
;
827 skb_queue_walk(&sk
->sk_write_queue
, skb
) {
828 csum
= csum_add(csum
, skb
->csum
);
831 uh
->check
= csum_ipv6_magic(saddr
, daddr
, len
, IPPROTO_UDP
,
834 uh
->check
= CSUM_MANGLED_0
;
842 static int udp_v6_push_pending_frames(struct sock
*sk
)
846 struct udp_sock
*up
= udp_sk(sk
);
847 struct inet_sock
*inet
= inet_sk(sk
);
848 struct flowi
*fl
= &inet
->cork
.fl
;
850 int is_udplite
= IS_UDPLITE(sk
);
853 /* Grab the skbuff where UDP header space exists. */
854 if ((skb
= skb_peek(&sk
->sk_write_queue
)) == NULL
)
858 * Create a UDP header
861 uh
->source
= fl
->fl_ip_sport
;
862 uh
->dest
= fl
->fl_ip_dport
;
863 uh
->len
= htons(up
->len
);
867 csum
= udplite_csum_outgoing(sk
, skb
);
868 else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) { /* UDP hardware csum */
869 udp6_hwcsum_outgoing(sk
, skb
, &fl
->fl6_src
, &fl
->fl6_dst
,
873 csum
= udp_csum_outgoing(sk
, skb
);
875 /* add protocol-dependent pseudo-header */
876 uh
->check
= csum_ipv6_magic(&fl
->fl6_src
, &fl
->fl6_dst
,
877 up
->len
, fl
->proto
, csum
);
879 uh
->check
= CSUM_MANGLED_0
;
882 err
= ip6_push_pending_frames(sk
);
884 if (err
== -ENOBUFS
&& !inet6_sk(sk
)->recverr
) {
885 UDP6_INC_STATS_USER(sock_net(sk
),
886 UDP_MIB_SNDBUFERRORS
, is_udplite
);
890 UDP6_INC_STATS_USER(sock_net(sk
),
891 UDP_MIB_OUTDATAGRAMS
, is_udplite
);
898 int udpv6_sendmsg(struct kiocb
*iocb
, struct sock
*sk
,
899 struct msghdr
*msg
, size_t len
)
901 struct ipv6_txoptions opt_space
;
902 struct udp_sock
*up
= udp_sk(sk
);
903 struct inet_sock
*inet
= inet_sk(sk
);
904 struct ipv6_pinfo
*np
= inet6_sk(sk
);
905 struct sockaddr_in6
*sin6
= (struct sockaddr_in6
*) msg
->msg_name
;
906 struct in6_addr
*daddr
, *final_p
= NULL
, final
;
907 struct ipv6_txoptions
*opt
= NULL
;
908 struct ip6_flowlabel
*flowlabel
= NULL
;
910 struct dst_entry
*dst
;
911 int addr_len
= msg
->msg_namelen
;
915 int corkreq
= up
->corkflag
|| msg
->msg_flags
&MSG_MORE
;
918 int is_udplite
= IS_UDPLITE(sk
);
919 int (*getfrag
)(void *, char *, int, int, int, struct sk_buff
*);
921 /* destination address check */
923 if (addr_len
< offsetof(struct sockaddr
, sa_data
))
926 switch (sin6
->sin6_family
) {
928 if (addr_len
< SIN6_LEN_RFC2133
)
930 daddr
= &sin6
->sin6_addr
;
935 msg
->msg_name
= sin6
= NULL
;
936 msg
->msg_namelen
= addr_len
= 0;
942 } else if (!up
->pending
) {
943 if (sk
->sk_state
!= TCP_ESTABLISHED
)
944 return -EDESTADDRREQ
;
950 if (ipv6_addr_v4mapped(daddr
)) {
951 struct sockaddr_in sin
;
952 sin
.sin_family
= AF_INET
;
953 sin
.sin_port
= sin6
? sin6
->sin6_port
: inet
->inet_dport
;
954 sin
.sin_addr
.s_addr
= daddr
->s6_addr32
[3];
955 msg
->msg_name
= &sin
;
956 msg
->msg_namelen
= sizeof(sin
);
958 if (__ipv6_only_sock(sk
))
960 return udp_sendmsg(iocb
, sk
, msg
, len
);
964 if (up
->pending
== AF_INET
)
965 return udp_sendmsg(iocb
, sk
, msg
, len
);
967 /* Rough check on arithmetic overflow,
968 better check is made in ip6_append_data().
970 if (len
> INT_MAX
- sizeof(struct udphdr
))
975 * There are pending frames.
976 * The socket lock must be held while it's corked.
979 if (likely(up
->pending
)) {
980 if (unlikely(up
->pending
!= AF_INET6
)) {
982 return -EAFNOSUPPORT
;
989 ulen
+= sizeof(struct udphdr
);
991 memset(&fl
, 0, sizeof(fl
));
994 if (sin6
->sin6_port
== 0)
997 fl
.fl_ip_dport
= sin6
->sin6_port
;
998 daddr
= &sin6
->sin6_addr
;
1001 fl
.fl6_flowlabel
= sin6
->sin6_flowinfo
&IPV6_FLOWINFO_MASK
;
1002 if (fl
.fl6_flowlabel
&IPV6_FLOWLABEL_MASK
) {
1003 flowlabel
= fl6_sock_lookup(sk
, fl
.fl6_flowlabel
);
1004 if (flowlabel
== NULL
)
1006 daddr
= &flowlabel
->dst
;
1011 * Otherwise it will be difficult to maintain
1014 if (sk
->sk_state
== TCP_ESTABLISHED
&&
1015 ipv6_addr_equal(daddr
, &np
->daddr
))
1018 if (addr_len
>= sizeof(struct sockaddr_in6
) &&
1019 sin6
->sin6_scope_id
&&
1020 ipv6_addr_type(daddr
)&IPV6_ADDR_LINKLOCAL
)
1021 fl
.oif
= sin6
->sin6_scope_id
;
1023 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1024 return -EDESTADDRREQ
;
1026 fl
.fl_ip_dport
= inet
->inet_dport
;
1028 fl
.fl6_flowlabel
= np
->flow_label
;
1033 fl
.oif
= sk
->sk_bound_dev_if
;
1036 fl
.oif
= np
->sticky_pktinfo
.ipi6_ifindex
;
1038 fl
.mark
= sk
->sk_mark
;
1040 if (msg
->msg_controllen
) {
1042 memset(opt
, 0, sizeof(struct ipv6_txoptions
));
1043 opt
->tot_len
= sizeof(*opt
);
1045 err
= datagram_send_ctl(sock_net(sk
), msg
, &fl
, opt
, &hlimit
, &tclass
);
1047 fl6_sock_release(flowlabel
);
1050 if ((fl
.fl6_flowlabel
&IPV6_FLOWLABEL_MASK
) && !flowlabel
) {
1051 flowlabel
= fl6_sock_lookup(sk
, fl
.fl6_flowlabel
);
1052 if (flowlabel
== NULL
)
1055 if (!(opt
->opt_nflen
|opt
->opt_flen
))
1062 opt
= fl6_merge_options(&opt_space
, flowlabel
, opt
);
1063 opt
= ipv6_fixup_options(&opt_space
, opt
);
1065 fl
.proto
= sk
->sk_protocol
;
1066 if (!ipv6_addr_any(daddr
))
1067 ipv6_addr_copy(&fl
.fl6_dst
, daddr
);
1069 fl
.fl6_dst
.s6_addr
[15] = 0x1; /* :: means loopback (BSD'ism) */
1070 if (ipv6_addr_any(&fl
.fl6_src
) && !ipv6_addr_any(&np
->saddr
))
1071 ipv6_addr_copy(&fl
.fl6_src
, &np
->saddr
);
1072 fl
.fl_ip_sport
= inet
->inet_sport
;
1074 /* merge ip6_build_xmit from ip6_output */
1075 if (opt
&& opt
->srcrt
) {
1076 struct rt0_hdr
*rt0
= (struct rt0_hdr
*) opt
->srcrt
;
1077 ipv6_addr_copy(&final
, &fl
.fl6_dst
);
1078 ipv6_addr_copy(&fl
.fl6_dst
, rt0
->addr
);
1083 if (!fl
.oif
&& ipv6_addr_is_multicast(&fl
.fl6_dst
)) {
1084 fl
.oif
= np
->mcast_oif
;
1088 security_sk_classify_flow(sk
, &fl
);
1090 err
= ip6_sk_dst_lookup(sk
, &dst
, &fl
);
1094 ipv6_addr_copy(&fl
.fl6_dst
, final_p
);
1096 err
= __xfrm_lookup(sock_net(sk
), &dst
, &fl
, sk
, XFRM_LOOKUP_WAIT
);
1098 if (err
== -EREMOTE
)
1099 err
= ip6_dst_blackhole(sk
, &dst
, &fl
);
1105 if (ipv6_addr_is_multicast(&fl
.fl6_dst
))
1106 hlimit
= np
->mcast_hops
;
1108 hlimit
= np
->hop_limit
;
1110 hlimit
= ip6_dst_hoplimit(dst
);
1114 tclass
= np
->tclass
;
1116 if (msg
->msg_flags
&MSG_CONFIRM
)
1121 if (unlikely(up
->pending
)) {
1122 /* The socket is already corked while preparing it. */
1123 /* ... which is an evident application bug. --ANK */
1126 LIMIT_NETDEBUG(KERN_DEBUG
"udp cork app bug 2\n");
1131 up
->pending
= AF_INET6
;
1135 getfrag
= is_udplite
? udplite_getfrag
: ip_generic_getfrag
;
1136 err
= ip6_append_data(sk
, getfrag
, msg
->msg_iov
, ulen
,
1137 sizeof(struct udphdr
), hlimit
, tclass
, opt
, &fl
,
1138 (struct rt6_info
*)dst
,
1139 corkreq
? msg
->msg_flags
|MSG_MORE
: msg
->msg_flags
);
1141 udp_v6_flush_pending_frames(sk
);
1143 err
= udp_v6_push_pending_frames(sk
);
1144 else if (unlikely(skb_queue_empty(&sk
->sk_write_queue
)))
1149 ip6_dst_store(sk
, dst
,
1150 ipv6_addr_equal(&fl
.fl6_dst
, &np
->daddr
) ?
1152 #ifdef CONFIG_IPV6_SUBTREES
1153 ipv6_addr_equal(&fl
.fl6_src
, &np
->saddr
) ?
1164 err
= np
->recverr
? net_xmit_errno(err
) : 0;
1168 fl6_sock_release(flowlabel
);
1172 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1173 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1174 * we don't have a good statistic (IpOutDiscards but it can be too many
1175 * things). We could add another new stat but at least for now that
1176 * seems like overkill.
1178 if (err
== -ENOBUFS
|| test_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
)) {
1179 UDP6_INC_STATS_USER(sock_net(sk
),
1180 UDP_MIB_SNDBUFERRORS
, is_udplite
);
1186 if (!(msg
->msg_flags
&MSG_PROBE
) || len
)
1187 goto back_from_confirm
;
1192 void udpv6_destroy_sock(struct sock
*sk
)
1195 udp_v6_flush_pending_frames(sk
);
1198 inet6_destroy_sock(sk
);
1202 * Socket option code for UDP
1204 int udpv6_setsockopt(struct sock
*sk
, int level
, int optname
,
1205 char __user
*optval
, unsigned int optlen
)
1207 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1208 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1209 udp_v6_push_pending_frames
);
1210 return ipv6_setsockopt(sk
, level
, optname
, optval
, optlen
);
1213 #ifdef CONFIG_COMPAT
1214 int compat_udpv6_setsockopt(struct sock
*sk
, int level
, int optname
,
1215 char __user
*optval
, unsigned int optlen
)
1217 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1218 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1219 udp_v6_push_pending_frames
);
1220 return compat_ipv6_setsockopt(sk
, level
, optname
, optval
, optlen
);
1224 int udpv6_getsockopt(struct sock
*sk
, int level
, int optname
,
1225 char __user
*optval
, int __user
*optlen
)
1227 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1228 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1229 return ipv6_getsockopt(sk
, level
, optname
, optval
, optlen
);
1232 #ifdef CONFIG_COMPAT
1233 int compat_udpv6_getsockopt(struct sock
*sk
, int level
, int optname
,
1234 char __user
*optval
, int __user
*optlen
)
1236 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1237 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1238 return compat_ipv6_getsockopt(sk
, level
, optname
, optval
, optlen
);
1242 static int udp6_ufo_send_check(struct sk_buff
*skb
)
1244 struct ipv6hdr
*ipv6h
;
1247 if (!pskb_may_pull(skb
, sizeof(*uh
)))
1250 ipv6h
= ipv6_hdr(skb
);
1253 uh
->check
= ~csum_ipv6_magic(&ipv6h
->saddr
, &ipv6h
->daddr
, skb
->len
,
1255 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
1256 skb
->csum_offset
= offsetof(struct udphdr
, check
);
1257 skb
->ip_summed
= CHECKSUM_PARTIAL
;
1261 static struct sk_buff
*udp6_ufo_fragment(struct sk_buff
*skb
, int features
)
1263 struct sk_buff
*segs
= ERR_PTR(-EINVAL
);
1265 unsigned int unfrag_ip6hlen
, unfrag_len
;
1266 struct frag_hdr
*fptr
;
1267 u8
*mac_start
, *prevhdr
;
1269 u8 frag_hdr_sz
= sizeof(struct frag_hdr
);
1273 mss
= skb_shinfo(skb
)->gso_size
;
1274 if (unlikely(skb
->len
<= mss
))
1277 if (skb_gso_ok(skb
, features
| NETIF_F_GSO_ROBUST
)) {
1278 /* Packet is from an untrusted source, reset gso_segs. */
1279 int type
= skb_shinfo(skb
)->gso_type
;
1281 if (unlikely(type
& ~(SKB_GSO_UDP
| SKB_GSO_DODGY
) ||
1282 !(type
& (SKB_GSO_UDP
))))
1285 skb_shinfo(skb
)->gso_segs
= DIV_ROUND_UP(skb
->len
, mss
);
1291 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
1292 * do checksum of UDP packets sent as multiple IP fragments.
1294 offset
= skb
->csum_start
- skb_headroom(skb
);
1295 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1296 offset
+= skb
->csum_offset
;
1297 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1298 skb
->ip_summed
= CHECKSUM_NONE
;
1300 /* Check if there is enough headroom to insert fragment header. */
1301 if ((skb_headroom(skb
) < frag_hdr_sz
) &&
1302 pskb_expand_head(skb
, frag_hdr_sz
, 0, GFP_ATOMIC
))
1305 /* Find the unfragmentable header and shift it left by frag_hdr_sz
1306 * bytes to insert fragment header.
1308 unfrag_ip6hlen
= ip6_find_1stfragopt(skb
, &prevhdr
);
1310 *prevhdr
= NEXTHDR_FRAGMENT
;
1311 unfrag_len
= skb_network_header(skb
) - skb_mac_header(skb
) +
1313 mac_start
= skb_mac_header(skb
);
1314 memmove(mac_start
-frag_hdr_sz
, mac_start
, unfrag_len
);
1316 skb
->mac_header
-= frag_hdr_sz
;
1317 skb
->network_header
-= frag_hdr_sz
;
1319 fptr
= (struct frag_hdr
*)(skb_network_header(skb
) + unfrag_ip6hlen
);
1320 fptr
->nexthdr
= nexthdr
;
1322 ipv6_select_ident(fptr
);
1324 /* Fragment the skb. ipv6 header and the remaining fields of the
1325 * fragment header are updated in ipv6_gso_segment()
1327 segs
= skb_segment(skb
, features
);
1333 static const struct inet6_protocol udpv6_protocol
= {
1334 .handler
= udpv6_rcv
,
1335 .err_handler
= udpv6_err
,
1336 .gso_send_check
= udp6_ufo_send_check
,
1337 .gso_segment
= udp6_ufo_fragment
,
1338 .flags
= INET6_PROTO_NOPOLICY
|INET6_PROTO_FINAL
,
1341 /* ------------------------------------------------------------------------ */
1342 #ifdef CONFIG_PROC_FS
1344 static void udp6_sock_seq_show(struct seq_file
*seq
, struct sock
*sp
, int bucket
)
1346 struct inet_sock
*inet
= inet_sk(sp
);
1347 struct ipv6_pinfo
*np
= inet6_sk(sp
);
1348 struct in6_addr
*dest
, *src
;
1352 src
= &np
->rcv_saddr
;
1353 destp
= ntohs(inet
->inet_dport
);
1354 srcp
= ntohs(inet
->inet_sport
);
1356 "%5d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1357 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d\n",
1359 src
->s6_addr32
[0], src
->s6_addr32
[1],
1360 src
->s6_addr32
[2], src
->s6_addr32
[3], srcp
,
1361 dest
->s6_addr32
[0], dest
->s6_addr32
[1],
1362 dest
->s6_addr32
[2], dest
->s6_addr32
[3], destp
,
1364 sk_wmem_alloc_get(sp
),
1365 sk_rmem_alloc_get(sp
),
1369 atomic_read(&sp
->sk_refcnt
), sp
,
1370 atomic_read(&sp
->sk_drops
));
1373 int udp6_seq_show(struct seq_file
*seq
, void *v
)
1375 if (v
== SEQ_START_TOKEN
)
1380 "st tx_queue rx_queue tr tm->when retrnsmt"
1381 " uid timeout inode ref pointer drops\n");
1383 udp6_sock_seq_show(seq
, v
, ((struct udp_iter_state
*)seq
->private)->bucket
);
1387 static struct udp_seq_afinfo udp6_seq_afinfo
= {
1390 .udp_table
= &udp_table
,
1392 .owner
= THIS_MODULE
,
1395 .show
= udp6_seq_show
,
1399 int udp6_proc_init(struct net
*net
)
1401 return udp_proc_register(net
, &udp6_seq_afinfo
);
1404 void udp6_proc_exit(struct net
*net
) {
1405 udp_proc_unregister(net
, &udp6_seq_afinfo
);
1407 #endif /* CONFIG_PROC_FS */
1409 /* ------------------------------------------------------------------------ */
1411 struct proto udpv6_prot
= {
1413 .owner
= THIS_MODULE
,
1414 .close
= udp_lib_close
,
1415 .connect
= ip6_datagram_connect
,
1416 .disconnect
= udp_disconnect
,
1418 .destroy
= udpv6_destroy_sock
,
1419 .setsockopt
= udpv6_setsockopt
,
1420 .getsockopt
= udpv6_getsockopt
,
1421 .sendmsg
= udpv6_sendmsg
,
1422 .recvmsg
= udpv6_recvmsg
,
1423 .backlog_rcv
= udpv6_queue_rcv_skb
,
1424 .hash
= udp_lib_hash
,
1425 .unhash
= udp_lib_unhash
,
1426 .get_port
= udp_v6_get_port
,
1427 .memory_allocated
= &udp_memory_allocated
,
1428 .sysctl_mem
= sysctl_udp_mem
,
1429 .sysctl_wmem
= &sysctl_udp_wmem_min
,
1430 .sysctl_rmem
= &sysctl_udp_rmem_min
,
1431 .obj_size
= sizeof(struct udp6_sock
),
1432 .slab_flags
= SLAB_DESTROY_BY_RCU
,
1433 .h
.udp_table
= &udp_table
,
1434 #ifdef CONFIG_COMPAT
1435 .compat_setsockopt
= compat_udpv6_setsockopt
,
1436 .compat_getsockopt
= compat_udpv6_getsockopt
,
1440 static struct inet_protosw udpv6_protosw
= {
1442 .protocol
= IPPROTO_UDP
,
1443 .prot
= &udpv6_prot
,
1444 .ops
= &inet6_dgram_ops
,
1445 .no_check
= UDP_CSUM_DEFAULT
,
1446 .flags
= INET_PROTOSW_PERMANENT
,
1450 int __init
udpv6_init(void)
1454 ret
= inet6_add_protocol(&udpv6_protocol
, IPPROTO_UDP
);
1458 ret
= inet6_register_protosw(&udpv6_protosw
);
1460 goto out_udpv6_protocol
;
1465 inet6_del_protocol(&udpv6_protocol
, IPPROTO_UDP
);
1469 void udpv6_exit(void)
1471 inet6_unregister_protosw(&udpv6_protosw
);
1472 inet6_del_protocol(&udpv6_protocol
, IPPROTO_UDP
);