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 <linux/slab.h>
38 #include <asm/uaccess.h>
40 #include <net/ndisc.h>
41 #include <net/protocol.h>
42 #include <net/transp_v6.h>
43 #include <net/ip6_route.h>
45 #include <net/tcp_states.h>
46 #include <net/ip6_checksum.h>
49 #include <linux/proc_fs.h>
50 #include <linux/seq_file.h>
53 int ipv6_rcv_saddr_equal(const struct sock
*sk
, const struct sock
*sk2
)
55 const struct in6_addr
*sk_rcv_saddr6
= &inet6_sk(sk
)->rcv_saddr
;
56 const struct in6_addr
*sk2_rcv_saddr6
= inet6_rcv_saddr(sk2
);
57 __be32 sk1_rcv_saddr
= inet_sk(sk
)->inet_rcv_saddr
;
58 __be32 sk2_rcv_saddr
= inet_rcv_saddr(sk2
);
59 int sk_ipv6only
= ipv6_only_sock(sk
);
60 int sk2_ipv6only
= inet_v6_ipv6only(sk2
);
61 int addr_type
= ipv6_addr_type(sk_rcv_saddr6
);
62 int addr_type2
= sk2_rcv_saddr6
? ipv6_addr_type(sk2_rcv_saddr6
) : IPV6_ADDR_MAPPED
;
64 /* if both are mapped, treat as IPv4 */
65 if (addr_type
== IPV6_ADDR_MAPPED
&& addr_type2
== IPV6_ADDR_MAPPED
)
66 return (!sk2_ipv6only
&&
67 (!sk1_rcv_saddr
|| !sk2_rcv_saddr
||
68 sk1_rcv_saddr
== sk2_rcv_saddr
));
70 if (addr_type2
== IPV6_ADDR_ANY
&&
71 !(sk2_ipv6only
&& addr_type
== IPV6_ADDR_MAPPED
))
74 if (addr_type
== IPV6_ADDR_ANY
&&
75 !(sk_ipv6only
&& addr_type2
== IPV6_ADDR_MAPPED
))
79 ipv6_addr_equal(sk_rcv_saddr6
, sk2_rcv_saddr6
))
85 static unsigned int udp6_portaddr_hash(struct net
*net
,
86 const struct in6_addr
*addr6
,
89 unsigned int hash
, mix
= net_hash_mix(net
);
91 if (ipv6_addr_any(addr6
))
92 hash
= jhash_1word(0, mix
);
93 else if (ipv6_addr_v4mapped(addr6
))
94 hash
= jhash_1word((__force u32
)addr6
->s6_addr32
[3], mix
);
96 hash
= jhash2((__force u32
*)addr6
->s6_addr32
, 4, mix
);
102 int udp_v6_get_port(struct sock
*sk
, unsigned short snum
)
104 unsigned int hash2_nulladdr
=
105 udp6_portaddr_hash(sock_net(sk
), &in6addr_any
, snum
);
106 unsigned int hash2_partial
=
107 udp6_portaddr_hash(sock_net(sk
), &inet6_sk(sk
)->rcv_saddr
, 0);
109 /* precompute partial secondary hash */
110 udp_sk(sk
)->udp_portaddr_hash
= hash2_partial
;
111 return udp_lib_get_port(sk
, snum
, ipv6_rcv_saddr_equal
, hash2_nulladdr
);
114 static inline int compute_score(struct sock
*sk
, struct net
*net
,
116 struct in6_addr
*saddr
, __be16 sport
,
117 struct in6_addr
*daddr
, __be16 dport
,
122 if (net_eq(sock_net(sk
), net
) && udp_sk(sk
)->udp_port_hash
== hnum
&&
123 sk
->sk_family
== PF_INET6
) {
124 struct ipv6_pinfo
*np
= inet6_sk(sk
);
125 struct inet_sock
*inet
= inet_sk(sk
);
128 if (inet
->inet_dport
) {
129 if (inet
->inet_dport
!= sport
)
133 if (!ipv6_addr_any(&np
->rcv_saddr
)) {
134 if (!ipv6_addr_equal(&np
->rcv_saddr
, daddr
))
138 if (!ipv6_addr_any(&np
->daddr
)) {
139 if (!ipv6_addr_equal(&np
->daddr
, saddr
))
143 if (sk
->sk_bound_dev_if
) {
144 if (sk
->sk_bound_dev_if
!= dif
)
152 #define SCORE2_MAX (1 + 1 + 1)
153 static inline int compute_score2(struct sock
*sk
, struct net
*net
,
154 const struct in6_addr
*saddr
, __be16 sport
,
155 const struct in6_addr
*daddr
, unsigned short hnum
,
160 if (net_eq(sock_net(sk
), net
) && udp_sk(sk
)->udp_port_hash
== hnum
&&
161 sk
->sk_family
== PF_INET6
) {
162 struct ipv6_pinfo
*np
= inet6_sk(sk
);
163 struct inet_sock
*inet
= inet_sk(sk
);
165 if (!ipv6_addr_equal(&np
->rcv_saddr
, daddr
))
168 if (inet
->inet_dport
) {
169 if (inet
->inet_dport
!= sport
)
173 if (!ipv6_addr_any(&np
->daddr
)) {
174 if (!ipv6_addr_equal(&np
->daddr
, saddr
))
178 if (sk
->sk_bound_dev_if
) {
179 if (sk
->sk_bound_dev_if
!= dif
)
188 /* called with read_rcu_lock() */
189 static struct sock
*udp6_lib_lookup2(struct net
*net
,
190 const struct in6_addr
*saddr
, __be16 sport
,
191 const struct in6_addr
*daddr
, unsigned int hnum
, int dif
,
192 struct udp_hslot
*hslot2
, unsigned int slot2
)
194 struct sock
*sk
, *result
;
195 struct hlist_nulls_node
*node
;
201 udp_portaddr_for_each_entry_rcu(sk
, node
, &hslot2
->head
) {
202 score
= compute_score2(sk
, net
, saddr
, sport
,
204 if (score
> badness
) {
207 if (score
== SCORE2_MAX
)
212 * if the nulls value we got at the end of this lookup is
213 * not the expected one, we must restart lookup.
214 * We probably met an item that was moved to another chain.
216 if (get_nulls_value(node
) != slot2
)
221 if (unlikely(!atomic_inc_not_zero(&result
->sk_refcnt
)))
223 else if (unlikely(compute_score2(result
, net
, saddr
, sport
,
224 daddr
, hnum
, dif
) < badness
)) {
232 static struct sock
*__udp6_lib_lookup(struct net
*net
,
233 struct in6_addr
*saddr
, __be16 sport
,
234 struct in6_addr
*daddr
, __be16 dport
,
235 int dif
, struct udp_table
*udptable
)
237 struct sock
*sk
, *result
;
238 struct hlist_nulls_node
*node
;
239 unsigned short hnum
= ntohs(dport
);
240 unsigned int hash2
, slot2
, slot
= udp_hashfn(net
, hnum
, udptable
->mask
);
241 struct udp_hslot
*hslot2
, *hslot
= &udptable
->hash
[slot
];
245 if (hslot
->count
> 10) {
246 hash2
= udp6_portaddr_hash(net
, daddr
, hnum
);
247 slot2
= hash2
& udptable
->mask
;
248 hslot2
= &udptable
->hash2
[slot2
];
249 if (hslot
->count
< hslot2
->count
)
252 result
= udp6_lib_lookup2(net
, saddr
, sport
,
256 hash2
= udp6_portaddr_hash(net
, &in6addr_any
, hnum
);
257 slot2
= hash2
& udptable
->mask
;
258 hslot2
= &udptable
->hash2
[slot2
];
259 if (hslot
->count
< hslot2
->count
)
262 result
= udp6_lib_lookup2(net
, saddr
, sport
,
263 &in6addr_any
, hnum
, dif
,
272 sk_nulls_for_each_rcu(sk
, node
, &hslot
->head
) {
273 score
= compute_score(sk
, net
, hnum
, saddr
, sport
, daddr
, dport
, dif
);
274 if (score
> badness
) {
280 * if the nulls value we got at the end of this lookup is
281 * not the expected one, we must restart lookup.
282 * We probably met an item that was moved to another chain.
284 if (get_nulls_value(node
) != slot
)
288 if (unlikely(!atomic_inc_not_zero(&result
->sk_refcnt
)))
290 else if (unlikely(compute_score(result
, net
, hnum
, saddr
, sport
,
291 daddr
, dport
, dif
) < badness
)) {
300 static struct sock
*__udp6_lib_lookup_skb(struct sk_buff
*skb
,
301 __be16 sport
, __be16 dport
,
302 struct udp_table
*udptable
)
305 struct ipv6hdr
*iph
= ipv6_hdr(skb
);
307 if (unlikely(sk
= skb_steal_sock(skb
)))
309 return __udp6_lib_lookup(dev_net(skb_dst(skb
)->dev
), &iph
->saddr
, sport
,
310 &iph
->daddr
, dport
, inet6_iif(skb
),
315 * This should be easy, if there is something there we
316 * return it, otherwise we block.
319 int udpv6_recvmsg(struct kiocb
*iocb
, struct sock
*sk
,
320 struct msghdr
*msg
, size_t len
,
321 int noblock
, int flags
, int *addr_len
)
323 struct ipv6_pinfo
*np
= inet6_sk(sk
);
324 struct inet_sock
*inet
= inet_sk(sk
);
329 int is_udplite
= IS_UDPLITE(sk
);
333 *addr_len
=sizeof(struct sockaddr_in6
);
335 if (flags
& MSG_ERRQUEUE
)
336 return ipv6_recv_error(sk
, msg
, len
);
338 if (np
->rxpmtu
&& np
->rxopt
.bits
.rxpmtu
)
339 return ipv6_recv_rxpmtu(sk
, msg
, len
);
342 skb
= __skb_recv_datagram(sk
, flags
| (noblock
? MSG_DONTWAIT
: 0),
347 ulen
= skb
->len
- sizeof(struct udphdr
);
351 msg
->msg_flags
|= MSG_TRUNC
;
353 is_udp4
= (skb
->protocol
== htons(ETH_P_IP
));
356 * If checksum is needed at all, try to do it while copying the
357 * data. If the data is truncated, or if we only want a partial
358 * coverage checksum (UDP-Lite), do it before the copy.
361 if (len
< ulen
|| UDP_SKB_CB(skb
)->partial_cov
) {
362 if (udp_lib_checksum_complete(skb
))
366 if (skb_csum_unnecessary(skb
))
367 err
= skb_copy_datagram_iovec(skb
, sizeof(struct udphdr
),
370 err
= skb_copy_and_csum_datagram_iovec(skb
, sizeof(struct udphdr
), msg
->msg_iov
);
379 UDP_INC_STATS_USER(sock_net(sk
),
380 UDP_MIB_INDATAGRAMS
, is_udplite
);
382 UDP6_INC_STATS_USER(sock_net(sk
),
383 UDP_MIB_INDATAGRAMS
, is_udplite
);
386 sock_recv_ts_and_drops(msg
, sk
, skb
);
388 /* Copy the address. */
390 struct sockaddr_in6
*sin6
;
392 sin6
= (struct sockaddr_in6
*) msg
->msg_name
;
393 sin6
->sin6_family
= AF_INET6
;
394 sin6
->sin6_port
= udp_hdr(skb
)->source
;
395 sin6
->sin6_flowinfo
= 0;
396 sin6
->sin6_scope_id
= 0;
399 ipv6_addr_set_v4mapped(ip_hdr(skb
)->saddr
,
402 ipv6_addr_copy(&sin6
->sin6_addr
,
403 &ipv6_hdr(skb
)->saddr
);
404 if (ipv6_addr_type(&sin6
->sin6_addr
) & IPV6_ADDR_LINKLOCAL
)
405 sin6
->sin6_scope_id
= IP6CB(skb
)->iif
;
410 if (inet
->cmsg_flags
)
411 ip_cmsg_recv(msg
, skb
);
414 datagram_recv_ctl(sk
, msg
, skb
);
418 if (flags
& MSG_TRUNC
)
422 skb_free_datagram_locked(sk
, skb
);
428 if (!skb_kill_datagram(sk
, skb
, flags
)) {
430 UDP_INC_STATS_USER(sock_net(sk
),
431 UDP_MIB_INERRORS
, is_udplite
);
433 UDP6_INC_STATS_USER(sock_net(sk
),
434 UDP_MIB_INERRORS
, is_udplite
);
438 if (flags
& MSG_DONTWAIT
)
443 void __udp6_lib_err(struct sk_buff
*skb
, struct inet6_skb_parm
*opt
,
444 u8 type
, u8 code
, int offset
, __be32 info
,
445 struct udp_table
*udptable
)
447 struct ipv6_pinfo
*np
;
448 struct ipv6hdr
*hdr
= (struct ipv6hdr
*)skb
->data
;
449 struct in6_addr
*saddr
= &hdr
->saddr
;
450 struct in6_addr
*daddr
= &hdr
->daddr
;
451 struct udphdr
*uh
= (struct udphdr
*)(skb
->data
+offset
);
455 sk
= __udp6_lib_lookup(dev_net(skb
->dev
), daddr
, uh
->dest
,
456 saddr
, uh
->source
, inet6_iif(skb
), udptable
);
462 if (!icmpv6_err_convert(type
, code
, &err
) && !np
->recverr
)
465 if (sk
->sk_state
!= TCP_ESTABLISHED
&& !np
->recverr
)
469 ipv6_icmp_error(sk
, skb
, err
, uh
->dest
, ntohl(info
), (u8
*)(uh
+1));
472 sk
->sk_error_report(sk
);
477 static __inline__
void udpv6_err(struct sk_buff
*skb
,
478 struct inet6_skb_parm
*opt
, u8 type
,
479 u8 code
, int offset
, __be32 info
)
481 __udp6_lib_err(skb
, opt
, type
, code
, offset
, info
, &udp_table
);
484 int udpv6_queue_rcv_skb(struct sock
* sk
, struct sk_buff
*skb
)
486 struct udp_sock
*up
= udp_sk(sk
);
488 int is_udplite
= IS_UDPLITE(sk
);
490 if (!xfrm6_policy_check(sk
, XFRM_POLICY_IN
, skb
))
494 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
496 if ((is_udplite
& UDPLITE_RECV_CC
) && UDP_SKB_CB(skb
)->partial_cov
) {
498 if (up
->pcrlen
== 0) { /* full coverage was set */
499 LIMIT_NETDEBUG(KERN_WARNING
"UDPLITE6: partial coverage"
500 " %d while full coverage %d requested\n",
501 UDP_SKB_CB(skb
)->cscov
, skb
->len
);
504 if (UDP_SKB_CB(skb
)->cscov
< up
->pcrlen
) {
505 LIMIT_NETDEBUG(KERN_WARNING
"UDPLITE6: coverage %d "
506 "too small, need min %d\n",
507 UDP_SKB_CB(skb
)->cscov
, up
->pcrlen
);
513 if (udp_lib_checksum_complete(skb
))
517 if ((rc
= ip_queue_rcv_skb(sk
, skb
)) < 0) {
518 /* Note that an ENOMEM error is charged twice */
520 UDP6_INC_STATS_BH(sock_net(sk
),
521 UDP_MIB_RCVBUFERRORS
, is_udplite
);
522 goto drop_no_sk_drops_inc
;
527 atomic_inc(&sk
->sk_drops
);
528 drop_no_sk_drops_inc
:
529 UDP6_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
534 static struct sock
*udp_v6_mcast_next(struct net
*net
, struct sock
*sk
,
535 __be16 loc_port
, struct in6_addr
*loc_addr
,
536 __be16 rmt_port
, struct in6_addr
*rmt_addr
,
539 struct hlist_nulls_node
*node
;
541 unsigned short num
= ntohs(loc_port
);
543 sk_nulls_for_each_from(s
, node
) {
544 struct inet_sock
*inet
= inet_sk(s
);
546 if (!net_eq(sock_net(s
), net
))
549 if (udp_sk(s
)->udp_port_hash
== num
&&
550 s
->sk_family
== PF_INET6
) {
551 struct ipv6_pinfo
*np
= inet6_sk(s
);
552 if (inet
->inet_dport
) {
553 if (inet
->inet_dport
!= rmt_port
)
556 if (!ipv6_addr_any(&np
->daddr
) &&
557 !ipv6_addr_equal(&np
->daddr
, rmt_addr
))
560 if (s
->sk_bound_dev_if
&& s
->sk_bound_dev_if
!= dif
)
563 if (!ipv6_addr_any(&np
->rcv_saddr
)) {
564 if (!ipv6_addr_equal(&np
->rcv_saddr
, loc_addr
))
567 if (!inet6_mc_check(s
, loc_addr
, rmt_addr
))
575 static void flush_stack(struct sock
**stack
, unsigned int count
,
576 struct sk_buff
*skb
, unsigned int final
)
580 struct sk_buff
*skb1
;
582 for (i
= 0; i
< count
; i
++) {
583 skb1
= (i
== final
) ? skb
: skb_clone(skb
, GFP_ATOMIC
);
587 if (sk_rcvqueues_full(sk
, skb
)) {
592 if (!sock_owned_by_user(sk
))
593 udpv6_queue_rcv_skb(sk
, skb1
);
594 else if (sk_add_backlog(sk
, skb1
)) {
603 atomic_inc(&sk
->sk_drops
);
604 UDP6_INC_STATS_BH(sock_net(sk
),
605 UDP_MIB_RCVBUFERRORS
, IS_UDPLITE(sk
));
606 UDP6_INC_STATS_BH(sock_net(sk
),
607 UDP_MIB_INERRORS
, IS_UDPLITE(sk
));
611 * Note: called only from the BH handler context,
612 * so we don't need to lock the hashes.
614 static int __udp6_lib_mcast_deliver(struct net
*net
, struct sk_buff
*skb
,
615 struct in6_addr
*saddr
, struct in6_addr
*daddr
,
616 struct udp_table
*udptable
)
618 struct sock
*sk
, *stack
[256 / sizeof(struct sock
*)];
619 const struct udphdr
*uh
= udp_hdr(skb
);
620 struct udp_hslot
*hslot
= udp_hashslot(udptable
, net
, ntohs(uh
->dest
));
622 unsigned int i
, count
= 0;
624 spin_lock(&hslot
->lock
);
625 sk
= sk_nulls_head(&hslot
->head
);
626 dif
= inet6_iif(skb
);
627 sk
= udp_v6_mcast_next(net
, sk
, uh
->dest
, daddr
, uh
->source
, saddr
, dif
);
630 sk
= udp_v6_mcast_next(net
, sk_nulls_next(sk
), uh
->dest
, daddr
,
631 uh
->source
, saddr
, dif
);
632 if (unlikely(count
== ARRAY_SIZE(stack
))) {
635 flush_stack(stack
, count
, skb
, ~0);
640 * before releasing the lock, we must take reference on sockets
642 for (i
= 0; i
< count
; i
++)
645 spin_unlock(&hslot
->lock
);
648 flush_stack(stack
, count
, skb
, count
- 1);
650 for (i
= 0; i
< count
; i
++)
658 static inline int udp6_csum_init(struct sk_buff
*skb
, struct udphdr
*uh
,
663 UDP_SKB_CB(skb
)->partial_cov
= 0;
664 UDP_SKB_CB(skb
)->cscov
= skb
->len
;
666 if (proto
== IPPROTO_UDPLITE
) {
667 err
= udplite_checksum_init(skb
, uh
);
672 if (uh
->check
== 0) {
673 /* RFC 2460 section 8.1 says that we SHOULD log
674 this error. Well, it is reasonable.
676 LIMIT_NETDEBUG(KERN_INFO
"IPv6: udp checksum is 0\n");
679 if (skb
->ip_summed
== CHECKSUM_COMPLETE
&&
680 !csum_ipv6_magic(&ipv6_hdr(skb
)->saddr
, &ipv6_hdr(skb
)->daddr
,
681 skb
->len
, proto
, skb
->csum
))
682 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
684 if (!skb_csum_unnecessary(skb
))
685 skb
->csum
= ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb
)->saddr
,
686 &ipv6_hdr(skb
)->daddr
,
687 skb
->len
, proto
, 0));
692 int __udp6_lib_rcv(struct sk_buff
*skb
, struct udp_table
*udptable
,
695 struct net
*net
= dev_net(skb
->dev
);
698 struct in6_addr
*saddr
, *daddr
;
701 if (!pskb_may_pull(skb
, sizeof(struct udphdr
)))
704 saddr
= &ipv6_hdr(skb
)->saddr
;
705 daddr
= &ipv6_hdr(skb
)->daddr
;
708 ulen
= ntohs(uh
->len
);
712 if (proto
== IPPROTO_UDP
) {
713 /* UDP validates ulen. */
715 /* Check for jumbo payload */
719 if (ulen
< sizeof(*uh
))
722 if (ulen
< skb
->len
) {
723 if (pskb_trim_rcsum(skb
, ulen
))
725 saddr
= &ipv6_hdr(skb
)->saddr
;
726 daddr
= &ipv6_hdr(skb
)->daddr
;
731 if (udp6_csum_init(skb
, uh
, proto
))
735 * Multicast receive code
737 if (ipv6_addr_is_multicast(daddr
))
738 return __udp6_lib_mcast_deliver(net
, skb
,
739 saddr
, daddr
, udptable
);
744 * check socket cache ... must talk to Alan about his plans
745 * for sock caches... i'll skip this for now.
747 sk
= __udp6_lib_lookup_skb(skb
, uh
->source
, uh
->dest
, udptable
);
750 if (!xfrm6_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
753 if (udp_lib_checksum_complete(skb
))
755 UDP6_INC_STATS_BH(net
, UDP_MIB_NOPORTS
,
756 proto
== IPPROTO_UDPLITE
);
758 icmpv6_send(skb
, ICMPV6_DEST_UNREACH
, ICMPV6_PORT_UNREACH
, 0);
766 if (sk_rcvqueues_full(sk
, skb
)) {
771 if (!sock_owned_by_user(sk
))
772 udpv6_queue_rcv_skb(sk
, skb
);
773 else if (sk_add_backlog(sk
, skb
)) {
774 atomic_inc(&sk
->sk_drops
);
784 LIMIT_NETDEBUG(KERN_DEBUG
"UDP%sv6: short packet: %d/%u\n",
785 proto
== IPPROTO_UDPLITE
? "-Lite" : "",
789 UDP6_INC_STATS_BH(net
, UDP_MIB_INERRORS
, proto
== IPPROTO_UDPLITE
);
794 static __inline__
int udpv6_rcv(struct sk_buff
*skb
)
796 return __udp6_lib_rcv(skb
, &udp_table
, IPPROTO_UDP
);
800 * Throw away all pending data and cancel the corking. Socket is locked.
802 static void udp_v6_flush_pending_frames(struct sock
*sk
)
804 struct udp_sock
*up
= udp_sk(sk
);
806 if (up
->pending
== AF_INET
)
807 udp_flush_pending_frames(sk
);
808 else if (up
->pending
) {
811 ip6_flush_pending_frames(sk
);
816 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
817 * @sk: socket we are sending on
818 * @skb: sk_buff containing the filled-in UDP header
819 * (checksum field must be zeroed out)
821 static void udp6_hwcsum_outgoing(struct sock
*sk
, struct sk_buff
*skb
,
822 const struct in6_addr
*saddr
,
823 const struct in6_addr
*daddr
, int len
)
826 struct udphdr
*uh
= udp_hdr(skb
);
829 if (skb_queue_len(&sk
->sk_write_queue
) == 1) {
830 /* Only one fragment on the socket. */
831 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
832 skb
->csum_offset
= offsetof(struct udphdr
, check
);
833 uh
->check
= ~csum_ipv6_magic(saddr
, daddr
, len
, IPPROTO_UDP
, 0);
836 * HW-checksum won't work as there are two or more
837 * fragments on the socket so that all csums of sk_buffs
840 offset
= skb_transport_offset(skb
);
841 skb
->csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
843 skb
->ip_summed
= CHECKSUM_NONE
;
845 skb_queue_walk(&sk
->sk_write_queue
, skb
) {
846 csum
= csum_add(csum
, skb
->csum
);
849 uh
->check
= csum_ipv6_magic(saddr
, daddr
, len
, IPPROTO_UDP
,
852 uh
->check
= CSUM_MANGLED_0
;
860 static int udp_v6_push_pending_frames(struct sock
*sk
)
864 struct udp_sock
*up
= udp_sk(sk
);
865 struct inet_sock
*inet
= inet_sk(sk
);
866 struct flowi
*fl
= &inet
->cork
.fl
;
868 int is_udplite
= IS_UDPLITE(sk
);
871 /* Grab the skbuff where UDP header space exists. */
872 if ((skb
= skb_peek(&sk
->sk_write_queue
)) == NULL
)
876 * Create a UDP header
879 uh
->source
= fl
->fl_ip_sport
;
880 uh
->dest
= fl
->fl_ip_dport
;
881 uh
->len
= htons(up
->len
);
885 csum
= udplite_csum_outgoing(sk
, skb
);
886 else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) { /* UDP hardware csum */
887 udp6_hwcsum_outgoing(sk
, skb
, &fl
->fl6_src
, &fl
->fl6_dst
,
891 csum
= udp_csum_outgoing(sk
, skb
);
893 /* add protocol-dependent pseudo-header */
894 uh
->check
= csum_ipv6_magic(&fl
->fl6_src
, &fl
->fl6_dst
,
895 up
->len
, fl
->proto
, csum
);
897 uh
->check
= CSUM_MANGLED_0
;
900 err
= ip6_push_pending_frames(sk
);
902 if (err
== -ENOBUFS
&& !inet6_sk(sk
)->recverr
) {
903 UDP6_INC_STATS_USER(sock_net(sk
),
904 UDP_MIB_SNDBUFERRORS
, is_udplite
);
908 UDP6_INC_STATS_USER(sock_net(sk
),
909 UDP_MIB_OUTDATAGRAMS
, is_udplite
);
916 int udpv6_sendmsg(struct kiocb
*iocb
, struct sock
*sk
,
917 struct msghdr
*msg
, size_t len
)
919 struct ipv6_txoptions opt_space
;
920 struct udp_sock
*up
= udp_sk(sk
);
921 struct inet_sock
*inet
= inet_sk(sk
);
922 struct ipv6_pinfo
*np
= inet6_sk(sk
);
923 struct sockaddr_in6
*sin6
= (struct sockaddr_in6
*) msg
->msg_name
;
924 struct in6_addr
*daddr
, *final_p
= NULL
, final
;
925 struct ipv6_txoptions
*opt
= NULL
;
926 struct ip6_flowlabel
*flowlabel
= NULL
;
928 struct dst_entry
*dst
;
929 int addr_len
= msg
->msg_namelen
;
934 int corkreq
= up
->corkflag
|| msg
->msg_flags
&MSG_MORE
;
937 int is_udplite
= IS_UDPLITE(sk
);
938 int (*getfrag
)(void *, char *, int, int, int, struct sk_buff
*);
940 /* destination address check */
942 if (addr_len
< offsetof(struct sockaddr
, sa_data
))
945 switch (sin6
->sin6_family
) {
947 if (addr_len
< SIN6_LEN_RFC2133
)
949 daddr
= &sin6
->sin6_addr
;
954 msg
->msg_name
= sin6
= NULL
;
955 msg
->msg_namelen
= addr_len
= 0;
961 } else if (!up
->pending
) {
962 if (sk
->sk_state
!= TCP_ESTABLISHED
)
963 return -EDESTADDRREQ
;
969 if (ipv6_addr_v4mapped(daddr
)) {
970 struct sockaddr_in sin
;
971 sin
.sin_family
= AF_INET
;
972 sin
.sin_port
= sin6
? sin6
->sin6_port
: inet
->inet_dport
;
973 sin
.sin_addr
.s_addr
= daddr
->s6_addr32
[3];
974 msg
->msg_name
= &sin
;
975 msg
->msg_namelen
= sizeof(sin
);
977 if (__ipv6_only_sock(sk
))
979 return udp_sendmsg(iocb
, sk
, msg
, len
);
983 if (up
->pending
== AF_INET
)
984 return udp_sendmsg(iocb
, sk
, msg
, len
);
986 /* Rough check on arithmetic overflow,
987 better check is made in ip6_append_data().
989 if (len
> INT_MAX
- sizeof(struct udphdr
))
994 * There are pending frames.
995 * The socket lock must be held while it's corked.
998 if (likely(up
->pending
)) {
999 if (unlikely(up
->pending
!= AF_INET6
)) {
1001 return -EAFNOSUPPORT
;
1004 goto do_append_data
;
1008 ulen
+= sizeof(struct udphdr
);
1010 memset(&fl
, 0, sizeof(fl
));
1013 if (sin6
->sin6_port
== 0)
1016 fl
.fl_ip_dport
= sin6
->sin6_port
;
1017 daddr
= &sin6
->sin6_addr
;
1020 fl
.fl6_flowlabel
= sin6
->sin6_flowinfo
&IPV6_FLOWINFO_MASK
;
1021 if (fl
.fl6_flowlabel
&IPV6_FLOWLABEL_MASK
) {
1022 flowlabel
= fl6_sock_lookup(sk
, fl
.fl6_flowlabel
);
1023 if (flowlabel
== NULL
)
1025 daddr
= &flowlabel
->dst
;
1030 * Otherwise it will be difficult to maintain
1033 if (sk
->sk_state
== TCP_ESTABLISHED
&&
1034 ipv6_addr_equal(daddr
, &np
->daddr
))
1037 if (addr_len
>= sizeof(struct sockaddr_in6
) &&
1038 sin6
->sin6_scope_id
&&
1039 ipv6_addr_type(daddr
)&IPV6_ADDR_LINKLOCAL
)
1040 fl
.oif
= sin6
->sin6_scope_id
;
1042 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1043 return -EDESTADDRREQ
;
1045 fl
.fl_ip_dport
= inet
->inet_dport
;
1047 fl
.fl6_flowlabel
= np
->flow_label
;
1052 fl
.oif
= sk
->sk_bound_dev_if
;
1055 fl
.oif
= np
->sticky_pktinfo
.ipi6_ifindex
;
1057 fl
.mark
= sk
->sk_mark
;
1059 if (msg
->msg_controllen
) {
1061 memset(opt
, 0, sizeof(struct ipv6_txoptions
));
1062 opt
->tot_len
= sizeof(*opt
);
1064 err
= datagram_send_ctl(sock_net(sk
), msg
, &fl
, opt
, &hlimit
,
1065 &tclass
, &dontfrag
);
1067 fl6_sock_release(flowlabel
);
1070 if ((fl
.fl6_flowlabel
&IPV6_FLOWLABEL_MASK
) && !flowlabel
) {
1071 flowlabel
= fl6_sock_lookup(sk
, fl
.fl6_flowlabel
);
1072 if (flowlabel
== NULL
)
1075 if (!(opt
->opt_nflen
|opt
->opt_flen
))
1082 opt
= fl6_merge_options(&opt_space
, flowlabel
, opt
);
1083 opt
= ipv6_fixup_options(&opt_space
, opt
);
1085 fl
.proto
= sk
->sk_protocol
;
1086 if (!ipv6_addr_any(daddr
))
1087 ipv6_addr_copy(&fl
.fl6_dst
, daddr
);
1089 fl
.fl6_dst
.s6_addr
[15] = 0x1; /* :: means loopback (BSD'ism) */
1090 if (ipv6_addr_any(&fl
.fl6_src
) && !ipv6_addr_any(&np
->saddr
))
1091 ipv6_addr_copy(&fl
.fl6_src
, &np
->saddr
);
1092 fl
.fl_ip_sport
= inet
->inet_sport
;
1094 /* merge ip6_build_xmit from ip6_output */
1095 if (opt
&& opt
->srcrt
) {
1096 struct rt0_hdr
*rt0
= (struct rt0_hdr
*) opt
->srcrt
;
1097 ipv6_addr_copy(&final
, &fl
.fl6_dst
);
1098 ipv6_addr_copy(&fl
.fl6_dst
, rt0
->addr
);
1103 if (!fl
.oif
&& ipv6_addr_is_multicast(&fl
.fl6_dst
)) {
1104 fl
.oif
= np
->mcast_oif
;
1108 security_sk_classify_flow(sk
, &fl
);
1110 err
= ip6_sk_dst_lookup(sk
, &dst
, &fl
);
1114 ipv6_addr_copy(&fl
.fl6_dst
, final_p
);
1116 err
= __xfrm_lookup(sock_net(sk
), &dst
, &fl
, sk
, XFRM_LOOKUP_WAIT
);
1118 if (err
== -EREMOTE
)
1119 err
= ip6_dst_blackhole(sk
, &dst
, &fl
);
1125 if (ipv6_addr_is_multicast(&fl
.fl6_dst
))
1126 hlimit
= np
->mcast_hops
;
1128 hlimit
= np
->hop_limit
;
1130 hlimit
= ip6_dst_hoplimit(dst
);
1134 tclass
= np
->tclass
;
1137 dontfrag
= np
->dontfrag
;
1139 if (msg
->msg_flags
&MSG_CONFIRM
)
1144 if (unlikely(up
->pending
)) {
1145 /* The socket is already corked while preparing it. */
1146 /* ... which is an evident application bug. --ANK */
1149 LIMIT_NETDEBUG(KERN_DEBUG
"udp cork app bug 2\n");
1154 up
->pending
= AF_INET6
;
1158 getfrag
= is_udplite
? udplite_getfrag
: ip_generic_getfrag
;
1159 err
= ip6_append_data(sk
, getfrag
, msg
->msg_iov
, ulen
,
1160 sizeof(struct udphdr
), hlimit
, tclass
, opt
, &fl
,
1161 (struct rt6_info
*)dst
,
1162 corkreq
? msg
->msg_flags
|MSG_MORE
: msg
->msg_flags
, dontfrag
);
1164 udp_v6_flush_pending_frames(sk
);
1166 err
= udp_v6_push_pending_frames(sk
);
1167 else if (unlikely(skb_queue_empty(&sk
->sk_write_queue
)))
1172 ip6_dst_store(sk
, dst
,
1173 ipv6_addr_equal(&fl
.fl6_dst
, &np
->daddr
) ?
1175 #ifdef CONFIG_IPV6_SUBTREES
1176 ipv6_addr_equal(&fl
.fl6_src
, &np
->saddr
) ?
1187 err
= np
->recverr
? net_xmit_errno(err
) : 0;
1191 fl6_sock_release(flowlabel
);
1195 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1196 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1197 * we don't have a good statistic (IpOutDiscards but it can be too many
1198 * things). We could add another new stat but at least for now that
1199 * seems like overkill.
1201 if (err
== -ENOBUFS
|| test_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
)) {
1202 UDP6_INC_STATS_USER(sock_net(sk
),
1203 UDP_MIB_SNDBUFERRORS
, is_udplite
);
1209 if (!(msg
->msg_flags
&MSG_PROBE
) || len
)
1210 goto back_from_confirm
;
1215 void udpv6_destroy_sock(struct sock
*sk
)
1218 udp_v6_flush_pending_frames(sk
);
1221 inet6_destroy_sock(sk
);
1225 * Socket option code for UDP
1227 int udpv6_setsockopt(struct sock
*sk
, int level
, int optname
,
1228 char __user
*optval
, unsigned int optlen
)
1230 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1231 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1232 udp_v6_push_pending_frames
);
1233 return ipv6_setsockopt(sk
, level
, optname
, optval
, optlen
);
1236 #ifdef CONFIG_COMPAT
1237 int compat_udpv6_setsockopt(struct sock
*sk
, int level
, int optname
,
1238 char __user
*optval
, unsigned int optlen
)
1240 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1241 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1242 udp_v6_push_pending_frames
);
1243 return compat_ipv6_setsockopt(sk
, level
, optname
, optval
, optlen
);
1247 int udpv6_getsockopt(struct sock
*sk
, int level
, int optname
,
1248 char __user
*optval
, int __user
*optlen
)
1250 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1251 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1252 return ipv6_getsockopt(sk
, level
, optname
, optval
, optlen
);
1255 #ifdef CONFIG_COMPAT
1256 int compat_udpv6_getsockopt(struct sock
*sk
, int level
, int optname
,
1257 char __user
*optval
, int __user
*optlen
)
1259 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1260 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1261 return compat_ipv6_getsockopt(sk
, level
, optname
, optval
, optlen
);
1265 static int udp6_ufo_send_check(struct sk_buff
*skb
)
1267 struct ipv6hdr
*ipv6h
;
1270 if (!pskb_may_pull(skb
, sizeof(*uh
)))
1273 ipv6h
= ipv6_hdr(skb
);
1276 uh
->check
= ~csum_ipv6_magic(&ipv6h
->saddr
, &ipv6h
->daddr
, skb
->len
,
1278 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
1279 skb
->csum_offset
= offsetof(struct udphdr
, check
);
1280 skb
->ip_summed
= CHECKSUM_PARTIAL
;
1284 static struct sk_buff
*udp6_ufo_fragment(struct sk_buff
*skb
, int features
)
1286 struct sk_buff
*segs
= ERR_PTR(-EINVAL
);
1288 unsigned int unfrag_ip6hlen
, unfrag_len
;
1289 struct frag_hdr
*fptr
;
1290 u8
*mac_start
, *prevhdr
;
1292 u8 frag_hdr_sz
= sizeof(struct frag_hdr
);
1296 mss
= skb_shinfo(skb
)->gso_size
;
1297 if (unlikely(skb
->len
<= mss
))
1300 if (skb_gso_ok(skb
, features
| NETIF_F_GSO_ROBUST
)) {
1301 /* Packet is from an untrusted source, reset gso_segs. */
1302 int type
= skb_shinfo(skb
)->gso_type
;
1304 if (unlikely(type
& ~(SKB_GSO_UDP
| SKB_GSO_DODGY
) ||
1305 !(type
& (SKB_GSO_UDP
))))
1308 skb_shinfo(skb
)->gso_segs
= DIV_ROUND_UP(skb
->len
, mss
);
1314 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
1315 * do checksum of UDP packets sent as multiple IP fragments.
1317 offset
= skb
->csum_start
- skb_headroom(skb
);
1318 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1319 offset
+= skb
->csum_offset
;
1320 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1321 skb
->ip_summed
= CHECKSUM_NONE
;
1323 /* Check if there is enough headroom to insert fragment header. */
1324 if ((skb_headroom(skb
) < frag_hdr_sz
) &&
1325 pskb_expand_head(skb
, frag_hdr_sz
, 0, GFP_ATOMIC
))
1328 /* Find the unfragmentable header and shift it left by frag_hdr_sz
1329 * bytes to insert fragment header.
1331 unfrag_ip6hlen
= ip6_find_1stfragopt(skb
, &prevhdr
);
1333 *prevhdr
= NEXTHDR_FRAGMENT
;
1334 unfrag_len
= skb_network_header(skb
) - skb_mac_header(skb
) +
1336 mac_start
= skb_mac_header(skb
);
1337 memmove(mac_start
-frag_hdr_sz
, mac_start
, unfrag_len
);
1339 skb
->mac_header
-= frag_hdr_sz
;
1340 skb
->network_header
-= frag_hdr_sz
;
1342 fptr
= (struct frag_hdr
*)(skb_network_header(skb
) + unfrag_ip6hlen
);
1343 fptr
->nexthdr
= nexthdr
;
1345 ipv6_select_ident(fptr
);
1347 /* Fragment the skb. ipv6 header and the remaining fields of the
1348 * fragment header are updated in ipv6_gso_segment()
1350 segs
= skb_segment(skb
, features
);
1356 static const struct inet6_protocol udpv6_protocol
= {
1357 .handler
= udpv6_rcv
,
1358 .err_handler
= udpv6_err
,
1359 .gso_send_check
= udp6_ufo_send_check
,
1360 .gso_segment
= udp6_ufo_fragment
,
1361 .flags
= INET6_PROTO_NOPOLICY
|INET6_PROTO_FINAL
,
1364 /* ------------------------------------------------------------------------ */
1365 #ifdef CONFIG_PROC_FS
1367 static void udp6_sock_seq_show(struct seq_file
*seq
, struct sock
*sp
, int bucket
)
1369 struct inet_sock
*inet
= inet_sk(sp
);
1370 struct ipv6_pinfo
*np
= inet6_sk(sp
);
1371 struct in6_addr
*dest
, *src
;
1375 src
= &np
->rcv_saddr
;
1376 destp
= ntohs(inet
->inet_dport
);
1377 srcp
= ntohs(inet
->inet_sport
);
1379 "%5d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1380 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d\n",
1382 src
->s6_addr32
[0], src
->s6_addr32
[1],
1383 src
->s6_addr32
[2], src
->s6_addr32
[3], srcp
,
1384 dest
->s6_addr32
[0], dest
->s6_addr32
[1],
1385 dest
->s6_addr32
[2], dest
->s6_addr32
[3], destp
,
1387 sk_wmem_alloc_get(sp
),
1388 sk_rmem_alloc_get(sp
),
1392 atomic_read(&sp
->sk_refcnt
), sp
,
1393 atomic_read(&sp
->sk_drops
));
1396 int udp6_seq_show(struct seq_file
*seq
, void *v
)
1398 if (v
== SEQ_START_TOKEN
)
1403 "st tx_queue rx_queue tr tm->when retrnsmt"
1404 " uid timeout inode ref pointer drops\n");
1406 udp6_sock_seq_show(seq
, v
, ((struct udp_iter_state
*)seq
->private)->bucket
);
1410 static struct udp_seq_afinfo udp6_seq_afinfo
= {
1413 .udp_table
= &udp_table
,
1415 .owner
= THIS_MODULE
,
1418 .show
= udp6_seq_show
,
1422 int __net_init
udp6_proc_init(struct net
*net
)
1424 return udp_proc_register(net
, &udp6_seq_afinfo
);
1427 void udp6_proc_exit(struct net
*net
) {
1428 udp_proc_unregister(net
, &udp6_seq_afinfo
);
1430 #endif /* CONFIG_PROC_FS */
1432 /* ------------------------------------------------------------------------ */
1434 struct proto udpv6_prot
= {
1436 .owner
= THIS_MODULE
,
1437 .close
= udp_lib_close
,
1438 .connect
= ip6_datagram_connect
,
1439 .disconnect
= udp_disconnect
,
1441 .destroy
= udpv6_destroy_sock
,
1442 .setsockopt
= udpv6_setsockopt
,
1443 .getsockopt
= udpv6_getsockopt
,
1444 .sendmsg
= udpv6_sendmsg
,
1445 .recvmsg
= udpv6_recvmsg
,
1446 .backlog_rcv
= udpv6_queue_rcv_skb
,
1447 .hash
= udp_lib_hash
,
1448 .unhash
= udp_lib_unhash
,
1449 .get_port
= udp_v6_get_port
,
1450 .memory_allocated
= &udp_memory_allocated
,
1451 .sysctl_mem
= sysctl_udp_mem
,
1452 .sysctl_wmem
= &sysctl_udp_wmem_min
,
1453 .sysctl_rmem
= &sysctl_udp_rmem_min
,
1454 .obj_size
= sizeof(struct udp6_sock
),
1455 .slab_flags
= SLAB_DESTROY_BY_RCU
,
1456 .h
.udp_table
= &udp_table
,
1457 #ifdef CONFIG_COMPAT
1458 .compat_setsockopt
= compat_udpv6_setsockopt
,
1459 .compat_getsockopt
= compat_udpv6_getsockopt
,
1463 static struct inet_protosw udpv6_protosw
= {
1465 .protocol
= IPPROTO_UDP
,
1466 .prot
= &udpv6_prot
,
1467 .ops
= &inet6_dgram_ops
,
1468 .no_check
= UDP_CSUM_DEFAULT
,
1469 .flags
= INET_PROTOSW_PERMANENT
,
1473 int __init
udpv6_init(void)
1477 ret
= inet6_add_protocol(&udpv6_protocol
, IPPROTO_UDP
);
1481 ret
= inet6_register_protosw(&udpv6_protosw
);
1483 goto out_udpv6_protocol
;
1488 inet6_del_protocol(&udpv6_protocol
, IPPROTO_UDP
);
1492 void udpv6_exit(void)
1494 inet6_unregister_protosw(&udpv6_protosw
);
1495 inet6_del_protocol(&udpv6_protocol
, IPPROTO_UDP
);