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
= sk_rcv_saddr(sk
);
58 __be32 sk2_rcv_saddr
= sk_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 void udp_v6_rehash(struct sock
*sk
)
116 u16 new_hash
= udp6_portaddr_hash(sock_net(sk
),
117 &inet6_sk(sk
)->rcv_saddr
,
118 inet_sk(sk
)->inet_num
);
120 udp_lib_rehash(sk
, new_hash
);
123 static inline int compute_score(struct sock
*sk
, struct net
*net
,
125 const struct in6_addr
*saddr
, __be16 sport
,
126 const struct in6_addr
*daddr
, __be16 dport
,
131 if (net_eq(sock_net(sk
), net
) && udp_sk(sk
)->udp_port_hash
== hnum
&&
132 sk
->sk_family
== PF_INET6
) {
133 struct ipv6_pinfo
*np
= inet6_sk(sk
);
134 struct inet_sock
*inet
= inet_sk(sk
);
137 if (inet
->inet_dport
) {
138 if (inet
->inet_dport
!= sport
)
142 if (!ipv6_addr_any(&np
->rcv_saddr
)) {
143 if (!ipv6_addr_equal(&np
->rcv_saddr
, daddr
))
147 if (!ipv6_addr_any(&np
->daddr
)) {
148 if (!ipv6_addr_equal(&np
->daddr
, saddr
))
152 if (sk
->sk_bound_dev_if
) {
153 if (sk
->sk_bound_dev_if
!= dif
)
161 #define SCORE2_MAX (1 + 1 + 1)
162 static inline int compute_score2(struct sock
*sk
, struct net
*net
,
163 const struct in6_addr
*saddr
, __be16 sport
,
164 const struct in6_addr
*daddr
, unsigned short hnum
,
169 if (net_eq(sock_net(sk
), net
) && udp_sk(sk
)->udp_port_hash
== hnum
&&
170 sk
->sk_family
== PF_INET6
) {
171 struct ipv6_pinfo
*np
= inet6_sk(sk
);
172 struct inet_sock
*inet
= inet_sk(sk
);
174 if (!ipv6_addr_equal(&np
->rcv_saddr
, daddr
))
177 if (inet
->inet_dport
) {
178 if (inet
->inet_dport
!= sport
)
182 if (!ipv6_addr_any(&np
->daddr
)) {
183 if (!ipv6_addr_equal(&np
->daddr
, saddr
))
187 if (sk
->sk_bound_dev_if
) {
188 if (sk
->sk_bound_dev_if
!= dif
)
197 /* called with read_rcu_lock() */
198 static struct sock
*udp6_lib_lookup2(struct net
*net
,
199 const struct in6_addr
*saddr
, __be16 sport
,
200 const struct in6_addr
*daddr
, unsigned int hnum
, int dif
,
201 struct udp_hslot
*hslot2
, unsigned int slot2
)
203 struct sock
*sk
, *result
;
204 struct hlist_nulls_node
*node
;
210 udp_portaddr_for_each_entry_rcu(sk
, node
, &hslot2
->head
) {
211 score
= compute_score2(sk
, net
, saddr
, sport
,
213 if (score
> badness
) {
216 if (score
== SCORE2_MAX
)
221 * if the nulls value we got at the end of this lookup is
222 * not the expected one, we must restart lookup.
223 * We probably met an item that was moved to another chain.
225 if (get_nulls_value(node
) != slot2
)
230 if (unlikely(!atomic_inc_not_zero_hint(&result
->sk_refcnt
, 2)))
232 else if (unlikely(compute_score2(result
, net
, saddr
, sport
,
233 daddr
, hnum
, dif
) < badness
)) {
241 static struct sock
*__udp6_lib_lookup(struct net
*net
,
242 const struct in6_addr
*saddr
, __be16 sport
,
243 const struct in6_addr
*daddr
, __be16 dport
,
244 int dif
, struct udp_table
*udptable
)
246 struct sock
*sk
, *result
;
247 struct hlist_nulls_node
*node
;
248 unsigned short hnum
= ntohs(dport
);
249 unsigned int hash2
, slot2
, slot
= udp_hashfn(net
, hnum
, udptable
->mask
);
250 struct udp_hslot
*hslot2
, *hslot
= &udptable
->hash
[slot
];
254 if (hslot
->count
> 10) {
255 hash2
= udp6_portaddr_hash(net
, daddr
, hnum
);
256 slot2
= hash2
& udptable
->mask
;
257 hslot2
= &udptable
->hash2
[slot2
];
258 if (hslot
->count
< hslot2
->count
)
261 result
= udp6_lib_lookup2(net
, saddr
, sport
,
265 hash2
= udp6_portaddr_hash(net
, &in6addr_any
, hnum
);
266 slot2
= hash2
& udptable
->mask
;
267 hslot2
= &udptable
->hash2
[slot2
];
268 if (hslot
->count
< hslot2
->count
)
271 result
= udp6_lib_lookup2(net
, saddr
, sport
,
272 &in6addr_any
, hnum
, dif
,
281 sk_nulls_for_each_rcu(sk
, node
, &hslot
->head
) {
282 score
= compute_score(sk
, net
, hnum
, saddr
, sport
, daddr
, dport
, dif
);
283 if (score
> badness
) {
289 * if the nulls value we got at the end of this lookup is
290 * not the expected one, we must restart lookup.
291 * We probably met an item that was moved to another chain.
293 if (get_nulls_value(node
) != slot
)
297 if (unlikely(!atomic_inc_not_zero_hint(&result
->sk_refcnt
, 2)))
299 else if (unlikely(compute_score(result
, net
, hnum
, saddr
, sport
,
300 daddr
, dport
, dif
) < badness
)) {
309 static struct sock
*__udp6_lib_lookup_skb(struct sk_buff
*skb
,
310 __be16 sport
, __be16 dport
,
311 struct udp_table
*udptable
)
314 const struct ipv6hdr
*iph
= ipv6_hdr(skb
);
316 if (unlikely(sk
= skb_steal_sock(skb
)))
318 return __udp6_lib_lookup(dev_net(skb_dst(skb
)->dev
), &iph
->saddr
, sport
,
319 &iph
->daddr
, dport
, inet6_iif(skb
),
323 struct sock
*udp6_lib_lookup(struct net
*net
, const struct in6_addr
*saddr
, __be16 sport
,
324 const struct in6_addr
*daddr
, __be16 dport
, int dif
)
326 return __udp6_lib_lookup(net
, saddr
, sport
, daddr
, dport
, dif
, &udp_table
);
328 EXPORT_SYMBOL_GPL(udp6_lib_lookup
);
332 * This should be easy, if there is something there we
333 * return it, otherwise we block.
336 int udpv6_recvmsg(struct kiocb
*iocb
, struct sock
*sk
,
337 struct msghdr
*msg
, size_t len
,
338 int noblock
, int flags
, int *addr_len
)
340 struct ipv6_pinfo
*np
= inet6_sk(sk
);
341 struct inet_sock
*inet
= inet_sk(sk
);
346 int is_udplite
= IS_UDPLITE(sk
);
351 *addr_len
=sizeof(struct sockaddr_in6
);
353 if (flags
& MSG_ERRQUEUE
)
354 return ipv6_recv_error(sk
, msg
, len
);
356 if (np
->rxpmtu
&& np
->rxopt
.bits
.rxpmtu
)
357 return ipv6_recv_rxpmtu(sk
, msg
, len
);
360 skb
= __skb_recv_datagram(sk
, flags
| (noblock
? MSG_DONTWAIT
: 0),
365 ulen
= skb
->len
- sizeof(struct udphdr
);
369 msg
->msg_flags
|= MSG_TRUNC
;
371 is_udp4
= (skb
->protocol
== htons(ETH_P_IP
));
374 * If checksum is needed at all, try to do it while copying the
375 * data. If the data is truncated, or if we only want a partial
376 * coverage checksum (UDP-Lite), do it before the copy.
379 if (len
< ulen
|| UDP_SKB_CB(skb
)->partial_cov
) {
380 if (udp_lib_checksum_complete(skb
))
384 if (skb_csum_unnecessary(skb
))
385 err
= skb_copy_datagram_iovec(skb
, sizeof(struct udphdr
),
388 err
= skb_copy_and_csum_datagram_iovec(skb
, sizeof(struct udphdr
), msg
->msg_iov
);
397 UDP_INC_STATS_USER(sock_net(sk
),
398 UDP_MIB_INDATAGRAMS
, is_udplite
);
400 UDP6_INC_STATS_USER(sock_net(sk
),
401 UDP_MIB_INDATAGRAMS
, is_udplite
);
404 sock_recv_ts_and_drops(msg
, sk
, skb
);
406 /* Copy the address. */
408 struct sockaddr_in6
*sin6
;
410 sin6
= (struct sockaddr_in6
*) msg
->msg_name
;
411 sin6
->sin6_family
= AF_INET6
;
412 sin6
->sin6_port
= udp_hdr(skb
)->source
;
413 sin6
->sin6_flowinfo
= 0;
414 sin6
->sin6_scope_id
= 0;
417 ipv6_addr_set_v4mapped(ip_hdr(skb
)->saddr
,
420 ipv6_addr_copy(&sin6
->sin6_addr
,
421 &ipv6_hdr(skb
)->saddr
);
422 if (ipv6_addr_type(&sin6
->sin6_addr
) & IPV6_ADDR_LINKLOCAL
)
423 sin6
->sin6_scope_id
= IP6CB(skb
)->iif
;
428 if (inet
->cmsg_flags
)
429 ip_cmsg_recv(msg
, skb
);
432 datagram_recv_ctl(sk
, msg
, skb
);
436 if (flags
& MSG_TRUNC
)
440 skb_free_datagram_locked(sk
, skb
);
445 slow
= lock_sock_fast(sk
);
446 if (!skb_kill_datagram(sk
, skb
, flags
)) {
448 UDP_INC_STATS_USER(sock_net(sk
),
449 UDP_MIB_INERRORS
, is_udplite
);
451 UDP6_INC_STATS_USER(sock_net(sk
),
452 UDP_MIB_INERRORS
, is_udplite
);
454 unlock_sock_fast(sk
, slow
);
459 /* starting over for a new packet */
460 msg
->msg_flags
&= ~MSG_TRUNC
;
464 void __udp6_lib_err(struct sk_buff
*skb
, struct inet6_skb_parm
*opt
,
465 u8 type
, u8 code
, int offset
, __be32 info
,
466 struct udp_table
*udptable
)
468 struct ipv6_pinfo
*np
;
469 const struct ipv6hdr
*hdr
= (const struct ipv6hdr
*)skb
->data
;
470 const struct in6_addr
*saddr
= &hdr
->saddr
;
471 const struct in6_addr
*daddr
= &hdr
->daddr
;
472 struct udphdr
*uh
= (struct udphdr
*)(skb
->data
+offset
);
476 sk
= __udp6_lib_lookup(dev_net(skb
->dev
), daddr
, uh
->dest
,
477 saddr
, uh
->source
, inet6_iif(skb
), udptable
);
483 if (!icmpv6_err_convert(type
, code
, &err
) && !np
->recverr
)
486 if (sk
->sk_state
!= TCP_ESTABLISHED
&& !np
->recverr
)
490 ipv6_icmp_error(sk
, skb
, err
, uh
->dest
, ntohl(info
), (u8
*)(uh
+1));
493 sk
->sk_error_report(sk
);
498 static __inline__
void udpv6_err(struct sk_buff
*skb
,
499 struct inet6_skb_parm
*opt
, u8 type
,
500 u8 code
, int offset
, __be32 info
)
502 __udp6_lib_err(skb
, opt
, type
, code
, offset
, info
, &udp_table
);
505 int udpv6_queue_rcv_skb(struct sock
* sk
, struct sk_buff
*skb
)
507 struct udp_sock
*up
= udp_sk(sk
);
509 int is_udplite
= IS_UDPLITE(sk
);
511 if (!ipv6_addr_any(&inet6_sk(sk
)->daddr
))
512 sock_rps_save_rxhash(sk
, skb
->rxhash
);
514 if (!xfrm6_policy_check(sk
, XFRM_POLICY_IN
, skb
))
518 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
520 if ((is_udplite
& UDPLITE_RECV_CC
) && UDP_SKB_CB(skb
)->partial_cov
) {
522 if (up
->pcrlen
== 0) { /* full coverage was set */
523 LIMIT_NETDEBUG(KERN_WARNING
"UDPLITE6: partial coverage"
524 " %d while full coverage %d requested\n",
525 UDP_SKB_CB(skb
)->cscov
, skb
->len
);
528 if (UDP_SKB_CB(skb
)->cscov
< up
->pcrlen
) {
529 LIMIT_NETDEBUG(KERN_WARNING
"UDPLITE6: coverage %d "
530 "too small, need min %d\n",
531 UDP_SKB_CB(skb
)->cscov
, up
->pcrlen
);
536 if (rcu_dereference_raw(sk
->sk_filter
)) {
537 if (udp_lib_checksum_complete(skb
))
541 if ((rc
= ip_queue_rcv_skb(sk
, skb
)) < 0) {
542 /* Note that an ENOMEM error is charged twice */
544 UDP6_INC_STATS_BH(sock_net(sk
),
545 UDP_MIB_RCVBUFERRORS
, is_udplite
);
546 goto drop_no_sk_drops_inc
;
551 atomic_inc(&sk
->sk_drops
);
552 drop_no_sk_drops_inc
:
553 UDP6_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
558 static struct sock
*udp_v6_mcast_next(struct net
*net
, struct sock
*sk
,
559 __be16 loc_port
, const struct in6_addr
*loc_addr
,
560 __be16 rmt_port
, const struct in6_addr
*rmt_addr
,
563 struct hlist_nulls_node
*node
;
565 unsigned short num
= ntohs(loc_port
);
567 sk_nulls_for_each_from(s
, node
) {
568 struct inet_sock
*inet
= inet_sk(s
);
570 if (!net_eq(sock_net(s
), net
))
573 if (udp_sk(s
)->udp_port_hash
== num
&&
574 s
->sk_family
== PF_INET6
) {
575 struct ipv6_pinfo
*np
= inet6_sk(s
);
576 if (inet
->inet_dport
) {
577 if (inet
->inet_dport
!= rmt_port
)
580 if (!ipv6_addr_any(&np
->daddr
) &&
581 !ipv6_addr_equal(&np
->daddr
, rmt_addr
))
584 if (s
->sk_bound_dev_if
&& s
->sk_bound_dev_if
!= dif
)
587 if (!ipv6_addr_any(&np
->rcv_saddr
)) {
588 if (!ipv6_addr_equal(&np
->rcv_saddr
, loc_addr
))
591 if (!inet6_mc_check(s
, loc_addr
, rmt_addr
))
599 static void flush_stack(struct sock
**stack
, unsigned int count
,
600 struct sk_buff
*skb
, unsigned int final
)
604 struct sk_buff
*skb1
;
606 for (i
= 0; i
< count
; i
++) {
607 skb1
= (i
== final
) ? skb
: skb_clone(skb
, GFP_ATOMIC
);
611 if (sk_rcvqueues_full(sk
, skb1
)) {
616 if (!sock_owned_by_user(sk
))
617 udpv6_queue_rcv_skb(sk
, skb1
);
618 else if (sk_add_backlog(sk
, skb1
)) {
627 atomic_inc(&sk
->sk_drops
);
628 UDP6_INC_STATS_BH(sock_net(sk
),
629 UDP_MIB_RCVBUFERRORS
, IS_UDPLITE(sk
));
630 UDP6_INC_STATS_BH(sock_net(sk
),
631 UDP_MIB_INERRORS
, IS_UDPLITE(sk
));
635 * Note: called only from the BH handler context,
636 * so we don't need to lock the hashes.
638 static int __udp6_lib_mcast_deliver(struct net
*net
, struct sk_buff
*skb
,
639 const struct in6_addr
*saddr
, const struct in6_addr
*daddr
,
640 struct udp_table
*udptable
)
642 struct sock
*sk
, *stack
[256 / sizeof(struct sock
*)];
643 const struct udphdr
*uh
= udp_hdr(skb
);
644 struct udp_hslot
*hslot
= udp_hashslot(udptable
, net
, ntohs(uh
->dest
));
646 unsigned int i
, count
= 0;
648 spin_lock(&hslot
->lock
);
649 sk
= sk_nulls_head(&hslot
->head
);
650 dif
= inet6_iif(skb
);
651 sk
= udp_v6_mcast_next(net
, sk
, uh
->dest
, daddr
, uh
->source
, saddr
, dif
);
654 sk
= udp_v6_mcast_next(net
, sk_nulls_next(sk
), uh
->dest
, daddr
,
655 uh
->source
, saddr
, dif
);
656 if (unlikely(count
== ARRAY_SIZE(stack
))) {
659 flush_stack(stack
, count
, skb
, ~0);
664 * before releasing the lock, we must take reference on sockets
666 for (i
= 0; i
< count
; i
++)
669 spin_unlock(&hslot
->lock
);
672 flush_stack(stack
, count
, skb
, count
- 1);
674 for (i
= 0; i
< count
; i
++)
682 static inline int udp6_csum_init(struct sk_buff
*skb
, struct udphdr
*uh
,
687 UDP_SKB_CB(skb
)->partial_cov
= 0;
688 UDP_SKB_CB(skb
)->cscov
= skb
->len
;
690 if (proto
== IPPROTO_UDPLITE
) {
691 err
= udplite_checksum_init(skb
, uh
);
696 if (uh
->check
== 0) {
697 /* RFC 2460 section 8.1 says that we SHOULD log
698 this error. Well, it is reasonable.
700 LIMIT_NETDEBUG(KERN_INFO
"IPv6: udp checksum is 0\n");
703 if (skb
->ip_summed
== CHECKSUM_COMPLETE
&&
704 !csum_ipv6_magic(&ipv6_hdr(skb
)->saddr
, &ipv6_hdr(skb
)->daddr
,
705 skb
->len
, proto
, skb
->csum
))
706 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
708 if (!skb_csum_unnecessary(skb
))
709 skb
->csum
= ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb
)->saddr
,
710 &ipv6_hdr(skb
)->daddr
,
711 skb
->len
, proto
, 0));
716 int __udp6_lib_rcv(struct sk_buff
*skb
, struct udp_table
*udptable
,
719 struct net
*net
= dev_net(skb
->dev
);
722 const struct in6_addr
*saddr
, *daddr
;
725 if (!pskb_may_pull(skb
, sizeof(struct udphdr
)))
728 saddr
= &ipv6_hdr(skb
)->saddr
;
729 daddr
= &ipv6_hdr(skb
)->daddr
;
732 ulen
= ntohs(uh
->len
);
736 if (proto
== IPPROTO_UDP
) {
737 /* UDP validates ulen. */
739 /* Check for jumbo payload */
743 if (ulen
< sizeof(*uh
))
746 if (ulen
< skb
->len
) {
747 if (pskb_trim_rcsum(skb
, ulen
))
749 saddr
= &ipv6_hdr(skb
)->saddr
;
750 daddr
= &ipv6_hdr(skb
)->daddr
;
755 if (udp6_csum_init(skb
, uh
, proto
))
759 * Multicast receive code
761 if (ipv6_addr_is_multicast(daddr
))
762 return __udp6_lib_mcast_deliver(net
, skb
,
763 saddr
, daddr
, udptable
);
768 * check socket cache ... must talk to Alan about his plans
769 * for sock caches... i'll skip this for now.
771 sk
= __udp6_lib_lookup_skb(skb
, uh
->source
, uh
->dest
, udptable
);
774 if (!xfrm6_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
777 if (udp_lib_checksum_complete(skb
))
779 UDP6_INC_STATS_BH(net
, UDP_MIB_NOPORTS
,
780 proto
== IPPROTO_UDPLITE
);
782 icmpv6_send(skb
, ICMPV6_DEST_UNREACH
, ICMPV6_PORT_UNREACH
, 0);
790 if (sk_rcvqueues_full(sk
, skb
)) {
795 if (!sock_owned_by_user(sk
))
796 udpv6_queue_rcv_skb(sk
, skb
);
797 else if (sk_add_backlog(sk
, skb
)) {
798 atomic_inc(&sk
->sk_drops
);
808 LIMIT_NETDEBUG(KERN_DEBUG
"UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
809 proto
== IPPROTO_UDPLITE
? "-Lite" : "",
818 UDP6_INC_STATS_BH(net
, UDP_MIB_INERRORS
, proto
== IPPROTO_UDPLITE
);
823 static __inline__
int udpv6_rcv(struct sk_buff
*skb
)
825 return __udp6_lib_rcv(skb
, &udp_table
, IPPROTO_UDP
);
829 * Throw away all pending data and cancel the corking. Socket is locked.
831 static void udp_v6_flush_pending_frames(struct sock
*sk
)
833 struct udp_sock
*up
= udp_sk(sk
);
835 if (up
->pending
== AF_INET
)
836 udp_flush_pending_frames(sk
);
837 else if (up
->pending
) {
840 ip6_flush_pending_frames(sk
);
845 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
846 * @sk: socket we are sending on
847 * @skb: sk_buff containing the filled-in UDP header
848 * (checksum field must be zeroed out)
850 static void udp6_hwcsum_outgoing(struct sock
*sk
, struct sk_buff
*skb
,
851 const struct in6_addr
*saddr
,
852 const struct in6_addr
*daddr
, int len
)
855 struct udphdr
*uh
= udp_hdr(skb
);
858 if (skb_queue_len(&sk
->sk_write_queue
) == 1) {
859 /* Only one fragment on the socket. */
860 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
861 skb
->csum_offset
= offsetof(struct udphdr
, check
);
862 uh
->check
= ~csum_ipv6_magic(saddr
, daddr
, len
, IPPROTO_UDP
, 0);
865 * HW-checksum won't work as there are two or more
866 * fragments on the socket so that all csums of sk_buffs
869 offset
= skb_transport_offset(skb
);
870 skb
->csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
872 skb
->ip_summed
= CHECKSUM_NONE
;
874 skb_queue_walk(&sk
->sk_write_queue
, skb
) {
875 csum
= csum_add(csum
, skb
->csum
);
878 uh
->check
= csum_ipv6_magic(saddr
, daddr
, len
, IPPROTO_UDP
,
881 uh
->check
= CSUM_MANGLED_0
;
889 static int udp_v6_push_pending_frames(struct sock
*sk
)
893 struct udp_sock
*up
= udp_sk(sk
);
894 struct inet_sock
*inet
= inet_sk(sk
);
895 struct flowi6
*fl6
= &inet
->cork
.fl
.u
.ip6
;
897 int is_udplite
= IS_UDPLITE(sk
);
900 /* Grab the skbuff where UDP header space exists. */
901 if ((skb
= skb_peek(&sk
->sk_write_queue
)) == NULL
)
905 * Create a UDP header
908 uh
->source
= fl6
->fl6_sport
;
909 uh
->dest
= fl6
->fl6_dport
;
910 uh
->len
= htons(up
->len
);
914 csum
= udplite_csum_outgoing(sk
, skb
);
915 else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) { /* UDP hardware csum */
916 udp6_hwcsum_outgoing(sk
, skb
, &fl6
->saddr
, &fl6
->daddr
,
920 csum
= udp_csum_outgoing(sk
, skb
);
922 /* add protocol-dependent pseudo-header */
923 uh
->check
= csum_ipv6_magic(&fl6
->saddr
, &fl6
->daddr
,
924 up
->len
, fl6
->flowi6_proto
, csum
);
926 uh
->check
= CSUM_MANGLED_0
;
929 err
= ip6_push_pending_frames(sk
);
931 if (err
== -ENOBUFS
&& !inet6_sk(sk
)->recverr
) {
932 UDP6_INC_STATS_USER(sock_net(sk
),
933 UDP_MIB_SNDBUFERRORS
, is_udplite
);
937 UDP6_INC_STATS_USER(sock_net(sk
),
938 UDP_MIB_OUTDATAGRAMS
, is_udplite
);
945 int udpv6_sendmsg(struct kiocb
*iocb
, struct sock
*sk
,
946 struct msghdr
*msg
, size_t len
)
948 struct ipv6_txoptions opt_space
;
949 struct udp_sock
*up
= udp_sk(sk
);
950 struct inet_sock
*inet
= inet_sk(sk
);
951 struct ipv6_pinfo
*np
= inet6_sk(sk
);
952 struct sockaddr_in6
*sin6
= (struct sockaddr_in6
*) msg
->msg_name
;
953 struct in6_addr
*daddr
, *final_p
, final
;
954 struct ipv6_txoptions
*opt
= NULL
;
955 struct ip6_flowlabel
*flowlabel
= NULL
;
957 struct dst_entry
*dst
;
958 int addr_len
= msg
->msg_namelen
;
963 int corkreq
= up
->corkflag
|| msg
->msg_flags
&MSG_MORE
;
966 int is_udplite
= IS_UDPLITE(sk
);
967 int (*getfrag
)(void *, char *, int, int, int, struct sk_buff
*);
969 /* destination address check */
971 if (addr_len
< offsetof(struct sockaddr
, sa_data
))
974 switch (sin6
->sin6_family
) {
976 if (addr_len
< SIN6_LEN_RFC2133
)
978 daddr
= &sin6
->sin6_addr
;
983 msg
->msg_name
= sin6
= NULL
;
984 msg
->msg_namelen
= addr_len
= 0;
990 } else if (!up
->pending
) {
991 if (sk
->sk_state
!= TCP_ESTABLISHED
)
992 return -EDESTADDRREQ
;
998 if (ipv6_addr_v4mapped(daddr
)) {
999 struct sockaddr_in sin
;
1000 sin
.sin_family
= AF_INET
;
1001 sin
.sin_port
= sin6
? sin6
->sin6_port
: inet
->inet_dport
;
1002 sin
.sin_addr
.s_addr
= daddr
->s6_addr32
[3];
1003 msg
->msg_name
= &sin
;
1004 msg
->msg_namelen
= sizeof(sin
);
1006 if (__ipv6_only_sock(sk
))
1007 return -ENETUNREACH
;
1008 return udp_sendmsg(iocb
, sk
, msg
, len
);
1012 if (up
->pending
== AF_INET
)
1013 return udp_sendmsg(iocb
, sk
, msg
, len
);
1015 /* Rough check on arithmetic overflow,
1016 better check is made in ip6_append_data().
1018 if (len
> INT_MAX
- sizeof(struct udphdr
))
1023 * There are pending frames.
1024 * The socket lock must be held while it's corked.
1027 if (likely(up
->pending
)) {
1028 if (unlikely(up
->pending
!= AF_INET6
)) {
1030 return -EAFNOSUPPORT
;
1033 goto do_append_data
;
1037 ulen
+= sizeof(struct udphdr
);
1039 memset(&fl6
, 0, sizeof(fl6
));
1042 if (sin6
->sin6_port
== 0)
1045 fl6
.fl6_dport
= sin6
->sin6_port
;
1046 daddr
= &sin6
->sin6_addr
;
1049 fl6
.flowlabel
= sin6
->sin6_flowinfo
&IPV6_FLOWINFO_MASK
;
1050 if (fl6
.flowlabel
&IPV6_FLOWLABEL_MASK
) {
1051 flowlabel
= fl6_sock_lookup(sk
, fl6
.flowlabel
);
1052 if (flowlabel
== NULL
)
1054 daddr
= &flowlabel
->dst
;
1059 * Otherwise it will be difficult to maintain
1062 if (sk
->sk_state
== TCP_ESTABLISHED
&&
1063 ipv6_addr_equal(daddr
, &np
->daddr
))
1066 if (addr_len
>= sizeof(struct sockaddr_in6
) &&
1067 sin6
->sin6_scope_id
&&
1068 ipv6_addr_type(daddr
)&IPV6_ADDR_LINKLOCAL
)
1069 fl6
.flowi6_oif
= sin6
->sin6_scope_id
;
1071 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1072 return -EDESTADDRREQ
;
1074 fl6
.fl6_dport
= inet
->inet_dport
;
1076 fl6
.flowlabel
= np
->flow_label
;
1080 if (!fl6
.flowi6_oif
)
1081 fl6
.flowi6_oif
= sk
->sk_bound_dev_if
;
1083 if (!fl6
.flowi6_oif
)
1084 fl6
.flowi6_oif
= np
->sticky_pktinfo
.ipi6_ifindex
;
1086 fl6
.flowi6_mark
= sk
->sk_mark
;
1088 if (msg
->msg_controllen
) {
1090 memset(opt
, 0, sizeof(struct ipv6_txoptions
));
1091 opt
->tot_len
= sizeof(*opt
);
1093 err
= datagram_send_ctl(sock_net(sk
), msg
, &fl6
, opt
, &hlimit
,
1094 &tclass
, &dontfrag
);
1096 fl6_sock_release(flowlabel
);
1099 if ((fl6
.flowlabel
&IPV6_FLOWLABEL_MASK
) && !flowlabel
) {
1100 flowlabel
= fl6_sock_lookup(sk
, fl6
.flowlabel
);
1101 if (flowlabel
== NULL
)
1104 if (!(opt
->opt_nflen
|opt
->opt_flen
))
1111 opt
= fl6_merge_options(&opt_space
, flowlabel
, opt
);
1112 opt
= ipv6_fixup_options(&opt_space
, opt
);
1114 fl6
.flowi6_proto
= sk
->sk_protocol
;
1115 if (!ipv6_addr_any(daddr
))
1116 ipv6_addr_copy(&fl6
.daddr
, daddr
);
1118 fl6
.daddr
.s6_addr
[15] = 0x1; /* :: means loopback (BSD'ism) */
1119 if (ipv6_addr_any(&fl6
.saddr
) && !ipv6_addr_any(&np
->saddr
))
1120 ipv6_addr_copy(&fl6
.saddr
, &np
->saddr
);
1121 fl6
.fl6_sport
= inet
->inet_sport
;
1123 final_p
= fl6_update_dst(&fl6
, opt
, &final
);
1127 if (!fl6
.flowi6_oif
&& ipv6_addr_is_multicast(&fl6
.daddr
)) {
1128 fl6
.flowi6_oif
= np
->mcast_oif
;
1132 security_sk_classify_flow(sk
, flowi6_to_flowi(&fl6
));
1134 dst
= ip6_sk_dst_lookup_flow(sk
, &fl6
, final_p
, true);
1142 if (ipv6_addr_is_multicast(&fl6
.daddr
))
1143 hlimit
= np
->mcast_hops
;
1145 hlimit
= np
->hop_limit
;
1147 hlimit
= ip6_dst_hoplimit(dst
);
1151 tclass
= np
->tclass
;
1154 dontfrag
= np
->dontfrag
;
1156 if (msg
->msg_flags
&MSG_CONFIRM
)
1161 if (unlikely(up
->pending
)) {
1162 /* The socket is already corked while preparing it. */
1163 /* ... which is an evident application bug. --ANK */
1166 LIMIT_NETDEBUG(KERN_DEBUG
"udp cork app bug 2\n");
1171 up
->pending
= AF_INET6
;
1175 getfrag
= is_udplite
? udplite_getfrag
: ip_generic_getfrag
;
1176 err
= ip6_append_data(sk
, getfrag
, msg
->msg_iov
, ulen
,
1177 sizeof(struct udphdr
), hlimit
, tclass
, opt
, &fl6
,
1178 (struct rt6_info
*)dst
,
1179 corkreq
? msg
->msg_flags
|MSG_MORE
: msg
->msg_flags
, dontfrag
);
1181 udp_v6_flush_pending_frames(sk
);
1183 err
= udp_v6_push_pending_frames(sk
);
1184 else if (unlikely(skb_queue_empty(&sk
->sk_write_queue
)))
1189 ip6_dst_store(sk
, dst
,
1190 ipv6_addr_equal(&fl6
.daddr
, &np
->daddr
) ?
1192 #ifdef CONFIG_IPV6_SUBTREES
1193 ipv6_addr_equal(&fl6
.saddr
, &np
->saddr
) ?
1204 err
= np
->recverr
? net_xmit_errno(err
) : 0;
1208 fl6_sock_release(flowlabel
);
1212 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1213 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1214 * we don't have a good statistic (IpOutDiscards but it can be too many
1215 * things). We could add another new stat but at least for now that
1216 * seems like overkill.
1218 if (err
== -ENOBUFS
|| test_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
)) {
1219 UDP6_INC_STATS_USER(sock_net(sk
),
1220 UDP_MIB_SNDBUFERRORS
, is_udplite
);
1226 if (!(msg
->msg_flags
&MSG_PROBE
) || len
)
1227 goto back_from_confirm
;
1232 void udpv6_destroy_sock(struct sock
*sk
)
1235 udp_v6_flush_pending_frames(sk
);
1238 inet6_destroy_sock(sk
);
1242 * Socket option code for UDP
1244 int udpv6_setsockopt(struct sock
*sk
, int level
, int optname
,
1245 char __user
*optval
, unsigned int optlen
)
1247 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1248 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1249 udp_v6_push_pending_frames
);
1250 return ipv6_setsockopt(sk
, level
, optname
, optval
, optlen
);
1253 #ifdef CONFIG_COMPAT
1254 int compat_udpv6_setsockopt(struct sock
*sk
, int level
, int optname
,
1255 char __user
*optval
, unsigned int optlen
)
1257 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1258 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1259 udp_v6_push_pending_frames
);
1260 return compat_ipv6_setsockopt(sk
, level
, optname
, optval
, optlen
);
1264 int udpv6_getsockopt(struct sock
*sk
, int level
, int optname
,
1265 char __user
*optval
, int __user
*optlen
)
1267 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1268 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1269 return ipv6_getsockopt(sk
, level
, optname
, optval
, optlen
);
1272 #ifdef CONFIG_COMPAT
1273 int compat_udpv6_getsockopt(struct sock
*sk
, int level
, int optname
,
1274 char __user
*optval
, int __user
*optlen
)
1276 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1277 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1278 return compat_ipv6_getsockopt(sk
, level
, optname
, optval
, optlen
);
1282 static int udp6_ufo_send_check(struct sk_buff
*skb
)
1284 const struct ipv6hdr
*ipv6h
;
1287 if (!pskb_may_pull(skb
, sizeof(*uh
)))
1290 ipv6h
= ipv6_hdr(skb
);
1293 uh
->check
= ~csum_ipv6_magic(&ipv6h
->saddr
, &ipv6h
->daddr
, skb
->len
,
1295 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
1296 skb
->csum_offset
= offsetof(struct udphdr
, check
);
1297 skb
->ip_summed
= CHECKSUM_PARTIAL
;
1301 static struct sk_buff
*udp6_ufo_fragment(struct sk_buff
*skb
, u32 features
)
1303 struct sk_buff
*segs
= ERR_PTR(-EINVAL
);
1305 unsigned int unfrag_ip6hlen
, unfrag_len
;
1306 struct frag_hdr
*fptr
;
1307 u8
*mac_start
, *prevhdr
;
1309 u8 frag_hdr_sz
= sizeof(struct frag_hdr
);
1313 mss
= skb_shinfo(skb
)->gso_size
;
1314 if (unlikely(skb
->len
<= mss
))
1317 if (skb_gso_ok(skb
, features
| NETIF_F_GSO_ROBUST
)) {
1318 /* Packet is from an untrusted source, reset gso_segs. */
1319 int type
= skb_shinfo(skb
)->gso_type
;
1321 if (unlikely(type
& ~(SKB_GSO_UDP
| SKB_GSO_DODGY
) ||
1322 !(type
& (SKB_GSO_UDP
))))
1325 skb_shinfo(skb
)->gso_segs
= DIV_ROUND_UP(skb
->len
, mss
);
1331 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
1332 * do checksum of UDP packets sent as multiple IP fragments.
1334 offset
= skb_checksum_start_offset(skb
);
1335 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1336 offset
+= skb
->csum_offset
;
1337 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1338 skb
->ip_summed
= CHECKSUM_NONE
;
1340 /* Check if there is enough headroom to insert fragment header. */
1341 if ((skb_mac_header(skb
) < skb
->head
+ frag_hdr_sz
) &&
1342 pskb_expand_head(skb
, frag_hdr_sz
, 0, GFP_ATOMIC
))
1345 /* Find the unfragmentable header and shift it left by frag_hdr_sz
1346 * bytes to insert fragment header.
1348 unfrag_ip6hlen
= ip6_find_1stfragopt(skb
, &prevhdr
);
1350 *prevhdr
= NEXTHDR_FRAGMENT
;
1351 unfrag_len
= skb_network_header(skb
) - skb_mac_header(skb
) +
1353 mac_start
= skb_mac_header(skb
);
1354 memmove(mac_start
-frag_hdr_sz
, mac_start
, unfrag_len
);
1356 skb
->mac_header
-= frag_hdr_sz
;
1357 skb
->network_header
-= frag_hdr_sz
;
1359 fptr
= (struct frag_hdr
*)(skb_network_header(skb
) + unfrag_ip6hlen
);
1360 fptr
->nexthdr
= nexthdr
;
1362 ipv6_select_ident(fptr
, (struct rt6_info
*)skb_dst(skb
));
1364 /* Fragment the skb. ipv6 header and the remaining fields of the
1365 * fragment header are updated in ipv6_gso_segment()
1367 segs
= skb_segment(skb
, features
);
1373 static const struct inet6_protocol udpv6_protocol
= {
1374 .handler
= udpv6_rcv
,
1375 .err_handler
= udpv6_err
,
1376 .gso_send_check
= udp6_ufo_send_check
,
1377 .gso_segment
= udp6_ufo_fragment
,
1378 .flags
= INET6_PROTO_NOPOLICY
|INET6_PROTO_FINAL
,
1381 /* ------------------------------------------------------------------------ */
1382 #ifdef CONFIG_PROC_FS
1384 static void udp6_sock_seq_show(struct seq_file
*seq
, struct sock
*sp
, int bucket
)
1386 struct inet_sock
*inet
= inet_sk(sp
);
1387 struct ipv6_pinfo
*np
= inet6_sk(sp
);
1388 const struct in6_addr
*dest
, *src
;
1392 src
= &np
->rcv_saddr
;
1393 destp
= ntohs(inet
->inet_dport
);
1394 srcp
= ntohs(inet
->inet_sport
);
1396 "%5d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1397 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d\n",
1399 src
->s6_addr32
[0], src
->s6_addr32
[1],
1400 src
->s6_addr32
[2], src
->s6_addr32
[3], srcp
,
1401 dest
->s6_addr32
[0], dest
->s6_addr32
[1],
1402 dest
->s6_addr32
[2], dest
->s6_addr32
[3], destp
,
1404 sk_wmem_alloc_get(sp
),
1405 sk_rmem_alloc_get(sp
),
1409 atomic_read(&sp
->sk_refcnt
), sp
,
1410 atomic_read(&sp
->sk_drops
));
1413 int udp6_seq_show(struct seq_file
*seq
, void *v
)
1415 if (v
== SEQ_START_TOKEN
)
1420 "st tx_queue rx_queue tr tm->when retrnsmt"
1421 " uid timeout inode ref pointer drops\n");
1423 udp6_sock_seq_show(seq
, v
, ((struct udp_iter_state
*)seq
->private)->bucket
);
1427 static struct udp_seq_afinfo udp6_seq_afinfo
= {
1430 .udp_table
= &udp_table
,
1432 .owner
= THIS_MODULE
,
1435 .show
= udp6_seq_show
,
1439 int __net_init
udp6_proc_init(struct net
*net
)
1441 return udp_proc_register(net
, &udp6_seq_afinfo
);
1444 void udp6_proc_exit(struct net
*net
) {
1445 udp_proc_unregister(net
, &udp6_seq_afinfo
);
1447 #endif /* CONFIG_PROC_FS */
1449 /* ------------------------------------------------------------------------ */
1451 struct proto udpv6_prot
= {
1453 .owner
= THIS_MODULE
,
1454 .close
= udp_lib_close
,
1455 .connect
= ip6_datagram_connect
,
1456 .disconnect
= udp_disconnect
,
1458 .destroy
= udpv6_destroy_sock
,
1459 .setsockopt
= udpv6_setsockopt
,
1460 .getsockopt
= udpv6_getsockopt
,
1461 .sendmsg
= udpv6_sendmsg
,
1462 .recvmsg
= udpv6_recvmsg
,
1463 .backlog_rcv
= udpv6_queue_rcv_skb
,
1464 .hash
= udp_lib_hash
,
1465 .unhash
= udp_lib_unhash
,
1466 .rehash
= udp_v6_rehash
,
1467 .get_port
= udp_v6_get_port
,
1468 .memory_allocated
= &udp_memory_allocated
,
1469 .sysctl_mem
= sysctl_udp_mem
,
1470 .sysctl_wmem
= &sysctl_udp_wmem_min
,
1471 .sysctl_rmem
= &sysctl_udp_rmem_min
,
1472 .obj_size
= sizeof(struct udp6_sock
),
1473 .slab_flags
= SLAB_DESTROY_BY_RCU
,
1474 .h
.udp_table
= &udp_table
,
1475 #ifdef CONFIG_COMPAT
1476 .compat_setsockopt
= compat_udpv6_setsockopt
,
1477 .compat_getsockopt
= compat_udpv6_getsockopt
,
1479 .clear_sk
= sk_prot_clear_portaddr_nulls
,
1482 static struct inet_protosw udpv6_protosw
= {
1484 .protocol
= IPPROTO_UDP
,
1485 .prot
= &udpv6_prot
,
1486 .ops
= &inet6_dgram_ops
,
1487 .no_check
= UDP_CSUM_DEFAULT
,
1488 .flags
= INET_PROTOSW_PERMANENT
,
1492 int __init
udpv6_init(void)
1496 ret
= inet6_add_protocol(&udpv6_protocol
, IPPROTO_UDP
);
1500 ret
= inet6_register_protosw(&udpv6_protosw
);
1502 goto out_udpv6_protocol
;
1507 inet6_del_protocol(&udpv6_protocol
, IPPROTO_UDP
);
1511 void udpv6_exit(void)
1513 inet6_unregister_protosw(&udpv6_protosw
);
1514 inet6_del_protocol(&udpv6_protocol
, IPPROTO_UDP
);