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>
51 #include <trace/events/skb.h>
54 int ipv6_rcv_saddr_equal(const struct sock
*sk
, const struct sock
*sk2
)
56 const struct in6_addr
*sk_rcv_saddr6
= &inet6_sk(sk
)->rcv_saddr
;
57 const struct in6_addr
*sk2_rcv_saddr6
= inet6_rcv_saddr(sk2
);
58 __be32 sk1_rcv_saddr
= sk_rcv_saddr(sk
);
59 __be32 sk2_rcv_saddr
= sk_rcv_saddr(sk2
);
60 int sk_ipv6only
= ipv6_only_sock(sk
);
61 int sk2_ipv6only
= inet_v6_ipv6only(sk2
);
62 int addr_type
= ipv6_addr_type(sk_rcv_saddr6
);
63 int addr_type2
= sk2_rcv_saddr6
? ipv6_addr_type(sk2_rcv_saddr6
) : IPV6_ADDR_MAPPED
;
65 /* if both are mapped, treat as IPv4 */
66 if (addr_type
== IPV6_ADDR_MAPPED
&& addr_type2
== IPV6_ADDR_MAPPED
)
67 return (!sk2_ipv6only
&&
68 (!sk1_rcv_saddr
|| !sk2_rcv_saddr
||
69 sk1_rcv_saddr
== sk2_rcv_saddr
));
71 if (addr_type2
== IPV6_ADDR_ANY
&&
72 !(sk2_ipv6only
&& addr_type
== IPV6_ADDR_MAPPED
))
75 if (addr_type
== IPV6_ADDR_ANY
&&
76 !(sk_ipv6only
&& addr_type2
== IPV6_ADDR_MAPPED
))
80 ipv6_addr_equal(sk_rcv_saddr6
, sk2_rcv_saddr6
))
86 static unsigned int udp6_portaddr_hash(struct net
*net
,
87 const struct in6_addr
*addr6
,
90 unsigned int hash
, mix
= net_hash_mix(net
);
92 if (ipv6_addr_any(addr6
))
93 hash
= jhash_1word(0, mix
);
94 else if (ipv6_addr_v4mapped(addr6
))
95 hash
= jhash_1word((__force u32
)addr6
->s6_addr32
[3], mix
);
97 hash
= jhash2((__force u32
*)addr6
->s6_addr32
, 4, mix
);
103 int udp_v6_get_port(struct sock
*sk
, unsigned short snum
)
105 unsigned int hash2_nulladdr
=
106 udp6_portaddr_hash(sock_net(sk
), &in6addr_any
, snum
);
107 unsigned int hash2_partial
=
108 udp6_portaddr_hash(sock_net(sk
), &inet6_sk(sk
)->rcv_saddr
, 0);
110 /* precompute partial secondary hash */
111 udp_sk(sk
)->udp_portaddr_hash
= hash2_partial
;
112 return udp_lib_get_port(sk
, snum
, ipv6_rcv_saddr_equal
, hash2_nulladdr
);
115 static void udp_v6_rehash(struct sock
*sk
)
117 u16 new_hash
= udp6_portaddr_hash(sock_net(sk
),
118 &inet6_sk(sk
)->rcv_saddr
,
119 inet_sk(sk
)->inet_num
);
121 udp_lib_rehash(sk
, new_hash
);
124 static inline int compute_score(struct sock
*sk
, struct net
*net
,
126 const struct in6_addr
*saddr
, __be16 sport
,
127 const struct in6_addr
*daddr
, __be16 dport
,
132 if (net_eq(sock_net(sk
), net
) && udp_sk(sk
)->udp_port_hash
== hnum
&&
133 sk
->sk_family
== PF_INET6
) {
134 struct ipv6_pinfo
*np
= inet6_sk(sk
);
135 struct inet_sock
*inet
= inet_sk(sk
);
138 if (inet
->inet_dport
) {
139 if (inet
->inet_dport
!= sport
)
143 if (!ipv6_addr_any(&np
->rcv_saddr
)) {
144 if (!ipv6_addr_equal(&np
->rcv_saddr
, daddr
))
148 if (!ipv6_addr_any(&np
->daddr
)) {
149 if (!ipv6_addr_equal(&np
->daddr
, saddr
))
153 if (sk
->sk_bound_dev_if
) {
154 if (sk
->sk_bound_dev_if
!= dif
)
162 #define SCORE2_MAX (1 + 1 + 1)
163 static inline int compute_score2(struct sock
*sk
, struct net
*net
,
164 const struct in6_addr
*saddr
, __be16 sport
,
165 const struct in6_addr
*daddr
, unsigned short hnum
,
170 if (net_eq(sock_net(sk
), net
) && udp_sk(sk
)->udp_port_hash
== hnum
&&
171 sk
->sk_family
== PF_INET6
) {
172 struct ipv6_pinfo
*np
= inet6_sk(sk
);
173 struct inet_sock
*inet
= inet_sk(sk
);
175 if (!ipv6_addr_equal(&np
->rcv_saddr
, daddr
))
178 if (inet
->inet_dport
) {
179 if (inet
->inet_dport
!= sport
)
183 if (!ipv6_addr_any(&np
->daddr
)) {
184 if (!ipv6_addr_equal(&np
->daddr
, saddr
))
188 if (sk
->sk_bound_dev_if
) {
189 if (sk
->sk_bound_dev_if
!= dif
)
198 /* called with read_rcu_lock() */
199 static struct sock
*udp6_lib_lookup2(struct net
*net
,
200 const struct in6_addr
*saddr
, __be16 sport
,
201 const struct in6_addr
*daddr
, unsigned int hnum
, int dif
,
202 struct udp_hslot
*hslot2
, unsigned int slot2
)
204 struct sock
*sk
, *result
;
205 struct hlist_nulls_node
*node
;
211 udp_portaddr_for_each_entry_rcu(sk
, node
, &hslot2
->head
) {
212 score
= compute_score2(sk
, net
, saddr
, sport
,
214 if (score
> badness
) {
217 if (score
== SCORE2_MAX
)
222 * if the nulls value we got at the end of this lookup is
223 * not the expected one, we must restart lookup.
224 * We probably met an item that was moved to another chain.
226 if (get_nulls_value(node
) != slot2
)
231 if (unlikely(!atomic_inc_not_zero_hint(&result
->sk_refcnt
, 2)))
233 else if (unlikely(compute_score2(result
, net
, saddr
, sport
,
234 daddr
, hnum
, dif
) < badness
)) {
242 struct sock
*__udp6_lib_lookup(struct net
*net
,
243 const struct in6_addr
*saddr
, __be16 sport
,
244 const struct in6_addr
*daddr
, __be16 dport
,
245 int dif
, struct udp_table
*udptable
)
247 struct sock
*sk
, *result
;
248 struct hlist_nulls_node
*node
;
249 unsigned short hnum
= ntohs(dport
);
250 unsigned int hash2
, slot2
, slot
= udp_hashfn(net
, hnum
, udptable
->mask
);
251 struct udp_hslot
*hslot2
, *hslot
= &udptable
->hash
[slot
];
255 if (hslot
->count
> 10) {
256 hash2
= udp6_portaddr_hash(net
, daddr
, 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
,
266 hash2
= udp6_portaddr_hash(net
, &in6addr_any
, hnum
);
267 slot2
= hash2
& udptable
->mask
;
268 hslot2
= &udptable
->hash2
[slot2
];
269 if (hslot
->count
< hslot2
->count
)
272 result
= udp6_lib_lookup2(net
, saddr
, sport
,
273 &in6addr_any
, hnum
, dif
,
282 sk_nulls_for_each_rcu(sk
, node
, &hslot
->head
) {
283 score
= compute_score(sk
, net
, hnum
, saddr
, sport
, daddr
, dport
, dif
);
284 if (score
> badness
) {
290 * if the nulls value we got at the end of this lookup is
291 * not the expected one, we must restart lookup.
292 * We probably met an item that was moved to another chain.
294 if (get_nulls_value(node
) != slot
)
298 if (unlikely(!atomic_inc_not_zero_hint(&result
->sk_refcnt
, 2)))
300 else if (unlikely(compute_score(result
, net
, hnum
, saddr
, sport
,
301 daddr
, dport
, dif
) < badness
)) {
309 EXPORT_SYMBOL_GPL(__udp6_lib_lookup
);
311 static struct sock
*__udp6_lib_lookup_skb(struct sk_buff
*skb
,
312 __be16 sport
, __be16 dport
,
313 struct udp_table
*udptable
)
316 const struct ipv6hdr
*iph
= ipv6_hdr(skb
);
318 if (unlikely(sk
= skb_steal_sock(skb
)))
320 return __udp6_lib_lookup(dev_net(skb_dst(skb
)->dev
), &iph
->saddr
, sport
,
321 &iph
->daddr
, dport
, inet6_iif(skb
),
325 struct sock
*udp6_lib_lookup(struct net
*net
, const struct in6_addr
*saddr
, __be16 sport
,
326 const struct in6_addr
*daddr
, __be16 dport
, int dif
)
328 return __udp6_lib_lookup(net
, saddr
, sport
, daddr
, dport
, dif
, &udp_table
);
330 EXPORT_SYMBOL_GPL(udp6_lib_lookup
);
334 * This should be easy, if there is something there we
335 * return it, otherwise we block.
338 int udpv6_recvmsg(struct kiocb
*iocb
, struct sock
*sk
,
339 struct msghdr
*msg
, size_t len
,
340 int noblock
, int flags
, int *addr_len
)
342 struct ipv6_pinfo
*np
= inet6_sk(sk
);
343 struct inet_sock
*inet
= inet_sk(sk
);
345 unsigned int ulen
, copied
;
348 int is_udplite
= IS_UDPLITE(sk
);
353 *addr_len
= sizeof(struct sockaddr_in6
);
355 if (flags
& MSG_ERRQUEUE
)
356 return ipv6_recv_error(sk
, msg
, len
);
358 if (np
->rxpmtu
&& np
->rxopt
.bits
.rxpmtu
)
359 return ipv6_recv_rxpmtu(sk
, msg
, len
);
362 skb
= __skb_recv_datagram(sk
, flags
| (noblock
? MSG_DONTWAIT
: 0),
363 &peeked
, &off
, &err
);
367 ulen
= skb
->len
- sizeof(struct udphdr
);
371 else if (copied
< ulen
)
372 msg
->msg_flags
|= MSG_TRUNC
;
374 is_udp4
= (skb
->protocol
== htons(ETH_P_IP
));
377 * If checksum is needed at all, try to do it while copying the
378 * data. If the data is truncated, or if we only want a partial
379 * coverage checksum (UDP-Lite), do it before the copy.
382 if (copied
< ulen
|| UDP_SKB_CB(skb
)->partial_cov
) {
383 if (udp_lib_checksum_complete(skb
))
387 if (skb_csum_unnecessary(skb
))
388 err
= skb_copy_datagram_iovec(skb
, sizeof(struct udphdr
),
389 msg
->msg_iov
, copied
);
391 err
= skb_copy_and_csum_datagram_iovec(skb
, sizeof(struct udphdr
), msg
->msg_iov
);
396 trace_kfree_skb(skb
, udpv6_recvmsg
);
398 atomic_inc(&sk
->sk_drops
);
400 UDP_INC_STATS_USER(sock_net(sk
),
404 UDP6_INC_STATS_USER(sock_net(sk
),
412 UDP_INC_STATS_USER(sock_net(sk
),
413 UDP_MIB_INDATAGRAMS
, is_udplite
);
415 UDP6_INC_STATS_USER(sock_net(sk
),
416 UDP_MIB_INDATAGRAMS
, is_udplite
);
419 sock_recv_ts_and_drops(msg
, sk
, skb
);
421 /* Copy the address. */
423 struct sockaddr_in6
*sin6
;
425 sin6
= (struct sockaddr_in6
*) msg
->msg_name
;
426 sin6
->sin6_family
= AF_INET6
;
427 sin6
->sin6_port
= udp_hdr(skb
)->source
;
428 sin6
->sin6_flowinfo
= 0;
429 sin6
->sin6_scope_id
= 0;
432 ipv6_addr_set_v4mapped(ip_hdr(skb
)->saddr
,
435 sin6
->sin6_addr
= ipv6_hdr(skb
)->saddr
;
436 if (ipv6_addr_type(&sin6
->sin6_addr
) & IPV6_ADDR_LINKLOCAL
)
437 sin6
->sin6_scope_id
= IP6CB(skb
)->iif
;
442 if (inet
->cmsg_flags
)
443 ip_cmsg_recv(msg
, skb
);
446 datagram_recv_ctl(sk
, msg
, skb
);
450 if (flags
& MSG_TRUNC
)
454 skb_free_datagram_locked(sk
, skb
);
459 slow
= lock_sock_fast(sk
);
460 if (!skb_kill_datagram(sk
, skb
, flags
)) {
462 UDP_INC_STATS_USER(sock_net(sk
),
463 UDP_MIB_INERRORS
, is_udplite
);
465 UDP6_INC_STATS_USER(sock_net(sk
),
466 UDP_MIB_INERRORS
, is_udplite
);
468 unlock_sock_fast(sk
, slow
);
473 /* starting over for a new packet */
474 msg
->msg_flags
&= ~MSG_TRUNC
;
478 void __udp6_lib_err(struct sk_buff
*skb
, struct inet6_skb_parm
*opt
,
479 u8 type
, u8 code
, int offset
, __be32 info
,
480 struct udp_table
*udptable
)
482 struct ipv6_pinfo
*np
;
483 const struct ipv6hdr
*hdr
= (const struct ipv6hdr
*)skb
->data
;
484 const struct in6_addr
*saddr
= &hdr
->saddr
;
485 const struct in6_addr
*daddr
= &hdr
->daddr
;
486 struct udphdr
*uh
= (struct udphdr
*)(skb
->data
+offset
);
490 sk
= __udp6_lib_lookup(dev_net(skb
->dev
), daddr
, uh
->dest
,
491 saddr
, uh
->source
, inet6_iif(skb
), udptable
);
495 if (type
== ICMPV6_PKT_TOOBIG
)
496 ip6_sk_update_pmtu(skb
, sk
, info
);
497 if (type
== NDISC_REDIRECT
)
498 ip6_sk_redirect(skb
, sk
);
502 if (!icmpv6_err_convert(type
, code
, &err
) && !np
->recverr
)
505 if (sk
->sk_state
!= TCP_ESTABLISHED
&& !np
->recverr
)
509 ipv6_icmp_error(sk
, skb
, err
, uh
->dest
, ntohl(info
), (u8
*)(uh
+1));
512 sk
->sk_error_report(sk
);
517 static int __udpv6_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
521 if (!ipv6_addr_any(&inet6_sk(sk
)->daddr
))
522 sock_rps_save_rxhash(sk
, skb
);
524 rc
= sock_queue_rcv_skb(sk
, skb
);
526 int is_udplite
= IS_UDPLITE(sk
);
528 /* Note that an ENOMEM error is charged twice */
530 UDP6_INC_STATS_BH(sock_net(sk
),
531 UDP_MIB_RCVBUFERRORS
, is_udplite
);
532 UDP6_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
539 static __inline__
void udpv6_err(struct sk_buff
*skb
,
540 struct inet6_skb_parm
*opt
, u8 type
,
541 u8 code
, int offset
, __be32 info
)
543 __udp6_lib_err(skb
, opt
, type
, code
, offset
, info
, &udp_table
);
546 static struct static_key udpv6_encap_needed __read_mostly
;
547 void udpv6_encap_enable(void)
549 if (!static_key_enabled(&udpv6_encap_needed
))
550 static_key_slow_inc(&udpv6_encap_needed
);
552 EXPORT_SYMBOL(udpv6_encap_enable
);
554 int udpv6_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
556 struct udp_sock
*up
= udp_sk(sk
);
558 int is_udplite
= IS_UDPLITE(sk
);
560 if (!xfrm6_policy_check(sk
, XFRM_POLICY_IN
, skb
))
563 if (static_key_false(&udpv6_encap_needed
) && up
->encap_type
) {
564 int (*encap_rcv
)(struct sock
*sk
, struct sk_buff
*skb
);
567 * This is an encapsulation socket so pass the skb to
568 * the socket's udp_encap_rcv() hook. Otherwise, just
569 * fall through and pass this up the UDP socket.
570 * up->encap_rcv() returns the following value:
571 * =0 if skb was successfully passed to the encap
572 * handler or was discarded by it.
573 * >0 if skb should be passed on to UDP.
574 * <0 if skb should be resubmitted as proto -N
577 /* if we're overly short, let UDP handle it */
578 encap_rcv
= ACCESS_ONCE(up
->encap_rcv
);
579 if (skb
->len
> sizeof(struct udphdr
) && encap_rcv
!= NULL
) {
582 ret
= encap_rcv(sk
, skb
);
584 UDP_INC_STATS_BH(sock_net(sk
),
591 /* FALLTHROUGH -- it's a UDP Packet */
595 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
597 if ((is_udplite
& UDPLITE_RECV_CC
) && UDP_SKB_CB(skb
)->partial_cov
) {
599 if (up
->pcrlen
== 0) { /* full coverage was set */
600 LIMIT_NETDEBUG(KERN_WARNING
"UDPLITE6: partial coverage"
601 " %d while full coverage %d requested\n",
602 UDP_SKB_CB(skb
)->cscov
, skb
->len
);
605 if (UDP_SKB_CB(skb
)->cscov
< up
->pcrlen
) {
606 LIMIT_NETDEBUG(KERN_WARNING
"UDPLITE6: coverage %d "
607 "too small, need min %d\n",
608 UDP_SKB_CB(skb
)->cscov
, up
->pcrlen
);
613 if (rcu_access_pointer(sk
->sk_filter
)) {
614 if (udp_lib_checksum_complete(skb
))
618 if (sk_rcvqueues_full(sk
, skb
, sk
->sk_rcvbuf
))
625 if (!sock_owned_by_user(sk
))
626 rc
= __udpv6_queue_rcv_skb(sk
, skb
);
627 else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
635 UDP6_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
636 atomic_inc(&sk
->sk_drops
);
641 static struct sock
*udp_v6_mcast_next(struct net
*net
, struct sock
*sk
,
642 __be16 loc_port
, const struct in6_addr
*loc_addr
,
643 __be16 rmt_port
, const struct in6_addr
*rmt_addr
,
646 struct hlist_nulls_node
*node
;
648 unsigned short num
= ntohs(loc_port
);
650 sk_nulls_for_each_from(s
, node
) {
651 struct inet_sock
*inet
= inet_sk(s
);
653 if (!net_eq(sock_net(s
), net
))
656 if (udp_sk(s
)->udp_port_hash
== num
&&
657 s
->sk_family
== PF_INET6
) {
658 struct ipv6_pinfo
*np
= inet6_sk(s
);
659 if (inet
->inet_dport
) {
660 if (inet
->inet_dport
!= rmt_port
)
663 if (!ipv6_addr_any(&np
->daddr
) &&
664 !ipv6_addr_equal(&np
->daddr
, rmt_addr
))
667 if (s
->sk_bound_dev_if
&& s
->sk_bound_dev_if
!= dif
)
670 if (!ipv6_addr_any(&np
->rcv_saddr
)) {
671 if (!ipv6_addr_equal(&np
->rcv_saddr
, loc_addr
))
674 if (!inet6_mc_check(s
, loc_addr
, rmt_addr
))
682 static void flush_stack(struct sock
**stack
, unsigned int count
,
683 struct sk_buff
*skb
, unsigned int final
)
685 struct sk_buff
*skb1
= NULL
;
689 for (i
= 0; i
< count
; i
++) {
691 if (likely(skb1
== NULL
))
692 skb1
= (i
== final
) ? skb
: skb_clone(skb
, GFP_ATOMIC
);
694 atomic_inc(&sk
->sk_drops
);
695 UDP6_INC_STATS_BH(sock_net(sk
), UDP_MIB_RCVBUFERRORS
,
697 UDP6_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
,
701 if (skb1
&& udpv6_queue_rcv_skb(sk
, skb1
) <= 0)
708 * Note: called only from the BH handler context,
709 * so we don't need to lock the hashes.
711 static int __udp6_lib_mcast_deliver(struct net
*net
, struct sk_buff
*skb
,
712 const struct in6_addr
*saddr
, const struct in6_addr
*daddr
,
713 struct udp_table
*udptable
)
715 struct sock
*sk
, *stack
[256 / sizeof(struct sock
*)];
716 const struct udphdr
*uh
= udp_hdr(skb
);
717 struct udp_hslot
*hslot
= udp_hashslot(udptable
, net
, ntohs(uh
->dest
));
719 unsigned int i
, count
= 0;
721 spin_lock(&hslot
->lock
);
722 sk
= sk_nulls_head(&hslot
->head
);
723 dif
= inet6_iif(skb
);
724 sk
= udp_v6_mcast_next(net
, sk
, uh
->dest
, daddr
, uh
->source
, saddr
, dif
);
727 sk
= udp_v6_mcast_next(net
, sk_nulls_next(sk
), uh
->dest
, daddr
,
728 uh
->source
, saddr
, dif
);
729 if (unlikely(count
== ARRAY_SIZE(stack
))) {
732 flush_stack(stack
, count
, skb
, ~0);
737 * before releasing the lock, we must take reference on sockets
739 for (i
= 0; i
< count
; i
++)
742 spin_unlock(&hslot
->lock
);
745 flush_stack(stack
, count
, skb
, count
- 1);
747 for (i
= 0; i
< count
; i
++)
755 int __udp6_lib_rcv(struct sk_buff
*skb
, struct udp_table
*udptable
,
758 struct net
*net
= dev_net(skb
->dev
);
761 const struct in6_addr
*saddr
, *daddr
;
764 if (!pskb_may_pull(skb
, sizeof(struct udphdr
)))
767 saddr
= &ipv6_hdr(skb
)->saddr
;
768 daddr
= &ipv6_hdr(skb
)->daddr
;
771 ulen
= ntohs(uh
->len
);
775 if (proto
== IPPROTO_UDP
) {
776 /* UDP validates ulen. */
778 /* Check for jumbo payload */
782 if (ulen
< sizeof(*uh
))
785 if (ulen
< skb
->len
) {
786 if (pskb_trim_rcsum(skb
, ulen
))
788 saddr
= &ipv6_hdr(skb
)->saddr
;
789 daddr
= &ipv6_hdr(skb
)->daddr
;
794 if (udp6_csum_init(skb
, uh
, proto
))
798 * Multicast receive code
800 if (ipv6_addr_is_multicast(daddr
))
801 return __udp6_lib_mcast_deliver(net
, skb
,
802 saddr
, daddr
, udptable
);
807 * check socket cache ... must talk to Alan about his plans
808 * for sock caches... i'll skip this for now.
810 sk
= __udp6_lib_lookup_skb(skb
, uh
->source
, uh
->dest
, udptable
);
812 int ret
= udpv6_queue_rcv_skb(sk
, skb
);
815 /* a return value > 0 means to resubmit the input, but
816 * it wants the return to be -protocol, or 0
824 if (!xfrm6_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
827 if (udp_lib_checksum_complete(skb
))
830 UDP6_INC_STATS_BH(net
, UDP_MIB_NOPORTS
, proto
== IPPROTO_UDPLITE
);
831 icmpv6_send(skb
, ICMPV6_DEST_UNREACH
, ICMPV6_PORT_UNREACH
, 0);
837 LIMIT_NETDEBUG(KERN_DEBUG
"UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
838 proto
== IPPROTO_UDPLITE
? "-Lite" : "",
847 UDP6_INC_STATS_BH(net
, UDP_MIB_INERRORS
, proto
== IPPROTO_UDPLITE
);
852 static __inline__
int udpv6_rcv(struct sk_buff
*skb
)
854 return __udp6_lib_rcv(skb
, &udp_table
, IPPROTO_UDP
);
858 * Throw away all pending data and cancel the corking. Socket is locked.
860 static void udp_v6_flush_pending_frames(struct sock
*sk
)
862 struct udp_sock
*up
= udp_sk(sk
);
864 if (up
->pending
== AF_INET
)
865 udp_flush_pending_frames(sk
);
866 else if (up
->pending
) {
869 ip6_flush_pending_frames(sk
);
874 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
875 * @sk: socket we are sending on
876 * @skb: sk_buff containing the filled-in UDP header
877 * (checksum field must be zeroed out)
879 static void udp6_hwcsum_outgoing(struct sock
*sk
, struct sk_buff
*skb
,
880 const struct in6_addr
*saddr
,
881 const struct in6_addr
*daddr
, int len
)
884 struct udphdr
*uh
= udp_hdr(skb
);
887 if (skb_queue_len(&sk
->sk_write_queue
) == 1) {
888 /* Only one fragment on the socket. */
889 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
890 skb
->csum_offset
= offsetof(struct udphdr
, check
);
891 uh
->check
= ~csum_ipv6_magic(saddr
, daddr
, len
, IPPROTO_UDP
, 0);
894 * HW-checksum won't work as there are two or more
895 * fragments on the socket so that all csums of sk_buffs
898 offset
= skb_transport_offset(skb
);
899 skb
->csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
901 skb
->ip_summed
= CHECKSUM_NONE
;
903 skb_queue_walk(&sk
->sk_write_queue
, skb
) {
904 csum
= csum_add(csum
, skb
->csum
);
907 uh
->check
= csum_ipv6_magic(saddr
, daddr
, len
, IPPROTO_UDP
,
910 uh
->check
= CSUM_MANGLED_0
;
918 static int udp_v6_push_pending_frames(struct sock
*sk
)
922 struct udp_sock
*up
= udp_sk(sk
);
923 struct inet_sock
*inet
= inet_sk(sk
);
924 struct flowi6
*fl6
= &inet
->cork
.fl
.u
.ip6
;
926 int is_udplite
= IS_UDPLITE(sk
);
929 /* Grab the skbuff where UDP header space exists. */
930 if ((skb
= skb_peek(&sk
->sk_write_queue
)) == NULL
)
934 * Create a UDP header
937 uh
->source
= fl6
->fl6_sport
;
938 uh
->dest
= fl6
->fl6_dport
;
939 uh
->len
= htons(up
->len
);
943 csum
= udplite_csum_outgoing(sk
, skb
);
944 else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) { /* UDP hardware csum */
945 udp6_hwcsum_outgoing(sk
, skb
, &fl6
->saddr
, &fl6
->daddr
,
949 csum
= udp_csum_outgoing(sk
, skb
);
951 /* add protocol-dependent pseudo-header */
952 uh
->check
= csum_ipv6_magic(&fl6
->saddr
, &fl6
->daddr
,
953 up
->len
, fl6
->flowi6_proto
, csum
);
955 uh
->check
= CSUM_MANGLED_0
;
958 err
= ip6_push_pending_frames(sk
);
960 if (err
== -ENOBUFS
&& !inet6_sk(sk
)->recverr
) {
961 UDP6_INC_STATS_USER(sock_net(sk
),
962 UDP_MIB_SNDBUFERRORS
, is_udplite
);
966 UDP6_INC_STATS_USER(sock_net(sk
),
967 UDP_MIB_OUTDATAGRAMS
, is_udplite
);
974 int udpv6_sendmsg(struct kiocb
*iocb
, struct sock
*sk
,
975 struct msghdr
*msg
, size_t len
)
977 struct ipv6_txoptions opt_space
;
978 struct udp_sock
*up
= udp_sk(sk
);
979 struct inet_sock
*inet
= inet_sk(sk
);
980 struct ipv6_pinfo
*np
= inet6_sk(sk
);
981 struct sockaddr_in6
*sin6
= (struct sockaddr_in6
*) msg
->msg_name
;
982 struct in6_addr
*daddr
, *final_p
, final
;
983 struct ipv6_txoptions
*opt
= NULL
;
984 struct ip6_flowlabel
*flowlabel
= NULL
;
986 struct dst_entry
*dst
;
987 int addr_len
= msg
->msg_namelen
;
992 int corkreq
= up
->corkflag
|| msg
->msg_flags
&MSG_MORE
;
995 int is_udplite
= IS_UDPLITE(sk
);
996 int (*getfrag
)(void *, char *, int, int, int, struct sk_buff
*);
998 /* destination address check */
1000 if (addr_len
< offsetof(struct sockaddr
, sa_data
))
1003 switch (sin6
->sin6_family
) {
1005 if (addr_len
< SIN6_LEN_RFC2133
)
1007 daddr
= &sin6
->sin6_addr
;
1010 goto do_udp_sendmsg
;
1012 msg
->msg_name
= sin6
= NULL
;
1013 msg
->msg_namelen
= addr_len
= 0;
1019 } else if (!up
->pending
) {
1020 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1021 return -EDESTADDRREQ
;
1027 if (ipv6_addr_v4mapped(daddr
)) {
1028 struct sockaddr_in sin
;
1029 sin
.sin_family
= AF_INET
;
1030 sin
.sin_port
= sin6
? sin6
->sin6_port
: inet
->inet_dport
;
1031 sin
.sin_addr
.s_addr
= daddr
->s6_addr32
[3];
1032 msg
->msg_name
= &sin
;
1033 msg
->msg_namelen
= sizeof(sin
);
1035 if (__ipv6_only_sock(sk
))
1036 return -ENETUNREACH
;
1037 return udp_sendmsg(iocb
, sk
, msg
, len
);
1041 if (up
->pending
== AF_INET
)
1042 return udp_sendmsg(iocb
, sk
, msg
, len
);
1044 /* Rough check on arithmetic overflow,
1045 better check is made in ip6_append_data().
1047 if (len
> INT_MAX
- sizeof(struct udphdr
))
1052 * There are pending frames.
1053 * The socket lock must be held while it's corked.
1056 if (likely(up
->pending
)) {
1057 if (unlikely(up
->pending
!= AF_INET6
)) {
1059 return -EAFNOSUPPORT
;
1062 goto do_append_data
;
1066 ulen
+= sizeof(struct udphdr
);
1068 memset(&fl6
, 0, sizeof(fl6
));
1071 if (sin6
->sin6_port
== 0)
1074 fl6
.fl6_dport
= sin6
->sin6_port
;
1075 daddr
= &sin6
->sin6_addr
;
1078 fl6
.flowlabel
= sin6
->sin6_flowinfo
&IPV6_FLOWINFO_MASK
;
1079 if (fl6
.flowlabel
&IPV6_FLOWLABEL_MASK
) {
1080 flowlabel
= fl6_sock_lookup(sk
, fl6
.flowlabel
);
1081 if (flowlabel
== NULL
)
1083 daddr
= &flowlabel
->dst
;
1088 * Otherwise it will be difficult to maintain
1091 if (sk
->sk_state
== TCP_ESTABLISHED
&&
1092 ipv6_addr_equal(daddr
, &np
->daddr
))
1095 if (addr_len
>= sizeof(struct sockaddr_in6
) &&
1096 sin6
->sin6_scope_id
&&
1097 ipv6_addr_type(daddr
)&IPV6_ADDR_LINKLOCAL
)
1098 fl6
.flowi6_oif
= sin6
->sin6_scope_id
;
1100 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1101 return -EDESTADDRREQ
;
1103 fl6
.fl6_dport
= inet
->inet_dport
;
1105 fl6
.flowlabel
= np
->flow_label
;
1109 if (!fl6
.flowi6_oif
)
1110 fl6
.flowi6_oif
= sk
->sk_bound_dev_if
;
1112 if (!fl6
.flowi6_oif
)
1113 fl6
.flowi6_oif
= np
->sticky_pktinfo
.ipi6_ifindex
;
1115 fl6
.flowi6_mark
= sk
->sk_mark
;
1117 if (msg
->msg_controllen
) {
1119 memset(opt
, 0, sizeof(struct ipv6_txoptions
));
1120 opt
->tot_len
= sizeof(*opt
);
1122 err
= datagram_send_ctl(sock_net(sk
), sk
, msg
, &fl6
, opt
,
1123 &hlimit
, &tclass
, &dontfrag
);
1125 fl6_sock_release(flowlabel
);
1128 if ((fl6
.flowlabel
&IPV6_FLOWLABEL_MASK
) && !flowlabel
) {
1129 flowlabel
= fl6_sock_lookup(sk
, fl6
.flowlabel
);
1130 if (flowlabel
== NULL
)
1133 if (!(opt
->opt_nflen
|opt
->opt_flen
))
1140 opt
= fl6_merge_options(&opt_space
, flowlabel
, opt
);
1141 opt
= ipv6_fixup_options(&opt_space
, opt
);
1143 fl6
.flowi6_proto
= sk
->sk_protocol
;
1144 if (!ipv6_addr_any(daddr
))
1147 fl6
.daddr
.s6_addr
[15] = 0x1; /* :: means loopback (BSD'ism) */
1148 if (ipv6_addr_any(&fl6
.saddr
) && !ipv6_addr_any(&np
->saddr
))
1149 fl6
.saddr
= np
->saddr
;
1150 fl6
.fl6_sport
= inet
->inet_sport
;
1152 final_p
= fl6_update_dst(&fl6
, opt
, &final
);
1156 if (!fl6
.flowi6_oif
&& ipv6_addr_is_multicast(&fl6
.daddr
)) {
1157 fl6
.flowi6_oif
= np
->mcast_oif
;
1159 } else if (!fl6
.flowi6_oif
)
1160 fl6
.flowi6_oif
= np
->ucast_oif
;
1162 security_sk_classify_flow(sk
, flowi6_to_flowi(&fl6
));
1164 dst
= ip6_sk_dst_lookup_flow(sk
, &fl6
, final_p
, true);
1172 if (ipv6_addr_is_multicast(&fl6
.daddr
))
1173 hlimit
= np
->mcast_hops
;
1175 hlimit
= np
->hop_limit
;
1177 hlimit
= ip6_dst_hoplimit(dst
);
1181 tclass
= np
->tclass
;
1184 dontfrag
= np
->dontfrag
;
1186 if (msg
->msg_flags
&MSG_CONFIRM
)
1191 if (unlikely(up
->pending
)) {
1192 /* The socket is already corked while preparing it. */
1193 /* ... which is an evident application bug. --ANK */
1196 LIMIT_NETDEBUG(KERN_DEBUG
"udp cork app bug 2\n");
1201 up
->pending
= AF_INET6
;
1205 getfrag
= is_udplite
? udplite_getfrag
: ip_generic_getfrag
;
1206 err
= ip6_append_data(sk
, getfrag
, msg
->msg_iov
, ulen
,
1207 sizeof(struct udphdr
), hlimit
, tclass
, opt
, &fl6
,
1208 (struct rt6_info
*)dst
,
1209 corkreq
? msg
->msg_flags
|MSG_MORE
: msg
->msg_flags
, dontfrag
);
1211 udp_v6_flush_pending_frames(sk
);
1213 err
= udp_v6_push_pending_frames(sk
);
1214 else if (unlikely(skb_queue_empty(&sk
->sk_write_queue
)))
1219 ip6_dst_store(sk
, dst
,
1220 ipv6_addr_equal(&fl6
.daddr
, &np
->daddr
) ?
1222 #ifdef CONFIG_IPV6_SUBTREES
1223 ipv6_addr_equal(&fl6
.saddr
, &np
->saddr
) ?
1234 err
= np
->recverr
? net_xmit_errno(err
) : 0;
1238 fl6_sock_release(flowlabel
);
1242 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1243 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1244 * we don't have a good statistic (IpOutDiscards but it can be too many
1245 * things). We could add another new stat but at least for now that
1246 * seems like overkill.
1248 if (err
== -ENOBUFS
|| test_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
)) {
1249 UDP6_INC_STATS_USER(sock_net(sk
),
1250 UDP_MIB_SNDBUFERRORS
, is_udplite
);
1256 if (!(msg
->msg_flags
&MSG_PROBE
) || len
)
1257 goto back_from_confirm
;
1262 void udpv6_destroy_sock(struct sock
*sk
)
1265 udp_v6_flush_pending_frames(sk
);
1268 inet6_destroy_sock(sk
);
1272 * Socket option code for UDP
1274 int udpv6_setsockopt(struct sock
*sk
, int level
, int optname
,
1275 char __user
*optval
, unsigned int optlen
)
1277 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1278 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1279 udp_v6_push_pending_frames
);
1280 return ipv6_setsockopt(sk
, level
, optname
, optval
, optlen
);
1283 #ifdef CONFIG_COMPAT
1284 int compat_udpv6_setsockopt(struct sock
*sk
, int level
, int optname
,
1285 char __user
*optval
, unsigned int optlen
)
1287 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1288 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1289 udp_v6_push_pending_frames
);
1290 return compat_ipv6_setsockopt(sk
, level
, optname
, optval
, optlen
);
1294 int udpv6_getsockopt(struct sock
*sk
, int level
, int optname
,
1295 char __user
*optval
, int __user
*optlen
)
1297 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1298 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1299 return ipv6_getsockopt(sk
, level
, optname
, optval
, optlen
);
1302 #ifdef CONFIG_COMPAT
1303 int compat_udpv6_getsockopt(struct sock
*sk
, int level
, int optname
,
1304 char __user
*optval
, int __user
*optlen
)
1306 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1307 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1308 return compat_ipv6_getsockopt(sk
, level
, optname
, optval
, optlen
);
1312 static const struct inet6_protocol udpv6_protocol
= {
1313 .handler
= udpv6_rcv
,
1314 .err_handler
= udpv6_err
,
1315 .flags
= INET6_PROTO_NOPOLICY
|INET6_PROTO_FINAL
,
1318 /* ------------------------------------------------------------------------ */
1319 #ifdef CONFIG_PROC_FS
1321 static void udp6_sock_seq_show(struct seq_file
*seq
, struct sock
*sp
, int bucket
)
1323 struct inet_sock
*inet
= inet_sk(sp
);
1324 struct ipv6_pinfo
*np
= inet6_sk(sp
);
1325 const struct in6_addr
*dest
, *src
;
1329 src
= &np
->rcv_saddr
;
1330 destp
= ntohs(inet
->inet_dport
);
1331 srcp
= ntohs(inet
->inet_sport
);
1333 "%5d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1334 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d\n",
1336 src
->s6_addr32
[0], src
->s6_addr32
[1],
1337 src
->s6_addr32
[2], src
->s6_addr32
[3], srcp
,
1338 dest
->s6_addr32
[0], dest
->s6_addr32
[1],
1339 dest
->s6_addr32
[2], dest
->s6_addr32
[3], destp
,
1341 sk_wmem_alloc_get(sp
),
1342 sk_rmem_alloc_get(sp
),
1344 from_kuid_munged(seq_user_ns(seq
), sock_i_uid(sp
)),
1347 atomic_read(&sp
->sk_refcnt
), sp
,
1348 atomic_read(&sp
->sk_drops
));
1351 int udp6_seq_show(struct seq_file
*seq
, void *v
)
1353 if (v
== SEQ_START_TOKEN
)
1358 "st tx_queue rx_queue tr tm->when retrnsmt"
1359 " uid timeout inode ref pointer drops\n");
1361 udp6_sock_seq_show(seq
, v
, ((struct udp_iter_state
*)seq
->private)->bucket
);
1365 static const struct file_operations udp6_afinfo_seq_fops
= {
1366 .owner
= THIS_MODULE
,
1367 .open
= udp_seq_open
,
1369 .llseek
= seq_lseek
,
1370 .release
= seq_release_net
1373 static struct udp_seq_afinfo udp6_seq_afinfo
= {
1376 .udp_table
= &udp_table
,
1377 .seq_fops
= &udp6_afinfo_seq_fops
,
1379 .show
= udp6_seq_show
,
1383 int __net_init
udp6_proc_init(struct net
*net
)
1385 return udp_proc_register(net
, &udp6_seq_afinfo
);
1388 void udp6_proc_exit(struct net
*net
) {
1389 udp_proc_unregister(net
, &udp6_seq_afinfo
);
1391 #endif /* CONFIG_PROC_FS */
1393 /* ------------------------------------------------------------------------ */
1395 struct proto udpv6_prot
= {
1397 .owner
= THIS_MODULE
,
1398 .close
= udp_lib_close
,
1399 .connect
= ip6_datagram_connect
,
1400 .disconnect
= udp_disconnect
,
1402 .destroy
= udpv6_destroy_sock
,
1403 .setsockopt
= udpv6_setsockopt
,
1404 .getsockopt
= udpv6_getsockopt
,
1405 .sendmsg
= udpv6_sendmsg
,
1406 .recvmsg
= udpv6_recvmsg
,
1407 .backlog_rcv
= __udpv6_queue_rcv_skb
,
1408 .hash
= udp_lib_hash
,
1409 .unhash
= udp_lib_unhash
,
1410 .rehash
= udp_v6_rehash
,
1411 .get_port
= udp_v6_get_port
,
1412 .memory_allocated
= &udp_memory_allocated
,
1413 .sysctl_mem
= sysctl_udp_mem
,
1414 .sysctl_wmem
= &sysctl_udp_wmem_min
,
1415 .sysctl_rmem
= &sysctl_udp_rmem_min
,
1416 .obj_size
= sizeof(struct udp6_sock
),
1417 .slab_flags
= SLAB_DESTROY_BY_RCU
,
1418 .h
.udp_table
= &udp_table
,
1419 #ifdef CONFIG_COMPAT
1420 .compat_setsockopt
= compat_udpv6_setsockopt
,
1421 .compat_getsockopt
= compat_udpv6_getsockopt
,
1423 .clear_sk
= sk_prot_clear_portaddr_nulls
,
1426 static struct inet_protosw udpv6_protosw
= {
1428 .protocol
= IPPROTO_UDP
,
1429 .prot
= &udpv6_prot
,
1430 .ops
= &inet6_dgram_ops
,
1431 .no_check
= UDP_CSUM_DEFAULT
,
1432 .flags
= INET_PROTOSW_PERMANENT
,
1436 int __init
udpv6_init(void)
1440 ret
= inet6_add_protocol(&udpv6_protocol
, IPPROTO_UDP
);
1444 ret
= inet6_register_protosw(&udpv6_protosw
);
1446 goto out_udpv6_protocol
;
1451 inet6_del_protocol(&udpv6_protocol
, IPPROTO_UDP
);
1455 void udpv6_exit(void)
1457 inet6_unregister_protosw(&udpv6_protosw
);
1458 inet6_del_protocol(&udpv6_protocol
, IPPROTO_UDP
);