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>
48 #include <net/inet6_hashtables.h>
49 #include <net/busy_poll.h>
51 #include <linux/proc_fs.h>
52 #include <linux/seq_file.h>
53 #include <trace/events/skb.h>
56 static unsigned int udp6_ehashfn(struct net
*net
,
57 const struct in6_addr
*laddr
,
59 const struct in6_addr
*faddr
,
62 static u32 udp6_ehash_secret __read_mostly
;
63 static u32 udp_ipv6_hash_secret __read_mostly
;
67 net_get_random_once(&udp6_ehash_secret
,
68 sizeof(udp6_ehash_secret
));
69 net_get_random_once(&udp_ipv6_hash_secret
,
70 sizeof(udp_ipv6_hash_secret
));
72 lhash
= (__force u32
)laddr
->s6_addr32
[3];
73 fhash
= __ipv6_addr_jhash(faddr
, udp_ipv6_hash_secret
);
75 return __inet6_ehashfn(lhash
, lport
, fhash
, fport
,
76 udp_ipv6_hash_secret
+ net_hash_mix(net
));
79 int ipv6_rcv_saddr_equal(const struct sock
*sk
, const struct sock
*sk2
)
81 const struct in6_addr
*sk2_rcv_saddr6
= inet6_rcv_saddr(sk2
);
82 int sk_ipv6only
= ipv6_only_sock(sk
);
83 int sk2_ipv6only
= inet_v6_ipv6only(sk2
);
84 int addr_type
= ipv6_addr_type(&sk
->sk_v6_rcv_saddr
);
85 int addr_type2
= sk2_rcv_saddr6
? ipv6_addr_type(sk2_rcv_saddr6
) : IPV6_ADDR_MAPPED
;
87 /* if both are mapped, treat as IPv4 */
88 if (addr_type
== IPV6_ADDR_MAPPED
&& addr_type2
== IPV6_ADDR_MAPPED
)
89 return (!sk2_ipv6only
&&
90 (!sk
->sk_rcv_saddr
|| !sk2
->sk_rcv_saddr
||
91 sk
->sk_rcv_saddr
== sk2
->sk_rcv_saddr
));
93 if (addr_type2
== IPV6_ADDR_ANY
&&
94 !(sk2_ipv6only
&& addr_type
== IPV6_ADDR_MAPPED
))
97 if (addr_type
== IPV6_ADDR_ANY
&&
98 !(sk_ipv6only
&& addr_type2
== IPV6_ADDR_MAPPED
))
101 if (sk2_rcv_saddr6
&&
102 ipv6_addr_equal(&sk
->sk_v6_rcv_saddr
, sk2_rcv_saddr6
))
108 static unsigned int udp6_portaddr_hash(struct net
*net
,
109 const struct in6_addr
*addr6
,
112 unsigned int hash
, mix
= net_hash_mix(net
);
114 if (ipv6_addr_any(addr6
))
115 hash
= jhash_1word(0, mix
);
116 else if (ipv6_addr_v4mapped(addr6
))
117 hash
= jhash_1word((__force u32
)addr6
->s6_addr32
[3], mix
);
119 hash
= jhash2((__force u32
*)addr6
->s6_addr32
, 4, mix
);
125 int udp_v6_get_port(struct sock
*sk
, unsigned short snum
)
127 unsigned int hash2_nulladdr
=
128 udp6_portaddr_hash(sock_net(sk
), &in6addr_any
, snum
);
129 unsigned int hash2_partial
=
130 udp6_portaddr_hash(sock_net(sk
), &sk
->sk_v6_rcv_saddr
, 0);
132 /* precompute partial secondary hash */
133 udp_sk(sk
)->udp_portaddr_hash
= hash2_partial
;
134 return udp_lib_get_port(sk
, snum
, ipv6_rcv_saddr_equal
, hash2_nulladdr
);
137 static void udp_v6_rehash(struct sock
*sk
)
139 u16 new_hash
= udp6_portaddr_hash(sock_net(sk
),
140 &sk
->sk_v6_rcv_saddr
,
141 inet_sk(sk
)->inet_num
);
143 udp_lib_rehash(sk
, new_hash
);
146 static inline int compute_score(struct sock
*sk
, struct net
*net
,
148 const struct in6_addr
*saddr
, __be16 sport
,
149 const struct in6_addr
*daddr
, __be16 dport
,
154 if (net_eq(sock_net(sk
), net
) && udp_sk(sk
)->udp_port_hash
== hnum
&&
155 sk
->sk_family
== PF_INET6
) {
156 struct inet_sock
*inet
= inet_sk(sk
);
159 if (inet
->inet_dport
) {
160 if (inet
->inet_dport
!= sport
)
164 if (!ipv6_addr_any(&sk
->sk_v6_rcv_saddr
)) {
165 if (!ipv6_addr_equal(&sk
->sk_v6_rcv_saddr
, daddr
))
169 if (!ipv6_addr_any(&sk
->sk_v6_daddr
)) {
170 if (!ipv6_addr_equal(&sk
->sk_v6_daddr
, saddr
))
174 if (sk
->sk_bound_dev_if
) {
175 if (sk
->sk_bound_dev_if
!= dif
)
183 #define SCORE2_MAX (1 + 1 + 1)
184 static inline int compute_score2(struct sock
*sk
, struct net
*net
,
185 const struct in6_addr
*saddr
, __be16 sport
,
186 const struct in6_addr
*daddr
, unsigned short hnum
,
191 if (net_eq(sock_net(sk
), net
) && udp_sk(sk
)->udp_port_hash
== hnum
&&
192 sk
->sk_family
== PF_INET6
) {
193 struct inet_sock
*inet
= inet_sk(sk
);
195 if (!ipv6_addr_equal(&sk
->sk_v6_rcv_saddr
, daddr
))
198 if (inet
->inet_dport
) {
199 if (inet
->inet_dport
!= sport
)
203 if (!ipv6_addr_any(&sk
->sk_v6_daddr
)) {
204 if (!ipv6_addr_equal(&sk
->sk_v6_daddr
, saddr
))
208 if (sk
->sk_bound_dev_if
) {
209 if (sk
->sk_bound_dev_if
!= dif
)
218 /* called with read_rcu_lock() */
219 static struct sock
*udp6_lib_lookup2(struct net
*net
,
220 const struct in6_addr
*saddr
, __be16 sport
,
221 const struct in6_addr
*daddr
, unsigned int hnum
, int dif
,
222 struct udp_hslot
*hslot2
, unsigned int slot2
)
224 struct sock
*sk
, *result
;
225 struct hlist_nulls_node
*node
;
226 int score
, badness
, matches
= 0, reuseport
= 0;
232 udp_portaddr_for_each_entry_rcu(sk
, node
, &hslot2
->head
) {
233 score
= compute_score2(sk
, net
, saddr
, sport
,
235 if (score
> badness
) {
238 reuseport
= sk
->sk_reuseport
;
240 hash
= udp6_ehashfn(net
, daddr
, hnum
,
243 } else if (score
== SCORE2_MAX
)
245 } else if (score
== badness
&& reuseport
) {
247 if (((u64
)hash
* matches
) >> 32 == 0)
249 hash
= next_pseudo_random32(hash
);
253 * if the nulls value we got at the end of this lookup is
254 * not the expected one, we must restart lookup.
255 * We probably met an item that was moved to another chain.
257 if (get_nulls_value(node
) != slot2
)
262 if (unlikely(!atomic_inc_not_zero_hint(&result
->sk_refcnt
, 2)))
264 else if (unlikely(compute_score2(result
, net
, saddr
, sport
,
265 daddr
, hnum
, dif
) < badness
)) {
273 struct sock
*__udp6_lib_lookup(struct net
*net
,
274 const struct in6_addr
*saddr
, __be16 sport
,
275 const struct in6_addr
*daddr
, __be16 dport
,
276 int dif
, struct udp_table
*udptable
)
278 struct sock
*sk
, *result
;
279 struct hlist_nulls_node
*node
;
280 unsigned short hnum
= ntohs(dport
);
281 unsigned int hash2
, slot2
, slot
= udp_hashfn(net
, hnum
, udptable
->mask
);
282 struct udp_hslot
*hslot2
, *hslot
= &udptable
->hash
[slot
];
283 int score
, badness
, matches
= 0, reuseport
= 0;
287 if (hslot
->count
> 10) {
288 hash2
= udp6_portaddr_hash(net
, daddr
, hnum
);
289 slot2
= hash2
& udptable
->mask
;
290 hslot2
= &udptable
->hash2
[slot2
];
291 if (hslot
->count
< hslot2
->count
)
294 result
= udp6_lib_lookup2(net
, saddr
, sport
,
298 hash2
= udp6_portaddr_hash(net
, &in6addr_any
, hnum
);
299 slot2
= hash2
& udptable
->mask
;
300 hslot2
= &udptable
->hash2
[slot2
];
301 if (hslot
->count
< hslot2
->count
)
304 result
= udp6_lib_lookup2(net
, saddr
, sport
,
305 &in6addr_any
, hnum
, dif
,
314 sk_nulls_for_each_rcu(sk
, node
, &hslot
->head
) {
315 score
= compute_score(sk
, net
, hnum
, saddr
, sport
, daddr
, dport
, dif
);
316 if (score
> badness
) {
319 reuseport
= sk
->sk_reuseport
;
321 hash
= udp6_ehashfn(net
, daddr
, hnum
,
325 } else if (score
== badness
&& reuseport
) {
327 if (((u64
)hash
* matches
) >> 32 == 0)
329 hash
= next_pseudo_random32(hash
);
333 * if the nulls value we got at the end of this lookup is
334 * not the expected one, we must restart lookup.
335 * We probably met an item that was moved to another chain.
337 if (get_nulls_value(node
) != slot
)
341 if (unlikely(!atomic_inc_not_zero_hint(&result
->sk_refcnt
, 2)))
343 else if (unlikely(compute_score(result
, net
, hnum
, saddr
, sport
,
344 daddr
, dport
, dif
) < badness
)) {
352 EXPORT_SYMBOL_GPL(__udp6_lib_lookup
);
354 static struct sock
*__udp6_lib_lookup_skb(struct sk_buff
*skb
,
355 __be16 sport
, __be16 dport
,
356 struct udp_table
*udptable
)
359 const struct ipv6hdr
*iph
= ipv6_hdr(skb
);
361 if (unlikely(sk
= skb_steal_sock(skb
)))
363 return __udp6_lib_lookup(dev_net(skb_dst(skb
)->dev
), &iph
->saddr
, sport
,
364 &iph
->daddr
, dport
, inet6_iif(skb
),
368 struct sock
*udp6_lib_lookup(struct net
*net
, const struct in6_addr
*saddr
, __be16 sport
,
369 const struct in6_addr
*daddr
, __be16 dport
, int dif
)
371 return __udp6_lib_lookup(net
, saddr
, sport
, daddr
, dport
, dif
, &udp_table
);
373 EXPORT_SYMBOL_GPL(udp6_lib_lookup
);
377 * This should be easy, if there is something there we
378 * return it, otherwise we block.
381 int udpv6_recvmsg(struct kiocb
*iocb
, struct sock
*sk
,
382 struct msghdr
*msg
, size_t len
,
383 int noblock
, int flags
, int *addr_len
)
385 struct ipv6_pinfo
*np
= inet6_sk(sk
);
386 struct inet_sock
*inet
= inet_sk(sk
);
388 unsigned int ulen
, copied
;
391 int is_udplite
= IS_UDPLITE(sk
);
396 *addr_len
= sizeof(struct sockaddr_in6
);
398 if (flags
& MSG_ERRQUEUE
)
399 return ipv6_recv_error(sk
, msg
, len
);
401 if (np
->rxpmtu
&& np
->rxopt
.bits
.rxpmtu
)
402 return ipv6_recv_rxpmtu(sk
, msg
, len
);
405 skb
= __skb_recv_datagram(sk
, flags
| (noblock
? MSG_DONTWAIT
: 0),
406 &peeked
, &off
, &err
);
410 ulen
= skb
->len
- sizeof(struct udphdr
);
414 else if (copied
< ulen
)
415 msg
->msg_flags
|= MSG_TRUNC
;
417 is_udp4
= (skb
->protocol
== htons(ETH_P_IP
));
420 * If checksum is needed at all, try to do it while copying the
421 * data. If the data is truncated, or if we only want a partial
422 * coverage checksum (UDP-Lite), do it before the copy.
425 if (copied
< ulen
|| UDP_SKB_CB(skb
)->partial_cov
) {
426 if (udp_lib_checksum_complete(skb
))
430 if (skb_csum_unnecessary(skb
))
431 err
= skb_copy_datagram_iovec(skb
, sizeof(struct udphdr
),
432 msg
->msg_iov
, copied
);
434 err
= skb_copy_and_csum_datagram_iovec(skb
, sizeof(struct udphdr
), msg
->msg_iov
);
439 trace_kfree_skb(skb
, udpv6_recvmsg
);
441 atomic_inc(&sk
->sk_drops
);
443 UDP_INC_STATS_USER(sock_net(sk
),
447 UDP6_INC_STATS_USER(sock_net(sk
),
455 UDP_INC_STATS_USER(sock_net(sk
),
456 UDP_MIB_INDATAGRAMS
, is_udplite
);
458 UDP6_INC_STATS_USER(sock_net(sk
),
459 UDP_MIB_INDATAGRAMS
, is_udplite
);
462 sock_recv_ts_and_drops(msg
, sk
, skb
);
464 /* Copy the address. */
466 struct sockaddr_in6
*sin6
;
468 sin6
= (struct sockaddr_in6
*) msg
->msg_name
;
469 sin6
->sin6_family
= AF_INET6
;
470 sin6
->sin6_port
= udp_hdr(skb
)->source
;
471 sin6
->sin6_flowinfo
= 0;
474 ipv6_addr_set_v4mapped(ip_hdr(skb
)->saddr
,
476 sin6
->sin6_scope_id
= 0;
478 sin6
->sin6_addr
= ipv6_hdr(skb
)->saddr
;
479 sin6
->sin6_scope_id
=
480 ipv6_iface_scope_id(&sin6
->sin6_addr
,
486 if (inet
->cmsg_flags
)
487 ip_cmsg_recv(msg
, skb
);
490 ip6_datagram_recv_ctl(sk
, msg
, skb
);
494 if (flags
& MSG_TRUNC
)
498 skb_free_datagram_locked(sk
, skb
);
503 slow
= lock_sock_fast(sk
);
504 if (!skb_kill_datagram(sk
, skb
, flags
)) {
506 UDP_INC_STATS_USER(sock_net(sk
),
507 UDP_MIB_CSUMERRORS
, is_udplite
);
508 UDP_INC_STATS_USER(sock_net(sk
),
509 UDP_MIB_INERRORS
, is_udplite
);
511 UDP6_INC_STATS_USER(sock_net(sk
),
512 UDP_MIB_CSUMERRORS
, is_udplite
);
513 UDP6_INC_STATS_USER(sock_net(sk
),
514 UDP_MIB_INERRORS
, is_udplite
);
517 unlock_sock_fast(sk
, slow
);
522 /* starting over for a new packet */
523 msg
->msg_flags
&= ~MSG_TRUNC
;
527 void __udp6_lib_err(struct sk_buff
*skb
, struct inet6_skb_parm
*opt
,
528 u8 type
, u8 code
, int offset
, __be32 info
,
529 struct udp_table
*udptable
)
531 struct ipv6_pinfo
*np
;
532 const struct ipv6hdr
*hdr
= (const struct ipv6hdr
*)skb
->data
;
533 const struct in6_addr
*saddr
= &hdr
->saddr
;
534 const struct in6_addr
*daddr
= &hdr
->daddr
;
535 struct udphdr
*uh
= (struct udphdr
*)(skb
->data
+offset
);
539 sk
= __udp6_lib_lookup(dev_net(skb
->dev
), daddr
, uh
->dest
,
540 saddr
, uh
->source
, inet6_iif(skb
), udptable
);
544 if (type
== ICMPV6_PKT_TOOBIG
)
545 ip6_sk_update_pmtu(skb
, sk
, info
);
546 if (type
== NDISC_REDIRECT
) {
547 ip6_sk_redirect(skb
, sk
);
553 if (!icmpv6_err_convert(type
, code
, &err
) && !np
->recverr
)
556 if (sk
->sk_state
!= TCP_ESTABLISHED
&& !np
->recverr
)
560 ipv6_icmp_error(sk
, skb
, err
, uh
->dest
, ntohl(info
), (u8
*)(uh
+1));
563 sk
->sk_error_report(sk
);
568 static int __udpv6_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
572 if (!ipv6_addr_any(&sk
->sk_v6_daddr
)) {
573 sock_rps_save_rxhash(sk
, skb
);
574 sk_mark_napi_id(sk
, skb
);
577 rc
= sock_queue_rcv_skb(sk
, skb
);
579 int is_udplite
= IS_UDPLITE(sk
);
581 /* Note that an ENOMEM error is charged twice */
583 UDP6_INC_STATS_BH(sock_net(sk
),
584 UDP_MIB_RCVBUFERRORS
, is_udplite
);
585 UDP6_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
592 static __inline__
void udpv6_err(struct sk_buff
*skb
,
593 struct inet6_skb_parm
*opt
, u8 type
,
594 u8 code
, int offset
, __be32 info
)
596 __udp6_lib_err(skb
, opt
, type
, code
, offset
, info
, &udp_table
);
599 static struct static_key udpv6_encap_needed __read_mostly
;
600 void udpv6_encap_enable(void)
602 if (!static_key_enabled(&udpv6_encap_needed
))
603 static_key_slow_inc(&udpv6_encap_needed
);
605 EXPORT_SYMBOL(udpv6_encap_enable
);
607 int udpv6_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
609 struct udp_sock
*up
= udp_sk(sk
);
611 int is_udplite
= IS_UDPLITE(sk
);
613 if (!xfrm6_policy_check(sk
, XFRM_POLICY_IN
, skb
))
616 if (static_key_false(&udpv6_encap_needed
) && up
->encap_type
) {
617 int (*encap_rcv
)(struct sock
*sk
, struct sk_buff
*skb
);
620 * This is an encapsulation socket so pass the skb to
621 * the socket's udp_encap_rcv() hook. Otherwise, just
622 * fall through and pass this up the UDP socket.
623 * up->encap_rcv() returns the following value:
624 * =0 if skb was successfully passed to the encap
625 * handler or was discarded by it.
626 * >0 if skb should be passed on to UDP.
627 * <0 if skb should be resubmitted as proto -N
630 /* if we're overly short, let UDP handle it */
631 encap_rcv
= ACCESS_ONCE(up
->encap_rcv
);
632 if (skb
->len
> sizeof(struct udphdr
) && encap_rcv
!= NULL
) {
635 ret
= encap_rcv(sk
, skb
);
637 UDP_INC_STATS_BH(sock_net(sk
),
644 /* FALLTHROUGH -- it's a UDP Packet */
648 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
650 if ((is_udplite
& UDPLITE_RECV_CC
) && UDP_SKB_CB(skb
)->partial_cov
) {
652 if (up
->pcrlen
== 0) { /* full coverage was set */
653 LIMIT_NETDEBUG(KERN_WARNING
"UDPLITE6: partial coverage"
654 " %d while full coverage %d requested\n",
655 UDP_SKB_CB(skb
)->cscov
, skb
->len
);
658 if (UDP_SKB_CB(skb
)->cscov
< up
->pcrlen
) {
659 LIMIT_NETDEBUG(KERN_WARNING
"UDPLITE6: coverage %d "
660 "too small, need min %d\n",
661 UDP_SKB_CB(skb
)->cscov
, up
->pcrlen
);
666 if (rcu_access_pointer(sk
->sk_filter
)) {
667 if (udp_lib_checksum_complete(skb
))
671 if (sk_rcvqueues_full(sk
, skb
, sk
->sk_rcvbuf
))
678 if (!sock_owned_by_user(sk
))
679 rc
= __udpv6_queue_rcv_skb(sk
, skb
);
680 else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
688 UDP6_INC_STATS_BH(sock_net(sk
), UDP_MIB_CSUMERRORS
, is_udplite
);
690 UDP6_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
691 atomic_inc(&sk
->sk_drops
);
696 static struct sock
*udp_v6_mcast_next(struct net
*net
, struct sock
*sk
,
697 __be16 loc_port
, const struct in6_addr
*loc_addr
,
698 __be16 rmt_port
, const struct in6_addr
*rmt_addr
,
701 struct hlist_nulls_node
*node
;
703 unsigned short num
= ntohs(loc_port
);
705 sk_nulls_for_each_from(s
, node
) {
706 struct inet_sock
*inet
= inet_sk(s
);
708 if (!net_eq(sock_net(s
), net
))
711 if (udp_sk(s
)->udp_port_hash
== num
&&
712 s
->sk_family
== PF_INET6
) {
713 if (inet
->inet_dport
) {
714 if (inet
->inet_dport
!= rmt_port
)
717 if (!ipv6_addr_any(&sk
->sk_v6_daddr
) &&
718 !ipv6_addr_equal(&sk
->sk_v6_daddr
, rmt_addr
))
721 if (s
->sk_bound_dev_if
&& s
->sk_bound_dev_if
!= dif
)
724 if (!ipv6_addr_any(&sk
->sk_v6_rcv_saddr
)) {
725 if (!ipv6_addr_equal(&sk
->sk_v6_rcv_saddr
, loc_addr
))
728 if (!inet6_mc_check(s
, loc_addr
, rmt_addr
))
736 static void flush_stack(struct sock
**stack
, unsigned int count
,
737 struct sk_buff
*skb
, unsigned int final
)
739 struct sk_buff
*skb1
= NULL
;
743 for (i
= 0; i
< count
; i
++) {
745 if (likely(skb1
== NULL
))
746 skb1
= (i
== final
) ? skb
: skb_clone(skb
, GFP_ATOMIC
);
748 atomic_inc(&sk
->sk_drops
);
749 UDP6_INC_STATS_BH(sock_net(sk
), UDP_MIB_RCVBUFERRORS
,
751 UDP6_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
,
755 if (skb1
&& udpv6_queue_rcv_skb(sk
, skb1
) <= 0)
762 * Note: called only from the BH handler context,
763 * so we don't need to lock the hashes.
765 static int __udp6_lib_mcast_deliver(struct net
*net
, struct sk_buff
*skb
,
766 const struct in6_addr
*saddr
, const struct in6_addr
*daddr
,
767 struct udp_table
*udptable
)
769 struct sock
*sk
, *stack
[256 / sizeof(struct sock
*)];
770 const struct udphdr
*uh
= udp_hdr(skb
);
771 struct udp_hslot
*hslot
= udp_hashslot(udptable
, net
, ntohs(uh
->dest
));
773 unsigned int i
, count
= 0;
775 spin_lock(&hslot
->lock
);
776 sk
= sk_nulls_head(&hslot
->head
);
777 dif
= inet6_iif(skb
);
778 sk
= udp_v6_mcast_next(net
, sk
, uh
->dest
, daddr
, uh
->source
, saddr
, dif
);
781 sk
= udp_v6_mcast_next(net
, sk_nulls_next(sk
), uh
->dest
, daddr
,
782 uh
->source
, saddr
, dif
);
783 if (unlikely(count
== ARRAY_SIZE(stack
))) {
786 flush_stack(stack
, count
, skb
, ~0);
791 * before releasing the lock, we must take reference on sockets
793 for (i
= 0; i
< count
; i
++)
796 spin_unlock(&hslot
->lock
);
799 flush_stack(stack
, count
, skb
, count
- 1);
801 for (i
= 0; i
< count
; i
++)
809 int __udp6_lib_rcv(struct sk_buff
*skb
, struct udp_table
*udptable
,
812 struct net
*net
= dev_net(skb
->dev
);
815 const struct in6_addr
*saddr
, *daddr
;
818 if (!pskb_may_pull(skb
, sizeof(struct udphdr
)))
821 saddr
= &ipv6_hdr(skb
)->saddr
;
822 daddr
= &ipv6_hdr(skb
)->daddr
;
825 ulen
= ntohs(uh
->len
);
829 if (proto
== IPPROTO_UDP
) {
830 /* UDP validates ulen. */
832 /* Check for jumbo payload */
836 if (ulen
< sizeof(*uh
))
839 if (ulen
< skb
->len
) {
840 if (pskb_trim_rcsum(skb
, ulen
))
842 saddr
= &ipv6_hdr(skb
)->saddr
;
843 daddr
= &ipv6_hdr(skb
)->daddr
;
848 if (udp6_csum_init(skb
, uh
, proto
))
852 * Multicast receive code
854 if (ipv6_addr_is_multicast(daddr
))
855 return __udp6_lib_mcast_deliver(net
, skb
,
856 saddr
, daddr
, udptable
);
861 * check socket cache ... must talk to Alan about his plans
862 * for sock caches... i'll skip this for now.
864 sk
= __udp6_lib_lookup_skb(skb
, uh
->source
, uh
->dest
, udptable
);
868 ret
= udpv6_queue_rcv_skb(sk
, skb
);
871 /* a return value > 0 means to resubmit the input, but
872 * it wants the return to be -protocol, or 0
880 if (!xfrm6_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
883 if (udp_lib_checksum_complete(skb
))
886 UDP6_INC_STATS_BH(net
, UDP_MIB_NOPORTS
, proto
== IPPROTO_UDPLITE
);
887 icmpv6_send(skb
, ICMPV6_DEST_UNREACH
, ICMPV6_PORT_UNREACH
, 0);
893 LIMIT_NETDEBUG(KERN_DEBUG
"UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
894 proto
== IPPROTO_UDPLITE
? "-Lite" : "",
903 UDP6_INC_STATS_BH(net
, UDP_MIB_CSUMERRORS
, proto
== IPPROTO_UDPLITE
);
905 UDP6_INC_STATS_BH(net
, UDP_MIB_INERRORS
, proto
== IPPROTO_UDPLITE
);
910 static __inline__
int udpv6_rcv(struct sk_buff
*skb
)
912 return __udp6_lib_rcv(skb
, &udp_table
, IPPROTO_UDP
);
916 * Throw away all pending data and cancel the corking. Socket is locked.
918 static void udp_v6_flush_pending_frames(struct sock
*sk
)
920 struct udp_sock
*up
= udp_sk(sk
);
922 if (up
->pending
== AF_INET
)
923 udp_flush_pending_frames(sk
);
924 else if (up
->pending
) {
927 ip6_flush_pending_frames(sk
);
932 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
933 * @sk: socket we are sending on
934 * @skb: sk_buff containing the filled-in UDP header
935 * (checksum field must be zeroed out)
937 static void udp6_hwcsum_outgoing(struct sock
*sk
, struct sk_buff
*skb
,
938 const struct in6_addr
*saddr
,
939 const struct in6_addr
*daddr
, int len
)
942 struct udphdr
*uh
= udp_hdr(skb
);
945 if (skb_queue_len(&sk
->sk_write_queue
) == 1) {
946 /* Only one fragment on the socket. */
947 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
948 skb
->csum_offset
= offsetof(struct udphdr
, check
);
949 uh
->check
= ~csum_ipv6_magic(saddr
, daddr
, len
, IPPROTO_UDP
, 0);
952 * HW-checksum won't work as there are two or more
953 * fragments on the socket so that all csums of sk_buffs
956 offset
= skb_transport_offset(skb
);
957 skb
->csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
959 skb
->ip_summed
= CHECKSUM_NONE
;
961 skb_queue_walk(&sk
->sk_write_queue
, skb
) {
962 csum
= csum_add(csum
, skb
->csum
);
965 uh
->check
= csum_ipv6_magic(saddr
, daddr
, len
, IPPROTO_UDP
,
968 uh
->check
= CSUM_MANGLED_0
;
976 static int udp_v6_push_pending_frames(struct sock
*sk
)
980 struct udp_sock
*up
= udp_sk(sk
);
981 struct inet_sock
*inet
= inet_sk(sk
);
984 int is_udplite
= IS_UDPLITE(sk
);
987 if (up
->pending
== AF_INET
)
988 return udp_push_pending_frames(sk
);
990 fl6
= &inet
->cork
.fl
.u
.ip6
;
992 /* Grab the skbuff where UDP header space exists. */
993 if ((skb
= skb_peek(&sk
->sk_write_queue
)) == NULL
)
997 * Create a UDP header
1000 uh
->source
= fl6
->fl6_sport
;
1001 uh
->dest
= fl6
->fl6_dport
;
1002 uh
->len
= htons(up
->len
);
1006 csum
= udplite_csum_outgoing(sk
, skb
);
1007 else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) { /* UDP hardware csum */
1008 udp6_hwcsum_outgoing(sk
, skb
, &fl6
->saddr
, &fl6
->daddr
,
1012 csum
= udp_csum_outgoing(sk
, skb
);
1014 /* add protocol-dependent pseudo-header */
1015 uh
->check
= csum_ipv6_magic(&fl6
->saddr
, &fl6
->daddr
,
1016 up
->len
, fl6
->flowi6_proto
, csum
);
1018 uh
->check
= CSUM_MANGLED_0
;
1021 err
= ip6_push_pending_frames(sk
);
1023 if (err
== -ENOBUFS
&& !inet6_sk(sk
)->recverr
) {
1024 UDP6_INC_STATS_USER(sock_net(sk
),
1025 UDP_MIB_SNDBUFERRORS
, is_udplite
);
1029 UDP6_INC_STATS_USER(sock_net(sk
),
1030 UDP_MIB_OUTDATAGRAMS
, is_udplite
);
1037 int udpv6_sendmsg(struct kiocb
*iocb
, struct sock
*sk
,
1038 struct msghdr
*msg
, size_t len
)
1040 struct ipv6_txoptions opt_space
;
1041 struct udp_sock
*up
= udp_sk(sk
);
1042 struct inet_sock
*inet
= inet_sk(sk
);
1043 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1044 struct sockaddr_in6
*sin6
= (struct sockaddr_in6
*) msg
->msg_name
;
1045 struct in6_addr
*daddr
, *final_p
, final
;
1046 struct ipv6_txoptions
*opt
= NULL
;
1047 struct ip6_flowlabel
*flowlabel
= NULL
;
1049 struct dst_entry
*dst
;
1050 int addr_len
= msg
->msg_namelen
;
1055 int corkreq
= up
->corkflag
|| msg
->msg_flags
&MSG_MORE
;
1058 int is_udplite
= IS_UDPLITE(sk
);
1059 int (*getfrag
)(void *, char *, int, int, int, struct sk_buff
*);
1061 /* destination address check */
1063 if (addr_len
< offsetof(struct sockaddr
, sa_data
))
1066 switch (sin6
->sin6_family
) {
1068 if (addr_len
< SIN6_LEN_RFC2133
)
1070 daddr
= &sin6
->sin6_addr
;
1073 goto do_udp_sendmsg
;
1075 msg
->msg_name
= sin6
= NULL
;
1076 msg
->msg_namelen
= addr_len
= 0;
1082 } else if (!up
->pending
) {
1083 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1084 return -EDESTADDRREQ
;
1085 daddr
= &sk
->sk_v6_daddr
;
1090 if (ipv6_addr_v4mapped(daddr
)) {
1091 struct sockaddr_in sin
;
1092 sin
.sin_family
= AF_INET
;
1093 sin
.sin_port
= sin6
? sin6
->sin6_port
: inet
->inet_dport
;
1094 sin
.sin_addr
.s_addr
= daddr
->s6_addr32
[3];
1095 msg
->msg_name
= &sin
;
1096 msg
->msg_namelen
= sizeof(sin
);
1098 if (__ipv6_only_sock(sk
))
1099 return -ENETUNREACH
;
1100 return udp_sendmsg(iocb
, sk
, msg
, len
);
1104 if (up
->pending
== AF_INET
)
1105 return udp_sendmsg(iocb
, sk
, msg
, len
);
1107 /* Rough check on arithmetic overflow,
1108 better check is made in ip6_append_data().
1110 if (len
> INT_MAX
- sizeof(struct udphdr
))
1115 * There are pending frames.
1116 * The socket lock must be held while it's corked.
1119 if (likely(up
->pending
)) {
1120 if (unlikely(up
->pending
!= AF_INET6
)) {
1122 return -EAFNOSUPPORT
;
1125 goto do_append_data
;
1129 ulen
+= sizeof(struct udphdr
);
1131 memset(&fl6
, 0, sizeof(fl6
));
1134 if (sin6
->sin6_port
== 0)
1137 fl6
.fl6_dport
= sin6
->sin6_port
;
1138 daddr
= &sin6
->sin6_addr
;
1141 fl6
.flowlabel
= sin6
->sin6_flowinfo
&IPV6_FLOWINFO_MASK
;
1142 if (fl6
.flowlabel
&IPV6_FLOWLABEL_MASK
) {
1143 flowlabel
= fl6_sock_lookup(sk
, fl6
.flowlabel
);
1144 if (flowlabel
== NULL
)
1146 daddr
= &flowlabel
->dst
;
1151 * Otherwise it will be difficult to maintain
1154 if (sk
->sk_state
== TCP_ESTABLISHED
&&
1155 ipv6_addr_equal(daddr
, &sk
->sk_v6_daddr
))
1156 daddr
= &sk
->sk_v6_daddr
;
1158 if (addr_len
>= sizeof(struct sockaddr_in6
) &&
1159 sin6
->sin6_scope_id
&&
1160 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr
)))
1161 fl6
.flowi6_oif
= sin6
->sin6_scope_id
;
1163 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1164 return -EDESTADDRREQ
;
1166 fl6
.fl6_dport
= inet
->inet_dport
;
1167 daddr
= &sk
->sk_v6_daddr
;
1168 fl6
.flowlabel
= np
->flow_label
;
1172 if (!fl6
.flowi6_oif
)
1173 fl6
.flowi6_oif
= sk
->sk_bound_dev_if
;
1175 if (!fl6
.flowi6_oif
)
1176 fl6
.flowi6_oif
= np
->sticky_pktinfo
.ipi6_ifindex
;
1178 fl6
.flowi6_mark
= sk
->sk_mark
;
1180 if (msg
->msg_controllen
) {
1182 memset(opt
, 0, sizeof(struct ipv6_txoptions
));
1183 opt
->tot_len
= sizeof(*opt
);
1185 err
= ip6_datagram_send_ctl(sock_net(sk
), sk
, msg
, &fl6
, opt
,
1186 &hlimit
, &tclass
, &dontfrag
);
1188 fl6_sock_release(flowlabel
);
1191 if ((fl6
.flowlabel
&IPV6_FLOWLABEL_MASK
) && !flowlabel
) {
1192 flowlabel
= fl6_sock_lookup(sk
, fl6
.flowlabel
);
1193 if (flowlabel
== NULL
)
1196 if (!(opt
->opt_nflen
|opt
->opt_flen
))
1203 opt
= fl6_merge_options(&opt_space
, flowlabel
, opt
);
1204 opt
= ipv6_fixup_options(&opt_space
, opt
);
1206 fl6
.flowi6_proto
= sk
->sk_protocol
;
1207 if (!ipv6_addr_any(daddr
))
1210 fl6
.daddr
.s6_addr
[15] = 0x1; /* :: means loopback (BSD'ism) */
1211 if (ipv6_addr_any(&fl6
.saddr
) && !ipv6_addr_any(&np
->saddr
))
1212 fl6
.saddr
= np
->saddr
;
1213 fl6
.fl6_sport
= inet
->inet_sport
;
1215 final_p
= fl6_update_dst(&fl6
, opt
, &final
);
1219 if (!fl6
.flowi6_oif
&& ipv6_addr_is_multicast(&fl6
.daddr
)) {
1220 fl6
.flowi6_oif
= np
->mcast_oif
;
1222 } else if (!fl6
.flowi6_oif
)
1223 fl6
.flowi6_oif
= np
->ucast_oif
;
1225 security_sk_classify_flow(sk
, flowi6_to_flowi(&fl6
));
1227 dst
= ip6_sk_dst_lookup_flow(sk
, &fl6
, final_p
, true);
1235 if (ipv6_addr_is_multicast(&fl6
.daddr
))
1236 hlimit
= np
->mcast_hops
;
1238 hlimit
= np
->hop_limit
;
1240 hlimit
= ip6_dst_hoplimit(dst
);
1244 tclass
= np
->tclass
;
1247 dontfrag
= np
->dontfrag
;
1249 if (msg
->msg_flags
&MSG_CONFIRM
)
1254 if (unlikely(up
->pending
)) {
1255 /* The socket is already corked while preparing it. */
1256 /* ... which is an evident application bug. --ANK */
1259 LIMIT_NETDEBUG(KERN_DEBUG
"udp cork app bug 2\n");
1264 up
->pending
= AF_INET6
;
1268 getfrag
= is_udplite
? udplite_getfrag
: ip_generic_getfrag
;
1269 err
= ip6_append_data(sk
, getfrag
, msg
->msg_iov
, ulen
,
1270 sizeof(struct udphdr
), hlimit
, tclass
, opt
, &fl6
,
1271 (struct rt6_info
*)dst
,
1272 corkreq
? msg
->msg_flags
|MSG_MORE
: msg
->msg_flags
, dontfrag
);
1274 udp_v6_flush_pending_frames(sk
);
1276 err
= udp_v6_push_pending_frames(sk
);
1277 else if (unlikely(skb_queue_empty(&sk
->sk_write_queue
)))
1282 ip6_dst_store(sk
, dst
,
1283 ipv6_addr_equal(&fl6
.daddr
, &sk
->sk_v6_daddr
) ?
1284 &sk
->sk_v6_daddr
: NULL
,
1285 #ifdef CONFIG_IPV6_SUBTREES
1286 ipv6_addr_equal(&fl6
.saddr
, &np
->saddr
) ?
1297 err
= np
->recverr
? net_xmit_errno(err
) : 0;
1301 fl6_sock_release(flowlabel
);
1305 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1306 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1307 * we don't have a good statistic (IpOutDiscards but it can be too many
1308 * things). We could add another new stat but at least for now that
1309 * seems like overkill.
1311 if (err
== -ENOBUFS
|| test_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
)) {
1312 UDP6_INC_STATS_USER(sock_net(sk
),
1313 UDP_MIB_SNDBUFERRORS
, is_udplite
);
1319 if (!(msg
->msg_flags
&MSG_PROBE
) || len
)
1320 goto back_from_confirm
;
1325 void udpv6_destroy_sock(struct sock
*sk
)
1327 struct udp_sock
*up
= udp_sk(sk
);
1329 udp_v6_flush_pending_frames(sk
);
1332 if (static_key_false(&udpv6_encap_needed
) && up
->encap_type
) {
1333 void (*encap_destroy
)(struct sock
*sk
);
1334 encap_destroy
= ACCESS_ONCE(up
->encap_destroy
);
1339 inet6_destroy_sock(sk
);
1343 * Socket option code for UDP
1345 int udpv6_setsockopt(struct sock
*sk
, int level
, int optname
,
1346 char __user
*optval
, unsigned int optlen
)
1348 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1349 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1350 udp_v6_push_pending_frames
);
1351 return ipv6_setsockopt(sk
, level
, optname
, optval
, optlen
);
1354 #ifdef CONFIG_COMPAT
1355 int compat_udpv6_setsockopt(struct sock
*sk
, int level
, int optname
,
1356 char __user
*optval
, unsigned int optlen
)
1358 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1359 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1360 udp_v6_push_pending_frames
);
1361 return compat_ipv6_setsockopt(sk
, level
, optname
, optval
, optlen
);
1365 int udpv6_getsockopt(struct sock
*sk
, int level
, int optname
,
1366 char __user
*optval
, int __user
*optlen
)
1368 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1369 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1370 return ipv6_getsockopt(sk
, level
, optname
, optval
, optlen
);
1373 #ifdef CONFIG_COMPAT
1374 int compat_udpv6_getsockopt(struct sock
*sk
, int level
, int optname
,
1375 char __user
*optval
, int __user
*optlen
)
1377 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1378 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1379 return compat_ipv6_getsockopt(sk
, level
, optname
, optval
, optlen
);
1383 static const struct inet6_protocol udpv6_protocol
= {
1384 .handler
= udpv6_rcv
,
1385 .err_handler
= udpv6_err
,
1386 .flags
= INET6_PROTO_NOPOLICY
|INET6_PROTO_FINAL
,
1389 /* ------------------------------------------------------------------------ */
1390 #ifdef CONFIG_PROC_FS
1391 int udp6_seq_show(struct seq_file
*seq
, void *v
)
1393 if (v
== SEQ_START_TOKEN
) {
1394 seq_puts(seq
, IPV6_SEQ_DGRAM_HEADER
);
1396 int bucket
= ((struct udp_iter_state
*)seq
->private)->bucket
;
1397 struct inet_sock
*inet
= inet_sk(v
);
1398 __u16 srcp
= ntohs(inet
->inet_sport
);
1399 __u16 destp
= ntohs(inet
->inet_dport
);
1400 ip6_dgram_sock_seq_show(seq
, v
, srcp
, destp
, bucket
);
1405 static const struct file_operations udp6_afinfo_seq_fops
= {
1406 .owner
= THIS_MODULE
,
1407 .open
= udp_seq_open
,
1409 .llseek
= seq_lseek
,
1410 .release
= seq_release_net
1413 static struct udp_seq_afinfo udp6_seq_afinfo
= {
1416 .udp_table
= &udp_table
,
1417 .seq_fops
= &udp6_afinfo_seq_fops
,
1419 .show
= udp6_seq_show
,
1423 int __net_init
udp6_proc_init(struct net
*net
)
1425 return udp_proc_register(net
, &udp6_seq_afinfo
);
1428 void udp6_proc_exit(struct net
*net
) {
1429 udp_proc_unregister(net
, &udp6_seq_afinfo
);
1431 #endif /* CONFIG_PROC_FS */
1433 void udp_v6_clear_sk(struct sock
*sk
, int size
)
1435 struct inet_sock
*inet
= inet_sk(sk
);
1437 /* we do not want to clear pinet6 field, because of RCU lookups */
1438 sk_prot_clear_portaddr_nulls(sk
, offsetof(struct inet_sock
, pinet6
));
1440 size
-= offsetof(struct inet_sock
, pinet6
) + sizeof(inet
->pinet6
);
1441 memset(&inet
->pinet6
+ 1, 0, size
);
1444 /* ------------------------------------------------------------------------ */
1446 struct proto udpv6_prot
= {
1448 .owner
= THIS_MODULE
,
1449 .close
= udp_lib_close
,
1450 .connect
= ip6_datagram_connect
,
1451 .disconnect
= udp_disconnect
,
1453 .destroy
= udpv6_destroy_sock
,
1454 .setsockopt
= udpv6_setsockopt
,
1455 .getsockopt
= udpv6_getsockopt
,
1456 .sendmsg
= udpv6_sendmsg
,
1457 .recvmsg
= udpv6_recvmsg
,
1458 .backlog_rcv
= __udpv6_queue_rcv_skb
,
1459 .hash
= udp_lib_hash
,
1460 .unhash
= udp_lib_unhash
,
1461 .rehash
= udp_v6_rehash
,
1462 .get_port
= udp_v6_get_port
,
1463 .memory_allocated
= &udp_memory_allocated
,
1464 .sysctl_mem
= sysctl_udp_mem
,
1465 .sysctl_wmem
= &sysctl_udp_wmem_min
,
1466 .sysctl_rmem
= &sysctl_udp_rmem_min
,
1467 .obj_size
= sizeof(struct udp6_sock
),
1468 .slab_flags
= SLAB_DESTROY_BY_RCU
,
1469 .h
.udp_table
= &udp_table
,
1470 #ifdef CONFIG_COMPAT
1471 .compat_setsockopt
= compat_udpv6_setsockopt
,
1472 .compat_getsockopt
= compat_udpv6_getsockopt
,
1474 .clear_sk
= udp_v6_clear_sk
,
1477 static struct inet_protosw udpv6_protosw
= {
1479 .protocol
= IPPROTO_UDP
,
1480 .prot
= &udpv6_prot
,
1481 .ops
= &inet6_dgram_ops
,
1482 .no_check
= UDP_CSUM_DEFAULT
,
1483 .flags
= INET_PROTOSW_PERMANENT
,
1487 int __init
udpv6_init(void)
1491 ret
= inet6_add_protocol(&udpv6_protocol
, IPPROTO_UDP
);
1495 ret
= inet6_register_protosw(&udpv6_protosw
);
1497 goto out_udpv6_protocol
;
1502 inet6_del_protocol(&udpv6_protocol
, IPPROTO_UDP
);
1506 void udpv6_exit(void)
1508 inet6_unregister_protosw(&udpv6_protosw
);
1509 inet6_del_protocol(&udpv6_protocol
, IPPROTO_UDP
);