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/addrconf.h>
41 #include <net/ndisc.h>
42 #include <net/protocol.h>
43 #include <net/transp_v6.h>
44 #include <net/ip6_route.h>
46 #include <net/tcp_states.h>
47 #include <net/ip6_checksum.h>
49 #include <net/inet6_hashtables.h>
50 #include <net/busy_poll.h>
51 #include <net/sock_reuseport.h>
53 #include <linux/proc_fs.h>
54 #include <linux/seq_file.h>
55 #include <trace/events/skb.h>
58 static u32
udp6_ehashfn(const struct net
*net
,
59 const struct in6_addr
*laddr
,
61 const struct in6_addr
*faddr
,
64 static u32 udp6_ehash_secret __read_mostly
;
65 static u32 udp_ipv6_hash_secret __read_mostly
;
69 net_get_random_once(&udp6_ehash_secret
,
70 sizeof(udp6_ehash_secret
));
71 net_get_random_once(&udp_ipv6_hash_secret
,
72 sizeof(udp_ipv6_hash_secret
));
74 lhash
= (__force u32
)laddr
->s6_addr32
[3];
75 fhash
= __ipv6_addr_jhash(faddr
, udp_ipv6_hash_secret
);
77 return __inet6_ehashfn(lhash
, lport
, fhash
, fport
,
78 udp_ipv6_hash_secret
+ net_hash_mix(net
));
81 static u32
udp6_portaddr_hash(const struct net
*net
,
82 const struct in6_addr
*addr6
,
85 unsigned int hash
, mix
= net_hash_mix(net
);
87 if (ipv6_addr_any(addr6
))
88 hash
= jhash_1word(0, mix
);
89 else if (ipv6_addr_v4mapped(addr6
))
90 hash
= jhash_1word((__force u32
)addr6
->s6_addr32
[3], mix
);
92 hash
= jhash2((__force u32
*)addr6
->s6_addr32
, 4, mix
);
97 int udp_v6_get_port(struct sock
*sk
, unsigned short snum
)
99 unsigned int hash2_nulladdr
=
100 udp6_portaddr_hash(sock_net(sk
), &in6addr_any
, snum
);
101 unsigned int hash2_partial
=
102 udp6_portaddr_hash(sock_net(sk
), &sk
->sk_v6_rcv_saddr
, 0);
104 /* precompute partial secondary hash */
105 udp_sk(sk
)->udp_portaddr_hash
= hash2_partial
;
106 return udp_lib_get_port(sk
, snum
, ipv6_rcv_saddr_equal
, hash2_nulladdr
);
109 static void udp_v6_rehash(struct sock
*sk
)
111 u16 new_hash
= udp6_portaddr_hash(sock_net(sk
),
112 &sk
->sk_v6_rcv_saddr
,
113 inet_sk(sk
)->inet_num
);
115 udp_lib_rehash(sk
, new_hash
);
118 static int compute_score(struct sock
*sk
, struct net
*net
,
119 const struct in6_addr
*saddr
, __be16 sport
,
120 const struct in6_addr
*daddr
, unsigned short hnum
,
124 struct inet_sock
*inet
;
126 if (!net_eq(sock_net(sk
), net
) ||
127 udp_sk(sk
)->udp_port_hash
!= hnum
||
128 sk
->sk_family
!= PF_INET6
)
134 if (inet
->inet_dport
) {
135 if (inet
->inet_dport
!= sport
)
140 if (!ipv6_addr_any(&sk
->sk_v6_rcv_saddr
)) {
141 if (!ipv6_addr_equal(&sk
->sk_v6_rcv_saddr
, daddr
))
146 if (!ipv6_addr_any(&sk
->sk_v6_daddr
)) {
147 if (!ipv6_addr_equal(&sk
->sk_v6_daddr
, saddr
))
152 if (sk
->sk_bound_dev_if
) {
153 if (sk
->sk_bound_dev_if
!= dif
)
158 if (sk
->sk_incoming_cpu
== raw_smp_processor_id())
164 /* called with rcu_read_lock() */
165 static struct sock
*udp6_lib_lookup2(struct net
*net
,
166 const struct in6_addr
*saddr
, __be16 sport
,
167 const struct in6_addr
*daddr
, unsigned int hnum
, int dif
,
168 struct udp_hslot
*hslot2
,
171 struct sock
*sk
, *result
;
172 int score
, badness
, matches
= 0, reuseport
= 0;
177 udp_portaddr_for_each_entry_rcu(sk
, &hslot2
->head
) {
178 score
= compute_score(sk
, net
, saddr
, sport
,
180 if (score
> badness
) {
181 reuseport
= sk
->sk_reuseport
;
183 hash
= udp6_ehashfn(net
, daddr
, hnum
,
186 result
= reuseport_select_sock(sk
, hash
, skb
,
187 sizeof(struct udphdr
));
194 } else if (score
== badness
&& reuseport
) {
196 if (reciprocal_scale(hash
, matches
) == 0)
198 hash
= next_pseudo_random32(hash
);
204 /* rcu_read_lock() must be held */
205 struct sock
*__udp6_lib_lookup(struct net
*net
,
206 const struct in6_addr
*saddr
, __be16 sport
,
207 const struct in6_addr
*daddr
, __be16 dport
,
208 int dif
, struct udp_table
*udptable
,
211 struct sock
*sk
, *result
;
212 unsigned short hnum
= ntohs(dport
);
213 unsigned int hash2
, slot2
, slot
= udp_hashfn(net
, hnum
, udptable
->mask
);
214 struct udp_hslot
*hslot2
, *hslot
= &udptable
->hash
[slot
];
215 int score
, badness
, matches
= 0, reuseport
= 0;
218 if (hslot
->count
> 10) {
219 hash2
= udp6_portaddr_hash(net
, daddr
, hnum
);
220 slot2
= hash2
& udptable
->mask
;
221 hslot2
= &udptable
->hash2
[slot2
];
222 if (hslot
->count
< hslot2
->count
)
225 result
= udp6_lib_lookup2(net
, saddr
, sport
,
229 unsigned int old_slot2
= slot2
;
230 hash2
= udp6_portaddr_hash(net
, &in6addr_any
, hnum
);
231 slot2
= hash2
& udptable
->mask
;
232 /* avoid searching the same slot again. */
233 if (unlikely(slot2
== old_slot2
))
236 hslot2
= &udptable
->hash2
[slot2
];
237 if (hslot
->count
< hslot2
->count
)
240 result
= udp6_lib_lookup2(net
, saddr
, sport
,
249 sk_for_each_rcu(sk
, &hslot
->head
) {
250 score
= compute_score(sk
, net
, saddr
, sport
, daddr
, hnum
, dif
);
251 if (score
> badness
) {
252 reuseport
= sk
->sk_reuseport
;
254 hash
= udp6_ehashfn(net
, daddr
, hnum
,
256 result
= reuseport_select_sock(sk
, hash
, skb
,
257 sizeof(struct udphdr
));
264 } else if (score
== badness
&& reuseport
) {
266 if (reciprocal_scale(hash
, matches
) == 0)
268 hash
= next_pseudo_random32(hash
);
273 EXPORT_SYMBOL_GPL(__udp6_lib_lookup
);
275 static struct sock
*__udp6_lib_lookup_skb(struct sk_buff
*skb
,
276 __be16 sport
, __be16 dport
,
277 struct udp_table
*udptable
)
279 const struct ipv6hdr
*iph
= ipv6_hdr(skb
);
282 sk
= skb_steal_sock(skb
);
285 return __udp6_lib_lookup(dev_net(skb
->dev
), &iph
->saddr
, sport
,
286 &iph
->daddr
, dport
, inet6_iif(skb
),
290 struct sock
*udp6_lib_lookup_skb(struct sk_buff
*skb
,
291 __be16 sport
, __be16 dport
)
293 const struct ipv6hdr
*iph
= ipv6_hdr(skb
);
295 return __udp6_lib_lookup(dev_net(skb
->dev
), &iph
->saddr
, sport
,
296 &iph
->daddr
, dport
, inet6_iif(skb
),
299 EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb
);
301 /* Must be called under rcu_read_lock().
302 * Does increment socket refcount.
304 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
305 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY)
306 struct sock
*udp6_lib_lookup(struct net
*net
, const struct in6_addr
*saddr
, __be16 sport
,
307 const struct in6_addr
*daddr
, __be16 dport
, int dif
)
311 sk
= __udp6_lib_lookup(net
, saddr
, sport
, daddr
, dport
,
312 dif
, &udp_table
, NULL
);
313 if (sk
&& !atomic_inc_not_zero(&sk
->sk_refcnt
))
317 EXPORT_SYMBOL_GPL(udp6_lib_lookup
);
321 * This should be easy, if there is something there we
322 * return it, otherwise we block.
325 int udpv6_recvmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
,
326 int noblock
, int flags
, int *addr_len
)
328 struct ipv6_pinfo
*np
= inet6_sk(sk
);
329 struct inet_sock
*inet
= inet_sk(sk
);
331 unsigned int ulen
, copied
;
332 int peeked
, peeking
, off
;
334 int is_udplite
= IS_UDPLITE(sk
);
335 bool checksum_valid
= false;
339 if (flags
& MSG_ERRQUEUE
)
340 return ipv6_recv_error(sk
, msg
, len
, addr_len
);
342 if (np
->rxpmtu
&& np
->rxopt
.bits
.rxpmtu
)
343 return ipv6_recv_rxpmtu(sk
, msg
, len
, addr_len
);
346 peeking
= off
= sk_peek_offset(sk
, flags
);
347 skb
= __skb_recv_datagram(sk
, flags
| (noblock
? MSG_DONTWAIT
: 0),
348 &peeked
, &off
, &err
);
354 if (copied
> ulen
- off
)
356 else if (copied
< ulen
)
357 msg
->msg_flags
|= MSG_TRUNC
;
359 is_udp4
= (skb
->protocol
== htons(ETH_P_IP
));
362 * If checksum is needed at all, try to do it while copying the
363 * data. If the data is truncated, or if we only want a partial
364 * coverage checksum (UDP-Lite), do it before the copy.
367 if (copied
< ulen
|| UDP_SKB_CB(skb
)->partial_cov
|| peeking
) {
368 checksum_valid
= !udp_lib_checksum_complete(skb
);
373 if (checksum_valid
|| skb_csum_unnecessary(skb
))
374 err
= skb_copy_datagram_msg(skb
, off
, msg
, copied
);
376 err
= skb_copy_and_csum_datagram_msg(skb
, off
, msg
);
381 trace_kfree_skb(skb
, udpv6_recvmsg
);
383 atomic_inc(&sk
->sk_drops
);
385 UDP_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
,
388 UDP6_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
,
391 skb_free_datagram_locked(sk
, skb
);
396 UDP_INC_STATS(sock_net(sk
), UDP_MIB_INDATAGRAMS
,
399 UDP6_INC_STATS(sock_net(sk
), UDP_MIB_INDATAGRAMS
,
403 sock_recv_ts_and_drops(msg
, sk
, skb
);
405 /* Copy the address. */
407 DECLARE_SOCKADDR(struct sockaddr_in6
*, sin6
, msg
->msg_name
);
408 sin6
->sin6_family
= AF_INET6
;
409 sin6
->sin6_port
= udp_hdr(skb
)->source
;
410 sin6
->sin6_flowinfo
= 0;
413 ipv6_addr_set_v4mapped(ip_hdr(skb
)->saddr
,
415 sin6
->sin6_scope_id
= 0;
417 sin6
->sin6_addr
= ipv6_hdr(skb
)->saddr
;
418 sin6
->sin6_scope_id
=
419 ipv6_iface_scope_id(&sin6
->sin6_addr
,
422 *addr_len
= sizeof(*sin6
);
426 ip6_datagram_recv_common_ctl(sk
, msg
, skb
);
429 if (inet
->cmsg_flags
)
430 ip_cmsg_recv_offset(msg
, skb
,
431 sizeof(struct udphdr
), off
);
434 ip6_datagram_recv_specific_ctl(sk
, msg
, skb
);
438 if (flags
& MSG_TRUNC
)
441 __skb_free_datagram_locked(sk
, skb
, peeking
? -err
: err
);
445 slow
= lock_sock_fast(sk
);
446 if (!skb_kill_datagram(sk
, skb
, flags
)) {
448 UDP_INC_STATS(sock_net(sk
),
449 UDP_MIB_CSUMERRORS
, is_udplite
);
450 UDP_INC_STATS(sock_net(sk
),
451 UDP_MIB_INERRORS
, is_udplite
);
453 UDP6_INC_STATS(sock_net(sk
),
454 UDP_MIB_CSUMERRORS
, is_udplite
);
455 UDP6_INC_STATS(sock_net(sk
),
456 UDP_MIB_INERRORS
, is_udplite
);
459 unlock_sock_fast(sk
, slow
);
461 /* starting over for a new packet, but check if we need to yield */
463 msg
->msg_flags
&= ~MSG_TRUNC
;
467 void __udp6_lib_err(struct sk_buff
*skb
, struct inet6_skb_parm
*opt
,
468 u8 type
, u8 code
, int offset
, __be32 info
,
469 struct udp_table
*udptable
)
471 struct ipv6_pinfo
*np
;
472 const struct ipv6hdr
*hdr
= (const struct ipv6hdr
*)skb
->data
;
473 const struct in6_addr
*saddr
= &hdr
->saddr
;
474 const struct in6_addr
*daddr
= &hdr
->daddr
;
475 struct udphdr
*uh
= (struct udphdr
*)(skb
->data
+offset
);
479 struct net
*net
= dev_net(skb
->dev
);
481 sk
= __udp6_lib_lookup(net
, daddr
, uh
->dest
, saddr
, uh
->source
,
482 inet6_iif(skb
), udptable
, skb
);
484 __ICMP6_INC_STATS(net
, __in6_dev_get(skb
->dev
),
489 harderr
= icmpv6_err_convert(type
, code
, &err
);
492 if (type
== ICMPV6_PKT_TOOBIG
) {
493 if (!ip6_sk_accept_pmtu(sk
))
495 ip6_sk_update_pmtu(skb
, sk
, info
);
496 if (np
->pmtudisc
!= IPV6_PMTUDISC_DONT
)
499 if (type
== NDISC_REDIRECT
) {
500 ip6_sk_redirect(skb
, sk
);
505 if (!harderr
|| sk
->sk_state
!= TCP_ESTABLISHED
)
508 ipv6_icmp_error(sk
, skb
, err
, uh
->dest
, ntohl(info
), (u8
*)(uh
+1));
512 sk
->sk_error_report(sk
);
517 int __udpv6_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
521 if (!ipv6_addr_any(&sk
->sk_v6_daddr
)) {
522 sock_rps_save_rxhash(sk
, skb
);
523 sk_mark_napi_id(sk
, skb
);
524 sk_incoming_cpu_update(sk
);
527 rc
= __sock_queue_rcv_skb(sk
, skb
);
529 int is_udplite
= IS_UDPLITE(sk
);
531 /* Note that an ENOMEM error is charged twice */
533 UDP6_INC_STATS(sock_net(sk
),
534 UDP_MIB_RCVBUFERRORS
, is_udplite
);
535 UDP6_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
542 static __inline__
void udpv6_err(struct sk_buff
*skb
,
543 struct inet6_skb_parm
*opt
, u8 type
,
544 u8 code
, int offset
, __be32 info
)
546 __udp6_lib_err(skb
, opt
, type
, code
, offset
, info
, &udp_table
);
549 static struct static_key udpv6_encap_needed __read_mostly
;
550 void udpv6_encap_enable(void)
552 if (!static_key_enabled(&udpv6_encap_needed
))
553 static_key_slow_inc(&udpv6_encap_needed
);
555 EXPORT_SYMBOL(udpv6_encap_enable
);
557 int udpv6_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
559 struct udp_sock
*up
= udp_sk(sk
);
561 int is_udplite
= IS_UDPLITE(sk
);
563 if (!xfrm6_policy_check(sk
, XFRM_POLICY_IN
, skb
))
566 if (static_key_false(&udpv6_encap_needed
) && up
->encap_type
) {
567 int (*encap_rcv
)(struct sock
*sk
, struct sk_buff
*skb
);
570 * This is an encapsulation socket so pass the skb to
571 * the socket's udp_encap_rcv() hook. Otherwise, just
572 * fall through and pass this up the UDP socket.
573 * up->encap_rcv() returns the following value:
574 * =0 if skb was successfully passed to the encap
575 * handler or was discarded by it.
576 * >0 if skb should be passed on to UDP.
577 * <0 if skb should be resubmitted as proto -N
580 /* if we're overly short, let UDP handle it */
581 encap_rcv
= ACCESS_ONCE(up
->encap_rcv
);
585 /* Verify checksum before giving to encap */
586 if (udp_lib_checksum_complete(skb
))
589 ret
= encap_rcv(sk
, skb
);
591 __UDP_INC_STATS(sock_net(sk
),
598 /* FALLTHROUGH -- it's a UDP Packet */
602 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
604 if ((is_udplite
& UDPLITE_RECV_CC
) && UDP_SKB_CB(skb
)->partial_cov
) {
606 if (up
->pcrlen
== 0) { /* full coverage was set */
607 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
608 UDP_SKB_CB(skb
)->cscov
, skb
->len
);
611 if (UDP_SKB_CB(skb
)->cscov
< up
->pcrlen
) {
612 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
613 UDP_SKB_CB(skb
)->cscov
, up
->pcrlen
);
618 if (rcu_access_pointer(sk
->sk_filter
) &&
619 udp_lib_checksum_complete(skb
))
622 if (sk_filter_trim_cap(sk
, skb
, sizeof(struct udphdr
)))
625 udp_csum_pull_header(skb
);
626 if (sk_rcvqueues_full(sk
, sk
->sk_rcvbuf
)) {
627 __UDP6_INC_STATS(sock_net(sk
),
628 UDP_MIB_RCVBUFERRORS
, is_udplite
);
636 if (!sock_owned_by_user(sk
))
637 rc
= __udpv6_queue_rcv_skb(sk
, skb
);
638 else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
647 __UDP6_INC_STATS(sock_net(sk
), UDP_MIB_CSUMERRORS
, is_udplite
);
649 __UDP6_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
650 atomic_inc(&sk
->sk_drops
);
655 static bool __udp_v6_is_mcast_sock(struct net
*net
, struct sock
*sk
,
656 __be16 loc_port
, const struct in6_addr
*loc_addr
,
657 __be16 rmt_port
, const struct in6_addr
*rmt_addr
,
658 int dif
, unsigned short hnum
)
660 struct inet_sock
*inet
= inet_sk(sk
);
662 if (!net_eq(sock_net(sk
), net
))
665 if (udp_sk(sk
)->udp_port_hash
!= hnum
||
666 sk
->sk_family
!= PF_INET6
||
667 (inet
->inet_dport
&& inet
->inet_dport
!= rmt_port
) ||
668 (!ipv6_addr_any(&sk
->sk_v6_daddr
) &&
669 !ipv6_addr_equal(&sk
->sk_v6_daddr
, rmt_addr
)) ||
670 (sk
->sk_bound_dev_if
&& sk
->sk_bound_dev_if
!= dif
) ||
671 (!ipv6_addr_any(&sk
->sk_v6_rcv_saddr
) &&
672 !ipv6_addr_equal(&sk
->sk_v6_rcv_saddr
, loc_addr
)))
674 if (!inet6_mc_check(sk
, loc_addr
, rmt_addr
))
679 static void udp6_csum_zero_error(struct sk_buff
*skb
)
681 /* RFC 2460 section 8.1 says that we SHOULD log
682 * this error. Well, it is reasonable.
684 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
685 &ipv6_hdr(skb
)->saddr
, ntohs(udp_hdr(skb
)->source
),
686 &ipv6_hdr(skb
)->daddr
, ntohs(udp_hdr(skb
)->dest
));
690 * Note: called only from the BH handler context,
691 * so we don't need to lock the hashes.
693 static int __udp6_lib_mcast_deliver(struct net
*net
, struct sk_buff
*skb
,
694 const struct in6_addr
*saddr
, const struct in6_addr
*daddr
,
695 struct udp_table
*udptable
, int proto
)
697 struct sock
*sk
, *first
= NULL
;
698 const struct udphdr
*uh
= udp_hdr(skb
);
699 unsigned short hnum
= ntohs(uh
->dest
);
700 struct udp_hslot
*hslot
= udp_hashslot(udptable
, net
, hnum
);
701 unsigned int offset
= offsetof(typeof(*sk
), sk_node
);
702 unsigned int hash2
= 0, hash2_any
= 0, use_hash2
= (hslot
->count
> 10);
703 int dif
= inet6_iif(skb
);
704 struct hlist_node
*node
;
705 struct sk_buff
*nskb
;
708 hash2_any
= udp6_portaddr_hash(net
, &in6addr_any
, hnum
) &
710 hash2
= udp6_portaddr_hash(net
, daddr
, hnum
) & udptable
->mask
;
712 hslot
= &udptable
->hash2
[hash2
];
713 offset
= offsetof(typeof(*sk
), __sk_common
.skc_portaddr_node
);
716 sk_for_each_entry_offset_rcu(sk
, node
, &hslot
->head
, offset
) {
717 if (!__udp_v6_is_mcast_sock(net
, sk
, uh
->dest
, daddr
,
718 uh
->source
, saddr
, dif
, hnum
))
720 /* If zero checksum and no_check is not on for
721 * the socket then skip it.
723 if (!uh
->check
&& !udp_sk(sk
)->no_check6_rx
)
729 nskb
= skb_clone(skb
, GFP_ATOMIC
);
730 if (unlikely(!nskb
)) {
731 atomic_inc(&sk
->sk_drops
);
732 __UDP6_INC_STATS(net
, UDP_MIB_RCVBUFERRORS
,
734 __UDP6_INC_STATS(net
, UDP_MIB_INERRORS
,
739 if (udpv6_queue_rcv_skb(sk
, nskb
) > 0)
743 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
744 if (use_hash2
&& hash2
!= hash2_any
) {
750 if (udpv6_queue_rcv_skb(first
, skb
) > 0)
754 __UDP6_INC_STATS(net
, UDP_MIB_IGNOREDMULTI
,
755 proto
== IPPROTO_UDPLITE
);
760 int __udp6_lib_rcv(struct sk_buff
*skb
, struct udp_table
*udptable
,
763 const struct in6_addr
*saddr
, *daddr
;
764 struct net
*net
= dev_net(skb
->dev
);
769 if (!pskb_may_pull(skb
, sizeof(struct udphdr
)))
772 saddr
= &ipv6_hdr(skb
)->saddr
;
773 daddr
= &ipv6_hdr(skb
)->daddr
;
776 ulen
= ntohs(uh
->len
);
780 if (proto
== IPPROTO_UDP
) {
781 /* UDP validates ulen. */
783 /* Check for jumbo payload */
787 if (ulen
< sizeof(*uh
))
790 if (ulen
< skb
->len
) {
791 if (pskb_trim_rcsum(skb
, ulen
))
793 saddr
= &ipv6_hdr(skb
)->saddr
;
794 daddr
= &ipv6_hdr(skb
)->daddr
;
799 if (udp6_csum_init(skb
, uh
, proto
))
803 * Multicast receive code
805 if (ipv6_addr_is_multicast(daddr
))
806 return __udp6_lib_mcast_deliver(net
, skb
,
807 saddr
, daddr
, udptable
, proto
);
812 * check socket cache ... must talk to Alan about his plans
813 * for sock caches... i'll skip this for now.
815 sk
= __udp6_lib_lookup_skb(skb
, uh
->source
, uh
->dest
, udptable
);
819 if (!uh
->check
&& !udp_sk(sk
)->no_check6_rx
) {
820 udp6_csum_zero_error(skb
);
824 if (inet_get_convert_csum(sk
) && uh
->check
&& !IS_UDPLITE(sk
))
825 skb_checksum_try_convert(skb
, IPPROTO_UDP
, uh
->check
,
828 ret
= udpv6_queue_rcv_skb(sk
, skb
);
830 /* a return value > 0 means to resubmit the input */
838 udp6_csum_zero_error(skb
);
842 if (!xfrm6_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
845 if (udp_lib_checksum_complete(skb
))
848 __UDP6_INC_STATS(net
, UDP_MIB_NOPORTS
, proto
== IPPROTO_UDPLITE
);
849 icmpv6_send(skb
, ICMPV6_DEST_UNREACH
, ICMPV6_PORT_UNREACH
, 0);
855 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
856 proto
== IPPROTO_UDPLITE
? "-Lite" : "",
857 saddr
, ntohs(uh
->source
),
859 daddr
, ntohs(uh
->dest
));
862 __UDP6_INC_STATS(net
, UDP_MIB_CSUMERRORS
, proto
== IPPROTO_UDPLITE
);
864 __UDP6_INC_STATS(net
, UDP_MIB_INERRORS
, proto
== IPPROTO_UDPLITE
);
869 static __inline__
int udpv6_rcv(struct sk_buff
*skb
)
871 return __udp6_lib_rcv(skb
, &udp_table
, IPPROTO_UDP
);
875 * Throw away all pending data and cancel the corking. Socket is locked.
877 static void udp_v6_flush_pending_frames(struct sock
*sk
)
879 struct udp_sock
*up
= udp_sk(sk
);
881 if (up
->pending
== AF_INET
)
882 udp_flush_pending_frames(sk
);
883 else if (up
->pending
) {
886 ip6_flush_pending_frames(sk
);
891 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
892 * @sk: socket we are sending on
893 * @skb: sk_buff containing the filled-in UDP header
894 * (checksum field must be zeroed out)
896 static void udp6_hwcsum_outgoing(struct sock
*sk
, struct sk_buff
*skb
,
897 const struct in6_addr
*saddr
,
898 const struct in6_addr
*daddr
, int len
)
901 struct udphdr
*uh
= udp_hdr(skb
);
902 struct sk_buff
*frags
= skb_shinfo(skb
)->frag_list
;
906 /* Only one fragment on the socket. */
907 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
908 skb
->csum_offset
= offsetof(struct udphdr
, check
);
909 uh
->check
= ~csum_ipv6_magic(saddr
, daddr
, len
, IPPROTO_UDP
, 0);
912 * HW-checksum won't work as there are two or more
913 * fragments on the socket so that all csums of sk_buffs
916 offset
= skb_transport_offset(skb
);
917 skb
->csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
919 skb
->ip_summed
= CHECKSUM_NONE
;
922 csum
= csum_add(csum
, frags
->csum
);
923 } while ((frags
= frags
->next
));
925 uh
->check
= csum_ipv6_magic(saddr
, daddr
, len
, IPPROTO_UDP
,
928 uh
->check
= CSUM_MANGLED_0
;
936 static int udp_v6_send_skb(struct sk_buff
*skb
, struct flowi6
*fl6
)
938 struct sock
*sk
= skb
->sk
;
941 int is_udplite
= IS_UDPLITE(sk
);
943 int offset
= skb_transport_offset(skb
);
944 int len
= skb
->len
- offset
;
947 * Create a UDP header
950 uh
->source
= fl6
->fl6_sport
;
951 uh
->dest
= fl6
->fl6_dport
;
952 uh
->len
= htons(len
);
956 csum
= udplite_csum(skb
);
957 else if (udp_sk(sk
)->no_check6_tx
) { /* UDP csum disabled */
958 skb
->ip_summed
= CHECKSUM_NONE
;
960 } else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) { /* UDP hardware csum */
961 udp6_hwcsum_outgoing(sk
, skb
, &fl6
->saddr
, &fl6
->daddr
, len
);
964 csum
= udp_csum(skb
);
966 /* add protocol-dependent pseudo-header */
967 uh
->check
= csum_ipv6_magic(&fl6
->saddr
, &fl6
->daddr
,
968 len
, fl6
->flowi6_proto
, csum
);
970 uh
->check
= CSUM_MANGLED_0
;
973 err
= ip6_send_skb(skb
);
975 if (err
== -ENOBUFS
&& !inet6_sk(sk
)->recverr
) {
976 UDP6_INC_STATS(sock_net(sk
),
977 UDP_MIB_SNDBUFERRORS
, is_udplite
);
981 UDP6_INC_STATS(sock_net(sk
),
982 UDP_MIB_OUTDATAGRAMS
, is_udplite
);
987 static int udp_v6_push_pending_frames(struct sock
*sk
)
990 struct udp_sock
*up
= udp_sk(sk
);
994 if (up
->pending
== AF_INET
)
995 return udp_push_pending_frames(sk
);
997 /* ip6_finish_skb will release the cork, so make a copy of
1000 fl6
= inet_sk(sk
)->cork
.fl
.u
.ip6
;
1002 skb
= ip6_finish_skb(sk
);
1006 err
= udp_v6_send_skb(skb
, &fl6
);
1014 int udpv6_sendmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
)
1016 struct ipv6_txoptions opt_space
;
1017 struct udp_sock
*up
= udp_sk(sk
);
1018 struct inet_sock
*inet
= inet_sk(sk
);
1019 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1020 DECLARE_SOCKADDR(struct sockaddr_in6
*, sin6
, msg
->msg_name
);
1021 struct in6_addr
*daddr
, *final_p
, final
;
1022 struct ipv6_txoptions
*opt
= NULL
;
1023 struct ipv6_txoptions
*opt_to_free
= NULL
;
1024 struct ip6_flowlabel
*flowlabel
= NULL
;
1026 struct dst_entry
*dst
;
1027 struct ipcm6_cookie ipc6
;
1028 int addr_len
= msg
->msg_namelen
;
1030 int corkreq
= up
->corkflag
|| msg
->msg_flags
&MSG_MORE
;
1033 int is_udplite
= IS_UDPLITE(sk
);
1034 int (*getfrag
)(void *, char *, int, int, int, struct sk_buff
*);
1035 struct sockcm_cookie sockc
;
1041 /* destination address check */
1043 if (addr_len
< offsetof(struct sockaddr
, sa_data
))
1046 switch (sin6
->sin6_family
) {
1048 if (addr_len
< SIN6_LEN_RFC2133
)
1050 daddr
= &sin6
->sin6_addr
;
1053 goto do_udp_sendmsg
;
1055 msg
->msg_name
= sin6
= NULL
;
1056 msg
->msg_namelen
= addr_len
= 0;
1062 } else if (!up
->pending
) {
1063 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1064 return -EDESTADDRREQ
;
1065 daddr
= &sk
->sk_v6_daddr
;
1070 if (ipv6_addr_v4mapped(daddr
)) {
1071 struct sockaddr_in sin
;
1072 sin
.sin_family
= AF_INET
;
1073 sin
.sin_port
= sin6
? sin6
->sin6_port
: inet
->inet_dport
;
1074 sin
.sin_addr
.s_addr
= daddr
->s6_addr32
[3];
1075 msg
->msg_name
= &sin
;
1076 msg
->msg_namelen
= sizeof(sin
);
1078 if (__ipv6_only_sock(sk
))
1079 return -ENETUNREACH
;
1080 return udp_sendmsg(sk
, msg
, len
);
1084 if (up
->pending
== AF_INET
)
1085 return udp_sendmsg(sk
, msg
, len
);
1087 /* Rough check on arithmetic overflow,
1088 better check is made in ip6_append_data().
1090 if (len
> INT_MAX
- sizeof(struct udphdr
))
1093 getfrag
= is_udplite
? udplite_getfrag
: ip_generic_getfrag
;
1096 * There are pending frames.
1097 * The socket lock must be held while it's corked.
1100 if (likely(up
->pending
)) {
1101 if (unlikely(up
->pending
!= AF_INET6
)) {
1103 return -EAFNOSUPPORT
;
1106 goto do_append_data
;
1110 ulen
+= sizeof(struct udphdr
);
1112 memset(&fl6
, 0, sizeof(fl6
));
1115 if (sin6
->sin6_port
== 0)
1118 fl6
.fl6_dport
= sin6
->sin6_port
;
1119 daddr
= &sin6
->sin6_addr
;
1122 fl6
.flowlabel
= sin6
->sin6_flowinfo
&IPV6_FLOWINFO_MASK
;
1123 if (fl6
.flowlabel
&IPV6_FLOWLABEL_MASK
) {
1124 flowlabel
= fl6_sock_lookup(sk
, fl6
.flowlabel
);
1131 * Otherwise it will be difficult to maintain
1134 if (sk
->sk_state
== TCP_ESTABLISHED
&&
1135 ipv6_addr_equal(daddr
, &sk
->sk_v6_daddr
))
1136 daddr
= &sk
->sk_v6_daddr
;
1138 if (addr_len
>= sizeof(struct sockaddr_in6
) &&
1139 sin6
->sin6_scope_id
&&
1140 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr
)))
1141 fl6
.flowi6_oif
= sin6
->sin6_scope_id
;
1143 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1144 return -EDESTADDRREQ
;
1146 fl6
.fl6_dport
= inet
->inet_dport
;
1147 daddr
= &sk
->sk_v6_daddr
;
1148 fl6
.flowlabel
= np
->flow_label
;
1152 if (!fl6
.flowi6_oif
)
1153 fl6
.flowi6_oif
= sk
->sk_bound_dev_if
;
1155 if (!fl6
.flowi6_oif
)
1156 fl6
.flowi6_oif
= np
->sticky_pktinfo
.ipi6_ifindex
;
1158 fl6
.flowi6_mark
= sk
->sk_mark
;
1159 sockc
.tsflags
= sk
->sk_tsflags
;
1161 if (msg
->msg_controllen
) {
1163 memset(opt
, 0, sizeof(struct ipv6_txoptions
));
1164 opt
->tot_len
= sizeof(*opt
);
1167 err
= ip6_datagram_send_ctl(sock_net(sk
), sk
, msg
, &fl6
, &ipc6
, &sockc
);
1169 fl6_sock_release(flowlabel
);
1172 if ((fl6
.flowlabel
&IPV6_FLOWLABEL_MASK
) && !flowlabel
) {
1173 flowlabel
= fl6_sock_lookup(sk
, fl6
.flowlabel
);
1177 if (!(opt
->opt_nflen
|opt
->opt_flen
))
1182 opt
= txopt_get(np
);
1186 opt
= fl6_merge_options(&opt_space
, flowlabel
, opt
);
1187 opt
= ipv6_fixup_options(&opt_space
, opt
);
1190 fl6
.flowi6_proto
= sk
->sk_protocol
;
1191 if (!ipv6_addr_any(daddr
))
1194 fl6
.daddr
.s6_addr
[15] = 0x1; /* :: means loopback (BSD'ism) */
1195 if (ipv6_addr_any(&fl6
.saddr
) && !ipv6_addr_any(&np
->saddr
))
1196 fl6
.saddr
= np
->saddr
;
1197 fl6
.fl6_sport
= inet
->inet_sport
;
1199 final_p
= fl6_update_dst(&fl6
, opt
, &final
);
1203 if (!fl6
.flowi6_oif
&& ipv6_addr_is_multicast(&fl6
.daddr
)) {
1204 fl6
.flowi6_oif
= np
->mcast_oif
;
1206 } else if (!fl6
.flowi6_oif
)
1207 fl6
.flowi6_oif
= np
->ucast_oif
;
1209 security_sk_classify_flow(sk
, flowi6_to_flowi(&fl6
));
1211 if (ipc6
.tclass
< 0)
1212 ipc6
.tclass
= np
->tclass
;
1214 fl6
.flowlabel
= ip6_make_flowinfo(ipc6
.tclass
, fl6
.flowlabel
);
1216 dst
= ip6_sk_dst_lookup_flow(sk
, &fl6
, final_p
);
1223 if (ipc6
.hlimit
< 0)
1224 ipc6
.hlimit
= ip6_sk_dst_hoplimit(np
, &fl6
, dst
);
1226 if (msg
->msg_flags
&MSG_CONFIRM
)
1230 /* Lockless fast path for the non-corking case */
1232 struct sk_buff
*skb
;
1234 skb
= ip6_make_skb(sk
, getfrag
, msg
, ulen
,
1235 sizeof(struct udphdr
), &ipc6
,
1236 &fl6
, (struct rt6_info
*)dst
,
1237 msg
->msg_flags
, &sockc
);
1239 if (!IS_ERR_OR_NULL(skb
))
1240 err
= udp_v6_send_skb(skb
, &fl6
);
1245 if (unlikely(up
->pending
)) {
1246 /* The socket is already corked while preparing it. */
1247 /* ... which is an evident application bug. --ANK */
1250 net_dbg_ratelimited("udp cork app bug 2\n");
1255 up
->pending
= AF_INET6
;
1258 if (ipc6
.dontfrag
< 0)
1259 ipc6
.dontfrag
= np
->dontfrag
;
1261 err
= ip6_append_data(sk
, getfrag
, msg
, ulen
, sizeof(struct udphdr
),
1262 &ipc6
, &fl6
, (struct rt6_info
*)dst
,
1263 corkreq
? msg
->msg_flags
|MSG_MORE
: msg
->msg_flags
, &sockc
);
1265 udp_v6_flush_pending_frames(sk
);
1267 err
= udp_v6_push_pending_frames(sk
);
1268 else if (unlikely(skb_queue_empty(&sk
->sk_write_queue
)))
1272 err
= np
->recverr
? net_xmit_errno(err
) : 0;
1278 ip6_dst_store(sk
, dst
,
1279 ipv6_addr_equal(&fl6
.daddr
, &sk
->sk_v6_daddr
) ?
1280 &sk
->sk_v6_daddr
: NULL
,
1281 #ifdef CONFIG_IPV6_SUBTREES
1282 ipv6_addr_equal(&fl6
.saddr
, &np
->saddr
) ?
1294 fl6_sock_release(flowlabel
);
1295 txopt_put(opt_to_free
);
1299 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1300 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1301 * we don't have a good statistic (IpOutDiscards but it can be too many
1302 * things). We could add another new stat but at least for now that
1303 * seems like overkill.
1305 if (err
== -ENOBUFS
|| test_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
)) {
1306 UDP6_INC_STATS(sock_net(sk
),
1307 UDP_MIB_SNDBUFERRORS
, is_udplite
);
1313 if (!(msg
->msg_flags
&MSG_PROBE
) || len
)
1314 goto back_from_confirm
;
1319 void udpv6_destroy_sock(struct sock
*sk
)
1321 struct udp_sock
*up
= udp_sk(sk
);
1323 udp_v6_flush_pending_frames(sk
);
1326 if (static_key_false(&udpv6_encap_needed
) && up
->encap_type
) {
1327 void (*encap_destroy
)(struct sock
*sk
);
1328 encap_destroy
= ACCESS_ONCE(up
->encap_destroy
);
1333 inet6_destroy_sock(sk
);
1337 * Socket option code for UDP
1339 int udpv6_setsockopt(struct sock
*sk
, int level
, int optname
,
1340 char __user
*optval
, unsigned int optlen
)
1342 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1343 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1344 udp_v6_push_pending_frames
);
1345 return ipv6_setsockopt(sk
, level
, optname
, optval
, optlen
);
1348 #ifdef CONFIG_COMPAT
1349 int compat_udpv6_setsockopt(struct sock
*sk
, int level
, int optname
,
1350 char __user
*optval
, unsigned int optlen
)
1352 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1353 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1354 udp_v6_push_pending_frames
);
1355 return compat_ipv6_setsockopt(sk
, level
, optname
, optval
, optlen
);
1359 int udpv6_getsockopt(struct sock
*sk
, int level
, int optname
,
1360 char __user
*optval
, int __user
*optlen
)
1362 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1363 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1364 return ipv6_getsockopt(sk
, level
, optname
, optval
, optlen
);
1367 #ifdef CONFIG_COMPAT
1368 int compat_udpv6_getsockopt(struct sock
*sk
, int level
, int optname
,
1369 char __user
*optval
, int __user
*optlen
)
1371 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1372 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1373 return compat_ipv6_getsockopt(sk
, level
, optname
, optval
, optlen
);
1377 static const struct inet6_protocol udpv6_protocol
= {
1378 .handler
= udpv6_rcv
,
1379 .err_handler
= udpv6_err
,
1380 .flags
= INET6_PROTO_NOPOLICY
|INET6_PROTO_FINAL
,
1383 /* ------------------------------------------------------------------------ */
1384 #ifdef CONFIG_PROC_FS
1385 int udp6_seq_show(struct seq_file
*seq
, void *v
)
1387 if (v
== SEQ_START_TOKEN
) {
1388 seq_puts(seq
, IPV6_SEQ_DGRAM_HEADER
);
1390 int bucket
= ((struct udp_iter_state
*)seq
->private)->bucket
;
1391 struct inet_sock
*inet
= inet_sk(v
);
1392 __u16 srcp
= ntohs(inet
->inet_sport
);
1393 __u16 destp
= ntohs(inet
->inet_dport
);
1394 ip6_dgram_sock_seq_show(seq
, v
, srcp
, destp
, bucket
);
1399 static const struct file_operations udp6_afinfo_seq_fops
= {
1400 .owner
= THIS_MODULE
,
1401 .open
= udp_seq_open
,
1403 .llseek
= seq_lseek
,
1404 .release
= seq_release_net
1407 static struct udp_seq_afinfo udp6_seq_afinfo
= {
1410 .udp_table
= &udp_table
,
1411 .seq_fops
= &udp6_afinfo_seq_fops
,
1413 .show
= udp6_seq_show
,
1417 int __net_init
udp6_proc_init(struct net
*net
)
1419 return udp_proc_register(net
, &udp6_seq_afinfo
);
1422 void udp6_proc_exit(struct net
*net
)
1424 udp_proc_unregister(net
, &udp6_seq_afinfo
);
1426 #endif /* CONFIG_PROC_FS */
1428 /* ------------------------------------------------------------------------ */
1430 struct proto udpv6_prot
= {
1432 .owner
= THIS_MODULE
,
1433 .close
= udp_lib_close
,
1434 .connect
= ip6_datagram_connect
,
1435 .disconnect
= udp_disconnect
,
1437 .destroy
= udpv6_destroy_sock
,
1438 .setsockopt
= udpv6_setsockopt
,
1439 .getsockopt
= udpv6_getsockopt
,
1440 .sendmsg
= udpv6_sendmsg
,
1441 .recvmsg
= udpv6_recvmsg
,
1442 .backlog_rcv
= __udpv6_queue_rcv_skb
,
1443 .release_cb
= ip6_datagram_release_cb
,
1444 .hash
= udp_lib_hash
,
1445 .unhash
= udp_lib_unhash
,
1446 .rehash
= udp_v6_rehash
,
1447 .get_port
= udp_v6_get_port
,
1448 .memory_allocated
= &udp_memory_allocated
,
1449 .sysctl_mem
= sysctl_udp_mem
,
1450 .sysctl_wmem
= &sysctl_udp_wmem_min
,
1451 .sysctl_rmem
= &sysctl_udp_rmem_min
,
1452 .obj_size
= sizeof(struct udp6_sock
),
1453 .h
.udp_table
= &udp_table
,
1454 #ifdef CONFIG_COMPAT
1455 .compat_setsockopt
= compat_udpv6_setsockopt
,
1456 .compat_getsockopt
= compat_udpv6_getsockopt
,
1458 .diag_destroy
= udp_abort
,
1461 static struct inet_protosw udpv6_protosw
= {
1463 .protocol
= IPPROTO_UDP
,
1464 .prot
= &udpv6_prot
,
1465 .ops
= &inet6_dgram_ops
,
1466 .flags
= INET_PROTOSW_PERMANENT
,
1469 int __init
udpv6_init(void)
1473 ret
= inet6_add_protocol(&udpv6_protocol
, IPPROTO_UDP
);
1477 ret
= inet6_register_protosw(&udpv6_protosw
);
1479 goto out_udpv6_protocol
;
1484 inet6_del_protocol(&udpv6_protocol
, IPPROTO_UDP
);
1488 void udpv6_exit(void)
1490 inet6_unregister_protosw(&udpv6_protosw
);
1491 inet6_del_protocol(&udpv6_protocol
, IPPROTO_UDP
);