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 <linux/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 IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
307 struct sock
*udp6_lib_lookup(struct net
*net
, const struct in6_addr
*saddr
, __be16 sport
,
308 const struct in6_addr
*daddr
, __be16 dport
, int dif
)
312 sk
= __udp6_lib_lookup(net
, saddr
, sport
, daddr
, dport
,
313 dif
, &udp_table
, NULL
);
314 if (sk
&& !atomic_inc_not_zero(&sk
->sk_refcnt
))
318 EXPORT_SYMBOL_GPL(udp6_lib_lookup
);
322 * This should be easy, if there is something there we
323 * return it, otherwise we block.
326 int udpv6_recvmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
,
327 int noblock
, int flags
, int *addr_len
)
329 struct ipv6_pinfo
*np
= inet6_sk(sk
);
330 struct inet_sock
*inet
= inet_sk(sk
);
332 unsigned int ulen
, copied
;
333 int peeked
, peeking
, off
;
335 int is_udplite
= IS_UDPLITE(sk
);
336 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_udp(sk
, flags
, noblock
, &peeked
, &off
, &err
);
353 if (copied
> ulen
- off
)
355 else if (copied
< ulen
)
356 msg
->msg_flags
|= MSG_TRUNC
;
358 is_udp4
= (skb
->protocol
== htons(ETH_P_IP
));
361 * If checksum is needed at all, try to do it while copying the
362 * data. If the data is truncated, or if we only want a partial
363 * coverage checksum (UDP-Lite), do it before the copy.
366 if (copied
< ulen
|| peeking
||
367 (is_udplite
&& UDP_SKB_CB(skb
)->partial_cov
)) {
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
);
382 atomic_inc(&sk
->sk_drops
);
384 UDP_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
,
387 UDP6_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
,
395 UDP_INC_STATS(sock_net(sk
), UDP_MIB_INDATAGRAMS
,
398 UDP6_INC_STATS(sock_net(sk
), UDP_MIB_INDATAGRAMS
,
402 sock_recv_ts_and_drops(msg
, sk
, skb
);
404 /* Copy the address. */
406 DECLARE_SOCKADDR(struct sockaddr_in6
*, sin6
, msg
->msg_name
);
407 sin6
->sin6_family
= AF_INET6
;
408 sin6
->sin6_port
= udp_hdr(skb
)->source
;
409 sin6
->sin6_flowinfo
= 0;
412 ipv6_addr_set_v4mapped(ip_hdr(skb
)->saddr
,
414 sin6
->sin6_scope_id
= 0;
416 sin6
->sin6_addr
= ipv6_hdr(skb
)->saddr
;
417 sin6
->sin6_scope_id
=
418 ipv6_iface_scope_id(&sin6
->sin6_addr
,
421 *addr_len
= sizeof(*sin6
);
425 ip6_datagram_recv_common_ctl(sk
, msg
, skb
);
428 if (inet
->cmsg_flags
)
429 ip_cmsg_recv_offset(msg
, sk
, skb
,
430 sizeof(struct udphdr
), off
);
433 ip6_datagram_recv_specific_ctl(sk
, msg
, skb
);
437 if (flags
& MSG_TRUNC
)
440 skb_consume_udp(sk
, skb
, peeking
? -err
: err
);
444 if (!__sk_queue_drop_skb(sk
, skb
, flags
)) {
446 UDP_INC_STATS(sock_net(sk
),
447 UDP_MIB_CSUMERRORS
, is_udplite
);
448 UDP_INC_STATS(sock_net(sk
),
449 UDP_MIB_INERRORS
, is_udplite
);
451 UDP6_INC_STATS(sock_net(sk
),
452 UDP_MIB_CSUMERRORS
, is_udplite
);
453 UDP6_INC_STATS(sock_net(sk
),
454 UDP_MIB_INERRORS
, is_udplite
);
459 /* starting over for a new packet, but check if we need to yield */
461 msg
->msg_flags
&= ~MSG_TRUNC
;
465 void __udp6_lib_err(struct sk_buff
*skb
, struct inet6_skb_parm
*opt
,
466 u8 type
, u8 code
, int offset
, __be32 info
,
467 struct udp_table
*udptable
)
469 struct ipv6_pinfo
*np
;
470 const struct ipv6hdr
*hdr
= (const struct ipv6hdr
*)skb
->data
;
471 const struct in6_addr
*saddr
= &hdr
->saddr
;
472 const struct in6_addr
*daddr
= &hdr
->daddr
;
473 struct udphdr
*uh
= (struct udphdr
*)(skb
->data
+offset
);
477 struct net
*net
= dev_net(skb
->dev
);
479 sk
= __udp6_lib_lookup(net
, daddr
, uh
->dest
, saddr
, uh
->source
,
480 inet6_iif(skb
), udptable
, skb
);
482 __ICMP6_INC_STATS(net
, __in6_dev_get(skb
->dev
),
487 harderr
= icmpv6_err_convert(type
, code
, &err
);
490 if (type
== ICMPV6_PKT_TOOBIG
) {
491 if (!ip6_sk_accept_pmtu(sk
))
493 ip6_sk_update_pmtu(skb
, sk
, info
);
494 if (np
->pmtudisc
!= IPV6_PMTUDISC_DONT
)
497 if (type
== NDISC_REDIRECT
) {
498 ip6_sk_redirect(skb
, sk
);
503 if (!harderr
|| sk
->sk_state
!= TCP_ESTABLISHED
)
506 ipv6_icmp_error(sk
, skb
, err
, uh
->dest
, ntohl(info
), (u8
*)(uh
+1));
510 sk
->sk_error_report(sk
);
515 int __udpv6_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
519 if (!ipv6_addr_any(&sk
->sk_v6_daddr
)) {
520 sock_rps_save_rxhash(sk
, skb
);
521 sk_mark_napi_id(sk
, skb
);
522 sk_incoming_cpu_update(sk
);
524 sk_mark_napi_id_once(sk
, skb
);
527 rc
= __udp_enqueue_schedule_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
);
543 static __inline__
void udpv6_err(struct sk_buff
*skb
,
544 struct inet6_skb_parm
*opt
, u8 type
,
545 u8 code
, int offset
, __be32 info
)
547 __udp6_lib_err(skb
, opt
, type
, code
, offset
, info
, &udp_table
);
550 static struct static_key udpv6_encap_needed __read_mostly
;
551 void udpv6_encap_enable(void)
553 if (!static_key_enabled(&udpv6_encap_needed
))
554 static_key_slow_inc(&udpv6_encap_needed
);
556 EXPORT_SYMBOL(udpv6_encap_enable
);
558 int udpv6_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
560 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
);
629 return __udpv6_queue_rcv_skb(sk
, skb
);
632 __UDP6_INC_STATS(sock_net(sk
), UDP_MIB_CSUMERRORS
, is_udplite
);
634 __UDP6_INC_STATS(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
635 atomic_inc(&sk
->sk_drops
);
640 static bool __udp_v6_is_mcast_sock(struct net
*net
, struct sock
*sk
,
641 __be16 loc_port
, const struct in6_addr
*loc_addr
,
642 __be16 rmt_port
, const struct in6_addr
*rmt_addr
,
643 int dif
, unsigned short hnum
)
645 struct inet_sock
*inet
= inet_sk(sk
);
647 if (!net_eq(sock_net(sk
), net
))
650 if (udp_sk(sk
)->udp_port_hash
!= hnum
||
651 sk
->sk_family
!= PF_INET6
||
652 (inet
->inet_dport
&& inet
->inet_dport
!= rmt_port
) ||
653 (!ipv6_addr_any(&sk
->sk_v6_daddr
) &&
654 !ipv6_addr_equal(&sk
->sk_v6_daddr
, rmt_addr
)) ||
655 (sk
->sk_bound_dev_if
&& sk
->sk_bound_dev_if
!= dif
) ||
656 (!ipv6_addr_any(&sk
->sk_v6_rcv_saddr
) &&
657 !ipv6_addr_equal(&sk
->sk_v6_rcv_saddr
, loc_addr
)))
659 if (!inet6_mc_check(sk
, loc_addr
, rmt_addr
))
664 static void udp6_csum_zero_error(struct sk_buff
*skb
)
666 /* RFC 2460 section 8.1 says that we SHOULD log
667 * this error. Well, it is reasonable.
669 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
670 &ipv6_hdr(skb
)->saddr
, ntohs(udp_hdr(skb
)->source
),
671 &ipv6_hdr(skb
)->daddr
, ntohs(udp_hdr(skb
)->dest
));
675 * Note: called only from the BH handler context,
676 * so we don't need to lock the hashes.
678 static int __udp6_lib_mcast_deliver(struct net
*net
, struct sk_buff
*skb
,
679 const struct in6_addr
*saddr
, const struct in6_addr
*daddr
,
680 struct udp_table
*udptable
, int proto
)
682 struct sock
*sk
, *first
= NULL
;
683 const struct udphdr
*uh
= udp_hdr(skb
);
684 unsigned short hnum
= ntohs(uh
->dest
);
685 struct udp_hslot
*hslot
= udp_hashslot(udptable
, net
, hnum
);
686 unsigned int offset
= offsetof(typeof(*sk
), sk_node
);
687 unsigned int hash2
= 0, hash2_any
= 0, use_hash2
= (hslot
->count
> 10);
688 int dif
= inet6_iif(skb
);
689 struct hlist_node
*node
;
690 struct sk_buff
*nskb
;
693 hash2_any
= udp6_portaddr_hash(net
, &in6addr_any
, hnum
) &
695 hash2
= udp6_portaddr_hash(net
, daddr
, hnum
) & udptable
->mask
;
697 hslot
= &udptable
->hash2
[hash2
];
698 offset
= offsetof(typeof(*sk
), __sk_common
.skc_portaddr_node
);
701 sk_for_each_entry_offset_rcu(sk
, node
, &hslot
->head
, offset
) {
702 if (!__udp_v6_is_mcast_sock(net
, sk
, uh
->dest
, daddr
,
703 uh
->source
, saddr
, dif
, hnum
))
705 /* If zero checksum and no_check is not on for
706 * the socket then skip it.
708 if (!uh
->check
&& !udp_sk(sk
)->no_check6_rx
)
714 nskb
= skb_clone(skb
, GFP_ATOMIC
);
715 if (unlikely(!nskb
)) {
716 atomic_inc(&sk
->sk_drops
);
717 __UDP6_INC_STATS(net
, UDP_MIB_RCVBUFERRORS
,
719 __UDP6_INC_STATS(net
, UDP_MIB_INERRORS
,
724 if (udpv6_queue_rcv_skb(sk
, nskb
) > 0)
728 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
729 if (use_hash2
&& hash2
!= hash2_any
) {
735 if (udpv6_queue_rcv_skb(first
, skb
) > 0)
739 __UDP6_INC_STATS(net
, UDP_MIB_IGNOREDMULTI
,
740 proto
== IPPROTO_UDPLITE
);
745 int __udp6_lib_rcv(struct sk_buff
*skb
, struct udp_table
*udptable
,
748 const struct in6_addr
*saddr
, *daddr
;
749 struct net
*net
= dev_net(skb
->dev
);
754 if (!pskb_may_pull(skb
, sizeof(struct udphdr
)))
757 saddr
= &ipv6_hdr(skb
)->saddr
;
758 daddr
= &ipv6_hdr(skb
)->daddr
;
761 ulen
= ntohs(uh
->len
);
765 if (proto
== IPPROTO_UDP
) {
766 /* UDP validates ulen. */
768 /* Check for jumbo payload */
772 if (ulen
< sizeof(*uh
))
775 if (ulen
< skb
->len
) {
776 if (pskb_trim_rcsum(skb
, ulen
))
778 saddr
= &ipv6_hdr(skb
)->saddr
;
779 daddr
= &ipv6_hdr(skb
)->daddr
;
784 if (udp6_csum_init(skb
, uh
, proto
))
788 * Multicast receive code
790 if (ipv6_addr_is_multicast(daddr
))
791 return __udp6_lib_mcast_deliver(net
, skb
,
792 saddr
, daddr
, udptable
, proto
);
797 * check socket cache ... must talk to Alan about his plans
798 * for sock caches... i'll skip this for now.
800 sk
= __udp6_lib_lookup_skb(skb
, uh
->source
, uh
->dest
, udptable
);
804 if (!uh
->check
&& !udp_sk(sk
)->no_check6_rx
) {
805 udp6_csum_zero_error(skb
);
809 if (inet_get_convert_csum(sk
) && uh
->check
&& !IS_UDPLITE(sk
))
810 skb_checksum_try_convert(skb
, IPPROTO_UDP
, uh
->check
,
813 ret
= udpv6_queue_rcv_skb(sk
, skb
);
815 /* a return value > 0 means to resubmit the input */
823 udp6_csum_zero_error(skb
);
827 if (!xfrm6_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
830 if (udp_lib_checksum_complete(skb
))
833 __UDP6_INC_STATS(net
, UDP_MIB_NOPORTS
, proto
== IPPROTO_UDPLITE
);
834 icmpv6_send(skb
, ICMPV6_DEST_UNREACH
, ICMPV6_PORT_UNREACH
, 0);
840 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
841 proto
== IPPROTO_UDPLITE
? "-Lite" : "",
842 saddr
, ntohs(uh
->source
),
844 daddr
, ntohs(uh
->dest
));
847 __UDP6_INC_STATS(net
, UDP_MIB_CSUMERRORS
, proto
== IPPROTO_UDPLITE
);
849 __UDP6_INC_STATS(net
, UDP_MIB_INERRORS
, proto
== IPPROTO_UDPLITE
);
854 static __inline__
int udpv6_rcv(struct sk_buff
*skb
)
856 return __udp6_lib_rcv(skb
, &udp_table
, IPPROTO_UDP
);
860 * Throw away all pending data and cancel the corking. Socket is locked.
862 static void udp_v6_flush_pending_frames(struct sock
*sk
)
864 struct udp_sock
*up
= udp_sk(sk
);
866 if (up
->pending
== AF_INET
)
867 udp_flush_pending_frames(sk
);
868 else if (up
->pending
) {
871 ip6_flush_pending_frames(sk
);
876 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
877 * @sk: socket we are sending on
878 * @skb: sk_buff containing the filled-in UDP header
879 * (checksum field must be zeroed out)
881 static void udp6_hwcsum_outgoing(struct sock
*sk
, struct sk_buff
*skb
,
882 const struct in6_addr
*saddr
,
883 const struct in6_addr
*daddr
, int len
)
886 struct udphdr
*uh
= udp_hdr(skb
);
887 struct sk_buff
*frags
= skb_shinfo(skb
)->frag_list
;
891 /* Only one fragment on the socket. */
892 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
893 skb
->csum_offset
= offsetof(struct udphdr
, check
);
894 uh
->check
= ~csum_ipv6_magic(saddr
, daddr
, len
, IPPROTO_UDP
, 0);
897 * HW-checksum won't work as there are two or more
898 * fragments on the socket so that all csums of sk_buffs
901 offset
= skb_transport_offset(skb
);
902 skb
->csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
904 skb
->ip_summed
= CHECKSUM_NONE
;
907 csum
= csum_add(csum
, frags
->csum
);
908 } while ((frags
= frags
->next
));
910 uh
->check
= csum_ipv6_magic(saddr
, daddr
, len
, IPPROTO_UDP
,
913 uh
->check
= CSUM_MANGLED_0
;
921 static int udp_v6_send_skb(struct sk_buff
*skb
, struct flowi6
*fl6
)
923 struct sock
*sk
= skb
->sk
;
926 int is_udplite
= IS_UDPLITE(sk
);
928 int offset
= skb_transport_offset(skb
);
929 int len
= skb
->len
- offset
;
932 * Create a UDP header
935 uh
->source
= fl6
->fl6_sport
;
936 uh
->dest
= fl6
->fl6_dport
;
937 uh
->len
= htons(len
);
941 csum
= udplite_csum(skb
);
942 else if (udp_sk(sk
)->no_check6_tx
) { /* UDP csum disabled */
943 skb
->ip_summed
= CHECKSUM_NONE
;
945 } else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) { /* UDP hardware csum */
946 udp6_hwcsum_outgoing(sk
, skb
, &fl6
->saddr
, &fl6
->daddr
, len
);
949 csum
= udp_csum(skb
);
951 /* add protocol-dependent pseudo-header */
952 uh
->check
= csum_ipv6_magic(&fl6
->saddr
, &fl6
->daddr
,
953 len
, fl6
->flowi6_proto
, csum
);
955 uh
->check
= CSUM_MANGLED_0
;
958 err
= ip6_send_skb(skb
);
960 if (err
== -ENOBUFS
&& !inet6_sk(sk
)->recverr
) {
961 UDP6_INC_STATS(sock_net(sk
),
962 UDP_MIB_SNDBUFERRORS
, is_udplite
);
966 UDP6_INC_STATS(sock_net(sk
),
967 UDP_MIB_OUTDATAGRAMS
, is_udplite
);
972 static int udp_v6_push_pending_frames(struct sock
*sk
)
975 struct udp_sock
*up
= udp_sk(sk
);
979 if (up
->pending
== AF_INET
)
980 return udp_push_pending_frames(sk
);
982 /* ip6_finish_skb will release the cork, so make a copy of
985 fl6
= inet_sk(sk
)->cork
.fl
.u
.ip6
;
987 skb
= ip6_finish_skb(sk
);
991 err
= udp_v6_send_skb(skb
, &fl6
);
999 int udpv6_sendmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
)
1001 struct ipv6_txoptions opt_space
;
1002 struct udp_sock
*up
= udp_sk(sk
);
1003 struct inet_sock
*inet
= inet_sk(sk
);
1004 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1005 DECLARE_SOCKADDR(struct sockaddr_in6
*, sin6
, msg
->msg_name
);
1006 struct in6_addr
*daddr
, *final_p
, final
;
1007 struct ipv6_txoptions
*opt
= NULL
;
1008 struct ipv6_txoptions
*opt_to_free
= NULL
;
1009 struct ip6_flowlabel
*flowlabel
= NULL
;
1011 struct dst_entry
*dst
;
1012 struct ipcm6_cookie ipc6
;
1013 int addr_len
= msg
->msg_namelen
;
1015 int corkreq
= up
->corkflag
|| msg
->msg_flags
&MSG_MORE
;
1018 int is_udplite
= IS_UDPLITE(sk
);
1019 int (*getfrag
)(void *, char *, int, int, int, struct sk_buff
*);
1020 struct sockcm_cookie sockc
;
1026 /* destination address check */
1028 if (addr_len
< offsetof(struct sockaddr
, sa_data
))
1031 switch (sin6
->sin6_family
) {
1033 if (addr_len
< SIN6_LEN_RFC2133
)
1035 daddr
= &sin6
->sin6_addr
;
1038 goto do_udp_sendmsg
;
1040 msg
->msg_name
= sin6
= NULL
;
1041 msg
->msg_namelen
= addr_len
= 0;
1047 } else if (!up
->pending
) {
1048 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1049 return -EDESTADDRREQ
;
1050 daddr
= &sk
->sk_v6_daddr
;
1055 if (ipv6_addr_v4mapped(daddr
)) {
1056 struct sockaddr_in sin
;
1057 sin
.sin_family
= AF_INET
;
1058 sin
.sin_port
= sin6
? sin6
->sin6_port
: inet
->inet_dport
;
1059 sin
.sin_addr
.s_addr
= daddr
->s6_addr32
[3];
1060 msg
->msg_name
= &sin
;
1061 msg
->msg_namelen
= sizeof(sin
);
1063 if (__ipv6_only_sock(sk
))
1064 return -ENETUNREACH
;
1065 return udp_sendmsg(sk
, msg
, len
);
1069 if (up
->pending
== AF_INET
)
1070 return udp_sendmsg(sk
, msg
, len
);
1072 /* Rough check on arithmetic overflow,
1073 better check is made in ip6_append_data().
1075 if (len
> INT_MAX
- sizeof(struct udphdr
))
1078 getfrag
= is_udplite
? udplite_getfrag
: ip_generic_getfrag
;
1081 * There are pending frames.
1082 * The socket lock must be held while it's corked.
1085 if (likely(up
->pending
)) {
1086 if (unlikely(up
->pending
!= AF_INET6
)) {
1088 return -EAFNOSUPPORT
;
1091 goto do_append_data
;
1095 ulen
+= sizeof(struct udphdr
);
1097 memset(&fl6
, 0, sizeof(fl6
));
1100 if (sin6
->sin6_port
== 0)
1103 fl6
.fl6_dport
= sin6
->sin6_port
;
1104 daddr
= &sin6
->sin6_addr
;
1107 fl6
.flowlabel
= sin6
->sin6_flowinfo
&IPV6_FLOWINFO_MASK
;
1108 if (fl6
.flowlabel
&IPV6_FLOWLABEL_MASK
) {
1109 flowlabel
= fl6_sock_lookup(sk
, fl6
.flowlabel
);
1116 * Otherwise it will be difficult to maintain
1119 if (sk
->sk_state
== TCP_ESTABLISHED
&&
1120 ipv6_addr_equal(daddr
, &sk
->sk_v6_daddr
))
1121 daddr
= &sk
->sk_v6_daddr
;
1123 if (addr_len
>= sizeof(struct sockaddr_in6
) &&
1124 sin6
->sin6_scope_id
&&
1125 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr
)))
1126 fl6
.flowi6_oif
= sin6
->sin6_scope_id
;
1128 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1129 return -EDESTADDRREQ
;
1131 fl6
.fl6_dport
= inet
->inet_dport
;
1132 daddr
= &sk
->sk_v6_daddr
;
1133 fl6
.flowlabel
= np
->flow_label
;
1137 if (!fl6
.flowi6_oif
)
1138 fl6
.flowi6_oif
= sk
->sk_bound_dev_if
;
1140 if (!fl6
.flowi6_oif
)
1141 fl6
.flowi6_oif
= np
->sticky_pktinfo
.ipi6_ifindex
;
1143 fl6
.flowi6_mark
= sk
->sk_mark
;
1144 fl6
.flowi6_uid
= sk
->sk_uid
;
1145 sockc
.tsflags
= sk
->sk_tsflags
;
1147 if (msg
->msg_controllen
) {
1149 memset(opt
, 0, sizeof(struct ipv6_txoptions
));
1150 opt
->tot_len
= sizeof(*opt
);
1153 err
= ip6_datagram_send_ctl(sock_net(sk
), sk
, msg
, &fl6
, &ipc6
, &sockc
);
1155 fl6_sock_release(flowlabel
);
1158 if ((fl6
.flowlabel
&IPV6_FLOWLABEL_MASK
) && !flowlabel
) {
1159 flowlabel
= fl6_sock_lookup(sk
, fl6
.flowlabel
);
1163 if (!(opt
->opt_nflen
|opt
->opt_flen
))
1168 opt
= txopt_get(np
);
1172 opt
= fl6_merge_options(&opt_space
, flowlabel
, opt
);
1173 opt
= ipv6_fixup_options(&opt_space
, opt
);
1176 fl6
.flowi6_proto
= sk
->sk_protocol
;
1177 if (!ipv6_addr_any(daddr
))
1180 fl6
.daddr
.s6_addr
[15] = 0x1; /* :: means loopback (BSD'ism) */
1181 if (ipv6_addr_any(&fl6
.saddr
) && !ipv6_addr_any(&np
->saddr
))
1182 fl6
.saddr
= np
->saddr
;
1183 fl6
.fl6_sport
= inet
->inet_sport
;
1185 final_p
= fl6_update_dst(&fl6
, opt
, &final
);
1189 if (!fl6
.flowi6_oif
&& ipv6_addr_is_multicast(&fl6
.daddr
)) {
1190 fl6
.flowi6_oif
= np
->mcast_oif
;
1192 } else if (!fl6
.flowi6_oif
)
1193 fl6
.flowi6_oif
= np
->ucast_oif
;
1195 security_sk_classify_flow(sk
, flowi6_to_flowi(&fl6
));
1197 if (ipc6
.tclass
< 0)
1198 ipc6
.tclass
= np
->tclass
;
1200 fl6
.flowlabel
= ip6_make_flowinfo(ipc6
.tclass
, fl6
.flowlabel
);
1202 dst
= ip6_sk_dst_lookup_flow(sk
, &fl6
, final_p
);
1209 if (ipc6
.hlimit
< 0)
1210 ipc6
.hlimit
= ip6_sk_dst_hoplimit(np
, &fl6
, dst
);
1212 if (msg
->msg_flags
&MSG_CONFIRM
)
1216 /* Lockless fast path for the non-corking case */
1218 struct sk_buff
*skb
;
1220 skb
= ip6_make_skb(sk
, getfrag
, msg
, ulen
,
1221 sizeof(struct udphdr
), &ipc6
,
1222 &fl6
, (struct rt6_info
*)dst
,
1223 msg
->msg_flags
, &sockc
);
1225 if (!IS_ERR_OR_NULL(skb
))
1226 err
= udp_v6_send_skb(skb
, &fl6
);
1231 if (unlikely(up
->pending
)) {
1232 /* The socket is already corked while preparing it. */
1233 /* ... which is an evident application bug. --ANK */
1236 net_dbg_ratelimited("udp cork app bug 2\n");
1241 up
->pending
= AF_INET6
;
1244 if (ipc6
.dontfrag
< 0)
1245 ipc6
.dontfrag
= np
->dontfrag
;
1247 err
= ip6_append_data(sk
, getfrag
, msg
, ulen
, sizeof(struct udphdr
),
1248 &ipc6
, &fl6
, (struct rt6_info
*)dst
,
1249 corkreq
? msg
->msg_flags
|MSG_MORE
: msg
->msg_flags
, &sockc
);
1251 udp_v6_flush_pending_frames(sk
);
1253 err
= udp_v6_push_pending_frames(sk
);
1254 else if (unlikely(skb_queue_empty(&sk
->sk_write_queue
)))
1258 err
= np
->recverr
? net_xmit_errno(err
) : 0;
1264 ip6_dst_store(sk
, dst
,
1265 ipv6_addr_equal(&fl6
.daddr
, &sk
->sk_v6_daddr
) ?
1266 &sk
->sk_v6_daddr
: NULL
,
1267 #ifdef CONFIG_IPV6_SUBTREES
1268 ipv6_addr_equal(&fl6
.saddr
, &np
->saddr
) ?
1280 fl6_sock_release(flowlabel
);
1281 txopt_put(opt_to_free
);
1285 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1286 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1287 * we don't have a good statistic (IpOutDiscards but it can be too many
1288 * things). We could add another new stat but at least for now that
1289 * seems like overkill.
1291 if (err
== -ENOBUFS
|| test_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
)) {
1292 UDP6_INC_STATS(sock_net(sk
),
1293 UDP_MIB_SNDBUFERRORS
, is_udplite
);
1299 if (!(msg
->msg_flags
&MSG_PROBE
) || len
)
1300 goto back_from_confirm
;
1305 void udpv6_destroy_sock(struct sock
*sk
)
1307 struct udp_sock
*up
= udp_sk(sk
);
1309 udp_v6_flush_pending_frames(sk
);
1312 if (static_key_false(&udpv6_encap_needed
) && up
->encap_type
) {
1313 void (*encap_destroy
)(struct sock
*sk
);
1314 encap_destroy
= ACCESS_ONCE(up
->encap_destroy
);
1319 inet6_destroy_sock(sk
);
1323 * Socket option code for UDP
1325 int udpv6_setsockopt(struct sock
*sk
, int level
, int optname
,
1326 char __user
*optval
, unsigned int optlen
)
1328 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1329 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1330 udp_v6_push_pending_frames
);
1331 return ipv6_setsockopt(sk
, level
, optname
, optval
, optlen
);
1334 #ifdef CONFIG_COMPAT
1335 int compat_udpv6_setsockopt(struct sock
*sk
, int level
, int optname
,
1336 char __user
*optval
, unsigned int optlen
)
1338 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1339 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1340 udp_v6_push_pending_frames
);
1341 return compat_ipv6_setsockopt(sk
, level
, optname
, optval
, optlen
);
1345 int udpv6_getsockopt(struct sock
*sk
, int level
, int optname
,
1346 char __user
*optval
, int __user
*optlen
)
1348 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1349 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1350 return ipv6_getsockopt(sk
, level
, optname
, optval
, optlen
);
1353 #ifdef CONFIG_COMPAT
1354 int compat_udpv6_getsockopt(struct sock
*sk
, int level
, int optname
,
1355 char __user
*optval
, int __user
*optlen
)
1357 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1358 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1359 return compat_ipv6_getsockopt(sk
, level
, optname
, optval
, optlen
);
1363 static const struct inet6_protocol udpv6_protocol
= {
1364 .handler
= udpv6_rcv
,
1365 .err_handler
= udpv6_err
,
1366 .flags
= INET6_PROTO_NOPOLICY
|INET6_PROTO_FINAL
,
1369 /* ------------------------------------------------------------------------ */
1370 #ifdef CONFIG_PROC_FS
1371 int udp6_seq_show(struct seq_file
*seq
, void *v
)
1373 if (v
== SEQ_START_TOKEN
) {
1374 seq_puts(seq
, IPV6_SEQ_DGRAM_HEADER
);
1376 int bucket
= ((struct udp_iter_state
*)seq
->private)->bucket
;
1377 struct inet_sock
*inet
= inet_sk(v
);
1378 __u16 srcp
= ntohs(inet
->inet_sport
);
1379 __u16 destp
= ntohs(inet
->inet_dport
);
1380 ip6_dgram_sock_seq_show(seq
, v
, srcp
, destp
, bucket
);
1385 static const struct file_operations udp6_afinfo_seq_fops
= {
1386 .owner
= THIS_MODULE
,
1387 .open
= udp_seq_open
,
1389 .llseek
= seq_lseek
,
1390 .release
= seq_release_net
1393 static struct udp_seq_afinfo udp6_seq_afinfo
= {
1396 .udp_table
= &udp_table
,
1397 .seq_fops
= &udp6_afinfo_seq_fops
,
1399 .show
= udp6_seq_show
,
1403 int __net_init
udp6_proc_init(struct net
*net
)
1405 return udp_proc_register(net
, &udp6_seq_afinfo
);
1408 void udp6_proc_exit(struct net
*net
)
1410 udp_proc_unregister(net
, &udp6_seq_afinfo
);
1412 #endif /* CONFIG_PROC_FS */
1414 /* ------------------------------------------------------------------------ */
1416 struct proto udpv6_prot
= {
1418 .owner
= THIS_MODULE
,
1419 .close
= udp_lib_close
,
1420 .connect
= ip6_datagram_connect
,
1421 .disconnect
= udp_disconnect
,
1423 .init
= udp_init_sock
,
1424 .destroy
= udpv6_destroy_sock
,
1425 .setsockopt
= udpv6_setsockopt
,
1426 .getsockopt
= udpv6_getsockopt
,
1427 .sendmsg
= udpv6_sendmsg
,
1428 .recvmsg
= udpv6_recvmsg
,
1429 .release_cb
= ip6_datagram_release_cb
,
1430 .hash
= udp_lib_hash
,
1431 .unhash
= udp_lib_unhash
,
1432 .rehash
= udp_v6_rehash
,
1433 .get_port
= udp_v6_get_port
,
1434 .memory_allocated
= &udp_memory_allocated
,
1435 .sysctl_mem
= sysctl_udp_mem
,
1436 .sysctl_wmem
= &sysctl_udp_wmem_min
,
1437 .sysctl_rmem
= &sysctl_udp_rmem_min
,
1438 .obj_size
= sizeof(struct udp6_sock
),
1439 .h
.udp_table
= &udp_table
,
1440 #ifdef CONFIG_COMPAT
1441 .compat_setsockopt
= compat_udpv6_setsockopt
,
1442 .compat_getsockopt
= compat_udpv6_getsockopt
,
1444 .diag_destroy
= udp_abort
,
1447 static struct inet_protosw udpv6_protosw
= {
1449 .protocol
= IPPROTO_UDP
,
1450 .prot
= &udpv6_prot
,
1451 .ops
= &inet6_dgram_ops
,
1452 .flags
= INET_PROTOSW_PERMANENT
,
1455 int __init
udpv6_init(void)
1459 ret
= inet6_add_protocol(&udpv6_protocol
, IPPROTO_UDP
);
1463 ret
= inet6_register_protosw(&udpv6_protosw
);
1465 goto out_udpv6_protocol
;
1470 inet6_del_protocol(&udpv6_protocol
, IPPROTO_UDP
);
1474 void udpv6_exit(void)
1476 inet6_unregister_protosw(&udpv6_protosw
);
1477 inet6_del_protocol(&udpv6_protocol
, IPPROTO_UDP
);