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 const u32 lhash
= (__force u32
)laddr
->s6_addr32
[3];
63 const u32 fhash
= __ipv6_addr_jhash(faddr
, ipv6_hash_secret
);
64 return __inet6_ehashfn(lhash
, lport
, fhash
, fport
,
65 inet_ehash_secret
+ net_hash_mix(net
));
68 int ipv6_rcv_saddr_equal(const struct sock
*sk
, const struct sock
*sk2
)
70 const struct in6_addr
*sk2_rcv_saddr6
= inet6_rcv_saddr(sk2
);
71 int sk_ipv6only
= ipv6_only_sock(sk
);
72 int sk2_ipv6only
= inet_v6_ipv6only(sk2
);
73 int addr_type
= ipv6_addr_type(&sk
->sk_v6_rcv_saddr
);
74 int addr_type2
= sk2_rcv_saddr6
? ipv6_addr_type(sk2_rcv_saddr6
) : IPV6_ADDR_MAPPED
;
76 /* if both are mapped, treat as IPv4 */
77 if (addr_type
== IPV6_ADDR_MAPPED
&& addr_type2
== IPV6_ADDR_MAPPED
)
78 return (!sk2_ipv6only
&&
79 (!sk
->sk_rcv_saddr
|| !sk2
->sk_rcv_saddr
||
80 sk
->sk_rcv_saddr
== sk2
->sk_rcv_saddr
));
82 if (addr_type2
== IPV6_ADDR_ANY
&&
83 !(sk2_ipv6only
&& addr_type
== IPV6_ADDR_MAPPED
))
86 if (addr_type
== IPV6_ADDR_ANY
&&
87 !(sk_ipv6only
&& addr_type2
== IPV6_ADDR_MAPPED
))
91 ipv6_addr_equal(&sk
->sk_v6_rcv_saddr
, sk2_rcv_saddr6
))
97 static unsigned int udp6_portaddr_hash(struct net
*net
,
98 const struct in6_addr
*addr6
,
101 unsigned int hash
, mix
= net_hash_mix(net
);
103 if (ipv6_addr_any(addr6
))
104 hash
= jhash_1word(0, mix
);
105 else if (ipv6_addr_v4mapped(addr6
))
106 hash
= jhash_1word((__force u32
)addr6
->s6_addr32
[3], mix
);
108 hash
= jhash2((__force u32
*)addr6
->s6_addr32
, 4, mix
);
114 int udp_v6_get_port(struct sock
*sk
, unsigned short snum
)
116 unsigned int hash2_nulladdr
=
117 udp6_portaddr_hash(sock_net(sk
), &in6addr_any
, snum
);
118 unsigned int hash2_partial
=
119 udp6_portaddr_hash(sock_net(sk
), &sk
->sk_v6_rcv_saddr
, 0);
121 /* precompute partial secondary hash */
122 udp_sk(sk
)->udp_portaddr_hash
= hash2_partial
;
123 return udp_lib_get_port(sk
, snum
, ipv6_rcv_saddr_equal
, hash2_nulladdr
);
126 static void udp_v6_rehash(struct sock
*sk
)
128 u16 new_hash
= udp6_portaddr_hash(sock_net(sk
),
129 &sk
->sk_v6_rcv_saddr
,
130 inet_sk(sk
)->inet_num
);
132 udp_lib_rehash(sk
, new_hash
);
135 static inline int compute_score(struct sock
*sk
, struct net
*net
,
137 const struct in6_addr
*saddr
, __be16 sport
,
138 const struct in6_addr
*daddr
, __be16 dport
,
143 if (net_eq(sock_net(sk
), net
) && udp_sk(sk
)->udp_port_hash
== hnum
&&
144 sk
->sk_family
== PF_INET6
) {
145 struct inet_sock
*inet
= inet_sk(sk
);
148 if (inet
->inet_dport
) {
149 if (inet
->inet_dport
!= sport
)
153 if (!ipv6_addr_any(&sk
->sk_v6_rcv_saddr
)) {
154 if (!ipv6_addr_equal(&sk
->sk_v6_rcv_saddr
, daddr
))
158 if (!ipv6_addr_any(&sk
->sk_v6_daddr
)) {
159 if (!ipv6_addr_equal(&sk
->sk_v6_daddr
, saddr
))
163 if (sk
->sk_bound_dev_if
) {
164 if (sk
->sk_bound_dev_if
!= dif
)
172 #define SCORE2_MAX (1 + 1 + 1)
173 static inline int compute_score2(struct sock
*sk
, struct net
*net
,
174 const struct in6_addr
*saddr
, __be16 sport
,
175 const struct in6_addr
*daddr
, unsigned short hnum
,
180 if (net_eq(sock_net(sk
), net
) && udp_sk(sk
)->udp_port_hash
== hnum
&&
181 sk
->sk_family
== PF_INET6
) {
182 struct inet_sock
*inet
= inet_sk(sk
);
184 if (!ipv6_addr_equal(&sk
->sk_v6_rcv_saddr
, daddr
))
187 if (inet
->inet_dport
) {
188 if (inet
->inet_dport
!= sport
)
192 if (!ipv6_addr_any(&sk
->sk_v6_daddr
)) {
193 if (!ipv6_addr_equal(&sk
->sk_v6_daddr
, saddr
))
197 if (sk
->sk_bound_dev_if
) {
198 if (sk
->sk_bound_dev_if
!= dif
)
207 /* called with read_rcu_lock() */
208 static struct sock
*udp6_lib_lookup2(struct net
*net
,
209 const struct in6_addr
*saddr
, __be16 sport
,
210 const struct in6_addr
*daddr
, unsigned int hnum
, int dif
,
211 struct udp_hslot
*hslot2
, unsigned int slot2
)
213 struct sock
*sk
, *result
;
214 struct hlist_nulls_node
*node
;
215 int score
, badness
, matches
= 0, reuseport
= 0;
221 udp_portaddr_for_each_entry_rcu(sk
, node
, &hslot2
->head
) {
222 score
= compute_score2(sk
, net
, saddr
, sport
,
224 if (score
> badness
) {
227 reuseport
= sk
->sk_reuseport
;
229 hash
= udp6_ehashfn(net
, daddr
, hnum
,
232 } else if (score
== SCORE2_MAX
)
234 } else if (score
== badness
&& reuseport
) {
236 if (((u64
)hash
* matches
) >> 32 == 0)
238 hash
= next_pseudo_random32(hash
);
242 * if the nulls value we got at the end of this lookup is
243 * not the expected one, we must restart lookup.
244 * We probably met an item that was moved to another chain.
246 if (get_nulls_value(node
) != slot2
)
251 if (unlikely(!atomic_inc_not_zero_hint(&result
->sk_refcnt
, 2)))
253 else if (unlikely(compute_score2(result
, net
, saddr
, sport
,
254 daddr
, hnum
, dif
) < badness
)) {
262 struct sock
*__udp6_lib_lookup(struct net
*net
,
263 const struct in6_addr
*saddr
, __be16 sport
,
264 const struct in6_addr
*daddr
, __be16 dport
,
265 int dif
, struct udp_table
*udptable
)
267 struct sock
*sk
, *result
;
268 struct hlist_nulls_node
*node
;
269 unsigned short hnum
= ntohs(dport
);
270 unsigned int hash2
, slot2
, slot
= udp_hashfn(net
, hnum
, udptable
->mask
);
271 struct udp_hslot
*hslot2
, *hslot
= &udptable
->hash
[slot
];
272 int score
, badness
, matches
= 0, reuseport
= 0;
276 if (hslot
->count
> 10) {
277 hash2
= udp6_portaddr_hash(net
, daddr
, hnum
);
278 slot2
= hash2
& udptable
->mask
;
279 hslot2
= &udptable
->hash2
[slot2
];
280 if (hslot
->count
< hslot2
->count
)
283 result
= udp6_lib_lookup2(net
, saddr
, sport
,
287 hash2
= udp6_portaddr_hash(net
, &in6addr_any
, hnum
);
288 slot2
= hash2
& udptable
->mask
;
289 hslot2
= &udptable
->hash2
[slot2
];
290 if (hslot
->count
< hslot2
->count
)
293 result
= udp6_lib_lookup2(net
, saddr
, sport
,
294 &in6addr_any
, hnum
, dif
,
303 sk_nulls_for_each_rcu(sk
, node
, &hslot
->head
) {
304 score
= compute_score(sk
, net
, hnum
, saddr
, sport
, daddr
, dport
, dif
);
305 if (score
> badness
) {
308 reuseport
= sk
->sk_reuseport
;
310 hash
= udp6_ehashfn(net
, daddr
, hnum
,
314 } else if (score
== badness
&& reuseport
) {
316 if (((u64
)hash
* matches
) >> 32 == 0)
318 hash
= next_pseudo_random32(hash
);
322 * if the nulls value we got at the end of this lookup is
323 * not the expected one, we must restart lookup.
324 * We probably met an item that was moved to another chain.
326 if (get_nulls_value(node
) != slot
)
330 if (unlikely(!atomic_inc_not_zero_hint(&result
->sk_refcnt
, 2)))
332 else if (unlikely(compute_score(result
, net
, hnum
, saddr
, sport
,
333 daddr
, dport
, dif
) < badness
)) {
341 EXPORT_SYMBOL_GPL(__udp6_lib_lookup
);
343 static struct sock
*__udp6_lib_lookup_skb(struct sk_buff
*skb
,
344 __be16 sport
, __be16 dport
,
345 struct udp_table
*udptable
)
348 const struct ipv6hdr
*iph
= ipv6_hdr(skb
);
350 if (unlikely(sk
= skb_steal_sock(skb
)))
352 return __udp6_lib_lookup(dev_net(skb_dst(skb
)->dev
), &iph
->saddr
, sport
,
353 &iph
->daddr
, dport
, inet6_iif(skb
),
357 struct sock
*udp6_lib_lookup(struct net
*net
, const struct in6_addr
*saddr
, __be16 sport
,
358 const struct in6_addr
*daddr
, __be16 dport
, int dif
)
360 return __udp6_lib_lookup(net
, saddr
, sport
, daddr
, dport
, dif
, &udp_table
);
362 EXPORT_SYMBOL_GPL(udp6_lib_lookup
);
366 * This should be easy, if there is something there we
367 * return it, otherwise we block.
370 int udpv6_recvmsg(struct kiocb
*iocb
, struct sock
*sk
,
371 struct msghdr
*msg
, size_t len
,
372 int noblock
, int flags
, int *addr_len
)
374 struct ipv6_pinfo
*np
= inet6_sk(sk
);
375 struct inet_sock
*inet
= inet_sk(sk
);
377 unsigned int ulen
, copied
;
380 int is_udplite
= IS_UDPLITE(sk
);
385 *addr_len
= sizeof(struct sockaddr_in6
);
387 if (flags
& MSG_ERRQUEUE
)
388 return ipv6_recv_error(sk
, msg
, len
);
390 if (np
->rxpmtu
&& np
->rxopt
.bits
.rxpmtu
)
391 return ipv6_recv_rxpmtu(sk
, msg
, len
);
394 skb
= __skb_recv_datagram(sk
, flags
| (noblock
? MSG_DONTWAIT
: 0),
395 &peeked
, &off
, &err
);
399 ulen
= skb
->len
- sizeof(struct udphdr
);
403 else if (copied
< ulen
)
404 msg
->msg_flags
|= MSG_TRUNC
;
406 is_udp4
= (skb
->protocol
== htons(ETH_P_IP
));
409 * If checksum is needed at all, try to do it while copying the
410 * data. If the data is truncated, or if we only want a partial
411 * coverage checksum (UDP-Lite), do it before the copy.
414 if (copied
< ulen
|| UDP_SKB_CB(skb
)->partial_cov
) {
415 if (udp_lib_checksum_complete(skb
))
419 if (skb_csum_unnecessary(skb
))
420 err
= skb_copy_datagram_iovec(skb
, sizeof(struct udphdr
),
421 msg
->msg_iov
, copied
);
423 err
= skb_copy_and_csum_datagram_iovec(skb
, sizeof(struct udphdr
), msg
->msg_iov
);
428 trace_kfree_skb(skb
, udpv6_recvmsg
);
430 atomic_inc(&sk
->sk_drops
);
432 UDP_INC_STATS_USER(sock_net(sk
),
436 UDP6_INC_STATS_USER(sock_net(sk
),
444 UDP_INC_STATS_USER(sock_net(sk
),
445 UDP_MIB_INDATAGRAMS
, is_udplite
);
447 UDP6_INC_STATS_USER(sock_net(sk
),
448 UDP_MIB_INDATAGRAMS
, is_udplite
);
451 sock_recv_ts_and_drops(msg
, sk
, skb
);
453 /* Copy the address. */
455 struct sockaddr_in6
*sin6
;
457 sin6
= (struct sockaddr_in6
*) msg
->msg_name
;
458 sin6
->sin6_family
= AF_INET6
;
459 sin6
->sin6_port
= udp_hdr(skb
)->source
;
460 sin6
->sin6_flowinfo
= 0;
463 ipv6_addr_set_v4mapped(ip_hdr(skb
)->saddr
,
465 sin6
->sin6_scope_id
= 0;
467 sin6
->sin6_addr
= ipv6_hdr(skb
)->saddr
;
468 sin6
->sin6_scope_id
=
469 ipv6_iface_scope_id(&sin6
->sin6_addr
,
475 if (inet
->cmsg_flags
)
476 ip_cmsg_recv(msg
, skb
);
479 ip6_datagram_recv_ctl(sk
, msg
, skb
);
483 if (flags
& MSG_TRUNC
)
487 skb_free_datagram_locked(sk
, skb
);
492 slow
= lock_sock_fast(sk
);
493 if (!skb_kill_datagram(sk
, skb
, flags
)) {
495 UDP_INC_STATS_USER(sock_net(sk
),
496 UDP_MIB_CSUMERRORS
, is_udplite
);
497 UDP_INC_STATS_USER(sock_net(sk
),
498 UDP_MIB_INERRORS
, is_udplite
);
500 UDP6_INC_STATS_USER(sock_net(sk
),
501 UDP_MIB_CSUMERRORS
, is_udplite
);
502 UDP6_INC_STATS_USER(sock_net(sk
),
503 UDP_MIB_INERRORS
, is_udplite
);
506 unlock_sock_fast(sk
, slow
);
511 /* starting over for a new packet */
512 msg
->msg_flags
&= ~MSG_TRUNC
;
516 void __udp6_lib_err(struct sk_buff
*skb
, struct inet6_skb_parm
*opt
,
517 u8 type
, u8 code
, int offset
, __be32 info
,
518 struct udp_table
*udptable
)
520 struct ipv6_pinfo
*np
;
521 const struct ipv6hdr
*hdr
= (const struct ipv6hdr
*)skb
->data
;
522 const struct in6_addr
*saddr
= &hdr
->saddr
;
523 const struct in6_addr
*daddr
= &hdr
->daddr
;
524 struct udphdr
*uh
= (struct udphdr
*)(skb
->data
+offset
);
528 sk
= __udp6_lib_lookup(dev_net(skb
->dev
), daddr
, uh
->dest
,
529 saddr
, uh
->source
, inet6_iif(skb
), udptable
);
533 if (type
== ICMPV6_PKT_TOOBIG
)
534 ip6_sk_update_pmtu(skb
, sk
, info
);
535 if (type
== NDISC_REDIRECT
) {
536 ip6_sk_redirect(skb
, sk
);
542 if (!icmpv6_err_convert(type
, code
, &err
) && !np
->recverr
)
545 if (sk
->sk_state
!= TCP_ESTABLISHED
&& !np
->recverr
)
549 ipv6_icmp_error(sk
, skb
, err
, uh
->dest
, ntohl(info
), (u8
*)(uh
+1));
552 sk
->sk_error_report(sk
);
557 static int __udpv6_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
561 if (!ipv6_addr_any(&sk
->sk_v6_daddr
)) {
562 sock_rps_save_rxhash(sk
, skb
);
563 sk_mark_napi_id(sk
, skb
);
566 rc
= sock_queue_rcv_skb(sk
, skb
);
568 int is_udplite
= IS_UDPLITE(sk
);
570 /* Note that an ENOMEM error is charged twice */
572 UDP6_INC_STATS_BH(sock_net(sk
),
573 UDP_MIB_RCVBUFERRORS
, is_udplite
);
574 UDP6_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
581 static __inline__
void udpv6_err(struct sk_buff
*skb
,
582 struct inet6_skb_parm
*opt
, u8 type
,
583 u8 code
, int offset
, __be32 info
)
585 __udp6_lib_err(skb
, opt
, type
, code
, offset
, info
, &udp_table
);
588 static struct static_key udpv6_encap_needed __read_mostly
;
589 void udpv6_encap_enable(void)
591 if (!static_key_enabled(&udpv6_encap_needed
))
592 static_key_slow_inc(&udpv6_encap_needed
);
594 EXPORT_SYMBOL(udpv6_encap_enable
);
596 int udpv6_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
598 struct udp_sock
*up
= udp_sk(sk
);
600 int is_udplite
= IS_UDPLITE(sk
);
602 if (!xfrm6_policy_check(sk
, XFRM_POLICY_IN
, skb
))
605 if (static_key_false(&udpv6_encap_needed
) && up
->encap_type
) {
606 int (*encap_rcv
)(struct sock
*sk
, struct sk_buff
*skb
);
609 * This is an encapsulation socket so pass the skb to
610 * the socket's udp_encap_rcv() hook. Otherwise, just
611 * fall through and pass this up the UDP socket.
612 * up->encap_rcv() returns the following value:
613 * =0 if skb was successfully passed to the encap
614 * handler or was discarded by it.
615 * >0 if skb should be passed on to UDP.
616 * <0 if skb should be resubmitted as proto -N
619 /* if we're overly short, let UDP handle it */
620 encap_rcv
= ACCESS_ONCE(up
->encap_rcv
);
621 if (skb
->len
> sizeof(struct udphdr
) && encap_rcv
!= NULL
) {
624 ret
= encap_rcv(sk
, skb
);
626 UDP_INC_STATS_BH(sock_net(sk
),
633 /* FALLTHROUGH -- it's a UDP Packet */
637 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
639 if ((is_udplite
& UDPLITE_RECV_CC
) && UDP_SKB_CB(skb
)->partial_cov
) {
641 if (up
->pcrlen
== 0) { /* full coverage was set */
642 LIMIT_NETDEBUG(KERN_WARNING
"UDPLITE6: partial coverage"
643 " %d while full coverage %d requested\n",
644 UDP_SKB_CB(skb
)->cscov
, skb
->len
);
647 if (UDP_SKB_CB(skb
)->cscov
< up
->pcrlen
) {
648 LIMIT_NETDEBUG(KERN_WARNING
"UDPLITE6: coverage %d "
649 "too small, need min %d\n",
650 UDP_SKB_CB(skb
)->cscov
, up
->pcrlen
);
655 if (rcu_access_pointer(sk
->sk_filter
)) {
656 if (udp_lib_checksum_complete(skb
))
660 if (sk_rcvqueues_full(sk
, skb
, sk
->sk_rcvbuf
))
667 if (!sock_owned_by_user(sk
))
668 rc
= __udpv6_queue_rcv_skb(sk
, skb
);
669 else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
677 UDP6_INC_STATS_BH(sock_net(sk
), UDP_MIB_CSUMERRORS
, is_udplite
);
679 UDP6_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
680 atomic_inc(&sk
->sk_drops
);
685 static struct sock
*udp_v6_mcast_next(struct net
*net
, struct sock
*sk
,
686 __be16 loc_port
, const struct in6_addr
*loc_addr
,
687 __be16 rmt_port
, const struct in6_addr
*rmt_addr
,
690 struct hlist_nulls_node
*node
;
692 unsigned short num
= ntohs(loc_port
);
694 sk_nulls_for_each_from(s
, node
) {
695 struct inet_sock
*inet
= inet_sk(s
);
697 if (!net_eq(sock_net(s
), net
))
700 if (udp_sk(s
)->udp_port_hash
== num
&&
701 s
->sk_family
== PF_INET6
) {
702 if (inet
->inet_dport
) {
703 if (inet
->inet_dport
!= rmt_port
)
706 if (!ipv6_addr_any(&sk
->sk_v6_daddr
) &&
707 !ipv6_addr_equal(&sk
->sk_v6_daddr
, rmt_addr
))
710 if (s
->sk_bound_dev_if
&& s
->sk_bound_dev_if
!= dif
)
713 if (!ipv6_addr_any(&sk
->sk_v6_rcv_saddr
)) {
714 if (!ipv6_addr_equal(&sk
->sk_v6_rcv_saddr
, loc_addr
))
717 if (!inet6_mc_check(s
, loc_addr
, rmt_addr
))
725 static void flush_stack(struct sock
**stack
, unsigned int count
,
726 struct sk_buff
*skb
, unsigned int final
)
728 struct sk_buff
*skb1
= NULL
;
732 for (i
= 0; i
< count
; i
++) {
734 if (likely(skb1
== NULL
))
735 skb1
= (i
== final
) ? skb
: skb_clone(skb
, GFP_ATOMIC
);
737 atomic_inc(&sk
->sk_drops
);
738 UDP6_INC_STATS_BH(sock_net(sk
), UDP_MIB_RCVBUFERRORS
,
740 UDP6_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
,
744 if (skb1
&& udpv6_queue_rcv_skb(sk
, skb1
) <= 0)
751 * Note: called only from the BH handler context,
752 * so we don't need to lock the hashes.
754 static int __udp6_lib_mcast_deliver(struct net
*net
, struct sk_buff
*skb
,
755 const struct in6_addr
*saddr
, const struct in6_addr
*daddr
,
756 struct udp_table
*udptable
)
758 struct sock
*sk
, *stack
[256 / sizeof(struct sock
*)];
759 const struct udphdr
*uh
= udp_hdr(skb
);
760 struct udp_hslot
*hslot
= udp_hashslot(udptable
, net
, ntohs(uh
->dest
));
762 unsigned int i
, count
= 0;
764 spin_lock(&hslot
->lock
);
765 sk
= sk_nulls_head(&hslot
->head
);
766 dif
= inet6_iif(skb
);
767 sk
= udp_v6_mcast_next(net
, sk
, uh
->dest
, daddr
, uh
->source
, saddr
, dif
);
770 sk
= udp_v6_mcast_next(net
, sk_nulls_next(sk
), uh
->dest
, daddr
,
771 uh
->source
, saddr
, dif
);
772 if (unlikely(count
== ARRAY_SIZE(stack
))) {
775 flush_stack(stack
, count
, skb
, ~0);
780 * before releasing the lock, we must take reference on sockets
782 for (i
= 0; i
< count
; i
++)
785 spin_unlock(&hslot
->lock
);
788 flush_stack(stack
, count
, skb
, count
- 1);
790 for (i
= 0; i
< count
; i
++)
798 int __udp6_lib_rcv(struct sk_buff
*skb
, struct udp_table
*udptable
,
801 struct net
*net
= dev_net(skb
->dev
);
804 const struct in6_addr
*saddr
, *daddr
;
807 if (!pskb_may_pull(skb
, sizeof(struct udphdr
)))
810 saddr
= &ipv6_hdr(skb
)->saddr
;
811 daddr
= &ipv6_hdr(skb
)->daddr
;
814 ulen
= ntohs(uh
->len
);
818 if (proto
== IPPROTO_UDP
) {
819 /* UDP validates ulen. */
821 /* Check for jumbo payload */
825 if (ulen
< sizeof(*uh
))
828 if (ulen
< skb
->len
) {
829 if (pskb_trim_rcsum(skb
, ulen
))
831 saddr
= &ipv6_hdr(skb
)->saddr
;
832 daddr
= &ipv6_hdr(skb
)->daddr
;
837 if (udp6_csum_init(skb
, uh
, proto
))
841 * Multicast receive code
843 if (ipv6_addr_is_multicast(daddr
))
844 return __udp6_lib_mcast_deliver(net
, skb
,
845 saddr
, daddr
, udptable
);
850 * check socket cache ... must talk to Alan about his plans
851 * for sock caches... i'll skip this for now.
853 sk
= __udp6_lib_lookup_skb(skb
, uh
->source
, uh
->dest
, udptable
);
857 ret
= udpv6_queue_rcv_skb(sk
, skb
);
860 /* a return value > 0 means to resubmit the input, but
861 * it wants the return to be -protocol, or 0
869 if (!xfrm6_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
872 if (udp_lib_checksum_complete(skb
))
875 UDP6_INC_STATS_BH(net
, UDP_MIB_NOPORTS
, proto
== IPPROTO_UDPLITE
);
876 icmpv6_send(skb
, ICMPV6_DEST_UNREACH
, ICMPV6_PORT_UNREACH
, 0);
882 LIMIT_NETDEBUG(KERN_DEBUG
"UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
883 proto
== IPPROTO_UDPLITE
? "-Lite" : "",
892 UDP6_INC_STATS_BH(net
, UDP_MIB_CSUMERRORS
, proto
== IPPROTO_UDPLITE
);
894 UDP6_INC_STATS_BH(net
, UDP_MIB_INERRORS
, proto
== IPPROTO_UDPLITE
);
899 static __inline__
int udpv6_rcv(struct sk_buff
*skb
)
901 return __udp6_lib_rcv(skb
, &udp_table
, IPPROTO_UDP
);
905 * Throw away all pending data and cancel the corking. Socket is locked.
907 static void udp_v6_flush_pending_frames(struct sock
*sk
)
909 struct udp_sock
*up
= udp_sk(sk
);
911 if (up
->pending
== AF_INET
)
912 udp_flush_pending_frames(sk
);
913 else if (up
->pending
) {
916 ip6_flush_pending_frames(sk
);
921 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
922 * @sk: socket we are sending on
923 * @skb: sk_buff containing the filled-in UDP header
924 * (checksum field must be zeroed out)
926 static void udp6_hwcsum_outgoing(struct sock
*sk
, struct sk_buff
*skb
,
927 const struct in6_addr
*saddr
,
928 const struct in6_addr
*daddr
, int len
)
931 struct udphdr
*uh
= udp_hdr(skb
);
934 if (skb_queue_len(&sk
->sk_write_queue
) == 1) {
935 /* Only one fragment on the socket. */
936 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
937 skb
->csum_offset
= offsetof(struct udphdr
, check
);
938 uh
->check
= ~csum_ipv6_magic(saddr
, daddr
, len
, IPPROTO_UDP
, 0);
941 * HW-checksum won't work as there are two or more
942 * fragments on the socket so that all csums of sk_buffs
945 offset
= skb_transport_offset(skb
);
946 skb
->csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
948 skb
->ip_summed
= CHECKSUM_NONE
;
950 skb_queue_walk(&sk
->sk_write_queue
, skb
) {
951 csum
= csum_add(csum
, skb
->csum
);
954 uh
->check
= csum_ipv6_magic(saddr
, daddr
, len
, IPPROTO_UDP
,
957 uh
->check
= CSUM_MANGLED_0
;
965 static int udp_v6_push_pending_frames(struct sock
*sk
)
969 struct udp_sock
*up
= udp_sk(sk
);
970 struct inet_sock
*inet
= inet_sk(sk
);
973 int is_udplite
= IS_UDPLITE(sk
);
976 if (up
->pending
== AF_INET
)
977 return udp_push_pending_frames(sk
);
979 fl6
= &inet
->cork
.fl
.u
.ip6
;
981 /* Grab the skbuff where UDP header space exists. */
982 if ((skb
= skb_peek(&sk
->sk_write_queue
)) == NULL
)
986 * Create a UDP header
989 uh
->source
= fl6
->fl6_sport
;
990 uh
->dest
= fl6
->fl6_dport
;
991 uh
->len
= htons(up
->len
);
995 csum
= udplite_csum_outgoing(sk
, skb
);
996 else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) { /* UDP hardware csum */
997 udp6_hwcsum_outgoing(sk
, skb
, &fl6
->saddr
, &fl6
->daddr
,
1001 csum
= udp_csum_outgoing(sk
, skb
);
1003 /* add protocol-dependent pseudo-header */
1004 uh
->check
= csum_ipv6_magic(&fl6
->saddr
, &fl6
->daddr
,
1005 up
->len
, fl6
->flowi6_proto
, csum
);
1007 uh
->check
= CSUM_MANGLED_0
;
1010 err
= ip6_push_pending_frames(sk
);
1012 if (err
== -ENOBUFS
&& !inet6_sk(sk
)->recverr
) {
1013 UDP6_INC_STATS_USER(sock_net(sk
),
1014 UDP_MIB_SNDBUFERRORS
, is_udplite
);
1018 UDP6_INC_STATS_USER(sock_net(sk
),
1019 UDP_MIB_OUTDATAGRAMS
, is_udplite
);
1026 int udpv6_sendmsg(struct kiocb
*iocb
, struct sock
*sk
,
1027 struct msghdr
*msg
, size_t len
)
1029 struct ipv6_txoptions opt_space
;
1030 struct udp_sock
*up
= udp_sk(sk
);
1031 struct inet_sock
*inet
= inet_sk(sk
);
1032 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1033 struct sockaddr_in6
*sin6
= (struct sockaddr_in6
*) msg
->msg_name
;
1034 struct in6_addr
*daddr
, *final_p
, final
;
1035 struct ipv6_txoptions
*opt
= NULL
;
1036 struct ip6_flowlabel
*flowlabel
= NULL
;
1038 struct dst_entry
*dst
;
1039 int addr_len
= msg
->msg_namelen
;
1044 int corkreq
= up
->corkflag
|| msg
->msg_flags
&MSG_MORE
;
1047 int is_udplite
= IS_UDPLITE(sk
);
1048 int (*getfrag
)(void *, char *, int, int, int, struct sk_buff
*);
1050 /* destination address check */
1052 if (addr_len
< offsetof(struct sockaddr
, sa_data
))
1055 switch (sin6
->sin6_family
) {
1057 if (addr_len
< SIN6_LEN_RFC2133
)
1059 daddr
= &sin6
->sin6_addr
;
1062 goto do_udp_sendmsg
;
1064 msg
->msg_name
= sin6
= NULL
;
1065 msg
->msg_namelen
= addr_len
= 0;
1071 } else if (!up
->pending
) {
1072 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1073 return -EDESTADDRREQ
;
1074 daddr
= &sk
->sk_v6_daddr
;
1079 if (ipv6_addr_v4mapped(daddr
)) {
1080 struct sockaddr_in sin
;
1081 sin
.sin_family
= AF_INET
;
1082 sin
.sin_port
= sin6
? sin6
->sin6_port
: inet
->inet_dport
;
1083 sin
.sin_addr
.s_addr
= daddr
->s6_addr32
[3];
1084 msg
->msg_name
= &sin
;
1085 msg
->msg_namelen
= sizeof(sin
);
1087 if (__ipv6_only_sock(sk
))
1088 return -ENETUNREACH
;
1089 return udp_sendmsg(iocb
, sk
, msg
, len
);
1093 if (up
->pending
== AF_INET
)
1094 return udp_sendmsg(iocb
, sk
, msg
, len
);
1096 /* Rough check on arithmetic overflow,
1097 better check is made in ip6_append_data().
1099 if (len
> INT_MAX
- sizeof(struct udphdr
))
1104 * There are pending frames.
1105 * The socket lock must be held while it's corked.
1108 if (likely(up
->pending
)) {
1109 if (unlikely(up
->pending
!= AF_INET6
)) {
1111 return -EAFNOSUPPORT
;
1114 goto do_append_data
;
1118 ulen
+= sizeof(struct udphdr
);
1120 memset(&fl6
, 0, sizeof(fl6
));
1123 if (sin6
->sin6_port
== 0)
1126 fl6
.fl6_dport
= sin6
->sin6_port
;
1127 daddr
= &sin6
->sin6_addr
;
1130 fl6
.flowlabel
= sin6
->sin6_flowinfo
&IPV6_FLOWINFO_MASK
;
1131 if (fl6
.flowlabel
&IPV6_FLOWLABEL_MASK
) {
1132 flowlabel
= fl6_sock_lookup(sk
, fl6
.flowlabel
);
1133 if (flowlabel
== NULL
)
1135 daddr
= &flowlabel
->dst
;
1140 * Otherwise it will be difficult to maintain
1143 if (sk
->sk_state
== TCP_ESTABLISHED
&&
1144 ipv6_addr_equal(daddr
, &sk
->sk_v6_daddr
))
1145 daddr
= &sk
->sk_v6_daddr
;
1147 if (addr_len
>= sizeof(struct sockaddr_in6
) &&
1148 sin6
->sin6_scope_id
&&
1149 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr
)))
1150 fl6
.flowi6_oif
= sin6
->sin6_scope_id
;
1152 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1153 return -EDESTADDRREQ
;
1155 fl6
.fl6_dport
= inet
->inet_dport
;
1156 daddr
= &sk
->sk_v6_daddr
;
1157 fl6
.flowlabel
= np
->flow_label
;
1161 if (!fl6
.flowi6_oif
)
1162 fl6
.flowi6_oif
= sk
->sk_bound_dev_if
;
1164 if (!fl6
.flowi6_oif
)
1165 fl6
.flowi6_oif
= np
->sticky_pktinfo
.ipi6_ifindex
;
1167 fl6
.flowi6_mark
= sk
->sk_mark
;
1169 if (msg
->msg_controllen
) {
1171 memset(opt
, 0, sizeof(struct ipv6_txoptions
));
1172 opt
->tot_len
= sizeof(*opt
);
1174 err
= ip6_datagram_send_ctl(sock_net(sk
), sk
, msg
, &fl6
, opt
,
1175 &hlimit
, &tclass
, &dontfrag
);
1177 fl6_sock_release(flowlabel
);
1180 if ((fl6
.flowlabel
&IPV6_FLOWLABEL_MASK
) && !flowlabel
) {
1181 flowlabel
= fl6_sock_lookup(sk
, fl6
.flowlabel
);
1182 if (flowlabel
== NULL
)
1185 if (!(opt
->opt_nflen
|opt
->opt_flen
))
1192 opt
= fl6_merge_options(&opt_space
, flowlabel
, opt
);
1193 opt
= ipv6_fixup_options(&opt_space
, opt
);
1195 fl6
.flowi6_proto
= sk
->sk_protocol
;
1196 if (!ipv6_addr_any(daddr
))
1199 fl6
.daddr
.s6_addr
[15] = 0x1; /* :: means loopback (BSD'ism) */
1200 if (ipv6_addr_any(&fl6
.saddr
) && !ipv6_addr_any(&np
->saddr
))
1201 fl6
.saddr
= np
->saddr
;
1202 fl6
.fl6_sport
= inet
->inet_sport
;
1204 final_p
= fl6_update_dst(&fl6
, opt
, &final
);
1208 if (!fl6
.flowi6_oif
&& ipv6_addr_is_multicast(&fl6
.daddr
)) {
1209 fl6
.flowi6_oif
= np
->mcast_oif
;
1211 } else if (!fl6
.flowi6_oif
)
1212 fl6
.flowi6_oif
= np
->ucast_oif
;
1214 security_sk_classify_flow(sk
, flowi6_to_flowi(&fl6
));
1216 dst
= ip6_sk_dst_lookup_flow(sk
, &fl6
, final_p
, true);
1224 if (ipv6_addr_is_multicast(&fl6
.daddr
))
1225 hlimit
= np
->mcast_hops
;
1227 hlimit
= np
->hop_limit
;
1229 hlimit
= ip6_dst_hoplimit(dst
);
1233 tclass
= np
->tclass
;
1236 dontfrag
= np
->dontfrag
;
1238 if (msg
->msg_flags
&MSG_CONFIRM
)
1243 if (unlikely(up
->pending
)) {
1244 /* The socket is already corked while preparing it. */
1245 /* ... which is an evident application bug. --ANK */
1248 LIMIT_NETDEBUG(KERN_DEBUG
"udp cork app bug 2\n");
1253 up
->pending
= AF_INET6
;
1257 getfrag
= is_udplite
? udplite_getfrag
: ip_generic_getfrag
;
1258 err
= ip6_append_data(sk
, getfrag
, msg
->msg_iov
, ulen
,
1259 sizeof(struct udphdr
), hlimit
, tclass
, opt
, &fl6
,
1260 (struct rt6_info
*)dst
,
1261 corkreq
? msg
->msg_flags
|MSG_MORE
: msg
->msg_flags
, dontfrag
);
1263 udp_v6_flush_pending_frames(sk
);
1265 err
= udp_v6_push_pending_frames(sk
);
1266 else if (unlikely(skb_queue_empty(&sk
->sk_write_queue
)))
1271 ip6_dst_store(sk
, dst
,
1272 ipv6_addr_equal(&fl6
.daddr
, &sk
->sk_v6_daddr
) ?
1273 &sk
->sk_v6_daddr
: NULL
,
1274 #ifdef CONFIG_IPV6_SUBTREES
1275 ipv6_addr_equal(&fl6
.saddr
, &np
->saddr
) ?
1286 err
= np
->recverr
? net_xmit_errno(err
) : 0;
1290 fl6_sock_release(flowlabel
);
1294 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1295 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1296 * we don't have a good statistic (IpOutDiscards but it can be too many
1297 * things). We could add another new stat but at least for now that
1298 * seems like overkill.
1300 if (err
== -ENOBUFS
|| test_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
)) {
1301 UDP6_INC_STATS_USER(sock_net(sk
),
1302 UDP_MIB_SNDBUFERRORS
, is_udplite
);
1308 if (!(msg
->msg_flags
&MSG_PROBE
) || len
)
1309 goto back_from_confirm
;
1314 void udpv6_destroy_sock(struct sock
*sk
)
1316 struct udp_sock
*up
= udp_sk(sk
);
1318 udp_v6_flush_pending_frames(sk
);
1321 if (static_key_false(&udpv6_encap_needed
) && up
->encap_type
) {
1322 void (*encap_destroy
)(struct sock
*sk
);
1323 encap_destroy
= ACCESS_ONCE(up
->encap_destroy
);
1328 inet6_destroy_sock(sk
);
1332 * Socket option code for UDP
1334 int udpv6_setsockopt(struct sock
*sk
, int level
, int optname
,
1335 char __user
*optval
, unsigned int optlen
)
1337 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1338 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1339 udp_v6_push_pending_frames
);
1340 return ipv6_setsockopt(sk
, level
, optname
, optval
, optlen
);
1343 #ifdef CONFIG_COMPAT
1344 int compat_udpv6_setsockopt(struct sock
*sk
, int level
, int optname
,
1345 char __user
*optval
, unsigned int optlen
)
1347 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1348 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1349 udp_v6_push_pending_frames
);
1350 return compat_ipv6_setsockopt(sk
, level
, optname
, optval
, optlen
);
1354 int 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 ipv6_getsockopt(sk
, level
, optname
, optval
, optlen
);
1362 #ifdef CONFIG_COMPAT
1363 int compat_udpv6_getsockopt(struct sock
*sk
, int level
, int optname
,
1364 char __user
*optval
, int __user
*optlen
)
1366 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1367 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1368 return compat_ipv6_getsockopt(sk
, level
, optname
, optval
, optlen
);
1372 static const struct inet6_protocol udpv6_protocol
= {
1373 .handler
= udpv6_rcv
,
1374 .err_handler
= udpv6_err
,
1375 .flags
= INET6_PROTO_NOPOLICY
|INET6_PROTO_FINAL
,
1378 /* ------------------------------------------------------------------------ */
1379 #ifdef CONFIG_PROC_FS
1380 int udp6_seq_show(struct seq_file
*seq
, void *v
)
1382 if (v
== SEQ_START_TOKEN
) {
1383 seq_puts(seq
, IPV6_SEQ_DGRAM_HEADER
);
1385 int bucket
= ((struct udp_iter_state
*)seq
->private)->bucket
;
1386 struct inet_sock
*inet
= inet_sk(v
);
1387 __u16 srcp
= ntohs(inet
->inet_sport
);
1388 __u16 destp
= ntohs(inet
->inet_dport
);
1389 ip6_dgram_sock_seq_show(seq
, v
, srcp
, destp
, bucket
);
1394 static const struct file_operations udp6_afinfo_seq_fops
= {
1395 .owner
= THIS_MODULE
,
1396 .open
= udp_seq_open
,
1398 .llseek
= seq_lseek
,
1399 .release
= seq_release_net
1402 static struct udp_seq_afinfo udp6_seq_afinfo
= {
1405 .udp_table
= &udp_table
,
1406 .seq_fops
= &udp6_afinfo_seq_fops
,
1408 .show
= udp6_seq_show
,
1412 int __net_init
udp6_proc_init(struct net
*net
)
1414 return udp_proc_register(net
, &udp6_seq_afinfo
);
1417 void udp6_proc_exit(struct net
*net
) {
1418 udp_proc_unregister(net
, &udp6_seq_afinfo
);
1420 #endif /* CONFIG_PROC_FS */
1422 void udp_v6_clear_sk(struct sock
*sk
, int size
)
1424 struct inet_sock
*inet
= inet_sk(sk
);
1426 /* we do not want to clear pinet6 field, because of RCU lookups */
1427 sk_prot_clear_portaddr_nulls(sk
, offsetof(struct inet_sock
, pinet6
));
1429 size
-= offsetof(struct inet_sock
, pinet6
) + sizeof(inet
->pinet6
);
1430 memset(&inet
->pinet6
+ 1, 0, size
);
1433 /* ------------------------------------------------------------------------ */
1435 struct proto udpv6_prot
= {
1437 .owner
= THIS_MODULE
,
1438 .close
= udp_lib_close
,
1439 .connect
= ip6_datagram_connect
,
1440 .disconnect
= udp_disconnect
,
1442 .destroy
= udpv6_destroy_sock
,
1443 .setsockopt
= udpv6_setsockopt
,
1444 .getsockopt
= udpv6_getsockopt
,
1445 .sendmsg
= udpv6_sendmsg
,
1446 .recvmsg
= udpv6_recvmsg
,
1447 .backlog_rcv
= __udpv6_queue_rcv_skb
,
1448 .hash
= udp_lib_hash
,
1449 .unhash
= udp_lib_unhash
,
1450 .rehash
= udp_v6_rehash
,
1451 .get_port
= udp_v6_get_port
,
1452 .memory_allocated
= &udp_memory_allocated
,
1453 .sysctl_mem
= sysctl_udp_mem
,
1454 .sysctl_wmem
= &sysctl_udp_wmem_min
,
1455 .sysctl_rmem
= &sysctl_udp_rmem_min
,
1456 .obj_size
= sizeof(struct udp6_sock
),
1457 .slab_flags
= SLAB_DESTROY_BY_RCU
,
1458 .h
.udp_table
= &udp_table
,
1459 #ifdef CONFIG_COMPAT
1460 .compat_setsockopt
= compat_udpv6_setsockopt
,
1461 .compat_getsockopt
= compat_udpv6_getsockopt
,
1463 .clear_sk
= udp_v6_clear_sk
,
1466 static struct inet_protosw udpv6_protosw
= {
1468 .protocol
= IPPROTO_UDP
,
1469 .prot
= &udpv6_prot
,
1470 .ops
= &inet6_dgram_ops
,
1471 .no_check
= UDP_CSUM_DEFAULT
,
1472 .flags
= INET_PROTOSW_PERMANENT
,
1476 int __init
udpv6_init(void)
1480 ret
= inet6_add_protocol(&udpv6_protocol
, IPPROTO_UDP
);
1484 ret
= inet6_register_protosw(&udpv6_protosw
);
1486 goto out_udpv6_protocol
;
1491 inet6_del_protocol(&udpv6_protocol
, IPPROTO_UDP
);
1495 void udpv6_exit(void)
1497 inet6_unregister_protosw(&udpv6_protosw
);
1498 inet6_del_protocol(&udpv6_protocol
, IPPROTO_UDP
);