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>
50 #include <linux/proc_fs.h>
51 #include <linux/seq_file.h>
52 #include <trace/events/skb.h>
55 int ipv6_rcv_saddr_equal(const struct sock
*sk
, const struct sock
*sk2
)
57 const struct in6_addr
*sk_rcv_saddr6
= &inet6_sk(sk
)->rcv_saddr
;
58 const struct in6_addr
*sk2_rcv_saddr6
= inet6_rcv_saddr(sk2
);
59 __be32 sk1_rcv_saddr
= sk_rcv_saddr(sk
);
60 __be32 sk2_rcv_saddr
= sk_rcv_saddr(sk2
);
61 int sk_ipv6only
= ipv6_only_sock(sk
);
62 int sk2_ipv6only
= inet_v6_ipv6only(sk2
);
63 int addr_type
= ipv6_addr_type(sk_rcv_saddr6
);
64 int addr_type2
= sk2_rcv_saddr6
? ipv6_addr_type(sk2_rcv_saddr6
) : IPV6_ADDR_MAPPED
;
66 /* if both are mapped, treat as IPv4 */
67 if (addr_type
== IPV6_ADDR_MAPPED
&& addr_type2
== IPV6_ADDR_MAPPED
)
68 return (!sk2_ipv6only
&&
69 (!sk1_rcv_saddr
|| !sk2_rcv_saddr
||
70 sk1_rcv_saddr
== sk2_rcv_saddr
));
72 if (addr_type2
== IPV6_ADDR_ANY
&&
73 !(sk2_ipv6only
&& addr_type
== IPV6_ADDR_MAPPED
))
76 if (addr_type
== IPV6_ADDR_ANY
&&
77 !(sk_ipv6only
&& addr_type2
== IPV6_ADDR_MAPPED
))
81 ipv6_addr_equal(sk_rcv_saddr6
, sk2_rcv_saddr6
))
87 static unsigned int udp6_portaddr_hash(struct net
*net
,
88 const struct in6_addr
*addr6
,
91 unsigned int hash
, mix
= net_hash_mix(net
);
93 if (ipv6_addr_any(addr6
))
94 hash
= jhash_1word(0, mix
);
95 else if (ipv6_addr_v4mapped(addr6
))
96 hash
= jhash_1word((__force u32
)addr6
->s6_addr32
[3], mix
);
98 hash
= jhash2((__force u32
*)addr6
->s6_addr32
, 4, mix
);
104 int udp_v6_get_port(struct sock
*sk
, unsigned short snum
)
106 unsigned int hash2_nulladdr
=
107 udp6_portaddr_hash(sock_net(sk
), &in6addr_any
, snum
);
108 unsigned int hash2_partial
=
109 udp6_portaddr_hash(sock_net(sk
), &inet6_sk(sk
)->rcv_saddr
, 0);
111 /* precompute partial secondary hash */
112 udp_sk(sk
)->udp_portaddr_hash
= hash2_partial
;
113 return udp_lib_get_port(sk
, snum
, ipv6_rcv_saddr_equal
, hash2_nulladdr
);
116 static void udp_v6_rehash(struct sock
*sk
)
118 u16 new_hash
= udp6_portaddr_hash(sock_net(sk
),
119 &inet6_sk(sk
)->rcv_saddr
,
120 inet_sk(sk
)->inet_num
);
122 udp_lib_rehash(sk
, new_hash
);
125 static inline int compute_score(struct sock
*sk
, struct net
*net
,
127 const struct in6_addr
*saddr
, __be16 sport
,
128 const struct in6_addr
*daddr
, __be16 dport
,
133 if (net_eq(sock_net(sk
), net
) && udp_sk(sk
)->udp_port_hash
== hnum
&&
134 sk
->sk_family
== PF_INET6
) {
135 struct ipv6_pinfo
*np
= inet6_sk(sk
);
136 struct inet_sock
*inet
= inet_sk(sk
);
139 if (inet
->inet_dport
) {
140 if (inet
->inet_dport
!= sport
)
144 if (!ipv6_addr_any(&np
->rcv_saddr
)) {
145 if (!ipv6_addr_equal(&np
->rcv_saddr
, daddr
))
149 if (!ipv6_addr_any(&np
->daddr
)) {
150 if (!ipv6_addr_equal(&np
->daddr
, saddr
))
154 if (sk
->sk_bound_dev_if
) {
155 if (sk
->sk_bound_dev_if
!= dif
)
163 #define SCORE2_MAX (1 + 1 + 1)
164 static inline int compute_score2(struct sock
*sk
, struct net
*net
,
165 const struct in6_addr
*saddr
, __be16 sport
,
166 const struct in6_addr
*daddr
, unsigned short hnum
,
171 if (net_eq(sock_net(sk
), net
) && udp_sk(sk
)->udp_port_hash
== hnum
&&
172 sk
->sk_family
== PF_INET6
) {
173 struct ipv6_pinfo
*np
= inet6_sk(sk
);
174 struct inet_sock
*inet
= inet_sk(sk
);
176 if (!ipv6_addr_equal(&np
->rcv_saddr
, daddr
))
179 if (inet
->inet_dport
) {
180 if (inet
->inet_dport
!= sport
)
184 if (!ipv6_addr_any(&np
->daddr
)) {
185 if (!ipv6_addr_equal(&np
->daddr
, saddr
))
189 if (sk
->sk_bound_dev_if
) {
190 if (sk
->sk_bound_dev_if
!= dif
)
199 /* called with read_rcu_lock() */
200 static struct sock
*udp6_lib_lookup2(struct net
*net
,
201 const struct in6_addr
*saddr
, __be16 sport
,
202 const struct in6_addr
*daddr
, unsigned int hnum
, int dif
,
203 struct udp_hslot
*hslot2
, unsigned int slot2
)
205 struct sock
*sk
, *result
;
206 struct hlist_nulls_node
*node
;
207 int score
, badness
, matches
= 0, reuseport
= 0;
213 udp_portaddr_for_each_entry_rcu(sk
, node
, &hslot2
->head
) {
214 score
= compute_score2(sk
, net
, saddr
, sport
,
216 if (score
> badness
) {
219 reuseport
= sk
->sk_reuseport
;
221 hash
= inet6_ehashfn(net
, daddr
, hnum
,
224 } else if (score
== SCORE2_MAX
)
226 } else if (score
== badness
&& reuseport
) {
228 if (((u64
)hash
* matches
) >> 32 == 0)
230 hash
= next_pseudo_random32(hash
);
234 * if the nulls value we got at the end of this lookup is
235 * not the expected one, we must restart lookup.
236 * We probably met an item that was moved to another chain.
238 if (get_nulls_value(node
) != slot2
)
243 if (unlikely(!atomic_inc_not_zero_hint(&result
->sk_refcnt
, 2)))
245 else if (unlikely(compute_score2(result
, net
, saddr
, sport
,
246 daddr
, hnum
, dif
) < badness
)) {
254 struct sock
*__udp6_lib_lookup(struct net
*net
,
255 const struct in6_addr
*saddr
, __be16 sport
,
256 const struct in6_addr
*daddr
, __be16 dport
,
257 int dif
, struct udp_table
*udptable
)
259 struct sock
*sk
, *result
;
260 struct hlist_nulls_node
*node
;
261 unsigned short hnum
= ntohs(dport
);
262 unsigned int hash2
, slot2
, slot
= udp_hashfn(net
, hnum
, udptable
->mask
);
263 struct udp_hslot
*hslot2
, *hslot
= &udptable
->hash
[slot
];
264 int score
, badness
, matches
= 0, reuseport
= 0;
268 if (hslot
->count
> 10) {
269 hash2
= udp6_portaddr_hash(net
, daddr
, hnum
);
270 slot2
= hash2
& udptable
->mask
;
271 hslot2
= &udptable
->hash2
[slot2
];
272 if (hslot
->count
< hslot2
->count
)
275 result
= udp6_lib_lookup2(net
, saddr
, sport
,
279 hash2
= udp6_portaddr_hash(net
, &in6addr_any
, hnum
);
280 slot2
= hash2
& udptable
->mask
;
281 hslot2
= &udptable
->hash2
[slot2
];
282 if (hslot
->count
< hslot2
->count
)
285 result
= udp6_lib_lookup2(net
, saddr
, sport
,
286 &in6addr_any
, hnum
, dif
,
295 sk_nulls_for_each_rcu(sk
, node
, &hslot
->head
) {
296 score
= compute_score(sk
, net
, hnum
, saddr
, sport
, daddr
, dport
, dif
);
297 if (score
> badness
) {
300 reuseport
= sk
->sk_reuseport
;
302 hash
= inet6_ehashfn(net
, daddr
, hnum
,
306 } else if (score
== badness
&& reuseport
) {
308 if (((u64
)hash
* matches
) >> 32 == 0)
310 hash
= next_pseudo_random32(hash
);
314 * if the nulls value we got at the end of this lookup is
315 * not the expected one, we must restart lookup.
316 * We probably met an item that was moved to another chain.
318 if (get_nulls_value(node
) != slot
)
322 if (unlikely(!atomic_inc_not_zero_hint(&result
->sk_refcnt
, 2)))
324 else if (unlikely(compute_score(result
, net
, hnum
, saddr
, sport
,
325 daddr
, dport
, dif
) < badness
)) {
333 EXPORT_SYMBOL_GPL(__udp6_lib_lookup
);
335 static struct sock
*__udp6_lib_lookup_skb(struct sk_buff
*skb
,
336 __be16 sport
, __be16 dport
,
337 struct udp_table
*udptable
)
340 const struct ipv6hdr
*iph
= ipv6_hdr(skb
);
342 if (unlikely(sk
= skb_steal_sock(skb
)))
344 return __udp6_lib_lookup(dev_net(skb_dst(skb
)->dev
), &iph
->saddr
, sport
,
345 &iph
->daddr
, dport
, inet6_iif(skb
),
349 struct sock
*udp6_lib_lookup(struct net
*net
, const struct in6_addr
*saddr
, __be16 sport
,
350 const struct in6_addr
*daddr
, __be16 dport
, int dif
)
352 return __udp6_lib_lookup(net
, saddr
, sport
, daddr
, dport
, dif
, &udp_table
);
354 EXPORT_SYMBOL_GPL(udp6_lib_lookup
);
358 * This should be easy, if there is something there we
359 * return it, otherwise we block.
362 int udpv6_recvmsg(struct kiocb
*iocb
, struct sock
*sk
,
363 struct msghdr
*msg
, size_t len
,
364 int noblock
, int flags
, int *addr_len
)
366 struct ipv6_pinfo
*np
= inet6_sk(sk
);
367 struct inet_sock
*inet
= inet_sk(sk
);
369 unsigned int ulen
, copied
;
372 int is_udplite
= IS_UDPLITE(sk
);
373 bool checksum_valid
= false;
377 if (flags
& MSG_ERRQUEUE
)
378 return ipv6_recv_error(sk
, msg
, len
, addr_len
);
380 if (np
->rxpmtu
&& np
->rxopt
.bits
.rxpmtu
)
381 return ipv6_recv_rxpmtu(sk
, msg
, len
, addr_len
);
384 skb
= __skb_recv_datagram(sk
, flags
| (noblock
? MSG_DONTWAIT
: 0),
385 &peeked
, &off
, &err
);
389 ulen
= skb
->len
- sizeof(struct udphdr
);
393 else if (copied
< ulen
)
394 msg
->msg_flags
|= MSG_TRUNC
;
396 is_udp4
= (skb
->protocol
== htons(ETH_P_IP
));
399 * If checksum is needed at all, try to do it while copying the
400 * data. If the data is truncated, or if we only want a partial
401 * coverage checksum (UDP-Lite), do it before the copy.
404 if (copied
< ulen
|| UDP_SKB_CB(skb
)->partial_cov
) {
405 checksum_valid
= !udp_lib_checksum_complete(skb
);
410 if (checksum_valid
|| skb_csum_unnecessary(skb
))
411 err
= skb_copy_datagram_iovec(skb
, sizeof(struct udphdr
),
412 msg
->msg_iov
, copied
);
414 err
= skb_copy_and_csum_datagram_iovec(skb
, sizeof(struct udphdr
), msg
->msg_iov
);
419 trace_kfree_skb(skb
, udpv6_recvmsg
);
421 atomic_inc(&sk
->sk_drops
);
423 UDP_INC_STATS_USER(sock_net(sk
),
427 UDP6_INC_STATS_USER(sock_net(sk
),
435 UDP_INC_STATS_USER(sock_net(sk
),
436 UDP_MIB_INDATAGRAMS
, is_udplite
);
438 UDP6_INC_STATS_USER(sock_net(sk
),
439 UDP_MIB_INDATAGRAMS
, is_udplite
);
442 sock_recv_ts_and_drops(msg
, sk
, skb
);
444 /* Copy the address. */
446 struct sockaddr_in6
*sin6
;
448 sin6
= (struct sockaddr_in6
*) msg
->msg_name
;
449 sin6
->sin6_family
= AF_INET6
;
450 sin6
->sin6_port
= udp_hdr(skb
)->source
;
451 sin6
->sin6_flowinfo
= 0;
454 ipv6_addr_set_v4mapped(ip_hdr(skb
)->saddr
,
456 sin6
->sin6_scope_id
= 0;
458 sin6
->sin6_addr
= ipv6_hdr(skb
)->saddr
;
459 sin6
->sin6_scope_id
=
460 ipv6_iface_scope_id(&sin6
->sin6_addr
,
463 *addr_len
= sizeof(*sin6
);
466 if (inet
->cmsg_flags
)
467 ip_cmsg_recv(msg
, skb
);
470 ip6_datagram_recv_ctl(sk
, msg
, skb
);
474 if (flags
& MSG_TRUNC
)
478 skb_free_datagram_locked(sk
, skb
);
483 slow
= lock_sock_fast(sk
);
484 if (!skb_kill_datagram(sk
, skb
, flags
)) {
486 UDP_INC_STATS_USER(sock_net(sk
),
487 UDP_MIB_CSUMERRORS
, is_udplite
);
488 UDP_INC_STATS_USER(sock_net(sk
),
489 UDP_MIB_INERRORS
, is_udplite
);
491 UDP6_INC_STATS_USER(sock_net(sk
),
492 UDP_MIB_CSUMERRORS
, is_udplite
);
493 UDP6_INC_STATS_USER(sock_net(sk
),
494 UDP_MIB_INERRORS
, is_udplite
);
497 unlock_sock_fast(sk
, slow
);
499 /* starting over for a new packet, but check if we need to yield */
501 msg
->msg_flags
&= ~MSG_TRUNC
;
505 void __udp6_lib_err(struct sk_buff
*skb
, struct inet6_skb_parm
*opt
,
506 u8 type
, u8 code
, int offset
, __be32 info
,
507 struct udp_table
*udptable
)
509 struct ipv6_pinfo
*np
;
510 const struct ipv6hdr
*hdr
= (const struct ipv6hdr
*)skb
->data
;
511 const struct in6_addr
*saddr
= &hdr
->saddr
;
512 const struct in6_addr
*daddr
= &hdr
->daddr
;
513 struct udphdr
*uh
= (struct udphdr
*)(skb
->data
+offset
);
517 sk
= __udp6_lib_lookup(dev_net(skb
->dev
), daddr
, uh
->dest
,
518 saddr
, uh
->source
, inet6_iif(skb
), udptable
);
522 if (type
== ICMPV6_PKT_TOOBIG
)
523 ip6_sk_update_pmtu(skb
, sk
, info
);
524 if (type
== NDISC_REDIRECT
)
525 ip6_sk_redirect(skb
, sk
);
529 if (!icmpv6_err_convert(type
, code
, &err
) && !np
->recverr
)
532 if (sk
->sk_state
!= TCP_ESTABLISHED
&& !np
->recverr
)
536 ipv6_icmp_error(sk
, skb
, err
, uh
->dest
, ntohl(info
), (u8
*)(uh
+1));
539 sk
->sk_error_report(sk
);
544 static int __udpv6_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
548 if (!ipv6_addr_any(&inet6_sk(sk
)->daddr
))
549 sock_rps_save_rxhash(sk
, skb
);
551 rc
= sock_queue_rcv_skb(sk
, skb
);
553 int is_udplite
= IS_UDPLITE(sk
);
555 /* Note that an ENOMEM error is charged twice */
557 UDP6_INC_STATS_BH(sock_net(sk
),
558 UDP_MIB_RCVBUFERRORS
, is_udplite
);
559 UDP6_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
566 static __inline__
void udpv6_err(struct sk_buff
*skb
,
567 struct inet6_skb_parm
*opt
, u8 type
,
568 u8 code
, int offset
, __be32 info
)
570 __udp6_lib_err(skb
, opt
, type
, code
, offset
, info
, &udp_table
);
573 static struct static_key udpv6_encap_needed __read_mostly
;
574 void udpv6_encap_enable(void)
576 if (!static_key_enabled(&udpv6_encap_needed
))
577 static_key_slow_inc(&udpv6_encap_needed
);
579 EXPORT_SYMBOL(udpv6_encap_enable
);
581 int udpv6_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
583 struct udp_sock
*up
= udp_sk(sk
);
585 int is_udplite
= IS_UDPLITE(sk
);
587 if (!xfrm6_policy_check(sk
, XFRM_POLICY_IN
, skb
))
590 if (static_key_false(&udpv6_encap_needed
) && up
->encap_type
) {
591 int (*encap_rcv
)(struct sock
*sk
, struct sk_buff
*skb
);
594 * This is an encapsulation socket so pass the skb to
595 * the socket's udp_encap_rcv() hook. Otherwise, just
596 * fall through and pass this up the UDP socket.
597 * up->encap_rcv() returns the following value:
598 * =0 if skb was successfully passed to the encap
599 * handler or was discarded by it.
600 * >0 if skb should be passed on to UDP.
601 * <0 if skb should be resubmitted as proto -N
604 /* if we're overly short, let UDP handle it */
605 encap_rcv
= ACCESS_ONCE(up
->encap_rcv
);
606 if (skb
->len
> sizeof(struct udphdr
) && encap_rcv
!= NULL
) {
609 ret
= encap_rcv(sk
, skb
);
611 UDP_INC_STATS_BH(sock_net(sk
),
618 /* FALLTHROUGH -- it's a UDP Packet */
622 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
624 if ((is_udplite
& UDPLITE_RECV_CC
) && UDP_SKB_CB(skb
)->partial_cov
) {
626 if (up
->pcrlen
== 0) { /* full coverage was set */
627 LIMIT_NETDEBUG(KERN_WARNING
"UDPLITE6: partial coverage"
628 " %d while full coverage %d requested\n",
629 UDP_SKB_CB(skb
)->cscov
, skb
->len
);
632 if (UDP_SKB_CB(skb
)->cscov
< up
->pcrlen
) {
633 LIMIT_NETDEBUG(KERN_WARNING
"UDPLITE6: coverage %d "
634 "too small, need min %d\n",
635 UDP_SKB_CB(skb
)->cscov
, up
->pcrlen
);
640 if (rcu_access_pointer(sk
->sk_filter
)) {
641 if (udp_lib_checksum_complete(skb
))
645 if (sk_rcvqueues_full(sk
, skb
, sk
->sk_rcvbuf
))
652 if (!sock_owned_by_user(sk
))
653 rc
= __udpv6_queue_rcv_skb(sk
, skb
);
654 else if (sk_add_backlog(sk
, skb
, sk
->sk_rcvbuf
)) {
662 UDP6_INC_STATS_BH(sock_net(sk
), UDP_MIB_CSUMERRORS
, is_udplite
);
664 UDP6_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
665 atomic_inc(&sk
->sk_drops
);
670 static struct sock
*udp_v6_mcast_next(struct net
*net
, struct sock
*sk
,
671 __be16 loc_port
, const struct in6_addr
*loc_addr
,
672 __be16 rmt_port
, const struct in6_addr
*rmt_addr
,
675 struct hlist_nulls_node
*node
;
677 unsigned short num
= ntohs(loc_port
);
679 sk_nulls_for_each_from(s
, node
) {
680 struct inet_sock
*inet
= inet_sk(s
);
682 if (!net_eq(sock_net(s
), net
))
685 if (udp_sk(s
)->udp_port_hash
== num
&&
686 s
->sk_family
== PF_INET6
) {
687 struct ipv6_pinfo
*np
= inet6_sk(s
);
688 if (inet
->inet_dport
) {
689 if (inet
->inet_dport
!= rmt_port
)
692 if (!ipv6_addr_any(&np
->daddr
) &&
693 !ipv6_addr_equal(&np
->daddr
, rmt_addr
))
696 if (s
->sk_bound_dev_if
&& s
->sk_bound_dev_if
!= dif
)
699 if (!ipv6_addr_any(&np
->rcv_saddr
)) {
700 if (!ipv6_addr_equal(&np
->rcv_saddr
, loc_addr
))
703 if (!inet6_mc_check(s
, loc_addr
, rmt_addr
))
711 static void flush_stack(struct sock
**stack
, unsigned int count
,
712 struct sk_buff
*skb
, unsigned int final
)
714 struct sk_buff
*skb1
= NULL
;
718 for (i
= 0; i
< count
; i
++) {
720 if (likely(skb1
== NULL
))
721 skb1
= (i
== final
) ? skb
: skb_clone(skb
, GFP_ATOMIC
);
723 atomic_inc(&sk
->sk_drops
);
724 UDP6_INC_STATS_BH(sock_net(sk
), UDP_MIB_RCVBUFERRORS
,
726 UDP6_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
,
730 if (skb1
&& udpv6_queue_rcv_skb(sk
, skb1
) <= 0)
737 * Note: called only from the BH handler context,
738 * so we don't need to lock the hashes.
740 static int __udp6_lib_mcast_deliver(struct net
*net
, struct sk_buff
*skb
,
741 const struct in6_addr
*saddr
, const struct in6_addr
*daddr
,
742 struct udp_table
*udptable
)
744 struct sock
*sk
, *stack
[256 / sizeof(struct sock
*)];
745 const struct udphdr
*uh
= udp_hdr(skb
);
746 struct udp_hslot
*hslot
= udp_hashslot(udptable
, net
, ntohs(uh
->dest
));
748 unsigned int i
, count
= 0;
750 spin_lock(&hslot
->lock
);
751 sk
= sk_nulls_head(&hslot
->head
);
752 dif
= inet6_iif(skb
);
753 sk
= udp_v6_mcast_next(net
, sk
, uh
->dest
, daddr
, uh
->source
, saddr
, dif
);
756 sk
= udp_v6_mcast_next(net
, sk_nulls_next(sk
), uh
->dest
, daddr
,
757 uh
->source
, saddr
, dif
);
758 if (unlikely(count
== ARRAY_SIZE(stack
))) {
761 flush_stack(stack
, count
, skb
, ~0);
766 * before releasing the lock, we must take reference on sockets
768 for (i
= 0; i
< count
; i
++)
771 spin_unlock(&hslot
->lock
);
774 flush_stack(stack
, count
, skb
, count
- 1);
776 for (i
= 0; i
< count
; i
++)
784 int __udp6_lib_rcv(struct sk_buff
*skb
, struct udp_table
*udptable
,
787 struct net
*net
= dev_net(skb
->dev
);
790 const struct in6_addr
*saddr
, *daddr
;
793 if (!pskb_may_pull(skb
, sizeof(struct udphdr
)))
796 saddr
= &ipv6_hdr(skb
)->saddr
;
797 daddr
= &ipv6_hdr(skb
)->daddr
;
800 ulen
= ntohs(uh
->len
);
804 if (proto
== IPPROTO_UDP
) {
805 /* UDP validates ulen. */
807 /* Check for jumbo payload */
811 if (ulen
< sizeof(*uh
))
814 if (ulen
< skb
->len
) {
815 if (pskb_trim_rcsum(skb
, ulen
))
817 saddr
= &ipv6_hdr(skb
)->saddr
;
818 daddr
= &ipv6_hdr(skb
)->daddr
;
823 if (udp6_csum_init(skb
, uh
, proto
))
827 * Multicast receive code
829 if (ipv6_addr_is_multicast(daddr
))
830 return __udp6_lib_mcast_deliver(net
, skb
,
831 saddr
, daddr
, udptable
);
836 * check socket cache ... must talk to Alan about his plans
837 * for sock caches... i'll skip this for now.
839 sk
= __udp6_lib_lookup_skb(skb
, uh
->source
, uh
->dest
, udptable
);
841 int ret
= udpv6_queue_rcv_skb(sk
, skb
);
844 /* a return value > 0 means to resubmit the input, but
845 * it wants the return to be -protocol, or 0
853 if (!xfrm6_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
856 if (udp_lib_checksum_complete(skb
))
859 UDP6_INC_STATS_BH(net
, UDP_MIB_NOPORTS
, proto
== IPPROTO_UDPLITE
);
860 icmpv6_send(skb
, ICMPV6_DEST_UNREACH
, ICMPV6_PORT_UNREACH
, 0);
866 LIMIT_NETDEBUG(KERN_DEBUG
"UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
867 proto
== IPPROTO_UDPLITE
? "-Lite" : "",
876 UDP6_INC_STATS_BH(net
, UDP_MIB_CSUMERRORS
, proto
== IPPROTO_UDPLITE
);
878 UDP6_INC_STATS_BH(net
, UDP_MIB_INERRORS
, proto
== IPPROTO_UDPLITE
);
883 static __inline__
int udpv6_rcv(struct sk_buff
*skb
)
885 return __udp6_lib_rcv(skb
, &udp_table
, IPPROTO_UDP
);
889 * Throw away all pending data and cancel the corking. Socket is locked.
891 static void udp_v6_flush_pending_frames(struct sock
*sk
)
893 struct udp_sock
*up
= udp_sk(sk
);
895 if (up
->pending
== AF_INET
)
896 udp_flush_pending_frames(sk
);
897 else if (up
->pending
) {
900 ip6_flush_pending_frames(sk
);
905 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
906 * @sk: socket we are sending on
907 * @skb: sk_buff containing the filled-in UDP header
908 * (checksum field must be zeroed out)
910 static void udp6_hwcsum_outgoing(struct sock
*sk
, struct sk_buff
*skb
,
911 const struct in6_addr
*saddr
,
912 const struct in6_addr
*daddr
, int len
)
915 struct udphdr
*uh
= udp_hdr(skb
);
918 if (skb_queue_len(&sk
->sk_write_queue
) == 1) {
919 /* Only one fragment on the socket. */
920 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
921 skb
->csum_offset
= offsetof(struct udphdr
, check
);
922 uh
->check
= ~csum_ipv6_magic(saddr
, daddr
, len
, IPPROTO_UDP
, 0);
925 * HW-checksum won't work as there are two or more
926 * fragments on the socket so that all csums of sk_buffs
929 offset
= skb_transport_offset(skb
);
930 skb
->csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
932 skb
->ip_summed
= CHECKSUM_NONE
;
934 skb_queue_walk(&sk
->sk_write_queue
, skb
) {
935 csum
= csum_add(csum
, skb
->csum
);
938 uh
->check
= csum_ipv6_magic(saddr
, daddr
, len
, IPPROTO_UDP
,
941 uh
->check
= CSUM_MANGLED_0
;
949 static int udp_v6_push_pending_frames(struct sock
*sk
)
953 struct udp_sock
*up
= udp_sk(sk
);
954 struct inet_sock
*inet
= inet_sk(sk
);
957 int is_udplite
= IS_UDPLITE(sk
);
960 if (up
->pending
== AF_INET
)
961 return udp_push_pending_frames(sk
);
963 fl6
= &inet
->cork
.fl
.u
.ip6
;
965 /* Grab the skbuff where UDP header space exists. */
966 if ((skb
= skb_peek(&sk
->sk_write_queue
)) == NULL
)
970 * Create a UDP header
973 uh
->source
= fl6
->fl6_sport
;
974 uh
->dest
= fl6
->fl6_dport
;
975 uh
->len
= htons(up
->len
);
979 csum
= udplite_csum_outgoing(sk
, skb
);
980 else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) { /* UDP hardware csum */
981 udp6_hwcsum_outgoing(sk
, skb
, &fl6
->saddr
, &fl6
->daddr
,
985 csum
= udp_csum_outgoing(sk
, skb
);
987 /* add protocol-dependent pseudo-header */
988 uh
->check
= csum_ipv6_magic(&fl6
->saddr
, &fl6
->daddr
,
989 up
->len
, fl6
->flowi6_proto
, csum
);
991 uh
->check
= CSUM_MANGLED_0
;
994 err
= ip6_push_pending_frames(sk
);
996 if (err
== -ENOBUFS
&& !inet6_sk(sk
)->recverr
) {
997 UDP6_INC_STATS_USER(sock_net(sk
),
998 UDP_MIB_SNDBUFERRORS
, is_udplite
);
1002 UDP6_INC_STATS_USER(sock_net(sk
),
1003 UDP_MIB_OUTDATAGRAMS
, is_udplite
);
1010 int udpv6_sendmsg(struct kiocb
*iocb
, struct sock
*sk
,
1011 struct msghdr
*msg
, size_t len
)
1013 struct ipv6_txoptions opt_space
;
1014 struct udp_sock
*up
= udp_sk(sk
);
1015 struct inet_sock
*inet
= inet_sk(sk
);
1016 struct ipv6_pinfo
*np
= inet6_sk(sk
);
1017 struct sockaddr_in6
*sin6
= (struct sockaddr_in6
*) msg
->msg_name
;
1018 struct in6_addr
*daddr
, *final_p
, final
;
1019 struct ipv6_txoptions
*opt
= NULL
;
1020 struct ip6_flowlabel
*flowlabel
= NULL
;
1022 struct dst_entry
*dst
;
1023 int addr_len
= msg
->msg_namelen
;
1028 int corkreq
= up
->corkflag
|| msg
->msg_flags
&MSG_MORE
;
1031 int is_udplite
= IS_UDPLITE(sk
);
1032 int (*getfrag
)(void *, char *, int, int, int, struct sk_buff
*);
1034 /* destination address check */
1036 if (addr_len
< offsetof(struct sockaddr
, sa_data
))
1039 switch (sin6
->sin6_family
) {
1041 if (addr_len
< SIN6_LEN_RFC2133
)
1043 daddr
= &sin6
->sin6_addr
;
1046 goto do_udp_sendmsg
;
1048 msg
->msg_name
= sin6
= NULL
;
1049 msg
->msg_namelen
= addr_len
= 0;
1055 } else if (!up
->pending
) {
1056 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1057 return -EDESTADDRREQ
;
1063 if (ipv6_addr_v4mapped(daddr
)) {
1064 struct sockaddr_in sin
;
1065 sin
.sin_family
= AF_INET
;
1066 sin
.sin_port
= sin6
? sin6
->sin6_port
: inet
->inet_dport
;
1067 sin
.sin_addr
.s_addr
= daddr
->s6_addr32
[3];
1068 msg
->msg_name
= &sin
;
1069 msg
->msg_namelen
= sizeof(sin
);
1071 if (__ipv6_only_sock(sk
))
1072 return -ENETUNREACH
;
1073 return udp_sendmsg(iocb
, sk
, msg
, len
);
1077 if (up
->pending
== AF_INET
)
1078 return udp_sendmsg(iocb
, sk
, msg
, len
);
1080 /* Rough check on arithmetic overflow,
1081 better check is made in ip6_append_data().
1083 if (len
> INT_MAX
- sizeof(struct udphdr
))
1088 * There are pending frames.
1089 * The socket lock must be held while it's corked.
1092 if (likely(up
->pending
)) {
1093 if (unlikely(up
->pending
!= AF_INET6
)) {
1095 return -EAFNOSUPPORT
;
1098 goto do_append_data
;
1102 ulen
+= sizeof(struct udphdr
);
1104 memset(&fl6
, 0, sizeof(fl6
));
1107 if (sin6
->sin6_port
== 0)
1110 fl6
.fl6_dport
= sin6
->sin6_port
;
1111 daddr
= &sin6
->sin6_addr
;
1114 fl6
.flowlabel
= sin6
->sin6_flowinfo
&IPV6_FLOWINFO_MASK
;
1115 if (fl6
.flowlabel
&IPV6_FLOWLABEL_MASK
) {
1116 flowlabel
= fl6_sock_lookup(sk
, fl6
.flowlabel
);
1117 if (flowlabel
== NULL
)
1119 daddr
= &flowlabel
->dst
;
1124 * Otherwise it will be difficult to maintain
1127 if (sk
->sk_state
== TCP_ESTABLISHED
&&
1128 ipv6_addr_equal(daddr
, &np
->daddr
))
1131 if (addr_len
>= sizeof(struct sockaddr_in6
) &&
1132 sin6
->sin6_scope_id
&&
1133 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr
)))
1134 fl6
.flowi6_oif
= sin6
->sin6_scope_id
;
1136 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1137 return -EDESTADDRREQ
;
1139 fl6
.fl6_dport
= inet
->inet_dport
;
1141 fl6
.flowlabel
= np
->flow_label
;
1145 if (!fl6
.flowi6_oif
)
1146 fl6
.flowi6_oif
= sk
->sk_bound_dev_if
;
1148 if (!fl6
.flowi6_oif
)
1149 fl6
.flowi6_oif
= np
->sticky_pktinfo
.ipi6_ifindex
;
1151 fl6
.flowi6_mark
= sk
->sk_mark
;
1152 fl6
.flowi6_uid
= sock_i_uid(sk
);
1154 if (msg
->msg_controllen
) {
1156 memset(opt
, 0, sizeof(struct ipv6_txoptions
));
1157 opt
->tot_len
= sizeof(*opt
);
1159 err
= ip6_datagram_send_ctl(sock_net(sk
), sk
, msg
, &fl6
, opt
,
1160 &hlimit
, &tclass
, &dontfrag
);
1162 fl6_sock_release(flowlabel
);
1165 if ((fl6
.flowlabel
&IPV6_FLOWLABEL_MASK
) && !flowlabel
) {
1166 flowlabel
= fl6_sock_lookup(sk
, fl6
.flowlabel
);
1167 if (flowlabel
== NULL
)
1170 if (!(opt
->opt_nflen
|opt
->opt_flen
))
1177 opt
= fl6_merge_options(&opt_space
, flowlabel
, opt
);
1178 opt
= ipv6_fixup_options(&opt_space
, opt
);
1180 fl6
.flowi6_proto
= sk
->sk_protocol
;
1181 if (!ipv6_addr_any(daddr
))
1184 fl6
.daddr
.s6_addr
[15] = 0x1; /* :: means loopback (BSD'ism) */
1185 if (ipv6_addr_any(&fl6
.saddr
) && !ipv6_addr_any(&np
->saddr
))
1186 fl6
.saddr
= np
->saddr
;
1187 fl6
.fl6_sport
= inet
->inet_sport
;
1189 final_p
= fl6_update_dst(&fl6
, opt
, &final
);
1193 if (!fl6
.flowi6_oif
&& ipv6_addr_is_multicast(&fl6
.daddr
)) {
1194 fl6
.flowi6_oif
= np
->mcast_oif
;
1196 } else if (!fl6
.flowi6_oif
)
1197 fl6
.flowi6_oif
= np
->ucast_oif
;
1199 security_sk_classify_flow(sk
, flowi6_to_flowi(&fl6
));
1201 dst
= ip6_sk_dst_lookup_flow(sk
, &fl6
, final_p
, true);
1209 if (ipv6_addr_is_multicast(&fl6
.daddr
))
1210 hlimit
= np
->mcast_hops
;
1212 hlimit
= np
->hop_limit
;
1214 hlimit
= ip6_dst_hoplimit(dst
);
1218 tclass
= np
->tclass
;
1221 dontfrag
= np
->dontfrag
;
1223 if (msg
->msg_flags
&MSG_CONFIRM
)
1228 if (unlikely(up
->pending
)) {
1229 /* The socket is already corked while preparing it. */
1230 /* ... which is an evident application bug. --ANK */
1233 LIMIT_NETDEBUG(KERN_DEBUG
"udp cork app bug 2\n");
1238 up
->pending
= AF_INET6
;
1242 getfrag
= is_udplite
? udplite_getfrag
: ip_generic_getfrag
;
1243 err
= ip6_append_data(sk
, getfrag
, msg
->msg_iov
, ulen
,
1244 sizeof(struct udphdr
), hlimit
, tclass
, opt
, &fl6
,
1245 (struct rt6_info
*)dst
,
1246 corkreq
? msg
->msg_flags
|MSG_MORE
: msg
->msg_flags
, dontfrag
);
1248 udp_v6_flush_pending_frames(sk
);
1250 err
= udp_v6_push_pending_frames(sk
);
1251 else if (unlikely(skb_queue_empty(&sk
->sk_write_queue
)))
1256 ip6_dst_store(sk
, dst
,
1257 ipv6_addr_equal(&fl6
.daddr
, &np
->daddr
) ?
1259 #ifdef CONFIG_IPV6_SUBTREES
1260 ipv6_addr_equal(&fl6
.saddr
, &np
->saddr
) ?
1271 err
= np
->recverr
? net_xmit_errno(err
) : 0;
1275 fl6_sock_release(flowlabel
);
1279 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1280 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1281 * we don't have a good statistic (IpOutDiscards but it can be too many
1282 * things). We could add another new stat but at least for now that
1283 * seems like overkill.
1286 if (err
== -ENOBUFS
|| (sk
->sk_socket
&& test_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
)) ) {
1287 UDP6_INC_STATS_USER(sock_net(sk
),
1288 UDP_MIB_SNDBUFERRORS
, is_udplite
);
1294 if (!(msg
->msg_flags
&MSG_PROBE
) || len
)
1295 goto back_from_confirm
;
1300 void udpv6_destroy_sock(struct sock
*sk
)
1302 struct udp_sock
*up
= udp_sk(sk
);
1304 udp_v6_flush_pending_frames(sk
);
1307 if (static_key_false(&udpv6_encap_needed
) && up
->encap_type
) {
1308 void (*encap_destroy
)(struct sock
*sk
);
1309 encap_destroy
= ACCESS_ONCE(up
->encap_destroy
);
1314 inet6_destroy_sock(sk
);
1318 * Socket option code for UDP
1320 int udpv6_setsockopt(struct sock
*sk
, int level
, int optname
,
1321 char __user
*optval
, unsigned int optlen
)
1323 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1324 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1325 udp_v6_push_pending_frames
);
1326 return ipv6_setsockopt(sk
, level
, optname
, optval
, optlen
);
1329 #ifdef CONFIG_COMPAT
1330 int compat_udpv6_setsockopt(struct sock
*sk
, int level
, int optname
,
1331 char __user
*optval
, unsigned int optlen
)
1333 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1334 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1335 udp_v6_push_pending_frames
);
1336 return compat_ipv6_setsockopt(sk
, level
, optname
, optval
, optlen
);
1340 int udpv6_getsockopt(struct sock
*sk
, int level
, int optname
,
1341 char __user
*optval
, int __user
*optlen
)
1343 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1344 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1345 return ipv6_getsockopt(sk
, level
, optname
, optval
, optlen
);
1348 #ifdef CONFIG_COMPAT
1349 int compat_udpv6_getsockopt(struct sock
*sk
, int level
, int optname
,
1350 char __user
*optval
, int __user
*optlen
)
1352 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1353 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1354 return compat_ipv6_getsockopt(sk
, level
, optname
, optval
, optlen
);
1358 static const struct inet6_protocol udpv6_protocol
= {
1359 .handler
= udpv6_rcv
,
1360 .err_handler
= udpv6_err
,
1361 .flags
= INET6_PROTO_NOPOLICY
|INET6_PROTO_FINAL
,
1364 /* ------------------------------------------------------------------------ */
1365 #ifdef CONFIG_PROC_FS
1367 static void udp6_sock_seq_show(struct seq_file
*seq
, struct sock
*sp
, int bucket
)
1369 struct inet_sock
*inet
= inet_sk(sp
);
1370 struct ipv6_pinfo
*np
= inet6_sk(sp
);
1371 const struct in6_addr
*dest
, *src
;
1375 src
= &np
->rcv_saddr
;
1376 destp
= ntohs(inet
->inet_dport
);
1377 srcp
= ntohs(inet
->inet_sport
);
1379 "%5d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1380 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d\n",
1382 src
->s6_addr32
[0], src
->s6_addr32
[1],
1383 src
->s6_addr32
[2], src
->s6_addr32
[3], srcp
,
1384 dest
->s6_addr32
[0], dest
->s6_addr32
[1],
1385 dest
->s6_addr32
[2], dest
->s6_addr32
[3], destp
,
1387 sk_wmem_alloc_get(sp
),
1388 sk_rmem_alloc_get(sp
),
1390 from_kuid_munged(seq_user_ns(seq
), sock_i_uid(sp
)),
1393 atomic_read(&sp
->sk_refcnt
), sp
,
1394 atomic_read(&sp
->sk_drops
));
1397 int udp6_seq_show(struct seq_file
*seq
, void *v
)
1399 if (v
== SEQ_START_TOKEN
)
1404 "st tx_queue rx_queue tr tm->when retrnsmt"
1405 " uid timeout inode ref pointer drops\n");
1407 udp6_sock_seq_show(seq
, v
, ((struct udp_iter_state
*)seq
->private)->bucket
);
1411 static const struct file_operations udp6_afinfo_seq_fops
= {
1412 .owner
= THIS_MODULE
,
1413 .open
= udp_seq_open
,
1415 .llseek
= seq_lseek
,
1416 .release
= seq_release_net
1419 static struct udp_seq_afinfo udp6_seq_afinfo
= {
1422 .udp_table
= &udp_table
,
1423 .seq_fops
= &udp6_afinfo_seq_fops
,
1425 .show
= udp6_seq_show
,
1429 int __net_init
udp6_proc_init(struct net
*net
)
1431 return udp_proc_register(net
, &udp6_seq_afinfo
);
1434 void udp6_proc_exit(struct net
*net
) {
1435 udp_proc_unregister(net
, &udp6_seq_afinfo
);
1437 #endif /* CONFIG_PROC_FS */
1439 void udp_v6_clear_sk(struct sock
*sk
, int size
)
1441 struct inet_sock
*inet
= inet_sk(sk
);
1443 /* we do not want to clear pinet6 field, because of RCU lookups */
1444 sk_prot_clear_portaddr_nulls(sk
, offsetof(struct inet_sock
, pinet6
));
1446 size
-= offsetof(struct inet_sock
, pinet6
) + sizeof(inet
->pinet6
);
1447 memset(&inet
->pinet6
+ 1, 0, size
);
1450 /* ------------------------------------------------------------------------ */
1452 struct proto udpv6_prot
= {
1454 .owner
= THIS_MODULE
,
1455 .close
= udp_lib_close
,
1456 .connect
= ip6_datagram_connect
,
1457 .disconnect
= udp_disconnect
,
1459 .destroy
= udpv6_destroy_sock
,
1460 .setsockopt
= udpv6_setsockopt
,
1461 .getsockopt
= udpv6_getsockopt
,
1462 .sendmsg
= udpv6_sendmsg
,
1463 .recvmsg
= udpv6_recvmsg
,
1464 .backlog_rcv
= __udpv6_queue_rcv_skb
,
1465 .hash
= udp_lib_hash
,
1466 .unhash
= udp_lib_unhash
,
1467 .rehash
= udp_v6_rehash
,
1468 .get_port
= udp_v6_get_port
,
1469 .memory_allocated
= &udp_memory_allocated
,
1470 .sysctl_mem
= sysctl_udp_mem
,
1471 .sysctl_wmem
= &sysctl_udp_wmem_min
,
1472 .sysctl_rmem
= &sysctl_udp_rmem_min
,
1473 .obj_size
= sizeof(struct udp6_sock
),
1474 .slab_flags
= SLAB_DESTROY_BY_RCU
,
1475 .h
.udp_table
= &udp_table
,
1476 #ifdef CONFIG_COMPAT
1477 .compat_setsockopt
= compat_udpv6_setsockopt
,
1478 .compat_getsockopt
= compat_udpv6_getsockopt
,
1480 .clear_sk
= udp_v6_clear_sk
,
1483 static struct inet_protosw udpv6_protosw
= {
1485 .protocol
= IPPROTO_UDP
,
1486 .prot
= &udpv6_prot
,
1487 .ops
= &inet6_dgram_ops
,
1488 .no_check
= UDP_CSUM_DEFAULT
,
1489 .flags
= INET_PROTOSW_PERMANENT
,
1493 int __init
udpv6_init(void)
1497 ret
= inet6_add_protocol(&udpv6_protocol
, IPPROTO_UDP
);
1501 ret
= inet6_register_protosw(&udpv6_protosw
);
1503 goto out_udpv6_protocol
;
1508 inet6_del_protocol(&udpv6_protocol
, IPPROTO_UDP
);
1512 void udpv6_exit(void)
1514 inet6_unregister_protosw(&udpv6_protosw
);
1515 inet6_del_protocol(&udpv6_protocol
, IPPROTO_UDP
);