3 * Linux INET6 implementation
6 * Pedro Roque <roque@di.fc.ul.pt>
8 * Based on linux/ipv4/udp.c
11 * Hideaki YOSHIFUJI : sin6_scope_id support
12 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
13 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
14 * a single port at the same time.
15 * Kazunori MIYAZAWA @USAGI: change process style to use ip6_append_data
16 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/udp6 to seq_file.
18 * This program is free software; you can redistribute it and/or
19 * modify it under the terms of the GNU General Public License
20 * as published by the Free Software Foundation; either version
21 * 2 of the License, or (at your option) any later version.
24 #include <linux/errno.h>
25 #include <linux/types.h>
26 #include <linux/socket.h>
27 #include <linux/sockios.h>
28 #include <linux/net.h>
29 #include <linux/in6.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_arp.h>
32 #include <linux/ipv6.h>
33 #include <linux/icmpv6.h>
34 #include <linux/init.h>
35 #include <linux/module.h>
36 #include <linux/skbuff.h>
37 #include <linux/slab.h>
38 #include <asm/uaccess.h>
40 #include <net/ndisc.h>
41 #include <net/protocol.h>
42 #include <net/transp_v6.h>
43 #include <net/ip6_route.h>
45 #include <net/tcp_states.h>
46 #include <net/ip6_checksum.h>
48 #include <net/inet6_hashtables.h>
49 #include <net/busy_poll.h>
51 #include <linux/proc_fs.h>
52 #include <linux/seq_file.h>
53 #include <trace/events/skb.h>
56 static unsigned int udp6_ehashfn(struct net *net,
57 const struct in6_addr *laddr,
59 const struct in6_addr *faddr,
62 static u32 udp6_ehash_secret __read_mostly;
63 static u32 udp_ipv6_hash_secret __read_mostly;
67 net_get_random_once(&udp6_ehash_secret,
68 sizeof(udp6_ehash_secret));
69 net_get_random_once(&udp_ipv6_hash_secret,
70 sizeof(udp_ipv6_hash_secret));
72 lhash = (__force u32)laddr->s6_addr32[3];
73 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
75 return __inet6_ehashfn(lhash, lport, fhash, fport,
76 udp_ipv6_hash_secret + net_hash_mix(net));
79 int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2)
81 const struct in6_addr *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2);
82 int sk_ipv6only = ipv6_only_sock(sk);
83 int sk2_ipv6only = inet_v6_ipv6only(sk2);
84 int addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
85 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
87 /* if both are mapped, treat as IPv4 */
88 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED)
89 return (!sk2_ipv6only &&
90 (!sk->sk_rcv_saddr || !sk2->sk_rcv_saddr ||
91 sk->sk_rcv_saddr == sk2->sk_rcv_saddr));
93 if (addr_type2 == IPV6_ADDR_ANY &&
94 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
97 if (addr_type == IPV6_ADDR_ANY &&
98 !(sk_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
101 if (sk2_rcv_saddr6 &&
102 ipv6_addr_equal(&sk->sk_v6_rcv_saddr, sk2_rcv_saddr6))
108 static unsigned int udp6_portaddr_hash(struct net *net,
109 const struct in6_addr *addr6,
112 unsigned int hash, mix = net_hash_mix(net);
114 if (ipv6_addr_any(addr6))
115 hash = jhash_1word(0, mix);
116 else if (ipv6_addr_v4mapped(addr6))
117 hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix);
119 hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix);
125 int udp_v6_get_port(struct sock *sk, unsigned short snum)
127 unsigned int hash2_nulladdr =
128 udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
129 unsigned int hash2_partial =
130 udp6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
132 /* precompute partial secondary hash */
133 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
134 return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal, hash2_nulladdr);
137 static void udp_v6_rehash(struct sock *sk)
139 u16 new_hash = udp6_portaddr_hash(sock_net(sk),
140 &sk->sk_v6_rcv_saddr,
141 inet_sk(sk)->inet_num);
143 udp_lib_rehash(sk, new_hash);
146 static inline int compute_score(struct sock *sk, struct net *net,
148 const struct in6_addr *saddr, __be16 sport,
149 const struct in6_addr *daddr, __be16 dport,
154 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
155 sk->sk_family == PF_INET6) {
156 struct inet_sock *inet = inet_sk(sk);
159 if (inet->inet_dport) {
160 if (inet->inet_dport != sport)
164 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
165 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
169 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
170 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
174 if (sk->sk_bound_dev_if) {
175 if (sk->sk_bound_dev_if != dif)
183 #define SCORE2_MAX (1 + 1 + 1)
184 static inline int compute_score2(struct sock *sk, struct net *net,
185 const struct in6_addr *saddr, __be16 sport,
186 const struct in6_addr *daddr, unsigned short hnum,
191 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
192 sk->sk_family == PF_INET6) {
193 struct inet_sock *inet = inet_sk(sk);
195 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
198 if (inet->inet_dport) {
199 if (inet->inet_dport != sport)
203 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
204 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
208 if (sk->sk_bound_dev_if) {
209 if (sk->sk_bound_dev_if != dif)
218 /* called with read_rcu_lock() */
219 static struct sock *udp6_lib_lookup2(struct net *net,
220 const struct in6_addr *saddr, __be16 sport,
221 const struct in6_addr *daddr, unsigned int hnum, int dif,
222 struct udp_hslot *hslot2, unsigned int slot2)
224 struct sock *sk, *result;
225 struct hlist_nulls_node *node;
226 int score, badness, matches = 0, reuseport = 0;
232 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
233 score = compute_score2(sk, net, saddr, sport,
235 if (score > badness) {
238 reuseport = sk->sk_reuseport;
240 hash = udp6_ehashfn(net, daddr, hnum,
243 } else if (score == SCORE2_MAX)
245 } else if (score == badness && reuseport) {
247 if (((u64)hash * matches) >> 32 == 0)
249 hash = next_pseudo_random32(hash);
253 * if the nulls value we got at the end of this lookup is
254 * not the expected one, we must restart lookup.
255 * We probably met an item that was moved to another chain.
257 if (get_nulls_value(node) != slot2)
262 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
264 else if (unlikely(compute_score2(result, net, saddr, sport,
265 daddr, hnum, dif) < badness)) {
273 struct sock *__udp6_lib_lookup(struct net *net,
274 const struct in6_addr *saddr, __be16 sport,
275 const struct in6_addr *daddr, __be16 dport,
276 int dif, struct udp_table *udptable)
278 struct sock *sk, *result;
279 struct hlist_nulls_node *node;
280 unsigned short hnum = ntohs(dport);
281 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
282 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
283 int score, badness, matches = 0, reuseport = 0;
287 if (hslot->count > 10) {
288 hash2 = udp6_portaddr_hash(net, daddr, hnum);
289 slot2 = hash2 & udptable->mask;
290 hslot2 = &udptable->hash2[slot2];
291 if (hslot->count < hslot2->count)
294 result = udp6_lib_lookup2(net, saddr, sport,
298 hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
299 slot2 = hash2 & udptable->mask;
300 hslot2 = &udptable->hash2[slot2];
301 if (hslot->count < hslot2->count)
304 result = udp6_lib_lookup2(net, saddr, sport,
305 &in6addr_any, hnum, dif,
314 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
315 score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif);
316 if (score > badness) {
319 reuseport = sk->sk_reuseport;
321 hash = udp6_ehashfn(net, daddr, hnum,
325 } else if (score == badness && reuseport) {
327 if (((u64)hash * matches) >> 32 == 0)
329 hash = next_pseudo_random32(hash);
333 * if the nulls value we got at the end of this lookup is
334 * not the expected one, we must restart lookup.
335 * We probably met an item that was moved to another chain.
337 if (get_nulls_value(node) != slot)
341 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
343 else if (unlikely(compute_score(result, net, hnum, saddr, sport,
344 daddr, dport, dif) < badness)) {
352 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
354 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
355 __be16 sport, __be16 dport,
356 struct udp_table *udptable)
359 const struct ipv6hdr *iph = ipv6_hdr(skb);
361 if (unlikely(sk = skb_steal_sock(skb)))
363 return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport,
364 &iph->daddr, dport, inet6_iif(skb),
368 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
369 const struct in6_addr *daddr, __be16 dport, int dif)
371 return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
373 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
377 * This should be easy, if there is something there we
378 * return it, otherwise we block.
381 int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk,
382 struct msghdr *msg, size_t len,
383 int noblock, int flags, int *addr_len)
385 struct ipv6_pinfo *np = inet6_sk(sk);
386 struct inet_sock *inet = inet_sk(sk);
388 unsigned int ulen, copied;
391 int is_udplite = IS_UDPLITE(sk);
395 if (flags & MSG_ERRQUEUE)
396 return ipv6_recv_error(sk, msg, len, addr_len);
398 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
399 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
402 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
403 &peeked, &off, &err);
407 ulen = skb->len - sizeof(struct udphdr);
411 else if (copied < ulen)
412 msg->msg_flags |= MSG_TRUNC;
414 is_udp4 = (skb->protocol == htons(ETH_P_IP));
417 * If checksum is needed at all, try to do it while copying the
418 * data. If the data is truncated, or if we only want a partial
419 * coverage checksum (UDP-Lite), do it before the copy.
422 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
423 if (udp_lib_checksum_complete(skb))
427 if (skb_csum_unnecessary(skb))
428 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
429 msg->msg_iov, copied);
431 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
436 trace_kfree_skb(skb, udpv6_recvmsg);
438 atomic_inc(&sk->sk_drops);
440 UDP_INC_STATS_USER(sock_net(sk),
444 UDP6_INC_STATS_USER(sock_net(sk),
452 UDP_INC_STATS_USER(sock_net(sk),
453 UDP_MIB_INDATAGRAMS, is_udplite);
455 UDP6_INC_STATS_USER(sock_net(sk),
456 UDP_MIB_INDATAGRAMS, is_udplite);
459 sock_recv_ts_and_drops(msg, sk, skb);
461 /* Copy the address. */
463 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
464 sin6->sin6_family = AF_INET6;
465 sin6->sin6_port = udp_hdr(skb)->source;
466 sin6->sin6_flowinfo = 0;
469 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
471 sin6->sin6_scope_id = 0;
473 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
474 sin6->sin6_scope_id =
475 ipv6_iface_scope_id(&sin6->sin6_addr,
478 *addr_len = sizeof(*sin6);
482 ip6_datagram_recv_common_ctl(sk, msg, skb);
485 if (inet->cmsg_flags)
486 ip_cmsg_recv(msg, skb);
489 ip6_datagram_recv_specific_ctl(sk, msg, skb);
493 if (flags & MSG_TRUNC)
497 skb_free_datagram_locked(sk, skb);
502 slow = lock_sock_fast(sk);
503 if (!skb_kill_datagram(sk, skb, flags)) {
505 UDP_INC_STATS_USER(sock_net(sk),
506 UDP_MIB_CSUMERRORS, is_udplite);
507 UDP_INC_STATS_USER(sock_net(sk),
508 UDP_MIB_INERRORS, is_udplite);
510 UDP6_INC_STATS_USER(sock_net(sk),
511 UDP_MIB_CSUMERRORS, is_udplite);
512 UDP6_INC_STATS_USER(sock_net(sk),
513 UDP_MIB_INERRORS, is_udplite);
516 unlock_sock_fast(sk, slow);
521 /* starting over for a new packet */
522 msg->msg_flags &= ~MSG_TRUNC;
526 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
527 u8 type, u8 code, int offset, __be32 info,
528 struct udp_table *udptable)
530 struct ipv6_pinfo *np;
531 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
532 const struct in6_addr *saddr = &hdr->saddr;
533 const struct in6_addr *daddr = &hdr->daddr;
534 struct udphdr *uh = (struct udphdr*)(skb->data+offset);
538 sk = __udp6_lib_lookup(dev_net(skb->dev), daddr, uh->dest,
539 saddr, uh->source, inet6_iif(skb), udptable);
543 if (type == ICMPV6_PKT_TOOBIG) {
544 if (!ip6_sk_accept_pmtu(sk))
546 ip6_sk_update_pmtu(skb, sk, info);
548 if (type == NDISC_REDIRECT) {
549 ip6_sk_redirect(skb, sk);
555 if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
558 if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
562 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
565 sk->sk_error_report(sk);
570 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
574 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
575 sock_rps_save_rxhash(sk, skb);
576 sk_mark_napi_id(sk, skb);
579 rc = sock_queue_rcv_skb(sk, skb);
581 int is_udplite = IS_UDPLITE(sk);
583 /* Note that an ENOMEM error is charged twice */
585 UDP6_INC_STATS_BH(sock_net(sk),
586 UDP_MIB_RCVBUFERRORS, is_udplite);
587 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
594 static __inline__ void udpv6_err(struct sk_buff *skb,
595 struct inet6_skb_parm *opt, u8 type,
596 u8 code, int offset, __be32 info )
598 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
601 static struct static_key udpv6_encap_needed __read_mostly;
602 void udpv6_encap_enable(void)
604 if (!static_key_enabled(&udpv6_encap_needed))
605 static_key_slow_inc(&udpv6_encap_needed);
607 EXPORT_SYMBOL(udpv6_encap_enable);
609 int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
611 struct udp_sock *up = udp_sk(sk);
613 int is_udplite = IS_UDPLITE(sk);
615 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
618 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
619 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
622 * This is an encapsulation socket so pass the skb to
623 * the socket's udp_encap_rcv() hook. Otherwise, just
624 * fall through and pass this up the UDP socket.
625 * up->encap_rcv() returns the following value:
626 * =0 if skb was successfully passed to the encap
627 * handler or was discarded by it.
628 * >0 if skb should be passed on to UDP.
629 * <0 if skb should be resubmitted as proto -N
632 /* if we're overly short, let UDP handle it */
633 encap_rcv = ACCESS_ONCE(up->encap_rcv);
634 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
637 ret = encap_rcv(sk, skb);
639 UDP_INC_STATS_BH(sock_net(sk),
646 /* FALLTHROUGH -- it's a UDP Packet */
650 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
652 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
654 if (up->pcrlen == 0) { /* full coverage was set */
655 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage"
656 " %d while full coverage %d requested\n",
657 UDP_SKB_CB(skb)->cscov, skb->len);
660 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
661 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d "
662 "too small, need min %d\n",
663 UDP_SKB_CB(skb)->cscov, up->pcrlen);
668 if (rcu_access_pointer(sk->sk_filter)) {
669 if (udp_lib_checksum_complete(skb))
673 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf))
680 if (!sock_owned_by_user(sk))
681 rc = __udpv6_queue_rcv_skb(sk, skb);
682 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
690 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
692 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
693 atomic_inc(&sk->sk_drops);
698 static struct sock *udp_v6_mcast_next(struct net *net, struct sock *sk,
699 __be16 loc_port, const struct in6_addr *loc_addr,
700 __be16 rmt_port, const struct in6_addr *rmt_addr,
703 struct hlist_nulls_node *node;
705 unsigned short num = ntohs(loc_port);
707 sk_nulls_for_each_from(s, node) {
708 struct inet_sock *inet = inet_sk(s);
710 if (!net_eq(sock_net(s), net))
713 if (udp_sk(s)->udp_port_hash == num &&
714 s->sk_family == PF_INET6) {
715 if (inet->inet_dport) {
716 if (inet->inet_dport != rmt_port)
719 if (!ipv6_addr_any(&s->sk_v6_daddr) &&
720 !ipv6_addr_equal(&s->sk_v6_daddr, rmt_addr))
723 if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)
726 if (!ipv6_addr_any(&s->sk_v6_rcv_saddr)) {
727 if (!ipv6_addr_equal(&s->sk_v6_rcv_saddr, loc_addr))
730 if (!inet6_mc_check(s, loc_addr, rmt_addr))
738 static void flush_stack(struct sock **stack, unsigned int count,
739 struct sk_buff *skb, unsigned int final)
741 struct sk_buff *skb1 = NULL;
745 for (i = 0; i < count; i++) {
747 if (likely(skb1 == NULL))
748 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
750 atomic_inc(&sk->sk_drops);
751 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
753 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
757 if (skb1 && udpv6_queue_rcv_skb(sk, skb1) <= 0)
764 * Note: called only from the BH handler context,
765 * so we don't need to lock the hashes.
767 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
768 const struct in6_addr *saddr, const struct in6_addr *daddr,
769 struct udp_table *udptable)
771 struct sock *sk, *stack[256 / sizeof(struct sock *)];
772 const struct udphdr *uh = udp_hdr(skb);
773 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
775 unsigned int i, count = 0;
777 spin_lock(&hslot->lock);
778 sk = sk_nulls_head(&hslot->head);
779 dif = inet6_iif(skb);
780 sk = udp_v6_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
783 sk = udp_v6_mcast_next(net, sk_nulls_next(sk), uh->dest, daddr,
784 uh->source, saddr, dif);
785 if (unlikely(count == ARRAY_SIZE(stack))) {
788 flush_stack(stack, count, skb, ~0);
793 * before releasing the lock, we must take reference on sockets
795 for (i = 0; i < count; i++)
798 spin_unlock(&hslot->lock);
801 flush_stack(stack, count, skb, count - 1);
803 for (i = 0; i < count; i++)
811 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
814 struct net *net = dev_net(skb->dev);
817 const struct in6_addr *saddr, *daddr;
820 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
823 saddr = &ipv6_hdr(skb)->saddr;
824 daddr = &ipv6_hdr(skb)->daddr;
827 ulen = ntohs(uh->len);
831 if (proto == IPPROTO_UDP) {
832 /* UDP validates ulen. */
834 /* Check for jumbo payload */
838 if (ulen < sizeof(*uh))
841 if (ulen < skb->len) {
842 if (pskb_trim_rcsum(skb, ulen))
844 saddr = &ipv6_hdr(skb)->saddr;
845 daddr = &ipv6_hdr(skb)->daddr;
850 if (udp6_csum_init(skb, uh, proto))
854 * Multicast receive code
856 if (ipv6_addr_is_multicast(daddr))
857 return __udp6_lib_mcast_deliver(net, skb,
858 saddr, daddr, udptable);
863 * check socket cache ... must talk to Alan about his plans
864 * for sock caches... i'll skip this for now.
866 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
870 ret = udpv6_queue_rcv_skb(sk, skb);
873 /* a return value > 0 means to resubmit the input, but
874 * it wants the return to be -protocol, or 0
882 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
885 if (udp_lib_checksum_complete(skb))
888 UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
889 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
895 LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
896 proto == IPPROTO_UDPLITE ? "-Lite" : "",
905 UDP6_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
907 UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
912 static __inline__ int udpv6_rcv(struct sk_buff *skb)
914 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
918 * Throw away all pending data and cancel the corking. Socket is locked.
920 static void udp_v6_flush_pending_frames(struct sock *sk)
922 struct udp_sock *up = udp_sk(sk);
924 if (up->pending == AF_INET)
925 udp_flush_pending_frames(sk);
926 else if (up->pending) {
929 ip6_flush_pending_frames(sk);
934 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
935 * @sk: socket we are sending on
936 * @skb: sk_buff containing the filled-in UDP header
937 * (checksum field must be zeroed out)
939 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
940 const struct in6_addr *saddr,
941 const struct in6_addr *daddr, int len)
944 struct udphdr *uh = udp_hdr(skb);
947 if (skb_queue_len(&sk->sk_write_queue) == 1) {
948 /* Only one fragment on the socket. */
949 skb->csum_start = skb_transport_header(skb) - skb->head;
950 skb->csum_offset = offsetof(struct udphdr, check);
951 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
954 * HW-checksum won't work as there are two or more
955 * fragments on the socket so that all csums of sk_buffs
958 offset = skb_transport_offset(skb);
959 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
961 skb->ip_summed = CHECKSUM_NONE;
963 skb_queue_walk(&sk->sk_write_queue, skb) {
964 csum = csum_add(csum, skb->csum);
967 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
970 uh->check = CSUM_MANGLED_0;
978 static int udp_v6_push_pending_frames(struct sock *sk)
982 struct udp_sock *up = udp_sk(sk);
983 struct inet_sock *inet = inet_sk(sk);
986 int is_udplite = IS_UDPLITE(sk);
989 if (up->pending == AF_INET)
990 return udp_push_pending_frames(sk);
992 fl6 = &inet->cork.fl.u.ip6;
994 /* Grab the skbuff where UDP header space exists. */
995 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
999 * Create a UDP header
1002 uh->source = fl6->fl6_sport;
1003 uh->dest = fl6->fl6_dport;
1004 uh->len = htons(up->len);
1008 csum = udplite_csum_outgoing(sk, skb);
1009 else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1010 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr,
1014 csum = udp_csum_outgoing(sk, skb);
1016 /* add protocol-dependent pseudo-header */
1017 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1018 up->len, fl6->flowi6_proto, csum);
1020 uh->check = CSUM_MANGLED_0;
1023 err = ip6_push_pending_frames(sk);
1025 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1026 UDP6_INC_STATS_USER(sock_net(sk),
1027 UDP_MIB_SNDBUFERRORS, is_udplite);
1031 UDP6_INC_STATS_USER(sock_net(sk),
1032 UDP_MIB_OUTDATAGRAMS, is_udplite);
1039 int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
1040 struct msghdr *msg, size_t len)
1042 struct ipv6_txoptions opt_space;
1043 struct udp_sock *up = udp_sk(sk);
1044 struct inet_sock *inet = inet_sk(sk);
1045 struct ipv6_pinfo *np = inet6_sk(sk);
1046 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1047 struct in6_addr *daddr, *final_p, final;
1048 struct ipv6_txoptions *opt = NULL;
1049 struct ip6_flowlabel *flowlabel = NULL;
1051 struct dst_entry *dst;
1052 int addr_len = msg->msg_namelen;
1057 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1060 int is_udplite = IS_UDPLITE(sk);
1061 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1063 /* destination address check */
1065 if (addr_len < offsetof(struct sockaddr, sa_data))
1068 switch (sin6->sin6_family) {
1070 if (addr_len < SIN6_LEN_RFC2133)
1072 daddr = &sin6->sin6_addr;
1075 goto do_udp_sendmsg;
1077 msg->msg_name = sin6 = NULL;
1078 msg->msg_namelen = addr_len = 0;
1084 } else if (!up->pending) {
1085 if (sk->sk_state != TCP_ESTABLISHED)
1086 return -EDESTADDRREQ;
1087 daddr = &sk->sk_v6_daddr;
1092 if (ipv6_addr_v4mapped(daddr)) {
1093 struct sockaddr_in sin;
1094 sin.sin_family = AF_INET;
1095 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1096 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1097 msg->msg_name = &sin;
1098 msg->msg_namelen = sizeof(sin);
1100 if (__ipv6_only_sock(sk))
1101 return -ENETUNREACH;
1102 return udp_sendmsg(iocb, sk, msg, len);
1106 if (up->pending == AF_INET)
1107 return udp_sendmsg(iocb, sk, msg, len);
1109 /* Rough check on arithmetic overflow,
1110 better check is made in ip6_append_data().
1112 if (len > INT_MAX - sizeof(struct udphdr))
1117 * There are pending frames.
1118 * The socket lock must be held while it's corked.
1121 if (likely(up->pending)) {
1122 if (unlikely(up->pending != AF_INET6)) {
1124 return -EAFNOSUPPORT;
1127 goto do_append_data;
1131 ulen += sizeof(struct udphdr);
1133 memset(&fl6, 0, sizeof(fl6));
1136 if (sin6->sin6_port == 0)
1139 fl6.fl6_dport = sin6->sin6_port;
1140 daddr = &sin6->sin6_addr;
1143 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1144 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1145 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1146 if (flowlabel == NULL)
1152 * Otherwise it will be difficult to maintain
1155 if (sk->sk_state == TCP_ESTABLISHED &&
1156 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1157 daddr = &sk->sk_v6_daddr;
1159 if (addr_len >= sizeof(struct sockaddr_in6) &&
1160 sin6->sin6_scope_id &&
1161 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1162 fl6.flowi6_oif = sin6->sin6_scope_id;
1164 if (sk->sk_state != TCP_ESTABLISHED)
1165 return -EDESTADDRREQ;
1167 fl6.fl6_dport = inet->inet_dport;
1168 daddr = &sk->sk_v6_daddr;
1169 fl6.flowlabel = np->flow_label;
1173 if (!fl6.flowi6_oif)
1174 fl6.flowi6_oif = sk->sk_bound_dev_if;
1176 if (!fl6.flowi6_oif)
1177 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1179 fl6.flowi6_mark = sk->sk_mark;
1181 if (msg->msg_controllen) {
1183 memset(opt, 0, sizeof(struct ipv6_txoptions));
1184 opt->tot_len = sizeof(*opt);
1186 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt,
1187 &hlimit, &tclass, &dontfrag);
1189 fl6_sock_release(flowlabel);
1192 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1193 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1194 if (flowlabel == NULL)
1197 if (!(opt->opt_nflen|opt->opt_flen))
1204 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1205 opt = ipv6_fixup_options(&opt_space, opt);
1207 fl6.flowi6_proto = sk->sk_protocol;
1208 if (!ipv6_addr_any(daddr))
1211 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1212 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1213 fl6.saddr = np->saddr;
1214 fl6.fl6_sport = inet->inet_sport;
1216 final_p = fl6_update_dst(&fl6, opt, &final);
1220 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1221 fl6.flowi6_oif = np->mcast_oif;
1223 } else if (!fl6.flowi6_oif)
1224 fl6.flowi6_oif = np->ucast_oif;
1226 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1228 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p);
1236 if (ipv6_addr_is_multicast(&fl6.daddr))
1237 hlimit = np->mcast_hops;
1239 hlimit = np->hop_limit;
1241 hlimit = ip6_dst_hoplimit(dst);
1245 tclass = np->tclass;
1247 if (msg->msg_flags&MSG_CONFIRM)
1252 if (unlikely(up->pending)) {
1253 /* The socket is already corked while preparing it. */
1254 /* ... which is an evident application bug. --ANK */
1257 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
1262 up->pending = AF_INET6;
1266 dontfrag = np->dontfrag;
1268 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1269 err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen,
1270 sizeof(struct udphdr), hlimit, tclass, opt, &fl6,
1271 (struct rt6_info*)dst,
1272 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag);
1274 udp_v6_flush_pending_frames(sk);
1276 err = udp_v6_push_pending_frames(sk);
1277 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1282 ip6_dst_store(sk, dst,
1283 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ?
1284 &sk->sk_v6_daddr : NULL,
1285 #ifdef CONFIG_IPV6_SUBTREES
1286 ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1297 err = np->recverr ? net_xmit_errno(err) : 0;
1301 fl6_sock_release(flowlabel);
1305 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1306 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1307 * we don't have a good statistic (IpOutDiscards but it can be too many
1308 * things). We could add another new stat but at least for now that
1309 * seems like overkill.
1311 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1312 UDP6_INC_STATS_USER(sock_net(sk),
1313 UDP_MIB_SNDBUFERRORS, is_udplite);
1319 if (!(msg->msg_flags&MSG_PROBE) || len)
1320 goto back_from_confirm;
1325 void udpv6_destroy_sock(struct sock *sk)
1327 struct udp_sock *up = udp_sk(sk);
1329 udp_v6_flush_pending_frames(sk);
1332 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
1333 void (*encap_destroy)(struct sock *sk);
1334 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1339 inet6_destroy_sock(sk);
1343 * Socket option code for UDP
1345 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1346 char __user *optval, unsigned int optlen)
1348 if (level == SOL_UDP || level == SOL_UDPLITE)
1349 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1350 udp_v6_push_pending_frames);
1351 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1354 #ifdef CONFIG_COMPAT
1355 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1356 char __user *optval, unsigned int optlen)
1358 if (level == SOL_UDP || level == SOL_UDPLITE)
1359 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1360 udp_v6_push_pending_frames);
1361 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1365 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1366 char __user *optval, int __user *optlen)
1368 if (level == SOL_UDP || level == SOL_UDPLITE)
1369 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1370 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1373 #ifdef CONFIG_COMPAT
1374 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1375 char __user *optval, int __user *optlen)
1377 if (level == SOL_UDP || level == SOL_UDPLITE)
1378 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1379 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1383 static const struct inet6_protocol udpv6_protocol = {
1384 .handler = udpv6_rcv,
1385 .err_handler = udpv6_err,
1386 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1389 /* ------------------------------------------------------------------------ */
1390 #ifdef CONFIG_PROC_FS
1391 int udp6_seq_show(struct seq_file *seq, void *v)
1393 if (v == SEQ_START_TOKEN) {
1394 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1396 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1397 struct inet_sock *inet = inet_sk(v);
1398 __u16 srcp = ntohs(inet->inet_sport);
1399 __u16 destp = ntohs(inet->inet_dport);
1400 ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket);
1405 static const struct file_operations udp6_afinfo_seq_fops = {
1406 .owner = THIS_MODULE,
1407 .open = udp_seq_open,
1409 .llseek = seq_lseek,
1410 .release = seq_release_net
1413 static struct udp_seq_afinfo udp6_seq_afinfo = {
1416 .udp_table = &udp_table,
1417 .seq_fops = &udp6_afinfo_seq_fops,
1419 .show = udp6_seq_show,
1423 int __net_init udp6_proc_init(struct net *net)
1425 return udp_proc_register(net, &udp6_seq_afinfo);
1428 void udp6_proc_exit(struct net *net) {
1429 udp_proc_unregister(net, &udp6_seq_afinfo);
1431 #endif /* CONFIG_PROC_FS */
1433 void udp_v6_clear_sk(struct sock *sk, int size)
1435 struct inet_sock *inet = inet_sk(sk);
1437 /* we do not want to clear pinet6 field, because of RCU lookups */
1438 sk_prot_clear_portaddr_nulls(sk, offsetof(struct inet_sock, pinet6));
1440 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1441 memset(&inet->pinet6 + 1, 0, size);
1444 /* ------------------------------------------------------------------------ */
1446 struct proto udpv6_prot = {
1448 .owner = THIS_MODULE,
1449 .close = udp_lib_close,
1450 .connect = ip6_datagram_connect,
1451 .disconnect = udp_disconnect,
1453 .destroy = udpv6_destroy_sock,
1454 .setsockopt = udpv6_setsockopt,
1455 .getsockopt = udpv6_getsockopt,
1456 .sendmsg = udpv6_sendmsg,
1457 .recvmsg = udpv6_recvmsg,
1458 .backlog_rcv = __udpv6_queue_rcv_skb,
1459 .hash = udp_lib_hash,
1460 .unhash = udp_lib_unhash,
1461 .rehash = udp_v6_rehash,
1462 .get_port = udp_v6_get_port,
1463 .memory_allocated = &udp_memory_allocated,
1464 .sysctl_mem = sysctl_udp_mem,
1465 .sysctl_wmem = &sysctl_udp_wmem_min,
1466 .sysctl_rmem = &sysctl_udp_rmem_min,
1467 .obj_size = sizeof(struct udp6_sock),
1468 .slab_flags = SLAB_DESTROY_BY_RCU,
1469 .h.udp_table = &udp_table,
1470 #ifdef CONFIG_COMPAT
1471 .compat_setsockopt = compat_udpv6_setsockopt,
1472 .compat_getsockopt = compat_udpv6_getsockopt,
1474 .clear_sk = udp_v6_clear_sk,
1477 static struct inet_protosw udpv6_protosw = {
1479 .protocol = IPPROTO_UDP,
1480 .prot = &udpv6_prot,
1481 .ops = &inet6_dgram_ops,
1482 .no_check = UDP_CSUM_DEFAULT,
1483 .flags = INET_PROTOSW_PERMANENT,
1487 int __init udpv6_init(void)
1491 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1495 ret = inet6_register_protosw(&udpv6_protosw);
1497 goto out_udpv6_protocol;
1502 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1506 void udpv6_exit(void)
1508 inet6_unregister_protosw(&udpv6_protosw);
1509 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);