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 /* Verify checksum before giving to encap */
638 if (udp_lib_checksum_complete(skb))
641 ret = encap_rcv(sk, skb);
643 UDP_INC_STATS_BH(sock_net(sk),
650 /* FALLTHROUGH -- it's a UDP Packet */
654 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
656 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
658 if (up->pcrlen == 0) { /* full coverage was set */
659 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage"
660 " %d while full coverage %d requested\n",
661 UDP_SKB_CB(skb)->cscov, skb->len);
664 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
665 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d "
666 "too small, need min %d\n",
667 UDP_SKB_CB(skb)->cscov, up->pcrlen);
672 if (rcu_access_pointer(sk->sk_filter)) {
673 if (udp_lib_checksum_complete(skb))
677 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf))
684 if (!sock_owned_by_user(sk))
685 rc = __udpv6_queue_rcv_skb(sk, skb);
686 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
694 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
696 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
697 atomic_inc(&sk->sk_drops);
702 static struct sock *udp_v6_mcast_next(struct net *net, struct sock *sk,
703 __be16 loc_port, const struct in6_addr *loc_addr,
704 __be16 rmt_port, const struct in6_addr *rmt_addr,
707 struct hlist_nulls_node *node;
709 unsigned short num = ntohs(loc_port);
711 sk_nulls_for_each_from(s, node) {
712 struct inet_sock *inet = inet_sk(s);
714 if (!net_eq(sock_net(s), net))
717 if (udp_sk(s)->udp_port_hash == num &&
718 s->sk_family == PF_INET6) {
719 if (inet->inet_dport) {
720 if (inet->inet_dport != rmt_port)
723 if (!ipv6_addr_any(&sk->sk_v6_daddr) &&
724 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr))
727 if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)
730 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
731 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))
734 if (!inet6_mc_check(s, loc_addr, rmt_addr))
742 static void flush_stack(struct sock **stack, unsigned int count,
743 struct sk_buff *skb, unsigned int final)
745 struct sk_buff *skb1 = NULL;
749 for (i = 0; i < count; i++) {
751 if (likely(skb1 == NULL))
752 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
754 atomic_inc(&sk->sk_drops);
755 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
757 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
761 if (skb1 && udpv6_queue_rcv_skb(sk, skb1) <= 0)
768 static void udp6_csum_zero_error(struct sk_buff *skb)
770 /* RFC 2460 section 8.1 says that we SHOULD log
771 * this error. Well, it is reasonable.
773 LIMIT_NETDEBUG(KERN_INFO "IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
774 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
775 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
779 * Note: called only from the BH handler context,
780 * so we don't need to lock the hashes.
782 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
783 const struct in6_addr *saddr, const struct in6_addr *daddr,
784 struct udp_table *udptable)
786 struct sock *sk, *stack[256 / sizeof(struct sock *)];
787 const struct udphdr *uh = udp_hdr(skb);
788 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
790 unsigned int i, count = 0;
792 spin_lock(&hslot->lock);
793 sk = sk_nulls_head(&hslot->head);
794 dif = inet6_iif(skb);
795 sk = udp_v6_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
797 /* If zero checksum and sk_no_check is not on for
798 * the socket then skip it.
800 if (uh->check || sk->sk_no_check)
803 sk = udp_v6_mcast_next(net, sk_nulls_next(sk), uh->dest, daddr,
804 uh->source, saddr, dif);
805 if (unlikely(count == ARRAY_SIZE(stack))) {
808 flush_stack(stack, count, skb, ~0);
813 * before releasing the lock, we must take reference on sockets
815 for (i = 0; i < count; i++)
818 spin_unlock(&hslot->lock);
821 flush_stack(stack, count, skb, count - 1);
823 for (i = 0; i < count; i++)
831 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
834 struct net *net = dev_net(skb->dev);
837 const struct in6_addr *saddr, *daddr;
840 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
843 saddr = &ipv6_hdr(skb)->saddr;
844 daddr = &ipv6_hdr(skb)->daddr;
847 ulen = ntohs(uh->len);
851 if (proto == IPPROTO_UDP) {
852 /* UDP validates ulen. */
854 /* Check for jumbo payload */
858 if (ulen < sizeof(*uh))
861 if (ulen < skb->len) {
862 if (pskb_trim_rcsum(skb, ulen))
864 saddr = &ipv6_hdr(skb)->saddr;
865 daddr = &ipv6_hdr(skb)->daddr;
870 if (udp6_csum_init(skb, uh, proto))
874 * Multicast receive code
876 if (ipv6_addr_is_multicast(daddr))
877 return __udp6_lib_mcast_deliver(net, skb,
878 saddr, daddr, udptable);
883 * check socket cache ... must talk to Alan about his plans
884 * for sock caches... i'll skip this for now.
886 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
890 if (!uh->check && !sk->sk_no_check) {
892 udp6_csum_zero_error(skb);
896 ret = udpv6_queue_rcv_skb(sk, skb);
899 /* a return value > 0 means to resubmit the input, but
900 * it wants the return to be -protocol, or 0
909 udp6_csum_zero_error(skb);
913 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
916 if (udp_lib_checksum_complete(skb))
919 UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
920 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
926 LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
927 proto == IPPROTO_UDPLITE ? "-Lite" : "",
936 UDP6_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
938 UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
943 static __inline__ int udpv6_rcv(struct sk_buff *skb)
945 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
949 * Throw away all pending data and cancel the corking. Socket is locked.
951 static void udp_v6_flush_pending_frames(struct sock *sk)
953 struct udp_sock *up = udp_sk(sk);
955 if (up->pending == AF_INET)
956 udp_flush_pending_frames(sk);
957 else if (up->pending) {
960 ip6_flush_pending_frames(sk);
965 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
966 * @sk: socket we are sending on
967 * @skb: sk_buff containing the filled-in UDP header
968 * (checksum field must be zeroed out)
970 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
971 const struct in6_addr *saddr,
972 const struct in6_addr *daddr, int len)
975 struct udphdr *uh = udp_hdr(skb);
978 if (skb_queue_len(&sk->sk_write_queue) == 1) {
979 /* Only one fragment on the socket. */
980 skb->csum_start = skb_transport_header(skb) - skb->head;
981 skb->csum_offset = offsetof(struct udphdr, check);
982 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
985 * HW-checksum won't work as there are two or more
986 * fragments on the socket so that all csums of sk_buffs
989 offset = skb_transport_offset(skb);
990 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
992 skb->ip_summed = CHECKSUM_NONE;
994 skb_queue_walk(&sk->sk_write_queue, skb) {
995 csum = csum_add(csum, skb->csum);
998 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1001 uh->check = CSUM_MANGLED_0;
1009 static int udp_v6_push_pending_frames(struct sock *sk)
1011 struct sk_buff *skb;
1013 struct udp_sock *up = udp_sk(sk);
1014 struct inet_sock *inet = inet_sk(sk);
1017 int is_udplite = IS_UDPLITE(sk);
1020 if (up->pending == AF_INET)
1021 return udp_push_pending_frames(sk);
1023 fl6 = &inet->cork.fl.u.ip6;
1025 /* Grab the skbuff where UDP header space exists. */
1026 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
1030 * Create a UDP header
1033 uh->source = fl6->fl6_sport;
1034 uh->dest = fl6->fl6_dport;
1035 uh->len = htons(up->len);
1039 csum = udplite_csum_outgoing(sk, skb);
1040 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
1041 skb->ip_summed = CHECKSUM_NONE;
1043 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1044 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr,
1048 csum = udp_csum_outgoing(sk, skb);
1050 /* add protocol-dependent pseudo-header */
1051 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1052 up->len, fl6->flowi6_proto, csum);
1054 uh->check = CSUM_MANGLED_0;
1057 err = ip6_push_pending_frames(sk);
1059 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1060 UDP6_INC_STATS_USER(sock_net(sk),
1061 UDP_MIB_SNDBUFERRORS, is_udplite);
1065 UDP6_INC_STATS_USER(sock_net(sk),
1066 UDP_MIB_OUTDATAGRAMS, is_udplite);
1073 int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
1074 struct msghdr *msg, size_t len)
1076 struct ipv6_txoptions opt_space;
1077 struct udp_sock *up = udp_sk(sk);
1078 struct inet_sock *inet = inet_sk(sk);
1079 struct ipv6_pinfo *np = inet6_sk(sk);
1080 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1081 struct in6_addr *daddr, *final_p, final;
1082 struct ipv6_txoptions *opt = NULL;
1083 struct ip6_flowlabel *flowlabel = NULL;
1085 struct dst_entry *dst;
1086 int addr_len = msg->msg_namelen;
1091 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1094 int is_udplite = IS_UDPLITE(sk);
1095 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1097 /* destination address check */
1099 if (addr_len < offsetof(struct sockaddr, sa_data))
1102 switch (sin6->sin6_family) {
1104 if (addr_len < SIN6_LEN_RFC2133)
1106 daddr = &sin6->sin6_addr;
1109 goto do_udp_sendmsg;
1111 msg->msg_name = sin6 = NULL;
1112 msg->msg_namelen = addr_len = 0;
1118 } else if (!up->pending) {
1119 if (sk->sk_state != TCP_ESTABLISHED)
1120 return -EDESTADDRREQ;
1121 daddr = &sk->sk_v6_daddr;
1126 if (ipv6_addr_v4mapped(daddr)) {
1127 struct sockaddr_in sin;
1128 sin.sin_family = AF_INET;
1129 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1130 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1131 msg->msg_name = &sin;
1132 msg->msg_namelen = sizeof(sin);
1134 if (__ipv6_only_sock(sk))
1135 return -ENETUNREACH;
1136 return udp_sendmsg(iocb, sk, msg, len);
1140 if (up->pending == AF_INET)
1141 return udp_sendmsg(iocb, sk, msg, len);
1143 /* Rough check on arithmetic overflow,
1144 better check is made in ip6_append_data().
1146 if (len > INT_MAX - sizeof(struct udphdr))
1151 * There are pending frames.
1152 * The socket lock must be held while it's corked.
1155 if (likely(up->pending)) {
1156 if (unlikely(up->pending != AF_INET6)) {
1158 return -EAFNOSUPPORT;
1161 goto do_append_data;
1165 ulen += sizeof(struct udphdr);
1167 memset(&fl6, 0, sizeof(fl6));
1170 if (sin6->sin6_port == 0)
1173 fl6.fl6_dport = sin6->sin6_port;
1174 daddr = &sin6->sin6_addr;
1177 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1178 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1179 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1180 if (flowlabel == NULL)
1186 * Otherwise it will be difficult to maintain
1189 if (sk->sk_state == TCP_ESTABLISHED &&
1190 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1191 daddr = &sk->sk_v6_daddr;
1193 if (addr_len >= sizeof(struct sockaddr_in6) &&
1194 sin6->sin6_scope_id &&
1195 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1196 fl6.flowi6_oif = sin6->sin6_scope_id;
1198 if (sk->sk_state != TCP_ESTABLISHED)
1199 return -EDESTADDRREQ;
1201 fl6.fl6_dport = inet->inet_dport;
1202 daddr = &sk->sk_v6_daddr;
1203 fl6.flowlabel = np->flow_label;
1207 if (!fl6.flowi6_oif)
1208 fl6.flowi6_oif = sk->sk_bound_dev_if;
1210 if (!fl6.flowi6_oif)
1211 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1213 fl6.flowi6_mark = sk->sk_mark;
1215 if (msg->msg_controllen) {
1217 memset(opt, 0, sizeof(struct ipv6_txoptions));
1218 opt->tot_len = sizeof(*opt);
1220 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt,
1221 &hlimit, &tclass, &dontfrag);
1223 fl6_sock_release(flowlabel);
1226 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1227 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1228 if (flowlabel == NULL)
1231 if (!(opt->opt_nflen|opt->opt_flen))
1238 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1239 opt = ipv6_fixup_options(&opt_space, opt);
1241 fl6.flowi6_proto = sk->sk_protocol;
1242 if (!ipv6_addr_any(daddr))
1245 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1246 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1247 fl6.saddr = np->saddr;
1248 fl6.fl6_sport = inet->inet_sport;
1250 final_p = fl6_update_dst(&fl6, opt, &final);
1254 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1255 fl6.flowi6_oif = np->mcast_oif;
1257 } else if (!fl6.flowi6_oif)
1258 fl6.flowi6_oif = np->ucast_oif;
1260 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1262 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p);
1270 hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1273 tclass = np->tclass;
1275 if (msg->msg_flags&MSG_CONFIRM)
1280 if (unlikely(up->pending)) {
1281 /* The socket is already corked while preparing it. */
1282 /* ... which is an evident application bug. --ANK */
1285 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
1290 up->pending = AF_INET6;
1294 dontfrag = np->dontfrag;
1296 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1297 err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen,
1298 sizeof(struct udphdr), hlimit, tclass, opt, &fl6,
1299 (struct rt6_info*)dst,
1300 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag);
1302 udp_v6_flush_pending_frames(sk);
1304 err = udp_v6_push_pending_frames(sk);
1305 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1310 ip6_dst_store(sk, dst,
1311 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ?
1312 &sk->sk_v6_daddr : NULL,
1313 #ifdef CONFIG_IPV6_SUBTREES
1314 ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1325 err = np->recverr ? net_xmit_errno(err) : 0;
1329 fl6_sock_release(flowlabel);
1333 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1334 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1335 * we don't have a good statistic (IpOutDiscards but it can be too many
1336 * things). We could add another new stat but at least for now that
1337 * seems like overkill.
1339 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1340 UDP6_INC_STATS_USER(sock_net(sk),
1341 UDP_MIB_SNDBUFERRORS, is_udplite);
1347 if (!(msg->msg_flags&MSG_PROBE) || len)
1348 goto back_from_confirm;
1353 void udpv6_destroy_sock(struct sock *sk)
1355 struct udp_sock *up = udp_sk(sk);
1357 udp_v6_flush_pending_frames(sk);
1360 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
1361 void (*encap_destroy)(struct sock *sk);
1362 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1367 inet6_destroy_sock(sk);
1371 * Socket option code for UDP
1373 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1374 char __user *optval, unsigned int optlen)
1376 if (level == SOL_UDP || level == SOL_UDPLITE)
1377 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1378 udp_v6_push_pending_frames);
1379 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1382 #ifdef CONFIG_COMPAT
1383 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1384 char __user *optval, unsigned int optlen)
1386 if (level == SOL_UDP || level == SOL_UDPLITE)
1387 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1388 udp_v6_push_pending_frames);
1389 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1393 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1394 char __user *optval, int __user *optlen)
1396 if (level == SOL_UDP || level == SOL_UDPLITE)
1397 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1398 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1401 #ifdef CONFIG_COMPAT
1402 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1403 char __user *optval, int __user *optlen)
1405 if (level == SOL_UDP || level == SOL_UDPLITE)
1406 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1407 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1411 static const struct inet6_protocol udpv6_protocol = {
1412 .handler = udpv6_rcv,
1413 .err_handler = udpv6_err,
1414 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1417 /* ------------------------------------------------------------------------ */
1418 #ifdef CONFIG_PROC_FS
1419 int udp6_seq_show(struct seq_file *seq, void *v)
1421 if (v == SEQ_START_TOKEN) {
1422 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1424 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1425 struct inet_sock *inet = inet_sk(v);
1426 __u16 srcp = ntohs(inet->inet_sport);
1427 __u16 destp = ntohs(inet->inet_dport);
1428 ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket);
1433 static const struct file_operations udp6_afinfo_seq_fops = {
1434 .owner = THIS_MODULE,
1435 .open = udp_seq_open,
1437 .llseek = seq_lseek,
1438 .release = seq_release_net
1441 static struct udp_seq_afinfo udp6_seq_afinfo = {
1444 .udp_table = &udp_table,
1445 .seq_fops = &udp6_afinfo_seq_fops,
1447 .show = udp6_seq_show,
1451 int __net_init udp6_proc_init(struct net *net)
1453 return udp_proc_register(net, &udp6_seq_afinfo);
1456 void udp6_proc_exit(struct net *net) {
1457 udp_proc_unregister(net, &udp6_seq_afinfo);
1459 #endif /* CONFIG_PROC_FS */
1461 void udp_v6_clear_sk(struct sock *sk, int size)
1463 struct inet_sock *inet = inet_sk(sk);
1465 /* we do not want to clear pinet6 field, because of RCU lookups */
1466 sk_prot_clear_portaddr_nulls(sk, offsetof(struct inet_sock, pinet6));
1468 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1469 memset(&inet->pinet6 + 1, 0, size);
1472 /* ------------------------------------------------------------------------ */
1474 struct proto udpv6_prot = {
1476 .owner = THIS_MODULE,
1477 .close = udp_lib_close,
1478 .connect = ip6_datagram_connect,
1479 .disconnect = udp_disconnect,
1481 .destroy = udpv6_destroy_sock,
1482 .setsockopt = udpv6_setsockopt,
1483 .getsockopt = udpv6_getsockopt,
1484 .sendmsg = udpv6_sendmsg,
1485 .recvmsg = udpv6_recvmsg,
1486 .backlog_rcv = __udpv6_queue_rcv_skb,
1487 .hash = udp_lib_hash,
1488 .unhash = udp_lib_unhash,
1489 .rehash = udp_v6_rehash,
1490 .get_port = udp_v6_get_port,
1491 .memory_allocated = &udp_memory_allocated,
1492 .sysctl_mem = sysctl_udp_mem,
1493 .sysctl_wmem = &sysctl_udp_wmem_min,
1494 .sysctl_rmem = &sysctl_udp_rmem_min,
1495 .obj_size = sizeof(struct udp6_sock),
1496 .slab_flags = SLAB_DESTROY_BY_RCU,
1497 .h.udp_table = &udp_table,
1498 #ifdef CONFIG_COMPAT
1499 .compat_setsockopt = compat_udpv6_setsockopt,
1500 .compat_getsockopt = compat_udpv6_getsockopt,
1502 .clear_sk = udp_v6_clear_sk,
1505 static struct inet_protosw udpv6_protosw = {
1507 .protocol = IPPROTO_UDP,
1508 .prot = &udpv6_prot,
1509 .ops = &inet6_dgram_ops,
1510 .no_check = UDP_CSUM_DEFAULT,
1511 .flags = INET_PROTOSW_PERMANENT,
1515 int __init udpv6_init(void)
1519 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1523 ret = inet6_register_protosw(&udpv6_protosw);
1525 goto out_udpv6_protocol;
1530 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1534 void udpv6_exit(void)
1536 inet6_unregister_protosw(&udpv6_protosw);
1537 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);