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 int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2)
58 const struct in6_addr *sk_rcv_saddr6 = &inet6_sk(sk)->rcv_saddr;
59 const struct in6_addr *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2);
60 __be32 sk1_rcv_saddr = sk_rcv_saddr(sk);
61 __be32 sk2_rcv_saddr = sk_rcv_saddr(sk2);
62 int sk_ipv6only = ipv6_only_sock(sk);
63 int sk2_ipv6only = inet_v6_ipv6only(sk2);
64 int addr_type = ipv6_addr_type(sk_rcv_saddr6);
65 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
67 /* if both are mapped, treat as IPv4 */
68 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED)
69 return (!sk2_ipv6only &&
70 (!sk1_rcv_saddr || !sk2_rcv_saddr ||
71 sk1_rcv_saddr == sk2_rcv_saddr));
73 if (addr_type2 == IPV6_ADDR_ANY &&
74 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
77 if (addr_type == IPV6_ADDR_ANY &&
78 !(sk_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
82 ipv6_addr_equal(sk_rcv_saddr6, sk2_rcv_saddr6))
88 static unsigned int udp6_portaddr_hash(struct net *net,
89 const struct in6_addr *addr6,
92 unsigned int hash, mix = net_hash_mix(net);
94 if (ipv6_addr_any(addr6))
95 hash = jhash_1word(0, mix);
96 else if (ipv6_addr_v4mapped(addr6))
97 hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix);
99 hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix);
105 int udp_v6_get_port(struct sock *sk, unsigned short snum)
107 unsigned int hash2_nulladdr =
108 udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
109 unsigned int hash2_partial =
110 udp6_portaddr_hash(sock_net(sk), &inet6_sk(sk)->rcv_saddr, 0);
112 /* precompute partial secondary hash */
113 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
114 return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal, hash2_nulladdr);
117 static void udp_v6_rehash(struct sock *sk)
119 u16 new_hash = udp6_portaddr_hash(sock_net(sk),
120 &inet6_sk(sk)->rcv_saddr,
121 inet_sk(sk)->inet_num);
123 udp_lib_rehash(sk, new_hash);
126 static inline int compute_score(struct sock *sk, struct net *net,
128 const struct in6_addr *saddr, __be16 sport,
129 const struct in6_addr *daddr, __be16 dport,
134 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
135 sk->sk_family == PF_INET6) {
136 struct ipv6_pinfo *np = inet6_sk(sk);
137 struct inet_sock *inet = inet_sk(sk);
140 if (inet->inet_dport) {
141 if (inet->inet_dport != sport)
145 if (!ipv6_addr_any(&np->rcv_saddr)) {
146 if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
150 if (!ipv6_addr_any(&np->daddr)) {
151 if (!ipv6_addr_equal(&np->daddr, saddr))
155 if (sk->sk_bound_dev_if) {
156 if (sk->sk_bound_dev_if != dif)
164 #define SCORE2_MAX (1 + 1 + 1)
165 static inline int compute_score2(struct sock *sk, struct net *net,
166 const struct in6_addr *saddr, __be16 sport,
167 const struct in6_addr *daddr, unsigned short hnum,
172 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
173 sk->sk_family == PF_INET6) {
174 struct ipv6_pinfo *np = inet6_sk(sk);
175 struct inet_sock *inet = inet_sk(sk);
177 if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
180 if (inet->inet_dport) {
181 if (inet->inet_dport != sport)
185 if (!ipv6_addr_any(&np->daddr)) {
186 if (!ipv6_addr_equal(&np->daddr, saddr))
190 if (sk->sk_bound_dev_if) {
191 if (sk->sk_bound_dev_if != dif)
200 /* called with read_rcu_lock() */
201 static struct sock *udp6_lib_lookup2(struct net *net,
202 const struct in6_addr *saddr, __be16 sport,
203 const struct in6_addr *daddr, unsigned int hnum, int dif,
204 struct udp_hslot *hslot2, unsigned int slot2)
206 struct sock *sk, *result;
207 struct hlist_nulls_node *node;
208 int score, badness, matches = 0, reuseport = 0;
214 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
215 score = compute_score2(sk, net, saddr, sport,
217 if (score > badness) {
220 reuseport = sk->sk_reuseport;
222 hash = inet6_ehashfn(net, daddr, hnum,
225 } else if (score == SCORE2_MAX)
227 } else if (score == badness && reuseport) {
229 if (((u64)hash * matches) >> 32 == 0)
231 hash = next_pseudo_random32(hash);
235 * if the nulls value we got at the end of this lookup is
236 * not the expected one, we must restart lookup.
237 * We probably met an item that was moved to another chain.
239 if (get_nulls_value(node) != slot2)
244 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
246 else if (unlikely(compute_score2(result, net, saddr, sport,
247 daddr, hnum, dif) < badness)) {
255 struct sock *__udp6_lib_lookup(struct net *net,
256 const struct in6_addr *saddr, __be16 sport,
257 const struct in6_addr *daddr, __be16 dport,
258 int dif, struct udp_table *udptable)
260 struct sock *sk, *result;
261 struct hlist_nulls_node *node;
262 unsigned short hnum = ntohs(dport);
263 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
264 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
265 int score, badness, matches = 0, reuseport = 0;
269 if (hslot->count > 10) {
270 hash2 = udp6_portaddr_hash(net, daddr, hnum);
271 slot2 = hash2 & udptable->mask;
272 hslot2 = &udptable->hash2[slot2];
273 if (hslot->count < hslot2->count)
276 result = udp6_lib_lookup2(net, saddr, sport,
280 hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
281 slot2 = hash2 & udptable->mask;
282 hslot2 = &udptable->hash2[slot2];
283 if (hslot->count < hslot2->count)
286 result = udp6_lib_lookup2(net, saddr, sport,
287 &in6addr_any, hnum, dif,
296 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
297 score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif);
298 if (score > badness) {
301 reuseport = sk->sk_reuseport;
303 hash = inet6_ehashfn(net, daddr, hnum,
307 } else if (score == badness && reuseport) {
309 if (((u64)hash * matches) >> 32 == 0)
311 hash = next_pseudo_random32(hash);
315 * if the nulls value we got at the end of this lookup is
316 * not the expected one, we must restart lookup.
317 * We probably met an item that was moved to another chain.
319 if (get_nulls_value(node) != slot)
323 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
325 else if (unlikely(compute_score(result, net, hnum, saddr, sport,
326 daddr, dport, dif) < badness)) {
334 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
336 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
337 __be16 sport, __be16 dport,
338 struct udp_table *udptable)
341 const struct ipv6hdr *iph = ipv6_hdr(skb);
343 if (unlikely(sk = skb_steal_sock(skb)))
345 return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport,
346 &iph->daddr, dport, inet6_iif(skb),
350 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
351 const struct in6_addr *daddr, __be16 dport, int dif)
353 return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
355 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
359 * This should be easy, if there is something there we
360 * return it, otherwise we block.
363 int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk,
364 struct msghdr *msg, size_t len,
365 int noblock, int flags, int *addr_len)
367 struct ipv6_pinfo *np = inet6_sk(sk);
368 struct inet_sock *inet = inet_sk(sk);
370 unsigned int ulen, copied;
373 int is_udplite = IS_UDPLITE(sk);
378 *addr_len = sizeof(struct sockaddr_in6);
380 if (flags & MSG_ERRQUEUE)
381 return ipv6_recv_error(sk, msg, len);
383 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
384 return ipv6_recv_rxpmtu(sk, msg, len);
387 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
388 &peeked, &off, &err);
392 ulen = skb->len - sizeof(struct udphdr);
396 else if (copied < ulen)
397 msg->msg_flags |= MSG_TRUNC;
399 is_udp4 = (skb->protocol == htons(ETH_P_IP));
402 * If checksum is needed at all, try to do it while copying the
403 * data. If the data is truncated, or if we only want a partial
404 * coverage checksum (UDP-Lite), do it before the copy.
407 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
408 if (udp_lib_checksum_complete(skb))
412 if (skb_csum_unnecessary(skb))
413 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
414 msg->msg_iov, copied);
416 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
421 trace_kfree_skb(skb, udpv6_recvmsg);
423 atomic_inc(&sk->sk_drops);
425 UDP_INC_STATS_USER(sock_net(sk),
429 UDP6_INC_STATS_USER(sock_net(sk),
437 UDP_INC_STATS_USER(sock_net(sk),
438 UDP_MIB_INDATAGRAMS, is_udplite);
440 UDP6_INC_STATS_USER(sock_net(sk),
441 UDP_MIB_INDATAGRAMS, is_udplite);
444 sock_recv_ts_and_drops(msg, sk, skb);
446 /* Copy the address. */
448 struct sockaddr_in6 *sin6;
450 sin6 = (struct sockaddr_in6 *) msg->msg_name;
451 sin6->sin6_family = AF_INET6;
452 sin6->sin6_port = udp_hdr(skb)->source;
453 sin6->sin6_flowinfo = 0;
456 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
458 sin6->sin6_scope_id = 0;
460 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
461 sin6->sin6_scope_id =
462 ipv6_iface_scope_id(&sin6->sin6_addr,
468 if (inet->cmsg_flags)
469 ip_cmsg_recv(msg, skb);
472 ip6_datagram_recv_ctl(sk, msg, skb);
476 if (flags & MSG_TRUNC)
480 skb_free_datagram_locked(sk, skb);
485 slow = lock_sock_fast(sk);
486 if (!skb_kill_datagram(sk, skb, flags)) {
488 UDP_INC_STATS_USER(sock_net(sk),
489 UDP_MIB_CSUMERRORS, is_udplite);
490 UDP_INC_STATS_USER(sock_net(sk),
491 UDP_MIB_INERRORS, is_udplite);
493 UDP6_INC_STATS_USER(sock_net(sk),
494 UDP_MIB_CSUMERRORS, is_udplite);
495 UDP6_INC_STATS_USER(sock_net(sk),
496 UDP_MIB_INERRORS, is_udplite);
499 unlock_sock_fast(sk, slow);
504 /* starting over for a new packet */
505 msg->msg_flags &= ~MSG_TRUNC;
509 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
510 u8 type, u8 code, int offset, __be32 info,
511 struct udp_table *udptable)
513 struct ipv6_pinfo *np;
514 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
515 const struct in6_addr *saddr = &hdr->saddr;
516 const struct in6_addr *daddr = &hdr->daddr;
517 struct udphdr *uh = (struct udphdr*)(skb->data+offset);
521 sk = __udp6_lib_lookup(dev_net(skb->dev), daddr, uh->dest,
522 saddr, uh->source, inet6_iif(skb), udptable);
526 if (type == ICMPV6_PKT_TOOBIG)
527 ip6_sk_update_pmtu(skb, sk, info);
528 if (type == NDISC_REDIRECT) {
529 ip6_sk_redirect(skb, sk);
535 if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
538 if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
542 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
545 sk->sk_error_report(sk);
550 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
554 if (!ipv6_addr_any(&inet6_sk(sk)->daddr))
555 sock_rps_save_rxhash(sk, skb);
557 rc = sock_queue_rcv_skb(sk, skb);
559 int is_udplite = IS_UDPLITE(sk);
561 /* Note that an ENOMEM error is charged twice */
563 UDP6_INC_STATS_BH(sock_net(sk),
564 UDP_MIB_RCVBUFERRORS, is_udplite);
565 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
572 static __inline__ void udpv6_err(struct sk_buff *skb,
573 struct inet6_skb_parm *opt, u8 type,
574 u8 code, int offset, __be32 info )
576 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
579 static struct static_key udpv6_encap_needed __read_mostly;
580 void udpv6_encap_enable(void)
582 if (!static_key_enabled(&udpv6_encap_needed))
583 static_key_slow_inc(&udpv6_encap_needed);
585 EXPORT_SYMBOL(udpv6_encap_enable);
587 int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
589 struct udp_sock *up = udp_sk(sk);
591 int is_udplite = IS_UDPLITE(sk);
593 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
596 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
597 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
600 * This is an encapsulation socket so pass the skb to
601 * the socket's udp_encap_rcv() hook. Otherwise, just
602 * fall through and pass this up the UDP socket.
603 * up->encap_rcv() returns the following value:
604 * =0 if skb was successfully passed to the encap
605 * handler or was discarded by it.
606 * >0 if skb should be passed on to UDP.
607 * <0 if skb should be resubmitted as proto -N
610 /* if we're overly short, let UDP handle it */
611 encap_rcv = ACCESS_ONCE(up->encap_rcv);
612 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
615 ret = encap_rcv(sk, skb);
617 UDP_INC_STATS_BH(sock_net(sk),
624 /* FALLTHROUGH -- it's a UDP Packet */
628 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
630 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
632 if (up->pcrlen == 0) { /* full coverage was set */
633 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage"
634 " %d while full coverage %d requested\n",
635 UDP_SKB_CB(skb)->cscov, skb->len);
638 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
639 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d "
640 "too small, need min %d\n",
641 UDP_SKB_CB(skb)->cscov, up->pcrlen);
646 if (rcu_access_pointer(sk->sk_filter)) {
647 if (udp_lib_checksum_complete(skb))
651 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf))
658 if (!sock_owned_by_user(sk))
659 rc = __udpv6_queue_rcv_skb(sk, skb);
660 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
668 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
670 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
671 atomic_inc(&sk->sk_drops);
676 static struct sock *udp_v6_mcast_next(struct net *net, struct sock *sk,
677 __be16 loc_port, const struct in6_addr *loc_addr,
678 __be16 rmt_port, const struct in6_addr *rmt_addr,
681 struct hlist_nulls_node *node;
683 unsigned short num = ntohs(loc_port);
685 sk_nulls_for_each_from(s, node) {
686 struct inet_sock *inet = inet_sk(s);
688 if (!net_eq(sock_net(s), net))
691 if (udp_sk(s)->udp_port_hash == num &&
692 s->sk_family == PF_INET6) {
693 struct ipv6_pinfo *np = inet6_sk(s);
694 if (inet->inet_dport) {
695 if (inet->inet_dport != rmt_port)
698 if (!ipv6_addr_any(&np->daddr) &&
699 !ipv6_addr_equal(&np->daddr, rmt_addr))
702 if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)
705 if (!ipv6_addr_any(&np->rcv_saddr)) {
706 if (!ipv6_addr_equal(&np->rcv_saddr, loc_addr))
709 if (!inet6_mc_check(s, loc_addr, rmt_addr))
717 static void flush_stack(struct sock **stack, unsigned int count,
718 struct sk_buff *skb, unsigned int final)
720 struct sk_buff *skb1 = NULL;
724 for (i = 0; i < count; i++) {
726 if (likely(skb1 == NULL))
727 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
729 atomic_inc(&sk->sk_drops);
730 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
732 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
736 if (skb1 && udpv6_queue_rcv_skb(sk, skb1) <= 0)
743 * Note: called only from the BH handler context,
744 * so we don't need to lock the hashes.
746 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
747 const struct in6_addr *saddr, const struct in6_addr *daddr,
748 struct udp_table *udptable)
750 struct sock *sk, *stack[256 / sizeof(struct sock *)];
751 const struct udphdr *uh = udp_hdr(skb);
752 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
754 unsigned int i, count = 0;
756 spin_lock(&hslot->lock);
757 sk = sk_nulls_head(&hslot->head);
758 dif = inet6_iif(skb);
759 sk = udp_v6_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
762 sk = udp_v6_mcast_next(net, sk_nulls_next(sk), uh->dest, daddr,
763 uh->source, saddr, dif);
764 if (unlikely(count == ARRAY_SIZE(stack))) {
767 flush_stack(stack, count, skb, ~0);
772 * before releasing the lock, we must take reference on sockets
774 for (i = 0; i < count; i++)
777 spin_unlock(&hslot->lock);
780 flush_stack(stack, count, skb, count - 1);
782 for (i = 0; i < count; i++)
790 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
793 struct net *net = dev_net(skb->dev);
796 const struct in6_addr *saddr, *daddr;
799 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
802 saddr = &ipv6_hdr(skb)->saddr;
803 daddr = &ipv6_hdr(skb)->daddr;
806 ulen = ntohs(uh->len);
810 if (proto == IPPROTO_UDP) {
811 /* UDP validates ulen. */
813 /* Check for jumbo payload */
817 if (ulen < sizeof(*uh))
820 if (ulen < skb->len) {
821 if (pskb_trim_rcsum(skb, ulen))
823 saddr = &ipv6_hdr(skb)->saddr;
824 daddr = &ipv6_hdr(skb)->daddr;
829 if (udp6_csum_init(skb, uh, proto))
833 * Multicast receive code
835 if (ipv6_addr_is_multicast(daddr))
836 return __udp6_lib_mcast_deliver(net, skb,
837 saddr, daddr, udptable);
842 * check socket cache ... must talk to Alan about his plans
843 * for sock caches... i'll skip this for now.
845 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
849 sk_mark_napi_id(sk, skb);
850 ret = udpv6_queue_rcv_skb(sk, skb);
853 /* a return value > 0 means to resubmit the input, but
854 * it wants the return to be -protocol, or 0
862 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
865 if (udp_lib_checksum_complete(skb))
868 UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
869 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
875 LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
876 proto == IPPROTO_UDPLITE ? "-Lite" : "",
885 UDP6_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
887 UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
892 static __inline__ int udpv6_rcv(struct sk_buff *skb)
894 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
898 * Throw away all pending data and cancel the corking. Socket is locked.
900 static void udp_v6_flush_pending_frames(struct sock *sk)
902 struct udp_sock *up = udp_sk(sk);
904 if (up->pending == AF_INET)
905 udp_flush_pending_frames(sk);
906 else if (up->pending) {
909 ip6_flush_pending_frames(sk);
914 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
915 * @sk: socket we are sending on
916 * @skb: sk_buff containing the filled-in UDP header
917 * (checksum field must be zeroed out)
919 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
920 const struct in6_addr *saddr,
921 const struct in6_addr *daddr, int len)
924 struct udphdr *uh = udp_hdr(skb);
927 if (skb_queue_len(&sk->sk_write_queue) == 1) {
928 /* Only one fragment on the socket. */
929 skb->csum_start = skb_transport_header(skb) - skb->head;
930 skb->csum_offset = offsetof(struct udphdr, check);
931 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
934 * HW-checksum won't work as there are two or more
935 * fragments on the socket so that all csums of sk_buffs
938 offset = skb_transport_offset(skb);
939 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
941 skb->ip_summed = CHECKSUM_NONE;
943 skb_queue_walk(&sk->sk_write_queue, skb) {
944 csum = csum_add(csum, skb->csum);
947 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
950 uh->check = CSUM_MANGLED_0;
958 static int udp_v6_push_pending_frames(struct sock *sk)
962 struct udp_sock *up = udp_sk(sk);
963 struct inet_sock *inet = inet_sk(sk);
966 int is_udplite = IS_UDPLITE(sk);
969 if (up->pending == AF_INET)
970 return udp_push_pending_frames(sk);
972 fl6 = &inet->cork.fl.u.ip6;
974 /* Grab the skbuff where UDP header space exists. */
975 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
979 * Create a UDP header
982 uh->source = fl6->fl6_sport;
983 uh->dest = fl6->fl6_dport;
984 uh->len = htons(up->len);
988 csum = udplite_csum_outgoing(sk, skb);
989 else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
990 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr,
994 csum = udp_csum_outgoing(sk, skb);
996 /* add protocol-dependent pseudo-header */
997 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
998 up->len, fl6->flowi6_proto, csum);
1000 uh->check = CSUM_MANGLED_0;
1003 err = ip6_push_pending_frames(sk);
1005 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1006 UDP6_INC_STATS_USER(sock_net(sk),
1007 UDP_MIB_SNDBUFERRORS, is_udplite);
1011 UDP6_INC_STATS_USER(sock_net(sk),
1012 UDP_MIB_OUTDATAGRAMS, is_udplite);
1019 int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
1020 struct msghdr *msg, size_t len)
1022 struct ipv6_txoptions opt_space;
1023 struct udp_sock *up = udp_sk(sk);
1024 struct inet_sock *inet = inet_sk(sk);
1025 struct ipv6_pinfo *np = inet6_sk(sk);
1026 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) msg->msg_name;
1027 struct in6_addr *daddr, *final_p, final;
1028 struct ipv6_txoptions *opt = NULL;
1029 struct ip6_flowlabel *flowlabel = NULL;
1031 struct dst_entry *dst;
1032 int addr_len = msg->msg_namelen;
1037 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1040 int is_udplite = IS_UDPLITE(sk);
1041 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1043 /* destination address check */
1045 if (addr_len < offsetof(struct sockaddr, sa_data))
1048 switch (sin6->sin6_family) {
1050 if (addr_len < SIN6_LEN_RFC2133)
1052 daddr = &sin6->sin6_addr;
1055 goto do_udp_sendmsg;
1057 msg->msg_name = sin6 = NULL;
1058 msg->msg_namelen = addr_len = 0;
1064 } else if (!up->pending) {
1065 if (sk->sk_state != TCP_ESTABLISHED)
1066 return -EDESTADDRREQ;
1072 if (ipv6_addr_v4mapped(daddr)) {
1073 struct sockaddr_in sin;
1074 sin.sin_family = AF_INET;
1075 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1076 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1077 msg->msg_name = &sin;
1078 msg->msg_namelen = sizeof(sin);
1080 if (__ipv6_only_sock(sk))
1081 return -ENETUNREACH;
1082 return udp_sendmsg(iocb, sk, msg, len);
1086 if (up->pending == AF_INET)
1087 return udp_sendmsg(iocb, sk, msg, len);
1089 /* Rough check on arithmetic overflow,
1090 better check is made in ip6_append_data().
1092 if (len > INT_MAX - sizeof(struct udphdr))
1097 * There are pending frames.
1098 * The socket lock must be held while it's corked.
1101 if (likely(up->pending)) {
1102 if (unlikely(up->pending != AF_INET6)) {
1104 return -EAFNOSUPPORT;
1107 goto do_append_data;
1111 ulen += sizeof(struct udphdr);
1113 memset(&fl6, 0, sizeof(fl6));
1116 if (sin6->sin6_port == 0)
1119 fl6.fl6_dport = sin6->sin6_port;
1120 daddr = &sin6->sin6_addr;
1123 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1124 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1125 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1126 if (flowlabel == NULL)
1128 daddr = &flowlabel->dst;
1133 * Otherwise it will be difficult to maintain
1136 if (sk->sk_state == TCP_ESTABLISHED &&
1137 ipv6_addr_equal(daddr, &np->daddr))
1140 if (addr_len >= sizeof(struct sockaddr_in6) &&
1141 sin6->sin6_scope_id &&
1142 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1143 fl6.flowi6_oif = sin6->sin6_scope_id;
1145 if (sk->sk_state != TCP_ESTABLISHED)
1146 return -EDESTADDRREQ;
1148 fl6.fl6_dport = inet->inet_dport;
1150 fl6.flowlabel = np->flow_label;
1154 if (!fl6.flowi6_oif)
1155 fl6.flowi6_oif = sk->sk_bound_dev_if;
1157 if (!fl6.flowi6_oif)
1158 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1160 fl6.flowi6_mark = sk->sk_mark;
1162 if (msg->msg_controllen) {
1164 memset(opt, 0, sizeof(struct ipv6_txoptions));
1165 opt->tot_len = sizeof(*opt);
1167 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt,
1168 &hlimit, &tclass, &dontfrag);
1170 fl6_sock_release(flowlabel);
1173 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1174 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1175 if (flowlabel == NULL)
1178 if (!(opt->opt_nflen|opt->opt_flen))
1185 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1186 opt = ipv6_fixup_options(&opt_space, opt);
1188 fl6.flowi6_proto = sk->sk_protocol;
1189 if (!ipv6_addr_any(daddr))
1192 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1193 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1194 fl6.saddr = np->saddr;
1195 fl6.fl6_sport = inet->inet_sport;
1197 final_p = fl6_update_dst(&fl6, opt, &final);
1201 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1202 fl6.flowi6_oif = np->mcast_oif;
1204 } else if (!fl6.flowi6_oif)
1205 fl6.flowi6_oif = np->ucast_oif;
1207 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1209 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, true);
1217 if (ipv6_addr_is_multicast(&fl6.daddr))
1218 hlimit = np->mcast_hops;
1220 hlimit = np->hop_limit;
1222 hlimit = ip6_dst_hoplimit(dst);
1226 tclass = np->tclass;
1229 dontfrag = np->dontfrag;
1231 if (msg->msg_flags&MSG_CONFIRM)
1236 if (unlikely(up->pending)) {
1237 /* The socket is already corked while preparing it. */
1238 /* ... which is an evident application bug. --ANK */
1241 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
1246 up->pending = AF_INET6;
1250 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1251 err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen,
1252 sizeof(struct udphdr), hlimit, tclass, opt, &fl6,
1253 (struct rt6_info*)dst,
1254 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag);
1256 udp_v6_flush_pending_frames(sk);
1258 err = udp_v6_push_pending_frames(sk);
1259 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1264 ip6_dst_store(sk, dst,
1265 ipv6_addr_equal(&fl6.daddr, &np->daddr) ?
1267 #ifdef CONFIG_IPV6_SUBTREES
1268 ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1279 err = np->recverr ? net_xmit_errno(err) : 0;
1283 fl6_sock_release(flowlabel);
1287 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1288 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1289 * we don't have a good statistic (IpOutDiscards but it can be too many
1290 * things). We could add another new stat but at least for now that
1291 * seems like overkill.
1293 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1294 UDP6_INC_STATS_USER(sock_net(sk),
1295 UDP_MIB_SNDBUFERRORS, is_udplite);
1301 if (!(msg->msg_flags&MSG_PROBE) || len)
1302 goto back_from_confirm;
1307 void udpv6_destroy_sock(struct sock *sk)
1309 struct udp_sock *up = udp_sk(sk);
1311 udp_v6_flush_pending_frames(sk);
1314 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
1315 void (*encap_destroy)(struct sock *sk);
1316 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1321 inet6_destroy_sock(sk);
1325 * Socket option code for UDP
1327 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1328 char __user *optval, unsigned int optlen)
1330 if (level == SOL_UDP || level == SOL_UDPLITE)
1331 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1332 udp_v6_push_pending_frames);
1333 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1336 #ifdef CONFIG_COMPAT
1337 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1338 char __user *optval, unsigned int optlen)
1340 if (level == SOL_UDP || level == SOL_UDPLITE)
1341 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1342 udp_v6_push_pending_frames);
1343 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1347 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1348 char __user *optval, int __user *optlen)
1350 if (level == SOL_UDP || level == SOL_UDPLITE)
1351 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1352 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1355 #ifdef CONFIG_COMPAT
1356 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1357 char __user *optval, int __user *optlen)
1359 if (level == SOL_UDP || level == SOL_UDPLITE)
1360 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1361 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1365 static const struct inet6_protocol udpv6_protocol = {
1366 .handler = udpv6_rcv,
1367 .err_handler = udpv6_err,
1368 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1371 /* ------------------------------------------------------------------------ */
1372 #ifdef CONFIG_PROC_FS
1373 int udp6_seq_show(struct seq_file *seq, void *v)
1375 if (v == SEQ_START_TOKEN) {
1376 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1378 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1379 struct inet_sock *inet = inet_sk(v);
1380 __u16 srcp = ntohs(inet->inet_sport);
1381 __u16 destp = ntohs(inet->inet_dport);
1382 ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket);
1387 static const struct file_operations udp6_afinfo_seq_fops = {
1388 .owner = THIS_MODULE,
1389 .open = udp_seq_open,
1391 .llseek = seq_lseek,
1392 .release = seq_release_net
1395 static struct udp_seq_afinfo udp6_seq_afinfo = {
1398 .udp_table = &udp_table,
1399 .seq_fops = &udp6_afinfo_seq_fops,
1401 .show = udp6_seq_show,
1405 int __net_init udp6_proc_init(struct net *net)
1407 return udp_proc_register(net, &udp6_seq_afinfo);
1410 void udp6_proc_exit(struct net *net) {
1411 udp_proc_unregister(net, &udp6_seq_afinfo);
1413 #endif /* CONFIG_PROC_FS */
1415 void udp_v6_clear_sk(struct sock *sk, int size)
1417 struct inet_sock *inet = inet_sk(sk);
1419 /* we do not want to clear pinet6 field, because of RCU lookups */
1420 sk_prot_clear_portaddr_nulls(sk, offsetof(struct inet_sock, pinet6));
1422 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1423 memset(&inet->pinet6 + 1, 0, size);
1426 /* ------------------------------------------------------------------------ */
1428 struct proto udpv6_prot = {
1430 .owner = THIS_MODULE,
1431 .close = udp_lib_close,
1432 .connect = ip6_datagram_connect,
1433 .disconnect = udp_disconnect,
1435 .destroy = udpv6_destroy_sock,
1436 .setsockopt = udpv6_setsockopt,
1437 .getsockopt = udpv6_getsockopt,
1438 .sendmsg = udpv6_sendmsg,
1439 .recvmsg = udpv6_recvmsg,
1440 .backlog_rcv = __udpv6_queue_rcv_skb,
1441 .hash = udp_lib_hash,
1442 .unhash = udp_lib_unhash,
1443 .rehash = udp_v6_rehash,
1444 .get_port = udp_v6_get_port,
1445 .memory_allocated = &udp_memory_allocated,
1446 .sysctl_mem = sysctl_udp_mem,
1447 .sysctl_wmem = &sysctl_udp_wmem_min,
1448 .sysctl_rmem = &sysctl_udp_rmem_min,
1449 .obj_size = sizeof(struct udp6_sock),
1450 .slab_flags = SLAB_DESTROY_BY_RCU,
1451 .h.udp_table = &udp_table,
1452 #ifdef CONFIG_COMPAT
1453 .compat_setsockopt = compat_udpv6_setsockopt,
1454 .compat_getsockopt = compat_udpv6_getsockopt,
1456 .clear_sk = udp_v6_clear_sk,
1459 static struct inet_protosw udpv6_protosw = {
1461 .protocol = IPPROTO_UDP,
1462 .prot = &udpv6_prot,
1463 .ops = &inet6_dgram_ops,
1464 .no_check = UDP_CSUM_DEFAULT,
1465 .flags = INET_PROTOSW_PERMANENT,
1469 int __init udpv6_init(void)
1473 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1477 ret = inet6_register_protosw(&udpv6_protosw);
1479 goto out_udpv6_protocol;
1484 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1488 void udpv6_exit(void)
1490 inet6_unregister_protosw(&udpv6_protosw);
1491 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);