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 *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2);
59 int sk_ipv6only = ipv6_only_sock(sk);
60 int sk2_ipv6only = inet_v6_ipv6only(sk2);
61 int addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
62 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
64 /* if both are mapped, treat as IPv4 */
65 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED)
66 return (!sk2_ipv6only &&
67 (!sk->sk_rcv_saddr || !sk2->sk_rcv_saddr ||
68 sk->sk_rcv_saddr == sk2->sk_rcv_saddr));
70 if (addr_type2 == IPV6_ADDR_ANY &&
71 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
74 if (addr_type == IPV6_ADDR_ANY &&
75 !(sk_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
79 ipv6_addr_equal(&sk->sk_v6_rcv_saddr, sk2_rcv_saddr6))
85 static unsigned int udp6_portaddr_hash(struct net *net,
86 const struct in6_addr *addr6,
89 unsigned int hash, mix = net_hash_mix(net);
91 if (ipv6_addr_any(addr6))
92 hash = jhash_1word(0, mix);
93 else if (ipv6_addr_v4mapped(addr6))
94 hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix);
96 hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix);
102 int udp_v6_get_port(struct sock *sk, unsigned short snum)
104 unsigned int hash2_nulladdr =
105 udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
106 unsigned int hash2_partial =
107 udp6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
109 /* precompute partial secondary hash */
110 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
111 return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal, hash2_nulladdr);
114 static void udp_v6_rehash(struct sock *sk)
116 u16 new_hash = udp6_portaddr_hash(sock_net(sk),
117 &sk->sk_v6_rcv_saddr,
118 inet_sk(sk)->inet_num);
120 udp_lib_rehash(sk, new_hash);
123 static inline int compute_score(struct sock *sk, struct net *net,
125 const struct in6_addr *saddr, __be16 sport,
126 const struct in6_addr *daddr, __be16 dport,
131 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
132 sk->sk_family == PF_INET6) {
133 struct inet_sock *inet = inet_sk(sk);
136 if (inet->inet_dport) {
137 if (inet->inet_dport != sport)
141 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
142 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
146 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
147 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
151 if (sk->sk_bound_dev_if) {
152 if (sk->sk_bound_dev_if != dif)
160 #define SCORE2_MAX (1 + 1 + 1)
161 static inline int compute_score2(struct sock *sk, struct net *net,
162 const struct in6_addr *saddr, __be16 sport,
163 const struct in6_addr *daddr, unsigned short hnum,
168 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
169 sk->sk_family == PF_INET6) {
170 struct inet_sock *inet = inet_sk(sk);
172 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
175 if (inet->inet_dport) {
176 if (inet->inet_dport != sport)
180 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
181 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
185 if (sk->sk_bound_dev_if) {
186 if (sk->sk_bound_dev_if != dif)
195 /* called with read_rcu_lock() */
196 static struct sock *udp6_lib_lookup2(struct net *net,
197 const struct in6_addr *saddr, __be16 sport,
198 const struct in6_addr *daddr, unsigned int hnum, int dif,
199 struct udp_hslot *hslot2, unsigned int slot2)
201 struct sock *sk, *result;
202 struct hlist_nulls_node *node;
203 int score, badness, matches = 0, reuseport = 0;
209 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
210 score = compute_score2(sk, net, saddr, sport,
212 if (score > badness) {
215 reuseport = sk->sk_reuseport;
217 hash = inet6_ehashfn(net, daddr, hnum,
220 } else if (score == SCORE2_MAX)
222 } else if (score == badness && reuseport) {
224 if (((u64)hash * matches) >> 32 == 0)
226 hash = next_pseudo_random32(hash);
230 * if the nulls value we got at the end of this lookup is
231 * not the expected one, we must restart lookup.
232 * We probably met an item that was moved to another chain.
234 if (get_nulls_value(node) != slot2)
239 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
241 else if (unlikely(compute_score2(result, net, saddr, sport,
242 daddr, hnum, dif) < badness)) {
250 struct sock *__udp6_lib_lookup(struct net *net,
251 const struct in6_addr *saddr, __be16 sport,
252 const struct in6_addr *daddr, __be16 dport,
253 int dif, struct udp_table *udptable)
255 struct sock *sk, *result;
256 struct hlist_nulls_node *node;
257 unsigned short hnum = ntohs(dport);
258 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
259 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
260 int score, badness, matches = 0, reuseport = 0;
264 if (hslot->count > 10) {
265 hash2 = udp6_portaddr_hash(net, daddr, hnum);
266 slot2 = hash2 & udptable->mask;
267 hslot2 = &udptable->hash2[slot2];
268 if (hslot->count < hslot2->count)
271 result = udp6_lib_lookup2(net, saddr, sport,
275 hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
276 slot2 = hash2 & udptable->mask;
277 hslot2 = &udptable->hash2[slot2];
278 if (hslot->count < hslot2->count)
281 result = udp6_lib_lookup2(net, saddr, sport,
282 &in6addr_any, hnum, dif,
291 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
292 score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif);
293 if (score > badness) {
296 reuseport = sk->sk_reuseport;
298 hash = inet6_ehashfn(net, daddr, hnum,
302 } else if (score == badness && reuseport) {
304 if (((u64)hash * matches) >> 32 == 0)
306 hash = next_pseudo_random32(hash);
310 * if the nulls value we got at the end of this lookup is
311 * not the expected one, we must restart lookup.
312 * We probably met an item that was moved to another chain.
314 if (get_nulls_value(node) != slot)
318 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
320 else if (unlikely(compute_score(result, net, hnum, saddr, sport,
321 daddr, dport, dif) < badness)) {
329 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
331 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
332 __be16 sport, __be16 dport,
333 struct udp_table *udptable)
336 const struct ipv6hdr *iph = ipv6_hdr(skb);
338 if (unlikely(sk = skb_steal_sock(skb)))
340 return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport,
341 &iph->daddr, dport, inet6_iif(skb),
345 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
346 const struct in6_addr *daddr, __be16 dport, int dif)
348 return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
350 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
354 * This should be easy, if there is something there we
355 * return it, otherwise we block.
358 int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk,
359 struct msghdr *msg, size_t len,
360 int noblock, int flags, int *addr_len)
362 struct ipv6_pinfo *np = inet6_sk(sk);
363 struct inet_sock *inet = inet_sk(sk);
365 unsigned int ulen, copied;
368 int is_udplite = IS_UDPLITE(sk);
373 *addr_len = sizeof(struct sockaddr_in6);
375 if (flags & MSG_ERRQUEUE)
376 return ipv6_recv_error(sk, msg, len);
378 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
379 return ipv6_recv_rxpmtu(sk, msg, len);
382 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
383 &peeked, &off, &err);
387 ulen = skb->len - sizeof(struct udphdr);
391 else if (copied < ulen)
392 msg->msg_flags |= MSG_TRUNC;
394 is_udp4 = (skb->protocol == htons(ETH_P_IP));
397 * If checksum is needed at all, try to do it while copying the
398 * data. If the data is truncated, or if we only want a partial
399 * coverage checksum (UDP-Lite), do it before the copy.
402 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
403 if (udp_lib_checksum_complete(skb))
407 if (skb_csum_unnecessary(skb))
408 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
409 msg->msg_iov, copied);
411 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
416 trace_kfree_skb(skb, udpv6_recvmsg);
418 atomic_inc(&sk->sk_drops);
420 UDP_INC_STATS_USER(sock_net(sk),
424 UDP6_INC_STATS_USER(sock_net(sk),
432 UDP_INC_STATS_USER(sock_net(sk),
433 UDP_MIB_INDATAGRAMS, is_udplite);
435 UDP6_INC_STATS_USER(sock_net(sk),
436 UDP_MIB_INDATAGRAMS, is_udplite);
439 sock_recv_ts_and_drops(msg, sk, skb);
441 /* Copy the address. */
443 struct sockaddr_in6 *sin6;
445 sin6 = (struct sockaddr_in6 *) msg->msg_name;
446 sin6->sin6_family = AF_INET6;
447 sin6->sin6_port = udp_hdr(skb)->source;
448 sin6->sin6_flowinfo = 0;
451 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
453 sin6->sin6_scope_id = 0;
455 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
456 sin6->sin6_scope_id =
457 ipv6_iface_scope_id(&sin6->sin6_addr,
463 if (inet->cmsg_flags)
464 ip_cmsg_recv(msg, skb);
467 ip6_datagram_recv_ctl(sk, msg, skb);
471 if (flags & MSG_TRUNC)
475 skb_free_datagram_locked(sk, skb);
480 slow = lock_sock_fast(sk);
481 if (!skb_kill_datagram(sk, skb, flags)) {
483 UDP_INC_STATS_USER(sock_net(sk),
484 UDP_MIB_CSUMERRORS, is_udplite);
485 UDP_INC_STATS_USER(sock_net(sk),
486 UDP_MIB_INERRORS, is_udplite);
488 UDP6_INC_STATS_USER(sock_net(sk),
489 UDP_MIB_CSUMERRORS, is_udplite);
490 UDP6_INC_STATS_USER(sock_net(sk),
491 UDP_MIB_INERRORS, is_udplite);
494 unlock_sock_fast(sk, slow);
499 /* starting over for a new packet */
500 msg->msg_flags &= ~MSG_TRUNC;
504 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
505 u8 type, u8 code, int offset, __be32 info,
506 struct udp_table *udptable)
508 struct ipv6_pinfo *np;
509 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
510 const struct in6_addr *saddr = &hdr->saddr;
511 const struct in6_addr *daddr = &hdr->daddr;
512 struct udphdr *uh = (struct udphdr*)(skb->data+offset);
516 sk = __udp6_lib_lookup(dev_net(skb->dev), daddr, uh->dest,
517 saddr, uh->source, inet6_iif(skb), udptable);
521 if (type == ICMPV6_PKT_TOOBIG)
522 ip6_sk_update_pmtu(skb, sk, info);
523 if (type == NDISC_REDIRECT) {
524 ip6_sk_redirect(skb, sk);
530 if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
533 if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
537 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
540 sk->sk_error_report(sk);
545 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
549 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
550 sock_rps_save_rxhash(sk, skb);
551 sk_mark_napi_id(sk, skb);
554 rc = sock_queue_rcv_skb(sk, skb);
556 int is_udplite = IS_UDPLITE(sk);
558 /* Note that an ENOMEM error is charged twice */
560 UDP6_INC_STATS_BH(sock_net(sk),
561 UDP_MIB_RCVBUFERRORS, is_udplite);
562 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
569 static __inline__ void udpv6_err(struct sk_buff *skb,
570 struct inet6_skb_parm *opt, u8 type,
571 u8 code, int offset, __be32 info )
573 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
576 static struct static_key udpv6_encap_needed __read_mostly;
577 void udpv6_encap_enable(void)
579 if (!static_key_enabled(&udpv6_encap_needed))
580 static_key_slow_inc(&udpv6_encap_needed);
582 EXPORT_SYMBOL(udpv6_encap_enable);
584 int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
586 struct udp_sock *up = udp_sk(sk);
588 int is_udplite = IS_UDPLITE(sk);
590 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
593 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
594 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
597 * This is an encapsulation socket so pass the skb to
598 * the socket's udp_encap_rcv() hook. Otherwise, just
599 * fall through and pass this up the UDP socket.
600 * up->encap_rcv() returns the following value:
601 * =0 if skb was successfully passed to the encap
602 * handler or was discarded by it.
603 * >0 if skb should be passed on to UDP.
604 * <0 if skb should be resubmitted as proto -N
607 /* if we're overly short, let UDP handle it */
608 encap_rcv = ACCESS_ONCE(up->encap_rcv);
609 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
612 ret = encap_rcv(sk, skb);
614 UDP_INC_STATS_BH(sock_net(sk),
621 /* FALLTHROUGH -- it's a UDP Packet */
625 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
627 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
629 if (up->pcrlen == 0) { /* full coverage was set */
630 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage"
631 " %d while full coverage %d requested\n",
632 UDP_SKB_CB(skb)->cscov, skb->len);
635 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
636 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d "
637 "too small, need min %d\n",
638 UDP_SKB_CB(skb)->cscov, up->pcrlen);
643 if (rcu_access_pointer(sk->sk_filter)) {
644 if (udp_lib_checksum_complete(skb))
648 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf))
655 if (!sock_owned_by_user(sk))
656 rc = __udpv6_queue_rcv_skb(sk, skb);
657 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
665 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
667 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
668 atomic_inc(&sk->sk_drops);
673 static struct sock *udp_v6_mcast_next(struct net *net, struct sock *sk,
674 __be16 loc_port, const struct in6_addr *loc_addr,
675 __be16 rmt_port, const struct in6_addr *rmt_addr,
678 struct hlist_nulls_node *node;
680 unsigned short num = ntohs(loc_port);
682 sk_nulls_for_each_from(s, node) {
683 struct inet_sock *inet = inet_sk(s);
685 if (!net_eq(sock_net(s), net))
688 if (udp_sk(s)->udp_port_hash == num &&
689 s->sk_family == PF_INET6) {
690 if (inet->inet_dport) {
691 if (inet->inet_dport != rmt_port)
694 if (!ipv6_addr_any(&sk->sk_v6_daddr) &&
695 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr))
698 if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)
701 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
702 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))
705 if (!inet6_mc_check(s, loc_addr, rmt_addr))
713 static void flush_stack(struct sock **stack, unsigned int count,
714 struct sk_buff *skb, unsigned int final)
716 struct sk_buff *skb1 = NULL;
720 for (i = 0; i < count; i++) {
722 if (likely(skb1 == NULL))
723 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
725 atomic_inc(&sk->sk_drops);
726 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
728 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
732 if (skb1 && udpv6_queue_rcv_skb(sk, skb1) <= 0)
739 * Note: called only from the BH handler context,
740 * so we don't need to lock the hashes.
742 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
743 const struct in6_addr *saddr, const struct in6_addr *daddr,
744 struct udp_table *udptable)
746 struct sock *sk, *stack[256 / sizeof(struct sock *)];
747 const struct udphdr *uh = udp_hdr(skb);
748 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
750 unsigned int i, count = 0;
752 spin_lock(&hslot->lock);
753 sk = sk_nulls_head(&hslot->head);
754 dif = inet6_iif(skb);
755 sk = udp_v6_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
758 sk = udp_v6_mcast_next(net, sk_nulls_next(sk), uh->dest, daddr,
759 uh->source, saddr, dif);
760 if (unlikely(count == ARRAY_SIZE(stack))) {
763 flush_stack(stack, count, skb, ~0);
768 * before releasing the lock, we must take reference on sockets
770 for (i = 0; i < count; i++)
773 spin_unlock(&hslot->lock);
776 flush_stack(stack, count, skb, count - 1);
778 for (i = 0; i < count; i++)
786 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
789 struct net *net = dev_net(skb->dev);
792 const struct in6_addr *saddr, *daddr;
795 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
798 saddr = &ipv6_hdr(skb)->saddr;
799 daddr = &ipv6_hdr(skb)->daddr;
802 ulen = ntohs(uh->len);
806 if (proto == IPPROTO_UDP) {
807 /* UDP validates ulen. */
809 /* Check for jumbo payload */
813 if (ulen < sizeof(*uh))
816 if (ulen < skb->len) {
817 if (pskb_trim_rcsum(skb, ulen))
819 saddr = &ipv6_hdr(skb)->saddr;
820 daddr = &ipv6_hdr(skb)->daddr;
825 if (udp6_csum_init(skb, uh, proto))
829 * Multicast receive code
831 if (ipv6_addr_is_multicast(daddr))
832 return __udp6_lib_mcast_deliver(net, skb,
833 saddr, daddr, udptable);
838 * check socket cache ... must talk to Alan about his plans
839 * for sock caches... i'll skip this for now.
841 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
845 ret = udpv6_queue_rcv_skb(sk, skb);
848 /* a return value > 0 means to resubmit the input, but
849 * it wants the return to be -protocol, or 0
857 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
860 if (udp_lib_checksum_complete(skb))
863 UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
864 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
870 LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
871 proto == IPPROTO_UDPLITE ? "-Lite" : "",
880 UDP6_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
882 UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
887 static __inline__ int udpv6_rcv(struct sk_buff *skb)
889 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
893 * Throw away all pending data and cancel the corking. Socket is locked.
895 static void udp_v6_flush_pending_frames(struct sock *sk)
897 struct udp_sock *up = udp_sk(sk);
899 if (up->pending == AF_INET)
900 udp_flush_pending_frames(sk);
901 else if (up->pending) {
904 ip6_flush_pending_frames(sk);
909 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
910 * @sk: socket we are sending on
911 * @skb: sk_buff containing the filled-in UDP header
912 * (checksum field must be zeroed out)
914 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
915 const struct in6_addr *saddr,
916 const struct in6_addr *daddr, int len)
919 struct udphdr *uh = udp_hdr(skb);
922 if (skb_queue_len(&sk->sk_write_queue) == 1) {
923 /* Only one fragment on the socket. */
924 skb->csum_start = skb_transport_header(skb) - skb->head;
925 skb->csum_offset = offsetof(struct udphdr, check);
926 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
929 * HW-checksum won't work as there are two or more
930 * fragments on the socket so that all csums of sk_buffs
933 offset = skb_transport_offset(skb);
934 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
936 skb->ip_summed = CHECKSUM_NONE;
938 skb_queue_walk(&sk->sk_write_queue, skb) {
939 csum = csum_add(csum, skb->csum);
942 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
945 uh->check = CSUM_MANGLED_0;
953 static int udp_v6_push_pending_frames(struct sock *sk)
957 struct udp_sock *up = udp_sk(sk);
958 struct inet_sock *inet = inet_sk(sk);
961 int is_udplite = IS_UDPLITE(sk);
964 if (up->pending == AF_INET)
965 return udp_push_pending_frames(sk);
967 fl6 = &inet->cork.fl.u.ip6;
969 /* Grab the skbuff where UDP header space exists. */
970 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
974 * Create a UDP header
977 uh->source = fl6->fl6_sport;
978 uh->dest = fl6->fl6_dport;
979 uh->len = htons(up->len);
983 csum = udplite_csum_outgoing(sk, skb);
984 else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
985 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr,
989 csum = udp_csum_outgoing(sk, skb);
991 /* add protocol-dependent pseudo-header */
992 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
993 up->len, fl6->flowi6_proto, csum);
995 uh->check = CSUM_MANGLED_0;
998 err = ip6_push_pending_frames(sk);
1000 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1001 UDP6_INC_STATS_USER(sock_net(sk),
1002 UDP_MIB_SNDBUFERRORS, is_udplite);
1006 UDP6_INC_STATS_USER(sock_net(sk),
1007 UDP_MIB_OUTDATAGRAMS, is_udplite);
1014 int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
1015 struct msghdr *msg, size_t len)
1017 struct ipv6_txoptions opt_space;
1018 struct udp_sock *up = udp_sk(sk);
1019 struct inet_sock *inet = inet_sk(sk);
1020 struct ipv6_pinfo *np = inet6_sk(sk);
1021 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) msg->msg_name;
1022 struct in6_addr *daddr, *final_p, final;
1023 struct ipv6_txoptions *opt = NULL;
1024 struct ip6_flowlabel *flowlabel = NULL;
1026 struct dst_entry *dst;
1027 int addr_len = msg->msg_namelen;
1032 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1035 int is_udplite = IS_UDPLITE(sk);
1036 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1038 /* destination address check */
1040 if (addr_len < offsetof(struct sockaddr, sa_data))
1043 switch (sin6->sin6_family) {
1045 if (addr_len < SIN6_LEN_RFC2133)
1047 daddr = &sin6->sin6_addr;
1050 goto do_udp_sendmsg;
1052 msg->msg_name = sin6 = NULL;
1053 msg->msg_namelen = addr_len = 0;
1059 } else if (!up->pending) {
1060 if (sk->sk_state != TCP_ESTABLISHED)
1061 return -EDESTADDRREQ;
1062 daddr = &sk->sk_v6_daddr;
1067 if (ipv6_addr_v4mapped(daddr)) {
1068 struct sockaddr_in sin;
1069 sin.sin_family = AF_INET;
1070 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1071 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1072 msg->msg_name = &sin;
1073 msg->msg_namelen = sizeof(sin);
1075 if (__ipv6_only_sock(sk))
1076 return -ENETUNREACH;
1077 return udp_sendmsg(iocb, sk, msg, len);
1081 if (up->pending == AF_INET)
1082 return udp_sendmsg(iocb, sk, msg, len);
1084 /* Rough check on arithmetic overflow,
1085 better check is made in ip6_append_data().
1087 if (len > INT_MAX - sizeof(struct udphdr))
1092 * There are pending frames.
1093 * The socket lock must be held while it's corked.
1096 if (likely(up->pending)) {
1097 if (unlikely(up->pending != AF_INET6)) {
1099 return -EAFNOSUPPORT;
1102 goto do_append_data;
1106 ulen += sizeof(struct udphdr);
1108 memset(&fl6, 0, sizeof(fl6));
1111 if (sin6->sin6_port == 0)
1114 fl6.fl6_dport = sin6->sin6_port;
1115 daddr = &sin6->sin6_addr;
1118 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1119 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1120 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1121 if (flowlabel == NULL)
1123 daddr = &flowlabel->dst;
1128 * Otherwise it will be difficult to maintain
1131 if (sk->sk_state == TCP_ESTABLISHED &&
1132 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1133 daddr = &sk->sk_v6_daddr;
1135 if (addr_len >= sizeof(struct sockaddr_in6) &&
1136 sin6->sin6_scope_id &&
1137 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1138 fl6.flowi6_oif = sin6->sin6_scope_id;
1140 if (sk->sk_state != TCP_ESTABLISHED)
1141 return -EDESTADDRREQ;
1143 fl6.fl6_dport = inet->inet_dport;
1144 daddr = &sk->sk_v6_daddr;
1145 fl6.flowlabel = np->flow_label;
1149 if (!fl6.flowi6_oif)
1150 fl6.flowi6_oif = sk->sk_bound_dev_if;
1152 if (!fl6.flowi6_oif)
1153 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1155 fl6.flowi6_mark = sk->sk_mark;
1157 if (msg->msg_controllen) {
1159 memset(opt, 0, sizeof(struct ipv6_txoptions));
1160 opt->tot_len = sizeof(*opt);
1162 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt,
1163 &hlimit, &tclass, &dontfrag);
1165 fl6_sock_release(flowlabel);
1168 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1169 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1170 if (flowlabel == NULL)
1173 if (!(opt->opt_nflen|opt->opt_flen))
1180 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1181 opt = ipv6_fixup_options(&opt_space, opt);
1183 fl6.flowi6_proto = sk->sk_protocol;
1184 if (!ipv6_addr_any(daddr))
1187 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1188 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1189 fl6.saddr = np->saddr;
1190 fl6.fl6_sport = inet->inet_sport;
1192 final_p = fl6_update_dst(&fl6, opt, &final);
1196 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1197 fl6.flowi6_oif = np->mcast_oif;
1199 } else if (!fl6.flowi6_oif)
1200 fl6.flowi6_oif = np->ucast_oif;
1202 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1204 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, true);
1212 if (ipv6_addr_is_multicast(&fl6.daddr))
1213 hlimit = np->mcast_hops;
1215 hlimit = np->hop_limit;
1217 hlimit = ip6_dst_hoplimit(dst);
1221 tclass = np->tclass;
1224 dontfrag = np->dontfrag;
1226 if (msg->msg_flags&MSG_CONFIRM)
1231 if (unlikely(up->pending)) {
1232 /* The socket is already corked while preparing it. */
1233 /* ... which is an evident application bug. --ANK */
1236 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
1241 up->pending = AF_INET6;
1245 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1246 err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen,
1247 sizeof(struct udphdr), hlimit, tclass, opt, &fl6,
1248 (struct rt6_info*)dst,
1249 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag);
1251 udp_v6_flush_pending_frames(sk);
1253 err = udp_v6_push_pending_frames(sk);
1254 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1259 ip6_dst_store(sk, dst,
1260 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ?
1261 &sk->sk_v6_daddr : NULL,
1262 #ifdef CONFIG_IPV6_SUBTREES
1263 ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1274 err = np->recverr ? net_xmit_errno(err) : 0;
1278 fl6_sock_release(flowlabel);
1282 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1283 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1284 * we don't have a good statistic (IpOutDiscards but it can be too many
1285 * things). We could add another new stat but at least for now that
1286 * seems like overkill.
1288 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1289 UDP6_INC_STATS_USER(sock_net(sk),
1290 UDP_MIB_SNDBUFERRORS, is_udplite);
1296 if (!(msg->msg_flags&MSG_PROBE) || len)
1297 goto back_from_confirm;
1302 void udpv6_destroy_sock(struct sock *sk)
1304 struct udp_sock *up = udp_sk(sk);
1306 udp_v6_flush_pending_frames(sk);
1309 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
1310 void (*encap_destroy)(struct sock *sk);
1311 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1316 inet6_destroy_sock(sk);
1320 * Socket option code for UDP
1322 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1323 char __user *optval, unsigned int optlen)
1325 if (level == SOL_UDP || level == SOL_UDPLITE)
1326 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1327 udp_v6_push_pending_frames);
1328 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1331 #ifdef CONFIG_COMPAT
1332 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1333 char __user *optval, unsigned int optlen)
1335 if (level == SOL_UDP || level == SOL_UDPLITE)
1336 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1337 udp_v6_push_pending_frames);
1338 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1342 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1343 char __user *optval, int __user *optlen)
1345 if (level == SOL_UDP || level == SOL_UDPLITE)
1346 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1347 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1350 #ifdef CONFIG_COMPAT
1351 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1352 char __user *optval, int __user *optlen)
1354 if (level == SOL_UDP || level == SOL_UDPLITE)
1355 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1356 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1360 static const struct inet6_protocol udpv6_protocol = {
1361 .handler = udpv6_rcv,
1362 .err_handler = udpv6_err,
1363 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1366 /* ------------------------------------------------------------------------ */
1367 #ifdef CONFIG_PROC_FS
1368 int udp6_seq_show(struct seq_file *seq, void *v)
1370 if (v == SEQ_START_TOKEN) {
1371 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1373 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1374 struct inet_sock *inet = inet_sk(v);
1375 __u16 srcp = ntohs(inet->inet_sport);
1376 __u16 destp = ntohs(inet->inet_dport);
1377 ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket);
1382 static const struct file_operations udp6_afinfo_seq_fops = {
1383 .owner = THIS_MODULE,
1384 .open = udp_seq_open,
1386 .llseek = seq_lseek,
1387 .release = seq_release_net
1390 static struct udp_seq_afinfo udp6_seq_afinfo = {
1393 .udp_table = &udp_table,
1394 .seq_fops = &udp6_afinfo_seq_fops,
1396 .show = udp6_seq_show,
1400 int __net_init udp6_proc_init(struct net *net)
1402 return udp_proc_register(net, &udp6_seq_afinfo);
1405 void udp6_proc_exit(struct net *net) {
1406 udp_proc_unregister(net, &udp6_seq_afinfo);
1408 #endif /* CONFIG_PROC_FS */
1410 void udp_v6_clear_sk(struct sock *sk, int size)
1412 struct inet_sock *inet = inet_sk(sk);
1414 /* we do not want to clear pinet6 field, because of RCU lookups */
1415 sk_prot_clear_portaddr_nulls(sk, offsetof(struct inet_sock, pinet6));
1417 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1418 memset(&inet->pinet6 + 1, 0, size);
1421 /* ------------------------------------------------------------------------ */
1423 struct proto udpv6_prot = {
1425 .owner = THIS_MODULE,
1426 .close = udp_lib_close,
1427 .connect = ip6_datagram_connect,
1428 .disconnect = udp_disconnect,
1430 .destroy = udpv6_destroy_sock,
1431 .setsockopt = udpv6_setsockopt,
1432 .getsockopt = udpv6_getsockopt,
1433 .sendmsg = udpv6_sendmsg,
1434 .recvmsg = udpv6_recvmsg,
1435 .backlog_rcv = __udpv6_queue_rcv_skb,
1436 .hash = udp_lib_hash,
1437 .unhash = udp_lib_unhash,
1438 .rehash = udp_v6_rehash,
1439 .get_port = udp_v6_get_port,
1440 .memory_allocated = &udp_memory_allocated,
1441 .sysctl_mem = sysctl_udp_mem,
1442 .sysctl_wmem = &sysctl_udp_wmem_min,
1443 .sysctl_rmem = &sysctl_udp_rmem_min,
1444 .obj_size = sizeof(struct udp6_sock),
1445 .slab_flags = SLAB_DESTROY_BY_RCU,
1446 .h.udp_table = &udp_table,
1447 #ifdef CONFIG_COMPAT
1448 .compat_setsockopt = compat_udpv6_setsockopt,
1449 .compat_getsockopt = compat_udpv6_getsockopt,
1451 .clear_sk = udp_v6_clear_sk,
1454 static struct inet_protosw udpv6_protosw = {
1456 .protocol = IPPROTO_UDP,
1457 .prot = &udpv6_prot,
1458 .ops = &inet6_dgram_ops,
1459 .no_check = UDP_CSUM_DEFAULT,
1460 .flags = INET_PROTOSW_PERMANENT,
1464 int __init udpv6_init(void)
1468 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1472 ret = inet6_register_protosw(&udpv6_protosw);
1474 goto out_udpv6_protocol;
1479 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1483 void udpv6_exit(void)
1485 inet6_unregister_protosw(&udpv6_protosw);
1486 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);