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 sk2_ipv6only = inet_v6_ipv6only(sk2);
83 int addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
84 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
86 /* if both are mapped, treat as IPv4 */
87 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED)
88 return (!sk2_ipv6only &&
89 (!sk->sk_rcv_saddr || !sk2->sk_rcv_saddr ||
90 sk->sk_rcv_saddr == sk2->sk_rcv_saddr));
92 if (addr_type2 == IPV6_ADDR_ANY &&
93 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
96 if (addr_type == IPV6_ADDR_ANY &&
97 !(ipv6_only_sock(sk) && addr_type2 == IPV6_ADDR_MAPPED))
100 if (sk2_rcv_saddr6 &&
101 ipv6_addr_equal(&sk->sk_v6_rcv_saddr, sk2_rcv_saddr6))
107 static unsigned int udp6_portaddr_hash(struct net *net,
108 const struct in6_addr *addr6,
111 unsigned int hash, mix = net_hash_mix(net);
113 if (ipv6_addr_any(addr6))
114 hash = jhash_1word(0, mix);
115 else if (ipv6_addr_v4mapped(addr6))
116 hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix);
118 hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix);
124 int udp_v6_get_port(struct sock *sk, unsigned short snum)
126 unsigned int hash2_nulladdr =
127 udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
128 unsigned int hash2_partial =
129 udp6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
131 /* precompute partial secondary hash */
132 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
133 return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal, hash2_nulladdr);
136 static void udp_v6_rehash(struct sock *sk)
138 u16 new_hash = udp6_portaddr_hash(sock_net(sk),
139 &sk->sk_v6_rcv_saddr,
140 inet_sk(sk)->inet_num);
142 udp_lib_rehash(sk, new_hash);
145 static inline int compute_score(struct sock *sk, struct net *net,
147 const struct in6_addr *saddr, __be16 sport,
148 const struct in6_addr *daddr, __be16 dport,
153 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
154 sk->sk_family == PF_INET6) {
155 struct inet_sock *inet = inet_sk(sk);
158 if (inet->inet_dport) {
159 if (inet->inet_dport != sport)
163 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
164 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
168 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
169 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
173 if (sk->sk_bound_dev_if) {
174 if (sk->sk_bound_dev_if != dif)
182 #define SCORE2_MAX (1 + 1 + 1)
183 static inline int compute_score2(struct sock *sk, struct net *net,
184 const struct in6_addr *saddr, __be16 sport,
185 const struct in6_addr *daddr, unsigned short hnum,
190 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
191 sk->sk_family == PF_INET6) {
192 struct inet_sock *inet = inet_sk(sk);
194 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
197 if (inet->inet_dport) {
198 if (inet->inet_dport != sport)
202 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
203 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
207 if (sk->sk_bound_dev_if) {
208 if (sk->sk_bound_dev_if != dif)
217 /* called with read_rcu_lock() */
218 static struct sock *udp6_lib_lookup2(struct net *net,
219 const struct in6_addr *saddr, __be16 sport,
220 const struct in6_addr *daddr, unsigned int hnum, int dif,
221 struct udp_hslot *hslot2, unsigned int slot2)
223 struct sock *sk, *result;
224 struct hlist_nulls_node *node;
225 int score, badness, matches = 0, reuseport = 0;
231 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
232 score = compute_score2(sk, net, saddr, sport,
234 if (score > badness) {
237 reuseport = sk->sk_reuseport;
239 hash = udp6_ehashfn(net, daddr, hnum,
242 } else if (score == SCORE2_MAX)
244 } else if (score == badness && reuseport) {
246 if (reciprocal_scale(hash, matches) == 0)
248 hash = next_pseudo_random32(hash);
252 * if the nulls value we got at the end of this lookup is
253 * not the expected one, we must restart lookup.
254 * We probably met an item that was moved to another chain.
256 if (get_nulls_value(node) != slot2)
261 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
263 else if (unlikely(compute_score2(result, net, saddr, sport,
264 daddr, hnum, dif) < badness)) {
272 struct sock *__udp6_lib_lookup(struct net *net,
273 const struct in6_addr *saddr, __be16 sport,
274 const struct in6_addr *daddr, __be16 dport,
275 int dif, struct udp_table *udptable)
277 struct sock *sk, *result;
278 struct hlist_nulls_node *node;
279 unsigned short hnum = ntohs(dport);
280 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
281 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
282 int score, badness, matches = 0, reuseport = 0;
286 if (hslot->count > 10) {
287 hash2 = udp6_portaddr_hash(net, daddr, hnum);
288 slot2 = hash2 & udptable->mask;
289 hslot2 = &udptable->hash2[slot2];
290 if (hslot->count < hslot2->count)
293 result = udp6_lib_lookup2(net, saddr, sport,
297 hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
298 slot2 = hash2 & udptable->mask;
299 hslot2 = &udptable->hash2[slot2];
300 if (hslot->count < hslot2->count)
303 result = udp6_lib_lookup2(net, saddr, sport,
304 &in6addr_any, hnum, dif,
313 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
314 score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif);
315 if (score > badness) {
318 reuseport = sk->sk_reuseport;
320 hash = udp6_ehashfn(net, daddr, hnum,
324 } else if (score == badness && reuseport) {
326 if (reciprocal_scale(hash, matches) == 0)
328 hash = next_pseudo_random32(hash);
332 * if the nulls value we got at the end of this lookup is
333 * not the expected one, we must restart lookup.
334 * We probably met an item that was moved to another chain.
336 if (get_nulls_value(node) != slot)
340 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
342 else if (unlikely(compute_score(result, net, hnum, saddr, sport,
343 daddr, dport, dif) < badness)) {
351 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
353 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
354 __be16 sport, __be16 dport,
355 struct udp_table *udptable)
358 const struct ipv6hdr *iph = ipv6_hdr(skb);
360 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_msg(skb, sizeof(struct udphdr),
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);
537 struct net *net = dev_net(skb->dev);
539 sk = __udp6_lib_lookup(net, daddr, uh->dest,
540 saddr, uh->source, inet6_iif(skb), udptable);
542 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
547 if (type == ICMPV6_PKT_TOOBIG) {
548 if (!ip6_sk_accept_pmtu(sk))
550 ip6_sk_update_pmtu(skb, sk, info);
552 if (type == NDISC_REDIRECT) {
553 ip6_sk_redirect(skb, sk);
559 if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
562 if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
566 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
569 sk->sk_error_report(sk);
574 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
578 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
579 sock_rps_save_rxhash(sk, skb);
580 sk_mark_napi_id(sk, skb);
581 sk_incoming_cpu_update(sk);
584 rc = sock_queue_rcv_skb(sk, skb);
586 int is_udplite = IS_UDPLITE(sk);
588 /* Note that an ENOMEM error is charged twice */
590 UDP6_INC_STATS_BH(sock_net(sk),
591 UDP_MIB_RCVBUFERRORS, is_udplite);
592 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
599 static __inline__ void udpv6_err(struct sk_buff *skb,
600 struct inet6_skb_parm *opt, u8 type,
601 u8 code, int offset, __be32 info)
603 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
606 static struct static_key udpv6_encap_needed __read_mostly;
607 void udpv6_encap_enable(void)
609 if (!static_key_enabled(&udpv6_encap_needed))
610 static_key_slow_inc(&udpv6_encap_needed);
612 EXPORT_SYMBOL(udpv6_encap_enable);
614 int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
616 struct udp_sock *up = udp_sk(sk);
618 int is_udplite = IS_UDPLITE(sk);
620 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
623 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
624 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
627 * This is an encapsulation socket so pass the skb to
628 * the socket's udp_encap_rcv() hook. Otherwise, just
629 * fall through and pass this up the UDP socket.
630 * up->encap_rcv() returns the following value:
631 * =0 if skb was successfully passed to the encap
632 * handler or was discarded by it.
633 * >0 if skb should be passed on to UDP.
634 * <0 if skb should be resubmitted as proto -N
637 /* if we're overly short, let UDP handle it */
638 encap_rcv = ACCESS_ONCE(up->encap_rcv);
639 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
642 /* Verify checksum before giving to encap */
643 if (udp_lib_checksum_complete(skb))
646 ret = encap_rcv(sk, skb);
648 UDP_INC_STATS_BH(sock_net(sk),
655 /* FALLTHROUGH -- it's a UDP Packet */
659 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
661 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
663 if (up->pcrlen == 0) { /* full coverage was set */
664 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
665 UDP_SKB_CB(skb)->cscov, skb->len);
668 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
669 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
670 UDP_SKB_CB(skb)->cscov, up->pcrlen);
675 if (rcu_access_pointer(sk->sk_filter)) {
676 if (udp_lib_checksum_complete(skb))
680 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
681 UDP6_INC_STATS_BH(sock_net(sk),
682 UDP_MIB_RCVBUFERRORS, is_udplite);
690 if (!sock_owned_by_user(sk))
691 rc = __udpv6_queue_rcv_skb(sk, skb);
692 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
701 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
703 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
704 atomic_inc(&sk->sk_drops);
709 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
710 __be16 loc_port, const struct in6_addr *loc_addr,
711 __be16 rmt_port, const struct in6_addr *rmt_addr,
712 int dif, unsigned short hnum)
714 struct inet_sock *inet = inet_sk(sk);
716 if (!net_eq(sock_net(sk), net))
719 if (udp_sk(sk)->udp_port_hash != hnum ||
720 sk->sk_family != PF_INET6 ||
721 (inet->inet_dport && inet->inet_dport != rmt_port) ||
722 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
723 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
724 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
726 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
731 static void flush_stack(struct sock **stack, unsigned int count,
732 struct sk_buff *skb, unsigned int final)
734 struct sk_buff *skb1 = NULL;
738 for (i = 0; i < count; i++) {
740 if (likely(skb1 == NULL))
741 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
743 atomic_inc(&sk->sk_drops);
744 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
746 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
750 if (skb1 && udpv6_queue_rcv_skb(sk, skb1) <= 0)
758 static void udp6_csum_zero_error(struct sk_buff *skb)
760 /* RFC 2460 section 8.1 says that we SHOULD log
761 * this error. Well, it is reasonable.
763 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
764 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
765 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
769 * Note: called only from the BH handler context,
770 * so we don't need to lock the hashes.
772 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
773 const struct in6_addr *saddr, const struct in6_addr *daddr,
774 struct udp_table *udptable, int proto)
776 struct sock *sk, *stack[256 / sizeof(struct sock *)];
777 const struct udphdr *uh = udp_hdr(skb);
778 struct hlist_nulls_node *node;
779 unsigned short hnum = ntohs(uh->dest);
780 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
781 int dif = inet6_iif(skb);
782 unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node);
783 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
784 bool inner_flushed = false;
787 hash2_any = udp6_portaddr_hash(net, &in6addr_any, hnum) &
789 hash2 = udp6_portaddr_hash(net, daddr, hnum) & udp_table.mask;
791 hslot = &udp_table.hash2[hash2];
792 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
795 spin_lock(&hslot->lock);
796 sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) {
797 if (__udp_v6_is_mcast_sock(net, sk,
801 /* If zero checksum and no_check is not on for
802 * the socket then skip it.
804 (uh->check || udp_sk(sk)->no_check6_rx)) {
805 if (unlikely(count == ARRAY_SIZE(stack))) {
806 flush_stack(stack, count, skb, ~0);
807 inner_flushed = true;
815 spin_unlock(&hslot->lock);
817 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
818 if (use_hash2 && hash2 != hash2_any) {
824 flush_stack(stack, count, skb, count - 1);
827 UDP_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI,
828 proto == IPPROTO_UDPLITE);
834 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
837 struct net *net = dev_net(skb->dev);
840 const struct in6_addr *saddr, *daddr;
843 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
846 saddr = &ipv6_hdr(skb)->saddr;
847 daddr = &ipv6_hdr(skb)->daddr;
850 ulen = ntohs(uh->len);
854 if (proto == IPPROTO_UDP) {
855 /* UDP validates ulen. */
857 /* Check for jumbo payload */
861 if (ulen < sizeof(*uh))
864 if (ulen < skb->len) {
865 if (pskb_trim_rcsum(skb, ulen))
867 saddr = &ipv6_hdr(skb)->saddr;
868 daddr = &ipv6_hdr(skb)->daddr;
873 if (udp6_csum_init(skb, uh, proto))
877 * Multicast receive code
879 if (ipv6_addr_is_multicast(daddr))
880 return __udp6_lib_mcast_deliver(net, skb,
881 saddr, daddr, udptable, proto);
886 * check socket cache ... must talk to Alan about his plans
887 * for sock caches... i'll skip this for now.
889 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
893 if (!uh->check && !udp_sk(sk)->no_check6_rx) {
895 udp6_csum_zero_error(skb);
899 if (udp_sk(sk)->convert_csum && uh->check && !IS_UDPLITE(sk))
900 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
903 ret = udpv6_queue_rcv_skb(sk, skb);
906 /* a return value > 0 means to resubmit the input, but
907 * it wants the return to be -protocol, or 0
916 udp6_csum_zero_error(skb);
920 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
923 if (udp_lib_checksum_complete(skb))
926 UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
927 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
933 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
934 proto == IPPROTO_UDPLITE ? "-Lite" : "",
935 saddr, ntohs(uh->source),
937 daddr, ntohs(uh->dest));
940 UDP6_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
942 UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
947 static __inline__ int udpv6_rcv(struct sk_buff *skb)
949 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
953 * Throw away all pending data and cancel the corking. Socket is locked.
955 static void udp_v6_flush_pending_frames(struct sock *sk)
957 struct udp_sock *up = udp_sk(sk);
959 if (up->pending == AF_INET)
960 udp_flush_pending_frames(sk);
961 else if (up->pending) {
964 ip6_flush_pending_frames(sk);
969 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
970 * @sk: socket we are sending on
971 * @skb: sk_buff containing the filled-in UDP header
972 * (checksum field must be zeroed out)
974 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
975 const struct in6_addr *saddr,
976 const struct in6_addr *daddr, int len)
979 struct udphdr *uh = udp_hdr(skb);
982 if (skb_queue_len(&sk->sk_write_queue) == 1) {
983 /* Only one fragment on the socket. */
984 skb->csum_start = skb_transport_header(skb) - skb->head;
985 skb->csum_offset = offsetof(struct udphdr, check);
986 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
989 * HW-checksum won't work as there are two or more
990 * fragments on the socket so that all csums of sk_buffs
993 offset = skb_transport_offset(skb);
994 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
996 skb->ip_summed = CHECKSUM_NONE;
998 skb_queue_walk(&sk->sk_write_queue, skb) {
999 csum = csum_add(csum, skb->csum);
1002 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1005 uh->check = CSUM_MANGLED_0;
1013 static int udp_v6_push_pending_frames(struct sock *sk)
1015 struct sk_buff *skb;
1017 struct udp_sock *up = udp_sk(sk);
1018 struct inet_sock *inet = inet_sk(sk);
1021 int is_udplite = IS_UDPLITE(sk);
1024 if (up->pending == AF_INET)
1025 return udp_push_pending_frames(sk);
1027 fl6 = &inet->cork.fl.u.ip6;
1029 /* Grab the skbuff where UDP header space exists. */
1030 skb = skb_peek(&sk->sk_write_queue);
1035 * Create a UDP header
1038 uh->source = fl6->fl6_sport;
1039 uh->dest = fl6->fl6_dport;
1040 uh->len = htons(up->len);
1044 csum = udplite_csum_outgoing(sk, skb);
1045 else if (up->no_check6_tx) { /* UDP csum disabled */
1046 skb->ip_summed = CHECKSUM_NONE;
1048 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1049 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr,
1053 csum = udp_csum_outgoing(sk, skb);
1055 /* add protocol-dependent pseudo-header */
1056 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1057 up->len, fl6->flowi6_proto, csum);
1059 uh->check = CSUM_MANGLED_0;
1062 err = ip6_push_pending_frames(sk);
1064 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1065 UDP6_INC_STATS_USER(sock_net(sk),
1066 UDP_MIB_SNDBUFERRORS, is_udplite);
1070 UDP6_INC_STATS_USER(sock_net(sk),
1071 UDP_MIB_OUTDATAGRAMS, is_udplite);
1078 int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
1079 struct msghdr *msg, size_t len)
1081 struct ipv6_txoptions opt_space;
1082 struct udp_sock *up = udp_sk(sk);
1083 struct inet_sock *inet = inet_sk(sk);
1084 struct ipv6_pinfo *np = inet6_sk(sk);
1085 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1086 struct in6_addr *daddr, *final_p, final;
1087 struct ipv6_txoptions *opt = NULL;
1088 struct ip6_flowlabel *flowlabel = NULL;
1090 struct dst_entry *dst;
1091 int addr_len = msg->msg_namelen;
1096 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1099 int is_udplite = IS_UDPLITE(sk);
1100 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1102 /* destination address check */
1104 if (addr_len < offsetof(struct sockaddr, sa_data))
1107 switch (sin6->sin6_family) {
1109 if (addr_len < SIN6_LEN_RFC2133)
1111 daddr = &sin6->sin6_addr;
1114 goto do_udp_sendmsg;
1116 msg->msg_name = sin6 = NULL;
1117 msg->msg_namelen = addr_len = 0;
1123 } else if (!up->pending) {
1124 if (sk->sk_state != TCP_ESTABLISHED)
1125 return -EDESTADDRREQ;
1126 daddr = &sk->sk_v6_daddr;
1131 if (ipv6_addr_v4mapped(daddr)) {
1132 struct sockaddr_in sin;
1133 sin.sin_family = AF_INET;
1134 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1135 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1136 msg->msg_name = &sin;
1137 msg->msg_namelen = sizeof(sin);
1139 if (__ipv6_only_sock(sk))
1140 return -ENETUNREACH;
1141 return udp_sendmsg(iocb, sk, msg, len);
1145 if (up->pending == AF_INET)
1146 return udp_sendmsg(iocb, sk, msg, len);
1148 /* Rough check on arithmetic overflow,
1149 better check is made in ip6_append_data().
1151 if (len > INT_MAX - sizeof(struct udphdr))
1156 * There are pending frames.
1157 * The socket lock must be held while it's corked.
1160 if (likely(up->pending)) {
1161 if (unlikely(up->pending != AF_INET6)) {
1163 return -EAFNOSUPPORT;
1166 goto do_append_data;
1170 ulen += sizeof(struct udphdr);
1172 memset(&fl6, 0, sizeof(fl6));
1175 if (sin6->sin6_port == 0)
1178 fl6.fl6_dport = sin6->sin6_port;
1179 daddr = &sin6->sin6_addr;
1182 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1183 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1184 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1185 if (flowlabel == NULL)
1191 * Otherwise it will be difficult to maintain
1194 if (sk->sk_state == TCP_ESTABLISHED &&
1195 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1196 daddr = &sk->sk_v6_daddr;
1198 if (addr_len >= sizeof(struct sockaddr_in6) &&
1199 sin6->sin6_scope_id &&
1200 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1201 fl6.flowi6_oif = sin6->sin6_scope_id;
1203 if (sk->sk_state != TCP_ESTABLISHED)
1204 return -EDESTADDRREQ;
1206 fl6.fl6_dport = inet->inet_dport;
1207 daddr = &sk->sk_v6_daddr;
1208 fl6.flowlabel = np->flow_label;
1212 if (!fl6.flowi6_oif)
1213 fl6.flowi6_oif = sk->sk_bound_dev_if;
1215 if (!fl6.flowi6_oif)
1216 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1218 fl6.flowi6_mark = sk->sk_mark;
1220 if (msg->msg_controllen) {
1222 memset(opt, 0, sizeof(struct ipv6_txoptions));
1223 opt->tot_len = sizeof(*opt);
1225 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt,
1226 &hlimit, &tclass, &dontfrag);
1228 fl6_sock_release(flowlabel);
1231 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1232 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1233 if (flowlabel == NULL)
1236 if (!(opt->opt_nflen|opt->opt_flen))
1243 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1244 opt = ipv6_fixup_options(&opt_space, opt);
1246 fl6.flowi6_proto = sk->sk_protocol;
1247 if (!ipv6_addr_any(daddr))
1250 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1251 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1252 fl6.saddr = np->saddr;
1253 fl6.fl6_sport = inet->inet_sport;
1255 final_p = fl6_update_dst(&fl6, opt, &final);
1259 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1260 fl6.flowi6_oif = np->mcast_oif;
1262 } else if (!fl6.flowi6_oif)
1263 fl6.flowi6_oif = np->ucast_oif;
1265 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1267 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p);
1275 hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1278 tclass = np->tclass;
1280 if (msg->msg_flags&MSG_CONFIRM)
1285 if (unlikely(up->pending)) {
1286 /* The socket is already corked while preparing it. */
1287 /* ... which is an evident application bug. --ANK */
1290 net_dbg_ratelimited("udp cork app bug 2\n");
1295 up->pending = AF_INET6;
1299 dontfrag = np->dontfrag;
1301 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1302 err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen,
1303 sizeof(struct udphdr), hlimit, tclass, opt, &fl6,
1304 (struct rt6_info *)dst,
1305 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag);
1307 udp_v6_flush_pending_frames(sk);
1309 err = udp_v6_push_pending_frames(sk);
1310 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1315 ip6_dst_store(sk, dst,
1316 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ?
1317 &sk->sk_v6_daddr : NULL,
1318 #ifdef CONFIG_IPV6_SUBTREES
1319 ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1330 err = np->recverr ? net_xmit_errno(err) : 0;
1334 fl6_sock_release(flowlabel);
1338 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1339 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1340 * we don't have a good statistic (IpOutDiscards but it can be too many
1341 * things). We could add another new stat but at least for now that
1342 * seems like overkill.
1344 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1345 UDP6_INC_STATS_USER(sock_net(sk),
1346 UDP_MIB_SNDBUFERRORS, is_udplite);
1352 if (!(msg->msg_flags&MSG_PROBE) || len)
1353 goto back_from_confirm;
1358 void udpv6_destroy_sock(struct sock *sk)
1360 struct udp_sock *up = udp_sk(sk);
1362 udp_v6_flush_pending_frames(sk);
1365 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
1366 void (*encap_destroy)(struct sock *sk);
1367 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1372 inet6_destroy_sock(sk);
1376 * Socket option code for UDP
1378 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1379 char __user *optval, unsigned int optlen)
1381 if (level == SOL_UDP || level == SOL_UDPLITE)
1382 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1383 udp_v6_push_pending_frames);
1384 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1387 #ifdef CONFIG_COMPAT
1388 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1389 char __user *optval, unsigned int optlen)
1391 if (level == SOL_UDP || level == SOL_UDPLITE)
1392 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1393 udp_v6_push_pending_frames);
1394 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1398 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1399 char __user *optval, int __user *optlen)
1401 if (level == SOL_UDP || level == SOL_UDPLITE)
1402 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1403 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1406 #ifdef CONFIG_COMPAT
1407 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1408 char __user *optval, int __user *optlen)
1410 if (level == SOL_UDP || level == SOL_UDPLITE)
1411 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1412 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1416 static const struct inet6_protocol udpv6_protocol = {
1417 .handler = udpv6_rcv,
1418 .err_handler = udpv6_err,
1419 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1422 /* ------------------------------------------------------------------------ */
1423 #ifdef CONFIG_PROC_FS
1424 int udp6_seq_show(struct seq_file *seq, void *v)
1426 if (v == SEQ_START_TOKEN) {
1427 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1429 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1430 struct inet_sock *inet = inet_sk(v);
1431 __u16 srcp = ntohs(inet->inet_sport);
1432 __u16 destp = ntohs(inet->inet_dport);
1433 ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket);
1438 static const struct file_operations udp6_afinfo_seq_fops = {
1439 .owner = THIS_MODULE,
1440 .open = udp_seq_open,
1442 .llseek = seq_lseek,
1443 .release = seq_release_net
1446 static struct udp_seq_afinfo udp6_seq_afinfo = {
1449 .udp_table = &udp_table,
1450 .seq_fops = &udp6_afinfo_seq_fops,
1452 .show = udp6_seq_show,
1456 int __net_init udp6_proc_init(struct net *net)
1458 return udp_proc_register(net, &udp6_seq_afinfo);
1461 void udp6_proc_exit(struct net *net) {
1462 udp_proc_unregister(net, &udp6_seq_afinfo);
1464 #endif /* CONFIG_PROC_FS */
1466 void udp_v6_clear_sk(struct sock *sk, int size)
1468 struct inet_sock *inet = inet_sk(sk);
1470 /* we do not want to clear pinet6 field, because of RCU lookups */
1471 sk_prot_clear_portaddr_nulls(sk, offsetof(struct inet_sock, pinet6));
1473 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1474 memset(&inet->pinet6 + 1, 0, size);
1477 /* ------------------------------------------------------------------------ */
1479 struct proto udpv6_prot = {
1481 .owner = THIS_MODULE,
1482 .close = udp_lib_close,
1483 .connect = ip6_datagram_connect,
1484 .disconnect = udp_disconnect,
1486 .destroy = udpv6_destroy_sock,
1487 .setsockopt = udpv6_setsockopt,
1488 .getsockopt = udpv6_getsockopt,
1489 .sendmsg = udpv6_sendmsg,
1490 .recvmsg = udpv6_recvmsg,
1491 .backlog_rcv = __udpv6_queue_rcv_skb,
1492 .hash = udp_lib_hash,
1493 .unhash = udp_lib_unhash,
1494 .rehash = udp_v6_rehash,
1495 .get_port = udp_v6_get_port,
1496 .memory_allocated = &udp_memory_allocated,
1497 .sysctl_mem = sysctl_udp_mem,
1498 .sysctl_wmem = &sysctl_udp_wmem_min,
1499 .sysctl_rmem = &sysctl_udp_rmem_min,
1500 .obj_size = sizeof(struct udp6_sock),
1501 .slab_flags = SLAB_DESTROY_BY_RCU,
1502 .h.udp_table = &udp_table,
1503 #ifdef CONFIG_COMPAT
1504 .compat_setsockopt = compat_udpv6_setsockopt,
1505 .compat_getsockopt = compat_udpv6_getsockopt,
1507 .clear_sk = udp_v6_clear_sk,
1510 static struct inet_protosw udpv6_protosw = {
1512 .protocol = IPPROTO_UDP,
1513 .prot = &udpv6_prot,
1514 .ops = &inet6_dgram_ops,
1515 .flags = INET_PROTOSW_PERMANENT,
1519 int __init udpv6_init(void)
1523 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1527 ret = inet6_register_protosw(&udpv6_protosw);
1529 goto out_udpv6_protocol;
1534 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1538 void udpv6_exit(void)
1540 inet6_unregister_protosw(&udpv6_protosw);
1541 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);