3 * Linux INET6 implementation
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,
152 struct inet_sock *inet;
154 if (!net_eq(sock_net(sk), net) ||
155 udp_sk(sk)->udp_port_hash != hnum ||
156 sk->sk_family != PF_INET6)
162 if (inet->inet_dport) {
163 if (inet->inet_dport != sport)
168 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
169 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
174 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
175 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
180 if (sk->sk_bound_dev_if) {
181 if (sk->sk_bound_dev_if != dif)
189 #define SCORE2_MAX (1 + 1 + 1)
190 static inline int compute_score2(struct sock *sk, struct net *net,
191 const struct in6_addr *saddr, __be16 sport,
192 const struct in6_addr *daddr,
193 unsigned short hnum, int dif)
196 struct inet_sock *inet;
198 if (!net_eq(sock_net(sk), net) ||
199 udp_sk(sk)->udp_port_hash != hnum ||
200 sk->sk_family != PF_INET6)
203 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
209 if (inet->inet_dport) {
210 if (inet->inet_dport != sport)
215 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
216 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
221 if (sk->sk_bound_dev_if) {
222 if (sk->sk_bound_dev_if != dif)
230 /* called with read_rcu_lock() */
231 static struct sock *udp6_lib_lookup2(struct net *net,
232 const struct in6_addr *saddr, __be16 sport,
233 const struct in6_addr *daddr, unsigned int hnum, int dif,
234 struct udp_hslot *hslot2, unsigned int slot2)
236 struct sock *sk, *result;
237 struct hlist_nulls_node *node;
238 int score, badness, matches = 0, reuseport = 0;
244 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
245 score = compute_score2(sk, net, saddr, sport,
247 if (score > badness) {
250 reuseport = sk->sk_reuseport;
252 hash = udp6_ehashfn(net, daddr, hnum,
255 } else if (score == SCORE2_MAX)
257 } else if (score == badness && reuseport) {
259 if (reciprocal_scale(hash, matches) == 0)
261 hash = next_pseudo_random32(hash);
265 * if the nulls value we got at the end of this lookup is
266 * not the expected one, we must restart lookup.
267 * We probably met an item that was moved to another chain.
269 if (get_nulls_value(node) != slot2)
274 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
276 else if (unlikely(compute_score2(result, net, saddr, sport,
277 daddr, hnum, dif) < badness)) {
285 struct sock *__udp6_lib_lookup(struct net *net,
286 const struct in6_addr *saddr, __be16 sport,
287 const struct in6_addr *daddr, __be16 dport,
288 int dif, struct udp_table *udptable)
290 struct sock *sk, *result;
291 struct hlist_nulls_node *node;
292 unsigned short hnum = ntohs(dport);
293 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
294 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
295 int score, badness, matches = 0, reuseport = 0;
299 if (hslot->count > 10) {
300 hash2 = udp6_portaddr_hash(net, daddr, hnum);
301 slot2 = hash2 & udptable->mask;
302 hslot2 = &udptable->hash2[slot2];
303 if (hslot->count < hslot2->count)
306 result = udp6_lib_lookup2(net, saddr, sport,
310 hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
311 slot2 = hash2 & udptable->mask;
312 hslot2 = &udptable->hash2[slot2];
313 if (hslot->count < hslot2->count)
316 result = udp6_lib_lookup2(net, saddr, sport,
317 &in6addr_any, hnum, dif,
326 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
327 score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif);
328 if (score > badness) {
331 reuseport = sk->sk_reuseport;
333 hash = udp6_ehashfn(net, daddr, hnum,
337 } else if (score == badness && reuseport) {
339 if (reciprocal_scale(hash, matches) == 0)
341 hash = next_pseudo_random32(hash);
345 * if the nulls value we got at the end of this lookup is
346 * not the expected one, we must restart lookup.
347 * We probably met an item that was moved to another chain.
349 if (get_nulls_value(node) != slot)
353 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
355 else if (unlikely(compute_score(result, net, hnum, saddr, sport,
356 daddr, dport, dif) < badness)) {
364 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
366 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
367 __be16 sport, __be16 dport,
368 struct udp_table *udptable)
371 const struct ipv6hdr *iph = ipv6_hdr(skb);
373 sk = skb_steal_sock(skb);
376 return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport,
377 &iph->daddr, dport, inet6_iif(skb),
381 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
382 const struct in6_addr *daddr, __be16 dport, int dif)
384 return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
386 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
390 * This should be easy, if there is something there we
391 * return it, otherwise we block.
394 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
395 int noblock, int flags, int *addr_len)
397 struct ipv6_pinfo *np = inet6_sk(sk);
398 struct inet_sock *inet = inet_sk(sk);
400 unsigned int ulen, copied;
403 int is_udplite = IS_UDPLITE(sk);
407 if (flags & MSG_ERRQUEUE)
408 return ipv6_recv_error(sk, msg, len, addr_len);
410 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
411 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
414 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
415 &peeked, &off, &err);
419 ulen = skb->len - sizeof(struct udphdr);
423 else if (copied < ulen)
424 msg->msg_flags |= MSG_TRUNC;
426 is_udp4 = (skb->protocol == htons(ETH_P_IP));
429 * If checksum is needed at all, try to do it while copying the
430 * data. If the data is truncated, or if we only want a partial
431 * coverage checksum (UDP-Lite), do it before the copy.
434 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
435 if (udp_lib_checksum_complete(skb))
439 if (skb_csum_unnecessary(skb))
440 err = skb_copy_datagram_msg(skb, sizeof(struct udphdr),
443 err = skb_copy_and_csum_datagram_msg(skb, sizeof(struct udphdr), msg);
448 trace_kfree_skb(skb, udpv6_recvmsg);
450 atomic_inc(&sk->sk_drops);
452 UDP_INC_STATS_USER(sock_net(sk),
456 UDP6_INC_STATS_USER(sock_net(sk),
464 UDP_INC_STATS_USER(sock_net(sk),
465 UDP_MIB_INDATAGRAMS, is_udplite);
467 UDP6_INC_STATS_USER(sock_net(sk),
468 UDP_MIB_INDATAGRAMS, is_udplite);
471 sock_recv_ts_and_drops(msg, sk, skb);
473 /* Copy the address. */
475 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
476 sin6->sin6_family = AF_INET6;
477 sin6->sin6_port = udp_hdr(skb)->source;
478 sin6->sin6_flowinfo = 0;
481 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
483 sin6->sin6_scope_id = 0;
485 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
486 sin6->sin6_scope_id =
487 ipv6_iface_scope_id(&sin6->sin6_addr,
490 *addr_len = sizeof(*sin6);
494 ip6_datagram_recv_common_ctl(sk, msg, skb);
497 if (inet->cmsg_flags)
498 ip_cmsg_recv(msg, skb);
501 ip6_datagram_recv_specific_ctl(sk, msg, skb);
505 if (flags & MSG_TRUNC)
509 skb_free_datagram_locked(sk, skb);
514 slow = lock_sock_fast(sk);
515 if (!skb_kill_datagram(sk, skb, flags)) {
517 UDP_INC_STATS_USER(sock_net(sk),
518 UDP_MIB_CSUMERRORS, is_udplite);
519 UDP_INC_STATS_USER(sock_net(sk),
520 UDP_MIB_INERRORS, is_udplite);
522 UDP6_INC_STATS_USER(sock_net(sk),
523 UDP_MIB_CSUMERRORS, is_udplite);
524 UDP6_INC_STATS_USER(sock_net(sk),
525 UDP_MIB_INERRORS, is_udplite);
528 unlock_sock_fast(sk, slow);
533 /* starting over for a new packet */
534 msg->msg_flags &= ~MSG_TRUNC;
538 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
539 u8 type, u8 code, int offset, __be32 info,
540 struct udp_table *udptable)
542 struct ipv6_pinfo *np;
543 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
544 const struct in6_addr *saddr = &hdr->saddr;
545 const struct in6_addr *daddr = &hdr->daddr;
546 struct udphdr *uh = (struct udphdr *)(skb->data+offset);
549 struct net *net = dev_net(skb->dev);
551 sk = __udp6_lib_lookup(net, daddr, uh->dest,
552 saddr, uh->source, inet6_iif(skb), udptable);
554 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
559 if (type == ICMPV6_PKT_TOOBIG) {
560 if (!ip6_sk_accept_pmtu(sk))
562 ip6_sk_update_pmtu(skb, sk, info);
564 if (type == NDISC_REDIRECT) {
565 ip6_sk_redirect(skb, sk);
571 if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
574 if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
578 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
581 sk->sk_error_report(sk);
586 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
590 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
591 sock_rps_save_rxhash(sk, skb);
592 sk_mark_napi_id(sk, skb);
593 sk_incoming_cpu_update(sk);
596 rc = sock_queue_rcv_skb(sk, skb);
598 int is_udplite = IS_UDPLITE(sk);
600 /* Note that an ENOMEM error is charged twice */
602 UDP6_INC_STATS_BH(sock_net(sk),
603 UDP_MIB_RCVBUFERRORS, is_udplite);
604 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
611 static __inline__ void udpv6_err(struct sk_buff *skb,
612 struct inet6_skb_parm *opt, u8 type,
613 u8 code, int offset, __be32 info)
615 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
618 static struct static_key udpv6_encap_needed __read_mostly;
619 void udpv6_encap_enable(void)
621 if (!static_key_enabled(&udpv6_encap_needed))
622 static_key_slow_inc(&udpv6_encap_needed);
624 EXPORT_SYMBOL(udpv6_encap_enable);
626 int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
628 struct udp_sock *up = udp_sk(sk);
630 int is_udplite = IS_UDPLITE(sk);
632 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
635 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
636 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
639 * This is an encapsulation socket so pass the skb to
640 * the socket's udp_encap_rcv() hook. Otherwise, just
641 * fall through and pass this up the UDP socket.
642 * up->encap_rcv() returns the following value:
643 * =0 if skb was successfully passed to the encap
644 * handler or was discarded by it.
645 * >0 if skb should be passed on to UDP.
646 * <0 if skb should be resubmitted as proto -N
649 /* if we're overly short, let UDP handle it */
650 encap_rcv = ACCESS_ONCE(up->encap_rcv);
651 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
654 /* Verify checksum before giving to encap */
655 if (udp_lib_checksum_complete(skb))
658 ret = encap_rcv(sk, skb);
660 UDP_INC_STATS_BH(sock_net(sk),
667 /* FALLTHROUGH -- it's a UDP Packet */
671 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
673 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
675 if (up->pcrlen == 0) { /* full coverage was set */
676 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
677 UDP_SKB_CB(skb)->cscov, skb->len);
680 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
681 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
682 UDP_SKB_CB(skb)->cscov, up->pcrlen);
687 if (rcu_access_pointer(sk->sk_filter)) {
688 if (udp_lib_checksum_complete(skb))
692 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
693 UDP6_INC_STATS_BH(sock_net(sk),
694 UDP_MIB_RCVBUFERRORS, is_udplite);
702 if (!sock_owned_by_user(sk))
703 rc = __udpv6_queue_rcv_skb(sk, skb);
704 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
713 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
715 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
716 atomic_inc(&sk->sk_drops);
721 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
722 __be16 loc_port, const struct in6_addr *loc_addr,
723 __be16 rmt_port, const struct in6_addr *rmt_addr,
724 int dif, unsigned short hnum)
726 struct inet_sock *inet = inet_sk(sk);
728 if (!net_eq(sock_net(sk), net))
731 if (udp_sk(sk)->udp_port_hash != hnum ||
732 sk->sk_family != PF_INET6 ||
733 (inet->inet_dport && inet->inet_dport != rmt_port) ||
734 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
735 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
736 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
738 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
743 static void flush_stack(struct sock **stack, unsigned int count,
744 struct sk_buff *skb, unsigned int final)
746 struct sk_buff *skb1 = NULL;
750 for (i = 0; i < count; i++) {
752 if (likely(skb1 == NULL))
753 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
755 atomic_inc(&sk->sk_drops);
756 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
758 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
762 if (skb1 && udpv6_queue_rcv_skb(sk, skb1) <= 0)
770 static void udp6_csum_zero_error(struct sk_buff *skb)
772 /* RFC 2460 section 8.1 says that we SHOULD log
773 * this error. Well, it is reasonable.
775 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
776 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
777 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
781 * Note: called only from the BH handler context,
782 * so we don't need to lock the hashes.
784 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
785 const struct in6_addr *saddr, const struct in6_addr *daddr,
786 struct udp_table *udptable, int proto)
788 struct sock *sk, *stack[256 / sizeof(struct sock *)];
789 const struct udphdr *uh = udp_hdr(skb);
790 struct hlist_nulls_node *node;
791 unsigned short hnum = ntohs(uh->dest);
792 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
793 int dif = inet6_iif(skb);
794 unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node);
795 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
796 bool inner_flushed = false;
799 hash2_any = udp6_portaddr_hash(net, &in6addr_any, hnum) &
801 hash2 = udp6_portaddr_hash(net, daddr, hnum) & udp_table.mask;
803 hslot = &udp_table.hash2[hash2];
804 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
807 spin_lock(&hslot->lock);
808 sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) {
809 if (__udp_v6_is_mcast_sock(net, sk,
813 /* If zero checksum and no_check is not on for
814 * the socket then skip it.
816 (uh->check || udp_sk(sk)->no_check6_rx)) {
817 if (unlikely(count == ARRAY_SIZE(stack))) {
818 flush_stack(stack, count, skb, ~0);
819 inner_flushed = true;
827 spin_unlock(&hslot->lock);
829 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
830 if (use_hash2 && hash2 != hash2_any) {
836 flush_stack(stack, count, skb, count - 1);
839 UDP_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI,
840 proto == IPPROTO_UDPLITE);
846 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
849 struct net *net = dev_net(skb->dev);
852 const struct in6_addr *saddr, *daddr;
855 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
858 saddr = &ipv6_hdr(skb)->saddr;
859 daddr = &ipv6_hdr(skb)->daddr;
862 ulen = ntohs(uh->len);
866 if (proto == IPPROTO_UDP) {
867 /* UDP validates ulen. */
869 /* Check for jumbo payload */
873 if (ulen < sizeof(*uh))
876 if (ulen < skb->len) {
877 if (pskb_trim_rcsum(skb, ulen))
879 saddr = &ipv6_hdr(skb)->saddr;
880 daddr = &ipv6_hdr(skb)->daddr;
885 if (udp6_csum_init(skb, uh, proto))
889 * Multicast receive code
891 if (ipv6_addr_is_multicast(daddr))
892 return __udp6_lib_mcast_deliver(net, skb,
893 saddr, daddr, udptable, proto);
898 * check socket cache ... must talk to Alan about his plans
899 * for sock caches... i'll skip this for now.
901 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
905 if (!uh->check && !udp_sk(sk)->no_check6_rx) {
907 udp6_csum_zero_error(skb);
911 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
912 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
915 ret = udpv6_queue_rcv_skb(sk, skb);
918 /* a return value > 0 means to resubmit the input, but
919 * it wants the return to be -protocol, or 0
928 udp6_csum_zero_error(skb);
932 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
935 if (udp_lib_checksum_complete(skb))
938 UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
939 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
945 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
946 proto == IPPROTO_UDPLITE ? "-Lite" : "",
947 saddr, ntohs(uh->source),
949 daddr, ntohs(uh->dest));
952 UDP6_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
954 UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
959 static __inline__ int udpv6_rcv(struct sk_buff *skb)
961 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
965 * Throw away all pending data and cancel the corking. Socket is locked.
967 static void udp_v6_flush_pending_frames(struct sock *sk)
969 struct udp_sock *up = udp_sk(sk);
971 if (up->pending == AF_INET)
972 udp_flush_pending_frames(sk);
973 else if (up->pending) {
976 ip6_flush_pending_frames(sk);
981 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
982 * @sk: socket we are sending on
983 * @skb: sk_buff containing the filled-in UDP header
984 * (checksum field must be zeroed out)
986 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
987 const struct in6_addr *saddr,
988 const struct in6_addr *daddr, int len)
991 struct udphdr *uh = udp_hdr(skb);
992 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
996 /* Only one fragment on the socket. */
997 skb->csum_start = skb_transport_header(skb) - skb->head;
998 skb->csum_offset = offsetof(struct udphdr, check);
999 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1002 * HW-checksum won't work as there are two or more
1003 * fragments on the socket so that all csums of sk_buffs
1004 * should be together
1006 offset = skb_transport_offset(skb);
1007 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1009 skb->ip_summed = CHECKSUM_NONE;
1012 csum = csum_add(csum, frags->csum);
1013 } while ((frags = frags->next));
1015 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1018 uh->check = CSUM_MANGLED_0;
1026 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6)
1028 struct sock *sk = skb->sk;
1031 int is_udplite = IS_UDPLITE(sk);
1033 int offset = skb_transport_offset(skb);
1034 int len = skb->len - offset;
1037 * Create a UDP header
1040 uh->source = fl6->fl6_sport;
1041 uh->dest = fl6->fl6_dport;
1042 uh->len = htons(len);
1046 csum = udplite_csum(skb);
1047 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */
1048 skb->ip_summed = CHECKSUM_NONE;
1050 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1051 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1054 csum = udp_csum(skb);
1056 /* add protocol-dependent pseudo-header */
1057 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1058 len, fl6->flowi6_proto, csum);
1060 uh->check = CSUM_MANGLED_0;
1063 err = ip6_send_skb(skb);
1065 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1066 UDP6_INC_STATS_USER(sock_net(sk),
1067 UDP_MIB_SNDBUFERRORS, is_udplite);
1071 UDP6_INC_STATS_USER(sock_net(sk),
1072 UDP_MIB_OUTDATAGRAMS, is_udplite);
1076 static int udp_v6_push_pending_frames(struct sock *sk)
1078 struct sk_buff *skb;
1079 struct udp_sock *up = udp_sk(sk);
1083 if (up->pending == AF_INET)
1084 return udp_push_pending_frames(sk);
1086 /* ip6_finish_skb will release the cork, so make a copy of
1089 fl6 = inet_sk(sk)->cork.fl.u.ip6;
1091 skb = ip6_finish_skb(sk);
1095 err = udp_v6_send_skb(skb, &fl6);
1103 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1105 struct ipv6_txoptions opt_space;
1106 struct udp_sock *up = udp_sk(sk);
1107 struct inet_sock *inet = inet_sk(sk);
1108 struct ipv6_pinfo *np = inet6_sk(sk);
1109 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1110 struct in6_addr *daddr, *final_p, final;
1111 struct ipv6_txoptions *opt = NULL;
1112 struct ip6_flowlabel *flowlabel = NULL;
1114 struct dst_entry *dst;
1115 int addr_len = msg->msg_namelen;
1120 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1123 int is_udplite = IS_UDPLITE(sk);
1124 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1126 /* destination address check */
1128 if (addr_len < offsetof(struct sockaddr, sa_data))
1131 switch (sin6->sin6_family) {
1133 if (addr_len < SIN6_LEN_RFC2133)
1135 daddr = &sin6->sin6_addr;
1138 goto do_udp_sendmsg;
1140 msg->msg_name = sin6 = NULL;
1141 msg->msg_namelen = addr_len = 0;
1147 } else if (!up->pending) {
1148 if (sk->sk_state != TCP_ESTABLISHED)
1149 return -EDESTADDRREQ;
1150 daddr = &sk->sk_v6_daddr;
1155 if (ipv6_addr_v4mapped(daddr)) {
1156 struct sockaddr_in sin;
1157 sin.sin_family = AF_INET;
1158 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1159 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1160 msg->msg_name = &sin;
1161 msg->msg_namelen = sizeof(sin);
1163 if (__ipv6_only_sock(sk))
1164 return -ENETUNREACH;
1165 return udp_sendmsg(sk, msg, len);
1169 if (up->pending == AF_INET)
1170 return udp_sendmsg(sk, msg, len);
1172 /* Rough check on arithmetic overflow,
1173 better check is made in ip6_append_data().
1175 if (len > INT_MAX - sizeof(struct udphdr))
1178 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1181 * There are pending frames.
1182 * The socket lock must be held while it's corked.
1185 if (likely(up->pending)) {
1186 if (unlikely(up->pending != AF_INET6)) {
1188 return -EAFNOSUPPORT;
1191 goto do_append_data;
1195 ulen += sizeof(struct udphdr);
1197 memset(&fl6, 0, sizeof(fl6));
1200 if (sin6->sin6_port == 0)
1203 fl6.fl6_dport = sin6->sin6_port;
1204 daddr = &sin6->sin6_addr;
1207 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1208 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1209 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1210 if (flowlabel == NULL)
1216 * Otherwise it will be difficult to maintain
1219 if (sk->sk_state == TCP_ESTABLISHED &&
1220 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1221 daddr = &sk->sk_v6_daddr;
1223 if (addr_len >= sizeof(struct sockaddr_in6) &&
1224 sin6->sin6_scope_id &&
1225 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1226 fl6.flowi6_oif = sin6->sin6_scope_id;
1228 if (sk->sk_state != TCP_ESTABLISHED)
1229 return -EDESTADDRREQ;
1231 fl6.fl6_dport = inet->inet_dport;
1232 daddr = &sk->sk_v6_daddr;
1233 fl6.flowlabel = np->flow_label;
1237 if (!fl6.flowi6_oif)
1238 fl6.flowi6_oif = sk->sk_bound_dev_if;
1240 if (!fl6.flowi6_oif)
1241 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1243 fl6.flowi6_mark = sk->sk_mark;
1245 if (msg->msg_controllen) {
1247 memset(opt, 0, sizeof(struct ipv6_txoptions));
1248 opt->tot_len = sizeof(*opt);
1250 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt,
1251 &hlimit, &tclass, &dontfrag);
1253 fl6_sock_release(flowlabel);
1256 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1257 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1258 if (flowlabel == NULL)
1261 if (!(opt->opt_nflen|opt->opt_flen))
1268 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1269 opt = ipv6_fixup_options(&opt_space, opt);
1271 fl6.flowi6_proto = sk->sk_protocol;
1272 if (!ipv6_addr_any(daddr))
1275 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1276 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1277 fl6.saddr = np->saddr;
1278 fl6.fl6_sport = inet->inet_sport;
1280 final_p = fl6_update_dst(&fl6, opt, &final);
1284 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1285 fl6.flowi6_oif = np->mcast_oif;
1287 } else if (!fl6.flowi6_oif)
1288 fl6.flowi6_oif = np->ucast_oif;
1290 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1292 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p);
1300 hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1303 tclass = np->tclass;
1305 if (msg->msg_flags&MSG_CONFIRM)
1309 /* Lockless fast path for the non-corking case */
1311 struct sk_buff *skb;
1313 skb = ip6_make_skb(sk, getfrag, msg, ulen,
1314 sizeof(struct udphdr), hlimit, tclass, opt,
1315 &fl6, (struct rt6_info *)dst,
1316 msg->msg_flags, dontfrag);
1318 if (!IS_ERR_OR_NULL(skb))
1319 err = udp_v6_send_skb(skb, &fl6);
1324 if (unlikely(up->pending)) {
1325 /* The socket is already corked while preparing it. */
1326 /* ... which is an evident application bug. --ANK */
1329 net_dbg_ratelimited("udp cork app bug 2\n");
1334 up->pending = AF_INET6;
1338 dontfrag = np->dontfrag;
1340 err = ip6_append_data(sk, getfrag, msg, ulen,
1341 sizeof(struct udphdr), hlimit, tclass, opt, &fl6,
1342 (struct rt6_info *)dst,
1343 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag);
1345 udp_v6_flush_pending_frames(sk);
1347 err = udp_v6_push_pending_frames(sk);
1348 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1352 err = np->recverr ? net_xmit_errno(err) : 0;
1358 ip6_dst_store(sk, dst,
1359 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ?
1360 &sk->sk_v6_daddr : NULL,
1361 #ifdef CONFIG_IPV6_SUBTREES
1362 ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1374 fl6_sock_release(flowlabel);
1378 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1379 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1380 * we don't have a good statistic (IpOutDiscards but it can be too many
1381 * things). We could add another new stat but at least for now that
1382 * seems like overkill.
1384 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1385 UDP6_INC_STATS_USER(sock_net(sk),
1386 UDP_MIB_SNDBUFERRORS, is_udplite);
1392 if (!(msg->msg_flags&MSG_PROBE) || len)
1393 goto back_from_confirm;
1398 void udpv6_destroy_sock(struct sock *sk)
1400 struct udp_sock *up = udp_sk(sk);
1402 udp_v6_flush_pending_frames(sk);
1405 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
1406 void (*encap_destroy)(struct sock *sk);
1407 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1412 inet6_destroy_sock(sk);
1416 * Socket option code for UDP
1418 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1419 char __user *optval, unsigned int optlen)
1421 if (level == SOL_UDP || level == SOL_UDPLITE)
1422 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1423 udp_v6_push_pending_frames);
1424 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1427 #ifdef CONFIG_COMPAT
1428 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1429 char __user *optval, unsigned int optlen)
1431 if (level == SOL_UDP || level == SOL_UDPLITE)
1432 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1433 udp_v6_push_pending_frames);
1434 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1438 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1439 char __user *optval, int __user *optlen)
1441 if (level == SOL_UDP || level == SOL_UDPLITE)
1442 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1443 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1446 #ifdef CONFIG_COMPAT
1447 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1448 char __user *optval, int __user *optlen)
1450 if (level == SOL_UDP || level == SOL_UDPLITE)
1451 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1452 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1456 static const struct inet6_protocol udpv6_protocol = {
1457 .handler = udpv6_rcv,
1458 .err_handler = udpv6_err,
1459 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1462 /* ------------------------------------------------------------------------ */
1463 #ifdef CONFIG_PROC_FS
1464 int udp6_seq_show(struct seq_file *seq, void *v)
1466 if (v == SEQ_START_TOKEN) {
1467 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1469 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1470 struct inet_sock *inet = inet_sk(v);
1471 __u16 srcp = ntohs(inet->inet_sport);
1472 __u16 destp = ntohs(inet->inet_dport);
1473 ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket);
1478 static const struct file_operations udp6_afinfo_seq_fops = {
1479 .owner = THIS_MODULE,
1480 .open = udp_seq_open,
1482 .llseek = seq_lseek,
1483 .release = seq_release_net
1486 static struct udp_seq_afinfo udp6_seq_afinfo = {
1489 .udp_table = &udp_table,
1490 .seq_fops = &udp6_afinfo_seq_fops,
1492 .show = udp6_seq_show,
1496 int __net_init udp6_proc_init(struct net *net)
1498 return udp_proc_register(net, &udp6_seq_afinfo);
1501 void udp6_proc_exit(struct net *net) {
1502 udp_proc_unregister(net, &udp6_seq_afinfo);
1504 #endif /* CONFIG_PROC_FS */
1506 void udp_v6_clear_sk(struct sock *sk, int size)
1508 struct inet_sock *inet = inet_sk(sk);
1510 /* we do not want to clear pinet6 field, because of RCU lookups */
1511 sk_prot_clear_portaddr_nulls(sk, offsetof(struct inet_sock, pinet6));
1513 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1514 memset(&inet->pinet6 + 1, 0, size);
1517 /* ------------------------------------------------------------------------ */
1519 struct proto udpv6_prot = {
1521 .owner = THIS_MODULE,
1522 .close = udp_lib_close,
1523 .connect = ip6_datagram_connect,
1524 .disconnect = udp_disconnect,
1526 .destroy = udpv6_destroy_sock,
1527 .setsockopt = udpv6_setsockopt,
1528 .getsockopt = udpv6_getsockopt,
1529 .sendmsg = udpv6_sendmsg,
1530 .recvmsg = udpv6_recvmsg,
1531 .backlog_rcv = __udpv6_queue_rcv_skb,
1532 .hash = udp_lib_hash,
1533 .unhash = udp_lib_unhash,
1534 .rehash = udp_v6_rehash,
1535 .get_port = udp_v6_get_port,
1536 .memory_allocated = &udp_memory_allocated,
1537 .sysctl_mem = sysctl_udp_mem,
1538 .sysctl_wmem = &sysctl_udp_wmem_min,
1539 .sysctl_rmem = &sysctl_udp_rmem_min,
1540 .obj_size = sizeof(struct udp6_sock),
1541 .slab_flags = SLAB_DESTROY_BY_RCU,
1542 .h.udp_table = &udp_table,
1543 #ifdef CONFIG_COMPAT
1544 .compat_setsockopt = compat_udpv6_setsockopt,
1545 .compat_getsockopt = compat_udpv6_getsockopt,
1547 .clear_sk = udp_v6_clear_sk,
1550 static struct inet_protosw udpv6_protosw = {
1552 .protocol = IPPROTO_UDP,
1553 .prot = &udpv6_prot,
1554 .ops = &inet6_dgram_ops,
1555 .flags = INET_PROTOSW_PERMANENT,
1559 int __init udpv6_init(void)
1563 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1567 ret = inet6_register_protosw(&udpv6_protosw);
1569 goto out_udpv6_protocol;
1574 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1578 void udpv6_exit(void)
1580 inet6_unregister_protosw(&udpv6_protosw);
1581 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);