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 <asm/uaccess.h>
39 #include <net/ndisc.h>
40 #include <net/protocol.h>
41 #include <net/transp_v6.h>
42 #include <net/ip6_route.h>
44 #include <net/tcp_states.h>
45 #include <net/ip6_checksum.h>
48 #include <linux/proc_fs.h>
49 #include <linux/seq_file.h>
52 int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2)
54 const struct in6_addr *sk_rcv_saddr6 = &inet6_sk(sk)->rcv_saddr;
55 const struct in6_addr *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2);
56 __be32 sk1_rcv_saddr = inet_sk(sk)->inet_rcv_saddr;
57 __be32 sk2_rcv_saddr = inet_rcv_saddr(sk2);
58 int sk_ipv6only = ipv6_only_sock(sk);
59 int sk2_ipv6only = inet_v6_ipv6only(sk2);
60 int addr_type = ipv6_addr_type(sk_rcv_saddr6);
61 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
63 /* if both are mapped, treat as IPv4 */
64 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED)
65 return (!sk2_ipv6only &&
66 (!sk1_rcv_saddr || !sk2_rcv_saddr ||
67 sk1_rcv_saddr == sk2_rcv_saddr));
69 if (addr_type2 == IPV6_ADDR_ANY &&
70 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
73 if (addr_type == IPV6_ADDR_ANY &&
74 !(sk_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
78 ipv6_addr_equal(sk_rcv_saddr6, sk2_rcv_saddr6))
84 static unsigned int udp6_portaddr_hash(struct net *net,
85 const struct in6_addr *addr6,
88 unsigned int hash, mix = net_hash_mix(net);
90 if (ipv6_addr_any(addr6))
91 hash = jhash_1word(0, mix);
92 else if (ipv6_addr_type(addr6) == IPV6_ADDR_MAPPED)
93 hash = jhash_1word(addr6->s6_addr32[3], mix);
95 hash = jhash2(addr6->s6_addr32, 4, mix);
101 int udp_v6_get_port(struct sock *sk, unsigned short snum)
103 /* precompute partial secondary hash */
104 udp_sk(sk)->udp_portaddr_hash =
105 udp6_portaddr_hash(sock_net(sk),
106 &inet6_sk(sk)->rcv_saddr,
108 return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal);
111 static inline int compute_score(struct sock *sk, struct net *net,
113 struct in6_addr *saddr, __be16 sport,
114 struct in6_addr *daddr, __be16 dport,
119 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
120 sk->sk_family == PF_INET6) {
121 struct ipv6_pinfo *np = inet6_sk(sk);
122 struct inet_sock *inet = inet_sk(sk);
125 if (inet->inet_dport) {
126 if (inet->inet_dport != sport)
130 if (!ipv6_addr_any(&np->rcv_saddr)) {
131 if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
135 if (!ipv6_addr_any(&np->daddr)) {
136 if (!ipv6_addr_equal(&np->daddr, saddr))
140 if (sk->sk_bound_dev_if) {
141 if (sk->sk_bound_dev_if != dif)
149 #define SCORE2_MAX (1 + 1 + 1)
150 static inline int compute_score2(struct sock *sk, struct net *net,
151 const struct in6_addr *saddr, __be16 sport,
152 const struct in6_addr *daddr, unsigned short hnum,
157 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
158 sk->sk_family == PF_INET6) {
159 struct ipv6_pinfo *np = inet6_sk(sk);
160 struct inet_sock *inet = inet_sk(sk);
162 if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
165 if (inet->inet_dport) {
166 if (inet->inet_dport != sport)
170 if (!ipv6_addr_any(&np->daddr)) {
171 if (!ipv6_addr_equal(&np->daddr, saddr))
175 if (sk->sk_bound_dev_if) {
176 if (sk->sk_bound_dev_if != dif)
184 #define udp_portaddr_for_each_entry_rcu(__sk, node, list) \
185 hlist_nulls_for_each_entry_rcu(__sk, node, list, __sk_common.skc_portaddr_node)
187 /* called with read_rcu_lock() */
188 static struct sock *udp6_lib_lookup2(struct net *net,
189 const struct in6_addr *saddr, __be16 sport,
190 const struct in6_addr *daddr, unsigned int hnum, int dif,
191 struct udp_hslot *hslot2, unsigned int slot2)
193 struct sock *sk, *result;
194 struct hlist_nulls_node *node;
200 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
201 score = compute_score2(sk, net, saddr, sport,
203 if (score > badness) {
206 if (score == SCORE2_MAX)
211 * if the nulls value we got at the end of this lookup is
212 * not the expected one, we must restart lookup.
213 * We probably met an item that was moved to another chain.
215 if (get_nulls_value(node) != slot2)
220 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
222 else if (unlikely(compute_score2(result, net, saddr, sport,
223 daddr, hnum, dif) < badness)) {
231 static struct sock *__udp6_lib_lookup(struct net *net,
232 struct in6_addr *saddr, __be16 sport,
233 struct in6_addr *daddr, __be16 dport,
234 int dif, struct udp_table *udptable)
236 struct sock *sk, *result;
237 struct hlist_nulls_node *node;
238 unsigned short hnum = ntohs(dport);
239 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
240 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
244 if (hslot->count > 10) {
245 hash2 = udp6_portaddr_hash(net, daddr, hnum);
246 slot2 = hash2 & udptable->mask;
247 hslot2 = &udptable->hash2[slot2];
248 if (hslot->count < hslot2->count)
251 result = udp6_lib_lookup2(net, saddr, sport,
255 hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
256 slot2 = hash2 & udptable->mask;
257 hslot2 = &udptable->hash2[slot2];
258 if (hslot->count < hslot2->count)
261 result = udp6_lib_lookup2(net, &in6addr_any, sport,
271 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
272 score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif);
273 if (score > badness) {
279 * if the nulls value we got at the end of this lookup is
280 * not the expected one, we must restart lookup.
281 * We probably met an item that was moved to another chain.
283 if (get_nulls_value(node) != slot)
287 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
289 else if (unlikely(compute_score(result, net, hnum, saddr, sport,
290 daddr, dport, dif) < badness)) {
299 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
300 __be16 sport, __be16 dport,
301 struct udp_table *udptable)
304 struct ipv6hdr *iph = ipv6_hdr(skb);
306 if (unlikely(sk = skb_steal_sock(skb)))
308 return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport,
309 &iph->daddr, dport, inet6_iif(skb),
314 * This should be easy, if there is something there we
315 * return it, otherwise we block.
318 int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk,
319 struct msghdr *msg, size_t len,
320 int noblock, int flags, int *addr_len)
322 struct ipv6_pinfo *np = inet6_sk(sk);
323 struct inet_sock *inet = inet_sk(sk);
325 unsigned int ulen, copied;
328 int is_udplite = IS_UDPLITE(sk);
332 *addr_len=sizeof(struct sockaddr_in6);
334 if (flags & MSG_ERRQUEUE)
335 return ipv6_recv_error(sk, msg, len);
338 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
343 ulen = skb->len - sizeof(struct udphdr);
347 else if (copied < ulen)
348 msg->msg_flags |= MSG_TRUNC;
350 is_udp4 = (skb->protocol == htons(ETH_P_IP));
353 * If checksum is needed at all, try to do it while copying the
354 * data. If the data is truncated, or if we only want a partial
355 * coverage checksum (UDP-Lite), do it before the copy.
358 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
359 if (udp_lib_checksum_complete(skb))
363 if (skb_csum_unnecessary(skb))
364 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
365 msg->msg_iov, copied );
367 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
376 UDP_INC_STATS_USER(sock_net(sk),
377 UDP_MIB_INDATAGRAMS, is_udplite);
379 UDP6_INC_STATS_USER(sock_net(sk),
380 UDP_MIB_INDATAGRAMS, is_udplite);
383 sock_recv_ts_and_drops(msg, sk, skb);
385 /* Copy the address. */
387 struct sockaddr_in6 *sin6;
389 sin6 = (struct sockaddr_in6 *) msg->msg_name;
390 sin6->sin6_family = AF_INET6;
391 sin6->sin6_port = udp_hdr(skb)->source;
392 sin6->sin6_flowinfo = 0;
393 sin6->sin6_scope_id = 0;
396 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
399 ipv6_addr_copy(&sin6->sin6_addr,
400 &ipv6_hdr(skb)->saddr);
401 if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
402 sin6->sin6_scope_id = IP6CB(skb)->iif;
407 if (inet->cmsg_flags)
408 ip_cmsg_recv(msg, skb);
411 datagram_recv_ctl(sk, msg, skb);
415 if (flags & MSG_TRUNC)
419 skb_free_datagram_locked(sk, skb);
425 if (!skb_kill_datagram(sk, skb, flags)) {
427 UDP_INC_STATS_USER(sock_net(sk),
428 UDP_MIB_INERRORS, is_udplite);
430 UDP6_INC_STATS_USER(sock_net(sk),
431 UDP_MIB_INERRORS, is_udplite);
435 if (flags & MSG_DONTWAIT)
440 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
441 u8 type, u8 code, int offset, __be32 info,
442 struct udp_table *udptable)
444 struct ipv6_pinfo *np;
445 struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
446 struct in6_addr *saddr = &hdr->saddr;
447 struct in6_addr *daddr = &hdr->daddr;
448 struct udphdr *uh = (struct udphdr*)(skb->data+offset);
452 sk = __udp6_lib_lookup(dev_net(skb->dev), daddr, uh->dest,
453 saddr, uh->source, inet6_iif(skb), udptable);
459 if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
462 if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
466 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
469 sk->sk_error_report(sk);
474 static __inline__ void udpv6_err(struct sk_buff *skb,
475 struct inet6_skb_parm *opt, u8 type,
476 u8 code, int offset, __be32 info )
478 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
481 int udpv6_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
483 struct udp_sock *up = udp_sk(sk);
485 int is_udplite = IS_UDPLITE(sk);
487 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
491 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
493 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
495 if (up->pcrlen == 0) { /* full coverage was set */
496 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage"
497 " %d while full coverage %d requested\n",
498 UDP_SKB_CB(skb)->cscov, skb->len);
501 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
502 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d "
503 "too small, need min %d\n",
504 UDP_SKB_CB(skb)->cscov, up->pcrlen);
510 if (udp_lib_checksum_complete(skb))
514 if ((rc = sock_queue_rcv_skb(sk, skb)) < 0) {
515 /* Note that an ENOMEM error is charged twice */
517 UDP6_INC_STATS_BH(sock_net(sk),
518 UDP_MIB_RCVBUFERRORS, is_udplite);
519 goto drop_no_sk_drops_inc;
524 atomic_inc(&sk->sk_drops);
525 drop_no_sk_drops_inc:
526 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
531 static struct sock *udp_v6_mcast_next(struct net *net, struct sock *sk,
532 __be16 loc_port, struct in6_addr *loc_addr,
533 __be16 rmt_port, struct in6_addr *rmt_addr,
536 struct hlist_nulls_node *node;
538 unsigned short num = ntohs(loc_port);
540 sk_nulls_for_each_from(s, node) {
541 struct inet_sock *inet = inet_sk(s);
543 if (!net_eq(sock_net(s), net))
546 if (udp_sk(s)->udp_port_hash == num &&
547 s->sk_family == PF_INET6) {
548 struct ipv6_pinfo *np = inet6_sk(s);
549 if (inet->inet_dport) {
550 if (inet->inet_dport != rmt_port)
553 if (!ipv6_addr_any(&np->daddr) &&
554 !ipv6_addr_equal(&np->daddr, rmt_addr))
557 if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)
560 if (!ipv6_addr_any(&np->rcv_saddr)) {
561 if (!ipv6_addr_equal(&np->rcv_saddr, loc_addr))
564 if (!inet6_mc_check(s, loc_addr, rmt_addr))
572 static void flush_stack(struct sock **stack, unsigned int count,
573 struct sk_buff *skb, unsigned int final)
577 struct sk_buff *skb1;
579 for (i = 0; i < count; i++) {
580 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
585 if (!sock_owned_by_user(sk))
586 udpv6_queue_rcv_skb(sk, skb1);
588 sk_add_backlog(sk, skb1);
594 * Note: called only from the BH handler context,
595 * so we don't need to lock the hashes.
597 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
598 struct in6_addr *saddr, struct in6_addr *daddr,
599 struct udp_table *udptable)
601 struct sock *sk, *stack[256 / sizeof(struct sock *)];
602 const struct udphdr *uh = udp_hdr(skb);
603 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
605 unsigned int i, count = 0;
607 spin_lock(&hslot->lock);
608 sk = sk_nulls_head(&hslot->head);
609 dif = inet6_iif(skb);
610 sk = udp_v6_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
613 sk = udp_v6_mcast_next(net, sk_nulls_next(sk), uh->dest, daddr,
614 uh->source, saddr, dif);
615 if (unlikely(count == ARRAY_SIZE(stack))) {
618 flush_stack(stack, count, skb, ~0);
623 * before releasing the lock, we must take reference on sockets
625 for (i = 0; i < count; i++)
628 spin_unlock(&hslot->lock);
631 flush_stack(stack, count, skb, count - 1);
633 for (i = 0; i < count; i++)
641 static inline int udp6_csum_init(struct sk_buff *skb, struct udphdr *uh,
646 UDP_SKB_CB(skb)->partial_cov = 0;
647 UDP_SKB_CB(skb)->cscov = skb->len;
649 if (proto == IPPROTO_UDPLITE) {
650 err = udplite_checksum_init(skb, uh);
655 if (uh->check == 0) {
656 /* RFC 2460 section 8.1 says that we SHOULD log
657 this error. Well, it is reasonable.
659 LIMIT_NETDEBUG(KERN_INFO "IPv6: udp checksum is 0\n");
662 if (skb->ip_summed == CHECKSUM_COMPLETE &&
663 !csum_ipv6_magic(&ipv6_hdr(skb)->saddr, &ipv6_hdr(skb)->daddr,
664 skb->len, proto, skb->csum))
665 skb->ip_summed = CHECKSUM_UNNECESSARY;
667 if (!skb_csum_unnecessary(skb))
668 skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
669 &ipv6_hdr(skb)->daddr,
670 skb->len, proto, 0));
675 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
680 struct net_device *dev = skb->dev;
681 struct in6_addr *saddr, *daddr;
683 struct net *net = dev_net(skb->dev);
685 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
688 saddr = &ipv6_hdr(skb)->saddr;
689 daddr = &ipv6_hdr(skb)->daddr;
692 ulen = ntohs(uh->len);
696 if (proto == IPPROTO_UDP) {
697 /* UDP validates ulen. */
699 /* Check for jumbo payload */
703 if (ulen < sizeof(*uh))
706 if (ulen < skb->len) {
707 if (pskb_trim_rcsum(skb, ulen))
709 saddr = &ipv6_hdr(skb)->saddr;
710 daddr = &ipv6_hdr(skb)->daddr;
715 if (udp6_csum_init(skb, uh, proto))
719 * Multicast receive code
721 if (ipv6_addr_is_multicast(daddr))
722 return __udp6_lib_mcast_deliver(net, skb,
723 saddr, daddr, udptable);
728 * check socket cache ... must talk to Alan about his plans
729 * for sock caches... i'll skip this for now.
731 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
734 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
737 if (udp_lib_checksum_complete(skb))
739 UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS,
740 proto == IPPROTO_UDPLITE);
742 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0, dev);
751 if (!sock_owned_by_user(sk))
752 udpv6_queue_rcv_skb(sk, skb);
754 sk_add_backlog(sk, skb);
760 LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: %d/%u\n",
761 proto == IPPROTO_UDPLITE ? "-Lite" : "",
765 UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
770 static __inline__ int udpv6_rcv(struct sk_buff *skb)
772 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
776 * Throw away all pending data and cancel the corking. Socket is locked.
778 static void udp_v6_flush_pending_frames(struct sock *sk)
780 struct udp_sock *up = udp_sk(sk);
782 if (up->pending == AF_INET)
783 udp_flush_pending_frames(sk);
784 else if (up->pending) {
787 ip6_flush_pending_frames(sk);
792 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
793 * @sk: socket we are sending on
794 * @skb: sk_buff containing the filled-in UDP header
795 * (checksum field must be zeroed out)
797 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
798 const struct in6_addr *saddr,
799 const struct in6_addr *daddr, int len)
802 struct udphdr *uh = udp_hdr(skb);
805 if (skb_queue_len(&sk->sk_write_queue) == 1) {
806 /* Only one fragment on the socket. */
807 skb->csum_start = skb_transport_header(skb) - skb->head;
808 skb->csum_offset = offsetof(struct udphdr, check);
809 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
812 * HW-checksum won't work as there are two or more
813 * fragments on the socket so that all csums of sk_buffs
816 offset = skb_transport_offset(skb);
817 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
819 skb->ip_summed = CHECKSUM_NONE;
821 skb_queue_walk(&sk->sk_write_queue, skb) {
822 csum = csum_add(csum, skb->csum);
825 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
828 uh->check = CSUM_MANGLED_0;
836 static int udp_v6_push_pending_frames(struct sock *sk)
840 struct udp_sock *up = udp_sk(sk);
841 struct inet_sock *inet = inet_sk(sk);
842 struct flowi *fl = &inet->cork.fl;
844 int is_udplite = IS_UDPLITE(sk);
847 /* Grab the skbuff where UDP header space exists. */
848 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
852 * Create a UDP header
855 uh->source = fl->fl_ip_sport;
856 uh->dest = fl->fl_ip_dport;
857 uh->len = htons(up->len);
861 csum = udplite_csum_outgoing(sk, skb);
862 else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
863 udp6_hwcsum_outgoing(sk, skb, &fl->fl6_src, &fl->fl6_dst,
867 csum = udp_csum_outgoing(sk, skb);
869 /* add protocol-dependent pseudo-header */
870 uh->check = csum_ipv6_magic(&fl->fl6_src, &fl->fl6_dst,
871 up->len, fl->proto, csum );
873 uh->check = CSUM_MANGLED_0;
876 err = ip6_push_pending_frames(sk);
878 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
879 UDP6_INC_STATS_USER(sock_net(sk),
880 UDP_MIB_SNDBUFERRORS, is_udplite);
884 UDP6_INC_STATS_USER(sock_net(sk),
885 UDP_MIB_OUTDATAGRAMS, is_udplite);
892 int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
893 struct msghdr *msg, size_t len)
895 struct ipv6_txoptions opt_space;
896 struct udp_sock *up = udp_sk(sk);
897 struct inet_sock *inet = inet_sk(sk);
898 struct ipv6_pinfo *np = inet6_sk(sk);
899 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) msg->msg_name;
900 struct in6_addr *daddr, *final_p = NULL, final;
901 struct ipv6_txoptions *opt = NULL;
902 struct ip6_flowlabel *flowlabel = NULL;
904 struct dst_entry *dst;
905 int addr_len = msg->msg_namelen;
909 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
912 int is_udplite = IS_UDPLITE(sk);
913 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
915 /* destination address check */
917 if (addr_len < offsetof(struct sockaddr, sa_data))
920 switch (sin6->sin6_family) {
922 if (addr_len < SIN6_LEN_RFC2133)
924 daddr = &sin6->sin6_addr;
929 msg->msg_name = sin6 = NULL;
930 msg->msg_namelen = addr_len = 0;
936 } else if (!up->pending) {
937 if (sk->sk_state != TCP_ESTABLISHED)
938 return -EDESTADDRREQ;
944 if (ipv6_addr_v4mapped(daddr)) {
945 struct sockaddr_in sin;
946 sin.sin_family = AF_INET;
947 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
948 sin.sin_addr.s_addr = daddr->s6_addr32[3];
949 msg->msg_name = &sin;
950 msg->msg_namelen = sizeof(sin);
952 if (__ipv6_only_sock(sk))
954 return udp_sendmsg(iocb, sk, msg, len);
958 if (up->pending == AF_INET)
959 return udp_sendmsg(iocb, sk, msg, len);
961 /* Rough check on arithmetic overflow,
962 better check is made in ip6_append_data().
964 if (len > INT_MAX - sizeof(struct udphdr))
969 * There are pending frames.
970 * The socket lock must be held while it's corked.
973 if (likely(up->pending)) {
974 if (unlikely(up->pending != AF_INET6)) {
976 return -EAFNOSUPPORT;
983 ulen += sizeof(struct udphdr);
985 memset(&fl, 0, sizeof(fl));
988 if (sin6->sin6_port == 0)
991 fl.fl_ip_dport = sin6->sin6_port;
992 daddr = &sin6->sin6_addr;
995 fl.fl6_flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
996 if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
997 flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
998 if (flowlabel == NULL)
1000 daddr = &flowlabel->dst;
1005 * Otherwise it will be difficult to maintain
1008 if (sk->sk_state == TCP_ESTABLISHED &&
1009 ipv6_addr_equal(daddr, &np->daddr))
1012 if (addr_len >= sizeof(struct sockaddr_in6) &&
1013 sin6->sin6_scope_id &&
1014 ipv6_addr_type(daddr)&IPV6_ADDR_LINKLOCAL)
1015 fl.oif = sin6->sin6_scope_id;
1017 if (sk->sk_state != TCP_ESTABLISHED)
1018 return -EDESTADDRREQ;
1020 fl.fl_ip_dport = inet->inet_dport;
1022 fl.fl6_flowlabel = np->flow_label;
1027 fl.oif = sk->sk_bound_dev_if;
1030 fl.oif = np->sticky_pktinfo.ipi6_ifindex;
1032 fl.mark = sk->sk_mark;
1034 if (msg->msg_controllen) {
1036 memset(opt, 0, sizeof(struct ipv6_txoptions));
1037 opt->tot_len = sizeof(*opt);
1039 err = datagram_send_ctl(sock_net(sk), msg, &fl, opt, &hlimit, &tclass);
1041 fl6_sock_release(flowlabel);
1044 if ((fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1045 flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
1046 if (flowlabel == NULL)
1049 if (!(opt->opt_nflen|opt->opt_flen))
1056 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1057 opt = ipv6_fixup_options(&opt_space, opt);
1059 fl.proto = sk->sk_protocol;
1060 if (!ipv6_addr_any(daddr))
1061 ipv6_addr_copy(&fl.fl6_dst, daddr);
1063 fl.fl6_dst.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1064 if (ipv6_addr_any(&fl.fl6_src) && !ipv6_addr_any(&np->saddr))
1065 ipv6_addr_copy(&fl.fl6_src, &np->saddr);
1066 fl.fl_ip_sport = inet->inet_sport;
1068 /* merge ip6_build_xmit from ip6_output */
1069 if (opt && opt->srcrt) {
1070 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
1071 ipv6_addr_copy(&final, &fl.fl6_dst);
1072 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1077 if (!fl.oif && ipv6_addr_is_multicast(&fl.fl6_dst)) {
1078 fl.oif = np->mcast_oif;
1082 security_sk_classify_flow(sk, &fl);
1084 err = ip6_sk_dst_lookup(sk, &dst, &fl);
1088 ipv6_addr_copy(&fl.fl6_dst, final_p);
1090 err = __xfrm_lookup(sock_net(sk), &dst, &fl, sk, XFRM_LOOKUP_WAIT);
1092 if (err == -EREMOTE)
1093 err = ip6_dst_blackhole(sk, &dst, &fl);
1099 if (ipv6_addr_is_multicast(&fl.fl6_dst))
1100 hlimit = np->mcast_hops;
1102 hlimit = np->hop_limit;
1104 hlimit = ip6_dst_hoplimit(dst);
1108 tclass = np->tclass;
1110 if (msg->msg_flags&MSG_CONFIRM)
1115 if (unlikely(up->pending)) {
1116 /* The socket is already corked while preparing it. */
1117 /* ... which is an evident application bug. --ANK */
1120 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
1125 up->pending = AF_INET6;
1129 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1130 err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen,
1131 sizeof(struct udphdr), hlimit, tclass, opt, &fl,
1132 (struct rt6_info*)dst,
1133 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1135 udp_v6_flush_pending_frames(sk);
1137 err = udp_v6_push_pending_frames(sk);
1138 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1143 ip6_dst_store(sk, dst,
1144 ipv6_addr_equal(&fl.fl6_dst, &np->daddr) ?
1146 #ifdef CONFIG_IPV6_SUBTREES
1147 ipv6_addr_equal(&fl.fl6_src, &np->saddr) ?
1158 err = np->recverr ? net_xmit_errno(err) : 0;
1162 fl6_sock_release(flowlabel);
1166 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1167 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1168 * we don't have a good statistic (IpOutDiscards but it can be too many
1169 * things). We could add another new stat but at least for now that
1170 * seems like overkill.
1172 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1173 UDP6_INC_STATS_USER(sock_net(sk),
1174 UDP_MIB_SNDBUFERRORS, is_udplite);
1180 if (!(msg->msg_flags&MSG_PROBE) || len)
1181 goto back_from_confirm;
1186 void udpv6_destroy_sock(struct sock *sk)
1189 udp_v6_flush_pending_frames(sk);
1192 inet6_destroy_sock(sk);
1196 * Socket option code for UDP
1198 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1199 char __user *optval, unsigned int optlen)
1201 if (level == SOL_UDP || level == SOL_UDPLITE)
1202 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1203 udp_v6_push_pending_frames);
1204 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1207 #ifdef CONFIG_COMPAT
1208 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1209 char __user *optval, unsigned int optlen)
1211 if (level == SOL_UDP || level == SOL_UDPLITE)
1212 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1213 udp_v6_push_pending_frames);
1214 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1218 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1219 char __user *optval, int __user *optlen)
1221 if (level == SOL_UDP || level == SOL_UDPLITE)
1222 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1223 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1226 #ifdef CONFIG_COMPAT
1227 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1228 char __user *optval, int __user *optlen)
1230 if (level == SOL_UDP || level == SOL_UDPLITE)
1231 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1232 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1236 static int udp6_ufo_send_check(struct sk_buff *skb)
1238 struct ipv6hdr *ipv6h;
1241 if (!pskb_may_pull(skb, sizeof(*uh)))
1244 ipv6h = ipv6_hdr(skb);
1247 uh->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, skb->len,
1249 skb->csum_start = skb_transport_header(skb) - skb->head;
1250 skb->csum_offset = offsetof(struct udphdr, check);
1251 skb->ip_summed = CHECKSUM_PARTIAL;
1255 static struct sk_buff *udp6_ufo_fragment(struct sk_buff *skb, int features)
1257 struct sk_buff *segs = ERR_PTR(-EINVAL);
1259 unsigned int unfrag_ip6hlen, unfrag_len;
1260 struct frag_hdr *fptr;
1261 u8 *mac_start, *prevhdr;
1263 u8 frag_hdr_sz = sizeof(struct frag_hdr);
1267 mss = skb_shinfo(skb)->gso_size;
1268 if (unlikely(skb->len <= mss))
1271 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
1272 /* Packet is from an untrusted source, reset gso_segs. */
1273 int type = skb_shinfo(skb)->gso_type;
1275 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
1276 !(type & (SKB_GSO_UDP))))
1279 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
1285 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
1286 * do checksum of UDP packets sent as multiple IP fragments.
1288 offset = skb->csum_start - skb_headroom(skb);
1289 csum = skb_checksum(skb, offset, skb->len- offset, 0);
1290 offset += skb->csum_offset;
1291 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1292 skb->ip_summed = CHECKSUM_NONE;
1294 /* Check if there is enough headroom to insert fragment header. */
1295 if ((skb_headroom(skb) < frag_hdr_sz) &&
1296 pskb_expand_head(skb, frag_hdr_sz, 0, GFP_ATOMIC))
1299 /* Find the unfragmentable header and shift it left by frag_hdr_sz
1300 * bytes to insert fragment header.
1302 unfrag_ip6hlen = ip6_find_1stfragopt(skb, &prevhdr);
1304 *prevhdr = NEXTHDR_FRAGMENT;
1305 unfrag_len = skb_network_header(skb) - skb_mac_header(skb) +
1307 mac_start = skb_mac_header(skb);
1308 memmove(mac_start-frag_hdr_sz, mac_start, unfrag_len);
1310 skb->mac_header -= frag_hdr_sz;
1311 skb->network_header -= frag_hdr_sz;
1313 fptr = (struct frag_hdr *)(skb_network_header(skb) + unfrag_ip6hlen);
1314 fptr->nexthdr = nexthdr;
1316 ipv6_select_ident(fptr);
1318 /* Fragment the skb. ipv6 header and the remaining fields of the
1319 * fragment header are updated in ipv6_gso_segment()
1321 segs = skb_segment(skb, features);
1327 static const struct inet6_protocol udpv6_protocol = {
1328 .handler = udpv6_rcv,
1329 .err_handler = udpv6_err,
1330 .gso_send_check = udp6_ufo_send_check,
1331 .gso_segment = udp6_ufo_fragment,
1332 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1335 /* ------------------------------------------------------------------------ */
1336 #ifdef CONFIG_PROC_FS
1338 static void udp6_sock_seq_show(struct seq_file *seq, struct sock *sp, int bucket)
1340 struct inet_sock *inet = inet_sk(sp);
1341 struct ipv6_pinfo *np = inet6_sk(sp);
1342 struct in6_addr *dest, *src;
1346 src = &np->rcv_saddr;
1347 destp = ntohs(inet->inet_dport);
1348 srcp = ntohs(inet->inet_sport);
1350 "%5d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1351 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d\n",
1353 src->s6_addr32[0], src->s6_addr32[1],
1354 src->s6_addr32[2], src->s6_addr32[3], srcp,
1355 dest->s6_addr32[0], dest->s6_addr32[1],
1356 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1358 sk_wmem_alloc_get(sp),
1359 sk_rmem_alloc_get(sp),
1363 atomic_read(&sp->sk_refcnt), sp,
1364 atomic_read(&sp->sk_drops));
1367 int udp6_seq_show(struct seq_file *seq, void *v)
1369 if (v == SEQ_START_TOKEN)
1374 "st tx_queue rx_queue tr tm->when retrnsmt"
1375 " uid timeout inode ref pointer drops\n");
1377 udp6_sock_seq_show(seq, v, ((struct udp_iter_state *)seq->private)->bucket);
1381 static struct udp_seq_afinfo udp6_seq_afinfo = {
1384 .udp_table = &udp_table,
1386 .owner = THIS_MODULE,
1389 .show = udp6_seq_show,
1393 int udp6_proc_init(struct net *net)
1395 return udp_proc_register(net, &udp6_seq_afinfo);
1398 void udp6_proc_exit(struct net *net) {
1399 udp_proc_unregister(net, &udp6_seq_afinfo);
1401 #endif /* CONFIG_PROC_FS */
1403 /* ------------------------------------------------------------------------ */
1405 struct proto udpv6_prot = {
1407 .owner = THIS_MODULE,
1408 .close = udp_lib_close,
1409 .connect = ip6_datagram_connect,
1410 .disconnect = udp_disconnect,
1412 .destroy = udpv6_destroy_sock,
1413 .setsockopt = udpv6_setsockopt,
1414 .getsockopt = udpv6_getsockopt,
1415 .sendmsg = udpv6_sendmsg,
1416 .recvmsg = udpv6_recvmsg,
1417 .backlog_rcv = udpv6_queue_rcv_skb,
1418 .hash = udp_lib_hash,
1419 .unhash = udp_lib_unhash,
1420 .get_port = udp_v6_get_port,
1421 .memory_allocated = &udp_memory_allocated,
1422 .sysctl_mem = sysctl_udp_mem,
1423 .sysctl_wmem = &sysctl_udp_wmem_min,
1424 .sysctl_rmem = &sysctl_udp_rmem_min,
1425 .obj_size = sizeof(struct udp6_sock),
1426 .slab_flags = SLAB_DESTROY_BY_RCU,
1427 .h.udp_table = &udp_table,
1428 #ifdef CONFIG_COMPAT
1429 .compat_setsockopt = compat_udpv6_setsockopt,
1430 .compat_getsockopt = compat_udpv6_getsockopt,
1434 static struct inet_protosw udpv6_protosw = {
1436 .protocol = IPPROTO_UDP,
1437 .prot = &udpv6_prot,
1438 .ops = &inet6_dgram_ops,
1439 .no_check = UDP_CSUM_DEFAULT,
1440 .flags = INET_PROTOSW_PERMANENT,
1444 int __init udpv6_init(void)
1448 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1452 ret = inet6_register_protosw(&udpv6_protosw);
1454 goto out_udpv6_protocol;
1459 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1463 void udpv6_exit(void)
1465 inet6_unregister_protosw(&udpv6_protosw);
1466 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);