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>
49 #include <linux/proc_fs.h>
50 #include <linux/seq_file.h>
53 int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2)
55 const struct in6_addr *sk_rcv_saddr6 = &inet6_sk(sk)->rcv_saddr;
56 const struct in6_addr *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2);
57 __be32 sk1_rcv_saddr = sk_rcv_saddr(sk);
58 __be32 sk2_rcv_saddr = sk_rcv_saddr(sk2);
59 int sk_ipv6only = ipv6_only_sock(sk);
60 int sk2_ipv6only = inet_v6_ipv6only(sk2);
61 int addr_type = ipv6_addr_type(sk_rcv_saddr6);
62 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
64 /* if both are mapped, treat as IPv4 */
65 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED)
66 return (!sk2_ipv6only &&
67 (!sk1_rcv_saddr || !sk2_rcv_saddr ||
68 sk1_rcv_saddr == sk2_rcv_saddr));
70 if (addr_type2 == IPV6_ADDR_ANY &&
71 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
74 if (addr_type == IPV6_ADDR_ANY &&
75 !(sk_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
79 ipv6_addr_equal(sk_rcv_saddr6, sk2_rcv_saddr6))
85 static unsigned int udp6_portaddr_hash(struct net *net,
86 const struct in6_addr *addr6,
89 unsigned int hash, mix = net_hash_mix(net);
91 if (ipv6_addr_any(addr6))
92 hash = jhash_1word(0, mix);
93 else if (ipv6_addr_v4mapped(addr6))
94 hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix);
96 hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix);
102 int udp_v6_get_port(struct sock *sk, unsigned short snum)
104 unsigned int hash2_nulladdr =
105 udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
106 unsigned int hash2_partial =
107 udp6_portaddr_hash(sock_net(sk), &inet6_sk(sk)->rcv_saddr, 0);
109 /* precompute partial secondary hash */
110 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
111 return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal, hash2_nulladdr);
114 static void udp_v6_rehash(struct sock *sk)
116 u16 new_hash = udp6_portaddr_hash(sock_net(sk),
117 &inet6_sk(sk)->rcv_saddr,
118 inet_sk(sk)->inet_num);
120 udp_lib_rehash(sk, new_hash);
123 static inline int compute_score(struct sock *sk, struct net *net,
125 const struct in6_addr *saddr, __be16 sport,
126 const struct in6_addr *daddr, __be16 dport,
131 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
132 sk->sk_family == PF_INET6) {
133 struct ipv6_pinfo *np = inet6_sk(sk);
134 struct inet_sock *inet = inet_sk(sk);
137 if (inet->inet_dport) {
138 if (inet->inet_dport != sport)
142 if (!ipv6_addr_any(&np->rcv_saddr)) {
143 if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
147 if (!ipv6_addr_any(&np->daddr)) {
148 if (!ipv6_addr_equal(&np->daddr, saddr))
152 if (sk->sk_bound_dev_if) {
153 if (sk->sk_bound_dev_if != dif)
161 #define SCORE2_MAX (1 + 1 + 1)
162 static inline int compute_score2(struct sock *sk, struct net *net,
163 const struct in6_addr *saddr, __be16 sport,
164 const struct in6_addr *daddr, unsigned short hnum,
169 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
170 sk->sk_family == PF_INET6) {
171 struct ipv6_pinfo *np = inet6_sk(sk);
172 struct inet_sock *inet = inet_sk(sk);
174 if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
177 if (inet->inet_dport) {
178 if (inet->inet_dport != sport)
182 if (!ipv6_addr_any(&np->daddr)) {
183 if (!ipv6_addr_equal(&np->daddr, saddr))
187 if (sk->sk_bound_dev_if) {
188 if (sk->sk_bound_dev_if != dif)
197 /* called with read_rcu_lock() */
198 static struct sock *udp6_lib_lookup2(struct net *net,
199 const struct in6_addr *saddr, __be16 sport,
200 const struct in6_addr *daddr, unsigned int hnum, int dif,
201 struct udp_hslot *hslot2, unsigned int slot2)
203 struct sock *sk, *result;
204 struct hlist_nulls_node *node;
210 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
211 score = compute_score2(sk, net, saddr, sport,
213 if (score > badness) {
216 if (score == SCORE2_MAX)
221 * if the nulls value we got at the end of this lookup is
222 * not the expected one, we must restart lookup.
223 * We probably met an item that was moved to another chain.
225 if (get_nulls_value(node) != slot2)
230 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
232 else if (unlikely(compute_score2(result, net, saddr, sport,
233 daddr, hnum, dif) < badness)) {
241 static struct sock *__udp6_lib_lookup(struct net *net,
242 const struct in6_addr *saddr, __be16 sport,
243 const struct in6_addr *daddr, __be16 dport,
244 int dif, struct udp_table *udptable)
246 struct sock *sk, *result;
247 struct hlist_nulls_node *node;
248 unsigned short hnum = ntohs(dport);
249 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
250 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
254 if (hslot->count > 10) {
255 hash2 = udp6_portaddr_hash(net, daddr, hnum);
256 slot2 = hash2 & udptable->mask;
257 hslot2 = &udptable->hash2[slot2];
258 if (hslot->count < hslot2->count)
261 result = udp6_lib_lookup2(net, saddr, sport,
265 hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
266 slot2 = hash2 & udptable->mask;
267 hslot2 = &udptable->hash2[slot2];
268 if (hslot->count < hslot2->count)
271 result = udp6_lib_lookup2(net, saddr, sport,
272 &in6addr_any, hnum, dif,
281 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
282 score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif);
283 if (score > badness) {
289 * if the nulls value we got at the end of this lookup is
290 * not the expected one, we must restart lookup.
291 * We probably met an item that was moved to another chain.
293 if (get_nulls_value(node) != slot)
297 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
299 else if (unlikely(compute_score(result, net, hnum, saddr, sport,
300 daddr, dport, dif) < badness)) {
309 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
310 __be16 sport, __be16 dport,
311 struct udp_table *udptable)
314 const struct ipv6hdr *iph = ipv6_hdr(skb);
316 if (unlikely(sk = skb_steal_sock(skb)))
318 return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport,
319 &iph->daddr, dport, inet6_iif(skb),
323 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
324 const struct in6_addr *daddr, __be16 dport, int dif)
326 return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
328 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
332 * This should be easy, if there is something there we
333 * return it, otherwise we block.
336 int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk,
337 struct msghdr *msg, size_t len,
338 int noblock, int flags, int *addr_len)
340 struct ipv6_pinfo *np = inet6_sk(sk);
341 struct inet_sock *inet = inet_sk(sk);
343 unsigned int ulen, copied;
346 int is_udplite = IS_UDPLITE(sk);
350 if (flags & MSG_ERRQUEUE)
351 return ipv6_recv_error(sk, msg, len, addr_len);
353 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
354 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
357 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
362 ulen = skb->len - sizeof(struct udphdr);
366 else if (copied < ulen)
367 msg->msg_flags |= MSG_TRUNC;
369 is_udp4 = (skb->protocol == htons(ETH_P_IP));
372 * If checksum is needed at all, try to do it while copying the
373 * data. If the data is truncated, or if we only want a partial
374 * coverage checksum (UDP-Lite), do it before the copy.
377 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
378 if (udp_lib_checksum_complete(skb))
382 if (skb_csum_unnecessary(skb))
383 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
384 msg->msg_iov, copied );
386 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
395 UDP_INC_STATS_USER(sock_net(sk),
396 UDP_MIB_INDATAGRAMS, is_udplite);
398 UDP6_INC_STATS_USER(sock_net(sk),
399 UDP_MIB_INDATAGRAMS, is_udplite);
402 sock_recv_ts_and_drops(msg, sk, skb);
404 /* Copy the address. */
406 struct sockaddr_in6 *sin6;
408 sin6 = (struct sockaddr_in6 *) msg->msg_name;
409 sin6->sin6_family = AF_INET6;
410 sin6->sin6_port = udp_hdr(skb)->source;
411 sin6->sin6_flowinfo = 0;
412 sin6->sin6_scope_id = 0;
415 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
418 ipv6_addr_copy(&sin6->sin6_addr,
419 &ipv6_hdr(skb)->saddr);
420 if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
421 sin6->sin6_scope_id = IP6CB(skb)->iif;
423 *addr_len = sizeof(*sin6);
426 if (inet->cmsg_flags)
427 ip_cmsg_recv(msg, skb);
430 datagram_recv_ctl(sk, msg, skb);
434 if (flags & MSG_TRUNC)
438 skb_free_datagram_locked(sk, skb);
443 slow = lock_sock_fast(sk);
444 if (!skb_kill_datagram(sk, skb, flags)) {
446 UDP_INC_STATS_USER(sock_net(sk),
447 UDP_MIB_INERRORS, is_udplite);
449 UDP6_INC_STATS_USER(sock_net(sk),
450 UDP_MIB_INERRORS, is_udplite);
452 unlock_sock_fast(sk, slow);
454 /* starting over for a new packet, but check if we need to yield */
456 msg->msg_flags &= ~MSG_TRUNC;
460 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
461 u8 type, u8 code, int offset, __be32 info,
462 struct udp_table *udptable)
464 struct ipv6_pinfo *np;
465 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
466 const struct in6_addr *saddr = &hdr->saddr;
467 const struct in6_addr *daddr = &hdr->daddr;
468 struct udphdr *uh = (struct udphdr*)(skb->data+offset);
472 sk = __udp6_lib_lookup(dev_net(skb->dev), daddr, uh->dest,
473 saddr, uh->source, inet6_iif(skb), udptable);
479 if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
482 if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
486 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
489 sk->sk_error_report(sk);
494 static __inline__ void udpv6_err(struct sk_buff *skb,
495 struct inet6_skb_parm *opt, u8 type,
496 u8 code, int offset, __be32 info )
498 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
501 int udpv6_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
503 struct udp_sock *up = udp_sk(sk);
505 int is_udplite = IS_UDPLITE(sk);
507 if (!ipv6_addr_any(&inet6_sk(sk)->daddr))
508 sock_rps_save_rxhash(sk, skb);
510 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
514 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
516 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
518 if (up->pcrlen == 0) { /* full coverage was set */
519 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage"
520 " %d while full coverage %d requested\n",
521 UDP_SKB_CB(skb)->cscov, skb->len);
524 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
525 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d "
526 "too small, need min %d\n",
527 UDP_SKB_CB(skb)->cscov, up->pcrlen);
532 if (rcu_access_pointer(sk->sk_filter)) {
533 if (udp_lib_checksum_complete(skb))
537 if ((rc = ip_queue_rcv_skb(sk, skb)) < 0) {
538 /* Note that an ENOMEM error is charged twice */
540 UDP6_INC_STATS_BH(sock_net(sk),
541 UDP_MIB_RCVBUFERRORS, is_udplite);
542 goto drop_no_sk_drops_inc;
547 atomic_inc(&sk->sk_drops);
548 drop_no_sk_drops_inc:
549 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
554 static struct sock *udp_v6_mcast_next(struct net *net, struct sock *sk,
555 __be16 loc_port, const struct in6_addr *loc_addr,
556 __be16 rmt_port, const struct in6_addr *rmt_addr,
559 struct hlist_nulls_node *node;
561 unsigned short num = ntohs(loc_port);
563 sk_nulls_for_each_from(s, node) {
564 struct inet_sock *inet = inet_sk(s);
566 if (!net_eq(sock_net(s), net))
569 if (udp_sk(s)->udp_port_hash == num &&
570 s->sk_family == PF_INET6) {
571 struct ipv6_pinfo *np = inet6_sk(s);
572 if (inet->inet_dport) {
573 if (inet->inet_dport != rmt_port)
576 if (!ipv6_addr_any(&np->daddr) &&
577 !ipv6_addr_equal(&np->daddr, rmt_addr))
580 if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)
583 if (!ipv6_addr_any(&np->rcv_saddr)) {
584 if (!ipv6_addr_equal(&np->rcv_saddr, loc_addr))
587 if (!inet6_mc_check(s, loc_addr, rmt_addr))
595 static void flush_stack(struct sock **stack, unsigned int count,
596 struct sk_buff *skb, unsigned int final)
600 struct sk_buff *skb1;
602 for (i = 0; i < count; i++) {
603 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
607 if (sk_rcvqueues_full(sk, skb1)) {
612 if (!sock_owned_by_user(sk))
613 udpv6_queue_rcv_skb(sk, skb1);
614 else if (sk_add_backlog(sk, skb1)) {
623 atomic_inc(&sk->sk_drops);
624 UDP6_INC_STATS_BH(sock_net(sk),
625 UDP_MIB_RCVBUFERRORS, IS_UDPLITE(sk));
626 UDP6_INC_STATS_BH(sock_net(sk),
627 UDP_MIB_INERRORS, IS_UDPLITE(sk));
631 * Note: called only from the BH handler context,
632 * so we don't need to lock the hashes.
634 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
635 const struct in6_addr *saddr, const struct in6_addr *daddr,
636 struct udp_table *udptable)
638 struct sock *sk, *stack[256 / sizeof(struct sock *)];
639 const struct udphdr *uh = udp_hdr(skb);
640 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
642 unsigned int i, count = 0;
644 spin_lock(&hslot->lock);
645 sk = sk_nulls_head(&hslot->head);
646 dif = inet6_iif(skb);
647 sk = udp_v6_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
650 sk = udp_v6_mcast_next(net, sk_nulls_next(sk), uh->dest, daddr,
651 uh->source, saddr, dif);
652 if (unlikely(count == ARRAY_SIZE(stack))) {
655 flush_stack(stack, count, skb, ~0);
660 * before releasing the lock, we must take reference on sockets
662 for (i = 0; i < count; i++)
665 spin_unlock(&hslot->lock);
668 flush_stack(stack, count, skb, count - 1);
670 for (i = 0; i < count; i++)
678 static inline int udp6_csum_init(struct sk_buff *skb, struct udphdr *uh,
683 UDP_SKB_CB(skb)->partial_cov = 0;
684 UDP_SKB_CB(skb)->cscov = skb->len;
686 if (proto == IPPROTO_UDPLITE) {
687 err = udplite_checksum_init(skb, uh);
692 if (uh->check == 0) {
693 /* RFC 2460 section 8.1 says that we SHOULD log
694 this error. Well, it is reasonable.
696 LIMIT_NETDEBUG(KERN_INFO "IPv6: udp checksum is 0\n");
699 if (skb->ip_summed == CHECKSUM_COMPLETE &&
700 !csum_ipv6_magic(&ipv6_hdr(skb)->saddr, &ipv6_hdr(skb)->daddr,
701 skb->len, proto, skb->csum))
702 skb->ip_summed = CHECKSUM_UNNECESSARY;
704 if (!skb_csum_unnecessary(skb))
705 skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
706 &ipv6_hdr(skb)->daddr,
707 skb->len, proto, 0));
712 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
715 struct net *net = dev_net(skb->dev);
718 const struct in6_addr *saddr, *daddr;
721 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
724 saddr = &ipv6_hdr(skb)->saddr;
725 daddr = &ipv6_hdr(skb)->daddr;
728 ulen = ntohs(uh->len);
732 if (proto == IPPROTO_UDP) {
733 /* UDP validates ulen. */
735 /* Check for jumbo payload */
739 if (ulen < sizeof(*uh))
742 if (ulen < skb->len) {
743 if (pskb_trim_rcsum(skb, ulen))
745 saddr = &ipv6_hdr(skb)->saddr;
746 daddr = &ipv6_hdr(skb)->daddr;
751 if (udp6_csum_init(skb, uh, proto))
755 * Multicast receive code
757 if (ipv6_addr_is_multicast(daddr))
758 return __udp6_lib_mcast_deliver(net, skb,
759 saddr, daddr, udptable);
764 * check socket cache ... must talk to Alan about his plans
765 * for sock caches... i'll skip this for now.
767 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
770 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
773 if (udp_lib_checksum_complete(skb))
775 UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS,
776 proto == IPPROTO_UDPLITE);
778 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
786 if (sk_rcvqueues_full(sk, skb)) {
791 if (!sock_owned_by_user(sk))
792 udpv6_queue_rcv_skb(sk, skb);
793 else if (sk_add_backlog(sk, skb)) {
794 atomic_inc(&sk->sk_drops);
804 LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
805 proto == IPPROTO_UDPLITE ? "-Lite" : "",
814 UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
819 static __inline__ int udpv6_rcv(struct sk_buff *skb)
821 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
825 * Throw away all pending data and cancel the corking. Socket is locked.
827 static void udp_v6_flush_pending_frames(struct sock *sk)
829 struct udp_sock *up = udp_sk(sk);
831 if (up->pending == AF_INET)
832 udp_flush_pending_frames(sk);
833 else if (up->pending) {
836 ip6_flush_pending_frames(sk);
841 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
842 * @sk: socket we are sending on
843 * @skb: sk_buff containing the filled-in UDP header
844 * (checksum field must be zeroed out)
846 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
847 const struct in6_addr *saddr,
848 const struct in6_addr *daddr, int len)
851 struct udphdr *uh = udp_hdr(skb);
854 if (skb_queue_len(&sk->sk_write_queue) == 1) {
855 /* Only one fragment on the socket. */
856 skb->csum_start = skb_transport_header(skb) - skb->head;
857 skb->csum_offset = offsetof(struct udphdr, check);
858 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
861 * HW-checksum won't work as there are two or more
862 * fragments on the socket so that all csums of sk_buffs
865 offset = skb_transport_offset(skb);
866 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
868 skb->ip_summed = CHECKSUM_NONE;
870 skb_queue_walk(&sk->sk_write_queue, skb) {
871 csum = csum_add(csum, skb->csum);
874 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
877 uh->check = CSUM_MANGLED_0;
885 static int udp_v6_push_pending_frames(struct sock *sk)
889 struct udp_sock *up = udp_sk(sk);
890 struct inet_sock *inet = inet_sk(sk);
893 int is_udplite = IS_UDPLITE(sk);
896 if (up->pending == AF_INET)
897 return udp_push_pending_frames(sk);
899 fl6 = &inet->cork.fl.u.ip6;
901 /* Grab the skbuff where UDP header space exists. */
902 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
906 * Create a UDP header
909 uh->source = fl6->fl6_sport;
910 uh->dest = fl6->fl6_dport;
911 uh->len = htons(up->len);
915 csum = udplite_csum_outgoing(sk, skb);
916 else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
917 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr,
921 csum = udp_csum_outgoing(sk, skb);
923 /* add protocol-dependent pseudo-header */
924 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
925 up->len, fl6->flowi6_proto, csum);
927 uh->check = CSUM_MANGLED_0;
930 err = ip6_push_pending_frames(sk);
932 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
933 UDP6_INC_STATS_USER(sock_net(sk),
934 UDP_MIB_SNDBUFERRORS, is_udplite);
938 UDP6_INC_STATS_USER(sock_net(sk),
939 UDP_MIB_OUTDATAGRAMS, is_udplite);
946 int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
947 struct msghdr *msg, size_t len)
949 struct ipv6_txoptions opt_space;
950 struct udp_sock *up = udp_sk(sk);
951 struct inet_sock *inet = inet_sk(sk);
952 struct ipv6_pinfo *np = inet6_sk(sk);
953 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) msg->msg_name;
954 struct in6_addr *daddr, *final_p, final;
955 struct ipv6_txoptions *opt = NULL;
956 struct ipv6_txoptions *opt_to_free = NULL;
957 struct ip6_flowlabel *flowlabel = NULL;
959 struct dst_entry *dst;
960 int addr_len = msg->msg_namelen;
965 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
968 int is_udplite = IS_UDPLITE(sk);
969 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
971 /* destination address check */
973 if (addr_len < offsetof(struct sockaddr, sa_data))
976 switch (sin6->sin6_family) {
978 if (addr_len < SIN6_LEN_RFC2133)
980 daddr = &sin6->sin6_addr;
985 msg->msg_name = sin6 = NULL;
986 msg->msg_namelen = addr_len = 0;
992 } else if (!up->pending) {
993 if (sk->sk_state != TCP_ESTABLISHED)
994 return -EDESTADDRREQ;
1000 if (ipv6_addr_v4mapped(daddr)) {
1001 struct sockaddr_in sin;
1002 sin.sin_family = AF_INET;
1003 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1004 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1005 msg->msg_name = &sin;
1006 msg->msg_namelen = sizeof(sin);
1008 if (__ipv6_only_sock(sk))
1009 return -ENETUNREACH;
1010 return udp_sendmsg(iocb, sk, msg, len);
1014 if (up->pending == AF_INET)
1015 return udp_sendmsg(iocb, sk, msg, len);
1017 /* Rough check on arithmetic overflow,
1018 better check is made in ip6_append_data().
1020 if (len > INT_MAX - sizeof(struct udphdr))
1025 * There are pending frames.
1026 * The socket lock must be held while it's corked.
1029 if (likely(up->pending)) {
1030 if (unlikely(up->pending != AF_INET6)) {
1032 return -EAFNOSUPPORT;
1035 goto do_append_data;
1039 ulen += sizeof(struct udphdr);
1041 memset(&fl6, 0, sizeof(fl6));
1044 if (sin6->sin6_port == 0)
1047 fl6.fl6_dport = sin6->sin6_port;
1048 daddr = &sin6->sin6_addr;
1051 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1052 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1053 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1054 if (flowlabel == NULL)
1056 daddr = &flowlabel->dst;
1061 * Otherwise it will be difficult to maintain
1064 if (sk->sk_state == TCP_ESTABLISHED &&
1065 ipv6_addr_equal(daddr, &np->daddr))
1068 if (addr_len >= sizeof(struct sockaddr_in6) &&
1069 sin6->sin6_scope_id &&
1070 ipv6_addr_type(daddr)&IPV6_ADDR_LINKLOCAL)
1071 fl6.flowi6_oif = sin6->sin6_scope_id;
1073 if (sk->sk_state != TCP_ESTABLISHED)
1074 return -EDESTADDRREQ;
1076 fl6.fl6_dport = inet->inet_dport;
1078 fl6.flowlabel = np->flow_label;
1082 if (!fl6.flowi6_oif)
1083 fl6.flowi6_oif = sk->sk_bound_dev_if;
1085 if (!fl6.flowi6_oif)
1086 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1088 fl6.flowi6_mark = sk->sk_mark;
1090 if (msg->msg_controllen) {
1092 memset(opt, 0, sizeof(struct ipv6_txoptions));
1093 opt->tot_len = sizeof(*opt);
1095 err = datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt,
1096 &hlimit, &tclass, &dontfrag);
1098 fl6_sock_release(flowlabel);
1101 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1102 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1103 if (flowlabel == NULL)
1106 if (!(opt->opt_nflen|opt->opt_flen))
1111 opt = txopt_get(np);
1115 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1116 opt = ipv6_fixup_options(&opt_space, opt);
1118 fl6.flowi6_proto = sk->sk_protocol;
1119 if (!ipv6_addr_any(daddr))
1120 ipv6_addr_copy(&fl6.daddr, daddr);
1122 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1123 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1124 ipv6_addr_copy(&fl6.saddr, &np->saddr);
1125 fl6.fl6_sport = inet->inet_sport;
1127 final_p = fl6_update_dst(&fl6, opt, &final);
1131 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1132 fl6.flowi6_oif = np->mcast_oif;
1136 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1138 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, true);
1146 if (ipv6_addr_is_multicast(&fl6.daddr))
1147 hlimit = np->mcast_hops;
1149 hlimit = np->hop_limit;
1151 hlimit = ip6_dst_hoplimit(dst);
1155 tclass = np->tclass;
1158 dontfrag = np->dontfrag;
1160 if (msg->msg_flags&MSG_CONFIRM)
1165 if (unlikely(up->pending)) {
1166 /* The socket is already corked while preparing it. */
1167 /* ... which is an evident application bug. --ANK */
1170 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
1175 up->pending = AF_INET6;
1179 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1180 err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen,
1181 sizeof(struct udphdr), hlimit, tclass, opt, &fl6,
1182 (struct rt6_info*)dst,
1183 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag);
1185 udp_v6_flush_pending_frames(sk);
1187 err = udp_v6_push_pending_frames(sk);
1188 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1193 ip6_dst_store(sk, dst,
1194 ipv6_addr_equal(&fl6.daddr, &np->daddr) ?
1196 #ifdef CONFIG_IPV6_SUBTREES
1197 ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1208 err = np->recverr ? net_xmit_errno(err) : 0;
1212 fl6_sock_release(flowlabel);
1213 txopt_put(opt_to_free);
1217 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1218 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1219 * we don't have a good statistic (IpOutDiscards but it can be too many
1220 * things). We could add another new stat but at least for now that
1221 * seems like overkill.
1223 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1224 UDP6_INC_STATS_USER(sock_net(sk),
1225 UDP_MIB_SNDBUFERRORS, is_udplite);
1231 if (!(msg->msg_flags&MSG_PROBE) || len)
1232 goto back_from_confirm;
1237 void udpv6_destroy_sock(struct sock *sk)
1240 udp_v6_flush_pending_frames(sk);
1243 inet6_destroy_sock(sk);
1247 * Socket option code for UDP
1249 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1250 char __user *optval, unsigned int optlen)
1252 if (level == SOL_UDP || level == SOL_UDPLITE)
1253 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1254 udp_v6_push_pending_frames);
1255 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1258 #ifdef CONFIG_COMPAT
1259 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1260 char __user *optval, unsigned int optlen)
1262 if (level == SOL_UDP || level == SOL_UDPLITE)
1263 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1264 udp_v6_push_pending_frames);
1265 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1269 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1270 char __user *optval, int __user *optlen)
1272 if (level == SOL_UDP || level == SOL_UDPLITE)
1273 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1274 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1277 #ifdef CONFIG_COMPAT
1278 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1279 char __user *optval, int __user *optlen)
1281 if (level == SOL_UDP || level == SOL_UDPLITE)
1282 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1283 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1287 static int udp6_ufo_send_check(struct sk_buff *skb)
1289 const struct ipv6hdr *ipv6h;
1292 if (!pskb_may_pull(skb, sizeof(*uh)))
1295 ipv6h = ipv6_hdr(skb);
1298 uh->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, skb->len,
1300 skb->csum_start = skb_transport_header(skb) - skb->head;
1301 skb->csum_offset = offsetof(struct udphdr, check);
1302 skb->ip_summed = CHECKSUM_PARTIAL;
1306 static struct sk_buff *udp6_ufo_fragment(struct sk_buff *skb, u32 features)
1308 struct sk_buff *segs = ERR_PTR(-EINVAL);
1310 unsigned int unfrag_ip6hlen, unfrag_len;
1311 struct frag_hdr *fptr;
1312 u8 *mac_start, *prevhdr;
1314 u8 frag_hdr_sz = sizeof(struct frag_hdr);
1318 mss = skb_shinfo(skb)->gso_size;
1319 if (unlikely(skb->len <= mss))
1322 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
1323 /* Packet is from an untrusted source, reset gso_segs. */
1324 int type = skb_shinfo(skb)->gso_type;
1326 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
1327 !(type & (SKB_GSO_UDP))))
1330 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
1336 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
1337 * do checksum of UDP packets sent as multiple IP fragments.
1339 offset = skb_checksum_start_offset(skb);
1340 csum = skb_checksum(skb, offset, skb->len- offset, 0);
1341 offset += skb->csum_offset;
1342 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1343 skb->ip_summed = CHECKSUM_NONE;
1345 /* Check if there is enough headroom to insert fragment header. */
1346 if ((skb_mac_header(skb) < skb->head + frag_hdr_sz) &&
1347 pskb_expand_head(skb, frag_hdr_sz, 0, GFP_ATOMIC))
1350 /* Find the unfragmentable header and shift it left by frag_hdr_sz
1351 * bytes to insert fragment header.
1353 unfrag_ip6hlen = ip6_find_1stfragopt(skb, &prevhdr);
1355 *prevhdr = NEXTHDR_FRAGMENT;
1356 unfrag_len = skb_network_header(skb) - skb_mac_header(skb) +
1358 mac_start = skb_mac_header(skb);
1359 memmove(mac_start-frag_hdr_sz, mac_start, unfrag_len);
1361 skb->mac_header -= frag_hdr_sz;
1362 skb->network_header -= frag_hdr_sz;
1364 fptr = (struct frag_hdr *)(skb_network_header(skb) + unfrag_ip6hlen);
1365 fptr->nexthdr = nexthdr;
1367 fptr->identification = skb_shinfo(skb)->ip6_frag_id;
1369 /* Fragment the skb. ipv6 header and the remaining fields of the
1370 * fragment header are updated in ipv6_gso_segment()
1372 segs = skb_segment(skb, features);
1378 static const struct inet6_protocol udpv6_protocol = {
1379 .handler = udpv6_rcv,
1380 .err_handler = udpv6_err,
1381 .gso_send_check = udp6_ufo_send_check,
1382 .gso_segment = udp6_ufo_fragment,
1383 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1386 /* ------------------------------------------------------------------------ */
1387 #ifdef CONFIG_PROC_FS
1389 static void udp6_sock_seq_show(struct seq_file *seq, struct sock *sp, int bucket)
1391 struct inet_sock *inet = inet_sk(sp);
1392 struct ipv6_pinfo *np = inet6_sk(sp);
1393 const struct in6_addr *dest, *src;
1397 src = &np->rcv_saddr;
1398 destp = ntohs(inet->inet_dport);
1399 srcp = ntohs(inet->inet_sport);
1401 "%5d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1402 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d\n",
1404 src->s6_addr32[0], src->s6_addr32[1],
1405 src->s6_addr32[2], src->s6_addr32[3], srcp,
1406 dest->s6_addr32[0], dest->s6_addr32[1],
1407 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1409 sk_wmem_alloc_get(sp),
1410 sk_rmem_alloc_get(sp),
1414 atomic_read(&sp->sk_refcnt), sp,
1415 atomic_read(&sp->sk_drops));
1418 int udp6_seq_show(struct seq_file *seq, void *v)
1420 if (v == SEQ_START_TOKEN)
1425 "st tx_queue rx_queue tr tm->when retrnsmt"
1426 " uid timeout inode ref pointer drops\n");
1428 udp6_sock_seq_show(seq, v, ((struct udp_iter_state *)seq->private)->bucket);
1432 static const struct file_operations udp6_afinfo_seq_fops = {
1433 .owner = THIS_MODULE,
1434 .open = udp_seq_open,
1436 .llseek = seq_lseek,
1437 .release = seq_release_net
1440 static struct udp_seq_afinfo udp6_seq_afinfo = {
1443 .udp_table = &udp_table,
1444 .seq_fops = &udp6_afinfo_seq_fops,
1446 .show = udp6_seq_show,
1450 int __net_init udp6_proc_init(struct net *net)
1452 return udp_proc_register(net, &udp6_seq_afinfo);
1455 void udp6_proc_exit(struct net *net) {
1456 udp_proc_unregister(net, &udp6_seq_afinfo);
1458 #endif /* CONFIG_PROC_FS */
1460 void udp_v6_clear_sk(struct sock *sk, int size)
1462 struct inet_sock *inet = inet_sk(sk);
1464 /* we do not want to clear pinet6 field, because of RCU lookups */
1465 sk_prot_clear_portaddr_nulls(sk, offsetof(struct inet_sock, pinet6));
1467 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1468 memset(&inet->pinet6 + 1, 0, size);
1471 /* ------------------------------------------------------------------------ */
1473 struct proto udpv6_prot = {
1475 .owner = THIS_MODULE,
1476 .close = udp_lib_close,
1477 .connect = ip6_datagram_connect,
1478 .disconnect = udp_disconnect,
1480 .destroy = udpv6_destroy_sock,
1481 .setsockopt = udpv6_setsockopt,
1482 .getsockopt = udpv6_getsockopt,
1483 .sendmsg = udpv6_sendmsg,
1484 .recvmsg = udpv6_recvmsg,
1485 .backlog_rcv = udpv6_queue_rcv_skb,
1486 .hash = udp_lib_hash,
1487 .unhash = udp_lib_unhash,
1488 .rehash = udp_v6_rehash,
1489 .get_port = udp_v6_get_port,
1490 .memory_allocated = &udp_memory_allocated,
1491 .sysctl_mem = sysctl_udp_mem,
1492 .sysctl_wmem = &sysctl_udp_wmem_min,
1493 .sysctl_rmem = &sysctl_udp_rmem_min,
1494 .obj_size = sizeof(struct udp6_sock),
1495 .slab_flags = SLAB_DESTROY_BY_RCU,
1496 .h.udp_table = &udp_table,
1497 #ifdef CONFIG_COMPAT
1498 .compat_setsockopt = compat_udpv6_setsockopt,
1499 .compat_getsockopt = compat_udpv6_getsockopt,
1501 .clear_sk = udp_v6_clear_sk,
1504 static struct inet_protosw udpv6_protosw = {
1506 .protocol = IPPROTO_UDP,
1507 .prot = &udpv6_prot,
1508 .ops = &inet6_dgram_ops,
1509 .no_check = UDP_CSUM_DEFAULT,
1510 .flags = INET_PROTOSW_PERMANENT,
1514 int __init udpv6_init(void)
1518 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1522 ret = inet6_register_protosw(&udpv6_protosw);
1524 goto out_udpv6_protocol;
1529 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1533 void udpv6_exit(void)
1535 inet6_unregister_protosw(&udpv6_protosw);
1536 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);