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),
387 msg->msg_iov, copied);
396 UDP_INC_STATS_USER(sock_net(sk),
397 UDP_MIB_INDATAGRAMS, is_udplite);
399 UDP6_INC_STATS_USER(sock_net(sk),
400 UDP_MIB_INDATAGRAMS, is_udplite);
403 sock_recv_ts_and_drops(msg, sk, skb);
405 /* Copy the address. */
407 struct sockaddr_in6 *sin6;
409 sin6 = (struct sockaddr_in6 *) msg->msg_name;
410 sin6->sin6_family = AF_INET6;
411 sin6->sin6_port = udp_hdr(skb)->source;
412 sin6->sin6_flowinfo = 0;
413 sin6->sin6_scope_id = 0;
416 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
419 ipv6_addr_copy(&sin6->sin6_addr,
420 &ipv6_hdr(skb)->saddr);
421 if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
422 sin6->sin6_scope_id = IP6CB(skb)->iif;
424 *addr_len = sizeof(*sin6);
427 if (inet->cmsg_flags)
428 ip_cmsg_recv(msg, skb);
431 datagram_recv_ctl(sk, msg, skb);
435 if (flags & MSG_TRUNC)
439 skb_free_datagram_locked(sk, skb);
444 slow = lock_sock_fast(sk);
445 if (!skb_kill_datagram(sk, skb, flags)) {
447 UDP_INC_STATS_USER(sock_net(sk),
448 UDP_MIB_INERRORS, is_udplite);
450 UDP6_INC_STATS_USER(sock_net(sk),
451 UDP_MIB_INERRORS, is_udplite);
453 unlock_sock_fast(sk, slow);
455 /* starting over for a new packet, but check if we need to yield */
457 msg->msg_flags &= ~MSG_TRUNC;
461 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
462 u8 type, u8 code, int offset, __be32 info,
463 struct udp_table *udptable)
465 struct ipv6_pinfo *np;
466 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
467 const struct in6_addr *saddr = &hdr->saddr;
468 const struct in6_addr *daddr = &hdr->daddr;
469 struct udphdr *uh = (struct udphdr*)(skb->data+offset);
473 sk = __udp6_lib_lookup(dev_net(skb->dev), daddr, uh->dest,
474 saddr, uh->source, inet6_iif(skb), udptable);
480 if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
483 if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
487 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
490 sk->sk_error_report(sk);
495 static __inline__ void udpv6_err(struct sk_buff *skb,
496 struct inet6_skb_parm *opt, u8 type,
497 u8 code, int offset, __be32 info )
499 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
502 int udpv6_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
504 struct udp_sock *up = udp_sk(sk);
506 int is_udplite = IS_UDPLITE(sk);
508 if (!ipv6_addr_any(&inet6_sk(sk)->daddr))
509 sock_rps_save_rxhash(sk, skb);
511 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
515 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
517 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
519 if (up->pcrlen == 0) { /* full coverage was set */
520 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage"
521 " %d while full coverage %d requested\n",
522 UDP_SKB_CB(skb)->cscov, skb->len);
525 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
526 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d "
527 "too small, need min %d\n",
528 UDP_SKB_CB(skb)->cscov, up->pcrlen);
533 if (rcu_access_pointer(sk->sk_filter)) {
534 if (udp_lib_checksum_complete(skb))
538 if ((rc = ip_queue_rcv_skb(sk, skb)) < 0) {
539 /* Note that an ENOMEM error is charged twice */
541 UDP6_INC_STATS_BH(sock_net(sk),
542 UDP_MIB_RCVBUFERRORS, is_udplite);
543 goto drop_no_sk_drops_inc;
548 atomic_inc(&sk->sk_drops);
549 drop_no_sk_drops_inc:
550 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
555 static struct sock *udp_v6_mcast_next(struct net *net, struct sock *sk,
556 __be16 loc_port, const struct in6_addr *loc_addr,
557 __be16 rmt_port, const struct in6_addr *rmt_addr,
560 struct hlist_nulls_node *node;
562 unsigned short num = ntohs(loc_port);
564 sk_nulls_for_each_from(s, node) {
565 struct inet_sock *inet = inet_sk(s);
567 if (!net_eq(sock_net(s), net))
570 if (udp_sk(s)->udp_port_hash == num &&
571 s->sk_family == PF_INET6) {
572 struct ipv6_pinfo *np = inet6_sk(s);
573 if (inet->inet_dport) {
574 if (inet->inet_dport != rmt_port)
577 if (!ipv6_addr_any(&np->daddr) &&
578 !ipv6_addr_equal(&np->daddr, rmt_addr))
581 if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)
584 if (!ipv6_addr_any(&np->rcv_saddr)) {
585 if (!ipv6_addr_equal(&np->rcv_saddr, loc_addr))
588 if (!inet6_mc_check(s, loc_addr, rmt_addr))
596 static void flush_stack(struct sock **stack, unsigned int count,
597 struct sk_buff *skb, unsigned int final)
601 struct sk_buff *skb1;
603 for (i = 0; i < count; i++) {
604 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
608 if (sk_rcvqueues_full(sk, skb1)) {
613 if (!sock_owned_by_user(sk))
614 udpv6_queue_rcv_skb(sk, skb1);
615 else if (sk_add_backlog(sk, skb1)) {
624 atomic_inc(&sk->sk_drops);
625 UDP6_INC_STATS_BH(sock_net(sk),
626 UDP_MIB_RCVBUFERRORS, IS_UDPLITE(sk));
627 UDP6_INC_STATS_BH(sock_net(sk),
628 UDP_MIB_INERRORS, IS_UDPLITE(sk));
632 * Note: called only from the BH handler context,
633 * so we don't need to lock the hashes.
635 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
636 const struct in6_addr *saddr, const struct in6_addr *daddr,
637 struct udp_table *udptable)
639 struct sock *sk, *stack[256 / sizeof(struct sock *)];
640 const struct udphdr *uh = udp_hdr(skb);
641 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
643 unsigned int i, count = 0;
645 spin_lock(&hslot->lock);
646 sk = sk_nulls_head(&hslot->head);
647 dif = inet6_iif(skb);
648 sk = udp_v6_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
651 sk = udp_v6_mcast_next(net, sk_nulls_next(sk), uh->dest, daddr,
652 uh->source, saddr, dif);
653 if (unlikely(count == ARRAY_SIZE(stack))) {
656 flush_stack(stack, count, skb, ~0);
661 * before releasing the lock, we must take reference on sockets
663 for (i = 0; i < count; i++)
666 spin_unlock(&hslot->lock);
669 flush_stack(stack, count, skb, count - 1);
671 for (i = 0; i < count; i++)
679 static inline int udp6_csum_init(struct sk_buff *skb, struct udphdr *uh,
684 UDP_SKB_CB(skb)->partial_cov = 0;
685 UDP_SKB_CB(skb)->cscov = skb->len;
687 if (proto == IPPROTO_UDPLITE) {
688 err = udplite_checksum_init(skb, uh);
693 if (uh->check == 0) {
694 /* RFC 2460 section 8.1 says that we SHOULD log
695 this error. Well, it is reasonable.
697 LIMIT_NETDEBUG(KERN_INFO "IPv6: udp checksum is 0\n");
700 if (skb->ip_summed == CHECKSUM_COMPLETE &&
701 !csum_ipv6_magic(&ipv6_hdr(skb)->saddr, &ipv6_hdr(skb)->daddr,
702 skb->len, proto, skb->csum))
703 skb->ip_summed = CHECKSUM_UNNECESSARY;
705 if (!skb_csum_unnecessary(skb))
706 skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
707 &ipv6_hdr(skb)->daddr,
708 skb->len, proto, 0));
713 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
716 struct net *net = dev_net(skb->dev);
719 const struct in6_addr *saddr, *daddr;
722 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
725 saddr = &ipv6_hdr(skb)->saddr;
726 daddr = &ipv6_hdr(skb)->daddr;
729 ulen = ntohs(uh->len);
733 if (proto == IPPROTO_UDP) {
734 /* UDP validates ulen. */
736 /* Check for jumbo payload */
740 if (ulen < sizeof(*uh))
743 if (ulen < skb->len) {
744 if (pskb_trim_rcsum(skb, ulen))
746 saddr = &ipv6_hdr(skb)->saddr;
747 daddr = &ipv6_hdr(skb)->daddr;
752 if (udp6_csum_init(skb, uh, proto))
756 * Multicast receive code
758 if (ipv6_addr_is_multicast(daddr))
759 return __udp6_lib_mcast_deliver(net, skb,
760 saddr, daddr, udptable);
765 * check socket cache ... must talk to Alan about his plans
766 * for sock caches... i'll skip this for now.
768 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
771 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
774 if (udp_lib_checksum_complete(skb))
776 UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS,
777 proto == IPPROTO_UDPLITE);
779 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
787 if (sk_rcvqueues_full(sk, skb)) {
792 if (!sock_owned_by_user(sk))
793 udpv6_queue_rcv_skb(sk, skb);
794 else if (sk_add_backlog(sk, skb)) {
795 atomic_inc(&sk->sk_drops);
805 LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
806 proto == IPPROTO_UDPLITE ? "-Lite" : "",
815 UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
820 static __inline__ int udpv6_rcv(struct sk_buff *skb)
822 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
826 * Throw away all pending data and cancel the corking. Socket is locked.
828 static void udp_v6_flush_pending_frames(struct sock *sk)
830 struct udp_sock *up = udp_sk(sk);
832 if (up->pending == AF_INET)
833 udp_flush_pending_frames(sk);
834 else if (up->pending) {
837 ip6_flush_pending_frames(sk);
842 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
843 * @sk: socket we are sending on
844 * @skb: sk_buff containing the filled-in UDP header
845 * (checksum field must be zeroed out)
847 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
848 const struct in6_addr *saddr,
849 const struct in6_addr *daddr, int len)
852 struct udphdr *uh = udp_hdr(skb);
855 if (skb_queue_len(&sk->sk_write_queue) == 1) {
856 /* Only one fragment on the socket. */
857 skb->csum_start = skb_transport_header(skb) - skb->head;
858 skb->csum_offset = offsetof(struct udphdr, check);
859 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
862 * HW-checksum won't work as there are two or more
863 * fragments on the socket so that all csums of sk_buffs
866 offset = skb_transport_offset(skb);
867 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
869 skb->ip_summed = CHECKSUM_NONE;
871 skb_queue_walk(&sk->sk_write_queue, skb) {
872 csum = csum_add(csum, skb->csum);
875 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
878 uh->check = CSUM_MANGLED_0;
886 static int udp_v6_push_pending_frames(struct sock *sk)
890 struct udp_sock *up = udp_sk(sk);
891 struct inet_sock *inet = inet_sk(sk);
894 int is_udplite = IS_UDPLITE(sk);
897 if (up->pending == AF_INET)
898 return udp_push_pending_frames(sk);
900 fl6 = &inet->cork.fl.u.ip6;
902 /* Grab the skbuff where UDP header space exists. */
903 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
907 * Create a UDP header
910 uh->source = fl6->fl6_sport;
911 uh->dest = fl6->fl6_dport;
912 uh->len = htons(up->len);
916 csum = udplite_csum_outgoing(sk, skb);
917 else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
918 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr,
922 csum = udp_csum_outgoing(sk, skb);
924 /* add protocol-dependent pseudo-header */
925 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
926 up->len, fl6->flowi6_proto, csum);
928 uh->check = CSUM_MANGLED_0;
931 err = ip6_push_pending_frames(sk);
933 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
934 UDP6_INC_STATS_USER(sock_net(sk),
935 UDP_MIB_SNDBUFERRORS, is_udplite);
939 UDP6_INC_STATS_USER(sock_net(sk),
940 UDP_MIB_OUTDATAGRAMS, is_udplite);
947 int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
948 struct msghdr *msg, size_t len)
950 struct ipv6_txoptions opt_space;
951 struct udp_sock *up = udp_sk(sk);
952 struct inet_sock *inet = inet_sk(sk);
953 struct ipv6_pinfo *np = inet6_sk(sk);
954 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) msg->msg_name;
955 struct in6_addr *daddr, *final_p, final;
956 struct ipv6_txoptions *opt = NULL;
957 struct ipv6_txoptions *opt_to_free = NULL;
958 struct ip6_flowlabel *flowlabel = NULL;
960 struct dst_entry *dst;
961 int addr_len = msg->msg_namelen;
966 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
969 int is_udplite = IS_UDPLITE(sk);
970 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
972 /* destination address check */
974 if (addr_len < offsetof(struct sockaddr, sa_data))
977 switch (sin6->sin6_family) {
979 if (addr_len < SIN6_LEN_RFC2133)
981 daddr = &sin6->sin6_addr;
986 msg->msg_name = sin6 = NULL;
987 msg->msg_namelen = addr_len = 0;
993 } else if (!up->pending) {
994 if (sk->sk_state != TCP_ESTABLISHED)
995 return -EDESTADDRREQ;
1001 if (ipv6_addr_v4mapped(daddr)) {
1002 struct sockaddr_in sin;
1003 sin.sin_family = AF_INET;
1004 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1005 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1006 msg->msg_name = &sin;
1007 msg->msg_namelen = sizeof(sin);
1009 if (__ipv6_only_sock(sk))
1010 return -ENETUNREACH;
1011 return udp_sendmsg(iocb, sk, msg, len);
1015 if (up->pending == AF_INET)
1016 return udp_sendmsg(iocb, sk, msg, len);
1018 /* Rough check on arithmetic overflow,
1019 better check is made in ip6_append_data().
1021 if (len > INT_MAX - sizeof(struct udphdr))
1026 * There are pending frames.
1027 * The socket lock must be held while it's corked.
1030 if (likely(up->pending)) {
1031 if (unlikely(up->pending != AF_INET6)) {
1033 return -EAFNOSUPPORT;
1036 goto do_append_data;
1040 ulen += sizeof(struct udphdr);
1042 memset(&fl6, 0, sizeof(fl6));
1045 if (sin6->sin6_port == 0)
1048 fl6.fl6_dport = sin6->sin6_port;
1049 daddr = &sin6->sin6_addr;
1052 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1053 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1054 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1055 if (flowlabel == NULL)
1057 daddr = &flowlabel->dst;
1062 * Otherwise it will be difficult to maintain
1065 if (sk->sk_state == TCP_ESTABLISHED &&
1066 ipv6_addr_equal(daddr, &np->daddr))
1069 if (addr_len >= sizeof(struct sockaddr_in6) &&
1070 sin6->sin6_scope_id &&
1071 ipv6_addr_type(daddr)&IPV6_ADDR_LINKLOCAL)
1072 fl6.flowi6_oif = sin6->sin6_scope_id;
1074 if (sk->sk_state != TCP_ESTABLISHED)
1075 return -EDESTADDRREQ;
1077 fl6.fl6_dport = inet->inet_dport;
1079 fl6.flowlabel = np->flow_label;
1083 if (!fl6.flowi6_oif)
1084 fl6.flowi6_oif = sk->sk_bound_dev_if;
1086 if (!fl6.flowi6_oif)
1087 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1089 fl6.flowi6_mark = sk->sk_mark;
1091 if (msg->msg_controllen) {
1093 memset(opt, 0, sizeof(struct ipv6_txoptions));
1094 opt->tot_len = sizeof(*opt);
1096 err = datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt,
1097 &hlimit, &tclass, &dontfrag);
1099 fl6_sock_release(flowlabel);
1102 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1103 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1104 if (flowlabel == NULL)
1107 if (!(opt->opt_nflen|opt->opt_flen))
1112 opt = txopt_get(np);
1116 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1117 opt = ipv6_fixup_options(&opt_space, opt);
1119 fl6.flowi6_proto = sk->sk_protocol;
1120 if (!ipv6_addr_any(daddr))
1121 ipv6_addr_copy(&fl6.daddr, daddr);
1123 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1124 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1125 ipv6_addr_copy(&fl6.saddr, &np->saddr);
1126 fl6.fl6_sport = inet->inet_sport;
1128 final_p = fl6_update_dst(&fl6, opt, &final);
1132 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1133 fl6.flowi6_oif = np->mcast_oif;
1137 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1139 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, true);
1147 if (ipv6_addr_is_multicast(&fl6.daddr))
1148 hlimit = np->mcast_hops;
1150 hlimit = np->hop_limit;
1152 hlimit = ip6_dst_hoplimit(dst);
1156 tclass = np->tclass;
1159 dontfrag = np->dontfrag;
1161 if (msg->msg_flags&MSG_CONFIRM)
1166 if (unlikely(up->pending)) {
1167 /* The socket is already corked while preparing it. */
1168 /* ... which is an evident application bug. --ANK */
1171 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
1176 up->pending = AF_INET6;
1180 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1181 err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen,
1182 sizeof(struct udphdr), hlimit, tclass, opt, &fl6,
1183 (struct rt6_info*)dst,
1184 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag);
1186 udp_v6_flush_pending_frames(sk);
1188 err = udp_v6_push_pending_frames(sk);
1189 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1194 ip6_dst_store(sk, dst,
1195 ipv6_addr_equal(&fl6.daddr, &np->daddr) ?
1197 #ifdef CONFIG_IPV6_SUBTREES
1198 ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1209 err = np->recverr ? net_xmit_errno(err) : 0;
1213 fl6_sock_release(flowlabel);
1214 txopt_put(opt_to_free);
1218 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1219 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1220 * we don't have a good statistic (IpOutDiscards but it can be too many
1221 * things). We could add another new stat but at least for now that
1222 * seems like overkill.
1224 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1225 UDP6_INC_STATS_USER(sock_net(sk),
1226 UDP_MIB_SNDBUFERRORS, is_udplite);
1232 if (!(msg->msg_flags&MSG_PROBE) || len)
1233 goto back_from_confirm;
1238 void udpv6_destroy_sock(struct sock *sk)
1241 udp_v6_flush_pending_frames(sk);
1244 inet6_destroy_sock(sk);
1248 * Socket option code for UDP
1250 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1251 char __user *optval, unsigned int optlen)
1253 if (level == SOL_UDP || level == SOL_UDPLITE)
1254 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1255 udp_v6_push_pending_frames);
1256 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1259 #ifdef CONFIG_COMPAT
1260 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1261 char __user *optval, unsigned int optlen)
1263 if (level == SOL_UDP || level == SOL_UDPLITE)
1264 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1265 udp_v6_push_pending_frames);
1266 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1270 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1271 char __user *optval, int __user *optlen)
1273 if (level == SOL_UDP || level == SOL_UDPLITE)
1274 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1275 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1278 #ifdef CONFIG_COMPAT
1279 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1280 char __user *optval, int __user *optlen)
1282 if (level == SOL_UDP || level == SOL_UDPLITE)
1283 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1284 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1288 static int udp6_ufo_send_check(struct sk_buff *skb)
1290 const struct ipv6hdr *ipv6h;
1293 if (!pskb_may_pull(skb, sizeof(*uh)))
1296 ipv6h = ipv6_hdr(skb);
1299 uh->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, skb->len,
1301 skb->csum_start = skb_transport_header(skb) - skb->head;
1302 skb->csum_offset = offsetof(struct udphdr, check);
1303 skb->ip_summed = CHECKSUM_PARTIAL;
1307 static struct sk_buff *udp6_ufo_fragment(struct sk_buff *skb, u32 features)
1309 struct sk_buff *segs = ERR_PTR(-EINVAL);
1311 unsigned int unfrag_ip6hlen, unfrag_len;
1312 struct frag_hdr *fptr;
1313 u8 *mac_start, *prevhdr;
1315 u8 frag_hdr_sz = sizeof(struct frag_hdr);
1319 mss = skb_shinfo(skb)->gso_size;
1320 if (unlikely(skb->len <= mss))
1323 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
1324 /* Packet is from an untrusted source, reset gso_segs. */
1325 int type = skb_shinfo(skb)->gso_type;
1327 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
1328 !(type & (SKB_GSO_UDP))))
1331 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
1337 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
1338 * do checksum of UDP packets sent as multiple IP fragments.
1340 offset = skb_checksum_start_offset(skb);
1341 csum = skb_checksum(skb, offset, skb->len- offset, 0);
1342 offset += skb->csum_offset;
1343 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1344 skb->ip_summed = CHECKSUM_NONE;
1346 /* Check if there is enough headroom to insert fragment header. */
1347 if ((skb_mac_header(skb) < skb->head + frag_hdr_sz) &&
1348 pskb_expand_head(skb, frag_hdr_sz, 0, GFP_ATOMIC))
1351 /* Find the unfragmentable header and shift it left by frag_hdr_sz
1352 * bytes to insert fragment header.
1354 unfrag_ip6hlen = ip6_find_1stfragopt(skb, &prevhdr);
1356 *prevhdr = NEXTHDR_FRAGMENT;
1357 unfrag_len = skb_network_header(skb) - skb_mac_header(skb) +
1359 mac_start = skb_mac_header(skb);
1360 memmove(mac_start-frag_hdr_sz, mac_start, unfrag_len);
1362 skb->mac_header -= frag_hdr_sz;
1363 skb->network_header -= frag_hdr_sz;
1365 fptr = (struct frag_hdr *)(skb_network_header(skb) + unfrag_ip6hlen);
1366 fptr->nexthdr = nexthdr;
1368 fptr->identification = skb_shinfo(skb)->ip6_frag_id;
1370 /* Fragment the skb. ipv6 header and the remaining fields of the
1371 * fragment header are updated in ipv6_gso_segment()
1373 segs = skb_segment(skb, features);
1379 static const struct inet6_protocol udpv6_protocol = {
1380 .handler = udpv6_rcv,
1381 .err_handler = udpv6_err,
1382 .gso_send_check = udp6_ufo_send_check,
1383 .gso_segment = udp6_ufo_fragment,
1384 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1387 /* ------------------------------------------------------------------------ */
1388 #ifdef CONFIG_PROC_FS
1390 static void udp6_sock_seq_show(struct seq_file *seq, struct sock *sp, int bucket)
1392 struct inet_sock *inet = inet_sk(sp);
1393 struct ipv6_pinfo *np = inet6_sk(sp);
1394 const struct in6_addr *dest, *src;
1398 src = &np->rcv_saddr;
1399 destp = ntohs(inet->inet_dport);
1400 srcp = ntohs(inet->inet_sport);
1402 "%5d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1403 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d\n",
1405 src->s6_addr32[0], src->s6_addr32[1],
1406 src->s6_addr32[2], src->s6_addr32[3], srcp,
1407 dest->s6_addr32[0], dest->s6_addr32[1],
1408 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1410 sk_wmem_alloc_get(sp),
1411 sk_rmem_alloc_get(sp),
1415 atomic_read(&sp->sk_refcnt), sp,
1416 atomic_read(&sp->sk_drops));
1419 int udp6_seq_show(struct seq_file *seq, void *v)
1421 if (v == SEQ_START_TOKEN)
1426 "st tx_queue rx_queue tr tm->when retrnsmt"
1427 " uid timeout inode ref pointer drops\n");
1429 udp6_sock_seq_show(seq, v, ((struct udp_iter_state *)seq->private)->bucket);
1433 static const struct file_operations udp6_afinfo_seq_fops = {
1434 .owner = THIS_MODULE,
1435 .open = udp_seq_open,
1437 .llseek = seq_lseek,
1438 .release = seq_release_net
1441 static struct udp_seq_afinfo udp6_seq_afinfo = {
1444 .udp_table = &udp_table,
1445 .seq_fops = &udp6_afinfo_seq_fops,
1447 .show = udp6_seq_show,
1451 int __net_init udp6_proc_init(struct net *net)
1453 return udp_proc_register(net, &udp6_seq_afinfo);
1456 void udp6_proc_exit(struct net *net) {
1457 udp_proc_unregister(net, &udp6_seq_afinfo);
1459 #endif /* CONFIG_PROC_FS */
1461 void udp_v6_clear_sk(struct sock *sk, int size)
1463 struct inet_sock *inet = inet_sk(sk);
1465 /* we do not want to clear pinet6 field, because of RCU lookups */
1466 sk_prot_clear_portaddr_nulls(sk, offsetof(struct inet_sock, pinet6));
1468 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1469 memset(&inet->pinet6 + 1, 0, size);
1472 /* ------------------------------------------------------------------------ */
1474 struct proto udpv6_prot = {
1476 .owner = THIS_MODULE,
1477 .close = udp_lib_close,
1478 .connect = ip6_datagram_connect,
1479 .disconnect = udp_disconnect,
1481 .destroy = udpv6_destroy_sock,
1482 .setsockopt = udpv6_setsockopt,
1483 .getsockopt = udpv6_getsockopt,
1484 .sendmsg = udpv6_sendmsg,
1485 .recvmsg = udpv6_recvmsg,
1486 .backlog_rcv = udpv6_queue_rcv_skb,
1487 .hash = udp_lib_hash,
1488 .unhash = udp_lib_unhash,
1489 .rehash = udp_v6_rehash,
1490 .get_port = udp_v6_get_port,
1491 .memory_allocated = &udp_memory_allocated,
1492 .sysctl_mem = sysctl_udp_mem,
1493 .sysctl_wmem = &sysctl_udp_wmem_min,
1494 .sysctl_rmem = &sysctl_udp_rmem_min,
1495 .obj_size = sizeof(struct udp6_sock),
1496 .slab_flags = SLAB_DESTROY_BY_RCU,
1497 .h.udp_table = &udp_table,
1498 #ifdef CONFIG_COMPAT
1499 .compat_setsockopt = compat_udpv6_setsockopt,
1500 .compat_getsockopt = compat_udpv6_getsockopt,
1502 .clear_sk = udp_v6_clear_sk,
1505 static struct inet_protosw udpv6_protosw = {
1507 .protocol = IPPROTO_UDP,
1508 .prot = &udpv6_prot,
1509 .ops = &inet6_dgram_ops,
1510 .no_check = UDP_CSUM_DEFAULT,
1511 .flags = INET_PROTOSW_PERMANENT,
1515 int __init udpv6_init(void)
1519 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1523 ret = inet6_register_protosw(&udpv6_protosw);
1525 goto out_udpv6_protocol;
1530 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1534 void udpv6_exit(void)
1536 inet6_unregister_protosw(&udpv6_protosw);
1537 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);