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 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);
346 int is_udplite = IS_UDPLITE(sk);
351 *addr_len=sizeof(struct sockaddr_in6);
353 if (flags & MSG_ERRQUEUE)
354 return ipv6_recv_error(sk, msg, len);
356 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
357 return ipv6_recv_rxpmtu(sk, msg, len);
360 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
365 ulen = skb->len - sizeof(struct udphdr);
369 msg->msg_flags |= MSG_TRUNC;
371 is_udp4 = (skb->protocol == htons(ETH_P_IP));
374 * If checksum is needed at all, try to do it while copying the
375 * data. If the data is truncated, or if we only want a partial
376 * coverage checksum (UDP-Lite), do it before the copy.
379 if (len < ulen || UDP_SKB_CB(skb)->partial_cov) {
380 if (udp_lib_checksum_complete(skb))
384 if (skb_csum_unnecessary(skb))
385 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
388 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
397 UDP_INC_STATS_USER(sock_net(sk),
398 UDP_MIB_INDATAGRAMS, is_udplite);
400 UDP6_INC_STATS_USER(sock_net(sk),
401 UDP_MIB_INDATAGRAMS, is_udplite);
404 sock_recv_ts_and_drops(msg, sk, skb);
406 /* Copy the address. */
408 struct sockaddr_in6 *sin6;
410 sin6 = (struct sockaddr_in6 *) msg->msg_name;
411 sin6->sin6_family = AF_INET6;
412 sin6->sin6_port = udp_hdr(skb)->source;
413 sin6->sin6_flowinfo = 0;
414 sin6->sin6_scope_id = 0;
417 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
420 ipv6_addr_copy(&sin6->sin6_addr,
421 &ipv6_hdr(skb)->saddr);
422 if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
423 sin6->sin6_scope_id = IP6CB(skb)->iif;
428 if (inet->cmsg_flags)
429 ip_cmsg_recv(msg, skb);
432 datagram_recv_ctl(sk, msg, skb);
436 if (flags & MSG_TRUNC)
440 skb_free_datagram_locked(sk, skb);
445 slow = lock_sock_fast(sk);
446 if (!skb_kill_datagram(sk, skb, flags)) {
448 UDP_INC_STATS_USER(sock_net(sk),
449 UDP_MIB_INERRORS, is_udplite);
451 UDP6_INC_STATS_USER(sock_net(sk),
452 UDP_MIB_INERRORS, is_udplite);
454 unlock_sock_fast(sk, slow);
456 if (flags & MSG_DONTWAIT)
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 struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
467 struct in6_addr *saddr = &hdr->saddr;
468 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 (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
512 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
514 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
516 if (up->pcrlen == 0) { /* full coverage was set */
517 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage"
518 " %d while full coverage %d requested\n",
519 UDP_SKB_CB(skb)->cscov, skb->len);
522 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
523 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d "
524 "too small, need min %d\n",
525 UDP_SKB_CB(skb)->cscov, up->pcrlen);
530 if (rcu_dereference_raw(sk->sk_filter)) {
531 if (udp_lib_checksum_complete(skb))
535 if ((rc = ip_queue_rcv_skb(sk, skb)) < 0) {
536 /* Note that an ENOMEM error is charged twice */
538 UDP6_INC_STATS_BH(sock_net(sk),
539 UDP_MIB_RCVBUFERRORS, is_udplite);
540 goto drop_no_sk_drops_inc;
545 atomic_inc(&sk->sk_drops);
546 drop_no_sk_drops_inc:
547 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
552 static struct sock *udp_v6_mcast_next(struct net *net, struct sock *sk,
553 __be16 loc_port, struct in6_addr *loc_addr,
554 __be16 rmt_port, struct in6_addr *rmt_addr,
557 struct hlist_nulls_node *node;
559 unsigned short num = ntohs(loc_port);
561 sk_nulls_for_each_from(s, node) {
562 struct inet_sock *inet = inet_sk(s);
564 if (!net_eq(sock_net(s), net))
567 if (udp_sk(s)->udp_port_hash == num &&
568 s->sk_family == PF_INET6) {
569 struct ipv6_pinfo *np = inet6_sk(s);
570 if (inet->inet_dport) {
571 if (inet->inet_dport != rmt_port)
574 if (!ipv6_addr_any(&np->daddr) &&
575 !ipv6_addr_equal(&np->daddr, rmt_addr))
578 if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)
581 if (!ipv6_addr_any(&np->rcv_saddr)) {
582 if (!ipv6_addr_equal(&np->rcv_saddr, loc_addr))
585 if (!inet6_mc_check(s, loc_addr, rmt_addr))
593 static void flush_stack(struct sock **stack, unsigned int count,
594 struct sk_buff *skb, unsigned int final)
598 struct sk_buff *skb1;
600 for (i = 0; i < count; i++) {
601 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
605 if (sk_rcvqueues_full(sk, skb1)) {
610 if (!sock_owned_by_user(sk))
611 udpv6_queue_rcv_skb(sk, skb1);
612 else if (sk_add_backlog(sk, skb1)) {
621 atomic_inc(&sk->sk_drops);
622 UDP6_INC_STATS_BH(sock_net(sk),
623 UDP_MIB_RCVBUFERRORS, IS_UDPLITE(sk));
624 UDP6_INC_STATS_BH(sock_net(sk),
625 UDP_MIB_INERRORS, IS_UDPLITE(sk));
629 * Note: called only from the BH handler context,
630 * so we don't need to lock the hashes.
632 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
633 struct in6_addr *saddr, struct in6_addr *daddr,
634 struct udp_table *udptable)
636 struct sock *sk, *stack[256 / sizeof(struct sock *)];
637 const struct udphdr *uh = udp_hdr(skb);
638 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
640 unsigned int i, count = 0;
642 spin_lock(&hslot->lock);
643 sk = sk_nulls_head(&hslot->head);
644 dif = inet6_iif(skb);
645 sk = udp_v6_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
648 sk = udp_v6_mcast_next(net, sk_nulls_next(sk), uh->dest, daddr,
649 uh->source, saddr, dif);
650 if (unlikely(count == ARRAY_SIZE(stack))) {
653 flush_stack(stack, count, skb, ~0);
658 * before releasing the lock, we must take reference on sockets
660 for (i = 0; i < count; i++)
663 spin_unlock(&hslot->lock);
666 flush_stack(stack, count, skb, count - 1);
668 for (i = 0; i < count; i++)
676 static inline int udp6_csum_init(struct sk_buff *skb, struct udphdr *uh,
681 UDP_SKB_CB(skb)->partial_cov = 0;
682 UDP_SKB_CB(skb)->cscov = skb->len;
684 if (proto == IPPROTO_UDPLITE) {
685 err = udplite_checksum_init(skb, uh);
690 if (uh->check == 0) {
691 /* RFC 2460 section 8.1 says that we SHOULD log
692 this error. Well, it is reasonable.
694 LIMIT_NETDEBUG(KERN_INFO "IPv6: udp checksum is 0\n");
697 if (skb->ip_summed == CHECKSUM_COMPLETE &&
698 !csum_ipv6_magic(&ipv6_hdr(skb)->saddr, &ipv6_hdr(skb)->daddr,
699 skb->len, proto, skb->csum))
700 skb->ip_summed = CHECKSUM_UNNECESSARY;
702 if (!skb_csum_unnecessary(skb))
703 skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
704 &ipv6_hdr(skb)->daddr,
705 skb->len, proto, 0));
710 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
713 struct net *net = dev_net(skb->dev);
716 struct in6_addr *saddr, *daddr;
719 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
722 saddr = &ipv6_hdr(skb)->saddr;
723 daddr = &ipv6_hdr(skb)->daddr;
726 ulen = ntohs(uh->len);
730 if (proto == IPPROTO_UDP) {
731 /* UDP validates ulen. */
733 /* Check for jumbo payload */
737 if (ulen < sizeof(*uh))
740 if (ulen < skb->len) {
741 if (pskb_trim_rcsum(skb, ulen))
743 saddr = &ipv6_hdr(skb)->saddr;
744 daddr = &ipv6_hdr(skb)->daddr;
749 if (udp6_csum_init(skb, uh, proto))
753 * Multicast receive code
755 if (ipv6_addr_is_multicast(daddr))
756 return __udp6_lib_mcast_deliver(net, skb,
757 saddr, daddr, udptable);
762 * check socket cache ... must talk to Alan about his plans
763 * for sock caches... i'll skip this for now.
765 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
768 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
771 if (udp_lib_checksum_complete(skb))
773 UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS,
774 proto == IPPROTO_UDPLITE);
776 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
784 if (sk_rcvqueues_full(sk, skb)) {
789 if (!sock_owned_by_user(sk))
790 udpv6_queue_rcv_skb(sk, skb);
791 else if (sk_add_backlog(sk, skb)) {
792 atomic_inc(&sk->sk_drops);
802 LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
803 proto == IPPROTO_UDPLITE ? "-Lite" : "",
812 UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
817 static __inline__ int udpv6_rcv(struct sk_buff *skb)
819 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
823 * Throw away all pending data and cancel the corking. Socket is locked.
825 static void udp_v6_flush_pending_frames(struct sock *sk)
827 struct udp_sock *up = udp_sk(sk);
829 if (up->pending == AF_INET)
830 udp_flush_pending_frames(sk);
831 else if (up->pending) {
834 ip6_flush_pending_frames(sk);
839 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
840 * @sk: socket we are sending on
841 * @skb: sk_buff containing the filled-in UDP header
842 * (checksum field must be zeroed out)
844 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
845 const struct in6_addr *saddr,
846 const struct in6_addr *daddr, int len)
849 struct udphdr *uh = udp_hdr(skb);
852 if (skb_queue_len(&sk->sk_write_queue) == 1) {
853 /* Only one fragment on the socket. */
854 skb->csum_start = skb_transport_header(skb) - skb->head;
855 skb->csum_offset = offsetof(struct udphdr, check);
856 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
859 * HW-checksum won't work as there are two or more
860 * fragments on the socket so that all csums of sk_buffs
863 offset = skb_transport_offset(skb);
864 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
866 skb->ip_summed = CHECKSUM_NONE;
868 skb_queue_walk(&sk->sk_write_queue, skb) {
869 csum = csum_add(csum, skb->csum);
872 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
875 uh->check = CSUM_MANGLED_0;
883 static int udp_v6_push_pending_frames(struct sock *sk)
887 struct udp_sock *up = udp_sk(sk);
888 struct inet_sock *inet = inet_sk(sk);
889 struct flowi *fl = &inet->cork.fl;
891 int is_udplite = IS_UDPLITE(sk);
894 /* Grab the skbuff where UDP header space exists. */
895 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
899 * Create a UDP header
902 uh->source = fl->fl_ip_sport;
903 uh->dest = fl->fl_ip_dport;
904 uh->len = htons(up->len);
908 csum = udplite_csum_outgoing(sk, skb);
909 else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
910 udp6_hwcsum_outgoing(sk, skb, &fl->fl6_src, &fl->fl6_dst,
914 csum = udp_csum_outgoing(sk, skb);
916 /* add protocol-dependent pseudo-header */
917 uh->check = csum_ipv6_magic(&fl->fl6_src, &fl->fl6_dst,
918 up->len, fl->proto, csum );
920 uh->check = CSUM_MANGLED_0;
923 err = ip6_push_pending_frames(sk);
925 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
926 UDP6_INC_STATS_USER(sock_net(sk),
927 UDP_MIB_SNDBUFERRORS, is_udplite);
931 UDP6_INC_STATS_USER(sock_net(sk),
932 UDP_MIB_OUTDATAGRAMS, is_udplite);
939 int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
940 struct msghdr *msg, size_t len)
942 struct ipv6_txoptions opt_space;
943 struct udp_sock *up = udp_sk(sk);
944 struct inet_sock *inet = inet_sk(sk);
945 struct ipv6_pinfo *np = inet6_sk(sk);
946 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) msg->msg_name;
947 struct in6_addr *daddr, *final_p, final;
948 struct ipv6_txoptions *opt = NULL;
949 struct ip6_flowlabel *flowlabel = NULL;
951 struct dst_entry *dst;
952 int addr_len = msg->msg_namelen;
957 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
960 int is_udplite = IS_UDPLITE(sk);
961 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
963 /* destination address check */
965 if (addr_len < offsetof(struct sockaddr, sa_data))
968 switch (sin6->sin6_family) {
970 if (addr_len < SIN6_LEN_RFC2133)
972 daddr = &sin6->sin6_addr;
977 msg->msg_name = sin6 = NULL;
978 msg->msg_namelen = addr_len = 0;
984 } else if (!up->pending) {
985 if (sk->sk_state != TCP_ESTABLISHED)
986 return -EDESTADDRREQ;
992 if (ipv6_addr_v4mapped(daddr)) {
993 struct sockaddr_in sin;
994 sin.sin_family = AF_INET;
995 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
996 sin.sin_addr.s_addr = daddr->s6_addr32[3];
997 msg->msg_name = &sin;
998 msg->msg_namelen = sizeof(sin);
1000 if (__ipv6_only_sock(sk))
1001 return -ENETUNREACH;
1002 return udp_sendmsg(iocb, sk, msg, len);
1006 if (up->pending == AF_INET)
1007 return udp_sendmsg(iocb, sk, msg, len);
1009 /* Rough check on arithmetic overflow,
1010 better check is made in ip6_append_data().
1012 if (len > INT_MAX - sizeof(struct udphdr))
1017 * There are pending frames.
1018 * The socket lock must be held while it's corked.
1021 if (likely(up->pending)) {
1022 if (unlikely(up->pending != AF_INET6)) {
1024 return -EAFNOSUPPORT;
1027 goto do_append_data;
1031 ulen += sizeof(struct udphdr);
1033 memset(&fl, 0, sizeof(fl));
1036 if (sin6->sin6_port == 0)
1039 fl.fl_ip_dport = sin6->sin6_port;
1040 daddr = &sin6->sin6_addr;
1043 fl.fl6_flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1044 if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
1045 flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
1046 if (flowlabel == NULL)
1048 daddr = &flowlabel->dst;
1053 * Otherwise it will be difficult to maintain
1056 if (sk->sk_state == TCP_ESTABLISHED &&
1057 ipv6_addr_equal(daddr, &np->daddr))
1060 if (addr_len >= sizeof(struct sockaddr_in6) &&
1061 sin6->sin6_scope_id &&
1062 ipv6_addr_type(daddr)&IPV6_ADDR_LINKLOCAL)
1063 fl.oif = sin6->sin6_scope_id;
1065 if (sk->sk_state != TCP_ESTABLISHED)
1066 return -EDESTADDRREQ;
1068 fl.fl_ip_dport = inet->inet_dport;
1070 fl.fl6_flowlabel = np->flow_label;
1075 fl.oif = sk->sk_bound_dev_if;
1078 fl.oif = np->sticky_pktinfo.ipi6_ifindex;
1080 fl.mark = sk->sk_mark;
1082 if (msg->msg_controllen) {
1084 memset(opt, 0, sizeof(struct ipv6_txoptions));
1085 opt->tot_len = sizeof(*opt);
1087 err = datagram_send_ctl(sock_net(sk), msg, &fl, opt, &hlimit,
1088 &tclass, &dontfrag);
1090 fl6_sock_release(flowlabel);
1093 if ((fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1094 flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
1095 if (flowlabel == NULL)
1098 if (!(opt->opt_nflen|opt->opt_flen))
1105 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1106 opt = ipv6_fixup_options(&opt_space, opt);
1108 fl.proto = sk->sk_protocol;
1109 if (!ipv6_addr_any(daddr))
1110 ipv6_addr_copy(&fl.fl6_dst, daddr);
1112 fl.fl6_dst.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1113 if (ipv6_addr_any(&fl.fl6_src) && !ipv6_addr_any(&np->saddr))
1114 ipv6_addr_copy(&fl.fl6_src, &np->saddr);
1115 fl.fl_ip_sport = inet->inet_sport;
1117 final_p = fl6_update_dst(&fl, opt, &final);
1121 if (!fl.oif && ipv6_addr_is_multicast(&fl.fl6_dst)) {
1122 fl.oif = np->mcast_oif;
1126 security_sk_classify_flow(sk, &fl);
1128 dst = ip6_sk_dst_lookup_flow(sk, &fl, final_p, true);
1136 if (ipv6_addr_is_multicast(&fl.fl6_dst))
1137 hlimit = np->mcast_hops;
1139 hlimit = np->hop_limit;
1141 hlimit = ip6_dst_hoplimit(dst);
1145 tclass = np->tclass;
1148 dontfrag = np->dontfrag;
1150 if (msg->msg_flags&MSG_CONFIRM)
1155 if (unlikely(up->pending)) {
1156 /* The socket is already corked while preparing it. */
1157 /* ... which is an evident application bug. --ANK */
1160 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
1165 up->pending = AF_INET6;
1169 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1170 err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen,
1171 sizeof(struct udphdr), hlimit, tclass, opt, &fl,
1172 (struct rt6_info*)dst,
1173 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag);
1175 udp_v6_flush_pending_frames(sk);
1177 err = udp_v6_push_pending_frames(sk);
1178 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1183 ip6_dst_store(sk, dst,
1184 ipv6_addr_equal(&fl.fl6_dst, &np->daddr) ?
1186 #ifdef CONFIG_IPV6_SUBTREES
1187 ipv6_addr_equal(&fl.fl6_src, &np->saddr) ?
1198 err = np->recverr ? net_xmit_errno(err) : 0;
1202 fl6_sock_release(flowlabel);
1206 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1207 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1208 * we don't have a good statistic (IpOutDiscards but it can be too many
1209 * things). We could add another new stat but at least for now that
1210 * seems like overkill.
1212 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1213 UDP6_INC_STATS_USER(sock_net(sk),
1214 UDP_MIB_SNDBUFERRORS, is_udplite);
1220 if (!(msg->msg_flags&MSG_PROBE) || len)
1221 goto back_from_confirm;
1226 void udpv6_destroy_sock(struct sock *sk)
1229 udp_v6_flush_pending_frames(sk);
1232 inet6_destroy_sock(sk);
1236 * Socket option code for UDP
1238 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1239 char __user *optval, unsigned int optlen)
1241 if (level == SOL_UDP || level == SOL_UDPLITE)
1242 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1243 udp_v6_push_pending_frames);
1244 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1247 #ifdef CONFIG_COMPAT
1248 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1249 char __user *optval, unsigned int optlen)
1251 if (level == SOL_UDP || level == SOL_UDPLITE)
1252 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1253 udp_v6_push_pending_frames);
1254 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1258 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1259 char __user *optval, int __user *optlen)
1261 if (level == SOL_UDP || level == SOL_UDPLITE)
1262 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1263 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1266 #ifdef CONFIG_COMPAT
1267 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1268 char __user *optval, int __user *optlen)
1270 if (level == SOL_UDP || level == SOL_UDPLITE)
1271 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1272 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1276 static int udp6_ufo_send_check(struct sk_buff *skb)
1278 struct ipv6hdr *ipv6h;
1281 if (!pskb_may_pull(skb, sizeof(*uh)))
1284 ipv6h = ipv6_hdr(skb);
1287 uh->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, skb->len,
1289 skb->csum_start = skb_transport_header(skb) - skb->head;
1290 skb->csum_offset = offsetof(struct udphdr, check);
1291 skb->ip_summed = CHECKSUM_PARTIAL;
1295 static struct sk_buff *udp6_ufo_fragment(struct sk_buff *skb, u32 features)
1297 struct sk_buff *segs = ERR_PTR(-EINVAL);
1299 unsigned int unfrag_ip6hlen, unfrag_len;
1300 struct frag_hdr *fptr;
1301 u8 *mac_start, *prevhdr;
1303 u8 frag_hdr_sz = sizeof(struct frag_hdr);
1307 mss = skb_shinfo(skb)->gso_size;
1308 if (unlikely(skb->len <= mss))
1311 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
1312 /* Packet is from an untrusted source, reset gso_segs. */
1313 int type = skb_shinfo(skb)->gso_type;
1315 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
1316 !(type & (SKB_GSO_UDP))))
1319 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
1325 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
1326 * do checksum of UDP packets sent as multiple IP fragments.
1328 offset = skb->csum_start - skb_headroom(skb);
1329 csum = skb_checksum(skb, offset, skb->len- offset, 0);
1330 offset += skb->csum_offset;
1331 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1332 skb->ip_summed = CHECKSUM_NONE;
1334 /* Check if there is enough headroom to insert fragment header. */
1335 if ((skb_headroom(skb) < frag_hdr_sz) &&
1336 pskb_expand_head(skb, frag_hdr_sz, 0, GFP_ATOMIC))
1339 /* Find the unfragmentable header and shift it left by frag_hdr_sz
1340 * bytes to insert fragment header.
1342 unfrag_ip6hlen = ip6_find_1stfragopt(skb, &prevhdr);
1344 *prevhdr = NEXTHDR_FRAGMENT;
1345 unfrag_len = skb_network_header(skb) - skb_mac_header(skb) +
1347 mac_start = skb_mac_header(skb);
1348 memmove(mac_start-frag_hdr_sz, mac_start, unfrag_len);
1350 skb->mac_header -= frag_hdr_sz;
1351 skb->network_header -= frag_hdr_sz;
1353 fptr = (struct frag_hdr *)(skb_network_header(skb) + unfrag_ip6hlen);
1354 fptr->nexthdr = nexthdr;
1356 ipv6_select_ident(fptr);
1358 /* Fragment the skb. ipv6 header and the remaining fields of the
1359 * fragment header are updated in ipv6_gso_segment()
1361 segs = skb_segment(skb, features);
1367 static const struct inet6_protocol udpv6_protocol = {
1368 .handler = udpv6_rcv,
1369 .err_handler = udpv6_err,
1370 .gso_send_check = udp6_ufo_send_check,
1371 .gso_segment = udp6_ufo_fragment,
1372 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1375 /* ------------------------------------------------------------------------ */
1376 #ifdef CONFIG_PROC_FS
1378 static void udp6_sock_seq_show(struct seq_file *seq, struct sock *sp, int bucket)
1380 struct inet_sock *inet = inet_sk(sp);
1381 struct ipv6_pinfo *np = inet6_sk(sp);
1382 struct in6_addr *dest, *src;
1386 src = &np->rcv_saddr;
1387 destp = ntohs(inet->inet_dport);
1388 srcp = ntohs(inet->inet_sport);
1390 "%5d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1391 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d\n",
1393 src->s6_addr32[0], src->s6_addr32[1],
1394 src->s6_addr32[2], src->s6_addr32[3], srcp,
1395 dest->s6_addr32[0], dest->s6_addr32[1],
1396 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1398 sk_wmem_alloc_get(sp),
1399 sk_rmem_alloc_get(sp),
1403 atomic_read(&sp->sk_refcnt), sp,
1404 atomic_read(&sp->sk_drops));
1407 int udp6_seq_show(struct seq_file *seq, void *v)
1409 if (v == SEQ_START_TOKEN)
1414 "st tx_queue rx_queue tr tm->when retrnsmt"
1415 " uid timeout inode ref pointer drops\n");
1417 udp6_sock_seq_show(seq, v, ((struct udp_iter_state *)seq->private)->bucket);
1421 static struct udp_seq_afinfo udp6_seq_afinfo = {
1424 .udp_table = &udp_table,
1426 .owner = THIS_MODULE,
1429 .show = udp6_seq_show,
1433 int __net_init udp6_proc_init(struct net *net)
1435 return udp_proc_register(net, &udp6_seq_afinfo);
1438 void udp6_proc_exit(struct net *net) {
1439 udp_proc_unregister(net, &udp6_seq_afinfo);
1441 #endif /* CONFIG_PROC_FS */
1443 /* ------------------------------------------------------------------------ */
1445 struct proto udpv6_prot = {
1447 .owner = THIS_MODULE,
1448 .close = udp_lib_close,
1449 .connect = ip6_datagram_connect,
1450 .disconnect = udp_disconnect,
1452 .destroy = udpv6_destroy_sock,
1453 .setsockopt = udpv6_setsockopt,
1454 .getsockopt = udpv6_getsockopt,
1455 .sendmsg = udpv6_sendmsg,
1456 .recvmsg = udpv6_recvmsg,
1457 .backlog_rcv = udpv6_queue_rcv_skb,
1458 .hash = udp_lib_hash,
1459 .unhash = udp_lib_unhash,
1460 .rehash = udp_v6_rehash,
1461 .get_port = udp_v6_get_port,
1462 .memory_allocated = &udp_memory_allocated,
1463 .sysctl_mem = sysctl_udp_mem,
1464 .sysctl_wmem = &sysctl_udp_wmem_min,
1465 .sysctl_rmem = &sysctl_udp_rmem_min,
1466 .obj_size = sizeof(struct udp6_sock),
1467 .slab_flags = SLAB_DESTROY_BY_RCU,
1468 .h.udp_table = &udp_table,
1469 #ifdef CONFIG_COMPAT
1470 .compat_setsockopt = compat_udpv6_setsockopt,
1471 .compat_getsockopt = compat_udpv6_getsockopt,
1473 .clear_sk = sk_prot_clear_portaddr_nulls,
1476 static struct inet_protosw udpv6_protosw = {
1478 .protocol = IPPROTO_UDP,
1479 .prot = &udpv6_prot,
1480 .ops = &inet6_dgram_ops,
1481 .no_check = UDP_CSUM_DEFAULT,
1482 .flags = INET_PROTOSW_PERMANENT,
1486 int __init udpv6_init(void)
1490 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1494 ret = inet6_register_protosw(&udpv6_protosw);
1496 goto out_udpv6_protocol;
1501 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1505 void udpv6_exit(void)
1507 inet6_unregister_protosw(&udpv6_protosw);
1508 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);