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>
48 #include <net/inet6_hashtables.h>
50 #include <linux/proc_fs.h>
51 #include <linux/seq_file.h>
52 #include <trace/events/skb.h>
55 int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2)
57 const struct in6_addr *sk_rcv_saddr6 = &inet6_sk(sk)->rcv_saddr;
58 const struct in6_addr *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2);
59 __be32 sk1_rcv_saddr = sk_rcv_saddr(sk);
60 __be32 sk2_rcv_saddr = sk_rcv_saddr(sk2);
61 int sk_ipv6only = ipv6_only_sock(sk);
62 int sk2_ipv6only = inet_v6_ipv6only(sk2);
63 int addr_type = ipv6_addr_type(sk_rcv_saddr6);
64 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
66 /* if both are mapped, treat as IPv4 */
67 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED)
68 return (!sk2_ipv6only &&
69 (!sk1_rcv_saddr || !sk2_rcv_saddr ||
70 sk1_rcv_saddr == sk2_rcv_saddr));
72 if (addr_type2 == IPV6_ADDR_ANY &&
73 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
76 if (addr_type == IPV6_ADDR_ANY &&
77 !(sk_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
81 ipv6_addr_equal(sk_rcv_saddr6, sk2_rcv_saddr6))
87 static unsigned int udp6_portaddr_hash(struct net *net,
88 const struct in6_addr *addr6,
91 unsigned int hash, mix = net_hash_mix(net);
93 if (ipv6_addr_any(addr6))
94 hash = jhash_1word(0, mix);
95 else if (ipv6_addr_v4mapped(addr6))
96 hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix);
98 hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix);
104 int udp_v6_get_port(struct sock *sk, unsigned short snum)
106 unsigned int hash2_nulladdr =
107 udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
108 unsigned int hash2_partial =
109 udp6_portaddr_hash(sock_net(sk), &inet6_sk(sk)->rcv_saddr, 0);
111 /* precompute partial secondary hash */
112 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
113 return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal, hash2_nulladdr);
116 static void udp_v6_rehash(struct sock *sk)
118 u16 new_hash = udp6_portaddr_hash(sock_net(sk),
119 &inet6_sk(sk)->rcv_saddr,
120 inet_sk(sk)->inet_num);
122 udp_lib_rehash(sk, new_hash);
125 static inline int compute_score(struct sock *sk, struct net *net,
127 const struct in6_addr *saddr, __be16 sport,
128 const struct in6_addr *daddr, __be16 dport,
133 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
134 sk->sk_family == PF_INET6) {
135 struct ipv6_pinfo *np = inet6_sk(sk);
136 struct inet_sock *inet = inet_sk(sk);
139 if (inet->inet_dport) {
140 if (inet->inet_dport != sport)
144 if (!ipv6_addr_any(&np->rcv_saddr)) {
145 if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
149 if (!ipv6_addr_any(&np->daddr)) {
150 if (!ipv6_addr_equal(&np->daddr, saddr))
154 if (sk->sk_bound_dev_if) {
155 if (sk->sk_bound_dev_if != dif)
163 #define SCORE2_MAX (1 + 1 + 1)
164 static inline int compute_score2(struct sock *sk, struct net *net,
165 const struct in6_addr *saddr, __be16 sport,
166 const struct in6_addr *daddr, unsigned short hnum,
171 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
172 sk->sk_family == PF_INET6) {
173 struct ipv6_pinfo *np = inet6_sk(sk);
174 struct inet_sock *inet = inet_sk(sk);
176 if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
179 if (inet->inet_dport) {
180 if (inet->inet_dport != sport)
184 if (!ipv6_addr_any(&np->daddr)) {
185 if (!ipv6_addr_equal(&np->daddr, saddr))
189 if (sk->sk_bound_dev_if) {
190 if (sk->sk_bound_dev_if != dif)
199 /* called with read_rcu_lock() */
200 static struct sock *udp6_lib_lookup2(struct net *net,
201 const struct in6_addr *saddr, __be16 sport,
202 const struct in6_addr *daddr, unsigned int hnum, int dif,
203 struct udp_hslot *hslot2, unsigned int slot2)
205 struct sock *sk, *result;
206 struct hlist_nulls_node *node;
207 int score, badness, matches = 0, reuseport = 0;
213 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
214 score = compute_score2(sk, net, saddr, sport,
216 if (score > badness) {
219 reuseport = sk->sk_reuseport;
221 hash = inet6_ehashfn(net, daddr, hnum,
224 } else if (score == SCORE2_MAX)
226 } else if (score == badness && reuseport) {
228 if (((u64)hash * matches) >> 32 == 0)
230 hash = next_pseudo_random32(hash);
234 * if the nulls value we got at the end of this lookup is
235 * not the expected one, we must restart lookup.
236 * We probably met an item that was moved to another chain.
238 if (get_nulls_value(node) != slot2)
243 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
245 else if (unlikely(compute_score2(result, net, saddr, sport,
246 daddr, hnum, dif) < badness)) {
254 struct sock *__udp6_lib_lookup(struct net *net,
255 const struct in6_addr *saddr, __be16 sport,
256 const struct in6_addr *daddr, __be16 dport,
257 int dif, struct udp_table *udptable)
259 struct sock *sk, *result;
260 struct hlist_nulls_node *node;
261 unsigned short hnum = ntohs(dport);
262 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
263 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
264 int score, badness, matches = 0, reuseport = 0;
268 if (hslot->count > 10) {
269 hash2 = udp6_portaddr_hash(net, daddr, hnum);
270 slot2 = hash2 & udptable->mask;
271 hslot2 = &udptable->hash2[slot2];
272 if (hslot->count < hslot2->count)
275 result = udp6_lib_lookup2(net, saddr, sport,
279 hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
280 slot2 = hash2 & udptable->mask;
281 hslot2 = &udptable->hash2[slot2];
282 if (hslot->count < hslot2->count)
285 result = udp6_lib_lookup2(net, saddr, sport,
286 &in6addr_any, hnum, dif,
295 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
296 score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif);
297 if (score > badness) {
300 reuseport = sk->sk_reuseport;
302 hash = inet6_ehashfn(net, daddr, hnum,
306 } else if (score == badness && reuseport) {
308 if (((u64)hash * matches) >> 32 == 0)
310 hash = next_pseudo_random32(hash);
314 * if the nulls value we got at the end of this lookup is
315 * not the expected one, we must restart lookup.
316 * We probably met an item that was moved to another chain.
318 if (get_nulls_value(node) != slot)
322 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
324 else if (unlikely(compute_score(result, net, hnum, saddr, sport,
325 daddr, dport, dif) < badness)) {
333 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
335 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
336 __be16 sport, __be16 dport,
337 struct udp_table *udptable)
340 const struct ipv6hdr *iph = ipv6_hdr(skb);
342 if (unlikely(sk = skb_steal_sock(skb)))
344 return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport,
345 &iph->daddr, dport, inet6_iif(skb),
349 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
350 const struct in6_addr *daddr, __be16 dport, int dif)
352 return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
354 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
358 * This should be easy, if there is something there we
359 * return it, otherwise we block.
362 int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk,
363 struct msghdr *msg, size_t len,
364 int noblock, int flags, int *addr_len)
366 struct ipv6_pinfo *np = inet6_sk(sk);
367 struct inet_sock *inet = inet_sk(sk);
369 unsigned int ulen, copied;
372 int is_udplite = IS_UDPLITE(sk);
377 *addr_len = sizeof(struct sockaddr_in6);
379 if (flags & MSG_ERRQUEUE)
380 return ipv6_recv_error(sk, msg, len);
382 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
383 return ipv6_recv_rxpmtu(sk, msg, len);
386 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
387 &peeked, &off, &err);
391 ulen = skb->len - sizeof(struct udphdr);
395 else if (copied < ulen)
396 msg->msg_flags |= MSG_TRUNC;
398 is_udp4 = (skb->protocol == htons(ETH_P_IP));
401 * If checksum is needed at all, try to do it while copying the
402 * data. If the data is truncated, or if we only want a partial
403 * coverage checksum (UDP-Lite), do it before the copy.
406 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
407 if (udp_lib_checksum_complete(skb))
411 if (skb_csum_unnecessary(skb))
412 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
413 msg->msg_iov, copied);
415 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
420 trace_kfree_skb(skb, udpv6_recvmsg);
422 atomic_inc(&sk->sk_drops);
424 UDP_INC_STATS_USER(sock_net(sk),
428 UDP6_INC_STATS_USER(sock_net(sk),
436 UDP_INC_STATS_USER(sock_net(sk),
437 UDP_MIB_INDATAGRAMS, is_udplite);
439 UDP6_INC_STATS_USER(sock_net(sk),
440 UDP_MIB_INDATAGRAMS, is_udplite);
443 sock_recv_ts_and_drops(msg, sk, skb);
445 /* Copy the address. */
447 struct sockaddr_in6 *sin6;
449 sin6 = (struct sockaddr_in6 *) msg->msg_name;
450 sin6->sin6_family = AF_INET6;
451 sin6->sin6_port = udp_hdr(skb)->source;
452 sin6->sin6_flowinfo = 0;
453 sin6->sin6_scope_id = 0;
456 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
459 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
460 if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
461 sin6->sin6_scope_id = IP6CB(skb)->iif;
466 if (inet->cmsg_flags)
467 ip_cmsg_recv(msg, skb);
470 ip6_datagram_recv_ctl(sk, msg, skb);
474 if (flags & MSG_TRUNC)
478 skb_free_datagram_locked(sk, skb);
483 slow = lock_sock_fast(sk);
484 if (!skb_kill_datagram(sk, skb, flags)) {
486 UDP_INC_STATS_USER(sock_net(sk),
487 UDP_MIB_INERRORS, is_udplite);
489 UDP6_INC_STATS_USER(sock_net(sk),
490 UDP_MIB_INERRORS, is_udplite);
492 unlock_sock_fast(sk, slow);
497 /* starting over for a new packet */
498 msg->msg_flags &= ~MSG_TRUNC;
502 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
503 u8 type, u8 code, int offset, __be32 info,
504 struct udp_table *udptable)
506 struct ipv6_pinfo *np;
507 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
508 const struct in6_addr *saddr = &hdr->saddr;
509 const struct in6_addr *daddr = &hdr->daddr;
510 struct udphdr *uh = (struct udphdr*)(skb->data+offset);
514 sk = __udp6_lib_lookup(dev_net(skb->dev), daddr, uh->dest,
515 saddr, uh->source, inet6_iif(skb), udptable);
519 if (type == ICMPV6_PKT_TOOBIG)
520 ip6_sk_update_pmtu(skb, sk, info);
521 if (type == NDISC_REDIRECT)
522 ip6_sk_redirect(skb, sk);
526 if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
529 if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
533 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
536 sk->sk_error_report(sk);
541 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
545 if (!ipv6_addr_any(&inet6_sk(sk)->daddr))
546 sock_rps_save_rxhash(sk, skb);
548 rc = sock_queue_rcv_skb(sk, skb);
550 int is_udplite = IS_UDPLITE(sk);
552 /* Note that an ENOMEM error is charged twice */
554 UDP6_INC_STATS_BH(sock_net(sk),
555 UDP_MIB_RCVBUFERRORS, is_udplite);
556 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
563 static __inline__ void udpv6_err(struct sk_buff *skb,
564 struct inet6_skb_parm *opt, u8 type,
565 u8 code, int offset, __be32 info )
567 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
570 static struct static_key udpv6_encap_needed __read_mostly;
571 void udpv6_encap_enable(void)
573 if (!static_key_enabled(&udpv6_encap_needed))
574 static_key_slow_inc(&udpv6_encap_needed);
576 EXPORT_SYMBOL(udpv6_encap_enable);
578 int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
580 struct udp_sock *up = udp_sk(sk);
582 int is_udplite = IS_UDPLITE(sk);
584 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
587 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
588 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
591 * This is an encapsulation socket so pass the skb to
592 * the socket's udp_encap_rcv() hook. Otherwise, just
593 * fall through and pass this up the UDP socket.
594 * up->encap_rcv() returns the following value:
595 * =0 if skb was successfully passed to the encap
596 * handler or was discarded by it.
597 * >0 if skb should be passed on to UDP.
598 * <0 if skb should be resubmitted as proto -N
601 /* if we're overly short, let UDP handle it */
602 encap_rcv = ACCESS_ONCE(up->encap_rcv);
603 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
606 ret = encap_rcv(sk, skb);
608 UDP_INC_STATS_BH(sock_net(sk),
615 /* FALLTHROUGH -- it's a UDP Packet */
619 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
621 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
623 if (up->pcrlen == 0) { /* full coverage was set */
624 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage"
625 " %d while full coverage %d requested\n",
626 UDP_SKB_CB(skb)->cscov, skb->len);
629 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
630 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d "
631 "too small, need min %d\n",
632 UDP_SKB_CB(skb)->cscov, up->pcrlen);
637 if (rcu_access_pointer(sk->sk_filter)) {
638 if (udp_lib_checksum_complete(skb))
642 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf))
649 if (!sock_owned_by_user(sk))
650 rc = __udpv6_queue_rcv_skb(sk, skb);
651 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
659 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
660 atomic_inc(&sk->sk_drops);
665 static struct sock *udp_v6_mcast_next(struct net *net, struct sock *sk,
666 __be16 loc_port, const struct in6_addr *loc_addr,
667 __be16 rmt_port, const struct in6_addr *rmt_addr,
670 struct hlist_nulls_node *node;
672 unsigned short num = ntohs(loc_port);
674 sk_nulls_for_each_from(s, node) {
675 struct inet_sock *inet = inet_sk(s);
677 if (!net_eq(sock_net(s), net))
680 if (udp_sk(s)->udp_port_hash == num &&
681 s->sk_family == PF_INET6) {
682 struct ipv6_pinfo *np = inet6_sk(s);
683 if (inet->inet_dport) {
684 if (inet->inet_dport != rmt_port)
687 if (!ipv6_addr_any(&np->daddr) &&
688 !ipv6_addr_equal(&np->daddr, rmt_addr))
691 if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)
694 if (!ipv6_addr_any(&np->rcv_saddr)) {
695 if (!ipv6_addr_equal(&np->rcv_saddr, loc_addr))
698 if (!inet6_mc_check(s, loc_addr, rmt_addr))
706 static void flush_stack(struct sock **stack, unsigned int count,
707 struct sk_buff *skb, unsigned int final)
709 struct sk_buff *skb1 = NULL;
713 for (i = 0; i < count; i++) {
715 if (likely(skb1 == NULL))
716 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
718 atomic_inc(&sk->sk_drops);
719 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
721 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
725 if (skb1 && udpv6_queue_rcv_skb(sk, skb1) <= 0)
732 * Note: called only from the BH handler context,
733 * so we don't need to lock the hashes.
735 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
736 const struct in6_addr *saddr, const struct in6_addr *daddr,
737 struct udp_table *udptable)
739 struct sock *sk, *stack[256 / sizeof(struct sock *)];
740 const struct udphdr *uh = udp_hdr(skb);
741 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
743 unsigned int i, count = 0;
745 spin_lock(&hslot->lock);
746 sk = sk_nulls_head(&hslot->head);
747 dif = inet6_iif(skb);
748 sk = udp_v6_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
751 sk = udp_v6_mcast_next(net, sk_nulls_next(sk), uh->dest, daddr,
752 uh->source, saddr, dif);
753 if (unlikely(count == ARRAY_SIZE(stack))) {
756 flush_stack(stack, count, skb, ~0);
761 * before releasing the lock, we must take reference on sockets
763 for (i = 0; i < count; i++)
766 spin_unlock(&hslot->lock);
769 flush_stack(stack, count, skb, count - 1);
771 for (i = 0; i < count; i++)
779 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
782 struct net *net = dev_net(skb->dev);
785 const struct in6_addr *saddr, *daddr;
788 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
791 saddr = &ipv6_hdr(skb)->saddr;
792 daddr = &ipv6_hdr(skb)->daddr;
795 ulen = ntohs(uh->len);
799 if (proto == IPPROTO_UDP) {
800 /* UDP validates ulen. */
802 /* Check for jumbo payload */
806 if (ulen < sizeof(*uh))
809 if (ulen < skb->len) {
810 if (pskb_trim_rcsum(skb, ulen))
812 saddr = &ipv6_hdr(skb)->saddr;
813 daddr = &ipv6_hdr(skb)->daddr;
818 if (udp6_csum_init(skb, uh, proto))
822 * Multicast receive code
824 if (ipv6_addr_is_multicast(daddr))
825 return __udp6_lib_mcast_deliver(net, skb,
826 saddr, daddr, udptable);
831 * check socket cache ... must talk to Alan about his plans
832 * for sock caches... i'll skip this for now.
834 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
836 int ret = udpv6_queue_rcv_skb(sk, skb);
839 /* a return value > 0 means to resubmit the input, but
840 * it wants the return to be -protocol, or 0
848 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
851 if (udp_lib_checksum_complete(skb))
854 UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
855 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
861 LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
862 proto == IPPROTO_UDPLITE ? "-Lite" : "",
871 UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
876 static __inline__ int udpv6_rcv(struct sk_buff *skb)
878 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
882 * Throw away all pending data and cancel the corking. Socket is locked.
884 static void udp_v6_flush_pending_frames(struct sock *sk)
886 struct udp_sock *up = udp_sk(sk);
888 if (up->pending == AF_INET)
889 udp_flush_pending_frames(sk);
890 else if (up->pending) {
893 ip6_flush_pending_frames(sk);
898 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
899 * @sk: socket we are sending on
900 * @skb: sk_buff containing the filled-in UDP header
901 * (checksum field must be zeroed out)
903 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
904 const struct in6_addr *saddr,
905 const struct in6_addr *daddr, int len)
908 struct udphdr *uh = udp_hdr(skb);
911 if (skb_queue_len(&sk->sk_write_queue) == 1) {
912 /* Only one fragment on the socket. */
913 skb->csum_start = skb_transport_header(skb) - skb->head;
914 skb->csum_offset = offsetof(struct udphdr, check);
915 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
918 * HW-checksum won't work as there are two or more
919 * fragments on the socket so that all csums of sk_buffs
922 offset = skb_transport_offset(skb);
923 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
925 skb->ip_summed = CHECKSUM_NONE;
927 skb_queue_walk(&sk->sk_write_queue, skb) {
928 csum = csum_add(csum, skb->csum);
931 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
934 uh->check = CSUM_MANGLED_0;
942 static int udp_v6_push_pending_frames(struct sock *sk)
946 struct udp_sock *up = udp_sk(sk);
947 struct inet_sock *inet = inet_sk(sk);
948 struct flowi6 *fl6 = &inet->cork.fl.u.ip6;
950 int is_udplite = IS_UDPLITE(sk);
953 /* Grab the skbuff where UDP header space exists. */
954 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
958 * Create a UDP header
961 uh->source = fl6->fl6_sport;
962 uh->dest = fl6->fl6_dport;
963 uh->len = htons(up->len);
967 csum = udplite_csum_outgoing(sk, skb);
968 else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
969 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr,
973 csum = udp_csum_outgoing(sk, skb);
975 /* add protocol-dependent pseudo-header */
976 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
977 up->len, fl6->flowi6_proto, csum);
979 uh->check = CSUM_MANGLED_0;
982 err = ip6_push_pending_frames(sk);
984 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
985 UDP6_INC_STATS_USER(sock_net(sk),
986 UDP_MIB_SNDBUFERRORS, is_udplite);
990 UDP6_INC_STATS_USER(sock_net(sk),
991 UDP_MIB_OUTDATAGRAMS, is_udplite);
998 int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
999 struct msghdr *msg, size_t len)
1001 struct ipv6_txoptions opt_space;
1002 struct udp_sock *up = udp_sk(sk);
1003 struct inet_sock *inet = inet_sk(sk);
1004 struct ipv6_pinfo *np = inet6_sk(sk);
1005 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) msg->msg_name;
1006 struct in6_addr *daddr, *final_p, final;
1007 struct ipv6_txoptions *opt = NULL;
1008 struct ip6_flowlabel *flowlabel = NULL;
1010 struct dst_entry *dst;
1011 int addr_len = msg->msg_namelen;
1016 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1019 int is_udplite = IS_UDPLITE(sk);
1020 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1022 /* destination address check */
1024 if (addr_len < offsetof(struct sockaddr, sa_data))
1027 switch (sin6->sin6_family) {
1029 if (addr_len < SIN6_LEN_RFC2133)
1031 daddr = &sin6->sin6_addr;
1034 goto do_udp_sendmsg;
1036 msg->msg_name = sin6 = NULL;
1037 msg->msg_namelen = addr_len = 0;
1043 } else if (!up->pending) {
1044 if (sk->sk_state != TCP_ESTABLISHED)
1045 return -EDESTADDRREQ;
1051 if (ipv6_addr_v4mapped(daddr)) {
1052 struct sockaddr_in sin;
1053 sin.sin_family = AF_INET;
1054 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1055 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1056 msg->msg_name = &sin;
1057 msg->msg_namelen = sizeof(sin);
1059 if (__ipv6_only_sock(sk))
1060 return -ENETUNREACH;
1061 return udp_sendmsg(iocb, sk, msg, len);
1065 if (up->pending == AF_INET)
1066 return udp_sendmsg(iocb, sk, msg, len);
1068 /* Rough check on arithmetic overflow,
1069 better check is made in ip6_append_data().
1071 if (len > INT_MAX - sizeof(struct udphdr))
1076 * There are pending frames.
1077 * The socket lock must be held while it's corked.
1080 if (likely(up->pending)) {
1081 if (unlikely(up->pending != AF_INET6)) {
1083 return -EAFNOSUPPORT;
1086 goto do_append_data;
1090 ulen += sizeof(struct udphdr);
1092 memset(&fl6, 0, sizeof(fl6));
1095 if (sin6->sin6_port == 0)
1098 fl6.fl6_dport = sin6->sin6_port;
1099 daddr = &sin6->sin6_addr;
1102 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1103 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1104 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1105 if (flowlabel == NULL)
1107 daddr = &flowlabel->dst;
1112 * Otherwise it will be difficult to maintain
1115 if (sk->sk_state == TCP_ESTABLISHED &&
1116 ipv6_addr_equal(daddr, &np->daddr))
1119 if (addr_len >= sizeof(struct sockaddr_in6) &&
1120 sin6->sin6_scope_id &&
1121 ipv6_addr_type(daddr)&IPV6_ADDR_LINKLOCAL)
1122 fl6.flowi6_oif = sin6->sin6_scope_id;
1124 if (sk->sk_state != TCP_ESTABLISHED)
1125 return -EDESTADDRREQ;
1127 fl6.fl6_dport = inet->inet_dport;
1129 fl6.flowlabel = np->flow_label;
1133 if (!fl6.flowi6_oif)
1134 fl6.flowi6_oif = sk->sk_bound_dev_if;
1136 if (!fl6.flowi6_oif)
1137 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1139 fl6.flowi6_mark = sk->sk_mark;
1141 if (msg->msg_controllen) {
1143 memset(opt, 0, sizeof(struct ipv6_txoptions));
1144 opt->tot_len = sizeof(*opt);
1146 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt,
1147 &hlimit, &tclass, &dontfrag);
1149 fl6_sock_release(flowlabel);
1152 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1153 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1154 if (flowlabel == NULL)
1157 if (!(opt->opt_nflen|opt->opt_flen))
1164 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1165 opt = ipv6_fixup_options(&opt_space, opt);
1167 fl6.flowi6_proto = sk->sk_protocol;
1168 if (!ipv6_addr_any(daddr))
1171 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1172 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1173 fl6.saddr = np->saddr;
1174 fl6.fl6_sport = inet->inet_sport;
1176 final_p = fl6_update_dst(&fl6, opt, &final);
1180 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1181 fl6.flowi6_oif = np->mcast_oif;
1183 } else if (!fl6.flowi6_oif)
1184 fl6.flowi6_oif = np->ucast_oif;
1186 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1188 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, true);
1196 if (ipv6_addr_is_multicast(&fl6.daddr))
1197 hlimit = np->mcast_hops;
1199 hlimit = np->hop_limit;
1201 hlimit = ip6_dst_hoplimit(dst);
1205 tclass = np->tclass;
1208 dontfrag = np->dontfrag;
1210 if (msg->msg_flags&MSG_CONFIRM)
1215 if (unlikely(up->pending)) {
1216 /* The socket is already corked while preparing it. */
1217 /* ... which is an evident application bug. --ANK */
1220 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
1225 up->pending = AF_INET6;
1229 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1230 err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen,
1231 sizeof(struct udphdr), hlimit, tclass, opt, &fl6,
1232 (struct rt6_info*)dst,
1233 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag);
1235 udp_v6_flush_pending_frames(sk);
1237 err = udp_v6_push_pending_frames(sk);
1238 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1243 ip6_dst_store(sk, dst,
1244 ipv6_addr_equal(&fl6.daddr, &np->daddr) ?
1246 #ifdef CONFIG_IPV6_SUBTREES
1247 ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1258 err = np->recverr ? net_xmit_errno(err) : 0;
1262 fl6_sock_release(flowlabel);
1266 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1267 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1268 * we don't have a good statistic (IpOutDiscards but it can be too many
1269 * things). We could add another new stat but at least for now that
1270 * seems like overkill.
1272 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1273 UDP6_INC_STATS_USER(sock_net(sk),
1274 UDP_MIB_SNDBUFERRORS, is_udplite);
1280 if (!(msg->msg_flags&MSG_PROBE) || len)
1281 goto back_from_confirm;
1286 void udpv6_destroy_sock(struct sock *sk)
1289 udp_v6_flush_pending_frames(sk);
1292 inet6_destroy_sock(sk);
1296 * Socket option code for UDP
1298 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1299 char __user *optval, unsigned int optlen)
1301 if (level == SOL_UDP || level == SOL_UDPLITE)
1302 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1303 udp_v6_push_pending_frames);
1304 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1307 #ifdef CONFIG_COMPAT
1308 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1309 char __user *optval, unsigned int optlen)
1311 if (level == SOL_UDP || level == SOL_UDPLITE)
1312 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1313 udp_v6_push_pending_frames);
1314 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1318 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1319 char __user *optval, int __user *optlen)
1321 if (level == SOL_UDP || level == SOL_UDPLITE)
1322 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1323 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1326 #ifdef CONFIG_COMPAT
1327 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1328 char __user *optval, int __user *optlen)
1330 if (level == SOL_UDP || level == SOL_UDPLITE)
1331 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1332 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1336 static const struct inet6_protocol udpv6_protocol = {
1337 .handler = udpv6_rcv,
1338 .err_handler = udpv6_err,
1339 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1342 /* ------------------------------------------------------------------------ */
1343 #ifdef CONFIG_PROC_FS
1345 static void udp6_sock_seq_show(struct seq_file *seq, struct sock *sp, int bucket)
1347 struct inet_sock *inet = inet_sk(sp);
1348 struct ipv6_pinfo *np = inet6_sk(sp);
1349 const struct in6_addr *dest, *src;
1353 src = &np->rcv_saddr;
1354 destp = ntohs(inet->inet_dport);
1355 srcp = ntohs(inet->inet_sport);
1357 "%5d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1358 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d\n",
1360 src->s6_addr32[0], src->s6_addr32[1],
1361 src->s6_addr32[2], src->s6_addr32[3], srcp,
1362 dest->s6_addr32[0], dest->s6_addr32[1],
1363 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1365 sk_wmem_alloc_get(sp),
1366 sk_rmem_alloc_get(sp),
1368 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1371 atomic_read(&sp->sk_refcnt), sp,
1372 atomic_read(&sp->sk_drops));
1375 int udp6_seq_show(struct seq_file *seq, void *v)
1377 if (v == SEQ_START_TOKEN)
1382 "st tx_queue rx_queue tr tm->when retrnsmt"
1383 " uid timeout inode ref pointer drops\n");
1385 udp6_sock_seq_show(seq, v, ((struct udp_iter_state *)seq->private)->bucket);
1389 static const struct file_operations udp6_afinfo_seq_fops = {
1390 .owner = THIS_MODULE,
1391 .open = udp_seq_open,
1393 .llseek = seq_lseek,
1394 .release = seq_release_net
1397 static struct udp_seq_afinfo udp6_seq_afinfo = {
1400 .udp_table = &udp_table,
1401 .seq_fops = &udp6_afinfo_seq_fops,
1403 .show = udp6_seq_show,
1407 int __net_init udp6_proc_init(struct net *net)
1409 return udp_proc_register(net, &udp6_seq_afinfo);
1412 void udp6_proc_exit(struct net *net) {
1413 udp_proc_unregister(net, &udp6_seq_afinfo);
1415 #endif /* CONFIG_PROC_FS */
1417 /* ------------------------------------------------------------------------ */
1419 struct proto udpv6_prot = {
1421 .owner = THIS_MODULE,
1422 .close = udp_lib_close,
1423 .connect = ip6_datagram_connect,
1424 .disconnect = udp_disconnect,
1426 .destroy = udpv6_destroy_sock,
1427 .setsockopt = udpv6_setsockopt,
1428 .getsockopt = udpv6_getsockopt,
1429 .sendmsg = udpv6_sendmsg,
1430 .recvmsg = udpv6_recvmsg,
1431 .backlog_rcv = __udpv6_queue_rcv_skb,
1432 .hash = udp_lib_hash,
1433 .unhash = udp_lib_unhash,
1434 .rehash = udp_v6_rehash,
1435 .get_port = udp_v6_get_port,
1436 .memory_allocated = &udp_memory_allocated,
1437 .sysctl_mem = sysctl_udp_mem,
1438 .sysctl_wmem = &sysctl_udp_wmem_min,
1439 .sysctl_rmem = &sysctl_udp_rmem_min,
1440 .obj_size = sizeof(struct udp6_sock),
1441 .slab_flags = SLAB_DESTROY_BY_RCU,
1442 .h.udp_table = &udp_table,
1443 #ifdef CONFIG_COMPAT
1444 .compat_setsockopt = compat_udpv6_setsockopt,
1445 .compat_getsockopt = compat_udpv6_getsockopt,
1447 .clear_sk = sk_prot_clear_portaddr_nulls,
1450 static struct inet_protosw udpv6_protosw = {
1452 .protocol = IPPROTO_UDP,
1453 .prot = &udpv6_prot,
1454 .ops = &inet6_dgram_ops,
1455 .no_check = UDP_CSUM_DEFAULT,
1456 .flags = INET_PROTOSW_PERMANENT,
1460 int __init udpv6_init(void)
1464 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1468 ret = inet6_register_protosw(&udpv6_protosw);
1470 goto out_udpv6_protocol;
1475 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1479 void udpv6_exit(void)
1481 inet6_unregister_protosw(&udpv6_protosw);
1482 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);