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/addrconf.h>
41 #include <net/ndisc.h>
42 #include <net/protocol.h>
43 #include <net/transp_v6.h>
44 #include <net/ip6_route.h>
46 #include <net/tcp_states.h>
47 #include <net/ip6_checksum.h>
49 #include <net/inet6_hashtables.h>
50 #include <net/busy_poll.h>
51 #include <net/sock_reuseport.h>
53 #include <linux/proc_fs.h>
54 #include <linux/seq_file.h>
55 #include <trace/events/skb.h>
58 static u32 udp6_ehashfn(const struct net *net,
59 const struct in6_addr *laddr,
61 const struct in6_addr *faddr,
64 static u32 udp6_ehash_secret __read_mostly;
65 static u32 udp_ipv6_hash_secret __read_mostly;
69 net_get_random_once(&udp6_ehash_secret,
70 sizeof(udp6_ehash_secret));
71 net_get_random_once(&udp_ipv6_hash_secret,
72 sizeof(udp_ipv6_hash_secret));
74 lhash = (__force u32)laddr->s6_addr32[3];
75 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
77 return __inet6_ehashfn(lhash, lport, fhash, fport,
78 udp_ipv6_hash_secret + net_hash_mix(net));
81 static u32 udp6_portaddr_hash(const struct net *net,
82 const struct in6_addr *addr6,
85 unsigned int hash, mix = net_hash_mix(net);
87 if (ipv6_addr_any(addr6))
88 hash = jhash_1word(0, mix);
89 else if (ipv6_addr_v4mapped(addr6))
90 hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix);
92 hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix);
97 int udp_v6_get_port(struct sock *sk, unsigned short snum)
99 unsigned int hash2_nulladdr =
100 udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
101 unsigned int hash2_partial =
102 udp6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
104 /* precompute partial secondary hash */
105 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
106 return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal, hash2_nulladdr);
109 static void udp_v6_rehash(struct sock *sk)
111 u16 new_hash = udp6_portaddr_hash(sock_net(sk),
112 &sk->sk_v6_rcv_saddr,
113 inet_sk(sk)->inet_num);
115 udp_lib_rehash(sk, new_hash);
118 static int compute_score(struct sock *sk, struct net *net,
119 const struct in6_addr *saddr, __be16 sport,
120 const struct in6_addr *daddr, unsigned short hnum,
124 struct inet_sock *inet;
126 if (!net_eq(sock_net(sk), net) ||
127 udp_sk(sk)->udp_port_hash != hnum ||
128 sk->sk_family != PF_INET6)
134 if (inet->inet_dport) {
135 if (inet->inet_dport != sport)
140 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
141 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
146 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
147 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
152 if (sk->sk_bound_dev_if) {
153 if (sk->sk_bound_dev_if != dif)
158 if (sk->sk_incoming_cpu == raw_smp_processor_id())
164 /* called with rcu_read_lock() */
165 static struct sock *udp6_lib_lookup2(struct net *net,
166 const struct in6_addr *saddr, __be16 sport,
167 const struct in6_addr *daddr, unsigned int hnum, int dif,
168 struct udp_hslot *hslot2,
171 struct sock *sk, *result;
172 int score, badness, matches = 0, reuseport = 0;
177 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
178 score = compute_score(sk, net, saddr, sport,
180 if (score > badness) {
181 reuseport = sk->sk_reuseport;
183 hash = udp6_ehashfn(net, daddr, hnum,
186 result = reuseport_select_sock(sk, hash, skb,
187 sizeof(struct udphdr));
194 } else if (score == badness && reuseport) {
196 if (reciprocal_scale(hash, matches) == 0)
198 hash = next_pseudo_random32(hash);
204 /* rcu_read_lock() must be held */
205 struct sock *__udp6_lib_lookup(struct net *net,
206 const struct in6_addr *saddr, __be16 sport,
207 const struct in6_addr *daddr, __be16 dport,
208 int dif, struct udp_table *udptable,
211 struct sock *sk, *result;
212 unsigned short hnum = ntohs(dport);
213 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
214 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
215 int score, badness, matches = 0, reuseport = 0;
218 if (hslot->count > 10) {
219 hash2 = udp6_portaddr_hash(net, daddr, hnum);
220 slot2 = hash2 & udptable->mask;
221 hslot2 = &udptable->hash2[slot2];
222 if (hslot->count < hslot2->count)
225 result = udp6_lib_lookup2(net, saddr, sport,
229 unsigned int old_slot2 = slot2;
230 hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
231 slot2 = hash2 & udptable->mask;
232 /* avoid searching the same slot again. */
233 if (unlikely(slot2 == old_slot2))
236 hslot2 = &udptable->hash2[slot2];
237 if (hslot->count < hslot2->count)
240 result = udp6_lib_lookup2(net, saddr, sport,
249 sk_for_each_rcu(sk, &hslot->head) {
250 score = compute_score(sk, net, saddr, sport, daddr, hnum, dif);
251 if (score > badness) {
252 reuseport = sk->sk_reuseport;
254 hash = udp6_ehashfn(net, daddr, hnum,
256 result = reuseport_select_sock(sk, hash, skb,
257 sizeof(struct udphdr));
264 } else if (score == badness && reuseport) {
266 if (reciprocal_scale(hash, matches) == 0)
268 hash = next_pseudo_random32(hash);
273 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
275 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
276 __be16 sport, __be16 dport,
277 struct udp_table *udptable)
279 const struct ipv6hdr *iph = ipv6_hdr(skb);
282 sk = skb_steal_sock(skb);
285 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
286 &iph->daddr, dport, inet6_iif(skb),
290 struct sock *udp6_lib_lookup_skb(struct sk_buff *skb,
291 __be16 sport, __be16 dport)
293 const struct ipv6hdr *iph = ipv6_hdr(skb);
295 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
296 &iph->daddr, dport, inet6_iif(skb),
299 EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb);
301 /* Must be called under rcu_read_lock().
302 * Does increment socket refcount.
304 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
305 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY)
306 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
307 const struct in6_addr *daddr, __be16 dport, int dif)
311 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport,
312 dif, &udp_table, NULL);
313 if (sk && !atomic_inc_not_zero(&sk->sk_refcnt))
317 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
321 * This should be easy, if there is something there we
322 * return it, otherwise we block.
325 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
326 int noblock, int flags, int *addr_len)
328 struct ipv6_pinfo *np = inet6_sk(sk);
329 struct inet_sock *inet = inet_sk(sk);
331 unsigned int ulen, copied;
332 int peeked, peeking, off;
334 int is_udplite = IS_UDPLITE(sk);
335 bool checksum_valid = false;
338 if (flags & MSG_ERRQUEUE)
339 return ipv6_recv_error(sk, msg, len, addr_len);
341 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
342 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
345 peeking = off = sk_peek_offset(sk, flags);
346 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
352 if (copied > ulen - off)
354 else if (copied < ulen)
355 msg->msg_flags |= MSG_TRUNC;
357 is_udp4 = (skb->protocol == htons(ETH_P_IP));
360 * If checksum is needed at all, try to do it while copying the
361 * data. If the data is truncated, or if we only want a partial
362 * coverage checksum (UDP-Lite), do it before the copy.
365 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov || peeking) {
366 checksum_valid = !udp_lib_checksum_complete(skb);
371 if (checksum_valid || skb_csum_unnecessary(skb))
372 err = skb_copy_datagram_msg(skb, off, msg, copied);
374 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
380 atomic_inc(&sk->sk_drops);
382 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
385 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
393 UDP_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS,
396 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS,
400 sock_recv_ts_and_drops(msg, sk, skb);
402 /* Copy the address. */
404 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
405 sin6->sin6_family = AF_INET6;
406 sin6->sin6_port = udp_hdr(skb)->source;
407 sin6->sin6_flowinfo = 0;
410 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
412 sin6->sin6_scope_id = 0;
414 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
415 sin6->sin6_scope_id =
416 ipv6_iface_scope_id(&sin6->sin6_addr,
419 *addr_len = sizeof(*sin6);
423 ip6_datagram_recv_common_ctl(sk, msg, skb);
426 if (inet->cmsg_flags)
427 ip_cmsg_recv_offset(msg, sk, skb,
428 sizeof(struct udphdr), off);
431 ip6_datagram_recv_specific_ctl(sk, msg, skb);
435 if (flags & MSG_TRUNC)
438 skb_consume_udp(sk, skb, peeking ? -err : err);
442 if (!__sk_queue_drop_skb(sk, skb, flags)) {
444 UDP_INC_STATS(sock_net(sk),
445 UDP_MIB_CSUMERRORS, is_udplite);
446 UDP_INC_STATS(sock_net(sk),
447 UDP_MIB_INERRORS, is_udplite);
449 UDP6_INC_STATS(sock_net(sk),
450 UDP_MIB_CSUMERRORS, is_udplite);
451 UDP6_INC_STATS(sock_net(sk),
452 UDP_MIB_INERRORS, is_udplite);
457 /* starting over for a new packet, but check if we need to yield */
459 msg->msg_flags &= ~MSG_TRUNC;
463 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
464 u8 type, u8 code, int offset, __be32 info,
465 struct udp_table *udptable)
467 struct ipv6_pinfo *np;
468 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
469 const struct in6_addr *saddr = &hdr->saddr;
470 const struct in6_addr *daddr = &hdr->daddr;
471 struct udphdr *uh = (struct udphdr *)(skb->data+offset);
475 struct net *net = dev_net(skb->dev);
477 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
478 inet6_iif(skb), udptable, skb);
480 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
485 harderr = icmpv6_err_convert(type, code, &err);
488 if (type == ICMPV6_PKT_TOOBIG) {
489 if (!ip6_sk_accept_pmtu(sk))
491 ip6_sk_update_pmtu(skb, sk, info);
492 if (np->pmtudisc != IPV6_PMTUDISC_DONT)
495 if (type == NDISC_REDIRECT) {
496 ip6_sk_redirect(skb, sk);
501 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
504 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
508 sk->sk_error_report(sk);
513 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
517 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
518 sock_rps_save_rxhash(sk, skb);
519 sk_mark_napi_id(sk, skb);
520 sk_incoming_cpu_update(sk);
523 rc = __udp_enqueue_schedule_skb(sk, skb);
525 int is_udplite = IS_UDPLITE(sk);
527 /* Note that an ENOMEM error is charged twice */
529 UDP6_INC_STATS(sock_net(sk),
530 UDP_MIB_RCVBUFERRORS, is_udplite);
531 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
539 static __inline__ void udpv6_err(struct sk_buff *skb,
540 struct inet6_skb_parm *opt, u8 type,
541 u8 code, int offset, __be32 info)
543 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
546 static struct static_key udpv6_encap_needed __read_mostly;
547 void udpv6_encap_enable(void)
549 if (!static_key_enabled(&udpv6_encap_needed))
550 static_key_slow_inc(&udpv6_encap_needed);
552 EXPORT_SYMBOL(udpv6_encap_enable);
554 int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
556 struct udp_sock *up = udp_sk(sk);
557 int is_udplite = IS_UDPLITE(sk);
559 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
562 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
563 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
566 * This is an encapsulation socket so pass the skb to
567 * the socket's udp_encap_rcv() hook. Otherwise, just
568 * fall through and pass this up the UDP socket.
569 * up->encap_rcv() returns the following value:
570 * =0 if skb was successfully passed to the encap
571 * handler or was discarded by it.
572 * >0 if skb should be passed on to UDP.
573 * <0 if skb should be resubmitted as proto -N
576 /* if we're overly short, let UDP handle it */
577 encap_rcv = ACCESS_ONCE(up->encap_rcv);
581 /* Verify checksum before giving to encap */
582 if (udp_lib_checksum_complete(skb))
585 ret = encap_rcv(sk, skb);
587 __UDP_INC_STATS(sock_net(sk),
594 /* FALLTHROUGH -- it's a UDP Packet */
598 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
600 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
602 if (up->pcrlen == 0) { /* full coverage was set */
603 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
604 UDP_SKB_CB(skb)->cscov, skb->len);
607 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
608 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
609 UDP_SKB_CB(skb)->cscov, up->pcrlen);
614 if (rcu_access_pointer(sk->sk_filter) &&
615 udp_lib_checksum_complete(skb))
618 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
621 udp_csum_pull_header(skb);
625 return __udpv6_queue_rcv_skb(sk, skb);
628 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
630 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
631 atomic_inc(&sk->sk_drops);
636 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
637 __be16 loc_port, const struct in6_addr *loc_addr,
638 __be16 rmt_port, const struct in6_addr *rmt_addr,
639 int dif, unsigned short hnum)
641 struct inet_sock *inet = inet_sk(sk);
643 if (!net_eq(sock_net(sk), net))
646 if (udp_sk(sk)->udp_port_hash != hnum ||
647 sk->sk_family != PF_INET6 ||
648 (inet->inet_dport && inet->inet_dport != rmt_port) ||
649 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
650 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
651 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif) ||
652 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
653 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
655 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
660 static void udp6_csum_zero_error(struct sk_buff *skb)
662 /* RFC 2460 section 8.1 says that we SHOULD log
663 * this error. Well, it is reasonable.
665 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
666 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
667 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
671 * Note: called only from the BH handler context,
672 * so we don't need to lock the hashes.
674 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
675 const struct in6_addr *saddr, const struct in6_addr *daddr,
676 struct udp_table *udptable, int proto)
678 struct sock *sk, *first = NULL;
679 const struct udphdr *uh = udp_hdr(skb);
680 unsigned short hnum = ntohs(uh->dest);
681 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
682 unsigned int offset = offsetof(typeof(*sk), sk_node);
683 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
684 int dif = inet6_iif(skb);
685 struct hlist_node *node;
686 struct sk_buff *nskb;
689 hash2_any = udp6_portaddr_hash(net, &in6addr_any, hnum) &
691 hash2 = udp6_portaddr_hash(net, daddr, hnum) & udp_table.mask;
693 hslot = &udp_table.hash2[hash2];
694 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
697 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
698 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
699 uh->source, saddr, dif, hnum))
701 /* If zero checksum and no_check is not on for
702 * the socket then skip it.
704 if (!uh->check && !udp_sk(sk)->no_check6_rx)
710 nskb = skb_clone(skb, GFP_ATOMIC);
711 if (unlikely(!nskb)) {
712 atomic_inc(&sk->sk_drops);
713 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
715 __UDP6_INC_STATS(net, UDP_MIB_INERRORS,
720 if (udpv6_queue_rcv_skb(sk, nskb) > 0)
724 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
725 if (use_hash2 && hash2 != hash2_any) {
731 if (udpv6_queue_rcv_skb(first, skb) > 0)
735 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
736 proto == IPPROTO_UDPLITE);
741 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
744 const struct in6_addr *saddr, *daddr;
745 struct net *net = dev_net(skb->dev);
750 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
753 saddr = &ipv6_hdr(skb)->saddr;
754 daddr = &ipv6_hdr(skb)->daddr;
757 ulen = ntohs(uh->len);
761 if (proto == IPPROTO_UDP) {
762 /* UDP validates ulen. */
764 /* Check for jumbo payload */
768 if (ulen < sizeof(*uh))
771 if (ulen < skb->len) {
772 if (pskb_trim_rcsum(skb, ulen))
774 saddr = &ipv6_hdr(skb)->saddr;
775 daddr = &ipv6_hdr(skb)->daddr;
780 if (udp6_csum_init(skb, uh, proto))
784 * Multicast receive code
786 if (ipv6_addr_is_multicast(daddr))
787 return __udp6_lib_mcast_deliver(net, skb,
788 saddr, daddr, udptable, proto);
793 * check socket cache ... must talk to Alan about his plans
794 * for sock caches... i'll skip this for now.
796 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
800 if (!uh->check && !udp_sk(sk)->no_check6_rx) {
801 udp6_csum_zero_error(skb);
805 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
806 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
809 ret = udpv6_queue_rcv_skb(sk, skb);
811 /* a return value > 0 means to resubmit the input */
819 udp6_csum_zero_error(skb);
823 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
826 if (udp_lib_checksum_complete(skb))
829 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
830 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
836 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
837 proto == IPPROTO_UDPLITE ? "-Lite" : "",
838 saddr, ntohs(uh->source),
840 daddr, ntohs(uh->dest));
843 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
845 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
850 static __inline__ int udpv6_rcv(struct sk_buff *skb)
852 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
856 * Throw away all pending data and cancel the corking. Socket is locked.
858 static void udp_v6_flush_pending_frames(struct sock *sk)
860 struct udp_sock *up = udp_sk(sk);
862 if (up->pending == AF_INET)
863 udp_flush_pending_frames(sk);
864 else if (up->pending) {
867 ip6_flush_pending_frames(sk);
872 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
873 * @sk: socket we are sending on
874 * @skb: sk_buff containing the filled-in UDP header
875 * (checksum field must be zeroed out)
877 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
878 const struct in6_addr *saddr,
879 const struct in6_addr *daddr, int len)
882 struct udphdr *uh = udp_hdr(skb);
883 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
887 /* Only one fragment on the socket. */
888 skb->csum_start = skb_transport_header(skb) - skb->head;
889 skb->csum_offset = offsetof(struct udphdr, check);
890 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
893 * HW-checksum won't work as there are two or more
894 * fragments on the socket so that all csums of sk_buffs
897 offset = skb_transport_offset(skb);
898 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
900 skb->ip_summed = CHECKSUM_NONE;
903 csum = csum_add(csum, frags->csum);
904 } while ((frags = frags->next));
906 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
909 uh->check = CSUM_MANGLED_0;
917 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6)
919 struct sock *sk = skb->sk;
922 int is_udplite = IS_UDPLITE(sk);
924 int offset = skb_transport_offset(skb);
925 int len = skb->len - offset;
928 * Create a UDP header
931 uh->source = fl6->fl6_sport;
932 uh->dest = fl6->fl6_dport;
933 uh->len = htons(len);
937 csum = udplite_csum(skb);
938 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */
939 skb->ip_summed = CHECKSUM_NONE;
941 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
942 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
945 csum = udp_csum(skb);
947 /* add protocol-dependent pseudo-header */
948 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
949 len, fl6->flowi6_proto, csum);
951 uh->check = CSUM_MANGLED_0;
954 err = ip6_send_skb(skb);
956 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
957 UDP6_INC_STATS(sock_net(sk),
958 UDP_MIB_SNDBUFERRORS, is_udplite);
962 UDP6_INC_STATS(sock_net(sk),
963 UDP_MIB_OUTDATAGRAMS, is_udplite);
968 static int udp_v6_push_pending_frames(struct sock *sk)
971 struct udp_sock *up = udp_sk(sk);
975 if (up->pending == AF_INET)
976 return udp_push_pending_frames(sk);
978 /* ip6_finish_skb will release the cork, so make a copy of
981 fl6 = inet_sk(sk)->cork.fl.u.ip6;
983 skb = ip6_finish_skb(sk);
987 err = udp_v6_send_skb(skb, &fl6);
995 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
997 struct ipv6_txoptions opt_space;
998 struct udp_sock *up = udp_sk(sk);
999 struct inet_sock *inet = inet_sk(sk);
1000 struct ipv6_pinfo *np = inet6_sk(sk);
1001 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1002 struct in6_addr *daddr, *final_p, final;
1003 struct ipv6_txoptions *opt = NULL;
1004 struct ipv6_txoptions *opt_to_free = NULL;
1005 struct ip6_flowlabel *flowlabel = NULL;
1007 struct dst_entry *dst;
1008 struct ipcm6_cookie ipc6;
1009 int addr_len = msg->msg_namelen;
1011 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1014 int is_udplite = IS_UDPLITE(sk);
1015 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1016 struct sockcm_cookie sockc;
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;
1046 daddr = &sk->sk_v6_daddr;
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(sk, msg, len);
1065 if (up->pending == AF_INET)
1066 return udp_sendmsg(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))
1074 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1077 * There are pending frames.
1078 * The socket lock must be held while it's corked.
1081 if (likely(up->pending)) {
1082 if (unlikely(up->pending != AF_INET6)) {
1084 return -EAFNOSUPPORT;
1087 goto do_append_data;
1091 ulen += sizeof(struct udphdr);
1093 memset(&fl6, 0, sizeof(fl6));
1096 if (sin6->sin6_port == 0)
1099 fl6.fl6_dport = sin6->sin6_port;
1100 daddr = &sin6->sin6_addr;
1103 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1104 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1105 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1112 * Otherwise it will be difficult to maintain
1115 if (sk->sk_state == TCP_ESTABLISHED &&
1116 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1117 daddr = &sk->sk_v6_daddr;
1119 if (addr_len >= sizeof(struct sockaddr_in6) &&
1120 sin6->sin6_scope_id &&
1121 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
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;
1128 daddr = &sk->sk_v6_daddr;
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;
1140 fl6.flowi6_uid = sk->sk_uid;
1141 sockc.tsflags = sk->sk_tsflags;
1143 if (msg->msg_controllen) {
1145 memset(opt, 0, sizeof(struct ipv6_txoptions));
1146 opt->tot_len = sizeof(*opt);
1149 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, &ipc6, &sockc);
1151 fl6_sock_release(flowlabel);
1154 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1155 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1159 if (!(opt->opt_nflen|opt->opt_flen))
1164 opt = txopt_get(np);
1168 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1169 opt = ipv6_fixup_options(&opt_space, opt);
1172 fl6.flowi6_proto = sk->sk_protocol;
1173 if (!ipv6_addr_any(daddr))
1176 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1177 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1178 fl6.saddr = np->saddr;
1179 fl6.fl6_sport = inet->inet_sport;
1181 final_p = fl6_update_dst(&fl6, opt, &final);
1185 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1186 fl6.flowi6_oif = np->mcast_oif;
1188 } else if (!fl6.flowi6_oif)
1189 fl6.flowi6_oif = np->ucast_oif;
1191 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1193 if (ipc6.tclass < 0)
1194 ipc6.tclass = np->tclass;
1196 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel);
1198 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p);
1205 if (ipc6.hlimit < 0)
1206 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1208 if (msg->msg_flags&MSG_CONFIRM)
1212 /* Lockless fast path for the non-corking case */
1214 struct sk_buff *skb;
1216 skb = ip6_make_skb(sk, getfrag, msg, ulen,
1217 sizeof(struct udphdr), &ipc6,
1218 &fl6, (struct rt6_info *)dst,
1219 msg->msg_flags, &sockc);
1221 if (!IS_ERR_OR_NULL(skb))
1222 err = udp_v6_send_skb(skb, &fl6);
1227 if (unlikely(up->pending)) {
1228 /* The socket is already corked while preparing it. */
1229 /* ... which is an evident application bug. --ANK */
1232 net_dbg_ratelimited("udp cork app bug 2\n");
1237 up->pending = AF_INET6;
1240 if (ipc6.dontfrag < 0)
1241 ipc6.dontfrag = np->dontfrag;
1243 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr),
1244 &ipc6, &fl6, (struct rt6_info *)dst,
1245 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, &sockc);
1247 udp_v6_flush_pending_frames(sk);
1249 err = udp_v6_push_pending_frames(sk);
1250 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1254 err = np->recverr ? net_xmit_errno(err) : 0;
1260 ip6_dst_store(sk, dst,
1261 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ?
1262 &sk->sk_v6_daddr : NULL,
1263 #ifdef CONFIG_IPV6_SUBTREES
1264 ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1276 fl6_sock_release(flowlabel);
1277 txopt_put(opt_to_free);
1281 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1282 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1283 * we don't have a good statistic (IpOutDiscards but it can be too many
1284 * things). We could add another new stat but at least for now that
1285 * seems like overkill.
1287 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1288 UDP6_INC_STATS(sock_net(sk),
1289 UDP_MIB_SNDBUFERRORS, is_udplite);
1295 if (!(msg->msg_flags&MSG_PROBE) || len)
1296 goto back_from_confirm;
1301 void udpv6_destroy_sock(struct sock *sk)
1303 struct udp_sock *up = udp_sk(sk);
1305 udp_v6_flush_pending_frames(sk);
1308 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
1309 void (*encap_destroy)(struct sock *sk);
1310 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1315 inet6_destroy_sock(sk);
1319 * Socket option code for UDP
1321 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1322 char __user *optval, unsigned int optlen)
1324 if (level == SOL_UDP || level == SOL_UDPLITE)
1325 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1326 udp_v6_push_pending_frames);
1327 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1330 #ifdef CONFIG_COMPAT
1331 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1332 char __user *optval, unsigned int optlen)
1334 if (level == SOL_UDP || level == SOL_UDPLITE)
1335 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1336 udp_v6_push_pending_frames);
1337 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1341 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1342 char __user *optval, int __user *optlen)
1344 if (level == SOL_UDP || level == SOL_UDPLITE)
1345 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1346 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1349 #ifdef CONFIG_COMPAT
1350 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1351 char __user *optval, int __user *optlen)
1353 if (level == SOL_UDP || level == SOL_UDPLITE)
1354 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1355 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1359 static const struct inet6_protocol udpv6_protocol = {
1360 .handler = udpv6_rcv,
1361 .err_handler = udpv6_err,
1362 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1365 /* ------------------------------------------------------------------------ */
1366 #ifdef CONFIG_PROC_FS
1367 int udp6_seq_show(struct seq_file *seq, void *v)
1369 if (v == SEQ_START_TOKEN) {
1370 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1372 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1373 struct inet_sock *inet = inet_sk(v);
1374 __u16 srcp = ntohs(inet->inet_sport);
1375 __u16 destp = ntohs(inet->inet_dport);
1376 ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket);
1381 static const struct file_operations udp6_afinfo_seq_fops = {
1382 .owner = THIS_MODULE,
1383 .open = udp_seq_open,
1385 .llseek = seq_lseek,
1386 .release = seq_release_net
1389 static struct udp_seq_afinfo udp6_seq_afinfo = {
1392 .udp_table = &udp_table,
1393 .seq_fops = &udp6_afinfo_seq_fops,
1395 .show = udp6_seq_show,
1399 int __net_init udp6_proc_init(struct net *net)
1401 return udp_proc_register(net, &udp6_seq_afinfo);
1404 void udp6_proc_exit(struct net *net)
1406 udp_proc_unregister(net, &udp6_seq_afinfo);
1408 #endif /* CONFIG_PROC_FS */
1410 /* ------------------------------------------------------------------------ */
1412 struct proto udpv6_prot = {
1414 .owner = THIS_MODULE,
1415 .close = udp_lib_close,
1416 .connect = ip6_datagram_connect,
1417 .disconnect = udp_disconnect,
1419 .init = udp_init_sock,
1420 .destroy = udpv6_destroy_sock,
1421 .setsockopt = udpv6_setsockopt,
1422 .getsockopt = udpv6_getsockopt,
1423 .sendmsg = udpv6_sendmsg,
1424 .recvmsg = udpv6_recvmsg,
1425 .release_cb = ip6_datagram_release_cb,
1426 .hash = udp_lib_hash,
1427 .unhash = udp_lib_unhash,
1428 .rehash = udp_v6_rehash,
1429 .get_port = udp_v6_get_port,
1430 .memory_allocated = &udp_memory_allocated,
1431 .sysctl_mem = sysctl_udp_mem,
1432 .sysctl_wmem = &sysctl_udp_wmem_min,
1433 .sysctl_rmem = &sysctl_udp_rmem_min,
1434 .obj_size = sizeof(struct udp6_sock),
1435 .h.udp_table = &udp_table,
1436 #ifdef CONFIG_COMPAT
1437 .compat_setsockopt = compat_udpv6_setsockopt,
1438 .compat_getsockopt = compat_udpv6_getsockopt,
1440 .diag_destroy = udp_abort,
1443 static struct inet_protosw udpv6_protosw = {
1445 .protocol = IPPROTO_UDP,
1446 .prot = &udpv6_prot,
1447 .ops = &inet6_dgram_ops,
1448 .flags = INET_PROTOSW_PERMANENT,
1451 int __init udpv6_init(void)
1455 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1459 ret = inet6_register_protosw(&udpv6_protosw);
1461 goto out_udpv6_protocol;
1466 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1470 void udpv6_exit(void)
1472 inet6_unregister_protosw(&udpv6_protosw);
1473 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);