3 * Linux INET6 implementation
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>
49 #include <net/busy_poll.h>
51 #include <linux/proc_fs.h>
52 #include <linux/seq_file.h>
53 #include <trace/events/skb.h>
56 static u32 udp6_ehashfn(const struct net *net,
57 const struct in6_addr *laddr,
59 const struct in6_addr *faddr,
62 static u32 udp6_ehash_secret __read_mostly;
63 static u32 udp_ipv6_hash_secret __read_mostly;
67 net_get_random_once(&udp6_ehash_secret,
68 sizeof(udp6_ehash_secret));
69 net_get_random_once(&udp_ipv6_hash_secret,
70 sizeof(udp_ipv6_hash_secret));
72 lhash = (__force u32)laddr->s6_addr32[3];
73 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
75 return __inet6_ehashfn(lhash, lport, fhash, fport,
76 udp_ipv6_hash_secret + net_hash_mix(net));
79 int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2)
81 const struct in6_addr *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2);
82 int sk2_ipv6only = inet_v6_ipv6only(sk2);
83 int addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
84 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
86 /* if both are mapped, treat as IPv4 */
87 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED)
88 return (!sk2_ipv6only &&
89 (!sk->sk_rcv_saddr || !sk2->sk_rcv_saddr ||
90 sk->sk_rcv_saddr == sk2->sk_rcv_saddr));
92 if (addr_type2 == IPV6_ADDR_ANY &&
93 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
96 if (addr_type == IPV6_ADDR_ANY &&
97 !(ipv6_only_sock(sk) && addr_type2 == IPV6_ADDR_MAPPED))
100 if (sk2_rcv_saddr6 &&
101 ipv6_addr_equal(&sk->sk_v6_rcv_saddr, sk2_rcv_saddr6))
107 static u32 udp6_portaddr_hash(const struct net *net,
108 const struct in6_addr *addr6,
111 unsigned int hash, mix = net_hash_mix(net);
113 if (ipv6_addr_any(addr6))
114 hash = jhash_1word(0, mix);
115 else if (ipv6_addr_v4mapped(addr6))
116 hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix);
118 hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix);
123 int udp_v6_get_port(struct sock *sk, unsigned short snum)
125 unsigned int hash2_nulladdr =
126 udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
127 unsigned int hash2_partial =
128 udp6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
130 /* precompute partial secondary hash */
131 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
132 return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal, hash2_nulladdr);
135 static void udp_v6_rehash(struct sock *sk)
137 u16 new_hash = udp6_portaddr_hash(sock_net(sk),
138 &sk->sk_v6_rcv_saddr,
139 inet_sk(sk)->inet_num);
141 udp_lib_rehash(sk, new_hash);
144 static inline int compute_score(struct sock *sk, struct net *net,
146 const struct in6_addr *saddr, __be16 sport,
147 const struct in6_addr *daddr, __be16 dport,
151 struct inet_sock *inet;
153 if (!net_eq(sock_net(sk), net) ||
154 udp_sk(sk)->udp_port_hash != hnum ||
155 sk->sk_family != PF_INET6)
161 if (inet->inet_dport) {
162 if (inet->inet_dport != sport)
167 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
168 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
173 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
174 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
179 if (sk->sk_bound_dev_if) {
180 if (sk->sk_bound_dev_if != dif)
188 #define SCORE2_MAX (1 + 1 + 1)
189 static inline int compute_score2(struct sock *sk, struct net *net,
190 const struct in6_addr *saddr, __be16 sport,
191 const struct in6_addr *daddr,
192 unsigned short hnum, int dif)
195 struct inet_sock *inet;
197 if (!net_eq(sock_net(sk), net) ||
198 udp_sk(sk)->udp_port_hash != hnum ||
199 sk->sk_family != PF_INET6)
202 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
208 if (inet->inet_dport) {
209 if (inet->inet_dport != sport)
214 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
215 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
220 if (sk->sk_bound_dev_if) {
221 if (sk->sk_bound_dev_if != dif)
229 /* called with read_rcu_lock() */
230 static struct sock *udp6_lib_lookup2(struct net *net,
231 const struct in6_addr *saddr, __be16 sport,
232 const struct in6_addr *daddr, unsigned int hnum, int dif,
233 struct udp_hslot *hslot2, unsigned int slot2)
235 struct sock *sk, *result;
236 struct hlist_nulls_node *node;
237 int score, badness, matches = 0, reuseport = 0;
243 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
244 score = compute_score2(sk, net, saddr, sport,
246 if (score > badness) {
249 reuseport = sk->sk_reuseport;
251 hash = udp6_ehashfn(net, daddr, hnum,
254 } else if (score == SCORE2_MAX)
256 } else if (score == badness && reuseport) {
258 if (reciprocal_scale(hash, matches) == 0)
260 hash = next_pseudo_random32(hash);
264 * if the nulls value we got at the end of this lookup is
265 * not the expected one, we must restart lookup.
266 * We probably met an item that was moved to another chain.
268 if (get_nulls_value(node) != slot2)
273 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
275 else if (unlikely(compute_score2(result, net, saddr, sport,
276 daddr, hnum, dif) < badness)) {
284 struct sock *__udp6_lib_lookup(struct net *net,
285 const struct in6_addr *saddr, __be16 sport,
286 const struct in6_addr *daddr, __be16 dport,
287 int dif, struct udp_table *udptable)
289 struct sock *sk, *result;
290 struct hlist_nulls_node *node;
291 unsigned short hnum = ntohs(dport);
292 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
293 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
294 int score, badness, matches = 0, reuseport = 0;
298 if (hslot->count > 10) {
299 hash2 = udp6_portaddr_hash(net, daddr, hnum);
300 slot2 = hash2 & udptable->mask;
301 hslot2 = &udptable->hash2[slot2];
302 if (hslot->count < hslot2->count)
305 result = udp6_lib_lookup2(net, saddr, sport,
309 hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
310 slot2 = hash2 & udptable->mask;
311 hslot2 = &udptable->hash2[slot2];
312 if (hslot->count < hslot2->count)
315 result = udp6_lib_lookup2(net, saddr, sport,
316 &in6addr_any, hnum, dif,
325 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
326 score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif);
327 if (score > badness) {
330 reuseport = sk->sk_reuseport;
332 hash = udp6_ehashfn(net, daddr, hnum,
336 } else if (score == badness && reuseport) {
338 if (reciprocal_scale(hash, matches) == 0)
340 hash = next_pseudo_random32(hash);
344 * if the nulls value we got at the end of this lookup is
345 * not the expected one, we must restart lookup.
346 * We probably met an item that was moved to another chain.
348 if (get_nulls_value(node) != slot)
352 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
354 else if (unlikely(compute_score(result, net, hnum, saddr, sport,
355 daddr, dport, dif) < badness)) {
363 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
365 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
366 __be16 sport, __be16 dport,
367 struct udp_table *udptable)
370 const struct ipv6hdr *iph = ipv6_hdr(skb);
372 sk = skb_steal_sock(skb);
375 return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport,
376 &iph->daddr, dport, inet6_iif(skb),
380 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
381 const struct in6_addr *daddr, __be16 dport, int dif)
383 return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
385 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
388 * This should be easy, if there is something there we
389 * return it, otherwise we block.
392 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
393 int noblock, int flags, int *addr_len)
395 struct ipv6_pinfo *np = inet6_sk(sk);
396 struct inet_sock *inet = inet_sk(sk);
398 unsigned int ulen, copied;
401 int is_udplite = IS_UDPLITE(sk);
405 if (flags & MSG_ERRQUEUE)
406 return ipv6_recv_error(sk, msg, len, addr_len);
408 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
409 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
412 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
413 &peeked, &off, &err);
417 ulen = skb->len - sizeof(struct udphdr);
421 else if (copied < ulen)
422 msg->msg_flags |= MSG_TRUNC;
424 is_udp4 = (skb->protocol == htons(ETH_P_IP));
427 * If checksum is needed at all, try to do it while copying the
428 * data. If the data is truncated, or if we only want a partial
429 * coverage checksum (UDP-Lite), do it before the copy.
432 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
433 if (udp_lib_checksum_complete(skb))
437 if (skb_csum_unnecessary(skb))
438 err = skb_copy_datagram_msg(skb, sizeof(struct udphdr),
441 err = skb_copy_and_csum_datagram_msg(skb, sizeof(struct udphdr), msg);
446 trace_kfree_skb(skb, udpv6_recvmsg);
448 atomic_inc(&sk->sk_drops);
450 UDP_INC_STATS_USER(sock_net(sk),
454 UDP6_INC_STATS_USER(sock_net(sk),
462 UDP_INC_STATS_USER(sock_net(sk),
463 UDP_MIB_INDATAGRAMS, is_udplite);
465 UDP6_INC_STATS_USER(sock_net(sk),
466 UDP_MIB_INDATAGRAMS, is_udplite);
469 sock_recv_ts_and_drops(msg, sk, skb);
471 /* Copy the address. */
473 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
474 sin6->sin6_family = AF_INET6;
475 sin6->sin6_port = udp_hdr(skb)->source;
476 sin6->sin6_flowinfo = 0;
479 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
481 sin6->sin6_scope_id = 0;
483 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
484 sin6->sin6_scope_id =
485 ipv6_iface_scope_id(&sin6->sin6_addr,
488 *addr_len = sizeof(*sin6);
492 ip6_datagram_recv_common_ctl(sk, msg, skb);
495 if (inet->cmsg_flags)
496 ip_cmsg_recv(msg, skb);
499 ip6_datagram_recv_specific_ctl(sk, msg, skb);
503 if (flags & MSG_TRUNC)
507 skb_free_datagram_locked(sk, skb);
512 slow = lock_sock_fast(sk);
513 if (!skb_kill_datagram(sk, skb, flags)) {
515 UDP_INC_STATS_USER(sock_net(sk),
516 UDP_MIB_CSUMERRORS, is_udplite);
517 UDP_INC_STATS_USER(sock_net(sk),
518 UDP_MIB_INERRORS, is_udplite);
520 UDP6_INC_STATS_USER(sock_net(sk),
521 UDP_MIB_CSUMERRORS, is_udplite);
522 UDP6_INC_STATS_USER(sock_net(sk),
523 UDP_MIB_INERRORS, is_udplite);
526 unlock_sock_fast(sk, slow);
528 /* starting over for a new packet, but check if we need to yield */
530 msg->msg_flags &= ~MSG_TRUNC;
534 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
535 u8 type, u8 code, int offset, __be32 info,
536 struct udp_table *udptable)
538 struct ipv6_pinfo *np;
539 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
540 const struct in6_addr *saddr = &hdr->saddr;
541 const struct in6_addr *daddr = &hdr->daddr;
542 struct udphdr *uh = (struct udphdr *)(skb->data+offset);
545 struct net *net = dev_net(skb->dev);
547 sk = __udp6_lib_lookup(net, daddr, uh->dest,
548 saddr, uh->source, inet6_iif(skb), udptable);
550 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
555 if (type == ICMPV6_PKT_TOOBIG) {
556 if (!ip6_sk_accept_pmtu(sk))
558 ip6_sk_update_pmtu(skb, sk, info);
560 if (type == NDISC_REDIRECT) {
561 ip6_sk_redirect(skb, sk);
567 if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
570 if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
574 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
577 sk->sk_error_report(sk);
582 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
586 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
587 sock_rps_save_rxhash(sk, skb);
588 sk_mark_napi_id(sk, skb);
589 sk_incoming_cpu_update(sk);
592 rc = sock_queue_rcv_skb(sk, skb);
594 int is_udplite = IS_UDPLITE(sk);
596 /* Note that an ENOMEM error is charged twice */
598 UDP6_INC_STATS_BH(sock_net(sk),
599 UDP_MIB_RCVBUFERRORS, is_udplite);
600 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
607 static __inline__ void udpv6_err(struct sk_buff *skb,
608 struct inet6_skb_parm *opt, u8 type,
609 u8 code, int offset, __be32 info)
611 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
614 static struct static_key udpv6_encap_needed __read_mostly;
615 void udpv6_encap_enable(void)
617 if (!static_key_enabled(&udpv6_encap_needed))
618 static_key_slow_inc(&udpv6_encap_needed);
620 EXPORT_SYMBOL(udpv6_encap_enable);
622 int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
624 struct udp_sock *up = udp_sk(sk);
626 int is_udplite = IS_UDPLITE(sk);
628 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
631 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
632 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
635 * This is an encapsulation socket so pass the skb to
636 * the socket's udp_encap_rcv() hook. Otherwise, just
637 * fall through and pass this up the UDP socket.
638 * up->encap_rcv() returns the following value:
639 * =0 if skb was successfully passed to the encap
640 * handler or was discarded by it.
641 * >0 if skb should be passed on to UDP.
642 * <0 if skb should be resubmitted as proto -N
645 /* if we're overly short, let UDP handle it */
646 encap_rcv = ACCESS_ONCE(up->encap_rcv);
647 if (skb->len > sizeof(struct udphdr) && encap_rcv) {
650 /* Verify checksum before giving to encap */
651 if (udp_lib_checksum_complete(skb))
654 ret = encap_rcv(sk, skb);
656 UDP_INC_STATS_BH(sock_net(sk),
663 /* FALLTHROUGH -- it's a UDP Packet */
667 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
669 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
671 if (up->pcrlen == 0) { /* full coverage was set */
672 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
673 UDP_SKB_CB(skb)->cscov, skb->len);
676 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
677 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
678 UDP_SKB_CB(skb)->cscov, up->pcrlen);
683 if (rcu_access_pointer(sk->sk_filter)) {
684 if (udp_lib_checksum_complete(skb))
688 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
689 UDP6_INC_STATS_BH(sock_net(sk),
690 UDP_MIB_RCVBUFERRORS, is_udplite);
698 if (!sock_owned_by_user(sk))
699 rc = __udpv6_queue_rcv_skb(sk, skb);
700 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
709 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
711 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
712 atomic_inc(&sk->sk_drops);
717 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
718 __be16 loc_port, const struct in6_addr *loc_addr,
719 __be16 rmt_port, const struct in6_addr *rmt_addr,
720 int dif, unsigned short hnum)
722 struct inet_sock *inet = inet_sk(sk);
724 if (!net_eq(sock_net(sk), net))
727 if (udp_sk(sk)->udp_port_hash != hnum ||
728 sk->sk_family != PF_INET6 ||
729 (inet->inet_dport && inet->inet_dport != rmt_port) ||
730 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
731 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
732 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif) ||
733 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
734 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
736 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
741 static void flush_stack(struct sock **stack, unsigned int count,
742 struct sk_buff *skb, unsigned int final)
744 struct sk_buff *skb1 = NULL;
748 for (i = 0; i < count; i++) {
751 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
753 atomic_inc(&sk->sk_drops);
754 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
756 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
760 if (skb1 && udpv6_queue_rcv_skb(sk, skb1) <= 0)
768 static void udp6_csum_zero_error(struct sk_buff *skb)
770 /* RFC 2460 section 8.1 says that we SHOULD log
771 * this error. Well, it is reasonable.
773 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
774 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
775 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
779 * Note: called only from the BH handler context,
780 * so we don't need to lock the hashes.
782 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
783 const struct in6_addr *saddr, const struct in6_addr *daddr,
784 struct udp_table *udptable, int proto)
786 struct sock *sk, *stack[256 / sizeof(struct sock *)];
787 const struct udphdr *uh = udp_hdr(skb);
788 struct hlist_nulls_node *node;
789 unsigned short hnum = ntohs(uh->dest);
790 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
791 int dif = inet6_iif(skb);
792 unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node);
793 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
794 bool inner_flushed = false;
797 hash2_any = udp6_portaddr_hash(net, &in6addr_any, hnum) &
799 hash2 = udp6_portaddr_hash(net, daddr, hnum) & udp_table.mask;
801 hslot = &udp_table.hash2[hash2];
802 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
805 spin_lock(&hslot->lock);
806 sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) {
807 if (__udp_v6_is_mcast_sock(net, sk,
811 /* If zero checksum and no_check is not on for
812 * the socket then skip it.
814 (uh->check || udp_sk(sk)->no_check6_rx)) {
815 if (unlikely(count == ARRAY_SIZE(stack))) {
816 flush_stack(stack, count, skb, ~0);
817 inner_flushed = true;
825 spin_unlock(&hslot->lock);
827 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
828 if (use_hash2 && hash2 != hash2_any) {
834 flush_stack(stack, count, skb, count - 1);
837 UDP_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI,
838 proto == IPPROTO_UDPLITE);
844 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
847 struct net *net = dev_net(skb->dev);
850 const struct in6_addr *saddr, *daddr;
853 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
856 saddr = &ipv6_hdr(skb)->saddr;
857 daddr = &ipv6_hdr(skb)->daddr;
860 ulen = ntohs(uh->len);
864 if (proto == IPPROTO_UDP) {
865 /* UDP validates ulen. */
867 /* Check for jumbo payload */
871 if (ulen < sizeof(*uh))
874 if (ulen < skb->len) {
875 if (pskb_trim_rcsum(skb, ulen))
877 saddr = &ipv6_hdr(skb)->saddr;
878 daddr = &ipv6_hdr(skb)->daddr;
883 if (udp6_csum_init(skb, uh, proto))
887 * Multicast receive code
889 if (ipv6_addr_is_multicast(daddr))
890 return __udp6_lib_mcast_deliver(net, skb,
891 saddr, daddr, udptable, proto);
896 * check socket cache ... must talk to Alan about his plans
897 * for sock caches... i'll skip this for now.
899 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
903 if (!uh->check && !udp_sk(sk)->no_check6_rx) {
905 udp6_csum_zero_error(skb);
909 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
910 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
913 ret = udpv6_queue_rcv_skb(sk, skb);
916 /* a return value > 0 means to resubmit the input, but
917 * it wants the return to be -protocol, or 0
926 udp6_csum_zero_error(skb);
930 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
933 if (udp_lib_checksum_complete(skb))
936 UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
937 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
943 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
944 proto == IPPROTO_UDPLITE ? "-Lite" : "",
945 saddr, ntohs(uh->source),
947 daddr, ntohs(uh->dest));
950 UDP6_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
952 UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
957 static __inline__ int udpv6_rcv(struct sk_buff *skb)
959 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
963 * Throw away all pending data and cancel the corking. Socket is locked.
965 static void udp_v6_flush_pending_frames(struct sock *sk)
967 struct udp_sock *up = udp_sk(sk);
969 if (up->pending == AF_INET)
970 udp_flush_pending_frames(sk);
971 else if (up->pending) {
974 ip6_flush_pending_frames(sk);
979 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
980 * @sk: socket we are sending on
981 * @skb: sk_buff containing the filled-in UDP header
982 * (checksum field must be zeroed out)
984 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
985 const struct in6_addr *saddr,
986 const struct in6_addr *daddr, int len)
989 struct udphdr *uh = udp_hdr(skb);
990 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
994 /* Only one fragment on the socket. */
995 skb->csum_start = skb_transport_header(skb) - skb->head;
996 skb->csum_offset = offsetof(struct udphdr, check);
997 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1000 * HW-checksum won't work as there are two or more
1001 * fragments on the socket so that all csums of sk_buffs
1002 * should be together
1004 offset = skb_transport_offset(skb);
1005 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1007 skb->ip_summed = CHECKSUM_NONE;
1010 csum = csum_add(csum, frags->csum);
1011 } while ((frags = frags->next));
1013 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1016 uh->check = CSUM_MANGLED_0;
1024 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6)
1026 struct sock *sk = skb->sk;
1029 int is_udplite = IS_UDPLITE(sk);
1031 int offset = skb_transport_offset(skb);
1032 int len = skb->len - offset;
1035 * Create a UDP header
1038 uh->source = fl6->fl6_sport;
1039 uh->dest = fl6->fl6_dport;
1040 uh->len = htons(len);
1044 csum = udplite_csum(skb);
1045 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */
1046 skb->ip_summed = CHECKSUM_NONE;
1048 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1049 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1052 csum = udp_csum(skb);
1054 /* add protocol-dependent pseudo-header */
1055 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1056 len, fl6->flowi6_proto, csum);
1058 uh->check = CSUM_MANGLED_0;
1061 err = ip6_send_skb(skb);
1063 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1064 UDP6_INC_STATS_USER(sock_net(sk),
1065 UDP_MIB_SNDBUFERRORS, is_udplite);
1069 UDP6_INC_STATS_USER(sock_net(sk),
1070 UDP_MIB_OUTDATAGRAMS, is_udplite);
1074 static int udp_v6_push_pending_frames(struct sock *sk)
1076 struct sk_buff *skb;
1077 struct udp_sock *up = udp_sk(sk);
1081 if (up->pending == AF_INET)
1082 return udp_push_pending_frames(sk);
1084 /* ip6_finish_skb will release the cork, so make a copy of
1087 fl6 = inet_sk(sk)->cork.fl.u.ip6;
1089 skb = ip6_finish_skb(sk);
1093 err = udp_v6_send_skb(skb, &fl6);
1101 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1103 struct ipv6_txoptions opt_space;
1104 struct udp_sock *up = udp_sk(sk);
1105 struct inet_sock *inet = inet_sk(sk);
1106 struct ipv6_pinfo *np = inet6_sk(sk);
1107 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1108 struct in6_addr *daddr, *final_p, final;
1109 struct ipv6_txoptions *opt = NULL;
1110 struct ip6_flowlabel *flowlabel = NULL;
1112 struct dst_entry *dst;
1113 int addr_len = msg->msg_namelen;
1118 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1121 int is_udplite = IS_UDPLITE(sk);
1122 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1124 /* destination address check */
1126 if (addr_len < offsetof(struct sockaddr, sa_data))
1129 switch (sin6->sin6_family) {
1131 if (addr_len < SIN6_LEN_RFC2133)
1133 daddr = &sin6->sin6_addr;
1136 goto do_udp_sendmsg;
1138 msg->msg_name = sin6 = NULL;
1139 msg->msg_namelen = addr_len = 0;
1145 } else if (!up->pending) {
1146 if (sk->sk_state != TCP_ESTABLISHED)
1147 return -EDESTADDRREQ;
1148 daddr = &sk->sk_v6_daddr;
1153 if (ipv6_addr_v4mapped(daddr)) {
1154 struct sockaddr_in sin;
1155 sin.sin_family = AF_INET;
1156 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1157 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1158 msg->msg_name = &sin;
1159 msg->msg_namelen = sizeof(sin);
1161 if (__ipv6_only_sock(sk))
1162 return -ENETUNREACH;
1163 return udp_sendmsg(sk, msg, len);
1167 if (up->pending == AF_INET)
1168 return udp_sendmsg(sk, msg, len);
1170 /* Rough check on arithmetic overflow,
1171 better check is made in ip6_append_data().
1173 if (len > INT_MAX - sizeof(struct udphdr))
1176 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1179 * There are pending frames.
1180 * The socket lock must be held while it's corked.
1183 if (likely(up->pending)) {
1184 if (unlikely(up->pending != AF_INET6)) {
1186 return -EAFNOSUPPORT;
1189 goto do_append_data;
1193 ulen += sizeof(struct udphdr);
1195 memset(&fl6, 0, sizeof(fl6));
1198 if (sin6->sin6_port == 0)
1201 fl6.fl6_dport = sin6->sin6_port;
1202 daddr = &sin6->sin6_addr;
1205 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1206 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1207 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1214 * Otherwise it will be difficult to maintain
1217 if (sk->sk_state == TCP_ESTABLISHED &&
1218 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1219 daddr = &sk->sk_v6_daddr;
1221 if (addr_len >= sizeof(struct sockaddr_in6) &&
1222 sin6->sin6_scope_id &&
1223 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1224 fl6.flowi6_oif = sin6->sin6_scope_id;
1226 if (sk->sk_state != TCP_ESTABLISHED)
1227 return -EDESTADDRREQ;
1229 fl6.fl6_dport = inet->inet_dport;
1230 daddr = &sk->sk_v6_daddr;
1231 fl6.flowlabel = np->flow_label;
1235 if (!fl6.flowi6_oif)
1236 fl6.flowi6_oif = sk->sk_bound_dev_if;
1238 if (!fl6.flowi6_oif)
1239 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1241 fl6.flowi6_mark = sk->sk_mark;
1243 if (msg->msg_controllen) {
1245 memset(opt, 0, sizeof(struct ipv6_txoptions));
1246 opt->tot_len = sizeof(*opt);
1248 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt,
1249 &hlimit, &tclass, &dontfrag);
1251 fl6_sock_release(flowlabel);
1254 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1255 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1259 if (!(opt->opt_nflen|opt->opt_flen))
1266 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1267 opt = ipv6_fixup_options(&opt_space, opt);
1269 fl6.flowi6_proto = sk->sk_protocol;
1270 if (!ipv6_addr_any(daddr))
1273 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1274 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1275 fl6.saddr = np->saddr;
1276 fl6.fl6_sport = inet->inet_sport;
1278 final_p = fl6_update_dst(&fl6, opt, &final);
1282 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1283 fl6.flowi6_oif = np->mcast_oif;
1285 } else if (!fl6.flowi6_oif)
1286 fl6.flowi6_oif = np->ucast_oif;
1288 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1290 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p);
1298 hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1301 tclass = np->tclass;
1303 if (msg->msg_flags&MSG_CONFIRM)
1307 /* Lockless fast path for the non-corking case */
1309 struct sk_buff *skb;
1311 skb = ip6_make_skb(sk, getfrag, msg, ulen,
1312 sizeof(struct udphdr), hlimit, tclass, opt,
1313 &fl6, (struct rt6_info *)dst,
1314 msg->msg_flags, dontfrag);
1316 if (!IS_ERR_OR_NULL(skb))
1317 err = udp_v6_send_skb(skb, &fl6);
1322 if (unlikely(up->pending)) {
1323 /* The socket is already corked while preparing it. */
1324 /* ... which is an evident application bug. --ANK */
1327 net_dbg_ratelimited("udp cork app bug 2\n");
1332 up->pending = AF_INET6;
1336 dontfrag = np->dontfrag;
1338 err = ip6_append_data(sk, getfrag, msg, ulen,
1339 sizeof(struct udphdr), hlimit, tclass, opt, &fl6,
1340 (struct rt6_info *)dst,
1341 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag);
1343 udp_v6_flush_pending_frames(sk);
1345 err = udp_v6_push_pending_frames(sk);
1346 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1350 err = np->recverr ? net_xmit_errno(err) : 0;
1356 ip6_dst_store(sk, dst,
1357 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ?
1358 &sk->sk_v6_daddr : NULL,
1359 #ifdef CONFIG_IPV6_SUBTREES
1360 ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1372 fl6_sock_release(flowlabel);
1376 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1377 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1378 * we don't have a good statistic (IpOutDiscards but it can be too many
1379 * things). We could add another new stat but at least for now that
1380 * seems like overkill.
1382 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1383 UDP6_INC_STATS_USER(sock_net(sk),
1384 UDP_MIB_SNDBUFERRORS, is_udplite);
1390 if (!(msg->msg_flags&MSG_PROBE) || len)
1391 goto back_from_confirm;
1396 void udpv6_destroy_sock(struct sock *sk)
1398 struct udp_sock *up = udp_sk(sk);
1400 udp_v6_flush_pending_frames(sk);
1403 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
1404 void (*encap_destroy)(struct sock *sk);
1405 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1410 inet6_destroy_sock(sk);
1414 * Socket option code for UDP
1416 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1417 char __user *optval, unsigned int optlen)
1419 if (level == SOL_UDP || level == SOL_UDPLITE)
1420 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1421 udp_v6_push_pending_frames);
1422 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1425 #ifdef CONFIG_COMPAT
1426 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1427 char __user *optval, unsigned int optlen)
1429 if (level == SOL_UDP || level == SOL_UDPLITE)
1430 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1431 udp_v6_push_pending_frames);
1432 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1436 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1437 char __user *optval, int __user *optlen)
1439 if (level == SOL_UDP || level == SOL_UDPLITE)
1440 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1441 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1444 #ifdef CONFIG_COMPAT
1445 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1446 char __user *optval, int __user *optlen)
1448 if (level == SOL_UDP || level == SOL_UDPLITE)
1449 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1450 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1454 static const struct inet6_protocol udpv6_protocol = {
1455 .handler = udpv6_rcv,
1456 .err_handler = udpv6_err,
1457 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1460 /* ------------------------------------------------------------------------ */
1461 #ifdef CONFIG_PROC_FS
1462 int udp6_seq_show(struct seq_file *seq, void *v)
1464 if (v == SEQ_START_TOKEN) {
1465 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1467 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1468 struct inet_sock *inet = inet_sk(v);
1469 __u16 srcp = ntohs(inet->inet_sport);
1470 __u16 destp = ntohs(inet->inet_dport);
1471 ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket);
1476 static const struct file_operations udp6_afinfo_seq_fops = {
1477 .owner = THIS_MODULE,
1478 .open = udp_seq_open,
1480 .llseek = seq_lseek,
1481 .release = seq_release_net
1484 static struct udp_seq_afinfo udp6_seq_afinfo = {
1487 .udp_table = &udp_table,
1488 .seq_fops = &udp6_afinfo_seq_fops,
1490 .show = udp6_seq_show,
1494 int __net_init udp6_proc_init(struct net *net)
1496 return udp_proc_register(net, &udp6_seq_afinfo);
1499 void udp6_proc_exit(struct net *net) {
1500 udp_proc_unregister(net, &udp6_seq_afinfo);
1502 #endif /* CONFIG_PROC_FS */
1504 void udp_v6_clear_sk(struct sock *sk, int size)
1506 struct inet_sock *inet = inet_sk(sk);
1508 /* we do not want to clear pinet6 field, because of RCU lookups */
1509 sk_prot_clear_portaddr_nulls(sk, offsetof(struct inet_sock, pinet6));
1511 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1512 memset(&inet->pinet6 + 1, 0, size);
1515 /* ------------------------------------------------------------------------ */
1517 struct proto udpv6_prot = {
1519 .owner = THIS_MODULE,
1520 .close = udp_lib_close,
1521 .connect = ip6_datagram_connect,
1522 .disconnect = udp_disconnect,
1524 .destroy = udpv6_destroy_sock,
1525 .setsockopt = udpv6_setsockopt,
1526 .getsockopt = udpv6_getsockopt,
1527 .sendmsg = udpv6_sendmsg,
1528 .recvmsg = udpv6_recvmsg,
1529 .backlog_rcv = __udpv6_queue_rcv_skb,
1530 .hash = udp_lib_hash,
1531 .unhash = udp_lib_unhash,
1532 .rehash = udp_v6_rehash,
1533 .get_port = udp_v6_get_port,
1534 .memory_allocated = &udp_memory_allocated,
1535 .sysctl_mem = sysctl_udp_mem,
1536 .sysctl_wmem = &sysctl_udp_wmem_min,
1537 .sysctl_rmem = &sysctl_udp_rmem_min,
1538 .obj_size = sizeof(struct udp6_sock),
1539 .slab_flags = SLAB_DESTROY_BY_RCU,
1540 .h.udp_table = &udp_table,
1541 #ifdef CONFIG_COMPAT
1542 .compat_setsockopt = compat_udpv6_setsockopt,
1543 .compat_getsockopt = compat_udpv6_getsockopt,
1545 .clear_sk = udp_v6_clear_sk,
1548 static struct inet_protosw udpv6_protosw = {
1550 .protocol = IPPROTO_UDP,
1551 .prot = &udpv6_prot,
1552 .ops = &inet6_dgram_ops,
1553 .flags = INET_PROTOSW_PERMANENT,
1556 int __init udpv6_init(void)
1560 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1564 ret = inet6_register_protosw(&udpv6_protosw);
1566 goto out_udpv6_protocol;
1571 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1575 void udpv6_exit(void)
1577 inet6_unregister_protosw(&udpv6_protosw);
1578 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);