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>
49 #include <net/busy_poll.h>
50 #include <net/sock_reuseport.h>
52 #include <linux/proc_fs.h>
53 #include <linux/seq_file.h>
54 #include <trace/events/skb.h>
57 static u32 udp6_ehashfn(const struct net *net,
58 const struct in6_addr *laddr,
60 const struct in6_addr *faddr,
63 static u32 udp6_ehash_secret __read_mostly;
64 static u32 udp_ipv6_hash_secret __read_mostly;
68 net_get_random_once(&udp6_ehash_secret,
69 sizeof(udp6_ehash_secret));
70 net_get_random_once(&udp_ipv6_hash_secret,
71 sizeof(udp_ipv6_hash_secret));
73 lhash = (__force u32)laddr->s6_addr32[3];
74 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
76 return __inet6_ehashfn(lhash, lport, fhash, fport,
77 udp_ipv6_hash_secret + net_hash_mix(net));
80 /* match_wildcard == true: IPV6_ADDR_ANY equals to any IPv6 addresses if IPv6
81 * only, and any IPv4 addresses if not IPv6 only
82 * match_wildcard == false: addresses must be exactly the same, i.e.
83 * IPV6_ADDR_ANY only equals to IPV6_ADDR_ANY,
84 * and 0.0.0.0 equals to 0.0.0.0 only
86 int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2,
89 const struct in6_addr *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2);
90 int sk2_ipv6only = inet_v6_ipv6only(sk2);
91 int addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
92 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
94 /* if both are mapped, treat as IPv4 */
95 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) {
97 if (sk->sk_rcv_saddr == sk2->sk_rcv_saddr)
99 if (!sk->sk_rcv_saddr || !sk2->sk_rcv_saddr)
100 return match_wildcard;
105 if (addr_type == IPV6_ADDR_ANY && addr_type2 == IPV6_ADDR_ANY)
108 if (addr_type2 == IPV6_ADDR_ANY && match_wildcard &&
109 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
112 if (addr_type == IPV6_ADDR_ANY && match_wildcard &&
113 !(ipv6_only_sock(sk) && addr_type2 == IPV6_ADDR_MAPPED))
116 if (sk2_rcv_saddr6 &&
117 ipv6_addr_equal(&sk->sk_v6_rcv_saddr, sk2_rcv_saddr6))
123 static u32 udp6_portaddr_hash(const struct net *net,
124 const struct in6_addr *addr6,
127 unsigned int hash, mix = net_hash_mix(net);
129 if (ipv6_addr_any(addr6))
130 hash = jhash_1word(0, mix);
131 else if (ipv6_addr_v4mapped(addr6))
132 hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix);
134 hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix);
139 int udp_v6_get_port(struct sock *sk, unsigned short snum)
141 unsigned int hash2_nulladdr =
142 udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
143 unsigned int hash2_partial =
144 udp6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
146 /* precompute partial secondary hash */
147 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
148 return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal, hash2_nulladdr);
151 static void udp_v6_rehash(struct sock *sk)
153 u16 new_hash = udp6_portaddr_hash(sock_net(sk),
154 &sk->sk_v6_rcv_saddr,
155 inet_sk(sk)->inet_num);
157 udp_lib_rehash(sk, new_hash);
160 static inline int compute_score(struct sock *sk, struct net *net,
162 const struct in6_addr *saddr, __be16 sport,
163 const struct in6_addr *daddr, __be16 dport,
167 struct inet_sock *inet;
169 if (!net_eq(sock_net(sk), net) ||
170 udp_sk(sk)->udp_port_hash != hnum ||
171 sk->sk_family != PF_INET6)
177 if (inet->inet_dport) {
178 if (inet->inet_dport != sport)
183 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
184 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
189 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
190 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
195 if (sk->sk_bound_dev_if) {
196 if (sk->sk_bound_dev_if != dif)
201 if (sk->sk_incoming_cpu == raw_smp_processor_id())
207 static inline int compute_score2(struct sock *sk, struct net *net,
208 const struct in6_addr *saddr, __be16 sport,
209 const struct in6_addr *daddr,
210 unsigned short hnum, int dif)
213 struct inet_sock *inet;
215 if (!net_eq(sock_net(sk), net) ||
216 udp_sk(sk)->udp_port_hash != hnum ||
217 sk->sk_family != PF_INET6)
220 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
226 if (inet->inet_dport) {
227 if (inet->inet_dport != sport)
232 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
233 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
238 if (sk->sk_bound_dev_if) {
239 if (sk->sk_bound_dev_if != dif)
244 if (sk->sk_incoming_cpu == raw_smp_processor_id())
250 /* called with read_rcu_lock() */
251 static struct sock *udp6_lib_lookup2(struct net *net,
252 const struct in6_addr *saddr, __be16 sport,
253 const struct in6_addr *daddr, unsigned int hnum, int dif,
254 struct udp_hslot *hslot2, unsigned int slot2)
256 struct sock *sk, *result;
257 struct hlist_nulls_node *node;
258 int score, badness, matches = 0, reuseport = 0;
264 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
265 score = compute_score2(sk, net, saddr, sport,
267 if (score > badness) {
270 reuseport = sk->sk_reuseport;
273 hash = udp6_ehashfn(net, daddr, hnum,
275 sk2 = reuseport_select_sock(sk, hash, NULL, 0);
282 } else if (score == badness && reuseport) {
284 if (reciprocal_scale(hash, matches) == 0)
286 hash = next_pseudo_random32(hash);
290 * if the nulls value we got at the end of this lookup is
291 * not the expected one, we must restart lookup.
292 * We probably met an item that was moved to another chain.
294 if (get_nulls_value(node) != slot2)
299 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
301 else if (unlikely(compute_score2(result, net, saddr, sport,
302 daddr, hnum, dif) < badness)) {
310 struct sock *__udp6_lib_lookup(struct net *net,
311 const struct in6_addr *saddr, __be16 sport,
312 const struct in6_addr *daddr, __be16 dport,
313 int dif, struct udp_table *udptable,
316 struct sock *sk, *result;
317 struct hlist_nulls_node *node;
318 unsigned short hnum = ntohs(dport);
319 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
320 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
321 int score, badness, matches = 0, reuseport = 0;
325 if (hslot->count > 10) {
326 hash2 = udp6_portaddr_hash(net, daddr, hnum);
327 slot2 = hash2 & udptable->mask;
328 hslot2 = &udptable->hash2[slot2];
329 if (hslot->count < hslot2->count)
332 result = udp6_lib_lookup2(net, saddr, sport,
336 hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
337 slot2 = hash2 & udptable->mask;
338 hslot2 = &udptable->hash2[slot2];
339 if (hslot->count < hslot2->count)
342 result = udp6_lib_lookup2(net, saddr, sport,
343 &in6addr_any, hnum, dif,
352 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
353 score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif);
354 if (score > badness) {
357 reuseport = sk->sk_reuseport;
360 hash = udp6_ehashfn(net, daddr, hnum,
362 sk2 = reuseport_select_sock(sk, hash, skb,
363 sizeof(struct udphdr));
370 } else if (score == badness && reuseport) {
372 if (reciprocal_scale(hash, matches) == 0)
374 hash = next_pseudo_random32(hash);
378 * if the nulls value we got at the end of this lookup is
379 * not the expected one, we must restart lookup.
380 * We probably met an item that was moved to another chain.
382 if (get_nulls_value(node) != slot)
387 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
389 else if (unlikely(compute_score(result, net, hnum, saddr, sport,
390 daddr, dport, dif) < badness)) {
398 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
400 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
401 __be16 sport, __be16 dport,
402 struct udp_table *udptable)
405 const struct ipv6hdr *iph = ipv6_hdr(skb);
407 sk = skb_steal_sock(skb);
410 return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport,
411 &iph->daddr, dport, inet6_iif(skb),
415 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
416 const struct in6_addr *daddr, __be16 dport, int dif)
418 return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table, NULL);
420 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
423 * This should be easy, if there is something there we
424 * return it, otherwise we block.
427 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
428 int noblock, int flags, int *addr_len)
430 struct ipv6_pinfo *np = inet6_sk(sk);
431 struct inet_sock *inet = inet_sk(sk);
433 unsigned int ulen, copied;
436 int is_udplite = IS_UDPLITE(sk);
437 bool checksum_valid = false;
441 if (flags & MSG_ERRQUEUE)
442 return ipv6_recv_error(sk, msg, len, addr_len);
444 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
445 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
448 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
449 &peeked, &off, &err);
453 ulen = skb->len - sizeof(struct udphdr);
457 else if (copied < ulen)
458 msg->msg_flags |= MSG_TRUNC;
460 is_udp4 = (skb->protocol == htons(ETH_P_IP));
463 * If checksum is needed at all, try to do it while copying the
464 * data. If the data is truncated, or if we only want a partial
465 * coverage checksum (UDP-Lite), do it before the copy.
468 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
469 checksum_valid = !udp_lib_checksum_complete(skb);
474 if (checksum_valid || skb_csum_unnecessary(skb))
475 err = skb_copy_datagram_msg(skb, sizeof(struct udphdr),
478 err = skb_copy_and_csum_datagram_msg(skb, sizeof(struct udphdr), msg);
483 trace_kfree_skb(skb, udpv6_recvmsg);
485 atomic_inc(&sk->sk_drops);
487 UDP_INC_STATS_USER(sock_net(sk),
491 UDP6_INC_STATS_USER(sock_net(sk),
499 UDP_INC_STATS_USER(sock_net(sk),
500 UDP_MIB_INDATAGRAMS, is_udplite);
502 UDP6_INC_STATS_USER(sock_net(sk),
503 UDP_MIB_INDATAGRAMS, is_udplite);
506 sock_recv_ts_and_drops(msg, sk, skb);
508 /* Copy the address. */
510 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
511 sin6->sin6_family = AF_INET6;
512 sin6->sin6_port = udp_hdr(skb)->source;
513 sin6->sin6_flowinfo = 0;
516 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
518 sin6->sin6_scope_id = 0;
520 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
521 sin6->sin6_scope_id =
522 ipv6_iface_scope_id(&sin6->sin6_addr,
525 *addr_len = sizeof(*sin6);
529 ip6_datagram_recv_common_ctl(sk, msg, skb);
532 if (inet->cmsg_flags)
533 ip_cmsg_recv(msg, skb);
536 ip6_datagram_recv_specific_ctl(sk, msg, skb);
540 if (flags & MSG_TRUNC)
544 skb_free_datagram_locked(sk, skb);
549 slow = lock_sock_fast(sk);
550 if (!skb_kill_datagram(sk, skb, flags)) {
552 UDP_INC_STATS_USER(sock_net(sk),
553 UDP_MIB_CSUMERRORS, is_udplite);
554 UDP_INC_STATS_USER(sock_net(sk),
555 UDP_MIB_INERRORS, is_udplite);
557 UDP6_INC_STATS_USER(sock_net(sk),
558 UDP_MIB_CSUMERRORS, is_udplite);
559 UDP6_INC_STATS_USER(sock_net(sk),
560 UDP_MIB_INERRORS, is_udplite);
563 unlock_sock_fast(sk, slow);
565 /* starting over for a new packet, but check if we need to yield */
567 msg->msg_flags &= ~MSG_TRUNC;
571 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
572 u8 type, u8 code, int offset, __be32 info,
573 struct udp_table *udptable)
575 struct ipv6_pinfo *np;
576 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
577 const struct in6_addr *saddr = &hdr->saddr;
578 const struct in6_addr *daddr = &hdr->daddr;
579 struct udphdr *uh = (struct udphdr *)(skb->data+offset);
582 struct net *net = dev_net(skb->dev);
584 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
585 inet6_iif(skb), udptable, skb);
587 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
592 if (type == ICMPV6_PKT_TOOBIG) {
593 if (!ip6_sk_accept_pmtu(sk))
595 ip6_sk_update_pmtu(skb, sk, info);
597 if (type == NDISC_REDIRECT) {
598 ip6_sk_redirect(skb, sk);
604 if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
607 if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
611 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
614 sk->sk_error_report(sk);
619 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
623 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
624 sock_rps_save_rxhash(sk, skb);
625 sk_mark_napi_id(sk, skb);
626 sk_incoming_cpu_update(sk);
629 rc = sock_queue_rcv_skb(sk, skb);
631 int is_udplite = IS_UDPLITE(sk);
633 /* Note that an ENOMEM error is charged twice */
635 UDP6_INC_STATS_BH(sock_net(sk),
636 UDP_MIB_RCVBUFERRORS, is_udplite);
637 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
644 static __inline__ void udpv6_err(struct sk_buff *skb,
645 struct inet6_skb_parm *opt, u8 type,
646 u8 code, int offset, __be32 info)
648 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
651 static struct static_key udpv6_encap_needed __read_mostly;
652 void udpv6_encap_enable(void)
654 if (!static_key_enabled(&udpv6_encap_needed))
655 static_key_slow_inc(&udpv6_encap_needed);
657 EXPORT_SYMBOL(udpv6_encap_enable);
659 int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
661 struct udp_sock *up = udp_sk(sk);
663 int is_udplite = IS_UDPLITE(sk);
665 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
668 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
669 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
672 * This is an encapsulation socket so pass the skb to
673 * the socket's udp_encap_rcv() hook. Otherwise, just
674 * fall through and pass this up the UDP socket.
675 * up->encap_rcv() returns the following value:
676 * =0 if skb was successfully passed to the encap
677 * handler or was discarded by it.
678 * >0 if skb should be passed on to UDP.
679 * <0 if skb should be resubmitted as proto -N
682 /* if we're overly short, let UDP handle it */
683 encap_rcv = ACCESS_ONCE(up->encap_rcv);
684 if (skb->len > sizeof(struct udphdr) && encap_rcv) {
687 /* Verify checksum before giving to encap */
688 if (udp_lib_checksum_complete(skb))
691 ret = encap_rcv(sk, skb);
693 UDP_INC_STATS_BH(sock_net(sk),
700 /* FALLTHROUGH -- it's a UDP Packet */
704 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
706 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
708 if (up->pcrlen == 0) { /* full coverage was set */
709 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
710 UDP_SKB_CB(skb)->cscov, skb->len);
713 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
714 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
715 UDP_SKB_CB(skb)->cscov, up->pcrlen);
720 if (rcu_access_pointer(sk->sk_filter)) {
721 if (udp_lib_checksum_complete(skb))
725 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
726 UDP6_INC_STATS_BH(sock_net(sk),
727 UDP_MIB_RCVBUFERRORS, is_udplite);
735 if (!sock_owned_by_user(sk))
736 rc = __udpv6_queue_rcv_skb(sk, skb);
737 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
746 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
748 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
749 atomic_inc(&sk->sk_drops);
754 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
755 __be16 loc_port, const struct in6_addr *loc_addr,
756 __be16 rmt_port, const struct in6_addr *rmt_addr,
757 int dif, unsigned short hnum)
759 struct inet_sock *inet = inet_sk(sk);
761 if (!net_eq(sock_net(sk), net))
764 if (udp_sk(sk)->udp_port_hash != hnum ||
765 sk->sk_family != PF_INET6 ||
766 (inet->inet_dport && inet->inet_dport != rmt_port) ||
767 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
768 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
769 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif) ||
770 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
771 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
773 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
778 static void flush_stack(struct sock **stack, unsigned int count,
779 struct sk_buff *skb, unsigned int final)
781 struct sk_buff *skb1 = NULL;
785 for (i = 0; i < count; i++) {
788 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
790 atomic_inc(&sk->sk_drops);
791 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
793 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
797 if (skb1 && udpv6_queue_rcv_skb(sk, skb1) <= 0)
805 static void udp6_csum_zero_error(struct sk_buff *skb)
807 /* RFC 2460 section 8.1 says that we SHOULD log
808 * this error. Well, it is reasonable.
810 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
811 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
812 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
816 * Note: called only from the BH handler context,
817 * so we don't need to lock the hashes.
819 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
820 const struct in6_addr *saddr, const struct in6_addr *daddr,
821 struct udp_table *udptable, int proto)
823 struct sock *sk, *stack[256 / sizeof(struct sock *)];
824 const struct udphdr *uh = udp_hdr(skb);
825 struct hlist_nulls_node *node;
826 unsigned short hnum = ntohs(uh->dest);
827 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
828 int dif = inet6_iif(skb);
829 unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node);
830 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
831 bool inner_flushed = false;
834 hash2_any = udp6_portaddr_hash(net, &in6addr_any, hnum) &
836 hash2 = udp6_portaddr_hash(net, daddr, hnum) & udp_table.mask;
838 hslot = &udp_table.hash2[hash2];
839 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
842 spin_lock(&hslot->lock);
843 sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) {
844 if (__udp_v6_is_mcast_sock(net, sk,
848 /* If zero checksum and no_check is not on for
849 * the socket then skip it.
851 (uh->check || udp_sk(sk)->no_check6_rx)) {
852 if (unlikely(count == ARRAY_SIZE(stack))) {
853 flush_stack(stack, count, skb, ~0);
854 inner_flushed = true;
862 spin_unlock(&hslot->lock);
864 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
865 if (use_hash2 && hash2 != hash2_any) {
871 flush_stack(stack, count, skb, count - 1);
874 UDP_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI,
875 proto == IPPROTO_UDPLITE);
881 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
884 struct net *net = dev_net(skb->dev);
887 const struct in6_addr *saddr, *daddr;
890 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
893 saddr = &ipv6_hdr(skb)->saddr;
894 daddr = &ipv6_hdr(skb)->daddr;
897 ulen = ntohs(uh->len);
901 if (proto == IPPROTO_UDP) {
902 /* UDP validates ulen. */
904 /* Check for jumbo payload */
908 if (ulen < sizeof(*uh))
911 if (ulen < skb->len) {
912 if (pskb_trim_rcsum(skb, ulen))
914 saddr = &ipv6_hdr(skb)->saddr;
915 daddr = &ipv6_hdr(skb)->daddr;
920 if (udp6_csum_init(skb, uh, proto))
924 * Multicast receive code
926 if (ipv6_addr_is_multicast(daddr))
927 return __udp6_lib_mcast_deliver(net, skb,
928 saddr, daddr, udptable, proto);
933 * check socket cache ... must talk to Alan about his plans
934 * for sock caches... i'll skip this for now.
936 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
940 if (!uh->check && !udp_sk(sk)->no_check6_rx) {
942 udp6_csum_zero_error(skb);
946 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
947 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
950 ret = udpv6_queue_rcv_skb(sk, skb);
953 /* a return value > 0 means to resubmit the input, but
954 * it wants the return to be -protocol, or 0
963 udp6_csum_zero_error(skb);
967 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
970 if (udp_lib_checksum_complete(skb))
973 UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
974 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
980 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
981 proto == IPPROTO_UDPLITE ? "-Lite" : "",
982 saddr, ntohs(uh->source),
984 daddr, ntohs(uh->dest));
987 UDP6_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
989 UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
994 static __inline__ int udpv6_rcv(struct sk_buff *skb)
996 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1000 * Throw away all pending data and cancel the corking. Socket is locked.
1002 static void udp_v6_flush_pending_frames(struct sock *sk)
1004 struct udp_sock *up = udp_sk(sk);
1006 if (up->pending == AF_INET)
1007 udp_flush_pending_frames(sk);
1008 else if (up->pending) {
1011 ip6_flush_pending_frames(sk);
1016 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
1017 * @sk: socket we are sending on
1018 * @skb: sk_buff containing the filled-in UDP header
1019 * (checksum field must be zeroed out)
1021 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
1022 const struct in6_addr *saddr,
1023 const struct in6_addr *daddr, int len)
1025 unsigned int offset;
1026 struct udphdr *uh = udp_hdr(skb);
1027 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
1031 /* Only one fragment on the socket. */
1032 skb->csum_start = skb_transport_header(skb) - skb->head;
1033 skb->csum_offset = offsetof(struct udphdr, check);
1034 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1037 * HW-checksum won't work as there are two or more
1038 * fragments on the socket so that all csums of sk_buffs
1039 * should be together
1041 offset = skb_transport_offset(skb);
1042 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1044 skb->ip_summed = CHECKSUM_NONE;
1047 csum = csum_add(csum, frags->csum);
1048 } while ((frags = frags->next));
1050 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1053 uh->check = CSUM_MANGLED_0;
1061 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6)
1063 struct sock *sk = skb->sk;
1066 int is_udplite = IS_UDPLITE(sk);
1068 int offset = skb_transport_offset(skb);
1069 int len = skb->len - offset;
1072 * Create a UDP header
1075 uh->source = fl6->fl6_sport;
1076 uh->dest = fl6->fl6_dport;
1077 uh->len = htons(len);
1081 csum = udplite_csum(skb);
1082 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */
1083 skb->ip_summed = CHECKSUM_NONE;
1085 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1086 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1089 csum = udp_csum(skb);
1091 /* add protocol-dependent pseudo-header */
1092 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1093 len, fl6->flowi6_proto, csum);
1095 uh->check = CSUM_MANGLED_0;
1098 err = ip6_send_skb(skb);
1100 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1101 UDP6_INC_STATS_USER(sock_net(sk),
1102 UDP_MIB_SNDBUFERRORS, is_udplite);
1106 UDP6_INC_STATS_USER(sock_net(sk),
1107 UDP_MIB_OUTDATAGRAMS, is_udplite);
1111 static int udp_v6_push_pending_frames(struct sock *sk)
1113 struct sk_buff *skb;
1114 struct udp_sock *up = udp_sk(sk);
1118 if (up->pending == AF_INET)
1119 return udp_push_pending_frames(sk);
1121 /* ip6_finish_skb will release the cork, so make a copy of
1124 fl6 = inet_sk(sk)->cork.fl.u.ip6;
1126 skb = ip6_finish_skb(sk);
1130 err = udp_v6_send_skb(skb, &fl6);
1138 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1140 struct ipv6_txoptions opt_space;
1141 struct udp_sock *up = udp_sk(sk);
1142 struct inet_sock *inet = inet_sk(sk);
1143 struct ipv6_pinfo *np = inet6_sk(sk);
1144 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1145 struct in6_addr *daddr, *final_p, final;
1146 struct ipv6_txoptions *opt = NULL;
1147 struct ipv6_txoptions *opt_to_free = NULL;
1148 struct ip6_flowlabel *flowlabel = NULL;
1150 struct dst_entry *dst;
1151 int addr_len = msg->msg_namelen;
1156 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1159 int is_udplite = IS_UDPLITE(sk);
1160 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1162 /* destination address check */
1164 if (addr_len < offsetof(struct sockaddr, sa_data))
1167 switch (sin6->sin6_family) {
1169 if (addr_len < SIN6_LEN_RFC2133)
1171 daddr = &sin6->sin6_addr;
1174 goto do_udp_sendmsg;
1176 msg->msg_name = sin6 = NULL;
1177 msg->msg_namelen = addr_len = 0;
1183 } else if (!up->pending) {
1184 if (sk->sk_state != TCP_ESTABLISHED)
1185 return -EDESTADDRREQ;
1186 daddr = &sk->sk_v6_daddr;
1191 if (ipv6_addr_v4mapped(daddr)) {
1192 struct sockaddr_in sin;
1193 sin.sin_family = AF_INET;
1194 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1195 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1196 msg->msg_name = &sin;
1197 msg->msg_namelen = sizeof(sin);
1199 if (__ipv6_only_sock(sk))
1200 return -ENETUNREACH;
1201 return udp_sendmsg(sk, msg, len);
1205 if (up->pending == AF_INET)
1206 return udp_sendmsg(sk, msg, len);
1208 /* Rough check on arithmetic overflow,
1209 better check is made in ip6_append_data().
1211 if (len > INT_MAX - sizeof(struct udphdr))
1214 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1217 * There are pending frames.
1218 * The socket lock must be held while it's corked.
1221 if (likely(up->pending)) {
1222 if (unlikely(up->pending != AF_INET6)) {
1224 return -EAFNOSUPPORT;
1227 goto do_append_data;
1231 ulen += sizeof(struct udphdr);
1233 memset(&fl6, 0, sizeof(fl6));
1236 if (sin6->sin6_port == 0)
1239 fl6.fl6_dport = sin6->sin6_port;
1240 daddr = &sin6->sin6_addr;
1243 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1244 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1245 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1252 * Otherwise it will be difficult to maintain
1255 if (sk->sk_state == TCP_ESTABLISHED &&
1256 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1257 daddr = &sk->sk_v6_daddr;
1259 if (addr_len >= sizeof(struct sockaddr_in6) &&
1260 sin6->sin6_scope_id &&
1261 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1262 fl6.flowi6_oif = sin6->sin6_scope_id;
1264 if (sk->sk_state != TCP_ESTABLISHED)
1265 return -EDESTADDRREQ;
1267 fl6.fl6_dport = inet->inet_dport;
1268 daddr = &sk->sk_v6_daddr;
1269 fl6.flowlabel = np->flow_label;
1273 if (!fl6.flowi6_oif)
1274 fl6.flowi6_oif = sk->sk_bound_dev_if;
1276 if (!fl6.flowi6_oif)
1277 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1279 fl6.flowi6_mark = sk->sk_mark;
1281 if (msg->msg_controllen) {
1283 memset(opt, 0, sizeof(struct ipv6_txoptions));
1284 opt->tot_len = sizeof(*opt);
1286 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt,
1287 &hlimit, &tclass, &dontfrag);
1289 fl6_sock_release(flowlabel);
1292 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1293 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1297 if (!(opt->opt_nflen|opt->opt_flen))
1302 opt = txopt_get(np);
1306 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1307 opt = ipv6_fixup_options(&opt_space, opt);
1309 fl6.flowi6_proto = sk->sk_protocol;
1310 if (!ipv6_addr_any(daddr))
1313 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1314 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1315 fl6.saddr = np->saddr;
1316 fl6.fl6_sport = inet->inet_sport;
1318 final_p = fl6_update_dst(&fl6, opt, &final);
1322 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1323 fl6.flowi6_oif = np->mcast_oif;
1325 } else if (!fl6.flowi6_oif)
1326 fl6.flowi6_oif = np->ucast_oif;
1328 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1330 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p);
1338 hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1341 tclass = np->tclass;
1343 if (msg->msg_flags&MSG_CONFIRM)
1347 /* Lockless fast path for the non-corking case */
1349 struct sk_buff *skb;
1351 skb = ip6_make_skb(sk, getfrag, msg, ulen,
1352 sizeof(struct udphdr), hlimit, tclass, opt,
1353 &fl6, (struct rt6_info *)dst,
1354 msg->msg_flags, dontfrag);
1356 if (!IS_ERR_OR_NULL(skb))
1357 err = udp_v6_send_skb(skb, &fl6);
1362 if (unlikely(up->pending)) {
1363 /* The socket is already corked while preparing it. */
1364 /* ... which is an evident application bug. --ANK */
1367 net_dbg_ratelimited("udp cork app bug 2\n");
1372 up->pending = AF_INET6;
1376 dontfrag = np->dontfrag;
1378 err = ip6_append_data(sk, getfrag, msg, ulen,
1379 sizeof(struct udphdr), hlimit, tclass, opt, &fl6,
1380 (struct rt6_info *)dst,
1381 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag);
1383 udp_v6_flush_pending_frames(sk);
1385 err = udp_v6_push_pending_frames(sk);
1386 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1390 err = np->recverr ? net_xmit_errno(err) : 0;
1396 ip6_dst_store(sk, dst,
1397 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ?
1398 &sk->sk_v6_daddr : NULL,
1399 #ifdef CONFIG_IPV6_SUBTREES
1400 ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1412 fl6_sock_release(flowlabel);
1413 txopt_put(opt_to_free);
1417 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1418 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1419 * we don't have a good statistic (IpOutDiscards but it can be too many
1420 * things). We could add another new stat but at least for now that
1421 * seems like overkill.
1423 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1424 UDP6_INC_STATS_USER(sock_net(sk),
1425 UDP_MIB_SNDBUFERRORS, is_udplite);
1431 if (!(msg->msg_flags&MSG_PROBE) || len)
1432 goto back_from_confirm;
1437 void udpv6_destroy_sock(struct sock *sk)
1439 struct udp_sock *up = udp_sk(sk);
1441 udp_v6_flush_pending_frames(sk);
1444 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
1445 void (*encap_destroy)(struct sock *sk);
1446 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1451 inet6_destroy_sock(sk);
1455 * Socket option code for UDP
1457 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1458 char __user *optval, unsigned int optlen)
1460 if (level == SOL_UDP || level == SOL_UDPLITE)
1461 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1462 udp_v6_push_pending_frames);
1463 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1466 #ifdef CONFIG_COMPAT
1467 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1468 char __user *optval, unsigned int optlen)
1470 if (level == SOL_UDP || level == SOL_UDPLITE)
1471 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1472 udp_v6_push_pending_frames);
1473 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1477 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1478 char __user *optval, int __user *optlen)
1480 if (level == SOL_UDP || level == SOL_UDPLITE)
1481 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1482 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1485 #ifdef CONFIG_COMPAT
1486 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1487 char __user *optval, int __user *optlen)
1489 if (level == SOL_UDP || level == SOL_UDPLITE)
1490 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1491 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1495 static const struct inet6_protocol udpv6_protocol = {
1496 .handler = udpv6_rcv,
1497 .err_handler = udpv6_err,
1498 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1501 /* ------------------------------------------------------------------------ */
1502 #ifdef CONFIG_PROC_FS
1503 int udp6_seq_show(struct seq_file *seq, void *v)
1505 if (v == SEQ_START_TOKEN) {
1506 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1508 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1509 struct inet_sock *inet = inet_sk(v);
1510 __u16 srcp = ntohs(inet->inet_sport);
1511 __u16 destp = ntohs(inet->inet_dport);
1512 ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket);
1517 static const struct file_operations udp6_afinfo_seq_fops = {
1518 .owner = THIS_MODULE,
1519 .open = udp_seq_open,
1521 .llseek = seq_lseek,
1522 .release = seq_release_net
1525 static struct udp_seq_afinfo udp6_seq_afinfo = {
1528 .udp_table = &udp_table,
1529 .seq_fops = &udp6_afinfo_seq_fops,
1531 .show = udp6_seq_show,
1535 int __net_init udp6_proc_init(struct net *net)
1537 return udp_proc_register(net, &udp6_seq_afinfo);
1540 void udp6_proc_exit(struct net *net)
1542 udp_proc_unregister(net, &udp6_seq_afinfo);
1544 #endif /* CONFIG_PROC_FS */
1546 void udp_v6_clear_sk(struct sock *sk, int size)
1548 struct inet_sock *inet = inet_sk(sk);
1550 /* we do not want to clear pinet6 field, because of RCU lookups */
1551 sk_prot_clear_portaddr_nulls(sk, offsetof(struct inet_sock, pinet6));
1553 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1554 memset(&inet->pinet6 + 1, 0, size);
1557 /* ------------------------------------------------------------------------ */
1559 struct proto udpv6_prot = {
1561 .owner = THIS_MODULE,
1562 .close = udp_lib_close,
1563 .connect = ip6_datagram_connect,
1564 .disconnect = udp_disconnect,
1566 .destroy = udpv6_destroy_sock,
1567 .setsockopt = udpv6_setsockopt,
1568 .getsockopt = udpv6_getsockopt,
1569 .sendmsg = udpv6_sendmsg,
1570 .recvmsg = udpv6_recvmsg,
1571 .backlog_rcv = __udpv6_queue_rcv_skb,
1572 .hash = udp_lib_hash,
1573 .unhash = udp_lib_unhash,
1574 .rehash = udp_v6_rehash,
1575 .get_port = udp_v6_get_port,
1576 .memory_allocated = &udp_memory_allocated,
1577 .sysctl_mem = sysctl_udp_mem,
1578 .sysctl_wmem = &sysctl_udp_wmem_min,
1579 .sysctl_rmem = &sysctl_udp_rmem_min,
1580 .obj_size = sizeof(struct udp6_sock),
1581 .slab_flags = SLAB_DESTROY_BY_RCU,
1582 .h.udp_table = &udp_table,
1583 #ifdef CONFIG_COMPAT
1584 .compat_setsockopt = compat_udpv6_setsockopt,
1585 .compat_getsockopt = compat_udpv6_getsockopt,
1587 .clear_sk = udp_v6_clear_sk,
1590 static struct inet_protosw udpv6_protosw = {
1592 .protocol = IPPROTO_UDP,
1593 .prot = &udpv6_prot,
1594 .ops = &inet6_dgram_ops,
1595 .flags = INET_PROTOSW_PERMANENT,
1598 int __init udpv6_init(void)
1602 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1606 ret = inet6_register_protosw(&udpv6_protosw);
1608 goto out_udpv6_protocol;
1613 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1617 void udpv6_exit(void)
1619 inet6_unregister_protosw(&udpv6_protosw);
1620 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);