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 <linux/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/inet_hashtables.h>
50 #include <net/inet6_hashtables.h>
51 #include <net/busy_poll.h>
52 #include <net/sock_reuseport.h>
54 #include <linux/proc_fs.h>
55 #include <linux/seq_file.h>
56 #include <trace/events/skb.h>
59 static bool udp6_lib_exact_dif_match(struct net *net, struct sk_buff *skb)
61 #if defined(CONFIG_NET_L3_MASTER_DEV)
62 if (!net->ipv4.sysctl_udp_l3mdev_accept &&
63 skb && ipv6_l3mdev_skb(IP6CB(skb)->flags))
69 static void udpv6_destruct_sock(struct sock *sk)
71 udp_destruct_common(sk);
72 inet6_sock_destruct(sk);
75 int udpv6_init_sock(struct sock *sk)
77 skb_queue_head_init(&udp_sk(sk)->reader_queue);
78 sk->sk_destruct = udpv6_destruct_sock;
82 static u32 udp6_ehashfn(const struct net *net,
83 const struct in6_addr *laddr,
85 const struct in6_addr *faddr,
88 static u32 udp6_ehash_secret __read_mostly;
89 static u32 udp_ipv6_hash_secret __read_mostly;
93 net_get_random_once(&udp6_ehash_secret,
94 sizeof(udp6_ehash_secret));
95 net_get_random_once(&udp_ipv6_hash_secret,
96 sizeof(udp_ipv6_hash_secret));
98 lhash = (__force u32)laddr->s6_addr32[3];
99 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
101 return __inet6_ehashfn(lhash, lport, fhash, fport,
102 udp6_ehash_secret + net_hash_mix(net));
105 static u32 udp6_portaddr_hash(const struct net *net,
106 const struct in6_addr *addr6,
109 unsigned int hash, mix = net_hash_mix(net);
111 if (ipv6_addr_any(addr6))
112 hash = jhash_1word(0, mix);
113 else if (ipv6_addr_v4mapped(addr6))
114 hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix);
116 hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix);
121 int udp_v6_get_port(struct sock *sk, unsigned short snum)
123 unsigned int hash2_nulladdr =
124 udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
125 unsigned int hash2_partial =
126 udp6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
128 /* precompute partial secondary hash */
129 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
130 return udp_lib_get_port(sk, snum, hash2_nulladdr);
133 static void udp_v6_rehash(struct sock *sk)
135 u16 new_hash = udp6_portaddr_hash(sock_net(sk),
136 &sk->sk_v6_rcv_saddr,
137 inet_sk(sk)->inet_num);
139 udp_lib_rehash(sk, new_hash);
142 static int compute_score(struct sock *sk, struct net *net,
143 const struct in6_addr *saddr, __be16 sport,
144 const struct in6_addr *daddr, unsigned short hnum,
145 int dif, int sdif, bool exact_dif)
148 struct inet_sock *inet;
150 if (!net_eq(sock_net(sk), net) ||
151 udp_sk(sk)->udp_port_hash != hnum ||
152 sk->sk_family != PF_INET6)
158 if (inet->inet_dport) {
159 if (inet->inet_dport != sport)
164 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
165 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
170 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
171 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
176 if (sk->sk_bound_dev_if || exact_dif) {
177 bool dev_match = (sk->sk_bound_dev_if == dif ||
178 sk->sk_bound_dev_if == sdif);
182 if (sk->sk_bound_dev_if)
186 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
192 /* called with rcu_read_lock() */
193 static struct sock *udp6_lib_lookup2(struct net *net,
194 const struct in6_addr *saddr, __be16 sport,
195 const struct in6_addr *daddr, unsigned int hnum,
196 int dif, int sdif, bool exact_dif,
197 struct udp_hslot *hslot2, struct sk_buff *skb)
199 struct sock *sk, *result;
200 int score, badness, matches = 0, reuseport = 0;
205 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
206 score = compute_score(sk, net, saddr, sport,
207 daddr, hnum, dif, sdif, exact_dif);
208 if (score > badness) {
209 reuseport = sk->sk_reuseport;
211 hash = udp6_ehashfn(net, daddr, hnum,
214 result = reuseport_select_sock(sk, hash, skb,
215 sizeof(struct udphdr));
222 } else if (score == badness && reuseport) {
224 if (reciprocal_scale(hash, matches) == 0)
226 hash = next_pseudo_random32(hash);
232 /* rcu_read_lock() must be held */
233 struct sock *__udp6_lib_lookup(struct net *net,
234 const struct in6_addr *saddr, __be16 sport,
235 const struct in6_addr *daddr, __be16 dport,
236 int dif, int sdif, struct udp_table *udptable,
239 struct sock *sk, *result;
240 unsigned short hnum = ntohs(dport);
241 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
242 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
243 bool exact_dif = udp6_lib_exact_dif_match(net, skb);
244 int score, badness, matches = 0, reuseport = 0;
247 if (hslot->count > 10) {
248 hash2 = udp6_portaddr_hash(net, daddr, hnum);
249 slot2 = hash2 & udptable->mask;
250 hslot2 = &udptable->hash2[slot2];
251 if (hslot->count < hslot2->count)
254 result = udp6_lib_lookup2(net, saddr, sport,
255 daddr, hnum, dif, sdif, exact_dif,
258 unsigned int old_slot2 = slot2;
259 hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
260 slot2 = hash2 & udptable->mask;
261 /* avoid searching the same slot again. */
262 if (unlikely(slot2 == old_slot2))
265 hslot2 = &udptable->hash2[slot2];
266 if (hslot->count < hslot2->count)
269 result = udp6_lib_lookup2(net, saddr, sport,
270 daddr, hnum, dif, sdif,
279 sk_for_each_rcu(sk, &hslot->head) {
280 score = compute_score(sk, net, saddr, sport, daddr, hnum, dif,
282 if (score > badness) {
283 reuseport = sk->sk_reuseport;
285 hash = udp6_ehashfn(net, daddr, hnum,
287 result = reuseport_select_sock(sk, hash, skb,
288 sizeof(struct udphdr));
295 } else if (score == badness && reuseport) {
297 if (reciprocal_scale(hash, matches) == 0)
299 hash = next_pseudo_random32(hash);
304 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
306 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
307 __be16 sport, __be16 dport,
308 struct udp_table *udptable)
310 const struct ipv6hdr *iph = ipv6_hdr(skb);
312 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
313 &iph->daddr, dport, inet6_iif(skb),
314 inet6_sdif(skb), udptable, skb);
317 struct sock *udp6_lib_lookup_skb(struct sk_buff *skb,
318 __be16 sport, __be16 dport)
320 const struct ipv6hdr *iph = ipv6_hdr(skb);
322 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
323 &iph->daddr, dport, inet6_iif(skb),
324 inet6_sdif(skb), &udp_table, NULL);
326 EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb);
328 /* Must be called under rcu_read_lock().
329 * Does increment socket refcount.
331 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
332 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \
333 IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
334 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
335 const struct in6_addr *daddr, __be16 dport, int dif)
339 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport,
340 dif, 0, &udp_table, NULL);
341 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
345 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
348 /* do not use the scratch area len for jumbogram: their length execeeds the
349 * scratch area space; note that the IP6CB flags is still in the first
350 * cacheline, so checking for jumbograms is cheap
352 static int udp6_skb_len(struct sk_buff *skb)
354 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb);
358 * This should be easy, if there is something there we
359 * return it, otherwise we block.
362 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
363 int noblock, int flags, int *addr_len)
365 struct ipv6_pinfo *np = inet6_sk(sk);
366 struct inet_sock *inet = inet_sk(sk);
368 unsigned int ulen, copied;
369 int peeked, peeking, off;
371 int is_udplite = IS_UDPLITE(sk);
372 bool checksum_valid = false;
375 if (flags & MSG_ERRQUEUE)
376 return ipv6_recv_error(sk, msg, len, addr_len);
378 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
379 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
382 peeking = flags & MSG_PEEK;
383 off = sk_peek_offset(sk, flags);
384 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
388 ulen = udp6_skb_len(skb);
390 if (copied > ulen - off)
392 else if (copied < ulen)
393 msg->msg_flags |= MSG_TRUNC;
395 is_udp4 = (skb->protocol == htons(ETH_P_IP));
398 * If checksum is needed at all, try to do it while copying the
399 * data. If the data is truncated, or if we only want a partial
400 * coverage checksum (UDP-Lite), do it before the copy.
403 if (copied < ulen || peeking ||
404 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
405 checksum_valid = udp_skb_csum_unnecessary(skb) ||
406 !__udp_lib_checksum_complete(skb);
411 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
412 if (udp_skb_is_linear(skb))
413 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
415 err = skb_copy_datagram_msg(skb, off, msg, copied);
417 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
423 atomic_inc(&sk->sk_drops);
425 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
428 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
436 UDP_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS,
439 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS,
443 sock_recv_ts_and_drops(msg, sk, skb);
445 /* Copy the address. */
447 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
448 sin6->sin6_family = AF_INET6;
449 sin6->sin6_port = udp_hdr(skb)->source;
450 sin6->sin6_flowinfo = 0;
453 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
455 sin6->sin6_scope_id = 0;
457 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
458 sin6->sin6_scope_id =
459 ipv6_iface_scope_id(&sin6->sin6_addr,
462 *addr_len = sizeof(*sin6);
466 ip6_datagram_recv_common_ctl(sk, msg, skb);
469 if (inet->cmsg_flags)
470 ip_cmsg_recv_offset(msg, sk, skb,
471 sizeof(struct udphdr), off);
474 ip6_datagram_recv_specific_ctl(sk, msg, skb);
478 if (flags & MSG_TRUNC)
481 skb_consume_udp(sk, skb, peeking ? -err : err);
485 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
486 udp_skb_destructor)) {
488 UDP_INC_STATS(sock_net(sk),
489 UDP_MIB_CSUMERRORS, is_udplite);
490 UDP_INC_STATS(sock_net(sk),
491 UDP_MIB_INERRORS, is_udplite);
493 UDP6_INC_STATS(sock_net(sk),
494 UDP_MIB_CSUMERRORS, is_udplite);
495 UDP6_INC_STATS(sock_net(sk),
496 UDP_MIB_INERRORS, is_udplite);
501 /* starting over for a new packet, but check if we need to yield */
503 msg->msg_flags &= ~MSG_TRUNC;
507 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
508 u8 type, u8 code, int offset, __be32 info,
509 struct udp_table *udptable)
511 struct ipv6_pinfo *np;
512 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
513 const struct in6_addr *saddr = &hdr->saddr;
514 const struct in6_addr *daddr = &hdr->daddr;
515 struct udphdr *uh = (struct udphdr *)(skb->data+offset);
519 struct net *net = dev_net(skb->dev);
521 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
522 inet6_iif(skb), 0, udptable, NULL);
524 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
529 harderr = icmpv6_err_convert(type, code, &err);
532 if (type == ICMPV6_PKT_TOOBIG) {
533 if (!ip6_sk_accept_pmtu(sk))
535 ip6_sk_update_pmtu(skb, sk, info);
536 if (np->pmtudisc != IPV6_PMTUDISC_DONT)
539 if (type == NDISC_REDIRECT) {
540 ip6_sk_redirect(skb, sk);
545 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
548 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
552 sk->sk_error_report(sk);
557 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
561 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
562 sock_rps_save_rxhash(sk, skb);
563 sk_mark_napi_id(sk, skb);
564 sk_incoming_cpu_update(sk);
566 sk_mark_napi_id_once(sk, skb);
569 rc = __udp_enqueue_schedule_skb(sk, skb);
571 int is_udplite = IS_UDPLITE(sk);
573 /* Note that an ENOMEM error is charged twice */
575 UDP6_INC_STATS(sock_net(sk),
576 UDP_MIB_RCVBUFERRORS, is_udplite);
577 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
585 static __inline__ void udpv6_err(struct sk_buff *skb,
586 struct inet6_skb_parm *opt, u8 type,
587 u8 code, int offset, __be32 info)
589 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
592 static struct static_key udpv6_encap_needed __read_mostly;
593 void udpv6_encap_enable(void)
595 static_key_enable(&udpv6_encap_needed);
597 EXPORT_SYMBOL(udpv6_encap_enable);
599 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
601 struct udp_sock *up = udp_sk(sk);
602 int is_udplite = IS_UDPLITE(sk);
604 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
607 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
608 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
611 * This is an encapsulation socket so pass the skb to
612 * the socket's udp_encap_rcv() hook. Otherwise, just
613 * fall through and pass this up the UDP socket.
614 * up->encap_rcv() returns the following value:
615 * =0 if skb was successfully passed to the encap
616 * handler or was discarded by it.
617 * >0 if skb should be passed on to UDP.
618 * <0 if skb should be resubmitted as proto -N
621 /* if we're overly short, let UDP handle it */
622 encap_rcv = ACCESS_ONCE(up->encap_rcv);
626 /* Verify checksum before giving to encap */
627 if (udp_lib_checksum_complete(skb))
630 ret = encap_rcv(sk, skb);
632 __UDP_INC_STATS(sock_net(sk),
639 /* FALLTHROUGH -- it's a UDP Packet */
643 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
645 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
647 if (up->pcrlen == 0) { /* full coverage was set */
648 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
649 UDP_SKB_CB(skb)->cscov, skb->len);
652 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
653 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
654 UDP_SKB_CB(skb)->cscov, up->pcrlen);
659 prefetch(&sk->sk_rmem_alloc);
660 if (rcu_access_pointer(sk->sk_filter) &&
661 udp_lib_checksum_complete(skb))
664 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
667 udp_csum_pull_header(skb);
671 return __udpv6_queue_rcv_skb(sk, skb);
674 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
676 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
677 atomic_inc(&sk->sk_drops);
682 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
683 __be16 loc_port, const struct in6_addr *loc_addr,
684 __be16 rmt_port, const struct in6_addr *rmt_addr,
685 int dif, unsigned short hnum)
687 struct inet_sock *inet = inet_sk(sk);
689 if (!net_eq(sock_net(sk), net))
692 if (udp_sk(sk)->udp_port_hash != hnum ||
693 sk->sk_family != PF_INET6 ||
694 (inet->inet_dport && inet->inet_dport != rmt_port) ||
695 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
696 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
697 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif) ||
698 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
699 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
701 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
706 static void udp6_csum_zero_error(struct sk_buff *skb)
708 /* RFC 2460 section 8.1 says that we SHOULD log
709 * this error. Well, it is reasonable.
711 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
712 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
713 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
717 * Note: called only from the BH handler context,
718 * so we don't need to lock the hashes.
720 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
721 const struct in6_addr *saddr, const struct in6_addr *daddr,
722 struct udp_table *udptable, int proto)
724 struct sock *sk, *first = NULL;
725 const struct udphdr *uh = udp_hdr(skb);
726 unsigned short hnum = ntohs(uh->dest);
727 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
728 unsigned int offset = offsetof(typeof(*sk), sk_node);
729 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
730 int dif = inet6_iif(skb);
731 struct hlist_node *node;
732 struct sk_buff *nskb;
735 hash2_any = udp6_portaddr_hash(net, &in6addr_any, hnum) &
737 hash2 = udp6_portaddr_hash(net, daddr, hnum) & udptable->mask;
739 hslot = &udptable->hash2[hash2];
740 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
743 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
744 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
745 uh->source, saddr, dif, hnum))
747 /* If zero checksum and no_check is not on for
748 * the socket then skip it.
750 if (!uh->check && !udp_sk(sk)->no_check6_rx)
756 nskb = skb_clone(skb, GFP_ATOMIC);
757 if (unlikely(!nskb)) {
758 atomic_inc(&sk->sk_drops);
759 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
761 __UDP6_INC_STATS(net, UDP_MIB_INERRORS,
766 if (udpv6_queue_rcv_skb(sk, nskb) > 0)
770 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
771 if (use_hash2 && hash2 != hash2_any) {
777 if (udpv6_queue_rcv_skb(first, skb) > 0)
781 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
782 proto == IPPROTO_UDPLITE);
787 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
789 if (udp_sk_rx_dst_set(sk, dst)) {
790 const struct rt6_info *rt = (const struct rt6_info *)dst;
792 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
796 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
797 * return code conversion for ip layer consumption
799 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
804 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
805 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
808 ret = udpv6_queue_rcv_skb(sk, skb);
810 /* a return value > 0 means to resubmit the input */
816 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
819 const struct in6_addr *saddr, *daddr;
820 struct net *net = dev_net(skb->dev);
825 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
828 saddr = &ipv6_hdr(skb)->saddr;
829 daddr = &ipv6_hdr(skb)->daddr;
832 ulen = ntohs(uh->len);
836 if (proto == IPPROTO_UDP) {
837 /* UDP validates ulen. */
839 /* Check for jumbo payload */
843 if (ulen < sizeof(*uh))
846 if (ulen < skb->len) {
847 if (pskb_trim_rcsum(skb, ulen))
849 saddr = &ipv6_hdr(skb)->saddr;
850 daddr = &ipv6_hdr(skb)->daddr;
855 if (udp6_csum_init(skb, uh, proto))
858 /* Check if the socket is already available, e.g. due to early demux */
859 sk = skb_steal_sock(skb);
861 struct dst_entry *dst = skb_dst(skb);
864 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
865 udp6_sk_rx_dst_set(sk, dst);
867 if (!uh->check && !udp_sk(sk)->no_check6_rx) {
869 goto report_csum_error;
872 ret = udp6_unicast_rcv_skb(sk, skb, uh);
878 * Multicast receive code
880 if (ipv6_addr_is_multicast(daddr))
881 return __udp6_lib_mcast_deliver(net, skb,
882 saddr, daddr, udptable, proto);
885 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
887 if (!uh->check && !udp_sk(sk)->no_check6_rx)
888 goto report_csum_error;
889 return udp6_unicast_rcv_skb(sk, skb, uh);
893 goto report_csum_error;
895 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
898 if (udp_lib_checksum_complete(skb))
901 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
902 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
908 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
909 proto == IPPROTO_UDPLITE ? "-Lite" : "",
910 saddr, ntohs(uh->source),
912 daddr, ntohs(uh->dest));
916 udp6_csum_zero_error(skb);
918 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
920 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
926 static struct sock *__udp6_lib_demux_lookup(struct net *net,
927 __be16 loc_port, const struct in6_addr *loc_addr,
928 __be16 rmt_port, const struct in6_addr *rmt_addr,
931 unsigned short hnum = ntohs(loc_port);
932 unsigned int hash2 = udp6_portaddr_hash(net, loc_addr, hnum);
933 unsigned int slot2 = hash2 & udp_table.mask;
934 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
935 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
938 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
939 if (sk->sk_state == TCP_ESTABLISHED &&
940 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif, sdif))
942 /* Only check first socket in chain */
948 void udp_v6_early_demux(struct sk_buff *skb)
950 struct net *net = dev_net(skb->dev);
951 const struct udphdr *uh;
953 struct dst_entry *dst;
954 int dif = skb->dev->ifindex;
955 int sdif = inet6_sdif(skb);
957 if (!pskb_may_pull(skb, skb_transport_offset(skb) +
958 sizeof(struct udphdr)))
963 if (skb->pkt_type == PACKET_HOST)
964 sk = __udp6_lib_demux_lookup(net, uh->dest,
965 &ipv6_hdr(skb)->daddr,
966 uh->source, &ipv6_hdr(skb)->saddr,
971 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
975 skb->destructor = sock_efree;
976 dst = rcu_dereference(sk->sk_rx_dst);
979 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
981 /* set noref for now.
982 * any place which wants to hold dst has to call
985 skb_dst_set_noref(skb, dst);
989 static __inline__ int udpv6_rcv(struct sk_buff *skb)
991 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
995 * Throw away all pending data and cancel the corking. Socket is locked.
997 static void udp_v6_flush_pending_frames(struct sock *sk)
999 struct udp_sock *up = udp_sk(sk);
1001 if (up->pending == AF_INET)
1002 udp_flush_pending_frames(sk);
1003 else if (up->pending) {
1006 ip6_flush_pending_frames(sk);
1011 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
1012 * @sk: socket we are sending on
1013 * @skb: sk_buff containing the filled-in UDP header
1014 * (checksum field must be zeroed out)
1016 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
1017 const struct in6_addr *saddr,
1018 const struct in6_addr *daddr, int len)
1020 unsigned int offset;
1021 struct udphdr *uh = udp_hdr(skb);
1022 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
1026 /* Only one fragment on the socket. */
1027 skb->csum_start = skb_transport_header(skb) - skb->head;
1028 skb->csum_offset = offsetof(struct udphdr, check);
1029 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1032 * HW-checksum won't work as there are two or more
1033 * fragments on the socket so that all csums of sk_buffs
1034 * should be together
1036 offset = skb_transport_offset(skb);
1037 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1040 skb->ip_summed = CHECKSUM_NONE;
1043 csum = csum_add(csum, frags->csum);
1044 } while ((frags = frags->next));
1046 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1049 uh->check = CSUM_MANGLED_0;
1057 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6)
1059 struct sock *sk = skb->sk;
1062 int is_udplite = IS_UDPLITE(sk);
1064 int offset = skb_transport_offset(skb);
1065 int len = skb->len - offset;
1068 * Create a UDP header
1071 uh->source = fl6->fl6_sport;
1072 uh->dest = fl6->fl6_dport;
1073 uh->len = htons(len);
1077 csum = udplite_csum(skb);
1078 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */
1079 skb->ip_summed = CHECKSUM_NONE;
1081 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1082 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1085 csum = udp_csum(skb);
1087 /* add protocol-dependent pseudo-header */
1088 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1089 len, fl6->flowi6_proto, csum);
1091 uh->check = CSUM_MANGLED_0;
1094 err = ip6_send_skb(skb);
1096 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1097 UDP6_INC_STATS(sock_net(sk),
1098 UDP_MIB_SNDBUFERRORS, is_udplite);
1102 UDP6_INC_STATS(sock_net(sk),
1103 UDP_MIB_OUTDATAGRAMS, is_udplite);
1108 static int udp_v6_push_pending_frames(struct sock *sk)
1110 struct sk_buff *skb;
1111 struct udp_sock *up = udp_sk(sk);
1115 if (up->pending == AF_INET)
1116 return udp_push_pending_frames(sk);
1118 /* ip6_finish_skb will release the cork, so make a copy of
1121 fl6 = inet_sk(sk)->cork.fl.u.ip6;
1123 skb = ip6_finish_skb(sk);
1127 err = udp_v6_send_skb(skb, &fl6);
1135 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1137 struct ipv6_txoptions opt_space;
1138 struct udp_sock *up = udp_sk(sk);
1139 struct inet_sock *inet = inet_sk(sk);
1140 struct ipv6_pinfo *np = inet6_sk(sk);
1141 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1142 struct in6_addr *daddr, *final_p, final;
1143 struct ipv6_txoptions *opt = NULL;
1144 struct ipv6_txoptions *opt_to_free = NULL;
1145 struct ip6_flowlabel *flowlabel = NULL;
1147 struct dst_entry *dst;
1148 struct ipcm6_cookie ipc6;
1149 int addr_len = msg->msg_namelen;
1151 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE;
1154 int is_udplite = IS_UDPLITE(sk);
1155 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1156 struct sockcm_cookie sockc;
1161 sockc.tsflags = sk->sk_tsflags;
1163 /* destination address check */
1165 if (addr_len < offsetof(struct sockaddr, sa_data))
1168 switch (sin6->sin6_family) {
1170 if (addr_len < SIN6_LEN_RFC2133)
1172 daddr = &sin6->sin6_addr;
1173 if (ipv6_addr_any(daddr) &&
1174 ipv6_addr_v4mapped(&np->saddr))
1175 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1179 goto do_udp_sendmsg;
1181 msg->msg_name = sin6 = NULL;
1182 msg->msg_namelen = addr_len = 0;
1188 } else if (!up->pending) {
1189 if (sk->sk_state != TCP_ESTABLISHED)
1190 return -EDESTADDRREQ;
1191 daddr = &sk->sk_v6_daddr;
1196 if (ipv6_addr_v4mapped(daddr)) {
1197 struct sockaddr_in sin;
1198 sin.sin_family = AF_INET;
1199 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1200 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1201 msg->msg_name = &sin;
1202 msg->msg_namelen = sizeof(sin);
1204 if (__ipv6_only_sock(sk))
1205 return -ENETUNREACH;
1206 return udp_sendmsg(sk, msg, len);
1210 if (up->pending == AF_INET)
1211 return udp_sendmsg(sk, msg, len);
1213 /* Rough check on arithmetic overflow,
1214 better check is made in ip6_append_data().
1216 if (len > INT_MAX - sizeof(struct udphdr))
1219 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1222 * There are pending frames.
1223 * The socket lock must be held while it's corked.
1226 if (likely(up->pending)) {
1227 if (unlikely(up->pending != AF_INET6)) {
1229 return -EAFNOSUPPORT;
1232 goto do_append_data;
1236 ulen += sizeof(struct udphdr);
1238 memset(&fl6, 0, sizeof(fl6));
1241 if (sin6->sin6_port == 0)
1244 fl6.fl6_dport = sin6->sin6_port;
1245 daddr = &sin6->sin6_addr;
1248 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1249 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1250 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1257 * Otherwise it will be difficult to maintain
1260 if (sk->sk_state == TCP_ESTABLISHED &&
1261 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1262 daddr = &sk->sk_v6_daddr;
1264 if (addr_len >= sizeof(struct sockaddr_in6) &&
1265 sin6->sin6_scope_id &&
1266 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1267 fl6.flowi6_oif = sin6->sin6_scope_id;
1269 if (sk->sk_state != TCP_ESTABLISHED)
1270 return -EDESTADDRREQ;
1272 fl6.fl6_dport = inet->inet_dport;
1273 daddr = &sk->sk_v6_daddr;
1274 fl6.flowlabel = np->flow_label;
1278 if (!fl6.flowi6_oif)
1279 fl6.flowi6_oif = sk->sk_bound_dev_if;
1281 if (!fl6.flowi6_oif)
1282 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1284 fl6.flowi6_mark = sk->sk_mark;
1285 fl6.flowi6_uid = sk->sk_uid;
1287 if (msg->msg_controllen) {
1289 memset(opt, 0, sizeof(struct ipv6_txoptions));
1290 opt->tot_len = sizeof(*opt);
1293 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, &ipc6, &sockc);
1295 fl6_sock_release(flowlabel);
1298 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1299 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1303 if (!(opt->opt_nflen|opt->opt_flen))
1308 opt = txopt_get(np);
1312 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1313 opt = ipv6_fixup_options(&opt_space, opt);
1316 fl6.flowi6_proto = sk->sk_protocol;
1317 if (!ipv6_addr_any(daddr))
1320 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1321 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1322 fl6.saddr = np->saddr;
1323 fl6.fl6_sport = inet->inet_sport;
1325 final_p = fl6_update_dst(&fl6, opt, &final);
1329 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1330 fl6.flowi6_oif = np->mcast_oif;
1332 } else if (!fl6.flowi6_oif)
1333 fl6.flowi6_oif = np->ucast_oif;
1335 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1337 if (ipc6.tclass < 0)
1338 ipc6.tclass = np->tclass;
1340 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel);
1342 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p);
1349 if (ipc6.hlimit < 0)
1350 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1352 if (msg->msg_flags&MSG_CONFIRM)
1356 /* Lockless fast path for the non-corking case */
1358 struct sk_buff *skb;
1360 skb = ip6_make_skb(sk, getfrag, msg, ulen,
1361 sizeof(struct udphdr), &ipc6,
1362 &fl6, (struct rt6_info *)dst,
1363 msg->msg_flags, &sockc);
1365 if (!IS_ERR_OR_NULL(skb))
1366 err = udp_v6_send_skb(skb, &fl6);
1371 if (unlikely(up->pending)) {
1372 /* The socket is already corked while preparing it. */
1373 /* ... which is an evident application bug. --ANK */
1376 net_dbg_ratelimited("udp cork app bug 2\n");
1381 up->pending = AF_INET6;
1384 if (ipc6.dontfrag < 0)
1385 ipc6.dontfrag = np->dontfrag;
1387 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr),
1388 &ipc6, &fl6, (struct rt6_info *)dst,
1389 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, &sockc);
1391 udp_v6_flush_pending_frames(sk);
1393 err = udp_v6_push_pending_frames(sk);
1394 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1398 err = np->recverr ? net_xmit_errno(err) : 0;
1404 ip6_dst_store(sk, dst,
1405 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ?
1406 &sk->sk_v6_daddr : NULL,
1407 #ifdef CONFIG_IPV6_SUBTREES
1408 ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1420 fl6_sock_release(flowlabel);
1421 txopt_put(opt_to_free);
1425 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1426 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1427 * we don't have a good statistic (IpOutDiscards but it can be too many
1428 * things). We could add another new stat but at least for now that
1429 * seems like overkill.
1431 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1432 UDP6_INC_STATS(sock_net(sk),
1433 UDP_MIB_SNDBUFERRORS, is_udplite);
1438 if (msg->msg_flags & MSG_PROBE)
1439 dst_confirm_neigh(dst, &fl6.daddr);
1440 if (!(msg->msg_flags&MSG_PROBE) || len)
1441 goto back_from_confirm;
1446 void udpv6_destroy_sock(struct sock *sk)
1448 struct udp_sock *up = udp_sk(sk);
1451 /* protects from races with udp_abort() */
1452 sock_set_flag(sk, SOCK_DEAD);
1453 udp_v6_flush_pending_frames(sk);
1456 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
1457 void (*encap_destroy)(struct sock *sk);
1458 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1465 * Socket option code for UDP
1467 int 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 ipv6_setsockopt(sk, level, optname, optval, optlen);
1476 #ifdef CONFIG_COMPAT
1477 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1478 char __user *optval, unsigned int optlen)
1480 if (level == SOL_UDP || level == SOL_UDPLITE)
1481 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1482 udp_v6_push_pending_frames);
1483 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1487 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1488 char __user *optval, int __user *optlen)
1490 if (level == SOL_UDP || level == SOL_UDPLITE)
1491 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1492 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1495 #ifdef CONFIG_COMPAT
1496 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1497 char __user *optval, int __user *optlen)
1499 if (level == SOL_UDP || level == SOL_UDPLITE)
1500 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1501 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1505 static const struct inet6_protocol udpv6_protocol = {
1506 .handler = udpv6_rcv,
1507 .err_handler = udpv6_err,
1508 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1511 /* ------------------------------------------------------------------------ */
1512 #ifdef CONFIG_PROC_FS
1513 int udp6_seq_show(struct seq_file *seq, void *v)
1515 if (v == SEQ_START_TOKEN) {
1516 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1518 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1519 struct inet_sock *inet = inet_sk(v);
1520 __u16 srcp = ntohs(inet->inet_sport);
1521 __u16 destp = ntohs(inet->inet_dport);
1522 __ip6_dgram_sock_seq_show(seq, v, srcp, destp,
1523 udp_rqueue_get(v), bucket);
1528 static const struct file_operations udp6_afinfo_seq_fops = {
1529 .owner = THIS_MODULE,
1530 .open = udp_seq_open,
1532 .llseek = seq_lseek,
1533 .release = seq_release_net
1536 static struct udp_seq_afinfo udp6_seq_afinfo = {
1539 .udp_table = &udp_table,
1540 .seq_fops = &udp6_afinfo_seq_fops,
1542 .show = udp6_seq_show,
1546 int __net_init udp6_proc_init(struct net *net)
1548 return udp_proc_register(net, &udp6_seq_afinfo);
1551 void udp6_proc_exit(struct net *net)
1553 udp_proc_unregister(net, &udp6_seq_afinfo);
1555 #endif /* CONFIG_PROC_FS */
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 .init = udpv6_init_sock,
1567 .destroy = udpv6_destroy_sock,
1568 .setsockopt = udpv6_setsockopt,
1569 .getsockopt = udpv6_getsockopt,
1570 .sendmsg = udpv6_sendmsg,
1571 .recvmsg = udpv6_recvmsg,
1572 .release_cb = ip6_datagram_release_cb,
1573 .hash = udp_lib_hash,
1574 .unhash = udp_lib_unhash,
1575 .rehash = udp_v6_rehash,
1576 .get_port = udp_v6_get_port,
1577 .memory_allocated = &udp_memory_allocated,
1578 .sysctl_mem = sysctl_udp_mem,
1579 .sysctl_wmem = &sysctl_udp_wmem_min,
1580 .sysctl_rmem = &sysctl_udp_rmem_min,
1581 .obj_size = sizeof(struct udp6_sock),
1582 .h.udp_table = &udp_table,
1583 #ifdef CONFIG_COMPAT
1584 .compat_setsockopt = compat_udpv6_setsockopt,
1585 .compat_getsockopt = compat_udpv6_getsockopt,
1587 .diag_destroy = udp_abort,
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);