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 u32 udp6_ehashfn(const struct net *net,
70 const struct in6_addr *laddr,
72 const struct in6_addr *faddr,
75 static u32 udp6_ehash_secret __read_mostly;
76 static u32 udp_ipv6_hash_secret __read_mostly;
80 net_get_random_once(&udp6_ehash_secret,
81 sizeof(udp6_ehash_secret));
82 net_get_random_once(&udp_ipv6_hash_secret,
83 sizeof(udp_ipv6_hash_secret));
85 lhash = (__force u32)laddr->s6_addr32[3];
86 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
88 return __inet6_ehashfn(lhash, lport, fhash, fport,
89 udp_ipv6_hash_secret + net_hash_mix(net));
92 static u32 udp6_portaddr_hash(const struct net *net,
93 const struct in6_addr *addr6,
96 unsigned int hash, mix = net_hash_mix(net);
98 if (ipv6_addr_any(addr6))
99 hash = jhash_1word(0, mix);
100 else if (ipv6_addr_v4mapped(addr6))
101 hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix);
103 hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix);
108 int udp_v6_get_port(struct sock *sk, unsigned short snum)
110 unsigned int hash2_nulladdr =
111 udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
112 unsigned int hash2_partial =
113 udp6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
115 /* precompute partial secondary hash */
116 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
117 return udp_lib_get_port(sk, snum, hash2_nulladdr);
120 static void udp_v6_rehash(struct sock *sk)
122 u16 new_hash = udp6_portaddr_hash(sock_net(sk),
123 &sk->sk_v6_rcv_saddr,
124 inet_sk(sk)->inet_num);
126 udp_lib_rehash(sk, new_hash);
129 static int compute_score(struct sock *sk, struct net *net,
130 const struct in6_addr *saddr, __be16 sport,
131 const struct in6_addr *daddr, unsigned short hnum,
132 int dif, int sdif, bool exact_dif)
135 struct inet_sock *inet;
137 if (!net_eq(sock_net(sk), net) ||
138 udp_sk(sk)->udp_port_hash != hnum ||
139 sk->sk_family != PF_INET6)
145 if (inet->inet_dport) {
146 if (inet->inet_dport != sport)
151 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
152 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
157 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
158 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
163 if (sk->sk_bound_dev_if || exact_dif) {
164 bool dev_match = (sk->sk_bound_dev_if == dif ||
165 sk->sk_bound_dev_if == sdif);
167 if (exact_dif && !dev_match)
169 if (sk->sk_bound_dev_if && dev_match)
173 if (sk->sk_incoming_cpu == raw_smp_processor_id())
179 /* called with rcu_read_lock() */
180 static struct sock *udp6_lib_lookup2(struct net *net,
181 const struct in6_addr *saddr, __be16 sport,
182 const struct in6_addr *daddr, unsigned int hnum,
183 int dif, int sdif, bool exact_dif,
184 struct udp_hslot *hslot2, struct sk_buff *skb)
186 struct sock *sk, *result;
187 int score, badness, matches = 0, reuseport = 0;
192 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
193 score = compute_score(sk, net, saddr, sport,
194 daddr, hnum, dif, sdif, exact_dif);
195 if (score > badness) {
196 reuseport = sk->sk_reuseport;
198 hash = udp6_ehashfn(net, daddr, hnum,
201 result = reuseport_select_sock(sk, hash, skb,
202 sizeof(struct udphdr));
209 } else if (score == badness && reuseport) {
211 if (reciprocal_scale(hash, matches) == 0)
213 hash = next_pseudo_random32(hash);
219 /* rcu_read_lock() must be held */
220 struct sock *__udp6_lib_lookup(struct net *net,
221 const struct in6_addr *saddr, __be16 sport,
222 const struct in6_addr *daddr, __be16 dport,
223 int dif, int sdif, struct udp_table *udptable,
226 struct sock *sk, *result;
227 unsigned short hnum = ntohs(dport);
228 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
229 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
230 bool exact_dif = udp6_lib_exact_dif_match(net, skb);
231 int score, badness, matches = 0, reuseport = 0;
234 if (hslot->count > 10) {
235 hash2 = udp6_portaddr_hash(net, daddr, hnum);
236 slot2 = hash2 & udptable->mask;
237 hslot2 = &udptable->hash2[slot2];
238 if (hslot->count < hslot2->count)
241 result = udp6_lib_lookup2(net, saddr, sport,
242 daddr, hnum, dif, sdif, exact_dif,
245 unsigned int old_slot2 = slot2;
246 hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
247 slot2 = hash2 & udptable->mask;
248 /* avoid searching the same slot again. */
249 if (unlikely(slot2 == old_slot2))
252 hslot2 = &udptable->hash2[slot2];
253 if (hslot->count < hslot2->count)
256 result = udp6_lib_lookup2(net, saddr, sport,
257 daddr, hnum, dif, sdif,
266 sk_for_each_rcu(sk, &hslot->head) {
267 score = compute_score(sk, net, saddr, sport, daddr, hnum, dif,
269 if (score > badness) {
270 reuseport = sk->sk_reuseport;
272 hash = udp6_ehashfn(net, daddr, hnum,
274 result = reuseport_select_sock(sk, hash, skb,
275 sizeof(struct udphdr));
282 } else if (score == badness && reuseport) {
284 if (reciprocal_scale(hash, matches) == 0)
286 hash = next_pseudo_random32(hash);
291 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
293 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
294 __be16 sport, __be16 dport,
295 struct udp_table *udptable)
297 const struct ipv6hdr *iph = ipv6_hdr(skb);
299 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
300 &iph->daddr, dport, inet6_iif(skb),
301 inet6_sdif(skb), udptable, skb);
304 struct sock *udp6_lib_lookup_skb(struct sk_buff *skb,
305 __be16 sport, __be16 dport)
307 const struct ipv6hdr *iph = ipv6_hdr(skb);
309 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
310 &iph->daddr, dport, inet6_iif(skb),
311 inet6_sdif(skb), &udp_table, skb);
313 EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb);
315 /* Must be called under rcu_read_lock().
316 * Does increment socket refcount.
318 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
319 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \
320 IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
321 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
322 const struct in6_addr *daddr, __be16 dport, int dif)
326 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport,
327 dif, 0, &udp_table, NULL);
328 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
332 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
335 /* do not use the scratch area len for jumbogram: their length execeeds the
336 * scratch area space; note that the IP6CB flags is still in the first
337 * cacheline, so checking for jumbograms is cheap
339 static int udp6_skb_len(struct sk_buff *skb)
341 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb);
345 * This should be easy, if there is something there we
346 * return it, otherwise we block.
349 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
350 int noblock, int flags, int *addr_len)
352 struct ipv6_pinfo *np = inet6_sk(sk);
353 struct inet_sock *inet = inet_sk(sk);
355 unsigned int ulen, copied;
356 int peeked, peeking, off;
358 int is_udplite = IS_UDPLITE(sk);
359 bool checksum_valid = false;
362 if (flags & MSG_ERRQUEUE)
363 return ipv6_recv_error(sk, msg, len, addr_len);
365 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
366 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
369 peeking = flags & MSG_PEEK;
370 off = sk_peek_offset(sk, flags);
371 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
375 ulen = udp6_skb_len(skb);
377 if (copied > ulen - off)
379 else if (copied < ulen)
380 msg->msg_flags |= MSG_TRUNC;
382 is_udp4 = (skb->protocol == htons(ETH_P_IP));
385 * If checksum is needed at all, try to do it while copying the
386 * data. If the data is truncated, or if we only want a partial
387 * coverage checksum (UDP-Lite), do it before the copy.
390 if (copied < ulen || peeking ||
391 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
392 checksum_valid = udp_skb_csum_unnecessary(skb) ||
393 !__udp_lib_checksum_complete(skb);
398 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
399 if (udp_skb_is_linear(skb))
400 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
402 err = skb_copy_datagram_msg(skb, off, msg, copied);
404 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
410 atomic_inc(&sk->sk_drops);
412 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
415 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
423 UDP_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS,
426 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS,
430 sock_recv_ts_and_drops(msg, sk, skb);
432 /* Copy the address. */
434 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
435 sin6->sin6_family = AF_INET6;
436 sin6->sin6_port = udp_hdr(skb)->source;
437 sin6->sin6_flowinfo = 0;
440 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
442 sin6->sin6_scope_id = 0;
444 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
445 sin6->sin6_scope_id =
446 ipv6_iface_scope_id(&sin6->sin6_addr,
449 *addr_len = sizeof(*sin6);
453 ip6_datagram_recv_common_ctl(sk, msg, skb);
456 if (inet->cmsg_flags)
457 ip_cmsg_recv_offset(msg, sk, skb,
458 sizeof(struct udphdr), off);
461 ip6_datagram_recv_specific_ctl(sk, msg, skb);
465 if (flags & MSG_TRUNC)
468 skb_consume_udp(sk, skb, peeking ? -err : err);
472 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
473 udp_skb_destructor)) {
475 UDP_INC_STATS(sock_net(sk),
476 UDP_MIB_CSUMERRORS, is_udplite);
477 UDP_INC_STATS(sock_net(sk),
478 UDP_MIB_INERRORS, is_udplite);
480 UDP6_INC_STATS(sock_net(sk),
481 UDP_MIB_CSUMERRORS, is_udplite);
482 UDP6_INC_STATS(sock_net(sk),
483 UDP_MIB_INERRORS, is_udplite);
488 /* starting over for a new packet, but check if we need to yield */
490 msg->msg_flags &= ~MSG_TRUNC;
494 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
495 u8 type, u8 code, int offset, __be32 info,
496 struct udp_table *udptable)
498 struct ipv6_pinfo *np;
499 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
500 const struct in6_addr *saddr = &hdr->saddr;
501 const struct in6_addr *daddr = &hdr->daddr;
502 struct udphdr *uh = (struct udphdr *)(skb->data+offset);
506 struct net *net = dev_net(skb->dev);
508 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
509 inet6_iif(skb), 0, udptable, skb);
511 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
516 harderr = icmpv6_err_convert(type, code, &err);
519 if (type == ICMPV6_PKT_TOOBIG) {
520 if (!ip6_sk_accept_pmtu(sk))
522 ip6_sk_update_pmtu(skb, sk, info);
523 if (np->pmtudisc != IPV6_PMTUDISC_DONT)
526 if (type == NDISC_REDIRECT) {
527 ip6_sk_redirect(skb, sk);
532 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
535 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
539 sk->sk_error_report(sk);
544 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
548 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
549 sock_rps_save_rxhash(sk, skb);
550 sk_mark_napi_id(sk, skb);
551 sk_incoming_cpu_update(sk);
553 sk_mark_napi_id_once(sk, skb);
556 rc = __udp_enqueue_schedule_skb(sk, skb);
558 int is_udplite = IS_UDPLITE(sk);
560 /* Note that an ENOMEM error is charged twice */
562 UDP6_INC_STATS(sock_net(sk),
563 UDP_MIB_RCVBUFERRORS, is_udplite);
564 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
572 static __inline__ void udpv6_err(struct sk_buff *skb,
573 struct inet6_skb_parm *opt, u8 type,
574 u8 code, int offset, __be32 info)
576 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
579 static struct static_key udpv6_encap_needed __read_mostly;
580 void udpv6_encap_enable(void)
582 if (!static_key_enabled(&udpv6_encap_needed))
583 static_key_slow_inc(&udpv6_encap_needed);
585 EXPORT_SYMBOL(udpv6_encap_enable);
587 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
589 struct udp_sock *up = udp_sk(sk);
590 int is_udplite = IS_UDPLITE(sk);
592 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
595 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
596 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
599 * This is an encapsulation socket so pass the skb to
600 * the socket's udp_encap_rcv() hook. Otherwise, just
601 * fall through and pass this up the UDP socket.
602 * up->encap_rcv() returns the following value:
603 * =0 if skb was successfully passed to the encap
604 * handler or was discarded by it.
605 * >0 if skb should be passed on to UDP.
606 * <0 if skb should be resubmitted as proto -N
609 /* if we're overly short, let UDP handle it */
610 encap_rcv = ACCESS_ONCE(up->encap_rcv);
614 /* Verify checksum before giving to encap */
615 if (udp_lib_checksum_complete(skb))
618 ret = encap_rcv(sk, skb);
620 __UDP_INC_STATS(sock_net(sk),
627 /* FALLTHROUGH -- it's a UDP Packet */
631 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
633 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
635 if (up->pcrlen == 0) { /* full coverage was set */
636 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
637 UDP_SKB_CB(skb)->cscov, skb->len);
640 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
641 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
642 UDP_SKB_CB(skb)->cscov, up->pcrlen);
647 prefetch(&sk->sk_rmem_alloc);
648 if (rcu_access_pointer(sk->sk_filter) &&
649 udp_lib_checksum_complete(skb))
652 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
655 udp_csum_pull_header(skb);
659 return __udpv6_queue_rcv_skb(sk, skb);
662 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
664 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
665 atomic_inc(&sk->sk_drops);
670 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
671 __be16 loc_port, const struct in6_addr *loc_addr,
672 __be16 rmt_port, const struct in6_addr *rmt_addr,
673 int dif, unsigned short hnum)
675 struct inet_sock *inet = inet_sk(sk);
677 if (!net_eq(sock_net(sk), net))
680 if (udp_sk(sk)->udp_port_hash != hnum ||
681 sk->sk_family != PF_INET6 ||
682 (inet->inet_dport && inet->inet_dport != rmt_port) ||
683 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
684 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
685 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif) ||
686 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
687 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
689 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
694 static void udp6_csum_zero_error(struct sk_buff *skb)
696 /* RFC 2460 section 8.1 says that we SHOULD log
697 * this error. Well, it is reasonable.
699 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
700 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
701 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
705 * Note: called only from the BH handler context,
706 * so we don't need to lock the hashes.
708 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
709 const struct in6_addr *saddr, const struct in6_addr *daddr,
710 struct udp_table *udptable, int proto)
712 struct sock *sk, *first = NULL;
713 const struct udphdr *uh = udp_hdr(skb);
714 unsigned short hnum = ntohs(uh->dest);
715 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
716 unsigned int offset = offsetof(typeof(*sk), sk_node);
717 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
718 int dif = inet6_iif(skb);
719 struct hlist_node *node;
720 struct sk_buff *nskb;
723 hash2_any = udp6_portaddr_hash(net, &in6addr_any, hnum) &
725 hash2 = udp6_portaddr_hash(net, daddr, hnum) & udptable->mask;
727 hslot = &udptable->hash2[hash2];
728 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
731 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
732 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
733 uh->source, saddr, dif, hnum))
735 /* If zero checksum and no_check is not on for
736 * the socket then skip it.
738 if (!uh->check && !udp_sk(sk)->no_check6_rx)
744 nskb = skb_clone(skb, GFP_ATOMIC);
745 if (unlikely(!nskb)) {
746 atomic_inc(&sk->sk_drops);
747 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
749 __UDP6_INC_STATS(net, UDP_MIB_INERRORS,
754 if (udpv6_queue_rcv_skb(sk, nskb) > 0)
758 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
759 if (use_hash2 && hash2 != hash2_any) {
765 if (udpv6_queue_rcv_skb(first, skb) > 0)
769 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
770 proto == IPPROTO_UDPLITE);
775 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
778 const struct in6_addr *saddr, *daddr;
779 struct net *net = dev_net(skb->dev);
784 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
787 saddr = &ipv6_hdr(skb)->saddr;
788 daddr = &ipv6_hdr(skb)->daddr;
791 ulen = ntohs(uh->len);
795 if (proto == IPPROTO_UDP) {
796 /* UDP validates ulen. */
798 /* Check for jumbo payload */
802 if (ulen < sizeof(*uh))
805 if (ulen < skb->len) {
806 if (pskb_trim_rcsum(skb, ulen))
808 saddr = &ipv6_hdr(skb)->saddr;
809 daddr = &ipv6_hdr(skb)->daddr;
814 if (udp6_csum_init(skb, uh, proto))
817 /* Check if the socket is already available, e.g. due to early demux */
818 sk = skb_steal_sock(skb);
820 struct dst_entry *dst = skb_dst(skb);
823 if (unlikely(sk->sk_rx_dst != dst))
824 udp_sk_rx_dst_set(sk, dst);
826 ret = udpv6_queue_rcv_skb(sk, skb);
829 /* a return value > 0 means to resubmit the input */
836 * Multicast receive code
838 if (ipv6_addr_is_multicast(daddr))
839 return __udp6_lib_mcast_deliver(net, skb,
840 saddr, daddr, udptable, proto);
843 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
847 if (!uh->check && !udp_sk(sk)->no_check6_rx) {
848 udp6_csum_zero_error(skb);
852 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
853 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
856 ret = udpv6_queue_rcv_skb(sk, skb);
858 /* a return value > 0 means to resubmit the input */
866 udp6_csum_zero_error(skb);
870 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
873 if (udp_lib_checksum_complete(skb))
876 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
877 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
883 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
884 proto == IPPROTO_UDPLITE ? "-Lite" : "",
885 saddr, ntohs(uh->source),
887 daddr, ntohs(uh->dest));
890 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
892 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
898 static struct sock *__udp6_lib_demux_lookup(struct net *net,
899 __be16 loc_port, const struct in6_addr *loc_addr,
900 __be16 rmt_port, const struct in6_addr *rmt_addr,
903 unsigned short hnum = ntohs(loc_port);
904 unsigned int hash2 = udp6_portaddr_hash(net, loc_addr, hnum);
905 unsigned int slot2 = hash2 & udp_table.mask;
906 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
907 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
910 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
911 if (sk->sk_state == TCP_ESTABLISHED &&
912 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif, sdif))
914 /* Only check first socket in chain */
920 static void udp_v6_early_demux(struct sk_buff *skb)
922 struct net *net = dev_net(skb->dev);
923 const struct udphdr *uh;
925 struct dst_entry *dst;
926 int dif = skb->dev->ifindex;
927 int sdif = inet6_sdif(skb);
929 if (!pskb_may_pull(skb, skb_transport_offset(skb) +
930 sizeof(struct udphdr)))
935 if (skb->pkt_type == PACKET_HOST)
936 sk = __udp6_lib_demux_lookup(net, uh->dest,
937 &ipv6_hdr(skb)->daddr,
938 uh->source, &ipv6_hdr(skb)->saddr,
943 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
947 skb->destructor = sock_efree;
948 dst = READ_ONCE(sk->sk_rx_dst);
951 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
953 /* set noref for now.
954 * any place which wants to hold dst has to call
957 skb_dst_set_noref(skb, dst);
961 static __inline__ int udpv6_rcv(struct sk_buff *skb)
963 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
967 * Throw away all pending data and cancel the corking. Socket is locked.
969 static void udp_v6_flush_pending_frames(struct sock *sk)
971 struct udp_sock *up = udp_sk(sk);
973 if (up->pending == AF_INET)
974 udp_flush_pending_frames(sk);
975 else if (up->pending) {
978 ip6_flush_pending_frames(sk);
983 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
984 * @sk: socket we are sending on
985 * @skb: sk_buff containing the filled-in UDP header
986 * (checksum field must be zeroed out)
988 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
989 const struct in6_addr *saddr,
990 const struct in6_addr *daddr, int len)
993 struct udphdr *uh = udp_hdr(skb);
994 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
998 /* Only one fragment on the socket. */
999 skb->csum_start = skb_transport_header(skb) - skb->head;
1000 skb->csum_offset = offsetof(struct udphdr, check);
1001 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1004 * HW-checksum won't work as there are two or more
1005 * fragments on the socket so that all csums of sk_buffs
1006 * should be together
1008 offset = skb_transport_offset(skb);
1009 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1011 skb->ip_summed = CHECKSUM_NONE;
1014 csum = csum_add(csum, frags->csum);
1015 } while ((frags = frags->next));
1017 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1020 uh->check = CSUM_MANGLED_0;
1028 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6)
1030 struct sock *sk = skb->sk;
1033 int is_udplite = IS_UDPLITE(sk);
1035 int offset = skb_transport_offset(skb);
1036 int len = skb->len - offset;
1039 * Create a UDP header
1042 uh->source = fl6->fl6_sport;
1043 uh->dest = fl6->fl6_dport;
1044 uh->len = htons(len);
1048 csum = udplite_csum(skb);
1049 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */
1050 skb->ip_summed = CHECKSUM_NONE;
1052 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1053 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1056 csum = udp_csum(skb);
1058 /* add protocol-dependent pseudo-header */
1059 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1060 len, fl6->flowi6_proto, csum);
1062 uh->check = CSUM_MANGLED_0;
1065 err = ip6_send_skb(skb);
1067 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1068 UDP6_INC_STATS(sock_net(sk),
1069 UDP_MIB_SNDBUFERRORS, is_udplite);
1073 UDP6_INC_STATS(sock_net(sk),
1074 UDP_MIB_OUTDATAGRAMS, is_udplite);
1079 static int udp_v6_push_pending_frames(struct sock *sk)
1081 struct sk_buff *skb;
1082 struct udp_sock *up = udp_sk(sk);
1086 if (up->pending == AF_INET)
1087 return udp_push_pending_frames(sk);
1089 /* ip6_finish_skb will release the cork, so make a copy of
1092 fl6 = inet_sk(sk)->cork.fl.u.ip6;
1094 skb = ip6_finish_skb(sk);
1098 err = udp_v6_send_skb(skb, &fl6);
1106 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1108 struct ipv6_txoptions opt_space;
1109 struct udp_sock *up = udp_sk(sk);
1110 struct inet_sock *inet = inet_sk(sk);
1111 struct ipv6_pinfo *np = inet6_sk(sk);
1112 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1113 struct in6_addr *daddr, *final_p, final;
1114 struct ipv6_txoptions *opt = NULL;
1115 struct ipv6_txoptions *opt_to_free = NULL;
1116 struct ip6_flowlabel *flowlabel = NULL;
1118 struct dst_entry *dst;
1119 struct ipcm6_cookie ipc6;
1120 int addr_len = msg->msg_namelen;
1122 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1125 int is_udplite = IS_UDPLITE(sk);
1126 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1127 struct sockcm_cookie sockc;
1132 sockc.tsflags = sk->sk_tsflags;
1134 /* destination address check */
1136 if (addr_len < offsetof(struct sockaddr, sa_data))
1139 switch (sin6->sin6_family) {
1141 if (addr_len < SIN6_LEN_RFC2133)
1143 daddr = &sin6->sin6_addr;
1144 if (ipv6_addr_any(daddr) &&
1145 ipv6_addr_v4mapped(&np->saddr))
1146 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1150 goto do_udp_sendmsg;
1152 msg->msg_name = sin6 = NULL;
1153 msg->msg_namelen = addr_len = 0;
1159 } else if (!up->pending) {
1160 if (sk->sk_state != TCP_ESTABLISHED)
1161 return -EDESTADDRREQ;
1162 daddr = &sk->sk_v6_daddr;
1167 if (ipv6_addr_v4mapped(daddr)) {
1168 struct sockaddr_in sin;
1169 sin.sin_family = AF_INET;
1170 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1171 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1172 msg->msg_name = &sin;
1173 msg->msg_namelen = sizeof(sin);
1175 if (__ipv6_only_sock(sk))
1176 return -ENETUNREACH;
1177 return udp_sendmsg(sk, msg, len);
1181 if (up->pending == AF_INET)
1182 return udp_sendmsg(sk, msg, len);
1184 /* Rough check on arithmetic overflow,
1185 better check is made in ip6_append_data().
1187 if (len > INT_MAX - sizeof(struct udphdr))
1190 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1193 * There are pending frames.
1194 * The socket lock must be held while it's corked.
1197 if (likely(up->pending)) {
1198 if (unlikely(up->pending != AF_INET6)) {
1200 return -EAFNOSUPPORT;
1203 goto do_append_data;
1207 ulen += sizeof(struct udphdr);
1209 memset(&fl6, 0, sizeof(fl6));
1212 if (sin6->sin6_port == 0)
1215 fl6.fl6_dport = sin6->sin6_port;
1216 daddr = &sin6->sin6_addr;
1219 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1220 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1221 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1228 * Otherwise it will be difficult to maintain
1231 if (sk->sk_state == TCP_ESTABLISHED &&
1232 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1233 daddr = &sk->sk_v6_daddr;
1235 if (addr_len >= sizeof(struct sockaddr_in6) &&
1236 sin6->sin6_scope_id &&
1237 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1238 fl6.flowi6_oif = sin6->sin6_scope_id;
1240 if (sk->sk_state != TCP_ESTABLISHED)
1241 return -EDESTADDRREQ;
1243 fl6.fl6_dport = inet->inet_dport;
1244 daddr = &sk->sk_v6_daddr;
1245 fl6.flowlabel = np->flow_label;
1249 if (!fl6.flowi6_oif)
1250 fl6.flowi6_oif = sk->sk_bound_dev_if;
1252 if (!fl6.flowi6_oif)
1253 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1255 fl6.flowi6_mark = sk->sk_mark;
1256 fl6.flowi6_uid = sk->sk_uid;
1258 if (msg->msg_controllen) {
1260 memset(opt, 0, sizeof(struct ipv6_txoptions));
1261 opt->tot_len = sizeof(*opt);
1264 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, &ipc6, &sockc);
1266 fl6_sock_release(flowlabel);
1269 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1270 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1274 if (!(opt->opt_nflen|opt->opt_flen))
1279 opt = txopt_get(np);
1283 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1284 opt = ipv6_fixup_options(&opt_space, opt);
1287 fl6.flowi6_proto = sk->sk_protocol;
1288 if (!ipv6_addr_any(daddr))
1291 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1292 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1293 fl6.saddr = np->saddr;
1294 fl6.fl6_sport = inet->inet_sport;
1296 final_p = fl6_update_dst(&fl6, opt, &final);
1300 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1301 fl6.flowi6_oif = np->mcast_oif;
1303 } else if (!fl6.flowi6_oif)
1304 fl6.flowi6_oif = np->ucast_oif;
1306 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1308 if (ipc6.tclass < 0)
1309 ipc6.tclass = np->tclass;
1311 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel);
1313 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p);
1320 if (ipc6.hlimit < 0)
1321 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1323 if (msg->msg_flags&MSG_CONFIRM)
1327 /* Lockless fast path for the non-corking case */
1329 struct sk_buff *skb;
1331 skb = ip6_make_skb(sk, getfrag, msg, ulen,
1332 sizeof(struct udphdr), &ipc6,
1333 &fl6, (struct rt6_info *)dst,
1334 msg->msg_flags, &sockc);
1336 if (!IS_ERR_OR_NULL(skb))
1337 err = udp_v6_send_skb(skb, &fl6);
1342 if (unlikely(up->pending)) {
1343 /* The socket is already corked while preparing it. */
1344 /* ... which is an evident application bug. --ANK */
1347 net_dbg_ratelimited("udp cork app bug 2\n");
1352 up->pending = AF_INET6;
1355 if (ipc6.dontfrag < 0)
1356 ipc6.dontfrag = np->dontfrag;
1358 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr),
1359 &ipc6, &fl6, (struct rt6_info *)dst,
1360 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, &sockc);
1362 udp_v6_flush_pending_frames(sk);
1364 err = udp_v6_push_pending_frames(sk);
1365 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1369 err = np->recverr ? net_xmit_errno(err) : 0;
1375 ip6_dst_store(sk, dst,
1376 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ?
1377 &sk->sk_v6_daddr : NULL,
1378 #ifdef CONFIG_IPV6_SUBTREES
1379 ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1391 fl6_sock_release(flowlabel);
1392 txopt_put(opt_to_free);
1396 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1397 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1398 * we don't have a good statistic (IpOutDiscards but it can be too many
1399 * things). We could add another new stat but at least for now that
1400 * seems like overkill.
1402 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1403 UDP6_INC_STATS(sock_net(sk),
1404 UDP_MIB_SNDBUFERRORS, is_udplite);
1409 if (msg->msg_flags & MSG_PROBE)
1410 dst_confirm_neigh(dst, &fl6.daddr);
1411 if (!(msg->msg_flags&MSG_PROBE) || len)
1412 goto back_from_confirm;
1417 void udpv6_destroy_sock(struct sock *sk)
1419 struct udp_sock *up = udp_sk(sk);
1421 udp_v6_flush_pending_frames(sk);
1424 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
1425 void (*encap_destroy)(struct sock *sk);
1426 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1431 inet6_destroy_sock(sk);
1435 * Socket option code for UDP
1437 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1438 char __user *optval, unsigned int optlen)
1440 if (level == SOL_UDP || level == SOL_UDPLITE)
1441 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1442 udp_v6_push_pending_frames);
1443 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1446 #ifdef CONFIG_COMPAT
1447 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1448 char __user *optval, unsigned int optlen)
1450 if (level == SOL_UDP || level == SOL_UDPLITE)
1451 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1452 udp_v6_push_pending_frames);
1453 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1457 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1458 char __user *optval, int __user *optlen)
1460 if (level == SOL_UDP || level == SOL_UDPLITE)
1461 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1462 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1465 #ifdef CONFIG_COMPAT
1466 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1467 char __user *optval, int __user *optlen)
1469 if (level == SOL_UDP || level == SOL_UDPLITE)
1470 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1471 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1475 /* thinking of making this const? Don't.
1476 * early_demux can change based on sysctl.
1478 static struct inet6_protocol udpv6_protocol = {
1479 .early_demux = udp_v6_early_demux,
1480 .early_demux_handler = udp_v6_early_demux,
1481 .handler = udpv6_rcv,
1482 .err_handler = udpv6_err,
1483 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1486 /* ------------------------------------------------------------------------ */
1487 #ifdef CONFIG_PROC_FS
1488 int udp6_seq_show(struct seq_file *seq, void *v)
1490 if (v == SEQ_START_TOKEN) {
1491 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1493 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1494 struct inet_sock *inet = inet_sk(v);
1495 __u16 srcp = ntohs(inet->inet_sport);
1496 __u16 destp = ntohs(inet->inet_dport);
1497 ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket);
1502 static const struct file_operations udp6_afinfo_seq_fops = {
1503 .owner = THIS_MODULE,
1504 .open = udp_seq_open,
1506 .llseek = seq_lseek,
1507 .release = seq_release_net
1510 static struct udp_seq_afinfo udp6_seq_afinfo = {
1513 .udp_table = &udp_table,
1514 .seq_fops = &udp6_afinfo_seq_fops,
1516 .show = udp6_seq_show,
1520 int __net_init udp6_proc_init(struct net *net)
1522 return udp_proc_register(net, &udp6_seq_afinfo);
1525 void udp6_proc_exit(struct net *net)
1527 udp_proc_unregister(net, &udp6_seq_afinfo);
1529 #endif /* CONFIG_PROC_FS */
1531 /* ------------------------------------------------------------------------ */
1533 struct proto udpv6_prot = {
1535 .owner = THIS_MODULE,
1536 .close = udp_lib_close,
1537 .connect = ip6_datagram_connect,
1538 .disconnect = udp_disconnect,
1540 .init = udp_init_sock,
1541 .destroy = udpv6_destroy_sock,
1542 .setsockopt = udpv6_setsockopt,
1543 .getsockopt = udpv6_getsockopt,
1544 .sendmsg = udpv6_sendmsg,
1545 .recvmsg = udpv6_recvmsg,
1546 .release_cb = ip6_datagram_release_cb,
1547 .hash = udp_lib_hash,
1548 .unhash = udp_lib_unhash,
1549 .rehash = udp_v6_rehash,
1550 .get_port = udp_v6_get_port,
1551 .memory_allocated = &udp_memory_allocated,
1552 .sysctl_mem = sysctl_udp_mem,
1553 .sysctl_wmem = &sysctl_udp_wmem_min,
1554 .sysctl_rmem = &sysctl_udp_rmem_min,
1555 .obj_size = sizeof(struct udp6_sock),
1556 .h.udp_table = &udp_table,
1557 #ifdef CONFIG_COMPAT
1558 .compat_setsockopt = compat_udpv6_setsockopt,
1559 .compat_getsockopt = compat_udpv6_getsockopt,
1561 .diag_destroy = udp_abort,
1564 static struct inet_protosw udpv6_protosw = {
1566 .protocol = IPPROTO_UDP,
1567 .prot = &udpv6_prot,
1568 .ops = &inet6_dgram_ops,
1569 .flags = INET_PROTOSW_PERMANENT,
1572 int __init udpv6_init(void)
1576 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1580 ret = inet6_register_protosw(&udpv6_protosw);
1582 goto out_udpv6_protocol;
1587 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1591 void udpv6_exit(void)
1593 inet6_unregister_protosw(&udpv6_protosw);
1594 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);