3 * Linux INET6 implementation
8 * Based on linux/ipv4/udp.c
11 * Hideaki YOSHIFUJI : sin6_scope_id support
12 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
13 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
14 * a single port at the same time.
15 * Kazunori MIYAZAWA @USAGI: change process style to use ip6_append_data
16 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/udp6 to seq_file.
18 * This program is free software; you can redistribute it and/or
19 * modify it under the terms of the GNU General Public License
20 * as published by the Free Software Foundation; either version
21 * 2 of the License, or (at your option) any later version.
24 #include <linux/errno.h>
25 #include <linux/types.h>
26 #include <linux/socket.h>
27 #include <linux/sockios.h>
28 #include <linux/net.h>
29 #include <linux/in6.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_arp.h>
32 #include <linux/ipv6.h>
33 #include <linux/icmpv6.h>
34 #include <linux/init.h>
35 #include <linux/module.h>
36 #include <linux/skbuff.h>
37 #include <linux/slab.h>
38 #include <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/inet6_hashtables.h>
50 #include <net/busy_poll.h>
51 #include <net/sock_reuseport.h>
53 #include <linux/proc_fs.h>
54 #include <linux/seq_file.h>
55 #include <trace/events/skb.h>
58 static bool udp6_lib_exact_dif_match(struct net *net, struct sk_buff *skb)
60 #if defined(CONFIG_NET_L3_MASTER_DEV)
61 if (!net->ipv4.sysctl_udp_l3mdev_accept &&
62 skb && ipv6_l3mdev_skb(IP6CB(skb)->flags))
68 static u32 udp6_ehashfn(const struct net *net,
69 const struct in6_addr *laddr,
71 const struct in6_addr *faddr,
74 static u32 udp6_ehash_secret __read_mostly;
75 static u32 udp_ipv6_hash_secret __read_mostly;
79 net_get_random_once(&udp6_ehash_secret,
80 sizeof(udp6_ehash_secret));
81 net_get_random_once(&udp_ipv6_hash_secret,
82 sizeof(udp_ipv6_hash_secret));
84 lhash = (__force u32)laddr->s6_addr32[3];
85 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
87 return __inet6_ehashfn(lhash, lport, fhash, fport,
88 udp_ipv6_hash_secret + net_hash_mix(net));
91 static u32 udp6_portaddr_hash(const struct net *net,
92 const struct in6_addr *addr6,
95 unsigned int hash, mix = net_hash_mix(net);
97 if (ipv6_addr_any(addr6))
98 hash = jhash_1word(0, mix);
99 else if (ipv6_addr_v4mapped(addr6))
100 hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix);
102 hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix);
107 int udp_v6_get_port(struct sock *sk, unsigned short snum)
109 unsigned int hash2_nulladdr =
110 udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
111 unsigned int hash2_partial =
112 udp6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
114 /* precompute partial secondary hash */
115 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
116 return udp_lib_get_port(sk, snum, hash2_nulladdr);
119 static void udp_v6_rehash(struct sock *sk)
121 u16 new_hash = udp6_portaddr_hash(sock_net(sk),
122 &sk->sk_v6_rcv_saddr,
123 inet_sk(sk)->inet_num);
125 udp_lib_rehash(sk, new_hash);
128 static int compute_score(struct sock *sk, struct net *net,
129 const struct in6_addr *saddr, __be16 sport,
130 const struct in6_addr *daddr, unsigned short hnum,
131 int dif, bool exact_dif)
134 struct inet_sock *inet;
136 if (!net_eq(sock_net(sk), net) ||
137 udp_sk(sk)->udp_port_hash != hnum ||
138 sk->sk_family != PF_INET6)
144 if (inet->inet_dport) {
145 if (inet->inet_dport != sport)
150 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
151 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
156 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
157 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
162 if (sk->sk_bound_dev_if || exact_dif) {
163 if (sk->sk_bound_dev_if != dif)
168 if (sk->sk_incoming_cpu == raw_smp_processor_id())
174 /* called with rcu_read_lock() */
175 static struct sock *udp6_lib_lookup2(struct net *net,
176 const struct in6_addr *saddr, __be16 sport,
177 const struct in6_addr *daddr, unsigned int hnum, int dif,
178 bool exact_dif, struct udp_hslot *hslot2,
181 struct sock *sk, *result;
182 int score, badness, matches = 0, reuseport = 0;
187 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
188 score = compute_score(sk, net, saddr, sport,
189 daddr, hnum, dif, exact_dif);
190 if (score > badness) {
191 reuseport = sk->sk_reuseport;
193 hash = udp6_ehashfn(net, daddr, hnum,
196 result = reuseport_select_sock(sk, hash, skb,
197 sizeof(struct udphdr));
204 } else if (score == badness && reuseport) {
206 if (reciprocal_scale(hash, matches) == 0)
208 hash = next_pseudo_random32(hash);
214 /* rcu_read_lock() must be held */
215 struct sock *__udp6_lib_lookup(struct net *net,
216 const struct in6_addr *saddr, __be16 sport,
217 const struct in6_addr *daddr, __be16 dport,
218 int dif, struct udp_table *udptable,
221 struct sock *sk, *result;
222 unsigned short hnum = ntohs(dport);
223 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
224 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
225 bool exact_dif = udp6_lib_exact_dif_match(net, skb);
226 int score, badness, matches = 0, reuseport = 0;
229 if (hslot->count > 10) {
230 hash2 = udp6_portaddr_hash(net, daddr, hnum);
231 slot2 = hash2 & udptable->mask;
232 hslot2 = &udptable->hash2[slot2];
233 if (hslot->count < hslot2->count)
236 result = udp6_lib_lookup2(net, saddr, sport,
237 daddr, hnum, dif, exact_dif,
240 unsigned int old_slot2 = slot2;
241 hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
242 slot2 = hash2 & udptable->mask;
243 /* avoid searching the same slot again. */
244 if (unlikely(slot2 == old_slot2))
247 hslot2 = &udptable->hash2[slot2];
248 if (hslot->count < hslot2->count)
251 result = udp6_lib_lookup2(net, saddr, sport,
261 sk_for_each_rcu(sk, &hslot->head) {
262 score = compute_score(sk, net, saddr, sport, daddr, hnum, dif,
264 if (score > badness) {
265 reuseport = sk->sk_reuseport;
267 hash = udp6_ehashfn(net, daddr, hnum,
269 result = reuseport_select_sock(sk, hash, skb,
270 sizeof(struct udphdr));
277 } else if (score == badness && reuseport) {
279 if (reciprocal_scale(hash, matches) == 0)
281 hash = next_pseudo_random32(hash);
286 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
288 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
289 __be16 sport, __be16 dport,
290 struct udp_table *udptable)
292 const struct ipv6hdr *iph = ipv6_hdr(skb);
295 sk = skb_steal_sock(skb);
298 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
299 &iph->daddr, dport, inet6_iif(skb),
303 struct sock *udp6_lib_lookup_skb(struct sk_buff *skb,
304 __be16 sport, __be16 dport)
306 const struct ipv6hdr *iph = ipv6_hdr(skb);
308 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
309 &iph->daddr, dport, inet6_iif(skb),
312 EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb);
314 /* Must be called under rcu_read_lock().
315 * Does increment socket refcount.
317 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
318 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \
319 IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
320 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
321 const struct in6_addr *daddr, __be16 dport, int dif)
325 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport,
326 dif, &udp_table, NULL);
327 if (sk && !atomic_inc_not_zero(&sk->sk_refcnt))
331 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
335 * This should be easy, if there is something there we
336 * return it, otherwise we block.
339 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
340 int noblock, int flags, int *addr_len)
342 struct ipv6_pinfo *np = inet6_sk(sk);
343 struct inet_sock *inet = inet_sk(sk);
345 unsigned int ulen, copied;
346 int peeked, peeking, off;
348 int is_udplite = IS_UDPLITE(sk);
349 bool checksum_valid = false;
352 if (flags & MSG_ERRQUEUE)
353 return ipv6_recv_error(sk, msg, len, addr_len);
355 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
356 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
359 peeking = off = sk_peek_offset(sk, flags);
360 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
366 if (copied > ulen - off)
368 else if (copied < ulen)
369 msg->msg_flags |= MSG_TRUNC;
371 is_udp4 = (skb->protocol == htons(ETH_P_IP));
374 * If checksum is needed at all, try to do it while copying the
375 * data. If the data is truncated, or if we only want a partial
376 * coverage checksum (UDP-Lite), do it before the copy.
379 if (copied < ulen || peeking ||
380 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
381 checksum_valid = !udp_lib_checksum_complete(skb);
386 if (checksum_valid || skb_csum_unnecessary(skb))
387 err = skb_copy_datagram_msg(skb, off, msg, copied);
389 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
395 atomic_inc(&sk->sk_drops);
397 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
400 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
408 UDP_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS,
411 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS,
415 sock_recv_ts_and_drops(msg, sk, skb);
417 /* Copy the address. */
419 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
420 sin6->sin6_family = AF_INET6;
421 sin6->sin6_port = udp_hdr(skb)->source;
422 sin6->sin6_flowinfo = 0;
425 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
427 sin6->sin6_scope_id = 0;
429 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
430 sin6->sin6_scope_id =
431 ipv6_iface_scope_id(&sin6->sin6_addr,
434 *addr_len = sizeof(*sin6);
438 ip6_datagram_recv_common_ctl(sk, msg, skb);
441 if (inet->cmsg_flags)
442 ip_cmsg_recv_offset(msg, sk, skb,
443 sizeof(struct udphdr), off);
446 ip6_datagram_recv_specific_ctl(sk, msg, skb);
450 if (flags & MSG_TRUNC)
453 skb_consume_udp(sk, skb, peeking ? -err : err);
457 if (!__sk_queue_drop_skb(sk, skb, flags, udp_skb_destructor)) {
459 UDP_INC_STATS(sock_net(sk),
460 UDP_MIB_CSUMERRORS, is_udplite);
461 UDP_INC_STATS(sock_net(sk),
462 UDP_MIB_INERRORS, is_udplite);
464 UDP6_INC_STATS(sock_net(sk),
465 UDP_MIB_CSUMERRORS, is_udplite);
466 UDP6_INC_STATS(sock_net(sk),
467 UDP_MIB_INERRORS, is_udplite);
472 /* starting over for a new packet, but check if we need to yield */
474 msg->msg_flags &= ~MSG_TRUNC;
478 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
479 u8 type, u8 code, int offset, __be32 info,
480 struct udp_table *udptable)
482 struct ipv6_pinfo *np;
483 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
484 const struct in6_addr *saddr = &hdr->saddr;
485 const struct in6_addr *daddr = &hdr->daddr;
486 struct udphdr *uh = (struct udphdr *)(skb->data+offset);
490 struct net *net = dev_net(skb->dev);
492 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
493 inet6_iif(skb), udptable, skb);
495 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
500 harderr = icmpv6_err_convert(type, code, &err);
503 if (type == ICMPV6_PKT_TOOBIG) {
504 if (!ip6_sk_accept_pmtu(sk))
506 ip6_sk_update_pmtu(skb, sk, info);
507 if (np->pmtudisc != IPV6_PMTUDISC_DONT)
510 if (type == NDISC_REDIRECT) {
511 ip6_sk_redirect(skb, sk);
516 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
519 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
523 sk->sk_error_report(sk);
528 int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
532 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
533 sock_rps_save_rxhash(sk, skb);
534 sk_mark_napi_id(sk, skb);
535 sk_incoming_cpu_update(sk);
537 sk_mark_napi_id_once(sk, skb);
540 rc = __udp_enqueue_schedule_skb(sk, skb);
542 int is_udplite = IS_UDPLITE(sk);
544 /* Note that an ENOMEM error is charged twice */
546 UDP6_INC_STATS(sock_net(sk),
547 UDP_MIB_RCVBUFERRORS, is_udplite);
548 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
556 static __inline__ void udpv6_err(struct sk_buff *skb,
557 struct inet6_skb_parm *opt, u8 type,
558 u8 code, int offset, __be32 info)
560 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
563 static struct static_key udpv6_encap_needed __read_mostly;
564 void udpv6_encap_enable(void)
566 if (!static_key_enabled(&udpv6_encap_needed))
567 static_key_slow_inc(&udpv6_encap_needed);
569 EXPORT_SYMBOL(udpv6_encap_enable);
571 int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
573 struct udp_sock *up = udp_sk(sk);
574 int is_udplite = IS_UDPLITE(sk);
576 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
579 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
580 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
583 * This is an encapsulation socket so pass the skb to
584 * the socket's udp_encap_rcv() hook. Otherwise, just
585 * fall through and pass this up the UDP socket.
586 * up->encap_rcv() returns the following value:
587 * =0 if skb was successfully passed to the encap
588 * handler or was discarded by it.
589 * >0 if skb should be passed on to UDP.
590 * <0 if skb should be resubmitted as proto -N
593 /* if we're overly short, let UDP handle it */
594 encap_rcv = ACCESS_ONCE(up->encap_rcv);
598 /* Verify checksum before giving to encap */
599 if (udp_lib_checksum_complete(skb))
602 ret = encap_rcv(sk, skb);
604 __UDP_INC_STATS(sock_net(sk),
611 /* FALLTHROUGH -- it's a UDP Packet */
615 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
617 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
619 if (up->pcrlen == 0) { /* full coverage was set */
620 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
621 UDP_SKB_CB(skb)->cscov, skb->len);
624 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
625 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
626 UDP_SKB_CB(skb)->cscov, up->pcrlen);
631 if (rcu_access_pointer(sk->sk_filter) &&
632 udp_lib_checksum_complete(skb))
635 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
638 udp_csum_pull_header(skb);
642 return __udpv6_queue_rcv_skb(sk, skb);
645 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
647 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
648 atomic_inc(&sk->sk_drops);
653 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
654 __be16 loc_port, const struct in6_addr *loc_addr,
655 __be16 rmt_port, const struct in6_addr *rmt_addr,
656 int dif, unsigned short hnum)
658 struct inet_sock *inet = inet_sk(sk);
660 if (!net_eq(sock_net(sk), net))
663 if (udp_sk(sk)->udp_port_hash != hnum ||
664 sk->sk_family != PF_INET6 ||
665 (inet->inet_dport && inet->inet_dport != rmt_port) ||
666 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
667 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
668 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif) ||
669 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
670 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
672 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
677 static void udp6_csum_zero_error(struct sk_buff *skb)
679 /* RFC 2460 section 8.1 says that we SHOULD log
680 * this error. Well, it is reasonable.
682 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
683 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
684 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
688 * Note: called only from the BH handler context,
689 * so we don't need to lock the hashes.
691 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
692 const struct in6_addr *saddr, const struct in6_addr *daddr,
693 struct udp_table *udptable, int proto)
695 struct sock *sk, *first = NULL;
696 const struct udphdr *uh = udp_hdr(skb);
697 unsigned short hnum = ntohs(uh->dest);
698 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
699 unsigned int offset = offsetof(typeof(*sk), sk_node);
700 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
701 int dif = inet6_iif(skb);
702 struct hlist_node *node;
703 struct sk_buff *nskb;
706 hash2_any = udp6_portaddr_hash(net, &in6addr_any, hnum) &
708 hash2 = udp6_portaddr_hash(net, daddr, hnum) & udptable->mask;
710 hslot = &udptable->hash2[hash2];
711 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
714 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
715 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
716 uh->source, saddr, dif, hnum))
718 /* If zero checksum and no_check is not on for
719 * the socket then skip it.
721 if (!uh->check && !udp_sk(sk)->no_check6_rx)
727 nskb = skb_clone(skb, GFP_ATOMIC);
728 if (unlikely(!nskb)) {
729 atomic_inc(&sk->sk_drops);
730 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
732 __UDP6_INC_STATS(net, UDP_MIB_INERRORS,
737 if (udpv6_queue_rcv_skb(sk, nskb) > 0)
741 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
742 if (use_hash2 && hash2 != hash2_any) {
748 if (udpv6_queue_rcv_skb(first, skb) > 0)
752 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
753 proto == IPPROTO_UDPLITE);
758 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
761 const struct in6_addr *saddr, *daddr;
762 struct net *net = dev_net(skb->dev);
767 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
770 saddr = &ipv6_hdr(skb)->saddr;
771 daddr = &ipv6_hdr(skb)->daddr;
774 ulen = ntohs(uh->len);
778 if (proto == IPPROTO_UDP) {
779 /* UDP validates ulen. */
781 /* Check for jumbo payload */
785 if (ulen < sizeof(*uh))
788 if (ulen < skb->len) {
789 if (pskb_trim_rcsum(skb, ulen))
791 saddr = &ipv6_hdr(skb)->saddr;
792 daddr = &ipv6_hdr(skb)->daddr;
797 if (udp6_csum_init(skb, uh, proto))
801 * Multicast receive code
803 if (ipv6_addr_is_multicast(daddr))
804 return __udp6_lib_mcast_deliver(net, skb,
805 saddr, daddr, udptable, proto);
810 * check socket cache ... must talk to Alan about his plans
811 * for sock caches... i'll skip this for now.
813 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
817 if (!uh->check && !udp_sk(sk)->no_check6_rx) {
818 udp6_csum_zero_error(skb);
822 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
823 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
826 ret = udpv6_queue_rcv_skb(sk, skb);
828 /* a return value > 0 means to resubmit the input */
836 udp6_csum_zero_error(skb);
840 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
843 if (udp_lib_checksum_complete(skb))
846 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
847 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
853 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
854 proto == IPPROTO_UDPLITE ? "-Lite" : "",
855 saddr, ntohs(uh->source),
857 daddr, ntohs(uh->dest));
860 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
862 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
867 static __inline__ int udpv6_rcv(struct sk_buff *skb)
869 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
873 * Throw away all pending data and cancel the corking. Socket is locked.
875 static void udp_v6_flush_pending_frames(struct sock *sk)
877 struct udp_sock *up = udp_sk(sk);
879 if (up->pending == AF_INET)
880 udp_flush_pending_frames(sk);
881 else if (up->pending) {
884 ip6_flush_pending_frames(sk);
889 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
890 * @sk: socket we are sending on
891 * @skb: sk_buff containing the filled-in UDP header
892 * (checksum field must be zeroed out)
894 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
895 const struct in6_addr *saddr,
896 const struct in6_addr *daddr, int len)
899 struct udphdr *uh = udp_hdr(skb);
900 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
904 /* Only one fragment on the socket. */
905 skb->csum_start = skb_transport_header(skb) - skb->head;
906 skb->csum_offset = offsetof(struct udphdr, check);
907 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
910 * HW-checksum won't work as there are two or more
911 * fragments on the socket so that all csums of sk_buffs
914 offset = skb_transport_offset(skb);
915 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
917 skb->ip_summed = CHECKSUM_NONE;
920 csum = csum_add(csum, frags->csum);
921 } while ((frags = frags->next));
923 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
926 uh->check = CSUM_MANGLED_0;
934 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6)
936 struct sock *sk = skb->sk;
939 int is_udplite = IS_UDPLITE(sk);
941 int offset = skb_transport_offset(skb);
942 int len = skb->len - offset;
945 * Create a UDP header
948 uh->source = fl6->fl6_sport;
949 uh->dest = fl6->fl6_dport;
950 uh->len = htons(len);
954 csum = udplite_csum(skb);
955 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */
956 skb->ip_summed = CHECKSUM_NONE;
958 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
959 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
962 csum = udp_csum(skb);
964 /* add protocol-dependent pseudo-header */
965 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
966 len, fl6->flowi6_proto, csum);
968 uh->check = CSUM_MANGLED_0;
971 err = ip6_send_skb(skb);
973 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
974 UDP6_INC_STATS(sock_net(sk),
975 UDP_MIB_SNDBUFERRORS, is_udplite);
979 UDP6_INC_STATS(sock_net(sk),
980 UDP_MIB_OUTDATAGRAMS, is_udplite);
985 static int udp_v6_push_pending_frames(struct sock *sk)
988 struct udp_sock *up = udp_sk(sk);
992 if (up->pending == AF_INET)
993 return udp_push_pending_frames(sk);
995 /* ip6_finish_skb will release the cork, so make a copy of
998 fl6 = inet_sk(sk)->cork.fl.u.ip6;
1000 skb = ip6_finish_skb(sk);
1004 err = udp_v6_send_skb(skb, &fl6);
1012 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1014 struct ipv6_txoptions opt_space;
1015 struct udp_sock *up = udp_sk(sk);
1016 struct inet_sock *inet = inet_sk(sk);
1017 struct ipv6_pinfo *np = inet6_sk(sk);
1018 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1019 struct in6_addr *daddr, *final_p, final;
1020 struct ipv6_txoptions *opt = NULL;
1021 struct ipv6_txoptions *opt_to_free = NULL;
1022 struct ip6_flowlabel *flowlabel = NULL;
1024 struct dst_entry *dst;
1025 struct ipcm6_cookie ipc6;
1026 int addr_len = msg->msg_namelen;
1028 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1031 int is_udplite = IS_UDPLITE(sk);
1032 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1033 struct sockcm_cookie sockc;
1038 sockc.tsflags = sk->sk_tsflags;
1040 /* destination address check */
1042 if (addr_len < offsetof(struct sockaddr, sa_data))
1045 switch (sin6->sin6_family) {
1047 if (addr_len < SIN6_LEN_RFC2133)
1049 daddr = &sin6->sin6_addr;
1050 if (ipv6_addr_any(daddr) &&
1051 ipv6_addr_v4mapped(&np->saddr))
1052 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1056 goto do_udp_sendmsg;
1058 msg->msg_name = sin6 = NULL;
1059 msg->msg_namelen = addr_len = 0;
1065 } else if (!up->pending) {
1066 if (sk->sk_state != TCP_ESTABLISHED)
1067 return -EDESTADDRREQ;
1068 daddr = &sk->sk_v6_daddr;
1073 if (ipv6_addr_v4mapped(daddr)) {
1074 struct sockaddr_in sin;
1075 sin.sin_family = AF_INET;
1076 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1077 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1078 msg->msg_name = &sin;
1079 msg->msg_namelen = sizeof(sin);
1081 if (__ipv6_only_sock(sk))
1082 return -ENETUNREACH;
1083 return udp_sendmsg(sk, msg, len);
1087 if (up->pending == AF_INET)
1088 return udp_sendmsg(sk, msg, len);
1090 /* Rough check on arithmetic overflow,
1091 better check is made in ip6_append_data().
1093 if (len > INT_MAX - sizeof(struct udphdr))
1096 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1099 * There are pending frames.
1100 * The socket lock must be held while it's corked.
1103 if (likely(up->pending)) {
1104 if (unlikely(up->pending != AF_INET6)) {
1106 return -EAFNOSUPPORT;
1109 goto do_append_data;
1113 ulen += sizeof(struct udphdr);
1115 memset(&fl6, 0, sizeof(fl6));
1118 if (sin6->sin6_port == 0)
1121 fl6.fl6_dport = sin6->sin6_port;
1122 daddr = &sin6->sin6_addr;
1125 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1126 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1127 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1134 * Otherwise it will be difficult to maintain
1137 if (sk->sk_state == TCP_ESTABLISHED &&
1138 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1139 daddr = &sk->sk_v6_daddr;
1141 if (addr_len >= sizeof(struct sockaddr_in6) &&
1142 sin6->sin6_scope_id &&
1143 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1144 fl6.flowi6_oif = sin6->sin6_scope_id;
1146 if (sk->sk_state != TCP_ESTABLISHED)
1147 return -EDESTADDRREQ;
1149 fl6.fl6_dport = inet->inet_dport;
1150 daddr = &sk->sk_v6_daddr;
1151 fl6.flowlabel = np->flow_label;
1155 if (!fl6.flowi6_oif)
1156 fl6.flowi6_oif = sk->sk_bound_dev_if;
1158 if (!fl6.flowi6_oif)
1159 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1161 fl6.flowi6_mark = sk->sk_mark;
1162 fl6.flowi6_uid = sk->sk_uid;
1164 if (msg->msg_controllen) {
1166 memset(opt, 0, sizeof(struct ipv6_txoptions));
1167 opt->tot_len = sizeof(*opt);
1170 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, &ipc6, &sockc);
1172 fl6_sock_release(flowlabel);
1175 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1176 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1180 if (!(opt->opt_nflen|opt->opt_flen))
1185 opt = txopt_get(np);
1189 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1190 opt = ipv6_fixup_options(&opt_space, opt);
1193 fl6.flowi6_proto = sk->sk_protocol;
1194 if (!ipv6_addr_any(daddr))
1197 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1198 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1199 fl6.saddr = np->saddr;
1200 fl6.fl6_sport = inet->inet_sport;
1202 final_p = fl6_update_dst(&fl6, opt, &final);
1206 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1207 fl6.flowi6_oif = np->mcast_oif;
1209 } else if (!fl6.flowi6_oif)
1210 fl6.flowi6_oif = np->ucast_oif;
1212 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1214 if (ipc6.tclass < 0)
1215 ipc6.tclass = np->tclass;
1217 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel);
1219 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p);
1226 if (ipc6.hlimit < 0)
1227 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1229 if (msg->msg_flags&MSG_CONFIRM)
1233 /* Lockless fast path for the non-corking case */
1235 struct sk_buff *skb;
1237 skb = ip6_make_skb(sk, getfrag, msg, ulen,
1238 sizeof(struct udphdr), &ipc6,
1239 &fl6, (struct rt6_info *)dst,
1240 msg->msg_flags, &sockc);
1242 if (!IS_ERR_OR_NULL(skb))
1243 err = udp_v6_send_skb(skb, &fl6);
1248 if (unlikely(up->pending)) {
1249 /* The socket is already corked while preparing it. */
1250 /* ... which is an evident application bug. --ANK */
1253 net_dbg_ratelimited("udp cork app bug 2\n");
1258 up->pending = AF_INET6;
1261 if (ipc6.dontfrag < 0)
1262 ipc6.dontfrag = np->dontfrag;
1264 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr),
1265 &ipc6, &fl6, (struct rt6_info *)dst,
1266 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, &sockc);
1268 udp_v6_flush_pending_frames(sk);
1270 err = udp_v6_push_pending_frames(sk);
1271 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1275 err = np->recverr ? net_xmit_errno(err) : 0;
1281 ip6_dst_store(sk, dst,
1282 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ?
1283 &sk->sk_v6_daddr : NULL,
1284 #ifdef CONFIG_IPV6_SUBTREES
1285 ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1297 fl6_sock_release(flowlabel);
1298 txopt_put(opt_to_free);
1302 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1303 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1304 * we don't have a good statistic (IpOutDiscards but it can be too many
1305 * things). We could add another new stat but at least for now that
1306 * seems like overkill.
1308 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1309 UDP6_INC_STATS(sock_net(sk),
1310 UDP_MIB_SNDBUFERRORS, is_udplite);
1315 if (msg->msg_flags & MSG_PROBE)
1316 dst_confirm_neigh(dst, &fl6.daddr);
1317 if (!(msg->msg_flags&MSG_PROBE) || len)
1318 goto back_from_confirm;
1323 void udpv6_destroy_sock(struct sock *sk)
1325 struct udp_sock *up = udp_sk(sk);
1327 udp_v6_flush_pending_frames(sk);
1330 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
1331 void (*encap_destroy)(struct sock *sk);
1332 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1337 inet6_destroy_sock(sk);
1341 * Socket option code for UDP
1343 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1344 char __user *optval, unsigned int optlen)
1346 if (level == SOL_UDP || level == SOL_UDPLITE)
1347 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1348 udp_v6_push_pending_frames);
1349 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1352 #ifdef CONFIG_COMPAT
1353 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1354 char __user *optval, unsigned int optlen)
1356 if (level == SOL_UDP || level == SOL_UDPLITE)
1357 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1358 udp_v6_push_pending_frames);
1359 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1363 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1364 char __user *optval, int __user *optlen)
1366 if (level == SOL_UDP || level == SOL_UDPLITE)
1367 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1368 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1371 #ifdef CONFIG_COMPAT
1372 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1373 char __user *optval, int __user *optlen)
1375 if (level == SOL_UDP || level == SOL_UDPLITE)
1376 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1377 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1381 static const struct inet6_protocol udpv6_protocol = {
1382 .handler = udpv6_rcv,
1383 .err_handler = udpv6_err,
1384 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1387 /* ------------------------------------------------------------------------ */
1388 #ifdef CONFIG_PROC_FS
1389 int udp6_seq_show(struct seq_file *seq, void *v)
1391 if (v == SEQ_START_TOKEN) {
1392 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1394 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1395 struct inet_sock *inet = inet_sk(v);
1396 __u16 srcp = ntohs(inet->inet_sport);
1397 __u16 destp = ntohs(inet->inet_dport);
1398 ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket);
1403 static const struct file_operations udp6_afinfo_seq_fops = {
1404 .owner = THIS_MODULE,
1405 .open = udp_seq_open,
1407 .llseek = seq_lseek,
1408 .release = seq_release_net
1411 static struct udp_seq_afinfo udp6_seq_afinfo = {
1414 .udp_table = &udp_table,
1415 .seq_fops = &udp6_afinfo_seq_fops,
1417 .show = udp6_seq_show,
1421 int __net_init udp6_proc_init(struct net *net)
1423 return udp_proc_register(net, &udp6_seq_afinfo);
1426 void udp6_proc_exit(struct net *net)
1428 udp_proc_unregister(net, &udp6_seq_afinfo);
1430 #endif /* CONFIG_PROC_FS */
1432 /* ------------------------------------------------------------------------ */
1434 struct proto udpv6_prot = {
1436 .owner = THIS_MODULE,
1437 .close = udp_lib_close,
1438 .connect = ip6_datagram_connect,
1439 .disconnect = udp_disconnect,
1441 .init = udp_init_sock,
1442 .destroy = udpv6_destroy_sock,
1443 .setsockopt = udpv6_setsockopt,
1444 .getsockopt = udpv6_getsockopt,
1445 .sendmsg = udpv6_sendmsg,
1446 .recvmsg = udpv6_recvmsg,
1447 .release_cb = ip6_datagram_release_cb,
1448 .hash = udp_lib_hash,
1449 .unhash = udp_lib_unhash,
1450 .rehash = udp_v6_rehash,
1451 .get_port = udp_v6_get_port,
1452 .memory_allocated = &udp_memory_allocated,
1453 .sysctl_mem = sysctl_udp_mem,
1454 .sysctl_wmem = &sysctl_udp_wmem_min,
1455 .sysctl_rmem = &sysctl_udp_rmem_min,
1456 .obj_size = sizeof(struct udp6_sock),
1457 .h.udp_table = &udp_table,
1458 #ifdef CONFIG_COMPAT
1459 .compat_setsockopt = compat_udpv6_setsockopt,
1460 .compat_getsockopt = compat_udpv6_getsockopt,
1462 .diag_destroy = udp_abort,
1465 static struct inet_protosw udpv6_protosw = {
1467 .protocol = IPPROTO_UDP,
1468 .prot = &udpv6_prot,
1469 .ops = &inet6_dgram_ops,
1470 .flags = INET_PROTOSW_PERMANENT,
1473 int __init udpv6_init(void)
1477 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1481 ret = inet6_register_protosw(&udpv6_protosw);
1483 goto out_udpv6_protocol;
1488 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1492 void udpv6_exit(void)
1494 inet6_unregister_protosw(&udpv6_protosw);
1495 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);