1 // SPDX-License-Identifier: GPL-2.0-or-later
4 * Linux INET6 implementation
9 * Based on linux/ipv4/udp.c
12 * Hideaki YOSHIFUJI : sin6_scope_id support
13 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
14 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
15 * a single port at the same time.
16 * Kazunori MIYAZAWA @USAGI: change process style to use ip6_append_data
17 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/udp6 to seq_file.
20 #include <linux/bpf-cgroup.h>
21 #include <linux/errno.h>
22 #include <linux/types.h>
23 #include <linux/socket.h>
24 #include <linux/sockios.h>
25 #include <linux/net.h>
26 #include <linux/in6.h>
27 #include <linux/netdevice.h>
28 #include <linux/if_arp.h>
29 #include <linux/ipv6.h>
30 #include <linux/icmpv6.h>
31 #include <linux/init.h>
32 #include <linux/module.h>
33 #include <linux/skbuff.h>
34 #include <linux/slab.h>
35 #include <linux/uaccess.h>
36 #include <linux/indirect_call_wrapper.h>
37 #include <trace/events/udp.h>
39 #include <net/addrconf.h>
40 #include <net/ndisc.h>
41 #include <net/protocol.h>
42 #include <net/transp_v6.h>
43 #include <net/ip6_route.h>
46 #include <net/tcp_states.h>
47 #include <net/ip6_checksum.h>
48 #include <net/ip6_tunnel.h>
50 #include <net/inet_hashtables.h>
51 #include <net/inet6_hashtables.h>
52 #include <net/busy_poll.h>
53 #include <net/sock_reuseport.h>
56 #include <linux/proc_fs.h>
57 #include <linux/seq_file.h>
58 #include <trace/events/skb.h>
61 static void udpv6_destruct_sock(struct sock *sk)
63 udp_destruct_common(sk);
64 inet6_sock_destruct(sk);
67 int udpv6_init_sock(struct sock *sk)
69 udp_lib_init_sock(sk);
70 sk->sk_destruct = udpv6_destruct_sock;
71 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
75 INDIRECT_CALLABLE_SCOPE
76 u32 udp6_ehashfn(const struct net *net,
77 const struct in6_addr *laddr,
79 const struct in6_addr *faddr,
84 net_get_random_once(&udp6_ehash_secret,
85 sizeof(udp6_ehash_secret));
86 net_get_random_once(&udp_ipv6_hash_secret,
87 sizeof(udp_ipv6_hash_secret));
89 lhash = (__force u32)laddr->s6_addr32[3];
90 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
92 return __inet6_ehashfn(lhash, lport, fhash, fport,
93 udp6_ehash_secret + net_hash_mix(net));
96 int udp_v6_get_port(struct sock *sk, unsigned short snum)
98 unsigned int hash2_nulladdr =
99 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
100 unsigned int hash2_partial =
101 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
103 /* precompute partial secondary hash */
104 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
105 return udp_lib_get_port(sk, snum, hash2_nulladdr);
108 void udp_v6_rehash(struct sock *sk)
110 u16 new_hash = ipv6_portaddr_hash(sock_net(sk),
111 &sk->sk_v6_rcv_saddr,
112 inet_sk(sk)->inet_num);
114 udp_lib_rehash(sk, new_hash);
117 static int compute_score(struct sock *sk, struct net *net,
118 const struct in6_addr *saddr, __be16 sport,
119 const struct in6_addr *daddr, unsigned short hnum,
122 int bound_dev_if, score;
123 struct inet_sock *inet;
126 if (!net_eq(sock_net(sk), net) ||
127 udp_sk(sk)->udp_port_hash != hnum ||
128 sk->sk_family != PF_INET6)
131 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
137 if (inet->inet_dport) {
138 if (inet->inet_dport != sport)
143 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
144 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
149 bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
150 dev_match = udp_sk_bound_dev_eq(net, bound_dev_if, dif, sdif);
156 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
162 /* called with rcu_read_lock() */
163 static struct sock *udp6_lib_lookup2(struct net *net,
164 const struct in6_addr *saddr, __be16 sport,
165 const struct in6_addr *daddr, unsigned int hnum,
166 int dif, int sdif, struct udp_hslot *hslot2,
169 struct sock *sk, *result;
175 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
176 need_rescore = false;
178 score = compute_score(need_rescore ? result : sk, net, saddr,
179 sport, daddr, hnum, dif, sdif);
180 if (score > badness) {
186 if (sk->sk_state == TCP_ESTABLISHED) {
191 result = inet6_lookup_reuseport(net, sk, skb, sizeof(struct udphdr),
192 saddr, sport, daddr, hnum, udp6_ehashfn);
198 /* Fall back to scoring if group has connections */
199 if (!reuseport_has_conns(sk))
202 /* Reuseport logic returned an error, keep original score. */
206 /* compute_score is too long of a function to be
207 * inlined, and calling it again here yields
208 * measureable overhead for some
209 * workloads. Work around it by jumping
210 * backwards to rescore 'result'.
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 unsigned short hnum = ntohs(dport);
227 unsigned int hash2, slot2;
228 struct udp_hslot *hslot2;
229 struct sock *result, *sk;
231 hash2 = ipv6_portaddr_hash(net, daddr, hnum);
232 slot2 = hash2 & udptable->mask;
233 hslot2 = &udptable->hash2[slot2];
235 /* Lookup connected or non-wildcard sockets */
236 result = udp6_lib_lookup2(net, saddr, sport,
237 daddr, hnum, dif, sdif,
239 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
242 /* Lookup redirect from BPF */
243 if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
244 udptable == net->ipv4.udp_table) {
245 sk = inet6_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr),
246 saddr, sport, daddr, hnum, dif,
254 /* Got non-wildcard socket or error on first lookup */
258 /* Lookup wildcard sockets */
259 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
260 slot2 = hash2 & udptable->mask;
261 hslot2 = &udptable->hash2[slot2];
263 result = udp6_lib_lookup2(net, saddr, sport,
264 &in6addr_any, hnum, dif, sdif,
271 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
273 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
274 __be16 sport, __be16 dport,
275 struct udp_table *udptable)
277 const struct ipv6hdr *iph = ipv6_hdr(skb);
279 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
280 &iph->daddr, dport, inet6_iif(skb),
281 inet6_sdif(skb), udptable, skb);
284 struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb,
285 __be16 sport, __be16 dport)
287 const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation];
288 const struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + offset);
289 struct net *net = dev_net(skb->dev);
292 inet6_get_iif_sdif(skb, &iif, &sdif);
294 return __udp6_lib_lookup(net, &iph->saddr, sport,
295 &iph->daddr, dport, iif,
296 sdif, net->ipv4.udp_table, NULL);
299 /* Must be called under rcu_read_lock().
300 * Does increment socket refcount.
302 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
303 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
304 const struct in6_addr *daddr, __be16 dport, int dif)
308 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport,
309 dif, 0, net->ipv4.udp_table, NULL);
310 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
314 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
317 /* do not use the scratch area len for jumbogram: their length execeeds the
318 * scratch area space; note that the IP6CB flags is still in the first
319 * cacheline, so checking for jumbograms is cheap
321 static int udp6_skb_len(struct sk_buff *skb)
323 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb);
327 * This should be easy, if there is something there we
328 * return it, otherwise we block.
331 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
332 int flags, int *addr_len)
334 struct ipv6_pinfo *np = inet6_sk(sk);
335 struct inet_sock *inet = inet_sk(sk);
337 unsigned int ulen, copied;
338 int off, err, peeking = flags & MSG_PEEK;
339 int is_udplite = IS_UDPLITE(sk);
340 struct udp_mib __percpu *mib;
341 bool checksum_valid = false;
344 if (flags & MSG_ERRQUEUE)
345 return ipv6_recv_error(sk, msg, len, addr_len);
347 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
348 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
351 off = sk_peek_offset(sk, flags);
352 skb = __skb_recv_udp(sk, flags, &off, &err);
356 ulen = udp6_skb_len(skb);
358 if (copied > ulen - off)
360 else if (copied < ulen)
361 msg->msg_flags |= MSG_TRUNC;
363 is_udp4 = (skb->protocol == htons(ETH_P_IP));
364 mib = __UDPX_MIB(sk, is_udp4);
367 * If checksum is needed at all, try to do it while copying the
368 * data. If the data is truncated, or if we only want a partial
369 * coverage checksum (UDP-Lite), do it before the copy.
372 if (copied < ulen || peeking ||
373 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
374 checksum_valid = udp_skb_csum_unnecessary(skb) ||
375 !__udp_lib_checksum_complete(skb);
380 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
381 if (udp_skb_is_linear(skb))
382 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
384 err = skb_copy_datagram_msg(skb, off, msg, copied);
386 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
392 atomic_inc(&sk->sk_drops);
393 SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
399 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS);
401 sock_recv_cmsgs(msg, sk, skb);
403 /* Copy the address. */
405 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
406 sin6->sin6_family = AF_INET6;
407 sin6->sin6_port = udp_hdr(skb)->source;
408 sin6->sin6_flowinfo = 0;
411 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
413 sin6->sin6_scope_id = 0;
415 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
416 sin6->sin6_scope_id =
417 ipv6_iface_scope_id(&sin6->sin6_addr,
420 *addr_len = sizeof(*sin6);
422 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk,
423 (struct sockaddr *)sin6,
427 if (udp_test_bit(GRO_ENABLED, sk))
428 udp_cmsg_recv(msg, sk, skb);
431 ip6_datagram_recv_common_ctl(sk, msg, skb);
434 if (inet_cmsg_flags(inet))
435 ip_cmsg_recv_offset(msg, sk, skb,
436 sizeof(struct udphdr), off);
439 ip6_datagram_recv_specific_ctl(sk, msg, skb);
443 if (flags & MSG_TRUNC)
446 skb_consume_udp(sk, skb, peeking ? -err : err);
450 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
451 udp_skb_destructor)) {
452 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS);
453 SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
457 /* starting over for a new packet, but check if we need to yield */
459 msg->msg_flags &= ~MSG_TRUNC;
463 DECLARE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
464 void udpv6_encap_enable(void)
466 static_branch_inc(&udpv6_encap_needed_key);
468 EXPORT_SYMBOL(udpv6_encap_enable);
470 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
471 * through error handlers in encapsulations looking for a match.
473 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb,
474 struct inet6_skb_parm *opt,
475 u8 type, u8 code, int offset, __be32 info)
479 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
480 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt,
481 u8 type, u8 code, int offset, __be32 info);
482 const struct ip6_tnl_encap_ops *encap;
484 encap = rcu_dereference(ip6tun_encaps[i]);
487 handler = encap->err_handler;
488 if (handler && !handler(skb, opt, type, code, offset, info))
495 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
496 * reversing source and destination port: this will match tunnels that force the
497 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
498 * lwtunnels might actually break this assumption by being configured with
499 * different destination ports on endpoints, in this case we won't be able to
500 * trace ICMP messages back to them.
502 * If this doesn't match any socket, probe tunnels with arbitrary destination
503 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
504 * we've sent packets to won't necessarily match the local destination port.
506 * Then ask the tunnel implementation to match the error against a valid
509 * Return an error if we can't find a match, the socket if we need further
510 * processing, zero otherwise.
512 static struct sock *__udp6_lib_err_encap(struct net *net,
513 const struct ipv6hdr *hdr, int offset,
515 struct udp_table *udptable,
518 struct inet6_skb_parm *opt,
519 u8 type, u8 code, __be32 info)
521 int (*lookup)(struct sock *sk, struct sk_buff *skb);
522 int network_offset, transport_offset;
525 network_offset = skb_network_offset(skb);
526 transport_offset = skb_transport_offset(skb);
528 /* Network header needs to point to the outer IPv6 header inside ICMP */
529 skb_reset_network_header(skb);
531 /* Transport header needs to point to the UDP header */
532 skb_set_transport_header(skb, offset);
537 lookup = READ_ONCE(up->encap_err_lookup);
538 if (lookup && lookup(sk, skb))
544 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source,
545 &hdr->saddr, uh->dest,
546 inet6_iif(skb), 0, udptable, skb);
550 lookup = READ_ONCE(up->encap_err_lookup);
551 if (!lookup || lookup(sk, skb))
557 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code,
561 skb_set_transport_header(skb, transport_offset);
562 skb_set_network_header(skb, network_offset);
567 int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
568 u8 type, u8 code, int offset, __be32 info,
569 struct udp_table *udptable)
571 struct ipv6_pinfo *np;
572 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
573 const struct in6_addr *saddr = &hdr->saddr;
574 const struct in6_addr *daddr = seg6_get_daddr(skb, opt) ? : &hdr->daddr;
575 struct udphdr *uh = (struct udphdr *)(skb->data+offset);
580 struct net *net = dev_net(skb->dev);
582 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
583 inet6_iif(skb), inet6_sdif(skb), udptable, NULL);
585 if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) {
586 /* No socket for error: try tunnels before discarding */
587 if (static_branch_unlikely(&udpv6_encap_needed_key)) {
588 sk = __udp6_lib_err_encap(net, hdr, offset, uh,
590 opt, type, code, info);
594 sk = ERR_PTR(-ENOENT);
597 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
605 harderr = icmpv6_err_convert(type, code, &err);
608 if (type == ICMPV6_PKT_TOOBIG) {
609 if (!ip6_sk_accept_pmtu(sk))
611 ip6_sk_update_pmtu(skb, sk, info);
612 if (READ_ONCE(np->pmtudisc) != IPV6_PMTUDISC_DONT)
615 if (type == NDISC_REDIRECT) {
617 ip6_redirect(skb, sock_net(sk), inet6_iif(skb),
618 READ_ONCE(sk->sk_mark), sk->sk_uid);
620 ip6_sk_redirect(skb, sk);
625 /* Tunnels don't have an application socket: don't pass errors back */
627 if (udp_sk(sk)->encap_err_rcv)
628 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest,
629 ntohl(info), (u8 *)(uh+1));
633 if (!inet6_test_bit(RECVERR6, sk)) {
634 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
637 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
646 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
650 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
651 sock_rps_save_rxhash(sk, skb);
652 sk_mark_napi_id(sk, skb);
653 sk_incoming_cpu_update(sk);
655 sk_mark_napi_id_once(sk, skb);
658 rc = __udp_enqueue_schedule_skb(sk, skb);
660 int is_udplite = IS_UDPLITE(sk);
661 enum skb_drop_reason drop_reason;
663 /* Note that an ENOMEM error is charged twice */
665 UDP6_INC_STATS(sock_net(sk),
666 UDP_MIB_RCVBUFERRORS, is_udplite);
667 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
669 UDP6_INC_STATS(sock_net(sk),
670 UDP_MIB_MEMERRORS, is_udplite);
671 drop_reason = SKB_DROP_REASON_PROTO_MEM;
673 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
674 trace_udp_fail_queue_rcv_skb(rc, sk, skb);
675 sk_skb_reason_drop(sk, skb, drop_reason);
682 static __inline__ int udpv6_err(struct sk_buff *skb,
683 struct inet6_skb_parm *opt, u8 type,
684 u8 code, int offset, __be32 info)
686 return __udp6_lib_err(skb, opt, type, code, offset, info,
687 dev_net(skb->dev)->ipv4.udp_table);
690 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
692 enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
693 struct udp_sock *up = udp_sk(sk);
694 int is_udplite = IS_UDPLITE(sk);
696 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) {
697 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
702 if (static_branch_unlikely(&udpv6_encap_needed_key) &&
703 READ_ONCE(up->encap_type)) {
704 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
707 * This is an encapsulation socket so pass the skb to
708 * the socket's udp_encap_rcv() hook. Otherwise, just
709 * fall through and pass this up the UDP socket.
710 * up->encap_rcv() returns the following value:
711 * =0 if skb was successfully passed to the encap
712 * handler or was discarded by it.
713 * >0 if skb should be passed on to UDP.
714 * <0 if skb should be resubmitted as proto -N
717 /* if we're overly short, let UDP handle it */
718 encap_rcv = READ_ONCE(up->encap_rcv);
722 /* Verify checksum before giving to encap */
723 if (udp_lib_checksum_complete(skb))
726 ret = encap_rcv(sk, skb);
728 __UDP6_INC_STATS(sock_net(sk),
735 /* FALLTHROUGH -- it's a UDP Packet */
739 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
741 if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) {
742 u16 pcrlen = READ_ONCE(up->pcrlen);
744 if (pcrlen == 0) { /* full coverage was set */
745 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
746 UDP_SKB_CB(skb)->cscov, skb->len);
749 if (UDP_SKB_CB(skb)->cscov < pcrlen) {
750 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
751 UDP_SKB_CB(skb)->cscov, pcrlen);
756 prefetch(&sk->sk_rmem_alloc);
757 if (rcu_access_pointer(sk->sk_filter) &&
758 udp_lib_checksum_complete(skb))
761 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
762 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
766 udp_csum_pull_header(skb);
770 return __udpv6_queue_rcv_skb(sk, skb);
773 drop_reason = SKB_DROP_REASON_UDP_CSUM;
774 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
776 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
777 atomic_inc(&sk->sk_drops);
778 sk_skb_reason_drop(sk, skb, drop_reason);
782 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
784 struct sk_buff *next, *segs;
787 if (likely(!udp_unexpected_gso(sk, skb)))
788 return udpv6_queue_rcv_one_skb(sk, skb);
790 __skb_push(skb, -skb_mac_offset(skb));
791 segs = udp_rcv_segment(sk, skb, false);
792 skb_list_walk_safe(segs, skb, next) {
793 __skb_pull(skb, skb_transport_offset(skb));
795 udp_post_segment_fix_csum(skb);
796 ret = udpv6_queue_rcv_one_skb(sk, skb);
798 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret,
804 static bool __udp_v6_is_mcast_sock(struct net *net, const struct sock *sk,
805 __be16 loc_port, const struct in6_addr *loc_addr,
806 __be16 rmt_port, const struct in6_addr *rmt_addr,
807 int dif, int sdif, unsigned short hnum)
809 const struct inet_sock *inet = inet_sk(sk);
811 if (!net_eq(sock_net(sk), net))
814 if (udp_sk(sk)->udp_port_hash != hnum ||
815 sk->sk_family != PF_INET6 ||
816 (inet->inet_dport && inet->inet_dport != rmt_port) ||
817 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
818 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
819 !udp_sk_bound_dev_eq(net, READ_ONCE(sk->sk_bound_dev_if), dif, sdif) ||
820 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
821 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
823 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
828 static void udp6_csum_zero_error(struct sk_buff *skb)
830 /* RFC 2460 section 8.1 says that we SHOULD log
831 * this error. Well, it is reasonable.
833 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
834 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
835 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
839 * Note: called only from the BH handler context,
840 * so we don't need to lock the hashes.
842 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
843 const struct in6_addr *saddr, const struct in6_addr *daddr,
844 struct udp_table *udptable, int proto)
846 struct sock *sk, *first = NULL;
847 const struct udphdr *uh = udp_hdr(skb);
848 unsigned short hnum = ntohs(uh->dest);
849 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
850 unsigned int offset = offsetof(typeof(*sk), sk_node);
851 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
852 int dif = inet6_iif(skb);
853 int sdif = inet6_sdif(skb);
854 struct hlist_node *node;
855 struct sk_buff *nskb;
858 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) &
860 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask;
862 hslot = &udptable->hash2[hash2];
863 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
866 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
867 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
868 uh->source, saddr, dif, sdif,
871 /* If zero checksum and no_check is not on for
872 * the socket then skip it.
874 if (!uh->check && !udp_get_no_check6_rx(sk))
880 nskb = skb_clone(skb, GFP_ATOMIC);
881 if (unlikely(!nskb)) {
882 atomic_inc(&sk->sk_drops);
883 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
885 __UDP6_INC_STATS(net, UDP_MIB_INERRORS,
890 if (udpv6_queue_rcv_skb(sk, nskb) > 0)
894 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
895 if (use_hash2 && hash2 != hash2_any) {
901 if (udpv6_queue_rcv_skb(first, skb) > 0)
905 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
906 proto == IPPROTO_UDPLITE);
911 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
913 if (udp_sk_rx_dst_set(sk, dst))
914 sk->sk_rx_dst_cookie = rt6_get_cookie(dst_rt6_info(dst));
917 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
918 * return code conversion for ip layer consumption
920 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
925 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
926 skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo);
928 ret = udpv6_queue_rcv_skb(sk, skb);
930 /* a return value > 0 means to resubmit the input */
936 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
939 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
940 const struct in6_addr *saddr, *daddr;
941 struct net *net = dev_net(skb->dev);
942 struct sock *sk = NULL;
947 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
950 saddr = &ipv6_hdr(skb)->saddr;
951 daddr = &ipv6_hdr(skb)->daddr;
954 ulen = ntohs(uh->len);
958 if (proto == IPPROTO_UDP) {
959 /* UDP validates ulen. */
961 /* Check for jumbo payload */
965 if (ulen < sizeof(*uh))
968 if (ulen < skb->len) {
969 if (pskb_trim_rcsum(skb, ulen))
971 saddr = &ipv6_hdr(skb)->saddr;
972 daddr = &ipv6_hdr(skb)->daddr;
977 if (udp6_csum_init(skb, uh, proto))
980 /* Check if the socket is already available, e.g. due to early demux */
981 sk = inet6_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest,
982 &refcounted, udp6_ehashfn);
987 struct dst_entry *dst = skb_dst(skb);
990 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
991 udp6_sk_rx_dst_set(sk, dst);
993 if (!uh->check && !udp_get_no_check6_rx(sk)) {
996 goto report_csum_error;
999 ret = udp6_unicast_rcv_skb(sk, skb, uh);
1006 * Multicast receive code
1008 if (ipv6_addr_is_multicast(daddr))
1009 return __udp6_lib_mcast_deliver(net, skb,
1010 saddr, daddr, udptable, proto);
1013 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1015 if (!uh->check && !udp_get_no_check6_rx(sk))
1016 goto report_csum_error;
1017 return udp6_unicast_rcv_skb(sk, skb, uh);
1020 reason = SKB_DROP_REASON_NO_SOCKET;
1023 goto report_csum_error;
1025 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1029 if (udp_lib_checksum_complete(skb))
1032 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1033 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
1035 sk_skb_reason_drop(sk, skb, reason);
1039 if (reason == SKB_DROP_REASON_NOT_SPECIFIED)
1040 reason = SKB_DROP_REASON_PKT_TOO_SMALL;
1041 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
1042 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1043 saddr, ntohs(uh->source),
1045 daddr, ntohs(uh->dest));
1049 udp6_csum_zero_error(skb);
1051 if (reason == SKB_DROP_REASON_NOT_SPECIFIED)
1052 reason = SKB_DROP_REASON_UDP_CSUM;
1053 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1055 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1056 sk_skb_reason_drop(sk, skb, reason);
1061 static struct sock *__udp6_lib_demux_lookup(struct net *net,
1062 __be16 loc_port, const struct in6_addr *loc_addr,
1063 __be16 rmt_port, const struct in6_addr *rmt_addr,
1066 struct udp_table *udptable = net->ipv4.udp_table;
1067 unsigned short hnum = ntohs(loc_port);
1068 unsigned int hash2, slot2;
1069 struct udp_hslot *hslot2;
1073 hash2 = ipv6_portaddr_hash(net, loc_addr, hnum);
1074 slot2 = hash2 & udptable->mask;
1075 hslot2 = &udptable->hash2[slot2];
1076 ports = INET_COMBINED_PORTS(rmt_port, hnum);
1078 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
1079 if (sk->sk_state == TCP_ESTABLISHED &&
1080 inet6_match(net, sk, rmt_addr, loc_addr, ports, dif, sdif))
1082 /* Only check first socket in chain */
1088 void udp_v6_early_demux(struct sk_buff *skb)
1090 struct net *net = dev_net(skb->dev);
1091 const struct udphdr *uh;
1093 struct dst_entry *dst;
1094 int dif = skb->dev->ifindex;
1095 int sdif = inet6_sdif(skb);
1097 if (!pskb_may_pull(skb, skb_transport_offset(skb) +
1098 sizeof(struct udphdr)))
1103 if (skb->pkt_type == PACKET_HOST)
1104 sk = __udp6_lib_demux_lookup(net, uh->dest,
1105 &ipv6_hdr(skb)->daddr,
1106 uh->source, &ipv6_hdr(skb)->saddr,
1115 DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk));
1116 skb->destructor = sock_pfree;
1117 dst = rcu_dereference(sk->sk_rx_dst);
1120 dst = dst_check(dst, sk->sk_rx_dst_cookie);
1122 /* set noref for now.
1123 * any place which wants to hold dst has to call
1126 skb_dst_set_noref(skb, dst);
1130 INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb)
1132 return __udp6_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
1136 * Throw away all pending data and cancel the corking. Socket is locked.
1138 static void udp_v6_flush_pending_frames(struct sock *sk)
1140 struct udp_sock *up = udp_sk(sk);
1142 if (up->pending == AF_INET)
1143 udp_flush_pending_frames(sk);
1144 else if (up->pending) {
1146 WRITE_ONCE(up->pending, 0);
1147 ip6_flush_pending_frames(sk);
1151 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
1154 if (addr_len < offsetofend(struct sockaddr, sa_family))
1156 /* The following checks are replicated from __ip6_datagram_connect()
1157 * and intended to prevent BPF program called below from accessing
1158 * bytes that are out of the bound specified by user in addr_len.
1160 if (uaddr->sa_family == AF_INET) {
1161 if (ipv6_only_sock(sk))
1162 return -EAFNOSUPPORT;
1163 return udp_pre_connect(sk, uaddr, addr_len);
1166 if (addr_len < SIN6_LEN_RFC2133)
1169 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr, &addr_len);
1173 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
1174 * @sk: socket we are sending on
1175 * @skb: sk_buff containing the filled-in UDP header
1176 * (checksum field must be zeroed out)
1177 * @saddr: source address
1178 * @daddr: destination address
1179 * @len: length of packet
1181 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
1182 const struct in6_addr *saddr,
1183 const struct in6_addr *daddr, int len)
1185 unsigned int offset;
1186 struct udphdr *uh = udp_hdr(skb);
1187 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
1191 /* Only one fragment on the socket. */
1192 skb->csum_start = skb_transport_header(skb) - skb->head;
1193 skb->csum_offset = offsetof(struct udphdr, check);
1194 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1197 * HW-checksum won't work as there are two or more
1198 * fragments on the socket so that all csums of sk_buffs
1199 * should be together
1201 offset = skb_transport_offset(skb);
1202 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1205 skb->ip_summed = CHECKSUM_NONE;
1208 csum = csum_add(csum, frags->csum);
1209 } while ((frags = frags->next));
1211 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1214 uh->check = CSUM_MANGLED_0;
1222 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6,
1223 struct inet_cork *cork)
1225 struct sock *sk = skb->sk;
1228 int is_udplite = IS_UDPLITE(sk);
1230 int offset = skb_transport_offset(skb);
1231 int len = skb->len - offset;
1232 int datalen = len - sizeof(*uh);
1235 * Create a UDP header
1238 uh->source = fl6->fl6_sport;
1239 uh->dest = fl6->fl6_dport;
1240 uh->len = htons(len);
1243 if (cork->gso_size) {
1244 const int hlen = skb_network_header_len(skb) +
1245 sizeof(struct udphdr);
1247 if (hlen + cork->gso_size > cork->fragsize) {
1251 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
1255 if (udp_get_no_check6_tx(sk)) {
1259 if (is_udplite || dst_xfrm(skb_dst(skb))) {
1264 if (datalen > cork->gso_size) {
1265 skb_shinfo(skb)->gso_size = cork->gso_size;
1266 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
1267 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
1274 csum = udplite_csum(skb);
1275 else if (udp_get_no_check6_tx(sk)) { /* UDP csum disabled */
1276 skb->ip_summed = CHECKSUM_NONE;
1278 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1280 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1283 csum = udp_csum(skb);
1285 /* add protocol-dependent pseudo-header */
1286 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1287 len, fl6->flowi6_proto, csum);
1289 uh->check = CSUM_MANGLED_0;
1292 err = ip6_send_skb(skb);
1294 if (err == -ENOBUFS && !inet6_test_bit(RECVERR6, sk)) {
1295 UDP6_INC_STATS(sock_net(sk),
1296 UDP_MIB_SNDBUFERRORS, is_udplite);
1300 UDP6_INC_STATS(sock_net(sk),
1301 UDP_MIB_OUTDATAGRAMS, is_udplite);
1306 static int udp_v6_push_pending_frames(struct sock *sk)
1308 struct sk_buff *skb;
1309 struct udp_sock *up = udp_sk(sk);
1312 if (up->pending == AF_INET)
1313 return udp_push_pending_frames(sk);
1315 skb = ip6_finish_skb(sk);
1319 err = udp_v6_send_skb(skb, &inet_sk(sk)->cork.fl.u.ip6,
1320 &inet_sk(sk)->cork.base);
1323 WRITE_ONCE(up->pending, 0);
1327 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1329 struct ipv6_txoptions opt_space;
1330 struct udp_sock *up = udp_sk(sk);
1331 struct inet_sock *inet = inet_sk(sk);
1332 struct ipv6_pinfo *np = inet6_sk(sk);
1333 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1334 struct in6_addr *daddr, *final_p, final;
1335 struct ipv6_txoptions *opt = NULL;
1336 struct ipv6_txoptions *opt_to_free = NULL;
1337 struct ip6_flowlabel *flowlabel = NULL;
1338 struct inet_cork_full cork;
1339 struct flowi6 *fl6 = &cork.fl.u.ip6;
1340 struct dst_entry *dst;
1341 struct ipcm6_cookie ipc6;
1342 int addr_len = msg->msg_namelen;
1343 bool connected = false;
1345 int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE;
1347 int is_udplite = IS_UDPLITE(sk);
1348 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1351 ipc6.gso_size = READ_ONCE(up->gso_size);
1352 ipc6.sockc.tsflags = READ_ONCE(sk->sk_tsflags);
1353 ipc6.sockc.mark = READ_ONCE(sk->sk_mark);
1355 /* destination address check */
1357 if (addr_len < offsetof(struct sockaddr, sa_data))
1360 switch (sin6->sin6_family) {
1362 if (addr_len < SIN6_LEN_RFC2133)
1364 daddr = &sin6->sin6_addr;
1365 if (ipv6_addr_any(daddr) &&
1366 ipv6_addr_v4mapped(&np->saddr))
1367 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1371 goto do_udp_sendmsg;
1373 msg->msg_name = sin6 = NULL;
1374 msg->msg_namelen = addr_len = 0;
1380 } else if (!READ_ONCE(up->pending)) {
1381 if (sk->sk_state != TCP_ESTABLISHED)
1382 return -EDESTADDRREQ;
1383 daddr = &sk->sk_v6_daddr;
1388 if (ipv6_addr_v4mapped(daddr)) {
1389 struct sockaddr_in sin;
1390 sin.sin_family = AF_INET;
1391 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1392 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1393 msg->msg_name = &sin;
1394 msg->msg_namelen = sizeof(sin);
1396 err = ipv6_only_sock(sk) ?
1397 -ENETUNREACH : udp_sendmsg(sk, msg, len);
1398 msg->msg_name = sin6;
1399 msg->msg_namelen = addr_len;
1404 /* Rough check on arithmetic overflow,
1405 better check is made in ip6_append_data().
1407 if (len > INT_MAX - sizeof(struct udphdr))
1410 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1411 if (READ_ONCE(up->pending)) {
1412 if (READ_ONCE(up->pending) == AF_INET)
1413 return udp_sendmsg(sk, msg, len);
1415 * There are pending frames.
1416 * The socket lock must be held while it's corked.
1419 if (likely(up->pending)) {
1420 if (unlikely(up->pending != AF_INET6)) {
1422 return -EAFNOSUPPORT;
1425 goto do_append_data;
1429 ulen += sizeof(struct udphdr);
1431 memset(fl6, 0, sizeof(*fl6));
1434 if (sin6->sin6_port == 0)
1437 fl6->fl6_dport = sin6->sin6_port;
1438 daddr = &sin6->sin6_addr;
1440 if (inet6_test_bit(SNDFLOW, sk)) {
1441 fl6->flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1442 if (fl6->flowlabel & IPV6_FLOWLABEL_MASK) {
1443 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel);
1444 if (IS_ERR(flowlabel))
1450 * Otherwise it will be difficult to maintain
1453 if (sk->sk_state == TCP_ESTABLISHED &&
1454 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1455 daddr = &sk->sk_v6_daddr;
1457 if (addr_len >= sizeof(struct sockaddr_in6) &&
1458 sin6->sin6_scope_id &&
1459 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1460 fl6->flowi6_oif = sin6->sin6_scope_id;
1462 if (sk->sk_state != TCP_ESTABLISHED)
1463 return -EDESTADDRREQ;
1465 fl6->fl6_dport = inet->inet_dport;
1466 daddr = &sk->sk_v6_daddr;
1467 fl6->flowlabel = np->flow_label;
1471 if (!fl6->flowi6_oif)
1472 fl6->flowi6_oif = READ_ONCE(sk->sk_bound_dev_if);
1474 if (!fl6->flowi6_oif)
1475 fl6->flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1477 fl6->flowi6_uid = sk->sk_uid;
1479 if (msg->msg_controllen) {
1481 memset(opt, 0, sizeof(struct ipv6_txoptions));
1482 opt->tot_len = sizeof(*opt);
1485 err = udp_cmsg_send(sk, msg, &ipc6.gso_size);
1487 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, fl6,
1492 fl6_sock_release(flowlabel);
1495 if ((fl6->flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1496 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel);
1497 if (IS_ERR(flowlabel))
1500 if (!(opt->opt_nflen|opt->opt_flen))
1504 opt = txopt_get(np);
1508 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1509 opt = ipv6_fixup_options(&opt_space, opt);
1512 fl6->flowi6_proto = sk->sk_protocol;
1513 fl6->flowi6_mark = ipc6.sockc.mark;
1514 fl6->daddr = *daddr;
1515 if (ipv6_addr_any(&fl6->saddr) && !ipv6_addr_any(&np->saddr))
1516 fl6->saddr = np->saddr;
1517 fl6->fl6_sport = inet->inet_sport;
1519 if (cgroup_bpf_enabled(CGROUP_UDP6_SENDMSG) && !connected) {
1520 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk,
1521 (struct sockaddr *)sin6,
1527 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
1528 /* BPF program rewrote IPv6-only by IPv4-mapped
1529 * IPv6. It's currently unsupported.
1534 if (sin6->sin6_port == 0) {
1535 /* BPF program set invalid port. Reject it. */
1539 fl6->fl6_dport = sin6->sin6_port;
1540 fl6->daddr = sin6->sin6_addr;
1544 if (ipv6_addr_any(&fl6->daddr))
1545 fl6->daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1547 final_p = fl6_update_dst(fl6, opt, &final);
1551 if (!fl6->flowi6_oif && ipv6_addr_is_multicast(&fl6->daddr)) {
1552 fl6->flowi6_oif = READ_ONCE(np->mcast_oif);
1554 } else if (!fl6->flowi6_oif)
1555 fl6->flowi6_oif = READ_ONCE(np->ucast_oif);
1557 security_sk_classify_flow(sk, flowi6_to_flowi_common(fl6));
1559 if (ipc6.tclass < 0)
1560 ipc6.tclass = np->tclass;
1562 fl6->flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6->flowlabel);
1564 dst = ip6_sk_dst_lookup_flow(sk, fl6, final_p, connected);
1571 if (ipc6.hlimit < 0)
1572 ipc6.hlimit = ip6_sk_dst_hoplimit(np, fl6, dst);
1574 if (msg->msg_flags&MSG_CONFIRM)
1578 /* Lockless fast path for the non-corking case */
1580 struct sk_buff *skb;
1582 skb = ip6_make_skb(sk, getfrag, msg, ulen,
1583 sizeof(struct udphdr), &ipc6,
1585 msg->msg_flags, &cork);
1587 if (!IS_ERR_OR_NULL(skb))
1588 err = udp_v6_send_skb(skb, fl6, &cork.base);
1589 /* ip6_make_skb steals dst reference */
1594 if (unlikely(up->pending)) {
1595 /* The socket is already corked while preparing it. */
1596 /* ... which is an evident application bug. --ANK */
1599 net_dbg_ratelimited("udp cork app bug 2\n");
1604 WRITE_ONCE(up->pending, AF_INET6);
1607 if (ipc6.dontfrag < 0)
1608 ipc6.dontfrag = inet6_test_bit(DONTFRAG, sk);
1610 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr),
1611 &ipc6, fl6, dst_rt6_info(dst),
1612 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1614 udp_v6_flush_pending_frames(sk);
1616 err = udp_v6_push_pending_frames(sk);
1617 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1618 WRITE_ONCE(up->pending, 0);
1621 err = inet6_test_bit(RECVERR6, sk) ? net_xmit_errno(err) : 0;
1627 fl6_sock_release(flowlabel);
1628 txopt_put(opt_to_free);
1632 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1633 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1634 * we don't have a good statistic (IpOutDiscards but it can be too many
1635 * things). We could add another new stat but at least for now that
1636 * seems like overkill.
1638 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1639 UDP6_INC_STATS(sock_net(sk),
1640 UDP_MIB_SNDBUFERRORS, is_udplite);
1645 if (msg->msg_flags & MSG_PROBE)
1646 dst_confirm_neigh(dst, &fl6->daddr);
1647 if (!(msg->msg_flags&MSG_PROBE) || len)
1648 goto back_from_confirm;
1652 EXPORT_SYMBOL(udpv6_sendmsg);
1654 static void udpv6_splice_eof(struct socket *sock)
1656 struct sock *sk = sock->sk;
1657 struct udp_sock *up = udp_sk(sk);
1659 if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk))
1663 if (up->pending && !udp_test_bit(CORK, sk))
1664 udp_v6_push_pending_frames(sk);
1668 void udpv6_destroy_sock(struct sock *sk)
1670 struct udp_sock *up = udp_sk(sk);
1673 /* protects from races with udp_abort() */
1674 sock_set_flag(sk, SOCK_DEAD);
1675 udp_v6_flush_pending_frames(sk);
1678 if (static_branch_unlikely(&udpv6_encap_needed_key)) {
1679 if (up->encap_type) {
1680 void (*encap_destroy)(struct sock *sk);
1681 encap_destroy = READ_ONCE(up->encap_destroy);
1685 if (udp_test_bit(ENCAP_ENABLED, sk)) {
1686 static_branch_dec(&udpv6_encap_needed_key);
1687 udp_encap_disable();
1693 * Socket option code for UDP
1695 int udpv6_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
1696 unsigned int optlen)
1698 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET)
1699 return udp_lib_setsockopt(sk, level, optname,
1701 udp_v6_push_pending_frames);
1702 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1705 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1706 char __user *optval, int __user *optlen)
1708 if (level == SOL_UDP || level == SOL_UDPLITE)
1709 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1710 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1714 /* ------------------------------------------------------------------------ */
1715 #ifdef CONFIG_PROC_FS
1716 int udp6_seq_show(struct seq_file *seq, void *v)
1718 if (v == SEQ_START_TOKEN) {
1719 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1721 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1722 const struct inet_sock *inet = inet_sk((const struct sock *)v);
1723 __u16 srcp = ntohs(inet->inet_sport);
1724 __u16 destp = ntohs(inet->inet_dport);
1725 __ip6_dgram_sock_seq_show(seq, v, srcp, destp,
1726 udp_rqueue_get(v), bucket);
1731 const struct seq_operations udp6_seq_ops = {
1732 .start = udp_seq_start,
1733 .next = udp_seq_next,
1734 .stop = udp_seq_stop,
1735 .show = udp6_seq_show,
1737 EXPORT_SYMBOL(udp6_seq_ops);
1739 static struct udp_seq_afinfo udp6_seq_afinfo = {
1744 int __net_init udp6_proc_init(struct net *net)
1746 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops,
1747 sizeof(struct udp_iter_state), &udp6_seq_afinfo))
1752 void udp6_proc_exit(struct net *net)
1754 remove_proc_entry("udp6", net->proc_net);
1756 #endif /* CONFIG_PROC_FS */
1758 /* ------------------------------------------------------------------------ */
1760 struct proto udpv6_prot = {
1762 .owner = THIS_MODULE,
1763 .close = udp_lib_close,
1764 .pre_connect = udpv6_pre_connect,
1765 .connect = ip6_datagram_connect,
1766 .disconnect = udp_disconnect,
1768 .init = udpv6_init_sock,
1769 .destroy = udpv6_destroy_sock,
1770 .setsockopt = udpv6_setsockopt,
1771 .getsockopt = udpv6_getsockopt,
1772 .sendmsg = udpv6_sendmsg,
1773 .recvmsg = udpv6_recvmsg,
1774 .splice_eof = udpv6_splice_eof,
1775 .release_cb = ip6_datagram_release_cb,
1776 .hash = udp_lib_hash,
1777 .unhash = udp_lib_unhash,
1778 .rehash = udp_v6_rehash,
1779 .get_port = udp_v6_get_port,
1780 .put_port = udp_lib_unhash,
1781 #ifdef CONFIG_BPF_SYSCALL
1782 .psock_update_sk_prot = udp_bpf_update_proto,
1785 .memory_allocated = &udp_memory_allocated,
1786 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
1788 .sysctl_mem = sysctl_udp_mem,
1789 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
1790 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
1791 .obj_size = sizeof(struct udp6_sock),
1792 .ipv6_pinfo_offset = offsetof(struct udp6_sock, inet6),
1793 .h.udp_table = NULL,
1794 .diag_destroy = udp_abort,
1797 static struct inet_protosw udpv6_protosw = {
1799 .protocol = IPPROTO_UDP,
1800 .prot = &udpv6_prot,
1801 .ops = &inet6_dgram_ops,
1802 .flags = INET_PROTOSW_PERMANENT,
1805 int __init udpv6_init(void)
1809 net_hotdata.udpv6_protocol = (struct inet6_protocol) {
1810 .handler = udpv6_rcv,
1811 .err_handler = udpv6_err,
1812 .flags = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL,
1814 ret = inet6_add_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP);
1818 ret = inet6_register_protosw(&udpv6_protosw);
1820 goto out_udpv6_protocol;
1825 inet6_del_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP);
1829 void udpv6_exit(void)
1831 inet6_unregister_protosw(&udpv6_protosw);
1832 inet6_del_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP);