projects/linux·blob master · refs

1980 lines · 50.3 KiB

1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * UDP over IPv6
4 * Linux INET6 implementation
5 *
6 * Authors:
7 * Pedro Roque <[email protected]>
8 *
9 * Based on linux/ipv4/udp.c
10 *
11 * Fixes:
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.
18 */
19
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>
38
39#include <net/addrconf.h>
40#include <net/aligned_data.h>
41#include <net/ndisc.h>
42#include <net/protocol.h>
43#include <net/transp_v6.h>
44#include <net/ip6_route.h>
45#include <net/raw.h>
46#include <net/seg6.h>
47#include <net/tcp_states.h>
48#include <net/ip6_checksum.h>
49#include <net/ip6_tunnel.h>
50#include <net/udp_tunnel.h>
51#include <net/xfrm.h>
52#include <net/inet_hashtables.h>
53#include <net/inet6_hashtables.h>
54#include <net/busy_poll.h>
55#include <net/sock_reuseport.h>
56#include <net/gro.h>
57
58#include <linux/proc_fs.h>
59#include <linux/seq_file.h>
60#include <trace/events/skb.h>
61
62static void udpv6_destruct_sock(struct sock *sk)
63{
64 udp_destruct_common(sk);
65 inet6_sock_destruct(sk);
66}
67
68static int udpv6_init_sock(struct sock *sk)
69{
70 int res = udp_lib_init_sock(sk);
71
72 sk->sk_destruct = udpv6_destruct_sock;
73 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
74 return res;
75}
76
77INDIRECT_CALLABLE_SCOPE
78u32 udp6_ehashfn(const struct net *net,
79 const struct in6_addr *laddr,
80 const u16 lport,
81 const struct in6_addr *faddr,
82 const __be16 fport)
83{
84 u32 lhash, fhash;
85
86 net_get_random_once(&udp6_ehash_secret,
87 sizeof(udp6_ehash_secret));
88 net_get_random_once(&udp_ipv6_hash_secret,
89 sizeof(udp_ipv6_hash_secret));
90
91 lhash = (__force u32)laddr->s6_addr32[3];
92 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
93
94 return __inet6_ehashfn(lhash, lport, fhash, fport,
95 udp6_ehash_secret + net_hash_mix(net));
96}
97
98static int udp_v6_get_port(struct sock *sk, unsigned short snum)
99{
100 unsigned int hash2_nulladdr =
101 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
102 unsigned int hash2_partial =
103 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
104
105 /* precompute partial secondary hash */
106 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
107 return udp_lib_get_port(sk, snum, hash2_nulladdr);
108}
109
110static void udp_v6_rehash(struct sock *sk)
111{
112 u16 new_hash = ipv6_portaddr_hash(sock_net(sk),
113 &sk->sk_v6_rcv_saddr,
114 inet_sk(sk)->inet_num);
115 u16 new_hash4;
116
117 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)) {
118 new_hash4 = udp_ehashfn(sock_net(sk),
119 sk->sk_rcv_saddr, sk->sk_num,
120 sk->sk_daddr, sk->sk_dport);
121 } else {
122 new_hash4 = udp6_ehashfn(sock_net(sk),
123 &sk->sk_v6_rcv_saddr, sk->sk_num,
124 &sk->sk_v6_daddr, sk->sk_dport);
125 }
126
127 udp_lib_rehash(sk, new_hash, new_hash4);
128}
129
130static __always_inline int
131compute_score(struct sock *sk, const struct net *net,
132 const struct in6_addr *saddr, __be16 sport,
133 const struct in6_addr *daddr, unsigned short hnum,
134 int dif, int sdif)
135{
136 int bound_dev_if, score;
137 struct inet_sock *inet;
138 bool dev_match;
139
140 if (!net_eq(sock_net(sk), net) ||
141 udp_sk(sk)->udp_port_hash != hnum ||
142 sk->sk_family != PF_INET6)
143 return -1;
144
145 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
146 return -1;
147
148 score = 0;
149 inet = inet_sk(sk);
150
151 if (inet->inet_dport) {
152 if (inet->inet_dport != sport)
153 return -1;
154 score++;
155 }
156
157 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
158 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
159 return -1;
160 score++;
161 }
162
163 bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
164 dev_match = udp_sk_bound_dev_eq(net, bound_dev_if, dif, sdif);
165 if (!dev_match)
166 return -1;
167 if (bound_dev_if)
168 score++;
169
170 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
171 score++;
172
173 return score;
174}
175
176/**
177 * udp6_lib_lookup1() - Simplified lookup using primary hash (destination port)
178 * @net: Network namespace
179 * @saddr: Source address, network order
180 * @sport: Source port, network order
181 * @daddr: Destination address, network order
182 * @hnum: Destination port, host order
183 * @dif: Destination interface index
184 * @sdif: Destination bridge port index, if relevant
185 * @udptable: Set of UDP hash tables
186 *
187 * Simplified lookup to be used as fallback if no sockets are found due to a
188 * potential race between (receive) address change, and lookup happening before
189 * the rehash operation. This function ignores SO_REUSEPORT groups while scoring
190 * result sockets, because if we have one, we don't need the fallback at all.
191 *
192 * Called under rcu_read_lock().
193 *
194 * Return: socket with highest matching score if any, NULL if none
195 */
196static struct sock *udp6_lib_lookup1(const struct net *net,
197 const struct in6_addr *saddr, __be16 sport,
198 const struct in6_addr *daddr,
199 unsigned int hnum, int dif, int sdif,
200 const struct udp_table *udptable)
201{
202 unsigned int slot = udp_hashfn(net, hnum, udptable->mask);
203 struct udp_hslot *hslot = &udptable->hash[slot];
204 struct sock *sk, *result = NULL;
205 int score, badness = 0;
206
207 sk_for_each_rcu(sk, &hslot->head) {
208 score = compute_score(sk, net,
209 saddr, sport, daddr, hnum, dif, sdif);
210 if (score > badness) {
211 result = sk;
212 badness = score;
213 }
214 }
215
216 return result;
217}
218
219/* called with rcu_read_lock() */
220static struct sock *udp6_lib_lookup2(const struct net *net,
221 const struct in6_addr *saddr, __be16 sport,
222 const struct in6_addr *daddr, unsigned int hnum,
223 int dif, int sdif, struct udp_hslot *hslot2,
224 struct sk_buff *skb)
225{
226 struct sock *sk, *result;
227 int score, badness;
228 bool need_rescore;
229
230 result = NULL;
231 badness = -1;
232 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
233 need_rescore = false;
234rescore:
235 score = compute_score(need_rescore ? result : sk, net, saddr,
236 sport, daddr, hnum, dif, sdif);
237 if (score > badness) {
238 badness = score;
239
240 if (need_rescore)
241 continue;
242
243 if (sk->sk_state == TCP_ESTABLISHED) {
244 result = sk;
245 continue;
246 }
247
248 result = inet6_lookup_reuseport(net, sk, skb, sizeof(struct udphdr),
249 saddr, sport, daddr, hnum, udp6_ehashfn);
250 if (!result) {
251 result = sk;
252 continue;
253 }
254
255 /* Fall back to scoring if group has connections */
256 if (!reuseport_has_conns(sk))
257 return result;
258
259 /* Reuseport logic returned an error, keep original score. */
260 if (IS_ERR(result))
261 continue;
262
263 /* compute_score is too long of a function to be
264 * inlined twice here, and calling it uninlined
265 * here yields measurable overhead for some
266 * workloads. Work around it by jumping
267 * backwards to rescore 'result'.
268 */
269 need_rescore = true;
270 goto rescore;
271 }
272 }
273 return result;
274}
275
276#if IS_ENABLED(CONFIG_BASE_SMALL)
277static struct sock *udp6_lib_lookup4(const struct net *net,
278 const struct in6_addr *saddr, __be16 sport,
279 const struct in6_addr *daddr,
280 unsigned int hnum, int dif, int sdif,
281 struct udp_table *udptable)
282{
283 return NULL;
284}
285
286static void udp6_hash4(struct sock *sk)
287{
288}
289#else /* !CONFIG_BASE_SMALL */
290static struct sock *udp6_lib_lookup4(const struct net *net,
291 const struct in6_addr *saddr, __be16 sport,
292 const struct in6_addr *daddr,
293 unsigned int hnum, int dif, int sdif,
294 struct udp_table *udptable)
295{
296 const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
297 const struct hlist_nulls_node *node;
298 struct udp_hslot *hslot4;
299 unsigned int hash4, slot;
300 struct udp_sock *up;
301 struct sock *sk;
302
303 hash4 = udp6_ehashfn(net, daddr, hnum, saddr, sport);
304 slot = hash4 & udptable->mask;
305 hslot4 = &udptable->hash4[slot];
306
307begin:
308 udp_lrpa_for_each_entry_rcu(up, node, &hslot4->nulls_head) {
309 sk = (struct sock *)up;
310 if (inet6_match(net, sk, saddr, daddr, ports, dif, sdif))
311 return sk;
312 }
313
314 /* if the nulls value we got at the end of this lookup is not the
315 * expected one, we must restart lookup. We probably met an item that
316 * was moved to another chain due to rehash.
317 */
318 if (get_nulls_value(node) != slot)
319 goto begin;
320
321 return NULL;
322}
323
324static void udp6_hash4(struct sock *sk)
325{
326 struct net *net = sock_net(sk);
327 unsigned int hash;
328
329 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)) {
330 udp4_hash4(sk);
331 return;
332 }
333
334 if (sk_unhashed(sk) || ipv6_addr_any(&sk->sk_v6_rcv_saddr))
335 return;
336
337 hash = udp6_ehashfn(net, &sk->sk_v6_rcv_saddr, sk->sk_num,
338 &sk->sk_v6_daddr, sk->sk_dport);
339
340 udp_lib_hash4(sk, hash);
341}
342#endif /* CONFIG_BASE_SMALL */
343
344/* rcu_read_lock() must be held */
345struct sock *__udp6_lib_lookup(const struct net *net,
346 const struct in6_addr *saddr, __be16 sport,
347 const struct in6_addr *daddr, __be16 dport,
348 int dif, int sdif, struct sk_buff *skb)
349{
350 struct udp_table *udptable = net->ipv4.udp_table;
351 unsigned short hnum = ntohs(dport);
352 struct udp_hslot *hslot2;
353 struct sock *result, *sk;
354 unsigned int hash2;
355
356 hash2 = ipv6_portaddr_hash(net, daddr, hnum);
357 hslot2 = udp_hashslot2(udptable, hash2);
358
359 if (udp_has_hash4(hslot2)) {
360 result = udp6_lib_lookup4(net, saddr, sport, daddr, hnum,
361 dif, sdif, udptable);
362 if (result) /* udp6_lib_lookup4 return sk or NULL */
363 return result;
364 }
365
366 /* Lookup connected or non-wildcard sockets */
367 result = udp6_lib_lookup2(net, saddr, sport,
368 daddr, hnum, dif, sdif,
369 hslot2, skb);
370 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
371 goto done;
372
373 /* Lookup redirect from BPF */
374 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
375 sk = inet6_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr),
376 saddr, sport, daddr, hnum, dif,
377 udp6_ehashfn);
378 if (sk) {
379 result = sk;
380 goto done;
381 }
382 }
383
384 /* Got non-wildcard socket or error on first lookup */
385 if (result)
386 goto done;
387
388 /* Lookup wildcard sockets */
389 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
390 hslot2 = udp_hashslot2(udptable, hash2);
391
392 result = udp6_lib_lookup2(net, saddr, sport,
393 &in6addr_any, hnum, dif, sdif,
394 hslot2, skb);
395 if (!IS_ERR_OR_NULL(result))
396 goto done;
397
398 /* Cover address change/lookup/rehash race: see __udp4_lib_lookup() */
399 result = udp6_lib_lookup1(net, saddr, sport, daddr, hnum, dif, sdif,
400 udptable);
401
402done:
403 if (IS_ERR(result))
404 return NULL;
405 return result;
406}
407EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
408
409static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
410 __be16 sport, __be16 dport)
411{
412 const struct ipv6hdr *iph = ipv6_hdr(skb);
413
414 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
415 &iph->daddr, dport, inet6_iif(skb),
416 inet6_sdif(skb), skb);
417}
418
419struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb,
420 __be16 sport, __be16 dport)
421{
422 const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation];
423 const struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + offset);
424 int iif, sdif;
425
426 inet6_get_iif_sdif(skb, &iif, &sdif);
427
428 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
429 &iph->daddr, dport, iif, sdif, NULL);
430}
431
432/* Must be called under rcu_read_lock().
433 * Does increment socket refcount.
434 */
435#if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
436struct sock *udp6_lib_lookup(const struct net *net, const struct in6_addr *saddr, __be16 sport,
437 const struct in6_addr *daddr, __be16 dport, int dif)
438{
439 struct sock *sk;
440
441 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, 0, NULL);
442 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
443 sk = NULL;
444 return sk;
445}
446EXPORT_SYMBOL_GPL(udp6_lib_lookup);
447#endif
448
449/* do not use the scratch area len for jumbogram: their length exceeds the
450 * scratch area space; note that the IP6CB flags is still in the first
451 * cacheline, so checking for jumbograms is cheap
452 */
453static int udp6_skb_len(struct sk_buff *skb)
454{
455 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb);
456}
457
458/*
459 * This should be easy, if there is something there we
460 * return it, otherwise we block.
461 */
462
463INDIRECT_CALLABLE_SCOPE
464int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
465 int flags)
466{
467 int off, is_udp4, err, peeking = flags & MSG_PEEK;
468 struct ipv6_pinfo *np = inet6_sk(sk);
469 struct inet_sock *inet = inet_sk(sk);
470 struct udp_mib __percpu *mib;
471 bool checksum_valid = false;
472 unsigned int ulen, copied;
473 struct sk_buff *skb;
474
475 if (flags & MSG_ERRQUEUE)
476 return ipv6_recv_error(sk, msg, len);
477
478 if (np->rxopt.bits.rxpmtu && READ_ONCE(np->rxpmtu))
479 return ipv6_recv_rxpmtu(sk, msg, len);
480
481try_again:
482 off = sk_peek_offset(sk, flags);
483 skb = __skb_recv_udp(sk, flags, &off, &err);
484 if (!skb)
485 return err;
486
487 ulen = udp6_skb_len(skb);
488 copied = len;
489 if (copied > ulen - off)
490 copied = ulen - off;
491 else if (copied < ulen)
492 msg->msg_flags |= MSG_TRUNC;
493
494 is_udp4 = (skb->protocol == htons(ETH_P_IP));
495 mib = __UDPX_MIB(sk, is_udp4);
496
497 /* If checksum is needed at all, try to do it while copying the
498 * data. If the data is truncated, do it before the copy.
499 */
500 if (copied < ulen || peeking) {
501 checksum_valid = udp_skb_csum_unnecessary(skb) ||
502 !__udp_lib_checksum_complete(skb);
503 if (!checksum_valid)
504 goto csum_copy_err;
505 }
506
507 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
508 if (udp_skb_is_linear(skb))
509 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
510 else
511 err = skb_copy_datagram_msg(skb, off, msg, copied);
512 } else {
513 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
514 if (err == -EINVAL)
515 goto csum_copy_err;
516 }
517 if (unlikely(err)) {
518 if (!peeking) {
519 udp_drops_inc(sk);
520 SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
521 }
522 kfree_skb(skb);
523 return err;
524 }
525 if (!peeking)
526 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS);
527
528 sock_recv_cmsgs(msg, sk, skb);
529
530 /* Copy the address. */
531 if (msg->msg_name) {
532 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
533 sin6->sin6_family = AF_INET6;
534 sin6->sin6_port = udp_hdr(skb)->source;
535 sin6->sin6_flowinfo = 0;
536
537 if (is_udp4) {
538 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
539 &sin6->sin6_addr);
540 sin6->sin6_scope_id = 0;
541 } else {
542 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
543 sin6->sin6_scope_id =
544 ipv6_iface_scope_id(&sin6->sin6_addr,
545 inet6_iif(skb));
546 }
547 msg->msg_namelen = sizeof(*sin6);
548
549 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk,
550 (struct sockaddr *)sin6,
551 &msg->msg_namelen);
552 }
553
554 if (udp_test_bit(GRO_ENABLED, sk))
555 udp_cmsg_recv(msg, sk, skb);
556
557 if (np->rxopt.all)
558 ip6_datagram_recv_common_ctl(sk, msg, skb);
559
560 if (is_udp4) {
561 if (inet_cmsg_flags(inet))
562 ip_cmsg_recv_offset(msg, sk, skb,
563 sizeof(struct udphdr), off);
564 } else {
565 if (np->rxopt.all)
566 ip6_datagram_recv_specific_ctl(sk, msg, skb);
567 }
568
569 err = copied;
570 if (flags & MSG_TRUNC)
571 err = ulen;
572
573 skb_consume_udp(sk, skb, peeking ? -err : err);
574 return err;
575
576csum_copy_err:
577 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
578 udp_skb_destructor)) {
579 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS);
580 SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
581 }
582 kfree_skb_reason(skb, SKB_DROP_REASON_UDP_CSUM);
583
584 /* starting over for a new packet, but check if we need to yield */
585 cond_resched();
586 msg->msg_flags &= ~MSG_TRUNC;
587 goto try_again;
588}
589
590DECLARE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
591void udpv6_encap_enable(void)
592{
593 static_branch_inc(&udpv6_encap_needed_key);
594}
595EXPORT_SYMBOL(udpv6_encap_enable);
596
597/* Handler for tunnels with arbitrary destination ports: no socket lookup, go
598 * through error handlers in encapsulations looking for a match.
599 */
600static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb,
601 struct inet6_skb_parm *opt,
602 u8 type, u8 code, int offset, __be32 info)
603{
604 int i;
605
606 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
607 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt,
608 u8 type, u8 code, int offset, __be32 info);
609 const struct ip6_tnl_encap_ops *encap;
610
611 encap = rcu_dereference(ip6tun_encaps[i]);
612 if (!encap)
613 continue;
614 handler = encap->err_handler;
615 if (handler && !handler(skb, opt, type, code, offset, info))
616 return 0;
617 }
618
619 return -ENOENT;
620}
621
622/* Try to match ICMP errors to UDP tunnels by looking up a socket without
623 * reversing source and destination port: this will match tunnels that force the
624 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
625 * lwtunnels might actually break this assumption by being configured with
626 * different destination ports on endpoints, in this case we won't be able to
627 * trace ICMP messages back to them.
628 *
629 * If this doesn't match any socket, probe tunnels with arbitrary destination
630 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
631 * we've sent packets to won't necessarily match the local destination port.
632 *
633 * Then ask the tunnel implementation to match the error against a valid
634 * association.
635 *
636 * Return an error if we can't find a match, the socket if we need further
637 * processing, zero otherwise.
638 */
639static struct sock *__udp6_lib_err_encap(struct net *net,
640 const struct ipv6hdr *hdr, int offset,
641 struct udphdr *uh,
642 struct sock *sk,
643 struct sk_buff *skb,
644 struct inet6_skb_parm *opt,
645 u8 type, u8 code, __be32 info)
646{
647 int (*lookup)(struct sock *sk, struct sk_buff *skb);
648 int network_offset, transport_offset;
649 struct udp_sock *up;
650
651 network_offset = skb_network_offset(skb);
652 transport_offset = skb_transport_offset(skb);
653
654 /* Network header needs to point to the outer IPv6 header inside ICMP */
655 skb_reset_network_header(skb);
656
657 /* Transport header needs to point to the UDP header */
658 skb_set_transport_header(skb, offset);
659
660 if (sk) {
661 up = udp_sk(sk);
662
663 lookup = READ_ONCE(up->encap_err_lookup);
664 if (lookup && lookup(sk, skb))
665 sk = NULL;
666
667 goto out;
668 }
669
670 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source,
671 &hdr->saddr, uh->dest,
672 inet6_iif(skb), 0, skb);
673 if (sk) {
674 up = udp_sk(sk);
675
676 lookup = READ_ONCE(up->encap_err_lookup);
677 if (!lookup || lookup(sk, skb))
678 sk = NULL;
679 }
680
681out:
682 if (!sk) {
683 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code,
684 offset, info));
685 }
686
687 skb_set_transport_header(skb, transport_offset);
688 skb_set_network_header(skb, network_offset);
689
690 return sk;
691}
692
693static void udpv6_err_update_exception(struct net *net, struct sk_buff *skb,
694 u8 type, __be32 info)
695{
696 if (type == ICMPV6_PKT_TOOBIG)
697 ip6_update_pmtu(skb, net, info, skb->dev->ifindex, 0,
698 sock_net_uid(net, NULL));
699 else if (type == NDISC_REDIRECT)
700 ip6_redirect(skb, net, skb->dev->ifindex, 0,
701 sock_net_uid(net, NULL));
702}
703
704static int udpv6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
705 u8 type, u8 code, int offset, __be32 info)
706{
707 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
708 struct udphdr *uh = (struct udphdr *)(skb->data + offset);
709 const struct in6_addr *saddr, *daddr;
710 struct net *net = dev_net(skb->dev);
711 struct ipv6_pinfo *np;
712 bool tunnel = false;
713 struct sock *sk;
714 int harderr;
715 int err;
716
717 udpv6_err_update_exception(net, skb, type, info);
718
719 daddr = seg6_get_daddr(skb, opt) ? : &hdr->daddr;
720 saddr = &hdr->saddr;
721 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
722 inet6_iif(skb), inet6_sdif(skb), NULL);
723
724 if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) {
725 /* No socket for error: try tunnels before discarding */
726 if (static_branch_unlikely(&udpv6_encap_needed_key)) {
727 sk = __udp6_lib_err_encap(net, hdr, offset, uh, sk, skb,
728 opt, type, code, info);
729 if (!sk)
730 return 0;
731 } else
732 sk = ERR_PTR(-ENOENT);
733
734 if (IS_ERR(sk)) {
735 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
736 ICMP6_MIB_INERRORS);
737 return PTR_ERR(sk);
738 }
739
740 tunnel = true;
741 }
742
743 harderr = icmpv6_err_convert(type, code, &err);
744 np = inet6_sk(sk);
745
746 if (type == ICMPV6_PKT_TOOBIG) {
747 if (!ip6_sk_accept_pmtu(sk))
748 goto out;
749 ip6_sk_update_pmtu(skb, sk, info);
750 if (READ_ONCE(np->pmtudisc) != IPV6_PMTUDISC_DONT)
751 harderr = 1;
752 }
753 if (type == NDISC_REDIRECT) {
754 if (tunnel) {
755 ip6_redirect(skb, sock_net(sk), inet6_iif(skb),
756 READ_ONCE(sk->sk_mark),
757 sk_uid(sk));
758 } else {
759 ip6_sk_redirect(skb, sk);
760 }
761 goto out;
762 }
763
764 /* Tunnels don't have an application socket: don't pass errors back */
765 if (tunnel) {
766 if (udp_sk(sk)->encap_err_rcv)
767 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest,
768 ntohl(info), (u8 *)(uh+1));
769 goto out;
770 }
771
772 if (!inet6_test_bit(RECVERR6, sk)) {
773 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
774 goto out;
775 } else {
776 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
777 }
778
779 sk->sk_err = err;
780 sk_error_report(sk);
781out:
782 return 0;
783}
784
785static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
786{
787 int rc;
788
789 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
790 sock_rps_save_rxhash(sk, skb);
791 sk_mark_napi_id(sk, skb);
792 sk_incoming_cpu_update(sk);
793 } else {
794 sk_mark_napi_id_once(sk, skb);
795 }
796
797 rc = __udp_enqueue_schedule_skb(sk, skb);
798 if (rc < 0) {
799 enum skb_drop_reason drop_reason;
800 struct net *net = sock_net(sk);
801
802 /* Note that an ENOMEM error is charged twice */
803 if (rc == -ENOMEM) {
804 UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS);
805 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
806 } else {
807 UDP6_INC_STATS(net, UDP_MIB_MEMERRORS);
808 drop_reason = SKB_DROP_REASON_PROTO_MEM;
809 }
810 UDP6_INC_STATS(net, UDP_MIB_INERRORS);
811 trace_udp_fail_queue_rcv_skb(rc, sk, skb);
812 sk_skb_reason_drop(sk, skb, drop_reason);
813 return -1;
814 }
815
816 return 0;
817}
818
819static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
820{
821 enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
822 struct udp_sock *up = udp_sk(sk);
823 struct net *net = sock_net(sk);
824
825 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) {
826 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
827 goto drop;
828 }
829 nf_reset_ct(skb);
830
831 if (static_branch_unlikely(&udpv6_encap_needed_key) &&
832 READ_ONCE(up->encap_type)) {
833 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
834
835 /*
836 * This is an encapsulation socket so pass the skb to
837 * the socket's udp_encap_rcv() hook. Otherwise, just
838 * fall through and pass this up the UDP socket.
839 * up->encap_rcv() returns the following value:
840 * =0 if skb was successfully passed to the encap
841 * handler or was discarded by it.
842 * >0 if skb should be passed on to UDP.
843 * <0 if skb should be resubmitted as proto -N
844 */
845
846 /* if we're overly short, let UDP handle it */
847 encap_rcv = READ_ONCE(up->encap_rcv);
848 if (encap_rcv) {
849 int ret;
850
851 /* Verify checksum before giving to encap */
852 if (udp_lib_checksum_complete(skb))
853 goto csum_error;
854
855 ret = encap_rcv(sk, skb);
856 if (ret <= 0) {
857 __UDP6_INC_STATS(net, UDP_MIB_INDATAGRAMS);
858 return -ret;
859 }
860 }
861
862 /* FALLTHROUGH -- it's a UDP Packet */
863 }
864
865 prefetch(&sk->sk_rmem_alloc);
866 if (rcu_access_pointer(sk->sk_filter) &&
867 udp_lib_checksum_complete(skb))
868 goto csum_error;
869
870 drop_reason = sk_filter_trim_cap(sk, skb, sizeof(struct udphdr));
871 if (drop_reason)
872 goto drop;
873
874 udp_csum_pull_header(skb);
875
876 skb_dst_drop(skb);
877
878 return __udpv6_queue_rcv_skb(sk, skb);
879
880csum_error:
881 drop_reason = SKB_DROP_REASON_UDP_CSUM;
882 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS);
883drop:
884 __UDP6_INC_STATS(net, UDP_MIB_INERRORS);
885 udp_drops_inc(sk);
886 sk_skb_reason_drop(sk, skb, drop_reason);
887 return -1;
888}
889
890static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
891{
892 struct sk_buff *next, *segs;
893 int ret;
894
895 if (likely(!udp_unexpected_gso(sk, skb)))
896 return udpv6_queue_rcv_one_skb(sk, skb);
897
898 __skb_push(skb, -skb_mac_offset(skb));
899 segs = udp_rcv_segment(sk, skb, false);
900 skb_list_walk_safe(segs, skb, next) {
901 __skb_pull(skb, skb_transport_offset(skb));
902
903 udp_post_segment_fix_csum(skb);
904 ret = udpv6_queue_rcv_one_skb(sk, skb);
905 if (ret > 0)
906 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret,
907 true);
908 }
909 return 0;
910}
911
912static bool __udp_v6_is_mcast_sock(struct net *net, const struct sock *sk,
913 __be16 loc_port, const struct in6_addr *loc_addr,
914 __be16 rmt_port, const struct in6_addr *rmt_addr,
915 int dif, int sdif, unsigned short hnum)
916{
917 const struct inet_sock *inet = inet_sk(sk);
918
919 if (!net_eq(sock_net(sk), net))
920 return false;
921
922 if (udp_sk(sk)->udp_port_hash != hnum ||
923 sk->sk_family != PF_INET6 ||
924 (inet->inet_dport && inet->inet_dport != rmt_port) ||
925 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
926 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
927 !udp_sk_bound_dev_eq(net, READ_ONCE(sk->sk_bound_dev_if), dif, sdif) ||
928 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
929 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
930 return false;
931 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
932 return false;
933 return true;
934}
935
936static void udp6_csum_zero_error(struct sk_buff *skb)
937{
938 /* RFC 2460 section 8.1 says that we SHOULD log
939 * this error. Well, it is reasonable.
940 */
941 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
942 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
943 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
944}
945
946/*
947 * Note: called only from the BH handler context,
948 * so we don't need to lock the hashes.
949 */
950static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
951 const struct in6_addr *saddr,
952 const struct in6_addr *daddr)
953{
954 struct udp_table *udptable = net->ipv4.udp_table;
955 const struct udphdr *uh = udp_hdr(skb);
956 unsigned int hash2, hash2_any, offset;
957 unsigned short hnum = ntohs(uh->dest);
958 struct sock *sk, *first = NULL;
959 int sdif = inet6_sdif(skb);
960 int dif = inet6_iif(skb);
961 struct hlist_node *node;
962 struct udp_hslot *hslot;
963 struct sk_buff *nskb;
964 bool use_hash2;
965 int ret;
966
967 hash2_any = 0;
968 hash2 = 0;
969 hslot = udp_hashslot(udptable, net, hnum);
970 use_hash2 = hslot->count > 10;
971 offset = offsetof(typeof(*sk), sk_node);
972
973 if (use_hash2) {
974 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) &
975 udptable->mask;
976 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask;
977start_lookup:
978 hslot = &udptable->hash2[hash2].hslot;
979 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
980 }
981
982 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
983 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
984 uh->source, saddr, dif, sdif,
985 hnum))
986 continue;
987 /* If zero checksum and no_check is not on for
988 * the socket then skip it.
989 */
990 if (!uh->check && !udp_get_no_check6_rx(sk))
991 continue;
992 if (!first) {
993 first = sk;
994 continue;
995 }
996 nskb = skb_clone(skb, GFP_ATOMIC);
997 if (unlikely(!nskb)) {
998 udp_drops_inc(sk);
999 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS);
1000 __UDP6_INC_STATS(net, UDP_MIB_INERRORS);
1001 continue;
1002 }
1003
1004 if (udpv6_queue_rcv_skb(sk, nskb) > 0)
1005 consume_skb(nskb);
1006 }
1007
1008 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
1009 if (use_hash2 && hash2 != hash2_any) {
1010 hash2 = hash2_any;
1011 goto start_lookup;
1012 }
1013
1014 if (first) {
1015 ret = udpv6_queue_rcv_skb(first, skb);
1016 if (ret > 0)
1017 return ret;
1018 } else {
1019 kfree_skb(skb);
1020 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI);
1021 }
1022 return 0;
1023}
1024
1025static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1026{
1027 if (udp_sk_rx_dst_set(sk, dst))
1028 sk->sk_rx_dst_cookie = rt6_get_cookie(dst_rt6_info(dst));
1029}
1030
1031/* wrapper for udp_queue_rcv_skb taking care of csum conversion and
1032 * return code conversion for ip layer consumption
1033 */
1034static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
1035 struct udphdr *uh)
1036{
1037 int ret;
1038
1039 if (inet_get_convert_csum(sk) && uh->check)
1040 skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo);
1041
1042 ret = udpv6_queue_rcv_skb(sk, skb);
1043
1044 /* a return value > 0 means to resubmit the input */
1045 if (ret > 0)
1046 return ret;
1047 return 0;
1048}
1049
1050static int udp6_csum_init(struct sk_buff *skb, struct udphdr *uh)
1051{
1052 int err;
1053
1054 /* To support RFC 6936 (allow zero checksum in UDP/IPV6 for tunnels)
1055 * we accept a checksum of zero here. When we find the socket
1056 * for the UDP packet we'll check if that socket allows zero checksum
1057 * for IPv6 (set by socket option).
1058 *
1059 * Note, we are only interested in != 0 or == 0, thus the
1060 * force to int.
1061 */
1062 err = (__force int)skb_checksum_init_zero_check(skb, IPPROTO_UDP, uh->check,
1063 ip6_compute_pseudo);
1064 if (err)
1065 return err;
1066
1067 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
1068 /* If SW calculated the value, we know it's bad */
1069 if (skb->csum_complete_sw)
1070 return 1;
1071
1072 /* HW says the value is bad. Let's validate that.
1073 * skb->csum is no longer the full packet checksum,
1074 * so don't treat is as such.
1075 */
1076 skb_checksum_complete_unset(skb);
1077 }
1078
1079 return 0;
1080}
1081
1082INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb)
1083{
1084 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1085 const struct in6_addr *saddr, *daddr;
1086 struct net *net = dev_net(skb->dev);
1087 struct sock *sk = NULL;
1088 struct udphdr *uh;
1089 bool refcounted;
1090 u32 ulen = 0;
1091
1092 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1093 goto discard;
1094
1095 saddr = &ipv6_hdr(skb)->saddr;
1096 daddr = &ipv6_hdr(skb)->daddr;
1097 uh = udp_hdr(skb);
1098
1099 ulen = udp_get_len(skb, uh, 0);
1100 if (ulen > skb->len)
1101 goto short_packet;
1102
1103 /* Check for jumbo payload */
1104 if (ulen == 0 && inet6_is_jumbogram(skb))
1105 ulen = skb->len;
1106
1107 if (ulen < sizeof(*uh))
1108 goto short_packet;
1109
1110 if (ulen < skb->len) {
1111 if (pskb_trim_rcsum(skb, ulen))
1112 goto short_packet;
1113
1114 saddr = &ipv6_hdr(skb)->saddr;
1115 daddr = &ipv6_hdr(skb)->daddr;
1116 uh = udp_hdr(skb);
1117 }
1118
1119 if (udp6_csum_init(skb, uh))
1120 goto csum_error;
1121
1122 /* Check if the socket is already available, e.g. due to early demux */
1123 sk = inet6_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest,
1124 &refcounted, udp6_ehashfn);
1125 if (IS_ERR(sk))
1126 goto no_sk;
1127
1128 if (sk) {
1129 struct dst_entry *dst = skb_dst(skb);
1130 int ret;
1131
1132 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
1133 udp6_sk_rx_dst_set(sk, dst);
1134
1135 if (!uh->check && !udp_get_no_check6_rx(sk)) {
1136 if (refcounted)
1137 sock_put(sk);
1138 goto report_csum_error;
1139 }
1140
1141 ret = udp6_unicast_rcv_skb(sk, skb, uh);
1142 if (refcounted)
1143 sock_put(sk);
1144 return ret;
1145 }
1146
1147 /*
1148 * Multicast receive code
1149 */
1150 if (ipv6_addr_is_multicast(daddr))
1151 return __udp6_lib_mcast_deliver(net, skb, saddr, daddr);
1152
1153 /* Unicast */
1154 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest);
1155 if (sk) {
1156 if (!uh->check && !udp_get_no_check6_rx(sk))
1157 goto report_csum_error;
1158 return udp6_unicast_rcv_skb(sk, skb, uh);
1159 }
1160no_sk:
1161 reason = SKB_DROP_REASON_NO_SOCKET;
1162
1163 if (!uh->check)
1164 goto report_csum_error;
1165
1166 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1167 goto discard;
1168 nf_reset_ct(skb);
1169
1170 if (udp_lib_checksum_complete(skb))
1171 goto csum_error;
1172
1173 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS);
1174 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
1175
1176 sk_skb_reason_drop(sk, skb, reason);
1177 return 0;
1178
1179short_packet:
1180 if (reason == SKB_DROP_REASON_NOT_SPECIFIED)
1181 reason = SKB_DROP_REASON_PKT_TOO_SMALL;
1182 net_dbg_ratelimited("UDPv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
1183 saddr, ntohs(uh->source),
1184 ulen, skb->len,
1185 daddr, ntohs(uh->dest));
1186 goto discard;
1187
1188report_csum_error:
1189 udp6_csum_zero_error(skb);
1190csum_error:
1191 if (reason == SKB_DROP_REASON_NOT_SPECIFIED)
1192 reason = SKB_DROP_REASON_UDP_CSUM;
1193 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS);
1194discard:
1195 __UDP6_INC_STATS(net, UDP_MIB_INERRORS);
1196 sk_skb_reason_drop(sk, skb, reason);
1197 return 0;
1198}
1199
1200
1201static struct sock *__udp6_lib_demux_lookup(struct net *net,
1202 __be16 loc_port, const struct in6_addr *loc_addr,
1203 __be16 rmt_port, const struct in6_addr *rmt_addr,
1204 int dif, int sdif)
1205{
1206 struct udp_table *udptable = net->ipv4.udp_table;
1207 unsigned short hnum = ntohs(loc_port);
1208 struct udp_hslot *hslot2;
1209 unsigned int hash2;
1210 __portpair ports;
1211 struct sock *sk;
1212
1213 hash2 = ipv6_portaddr_hash(net, loc_addr, hnum);
1214 hslot2 = udp_hashslot2(udptable, hash2);
1215 ports = INET_COMBINED_PORTS(rmt_port, hnum);
1216
1217 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
1218 if (sk->sk_state == TCP_ESTABLISHED &&
1219 inet6_match(net, sk, rmt_addr, loc_addr, ports, dif, sdif))
1220 return sk;
1221 /* Only check first socket in chain */
1222 break;
1223 }
1224 return NULL;
1225}
1226
1227void udp_v6_early_demux(struct sk_buff *skb)
1228{
1229 struct net *net = dev_net(skb->dev);
1230 const struct udphdr *uh;
1231 struct sock *sk;
1232 struct dst_entry *dst;
1233 int dif = skb->dev->ifindex;
1234 int sdif = inet6_sdif(skb);
1235
1236 if (!pskb_may_pull(skb, skb_transport_offset(skb) +
1237 sizeof(struct udphdr)))
1238 return;
1239
1240 uh = udp_hdr(skb);
1241
1242 if (skb->pkt_type == PACKET_HOST)
1243 sk = __udp6_lib_demux_lookup(net, uh->dest,
1244 &ipv6_hdr(skb)->daddr,
1245 uh->source, &ipv6_hdr(skb)->saddr,
1246 dif, sdif);
1247 else
1248 return;
1249
1250 if (!sk)
1251 return;
1252
1253 skb->sk = sk;
1254 DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk));
1255 skb->destructor = sock_pfree;
1256 dst = rcu_dereference(sk->sk_rx_dst);
1257
1258 if (dst)
1259 dst = dst_check(dst, sk->sk_rx_dst_cookie);
1260 if (dst) {
1261 /* set noref for now.
1262 * any place which wants to hold dst has to call
1263 * dst_hold_safe()
1264 */
1265 skb_dst_set_noref(skb, dst);
1266 }
1267}
1268
1269/*
1270 * Throw away all pending data and cancel the corking. Socket is locked.
1271 */
1272static void udp_v6_flush_pending_frames(struct sock *sk)
1273{
1274 struct udp_sock *up = udp_sk(sk);
1275
1276 if (up->pending == AF_INET)
1277 udp_flush_pending_frames(sk);
1278 else if (up->pending) {
1279 up->len = 0;
1280 WRITE_ONCE(up->pending, 0);
1281 ip6_flush_pending_frames(sk);
1282 }
1283}
1284
1285static int udpv6_pre_connect(struct sock *sk, struct sockaddr_unsized *uaddr,
1286 int addr_len)
1287{
1288 if (addr_len < offsetofend(struct sockaddr, sa_family))
1289 return -EINVAL;
1290 /* The following checks are replicated from __ip6_datagram_connect()
1291 * and intended to prevent BPF program called below from accessing
1292 * bytes that are out of the bound specified by user in addr_len.
1293 */
1294 if (uaddr->sa_family == AF_INET) {
1295 if (ipv6_only_sock(sk))
1296 return -EAFNOSUPPORT;
1297 return udp_pre_connect(sk, uaddr, addr_len);
1298 }
1299
1300 if (addr_len < SIN6_LEN_RFC2133)
1301 return -EINVAL;
1302
1303 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr, &addr_len);
1304}
1305
1306static int udpv6_connect(struct sock *sk, struct sockaddr_unsized *uaddr,
1307 int addr_len)
1308{
1309 int res;
1310
1311 lock_sock(sk);
1312 res = __ip6_datagram_connect(sk, uaddr, addr_len);
1313 if (!res)
1314 udp6_hash4(sk);
1315 release_sock(sk);
1316 return res;
1317}
1318
1319/**
1320 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
1321 * @sk: socket we are sending on
1322 * @skb: sk_buff containing the filled-in UDP header
1323 * (checksum field must be zeroed out)
1324 * @saddr: source address
1325 * @daddr: destination address
1326 * @len: length of packet
1327 */
1328static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
1329 const struct in6_addr *saddr,
1330 const struct in6_addr *daddr, int len)
1331{
1332 unsigned int offset;
1333 struct udphdr *uh = udp_hdr(skb);
1334 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
1335 __wsum csum = 0;
1336
1337 if (!frags) {
1338 /* Only one fragment on the socket. */
1339 skb->csum_start = skb_transport_header(skb) - skb->head;
1340 skb->csum_offset = offsetof(struct udphdr, check);
1341 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1342 } else {
1343 /*
1344 * HW-checksum won't work as there are two or more
1345 * fragments on the socket so that all csums of sk_buffs
1346 * should be together
1347 */
1348 offset = skb_transport_offset(skb);
1349 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1350 csum = skb->csum;
1351
1352 skb->ip_summed = CHECKSUM_NONE;
1353
1354 do {
1355 csum = csum_add(csum, frags->csum);
1356 } while ((frags = frags->next));
1357
1358 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1359 csum);
1360 if (uh->check == 0)
1361 uh->check = CSUM_MANGLED_0;
1362 }
1363}
1364
1365/*
1366 * Sending
1367 */
1368
1369static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6,
1370 struct inet_cork *cork)
1371{
1372 struct sock *sk = skb->sk;
1373 int offset, len, datalen;
1374 struct udphdr *uh;
1375 int err = 0;
1376
1377 offset = skb_transport_offset(skb);
1378 len = skb->len - offset;
1379 datalen = len - sizeof(*uh);
1380
1381 /*
1382 * Create a UDP header
1383 */
1384 uh = udp_hdr(skb);
1385 uh->source = fl6->fl6_sport;
1386 uh->dest = fl6->fl6_dport;
1387 /* Datagram length checked in udpv6_sendmsg. */
1388 udp_set_len_short(uh, len);
1389 uh->check = 0;
1390
1391 if (cork->gso_size) {
1392 const int hlen = skb_network_header_len(skb) +
1393 sizeof(struct udphdr);
1394
1395 if (hlen + min(datalen, cork->gso_size) > cork->fragsize) {
1396 kfree_skb(skb);
1397 return -EMSGSIZE;
1398 }
1399 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
1400 kfree_skb(skb);
1401 return -EINVAL;
1402 }
1403 if (udp_get_no_check6_tx(sk)) {
1404 kfree_skb(skb);
1405 return -EINVAL;
1406 }
1407 if (dst_xfrm(skb_dst(skb))) {
1408 kfree_skb(skb);
1409 return -EIO;
1410 }
1411
1412 if (datalen > cork->gso_size) {
1413 skb_shinfo(skb)->gso_size = cork->gso_size;
1414 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
1415 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
1416 cork->gso_size);
1417
1418 /* Don't checksum the payload, skb will get segmented */
1419 goto csum_partial;
1420 }
1421 }
1422
1423 if (udp_get_no_check6_tx(sk)) { /* UDP csum disabled */
1424 skb->ip_summed = CHECKSUM_NONE;
1425 goto send;
1426 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1427csum_partial:
1428 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1429 goto send;
1430 }
1431
1432 /* add protocol-dependent pseudo-header */
1433 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1434 len, IPPROTO_UDP, udp_csum(skb));
1435 if (uh->check == 0)
1436 uh->check = CSUM_MANGLED_0;
1437
1438send:
1439 err = ip6_send_skb(skb);
1440 if (unlikely(err)) {
1441 if (err == -ENOBUFS && !inet6_test_bit(RECVERR6, sk)) {
1442 UDP6_INC_STATS(sock_net(sk), UDP_MIB_SNDBUFERRORS);
1443 err = 0;
1444 }
1445 } else {
1446 UDP6_INC_STATS(sock_net(sk), UDP_MIB_OUTDATAGRAMS);
1447 }
1448 return err;
1449}
1450
1451static int udp_v6_push_pending_frames(struct sock *sk)
1452{
1453 struct sk_buff *skb;
1454 struct udp_sock *up = udp_sk(sk);
1455 int err = 0;
1456
1457 if (up->pending == AF_INET)
1458 return udp_push_pending_frames(sk);
1459
1460 skb = ip6_finish_skb(sk);
1461 if (!skb)
1462 goto out;
1463
1464 err = udp_v6_send_skb(skb, &inet_sk(sk)->cork.fl.u.ip6,
1465 &inet_sk(sk)->cork.base);
1466out:
1467 up->len = 0;
1468 WRITE_ONCE(up->pending, 0);
1469 return err;
1470}
1471
1472int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1473{
1474 int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE;
1475 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1476 struct ipv6_txoptions *opt_to_free = NULL;
1477 struct in6_addr *daddr, *final_p, final;
1478 struct ip6_flowlabel *flowlabel = NULL;
1479 struct inet_sock *inet = inet_sk(sk);
1480 struct ipv6_pinfo *np = inet6_sk(sk);
1481 struct ipv6_txoptions *opt = NULL;
1482 struct udp_sock *up = udp_sk(sk);
1483 struct ipv6_txoptions opt_space;
1484 int addr_len = msg->msg_namelen;
1485 struct inet_cork_full cork;
1486 struct ipcm6_cookie ipc6;
1487 bool connected = false;
1488 struct dst_entry *dst;
1489 struct flowi6 *fl6;
1490 int ulen = len;
1491 int err;
1492
1493 fl6 = &cork.fl.u.ip6;
1494 ipcm6_init_sk(&ipc6, sk);
1495 ipc6.gso_size = READ_ONCE(up->gso_size);
1496
1497 /* destination address check */
1498 if (sin6) {
1499 if (addr_len < offsetof(struct sockaddr, sa_data))
1500 return -EINVAL;
1501
1502 switch (sin6->sin6_family) {
1503 case AF_INET6:
1504 if (addr_len < SIN6_LEN_RFC2133)
1505 return -EINVAL;
1506 daddr = &sin6->sin6_addr;
1507 if (ipv6_addr_any(daddr) &&
1508 ipv6_addr_v4mapped(&np->saddr))
1509 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1510 daddr);
1511 break;
1512 case AF_INET:
1513 goto do_udp_sendmsg;
1514 case AF_UNSPEC:
1515 msg->msg_name = sin6 = NULL;
1516 msg->msg_namelen = addr_len = 0;
1517 daddr = NULL;
1518 break;
1519 default:
1520 return -EINVAL;
1521 }
1522 } else if (!READ_ONCE(up->pending)) {
1523 if (sk->sk_state != TCP_ESTABLISHED)
1524 return -EDESTADDRREQ;
1525 daddr = &sk->sk_v6_daddr;
1526 } else
1527 daddr = NULL;
1528
1529 if (daddr) {
1530 if (ipv6_addr_v4mapped(daddr)) {
1531 struct sockaddr_in sin;
1532 sin.sin_family = AF_INET;
1533 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1534 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1535 msg->msg_name = &sin;
1536 msg->msg_namelen = sizeof(sin);
1537do_udp_sendmsg:
1538 err = ipv6_only_sock(sk) ?
1539 -ENETUNREACH : udp_sendmsg(sk, msg, len);
1540 msg->msg_name = sin6;
1541 msg->msg_namelen = addr_len;
1542 return err;
1543 }
1544 }
1545
1546 /* Rough check on arithmetic overflow,
1547 better check is made in ip6_append_data().
1548 */
1549 if (len > INT_MAX - sizeof(struct udphdr))
1550 return -EMSGSIZE;
1551
1552 if (READ_ONCE(up->pending)) {
1553 if (READ_ONCE(up->pending) == AF_INET)
1554 return udp_sendmsg(sk, msg, len);
1555 /*
1556 * There are pending frames.
1557 * The socket lock must be held while it's corked.
1558 */
1559 lock_sock(sk);
1560 if (likely(up->pending)) {
1561 if (unlikely(up->pending != AF_INET6)) {
1562 release_sock(sk);
1563 return -EAFNOSUPPORT;
1564 }
1565 dst = NULL;
1566 goto do_append_data;
1567 }
1568 release_sock(sk);
1569 }
1570 ulen += sizeof(struct udphdr);
1571
1572 memset(fl6, 0, sizeof(*fl6));
1573
1574 if (sin6) {
1575 if (sin6->sin6_port == 0)
1576 return -EINVAL;
1577
1578 fl6->fl6_dport = sin6->sin6_port;
1579 daddr = &sin6->sin6_addr;
1580
1581 if (inet6_test_bit(SNDFLOW, sk)) {
1582 fl6->flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1583 if (fl6->flowlabel & IPV6_FLOWLABEL_MASK) {
1584 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel);
1585 if (IS_ERR(flowlabel))
1586 return -EINVAL;
1587 }
1588 }
1589
1590 /*
1591 * Otherwise it will be difficult to maintain
1592 * sk->sk_dst_cache.
1593 */
1594 if (sk->sk_state == TCP_ESTABLISHED &&
1595 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1596 daddr = &sk->sk_v6_daddr;
1597
1598 if (addr_len >= sizeof(struct sockaddr_in6) &&
1599 sin6->sin6_scope_id &&
1600 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1601 fl6->flowi6_oif = sin6->sin6_scope_id;
1602 } else {
1603 if (sk->sk_state != TCP_ESTABLISHED)
1604 return -EDESTADDRREQ;
1605
1606 fl6->fl6_dport = inet->inet_dport;
1607 daddr = &sk->sk_v6_daddr;
1608 fl6->flowlabel = np->flow_label;
1609 connected = true;
1610 }
1611
1612 if (!fl6->flowi6_oif)
1613 fl6->flowi6_oif = READ_ONCE(sk->sk_bound_dev_if);
1614
1615 if (!fl6->flowi6_oif)
1616 fl6->flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1617
1618 fl6->flowi6_uid = sk_uid(sk);
1619
1620 if (msg->msg_controllen) {
1621 opt = &opt_space;
1622 memset(opt, 0, sizeof(struct ipv6_txoptions));
1623 opt->tot_len = sizeof(*opt);
1624 ipc6.opt = opt;
1625
1626 err = udp_cmsg_send(sk, msg, &ipc6.gso_size);
1627 if (err > 0) {
1628 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, fl6,
1629 &ipc6);
1630 connected = false;
1631 }
1632 if (err < 0) {
1633 fl6_sock_release(flowlabel);
1634 return err;
1635 }
1636 if ((fl6->flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1637 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel);
1638 if (IS_ERR(flowlabel))
1639 return -EINVAL;
1640 }
1641 if (!(opt->opt_nflen|opt->opt_flen))
1642 opt = NULL;
1643 }
1644 if (!opt) {
1645 opt = txopt_get(np);
1646 opt_to_free = opt;
1647 }
1648 if (flowlabel)
1649 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1650 opt = ipv6_fixup_options(&opt_space, opt);
1651 ipc6.opt = opt;
1652
1653 fl6->flowi6_proto = IPPROTO_UDP;
1654 fl6->flowi6_mark = ipc6.sockc.mark;
1655 fl6->daddr = *daddr;
1656 if (ipv6_addr_any(&fl6->saddr) && !ipv6_addr_any(&np->saddr))
1657 fl6->saddr = np->saddr;
1658 fl6->fl6_sport = inet->inet_sport;
1659
1660 if (cgroup_bpf_enabled(CGROUP_UDP6_SENDMSG) && !connected) {
1661 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk,
1662 (struct sockaddr *)sin6,
1663 &addr_len,
1664 &fl6->saddr);
1665 if (err)
1666 goto out_no_dst;
1667 if (sin6) {
1668 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
1669 /* BPF program rewrote IPv6-only by IPv4-mapped
1670 * IPv6. It's currently unsupported.
1671 */
1672 err = -ENOTSUPP;
1673 goto out_no_dst;
1674 }
1675 if (sin6->sin6_port == 0) {
1676 /* BPF program set invalid port. Reject it. */
1677 err = -EINVAL;
1678 goto out_no_dst;
1679 }
1680 fl6->fl6_dport = sin6->sin6_port;
1681 fl6->daddr = sin6->sin6_addr;
1682 }
1683 }
1684
1685 if (ipv6_addr_any(&fl6->daddr))
1686 fl6->daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1687
1688 final_p = fl6_update_dst(fl6, opt, &final);
1689 if (final_p)
1690 connected = false;
1691
1692 if (!fl6->flowi6_oif && ipv6_addr_is_multicast(&fl6->daddr)) {
1693 fl6->flowi6_oif = READ_ONCE(np->mcast_oif);
1694 connected = false;
1695 } else if (!fl6->flowi6_oif)
1696 fl6->flowi6_oif = READ_ONCE(np->ucast_oif);
1697
1698 security_sk_classify_flow(sk, flowi6_to_flowi_common(fl6));
1699
1700 fl6->flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6->flowlabel);
1701
1702 dst = ip6_sk_dst_lookup_flow(sk, fl6, final_p, connected);
1703 if (IS_ERR(dst)) {
1704 err = PTR_ERR(dst);
1705 dst = NULL;
1706 goto out;
1707 }
1708
1709 if (ipc6.hlimit < 0)
1710 ipc6.hlimit = ip6_sk_dst_hoplimit(np, fl6, dst);
1711
1712 if (msg->msg_flags&MSG_CONFIRM)
1713 goto do_confirm;
1714back_from_confirm:
1715
1716 /* Lockless fast path for the non-corking case */
1717 if (!corkreq) {
1718 struct sk_buff *skb;
1719
1720 skb = ip6_make_skb(sk, ip_generic_getfrag, msg, ulen,
1721 sizeof(struct udphdr), &ipc6,
1722 dst_rt6_info(dst),
1723 msg->msg_flags, &cork);
1724 err = PTR_ERR(skb);
1725 if (!IS_ERR_OR_NULL(skb))
1726 err = udp_v6_send_skb(skb, fl6, &cork.base);
1727 /* ip6_make_skb steals dst reference */
1728 goto out_no_dst;
1729 }
1730
1731 lock_sock(sk);
1732 if (unlikely(up->pending)) {
1733 /* The socket is already corked while preparing it. */
1734 /* ... which is an evident application bug. --ANK */
1735 release_sock(sk);
1736
1737 net_dbg_ratelimited("udp cork app bug 2\n");
1738 err = -EINVAL;
1739 goto out;
1740 }
1741
1742 WRITE_ONCE(up->pending, AF_INET6);
1743
1744do_append_data:
1745 up->len += ulen;
1746 err = ip6_append_data(sk, ip_generic_getfrag, msg, ulen,
1747 sizeof(struct udphdr), &ipc6, fl6,
1748 dst_rt6_info(dst),
1749 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1750 if (err)
1751 udp_v6_flush_pending_frames(sk);
1752 else if (!corkreq)
1753 err = udp_v6_push_pending_frames(sk);
1754 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1755 WRITE_ONCE(up->pending, 0);
1756
1757 if (err > 0)
1758 err = inet6_test_bit(RECVERR6, sk) ? net_xmit_errno(err) : 0;
1759 release_sock(sk);
1760
1761out:
1762 dst_release(dst);
1763out_no_dst:
1764 fl6_sock_release(flowlabel);
1765 txopt_put(opt_to_free);
1766 if (!err)
1767 return len;
1768 /*
1769 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1770 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1771 * we don't have a good statistic (IpOutDiscards but it can be too many
1772 * things). We could add another new stat but at least for now that
1773 * seems like overkill.
1774 */
1775 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags))
1776 UDP6_INC_STATS(sock_net(sk), UDP_MIB_SNDBUFERRORS);
1777
1778 return err;
1779
1780do_confirm:
1781 if (msg->msg_flags & MSG_PROBE)
1782 dst_confirm_neigh(dst, &fl6->daddr);
1783 if (!(msg->msg_flags&MSG_PROBE) || len)
1784 goto back_from_confirm;
1785 err = 0;
1786 goto out;
1787}
1788EXPORT_SYMBOL(udpv6_sendmsg);
1789
1790static void udpv6_splice_eof(struct socket *sock)
1791{
1792 struct sock *sk = sock->sk;
1793 struct udp_sock *up = udp_sk(sk);
1794
1795 if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk))
1796 return;
1797
1798 lock_sock(sk);
1799 if (up->pending && !udp_test_bit(CORK, sk))
1800 udp_v6_push_pending_frames(sk);
1801 release_sock(sk);
1802}
1803
1804static void udpv6_destroy_sock(struct sock *sk)
1805{
1806 struct udp_sock *up = udp_sk(sk);
1807 lock_sock(sk);
1808
1809 /* protects from races with udp_abort() */
1810 sock_set_flag(sk, SOCK_DEAD);
1811 udp_v6_flush_pending_frames(sk);
1812 release_sock(sk);
1813
1814 if (static_branch_unlikely(&udpv6_encap_needed_key)) {
1815 if (up->encap_type) {
1816 void (*encap_destroy)(struct sock *sk);
1817 encap_destroy = READ_ONCE(up->encap_destroy);
1818 if (encap_destroy)
1819 encap_destroy(sk);
1820 }
1821 if (udp_test_bit(ENCAP_ENABLED, sk)) {
1822 static_branch_dec(&udpv6_encap_needed_key);
1823 udp_encap_disable();
1824 udp_tunnel_cleanup_gro(sk);
1825 }
1826 }
1827}
1828
1829/*
1830 * Socket option code for UDP
1831 */
1832static int udpv6_setsockopt(struct sock *sk, int level, int optname,
1833 sockptr_t optval, unsigned int optlen)
1834{
1835 if (level == SOL_UDP || level == SOL_SOCKET)
1836 return udp_lib_setsockopt(sk, level, optname,
1837 optval, optlen,
1838 udp_v6_push_pending_frames);
1839 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1840}
1841
1842static int udpv6_getsockopt(struct sock *sk, int level, int optname,
1843 char __user *optval, int __user *optlen)
1844{
1845 sockopt_t opt;
1846 int err;
1847
1848 if (level != SOL_UDP)
1849 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1850
1851 err = sockopt_init_user(&opt, optval, optlen);
1852 if (err)
1853 return err;
1854
1855 err = udp_lib_getsockopt(sk, level, optname, &opt);
1856 if (err)
1857 return err;
1858 if (put_user(opt.optlen, optlen))
1859 return -EFAULT;
1860 return 0;
1861}
1862
1863
1864/* ------------------------------------------------------------------------ */
1865#ifdef CONFIG_PROC_FS
1866static int udp6_seq_show(struct seq_file *seq, void *v)
1867{
1868 if (v == SEQ_START_TOKEN) {
1869 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1870 } else {
1871 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1872 const struct inet_sock *inet = inet_sk((const struct sock *)v);
1873 __u16 srcp = ntohs(inet->inet_sport);
1874 __u16 destp = ntohs(inet->inet_dport);
1875 __ip6_dgram_sock_seq_show(seq, v, srcp, destp,
1876 udp_rqueue_get(v), bucket);
1877 }
1878 return 0;
1879}
1880
1881static const struct seq_operations udp6_seq_ops = {
1882 .start = udp_seq_start,
1883 .next = udp_seq_next,
1884 .stop = udp_seq_stop,
1885 .show = udp6_seq_show,
1886};
1887
1888static struct udp_seq_afinfo udp6_seq_afinfo = {
1889 .family = AF_INET6,
1890};
1891
1892int __net_init udp6_proc_init(struct net *net)
1893{
1894 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops,
1895 sizeof(struct udp_iter_state), &udp6_seq_afinfo))
1896 return -ENOMEM;
1897 return 0;
1898}
1899
1900void udp6_proc_exit(struct net *net)
1901{
1902 remove_proc_entry("udp6", net->proc_net);
1903}
1904#endif /* CONFIG_PROC_FS */
1905
1906/* ------------------------------------------------------------------------ */
1907
1908struct proto udpv6_prot = {
1909 .name = "UDPv6",
1910 .owner = THIS_MODULE,
1911 .close = udp_lib_close,
1912 .pre_connect = udpv6_pre_connect,
1913 .connect = udpv6_connect,
1914 .disconnect = udp_disconnect,
1915 .ioctl = udp_ioctl,
1916 .init = udpv6_init_sock,
1917 .destroy = udpv6_destroy_sock,
1918 .setsockopt = udpv6_setsockopt,
1919 .getsockopt = udpv6_getsockopt,
1920 .sendmsg = udpv6_sendmsg,
1921 .recvmsg = udpv6_recvmsg,
1922 .splice_eof = udpv6_splice_eof,
1923 .release_cb = ip6_datagram_release_cb,
1924 .hash = udp_lib_hash,
1925 .unhash = udp_lib_unhash,
1926 .rehash = udp_v6_rehash,
1927 .get_port = udp_v6_get_port,
1928 .put_port = udp_lib_unhash,
1929#ifdef CONFIG_BPF_SYSCALL
1930 .psock_update_sk_prot = udp_bpf_update_proto,
1931#endif
1932
1933 .memory_allocated = &net_aligned_data.udp_memory_allocated,
1934 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
1935
1936 .sysctl_mem = sysctl_udp_mem,
1937 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
1938 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
1939 .obj_size = sizeof(struct udp6_sock),
1940 .ipv6_pinfo_offset = offsetof(struct udp6_sock, inet6),
1941 .diag_destroy = udp_abort,
1942};
1943
1944static struct inet_protosw udpv6_protosw = {
1945 .type = SOCK_DGRAM,
1946 .protocol = IPPROTO_UDP,
1947 .prot = &udpv6_prot,
1948 .ops = &inet6_dgram_ops,
1949 .flags = INET_PROTOSW_PERMANENT,
1950};
1951
1952int __init udpv6_init(void)
1953{
1954 int ret;
1955
1956 net_hotdata.udpv6_protocol = (struct inet6_protocol) {
1957 .handler = udpv6_rcv,
1958 .err_handler = udpv6_err,
1959 .flags = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL,
1960 };
1961 ret = inet6_add_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP);
1962 if (ret)
1963 goto out;
1964
1965 ret = inet6_register_protosw(&udpv6_protosw);
1966 if (ret)
1967 goto out_udpv6_protocol;
1968out:
1969 return ret;
1970
1971out_udpv6_protocol:
1972 inet6_del_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP);
1973 goto out;
1974}
1975
1976void udpv6_exit(void)
1977{
1978 inet6_unregister_protosw(&udpv6_protosw);
1979 inet6_del_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP);
1980}