Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * UDP over IPv6
4 * Linux INET6 implementation
5 *
6 * Authors:
7 * Pedro Roque <roque@di.fc.ul.pt>
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
38#include <net/addrconf.h>
39#include <net/ndisc.h>
40#include <net/protocol.h>
41#include <net/transp_v6.h>
42#include <net/ip6_route.h>
43#include <net/raw.h>
44#include <net/seg6.h>
45#include <net/tcp_states.h>
46#include <net/ip6_checksum.h>
47#include <net/ip6_tunnel.h>
48#include <trace/events/udp.h>
49#include <net/xfrm.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>
54#include <net/gro.h>
55
56#include <linux/proc_fs.h>
57#include <linux/seq_file.h>
58#include <trace/events/skb.h>
59#include "udp_impl.h"
60
61static void udpv6_destruct_sock(struct sock *sk)
62{
63 udp_destruct_common(sk);
64 inet6_sock_destruct(sk);
65}
66
67int udpv6_init_sock(struct sock *sk)
68{
69 udp_lib_init_sock(sk);
70 sk->sk_destruct = udpv6_destruct_sock;
71 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
72 return 0;
73}
74
75INDIRECT_CALLABLE_SCOPE
76u32 udp6_ehashfn(const struct net *net,
77 const struct in6_addr *laddr,
78 const u16 lport,
79 const struct in6_addr *faddr,
80 const __be16 fport)
81{
82 static u32 udp6_ehash_secret __read_mostly;
83 static u32 udp_ipv6_hash_secret __read_mostly;
84
85 u32 lhash, fhash;
86
87 net_get_random_once(&udp6_ehash_secret,
88 sizeof(udp6_ehash_secret));
89 net_get_random_once(&udp_ipv6_hash_secret,
90 sizeof(udp_ipv6_hash_secret));
91
92 lhash = (__force u32)laddr->s6_addr32[3];
93 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
94
95 return __inet6_ehashfn(lhash, lport, fhash, fport,
96 udp6_ehash_secret + net_hash_mix(net));
97}
98
99int udp_v6_get_port(struct sock *sk, unsigned short snum)
100{
101 unsigned int hash2_nulladdr =
102 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
103 unsigned int hash2_partial =
104 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
105
106 /* precompute partial secondary hash */
107 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
108 return udp_lib_get_port(sk, snum, hash2_nulladdr);
109}
110
111void udp_v6_rehash(struct sock *sk)
112{
113 u16 new_hash = ipv6_portaddr_hash(sock_net(sk),
114 &sk->sk_v6_rcv_saddr,
115 inet_sk(sk)->inet_num);
116
117 udp_lib_rehash(sk, new_hash);
118}
119
120static int compute_score(struct sock *sk, struct net *net,
121 const struct in6_addr *saddr, __be16 sport,
122 const struct in6_addr *daddr, unsigned short hnum,
123 int dif, int sdif)
124{
125 int bound_dev_if, score;
126 struct inet_sock *inet;
127 bool dev_match;
128
129 if (!net_eq(sock_net(sk), net) ||
130 udp_sk(sk)->udp_port_hash != hnum ||
131 sk->sk_family != PF_INET6)
132 return -1;
133
134 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
135 return -1;
136
137 score = 0;
138 inet = inet_sk(sk);
139
140 if (inet->inet_dport) {
141 if (inet->inet_dport != sport)
142 return -1;
143 score++;
144 }
145
146 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
147 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
148 return -1;
149 score++;
150 }
151
152 bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
153 dev_match = udp_sk_bound_dev_eq(net, bound_dev_if, dif, sdif);
154 if (!dev_match)
155 return -1;
156 if (bound_dev_if)
157 score++;
158
159 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
160 score++;
161
162 return score;
163}
164
165/* called with rcu_read_lock() */
166static struct sock *udp6_lib_lookup2(struct net *net,
167 const struct in6_addr *saddr, __be16 sport,
168 const struct in6_addr *daddr, unsigned int hnum,
169 int dif, int sdif, struct udp_hslot *hslot2,
170 struct sk_buff *skb)
171{
172 struct sock *sk, *result;
173 int score, badness;
174
175 result = NULL;
176 badness = -1;
177 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
178 score = compute_score(sk, net, saddr, sport,
179 daddr, hnum, dif, sdif);
180 if (score > badness) {
181 badness = score;
182
183 if (sk->sk_state == TCP_ESTABLISHED) {
184 result = sk;
185 continue;
186 }
187
188 result = inet6_lookup_reuseport(net, sk, skb, sizeof(struct udphdr),
189 saddr, sport, daddr, hnum, udp6_ehashfn);
190 if (!result) {
191 result = sk;
192 continue;
193 }
194
195 /* Fall back to scoring if group has connections */
196 if (!reuseport_has_conns(sk))
197 return result;
198
199 /* Reuseport logic returned an error, keep original score. */
200 if (IS_ERR(result))
201 continue;
202
203 badness = compute_score(sk, net, saddr, sport,
204 daddr, hnum, dif, sdif);
205 }
206 }
207 return result;
208}
209
210/* rcu_read_lock() must be held */
211struct sock *__udp6_lib_lookup(struct net *net,
212 const struct in6_addr *saddr, __be16 sport,
213 const struct in6_addr *daddr, __be16 dport,
214 int dif, int sdif, struct udp_table *udptable,
215 struct sk_buff *skb)
216{
217 unsigned short hnum = ntohs(dport);
218 unsigned int hash2, slot2;
219 struct udp_hslot *hslot2;
220 struct sock *result, *sk;
221
222 hash2 = ipv6_portaddr_hash(net, daddr, hnum);
223 slot2 = hash2 & udptable->mask;
224 hslot2 = &udptable->hash2[slot2];
225
226 /* Lookup connected or non-wildcard sockets */
227 result = udp6_lib_lookup2(net, saddr, sport,
228 daddr, hnum, dif, sdif,
229 hslot2, skb);
230 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
231 goto done;
232
233 /* Lookup redirect from BPF */
234 if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
235 udptable == net->ipv4.udp_table) {
236 sk = inet6_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr),
237 saddr, sport, daddr, hnum, dif,
238 udp6_ehashfn);
239 if (sk) {
240 result = sk;
241 goto done;
242 }
243 }
244
245 /* Got non-wildcard socket or error on first lookup */
246 if (result)
247 goto done;
248
249 /* Lookup wildcard sockets */
250 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
251 slot2 = hash2 & udptable->mask;
252 hslot2 = &udptable->hash2[slot2];
253
254 result = udp6_lib_lookup2(net, saddr, sport,
255 &in6addr_any, hnum, dif, sdif,
256 hslot2, skb);
257done:
258 if (IS_ERR(result))
259 return NULL;
260 return result;
261}
262EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
263
264static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
265 __be16 sport, __be16 dport,
266 struct udp_table *udptable)
267{
268 const struct ipv6hdr *iph = ipv6_hdr(skb);
269
270 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
271 &iph->daddr, dport, inet6_iif(skb),
272 inet6_sdif(skb), udptable, skb);
273}
274
275struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb,
276 __be16 sport, __be16 dport)
277{
278 const struct ipv6hdr *iph = ipv6_hdr(skb);
279 struct net *net = dev_net(skb->dev);
280 int iif, sdif;
281
282 inet6_get_iif_sdif(skb, &iif, &sdif);
283
284 return __udp6_lib_lookup(net, &iph->saddr, sport,
285 &iph->daddr, dport, iif,
286 sdif, net->ipv4.udp_table, NULL);
287}
288
289/* Must be called under rcu_read_lock().
290 * Does increment socket refcount.
291 */
292#if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
293struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
294 const struct in6_addr *daddr, __be16 dport, int dif)
295{
296 struct sock *sk;
297
298 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport,
299 dif, 0, net->ipv4.udp_table, NULL);
300 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
301 sk = NULL;
302 return sk;
303}
304EXPORT_SYMBOL_GPL(udp6_lib_lookup);
305#endif
306
307/* do not use the scratch area len for jumbogram: their length execeeds the
308 * scratch area space; note that the IP6CB flags is still in the first
309 * cacheline, so checking for jumbograms is cheap
310 */
311static int udp6_skb_len(struct sk_buff *skb)
312{
313 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb);
314}
315
316/*
317 * This should be easy, if there is something there we
318 * return it, otherwise we block.
319 */
320
321int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
322 int flags, int *addr_len)
323{
324 struct ipv6_pinfo *np = inet6_sk(sk);
325 struct inet_sock *inet = inet_sk(sk);
326 struct sk_buff *skb;
327 unsigned int ulen, copied;
328 int off, err, peeking = flags & MSG_PEEK;
329 int is_udplite = IS_UDPLITE(sk);
330 struct udp_mib __percpu *mib;
331 bool checksum_valid = false;
332 int is_udp4;
333
334 if (flags & MSG_ERRQUEUE)
335 return ipv6_recv_error(sk, msg, len, addr_len);
336
337 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
338 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
339
340try_again:
341 off = sk_peek_offset(sk, flags);
342 skb = __skb_recv_udp(sk, flags, &off, &err);
343 if (!skb)
344 return err;
345
346 ulen = udp6_skb_len(skb);
347 copied = len;
348 if (copied > ulen - off)
349 copied = ulen - off;
350 else if (copied < ulen)
351 msg->msg_flags |= MSG_TRUNC;
352
353 is_udp4 = (skb->protocol == htons(ETH_P_IP));
354 mib = __UDPX_MIB(sk, is_udp4);
355
356 /*
357 * If checksum is needed at all, try to do it while copying the
358 * data. If the data is truncated, or if we only want a partial
359 * coverage checksum (UDP-Lite), do it before the copy.
360 */
361
362 if (copied < ulen || peeking ||
363 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
364 checksum_valid = udp_skb_csum_unnecessary(skb) ||
365 !__udp_lib_checksum_complete(skb);
366 if (!checksum_valid)
367 goto csum_copy_err;
368 }
369
370 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
371 if (udp_skb_is_linear(skb))
372 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
373 else
374 err = skb_copy_datagram_msg(skb, off, msg, copied);
375 } else {
376 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
377 if (err == -EINVAL)
378 goto csum_copy_err;
379 }
380 if (unlikely(err)) {
381 if (!peeking) {
382 atomic_inc(&sk->sk_drops);
383 SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
384 }
385 kfree_skb(skb);
386 return err;
387 }
388 if (!peeking)
389 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS);
390
391 sock_recv_cmsgs(msg, sk, skb);
392
393 /* Copy the address. */
394 if (msg->msg_name) {
395 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
396 sin6->sin6_family = AF_INET6;
397 sin6->sin6_port = udp_hdr(skb)->source;
398 sin6->sin6_flowinfo = 0;
399
400 if (is_udp4) {
401 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
402 &sin6->sin6_addr);
403 sin6->sin6_scope_id = 0;
404 } else {
405 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
406 sin6->sin6_scope_id =
407 ipv6_iface_scope_id(&sin6->sin6_addr,
408 inet6_iif(skb));
409 }
410 *addr_len = sizeof(*sin6);
411
412 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk,
413 (struct sockaddr *)sin6,
414 addr_len);
415 }
416
417 if (udp_test_bit(GRO_ENABLED, sk))
418 udp_cmsg_recv(msg, sk, skb);
419
420 if (np->rxopt.all)
421 ip6_datagram_recv_common_ctl(sk, msg, skb);
422
423 if (is_udp4) {
424 if (inet_cmsg_flags(inet))
425 ip_cmsg_recv_offset(msg, sk, skb,
426 sizeof(struct udphdr), off);
427 } else {
428 if (np->rxopt.all)
429 ip6_datagram_recv_specific_ctl(sk, msg, skb);
430 }
431
432 err = copied;
433 if (flags & MSG_TRUNC)
434 err = ulen;
435
436 skb_consume_udp(sk, skb, peeking ? -err : err);
437 return err;
438
439csum_copy_err:
440 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
441 udp_skb_destructor)) {
442 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS);
443 SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
444 }
445 kfree_skb(skb);
446
447 /* starting over for a new packet, but check if we need to yield */
448 cond_resched();
449 msg->msg_flags &= ~MSG_TRUNC;
450 goto try_again;
451}
452
453DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
454void udpv6_encap_enable(void)
455{
456 static_branch_inc(&udpv6_encap_needed_key);
457}
458EXPORT_SYMBOL(udpv6_encap_enable);
459
460/* Handler for tunnels with arbitrary destination ports: no socket lookup, go
461 * through error handlers in encapsulations looking for a match.
462 */
463static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb,
464 struct inet6_skb_parm *opt,
465 u8 type, u8 code, int offset, __be32 info)
466{
467 int i;
468
469 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
470 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt,
471 u8 type, u8 code, int offset, __be32 info);
472 const struct ip6_tnl_encap_ops *encap;
473
474 encap = rcu_dereference(ip6tun_encaps[i]);
475 if (!encap)
476 continue;
477 handler = encap->err_handler;
478 if (handler && !handler(skb, opt, type, code, offset, info))
479 return 0;
480 }
481
482 return -ENOENT;
483}
484
485/* Try to match ICMP errors to UDP tunnels by looking up a socket without
486 * reversing source and destination port: this will match tunnels that force the
487 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
488 * lwtunnels might actually break this assumption by being configured with
489 * different destination ports on endpoints, in this case we won't be able to
490 * trace ICMP messages back to them.
491 *
492 * If this doesn't match any socket, probe tunnels with arbitrary destination
493 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
494 * we've sent packets to won't necessarily match the local destination port.
495 *
496 * Then ask the tunnel implementation to match the error against a valid
497 * association.
498 *
499 * Return an error if we can't find a match, the socket if we need further
500 * processing, zero otherwise.
501 */
502static struct sock *__udp6_lib_err_encap(struct net *net,
503 const struct ipv6hdr *hdr, int offset,
504 struct udphdr *uh,
505 struct udp_table *udptable,
506 struct sock *sk,
507 struct sk_buff *skb,
508 struct inet6_skb_parm *opt,
509 u8 type, u8 code, __be32 info)
510{
511 int (*lookup)(struct sock *sk, struct sk_buff *skb);
512 int network_offset, transport_offset;
513 struct udp_sock *up;
514
515 network_offset = skb_network_offset(skb);
516 transport_offset = skb_transport_offset(skb);
517
518 /* Network header needs to point to the outer IPv6 header inside ICMP */
519 skb_reset_network_header(skb);
520
521 /* Transport header needs to point to the UDP header */
522 skb_set_transport_header(skb, offset);
523
524 if (sk) {
525 up = udp_sk(sk);
526
527 lookup = READ_ONCE(up->encap_err_lookup);
528 if (lookup && lookup(sk, skb))
529 sk = NULL;
530
531 goto out;
532 }
533
534 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source,
535 &hdr->saddr, uh->dest,
536 inet6_iif(skb), 0, udptable, skb);
537 if (sk) {
538 up = udp_sk(sk);
539
540 lookup = READ_ONCE(up->encap_err_lookup);
541 if (!lookup || lookup(sk, skb))
542 sk = NULL;
543 }
544
545out:
546 if (!sk) {
547 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code,
548 offset, info));
549 }
550
551 skb_set_transport_header(skb, transport_offset);
552 skb_set_network_header(skb, network_offset);
553
554 return sk;
555}
556
557int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
558 u8 type, u8 code, int offset, __be32 info,
559 struct udp_table *udptable)
560{
561 struct ipv6_pinfo *np;
562 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
563 const struct in6_addr *saddr = &hdr->saddr;
564 const struct in6_addr *daddr = seg6_get_daddr(skb, opt) ? : &hdr->daddr;
565 struct udphdr *uh = (struct udphdr *)(skb->data+offset);
566 bool tunnel = false;
567 struct sock *sk;
568 int harderr;
569 int err;
570 struct net *net = dev_net(skb->dev);
571
572 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
573 inet6_iif(skb), inet6_sdif(skb), udptable, NULL);
574
575 if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) {
576 /* No socket for error: try tunnels before discarding */
577 if (static_branch_unlikely(&udpv6_encap_needed_key)) {
578 sk = __udp6_lib_err_encap(net, hdr, offset, uh,
579 udptable, sk, skb,
580 opt, type, code, info);
581 if (!sk)
582 return 0;
583 } else
584 sk = ERR_PTR(-ENOENT);
585
586 if (IS_ERR(sk)) {
587 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
588 ICMP6_MIB_INERRORS);
589 return PTR_ERR(sk);
590 }
591
592 tunnel = true;
593 }
594
595 harderr = icmpv6_err_convert(type, code, &err);
596 np = inet6_sk(sk);
597
598 if (type == ICMPV6_PKT_TOOBIG) {
599 if (!ip6_sk_accept_pmtu(sk))
600 goto out;
601 ip6_sk_update_pmtu(skb, sk, info);
602 if (READ_ONCE(np->pmtudisc) != IPV6_PMTUDISC_DONT)
603 harderr = 1;
604 }
605 if (type == NDISC_REDIRECT) {
606 if (tunnel) {
607 ip6_redirect(skb, sock_net(sk), inet6_iif(skb),
608 READ_ONCE(sk->sk_mark), sk->sk_uid);
609 } else {
610 ip6_sk_redirect(skb, sk);
611 }
612 goto out;
613 }
614
615 /* Tunnels don't have an application socket: don't pass errors back */
616 if (tunnel) {
617 if (udp_sk(sk)->encap_err_rcv)
618 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest,
619 ntohl(info), (u8 *)(uh+1));
620 goto out;
621 }
622
623 if (!inet6_test_bit(RECVERR6, sk)) {
624 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
625 goto out;
626 } else {
627 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
628 }
629
630 sk->sk_err = err;
631 sk_error_report(sk);
632out:
633 return 0;
634}
635
636static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
637{
638 int rc;
639
640 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
641 sock_rps_save_rxhash(sk, skb);
642 sk_mark_napi_id(sk, skb);
643 sk_incoming_cpu_update(sk);
644 } else {
645 sk_mark_napi_id_once(sk, skb);
646 }
647
648 rc = __udp_enqueue_schedule_skb(sk, skb);
649 if (rc < 0) {
650 int is_udplite = IS_UDPLITE(sk);
651 enum skb_drop_reason drop_reason;
652
653 /* Note that an ENOMEM error is charged twice */
654 if (rc == -ENOMEM) {
655 UDP6_INC_STATS(sock_net(sk),
656 UDP_MIB_RCVBUFERRORS, is_udplite);
657 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
658 } else {
659 UDP6_INC_STATS(sock_net(sk),
660 UDP_MIB_MEMERRORS, is_udplite);
661 drop_reason = SKB_DROP_REASON_PROTO_MEM;
662 }
663 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
664 kfree_skb_reason(skb, drop_reason);
665 trace_udp_fail_queue_rcv_skb(rc, sk);
666 return -1;
667 }
668
669 return 0;
670}
671
672static __inline__ int udpv6_err(struct sk_buff *skb,
673 struct inet6_skb_parm *opt, u8 type,
674 u8 code, int offset, __be32 info)
675{
676 return __udp6_lib_err(skb, opt, type, code, offset, info,
677 dev_net(skb->dev)->ipv4.udp_table);
678}
679
680static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
681{
682 enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
683 struct udp_sock *up = udp_sk(sk);
684 int is_udplite = IS_UDPLITE(sk);
685
686 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) {
687 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
688 goto drop;
689 }
690 nf_reset_ct(skb);
691
692 if (static_branch_unlikely(&udpv6_encap_needed_key) &&
693 READ_ONCE(up->encap_type)) {
694 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
695
696 /*
697 * This is an encapsulation socket so pass the skb to
698 * the socket's udp_encap_rcv() hook. Otherwise, just
699 * fall through and pass this up the UDP socket.
700 * up->encap_rcv() returns the following value:
701 * =0 if skb was successfully passed to the encap
702 * handler or was discarded by it.
703 * >0 if skb should be passed on to UDP.
704 * <0 if skb should be resubmitted as proto -N
705 */
706
707 /* if we're overly short, let UDP handle it */
708 encap_rcv = READ_ONCE(up->encap_rcv);
709 if (encap_rcv) {
710 int ret;
711
712 /* Verify checksum before giving to encap */
713 if (udp_lib_checksum_complete(skb))
714 goto csum_error;
715
716 ret = encap_rcv(sk, skb);
717 if (ret <= 0) {
718 __UDP6_INC_STATS(sock_net(sk),
719 UDP_MIB_INDATAGRAMS,
720 is_udplite);
721 return -ret;
722 }
723 }
724
725 /* FALLTHROUGH -- it's a UDP Packet */
726 }
727
728 /*
729 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
730 */
731 if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) {
732 u16 pcrlen = READ_ONCE(up->pcrlen);
733
734 if (pcrlen == 0) { /* full coverage was set */
735 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
736 UDP_SKB_CB(skb)->cscov, skb->len);
737 goto drop;
738 }
739 if (UDP_SKB_CB(skb)->cscov < pcrlen) {
740 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
741 UDP_SKB_CB(skb)->cscov, pcrlen);
742 goto drop;
743 }
744 }
745
746 prefetch(&sk->sk_rmem_alloc);
747 if (rcu_access_pointer(sk->sk_filter) &&
748 udp_lib_checksum_complete(skb))
749 goto csum_error;
750
751 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
752 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
753 goto drop;
754 }
755
756 udp_csum_pull_header(skb);
757
758 skb_dst_drop(skb);
759
760 return __udpv6_queue_rcv_skb(sk, skb);
761
762csum_error:
763 drop_reason = SKB_DROP_REASON_UDP_CSUM;
764 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
765drop:
766 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
767 atomic_inc(&sk->sk_drops);
768 kfree_skb_reason(skb, drop_reason);
769 return -1;
770}
771
772static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
773{
774 struct sk_buff *next, *segs;
775 int ret;
776
777 if (likely(!udp_unexpected_gso(sk, skb)))
778 return udpv6_queue_rcv_one_skb(sk, skb);
779
780 __skb_push(skb, -skb_mac_offset(skb));
781 segs = udp_rcv_segment(sk, skb, false);
782 skb_list_walk_safe(segs, skb, next) {
783 __skb_pull(skb, skb_transport_offset(skb));
784
785 udp_post_segment_fix_csum(skb);
786 ret = udpv6_queue_rcv_one_skb(sk, skb);
787 if (ret > 0)
788 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret,
789 true);
790 }
791 return 0;
792}
793
794static bool __udp_v6_is_mcast_sock(struct net *net, const struct sock *sk,
795 __be16 loc_port, const struct in6_addr *loc_addr,
796 __be16 rmt_port, const struct in6_addr *rmt_addr,
797 int dif, int sdif, unsigned short hnum)
798{
799 const struct inet_sock *inet = inet_sk(sk);
800
801 if (!net_eq(sock_net(sk), net))
802 return false;
803
804 if (udp_sk(sk)->udp_port_hash != hnum ||
805 sk->sk_family != PF_INET6 ||
806 (inet->inet_dport && inet->inet_dport != rmt_port) ||
807 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
808 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
809 !udp_sk_bound_dev_eq(net, READ_ONCE(sk->sk_bound_dev_if), dif, sdif) ||
810 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
811 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
812 return false;
813 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
814 return false;
815 return true;
816}
817
818static void udp6_csum_zero_error(struct sk_buff *skb)
819{
820 /* RFC 2460 section 8.1 says that we SHOULD log
821 * this error. Well, it is reasonable.
822 */
823 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
824 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
825 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
826}
827
828/*
829 * Note: called only from the BH handler context,
830 * so we don't need to lock the hashes.
831 */
832static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
833 const struct in6_addr *saddr, const struct in6_addr *daddr,
834 struct udp_table *udptable, int proto)
835{
836 struct sock *sk, *first = NULL;
837 const struct udphdr *uh = udp_hdr(skb);
838 unsigned short hnum = ntohs(uh->dest);
839 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
840 unsigned int offset = offsetof(typeof(*sk), sk_node);
841 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
842 int dif = inet6_iif(skb);
843 int sdif = inet6_sdif(skb);
844 struct hlist_node *node;
845 struct sk_buff *nskb;
846
847 if (use_hash2) {
848 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) &
849 udptable->mask;
850 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask;
851start_lookup:
852 hslot = &udptable->hash2[hash2];
853 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
854 }
855
856 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
857 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
858 uh->source, saddr, dif, sdif,
859 hnum))
860 continue;
861 /* If zero checksum and no_check is not on for
862 * the socket then skip it.
863 */
864 if (!uh->check && !udp_get_no_check6_rx(sk))
865 continue;
866 if (!first) {
867 first = sk;
868 continue;
869 }
870 nskb = skb_clone(skb, GFP_ATOMIC);
871 if (unlikely(!nskb)) {
872 atomic_inc(&sk->sk_drops);
873 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
874 IS_UDPLITE(sk));
875 __UDP6_INC_STATS(net, UDP_MIB_INERRORS,
876 IS_UDPLITE(sk));
877 continue;
878 }
879
880 if (udpv6_queue_rcv_skb(sk, nskb) > 0)
881 consume_skb(nskb);
882 }
883
884 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
885 if (use_hash2 && hash2 != hash2_any) {
886 hash2 = hash2_any;
887 goto start_lookup;
888 }
889
890 if (first) {
891 if (udpv6_queue_rcv_skb(first, skb) > 0)
892 consume_skb(skb);
893 } else {
894 kfree_skb(skb);
895 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
896 proto == IPPROTO_UDPLITE);
897 }
898 return 0;
899}
900
901static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
902{
903 if (udp_sk_rx_dst_set(sk, dst)) {
904 const struct rt6_info *rt = (const struct rt6_info *)dst;
905
906 sk->sk_rx_dst_cookie = rt6_get_cookie(rt);
907 }
908}
909
910/* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
911 * return code conversion for ip layer consumption
912 */
913static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
914 struct udphdr *uh)
915{
916 int ret;
917
918 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
919 skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo);
920
921 ret = udpv6_queue_rcv_skb(sk, skb);
922
923 /* a return value > 0 means to resubmit the input */
924 if (ret > 0)
925 return ret;
926 return 0;
927}
928
929int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
930 int proto)
931{
932 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
933 const struct in6_addr *saddr, *daddr;
934 struct net *net = dev_net(skb->dev);
935 struct udphdr *uh;
936 struct sock *sk;
937 bool refcounted;
938 u32 ulen = 0;
939
940 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
941 goto discard;
942
943 saddr = &ipv6_hdr(skb)->saddr;
944 daddr = &ipv6_hdr(skb)->daddr;
945 uh = udp_hdr(skb);
946
947 ulen = ntohs(uh->len);
948 if (ulen > skb->len)
949 goto short_packet;
950
951 if (proto == IPPROTO_UDP) {
952 /* UDP validates ulen. */
953
954 /* Check for jumbo payload */
955 if (ulen == 0)
956 ulen = skb->len;
957
958 if (ulen < sizeof(*uh))
959 goto short_packet;
960
961 if (ulen < skb->len) {
962 if (pskb_trim_rcsum(skb, ulen))
963 goto short_packet;
964 saddr = &ipv6_hdr(skb)->saddr;
965 daddr = &ipv6_hdr(skb)->daddr;
966 uh = udp_hdr(skb);
967 }
968 }
969
970 if (udp6_csum_init(skb, uh, proto))
971 goto csum_error;
972
973 /* Check if the socket is already available, e.g. due to early demux */
974 sk = inet6_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest,
975 &refcounted, udp6_ehashfn);
976 if (IS_ERR(sk))
977 goto no_sk;
978
979 if (sk) {
980 struct dst_entry *dst = skb_dst(skb);
981 int ret;
982
983 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
984 udp6_sk_rx_dst_set(sk, dst);
985
986 if (!uh->check && !udp_get_no_check6_rx(sk)) {
987 if (refcounted)
988 sock_put(sk);
989 goto report_csum_error;
990 }
991
992 ret = udp6_unicast_rcv_skb(sk, skb, uh);
993 if (refcounted)
994 sock_put(sk);
995 return ret;
996 }
997
998 /*
999 * Multicast receive code
1000 */
1001 if (ipv6_addr_is_multicast(daddr))
1002 return __udp6_lib_mcast_deliver(net, skb,
1003 saddr, daddr, udptable, proto);
1004
1005 /* Unicast */
1006 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1007 if (sk) {
1008 if (!uh->check && !udp_get_no_check6_rx(sk))
1009 goto report_csum_error;
1010 return udp6_unicast_rcv_skb(sk, skb, uh);
1011 }
1012no_sk:
1013 reason = SKB_DROP_REASON_NO_SOCKET;
1014
1015 if (!uh->check)
1016 goto report_csum_error;
1017
1018 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1019 goto discard;
1020 nf_reset_ct(skb);
1021
1022 if (udp_lib_checksum_complete(skb))
1023 goto csum_error;
1024
1025 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1026 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
1027
1028 kfree_skb_reason(skb, reason);
1029 return 0;
1030
1031short_packet:
1032 if (reason == SKB_DROP_REASON_NOT_SPECIFIED)
1033 reason = SKB_DROP_REASON_PKT_TOO_SMALL;
1034 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
1035 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1036 saddr, ntohs(uh->source),
1037 ulen, skb->len,
1038 daddr, ntohs(uh->dest));
1039 goto discard;
1040
1041report_csum_error:
1042 udp6_csum_zero_error(skb);
1043csum_error:
1044 if (reason == SKB_DROP_REASON_NOT_SPECIFIED)
1045 reason = SKB_DROP_REASON_UDP_CSUM;
1046 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1047discard:
1048 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1049 kfree_skb_reason(skb, reason);
1050 return 0;
1051}
1052
1053
1054static struct sock *__udp6_lib_demux_lookup(struct net *net,
1055 __be16 loc_port, const struct in6_addr *loc_addr,
1056 __be16 rmt_port, const struct in6_addr *rmt_addr,
1057 int dif, int sdif)
1058{
1059 struct udp_table *udptable = net->ipv4.udp_table;
1060 unsigned short hnum = ntohs(loc_port);
1061 unsigned int hash2, slot2;
1062 struct udp_hslot *hslot2;
1063 __portpair ports;
1064 struct sock *sk;
1065
1066 hash2 = ipv6_portaddr_hash(net, loc_addr, hnum);
1067 slot2 = hash2 & udptable->mask;
1068 hslot2 = &udptable->hash2[slot2];
1069 ports = INET_COMBINED_PORTS(rmt_port, hnum);
1070
1071 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
1072 if (sk->sk_state == TCP_ESTABLISHED &&
1073 inet6_match(net, sk, rmt_addr, loc_addr, ports, dif, sdif))
1074 return sk;
1075 /* Only check first socket in chain */
1076 break;
1077 }
1078 return NULL;
1079}
1080
1081void udp_v6_early_demux(struct sk_buff *skb)
1082{
1083 struct net *net = dev_net(skb->dev);
1084 const struct udphdr *uh;
1085 struct sock *sk;
1086 struct dst_entry *dst;
1087 int dif = skb->dev->ifindex;
1088 int sdif = inet6_sdif(skb);
1089
1090 if (!pskb_may_pull(skb, skb_transport_offset(skb) +
1091 sizeof(struct udphdr)))
1092 return;
1093
1094 uh = udp_hdr(skb);
1095
1096 if (skb->pkt_type == PACKET_HOST)
1097 sk = __udp6_lib_demux_lookup(net, uh->dest,
1098 &ipv6_hdr(skb)->daddr,
1099 uh->source, &ipv6_hdr(skb)->saddr,
1100 dif, sdif);
1101 else
1102 return;
1103
1104 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
1105 return;
1106
1107 skb->sk = sk;
1108 skb->destructor = sock_efree;
1109 dst = rcu_dereference(sk->sk_rx_dst);
1110
1111 if (dst)
1112 dst = dst_check(dst, sk->sk_rx_dst_cookie);
1113 if (dst) {
1114 /* set noref for now.
1115 * any place which wants to hold dst has to call
1116 * dst_hold_safe()
1117 */
1118 skb_dst_set_noref(skb, dst);
1119 }
1120}
1121
1122INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb)
1123{
1124 return __udp6_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
1125}
1126
1127/*
1128 * Throw away all pending data and cancel the corking. Socket is locked.
1129 */
1130static void udp_v6_flush_pending_frames(struct sock *sk)
1131{
1132 struct udp_sock *up = udp_sk(sk);
1133
1134 if (up->pending == AF_INET)
1135 udp_flush_pending_frames(sk);
1136 else if (up->pending) {
1137 up->len = 0;
1138 WRITE_ONCE(up->pending, 0);
1139 ip6_flush_pending_frames(sk);
1140 }
1141}
1142
1143static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
1144 int addr_len)
1145{
1146 if (addr_len < offsetofend(struct sockaddr, sa_family))
1147 return -EINVAL;
1148 /* The following checks are replicated from __ip6_datagram_connect()
1149 * and intended to prevent BPF program called below from accessing
1150 * bytes that are out of the bound specified by user in addr_len.
1151 */
1152 if (uaddr->sa_family == AF_INET) {
1153 if (ipv6_only_sock(sk))
1154 return -EAFNOSUPPORT;
1155 return udp_pre_connect(sk, uaddr, addr_len);
1156 }
1157
1158 if (addr_len < SIN6_LEN_RFC2133)
1159 return -EINVAL;
1160
1161 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr, &addr_len);
1162}
1163
1164/**
1165 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
1166 * @sk: socket we are sending on
1167 * @skb: sk_buff containing the filled-in UDP header
1168 * (checksum field must be zeroed out)
1169 * @saddr: source address
1170 * @daddr: destination address
1171 * @len: length of packet
1172 */
1173static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
1174 const struct in6_addr *saddr,
1175 const struct in6_addr *daddr, int len)
1176{
1177 unsigned int offset;
1178 struct udphdr *uh = udp_hdr(skb);
1179 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
1180 __wsum csum = 0;
1181
1182 if (!frags) {
1183 /* Only one fragment on the socket. */
1184 skb->csum_start = skb_transport_header(skb) - skb->head;
1185 skb->csum_offset = offsetof(struct udphdr, check);
1186 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1187 } else {
1188 /*
1189 * HW-checksum won't work as there are two or more
1190 * fragments on the socket so that all csums of sk_buffs
1191 * should be together
1192 */
1193 offset = skb_transport_offset(skb);
1194 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1195 csum = skb->csum;
1196
1197 skb->ip_summed = CHECKSUM_NONE;
1198
1199 do {
1200 csum = csum_add(csum, frags->csum);
1201 } while ((frags = frags->next));
1202
1203 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1204 csum);
1205 if (uh->check == 0)
1206 uh->check = CSUM_MANGLED_0;
1207 }
1208}
1209
1210/*
1211 * Sending
1212 */
1213
1214static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6,
1215 struct inet_cork *cork)
1216{
1217 struct sock *sk = skb->sk;
1218 struct udphdr *uh;
1219 int err = 0;
1220 int is_udplite = IS_UDPLITE(sk);
1221 __wsum csum = 0;
1222 int offset = skb_transport_offset(skb);
1223 int len = skb->len - offset;
1224 int datalen = len - sizeof(*uh);
1225
1226 /*
1227 * Create a UDP header
1228 */
1229 uh = udp_hdr(skb);
1230 uh->source = fl6->fl6_sport;
1231 uh->dest = fl6->fl6_dport;
1232 uh->len = htons(len);
1233 uh->check = 0;
1234
1235 if (cork->gso_size) {
1236 const int hlen = skb_network_header_len(skb) +
1237 sizeof(struct udphdr);
1238
1239 if (hlen + cork->gso_size > cork->fragsize) {
1240 kfree_skb(skb);
1241 return -EINVAL;
1242 }
1243 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
1244 kfree_skb(skb);
1245 return -EINVAL;
1246 }
1247 if (udp_get_no_check6_tx(sk)) {
1248 kfree_skb(skb);
1249 return -EINVAL;
1250 }
1251 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
1252 dst_xfrm(skb_dst(skb))) {
1253 kfree_skb(skb);
1254 return -EIO;
1255 }
1256
1257 if (datalen > cork->gso_size) {
1258 skb_shinfo(skb)->gso_size = cork->gso_size;
1259 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
1260 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
1261 cork->gso_size);
1262 }
1263 goto csum_partial;
1264 }
1265
1266 if (is_udplite)
1267 csum = udplite_csum(skb);
1268 else if (udp_get_no_check6_tx(sk)) { /* UDP csum disabled */
1269 skb->ip_summed = CHECKSUM_NONE;
1270 goto send;
1271 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1272csum_partial:
1273 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1274 goto send;
1275 } else
1276 csum = udp_csum(skb);
1277
1278 /* add protocol-dependent pseudo-header */
1279 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1280 len, fl6->flowi6_proto, csum);
1281 if (uh->check == 0)
1282 uh->check = CSUM_MANGLED_0;
1283
1284send:
1285 err = ip6_send_skb(skb);
1286 if (err) {
1287 if (err == -ENOBUFS && !inet6_test_bit(RECVERR6, sk)) {
1288 UDP6_INC_STATS(sock_net(sk),
1289 UDP_MIB_SNDBUFERRORS, is_udplite);
1290 err = 0;
1291 }
1292 } else {
1293 UDP6_INC_STATS(sock_net(sk),
1294 UDP_MIB_OUTDATAGRAMS, is_udplite);
1295 }
1296 return err;
1297}
1298
1299static int udp_v6_push_pending_frames(struct sock *sk)
1300{
1301 struct sk_buff *skb;
1302 struct udp_sock *up = udp_sk(sk);
1303 int err = 0;
1304
1305 if (up->pending == AF_INET)
1306 return udp_push_pending_frames(sk);
1307
1308 skb = ip6_finish_skb(sk);
1309 if (!skb)
1310 goto out;
1311
1312 err = udp_v6_send_skb(skb, &inet_sk(sk)->cork.fl.u.ip6,
1313 &inet_sk(sk)->cork.base);
1314out:
1315 up->len = 0;
1316 WRITE_ONCE(up->pending, 0);
1317 return err;
1318}
1319
1320int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1321{
1322 struct ipv6_txoptions opt_space;
1323 struct udp_sock *up = udp_sk(sk);
1324 struct inet_sock *inet = inet_sk(sk);
1325 struct ipv6_pinfo *np = inet6_sk(sk);
1326 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1327 struct in6_addr *daddr, *final_p, final;
1328 struct ipv6_txoptions *opt = NULL;
1329 struct ipv6_txoptions *opt_to_free = NULL;
1330 struct ip6_flowlabel *flowlabel = NULL;
1331 struct inet_cork_full cork;
1332 struct flowi6 *fl6 = &cork.fl.u.ip6;
1333 struct dst_entry *dst;
1334 struct ipcm6_cookie ipc6;
1335 int addr_len = msg->msg_namelen;
1336 bool connected = false;
1337 int ulen = len;
1338 int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE;
1339 int err;
1340 int is_udplite = IS_UDPLITE(sk);
1341 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1342
1343 ipcm6_init(&ipc6);
1344 ipc6.gso_size = READ_ONCE(up->gso_size);
1345 ipc6.sockc.tsflags = READ_ONCE(sk->sk_tsflags);
1346 ipc6.sockc.mark = READ_ONCE(sk->sk_mark);
1347
1348 /* destination address check */
1349 if (sin6) {
1350 if (addr_len < offsetof(struct sockaddr, sa_data))
1351 return -EINVAL;
1352
1353 switch (sin6->sin6_family) {
1354 case AF_INET6:
1355 if (addr_len < SIN6_LEN_RFC2133)
1356 return -EINVAL;
1357 daddr = &sin6->sin6_addr;
1358 if (ipv6_addr_any(daddr) &&
1359 ipv6_addr_v4mapped(&np->saddr))
1360 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1361 daddr);
1362 break;
1363 case AF_INET:
1364 goto do_udp_sendmsg;
1365 case AF_UNSPEC:
1366 msg->msg_name = sin6 = NULL;
1367 msg->msg_namelen = addr_len = 0;
1368 daddr = NULL;
1369 break;
1370 default:
1371 return -EINVAL;
1372 }
1373 } else if (!READ_ONCE(up->pending)) {
1374 if (sk->sk_state != TCP_ESTABLISHED)
1375 return -EDESTADDRREQ;
1376 daddr = &sk->sk_v6_daddr;
1377 } else
1378 daddr = NULL;
1379
1380 if (daddr) {
1381 if (ipv6_addr_v4mapped(daddr)) {
1382 struct sockaddr_in sin;
1383 sin.sin_family = AF_INET;
1384 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1385 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1386 msg->msg_name = &sin;
1387 msg->msg_namelen = sizeof(sin);
1388do_udp_sendmsg:
1389 err = ipv6_only_sock(sk) ?
1390 -ENETUNREACH : udp_sendmsg(sk, msg, len);
1391 msg->msg_name = sin6;
1392 msg->msg_namelen = addr_len;
1393 return err;
1394 }
1395 }
1396
1397 /* Rough check on arithmetic overflow,
1398 better check is made in ip6_append_data().
1399 */
1400 if (len > INT_MAX - sizeof(struct udphdr))
1401 return -EMSGSIZE;
1402
1403 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1404 if (READ_ONCE(up->pending)) {
1405 if (READ_ONCE(up->pending) == AF_INET)
1406 return udp_sendmsg(sk, msg, len);
1407 /*
1408 * There are pending frames.
1409 * The socket lock must be held while it's corked.
1410 */
1411 lock_sock(sk);
1412 if (likely(up->pending)) {
1413 if (unlikely(up->pending != AF_INET6)) {
1414 release_sock(sk);
1415 return -EAFNOSUPPORT;
1416 }
1417 dst = NULL;
1418 goto do_append_data;
1419 }
1420 release_sock(sk);
1421 }
1422 ulen += sizeof(struct udphdr);
1423
1424 memset(fl6, 0, sizeof(*fl6));
1425
1426 if (sin6) {
1427 if (sin6->sin6_port == 0)
1428 return -EINVAL;
1429
1430 fl6->fl6_dport = sin6->sin6_port;
1431 daddr = &sin6->sin6_addr;
1432
1433 if (inet6_test_bit(SNDFLOW, sk)) {
1434 fl6->flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1435 if (fl6->flowlabel & IPV6_FLOWLABEL_MASK) {
1436 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel);
1437 if (IS_ERR(flowlabel))
1438 return -EINVAL;
1439 }
1440 }
1441
1442 /*
1443 * Otherwise it will be difficult to maintain
1444 * sk->sk_dst_cache.
1445 */
1446 if (sk->sk_state == TCP_ESTABLISHED &&
1447 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1448 daddr = &sk->sk_v6_daddr;
1449
1450 if (addr_len >= sizeof(struct sockaddr_in6) &&
1451 sin6->sin6_scope_id &&
1452 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1453 fl6->flowi6_oif = sin6->sin6_scope_id;
1454 } else {
1455 if (sk->sk_state != TCP_ESTABLISHED)
1456 return -EDESTADDRREQ;
1457
1458 fl6->fl6_dport = inet->inet_dport;
1459 daddr = &sk->sk_v6_daddr;
1460 fl6->flowlabel = np->flow_label;
1461 connected = true;
1462 }
1463
1464 if (!fl6->flowi6_oif)
1465 fl6->flowi6_oif = READ_ONCE(sk->sk_bound_dev_if);
1466
1467 if (!fl6->flowi6_oif)
1468 fl6->flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1469
1470 fl6->flowi6_uid = sk->sk_uid;
1471
1472 if (msg->msg_controllen) {
1473 opt = &opt_space;
1474 memset(opt, 0, sizeof(struct ipv6_txoptions));
1475 opt->tot_len = sizeof(*opt);
1476 ipc6.opt = opt;
1477
1478 err = udp_cmsg_send(sk, msg, &ipc6.gso_size);
1479 if (err > 0)
1480 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, fl6,
1481 &ipc6);
1482 if (err < 0) {
1483 fl6_sock_release(flowlabel);
1484 return err;
1485 }
1486 if ((fl6->flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1487 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel);
1488 if (IS_ERR(flowlabel))
1489 return -EINVAL;
1490 }
1491 if (!(opt->opt_nflen|opt->opt_flen))
1492 opt = NULL;
1493 connected = false;
1494 }
1495 if (!opt) {
1496 opt = txopt_get(np);
1497 opt_to_free = opt;
1498 }
1499 if (flowlabel)
1500 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1501 opt = ipv6_fixup_options(&opt_space, opt);
1502 ipc6.opt = opt;
1503
1504 fl6->flowi6_proto = sk->sk_protocol;
1505 fl6->flowi6_mark = ipc6.sockc.mark;
1506 fl6->daddr = *daddr;
1507 if (ipv6_addr_any(&fl6->saddr) && !ipv6_addr_any(&np->saddr))
1508 fl6->saddr = np->saddr;
1509 fl6->fl6_sport = inet->inet_sport;
1510
1511 if (cgroup_bpf_enabled(CGROUP_UDP6_SENDMSG) && !connected) {
1512 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk,
1513 (struct sockaddr *)sin6,
1514 &addr_len,
1515 &fl6->saddr);
1516 if (err)
1517 goto out_no_dst;
1518 if (sin6) {
1519 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
1520 /* BPF program rewrote IPv6-only by IPv4-mapped
1521 * IPv6. It's currently unsupported.
1522 */
1523 err = -ENOTSUPP;
1524 goto out_no_dst;
1525 }
1526 if (sin6->sin6_port == 0) {
1527 /* BPF program set invalid port. Reject it. */
1528 err = -EINVAL;
1529 goto out_no_dst;
1530 }
1531 fl6->fl6_dport = sin6->sin6_port;
1532 fl6->daddr = sin6->sin6_addr;
1533 }
1534 }
1535
1536 if (ipv6_addr_any(&fl6->daddr))
1537 fl6->daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1538
1539 final_p = fl6_update_dst(fl6, opt, &final);
1540 if (final_p)
1541 connected = false;
1542
1543 if (!fl6->flowi6_oif && ipv6_addr_is_multicast(&fl6->daddr)) {
1544 fl6->flowi6_oif = READ_ONCE(np->mcast_oif);
1545 connected = false;
1546 } else if (!fl6->flowi6_oif)
1547 fl6->flowi6_oif = READ_ONCE(np->ucast_oif);
1548
1549 security_sk_classify_flow(sk, flowi6_to_flowi_common(fl6));
1550
1551 if (ipc6.tclass < 0)
1552 ipc6.tclass = np->tclass;
1553
1554 fl6->flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6->flowlabel);
1555
1556 dst = ip6_sk_dst_lookup_flow(sk, fl6, final_p, connected);
1557 if (IS_ERR(dst)) {
1558 err = PTR_ERR(dst);
1559 dst = NULL;
1560 goto out;
1561 }
1562
1563 if (ipc6.hlimit < 0)
1564 ipc6.hlimit = ip6_sk_dst_hoplimit(np, fl6, dst);
1565
1566 if (msg->msg_flags&MSG_CONFIRM)
1567 goto do_confirm;
1568back_from_confirm:
1569
1570 /* Lockless fast path for the non-corking case */
1571 if (!corkreq) {
1572 struct sk_buff *skb;
1573
1574 skb = ip6_make_skb(sk, getfrag, msg, ulen,
1575 sizeof(struct udphdr), &ipc6,
1576 (struct rt6_info *)dst,
1577 msg->msg_flags, &cork);
1578 err = PTR_ERR(skb);
1579 if (!IS_ERR_OR_NULL(skb))
1580 err = udp_v6_send_skb(skb, fl6, &cork.base);
1581 /* ip6_make_skb steals dst reference */
1582 goto out_no_dst;
1583 }
1584
1585 lock_sock(sk);
1586 if (unlikely(up->pending)) {
1587 /* The socket is already corked while preparing it. */
1588 /* ... which is an evident application bug. --ANK */
1589 release_sock(sk);
1590
1591 net_dbg_ratelimited("udp cork app bug 2\n");
1592 err = -EINVAL;
1593 goto out;
1594 }
1595
1596 WRITE_ONCE(up->pending, AF_INET6);
1597
1598do_append_data:
1599 if (ipc6.dontfrag < 0)
1600 ipc6.dontfrag = inet6_test_bit(DONTFRAG, sk);
1601 up->len += ulen;
1602 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr),
1603 &ipc6, fl6, (struct rt6_info *)dst,
1604 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1605 if (err)
1606 udp_v6_flush_pending_frames(sk);
1607 else if (!corkreq)
1608 err = udp_v6_push_pending_frames(sk);
1609 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1610 WRITE_ONCE(up->pending, 0);
1611
1612 if (err > 0)
1613 err = inet6_test_bit(RECVERR6, sk) ? net_xmit_errno(err) : 0;
1614 release_sock(sk);
1615
1616out:
1617 dst_release(dst);
1618out_no_dst:
1619 fl6_sock_release(flowlabel);
1620 txopt_put(opt_to_free);
1621 if (!err)
1622 return len;
1623 /*
1624 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1625 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1626 * we don't have a good statistic (IpOutDiscards but it can be too many
1627 * things). We could add another new stat but at least for now that
1628 * seems like overkill.
1629 */
1630 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1631 UDP6_INC_STATS(sock_net(sk),
1632 UDP_MIB_SNDBUFERRORS, is_udplite);
1633 }
1634 return err;
1635
1636do_confirm:
1637 if (msg->msg_flags & MSG_PROBE)
1638 dst_confirm_neigh(dst, &fl6->daddr);
1639 if (!(msg->msg_flags&MSG_PROBE) || len)
1640 goto back_from_confirm;
1641 err = 0;
1642 goto out;
1643}
1644EXPORT_SYMBOL(udpv6_sendmsg);
1645
1646static void udpv6_splice_eof(struct socket *sock)
1647{
1648 struct sock *sk = sock->sk;
1649 struct udp_sock *up = udp_sk(sk);
1650
1651 if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk))
1652 return;
1653
1654 lock_sock(sk);
1655 if (up->pending && !udp_test_bit(CORK, sk))
1656 udp_v6_push_pending_frames(sk);
1657 release_sock(sk);
1658}
1659
1660void udpv6_destroy_sock(struct sock *sk)
1661{
1662 struct udp_sock *up = udp_sk(sk);
1663 lock_sock(sk);
1664
1665 /* protects from races with udp_abort() */
1666 sock_set_flag(sk, SOCK_DEAD);
1667 udp_v6_flush_pending_frames(sk);
1668 release_sock(sk);
1669
1670 if (static_branch_unlikely(&udpv6_encap_needed_key)) {
1671 if (up->encap_type) {
1672 void (*encap_destroy)(struct sock *sk);
1673 encap_destroy = READ_ONCE(up->encap_destroy);
1674 if (encap_destroy)
1675 encap_destroy(sk);
1676 }
1677 if (udp_test_bit(ENCAP_ENABLED, sk)) {
1678 static_branch_dec(&udpv6_encap_needed_key);
1679 udp_encap_disable();
1680 }
1681 }
1682}
1683
1684/*
1685 * Socket option code for UDP
1686 */
1687int udpv6_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
1688 unsigned int optlen)
1689{
1690 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET)
1691 return udp_lib_setsockopt(sk, level, optname,
1692 optval, optlen,
1693 udp_v6_push_pending_frames);
1694 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1695}
1696
1697int udpv6_getsockopt(struct sock *sk, int level, int optname,
1698 char __user *optval, int __user *optlen)
1699{
1700 if (level == SOL_UDP || level == SOL_UDPLITE)
1701 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1702 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1703}
1704
1705static const struct inet6_protocol udpv6_protocol = {
1706 .handler = udpv6_rcv,
1707 .err_handler = udpv6_err,
1708 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1709};
1710
1711/* ------------------------------------------------------------------------ */
1712#ifdef CONFIG_PROC_FS
1713int udp6_seq_show(struct seq_file *seq, void *v)
1714{
1715 if (v == SEQ_START_TOKEN) {
1716 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1717 } else {
1718 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1719 const struct inet_sock *inet = inet_sk((const struct sock *)v);
1720 __u16 srcp = ntohs(inet->inet_sport);
1721 __u16 destp = ntohs(inet->inet_dport);
1722 __ip6_dgram_sock_seq_show(seq, v, srcp, destp,
1723 udp_rqueue_get(v), bucket);
1724 }
1725 return 0;
1726}
1727
1728const struct seq_operations udp6_seq_ops = {
1729 .start = udp_seq_start,
1730 .next = udp_seq_next,
1731 .stop = udp_seq_stop,
1732 .show = udp6_seq_show,
1733};
1734EXPORT_SYMBOL(udp6_seq_ops);
1735
1736static struct udp_seq_afinfo udp6_seq_afinfo = {
1737 .family = AF_INET6,
1738 .udp_table = NULL,
1739};
1740
1741int __net_init udp6_proc_init(struct net *net)
1742{
1743 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops,
1744 sizeof(struct udp_iter_state), &udp6_seq_afinfo))
1745 return -ENOMEM;
1746 return 0;
1747}
1748
1749void udp6_proc_exit(struct net *net)
1750{
1751 remove_proc_entry("udp6", net->proc_net);
1752}
1753#endif /* CONFIG_PROC_FS */
1754
1755/* ------------------------------------------------------------------------ */
1756
1757struct proto udpv6_prot = {
1758 .name = "UDPv6",
1759 .owner = THIS_MODULE,
1760 .close = udp_lib_close,
1761 .pre_connect = udpv6_pre_connect,
1762 .connect = ip6_datagram_connect,
1763 .disconnect = udp_disconnect,
1764 .ioctl = udp_ioctl,
1765 .init = udpv6_init_sock,
1766 .destroy = udpv6_destroy_sock,
1767 .setsockopt = udpv6_setsockopt,
1768 .getsockopt = udpv6_getsockopt,
1769 .sendmsg = udpv6_sendmsg,
1770 .recvmsg = udpv6_recvmsg,
1771 .splice_eof = udpv6_splice_eof,
1772 .release_cb = ip6_datagram_release_cb,
1773 .hash = udp_lib_hash,
1774 .unhash = udp_lib_unhash,
1775 .rehash = udp_v6_rehash,
1776 .get_port = udp_v6_get_port,
1777 .put_port = udp_lib_unhash,
1778#ifdef CONFIG_BPF_SYSCALL
1779 .psock_update_sk_prot = udp_bpf_update_proto,
1780#endif
1781
1782 .memory_allocated = &udp_memory_allocated,
1783 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
1784
1785 .sysctl_mem = sysctl_udp_mem,
1786 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
1787 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
1788 .obj_size = sizeof(struct udp6_sock),
1789 .ipv6_pinfo_offset = offsetof(struct udp6_sock, inet6),
1790 .h.udp_table = NULL,
1791 .diag_destroy = udp_abort,
1792};
1793
1794static struct inet_protosw udpv6_protosw = {
1795 .type = SOCK_DGRAM,
1796 .protocol = IPPROTO_UDP,
1797 .prot = &udpv6_prot,
1798 .ops = &inet6_dgram_ops,
1799 .flags = INET_PROTOSW_PERMANENT,
1800};
1801
1802int __init udpv6_init(void)
1803{
1804 int ret;
1805
1806 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1807 if (ret)
1808 goto out;
1809
1810 ret = inet6_register_protosw(&udpv6_protosw);
1811 if (ret)
1812 goto out_udpv6_protocol;
1813out:
1814 return ret;
1815
1816out_udpv6_protocol:
1817 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1818 goto out;
1819}
1820
1821void udpv6_exit(void)
1822{
1823 inet6_unregister_protosw(&udpv6_protosw);
1824 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1825}
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * UDP over IPv6
4 * Linux INET6 implementation
5 *
6 * Authors:
7 * Pedro Roque <roque@di.fc.ul.pt>
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
38#include <net/addrconf.h>
39#include <net/ndisc.h>
40#include <net/protocol.h>
41#include <net/transp_v6.h>
42#include <net/ip6_route.h>
43#include <net/raw.h>
44#include <net/seg6.h>
45#include <net/tcp_states.h>
46#include <net/ip6_checksum.h>
47#include <net/ip6_tunnel.h>
48#include <trace/events/udp.h>
49#include <net/xfrm.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>
54#include <net/gro.h>
55
56#include <linux/proc_fs.h>
57#include <linux/seq_file.h>
58#include <trace/events/skb.h>
59#include "udp_impl.h"
60
61static void udpv6_destruct_sock(struct sock *sk)
62{
63 udp_destruct_common(sk);
64 inet6_sock_destruct(sk);
65}
66
67int udpv6_init_sock(struct sock *sk)
68{
69 udp_lib_init_sock(sk);
70 sk->sk_destruct = udpv6_destruct_sock;
71 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
72 return 0;
73}
74
75INDIRECT_CALLABLE_SCOPE
76u32 udp6_ehashfn(const struct net *net,
77 const struct in6_addr *laddr,
78 const u16 lport,
79 const struct in6_addr *faddr,
80 const __be16 fport)
81{
82 u32 lhash, fhash;
83
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));
88
89 lhash = (__force u32)laddr->s6_addr32[3];
90 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
91
92 return __inet6_ehashfn(lhash, lport, fhash, fport,
93 udp6_ehash_secret + net_hash_mix(net));
94}
95
96int udp_v6_get_port(struct sock *sk, unsigned short snum)
97{
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);
102
103 /* precompute partial secondary hash */
104 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
105 return udp_lib_get_port(sk, snum, hash2_nulladdr);
106}
107
108void udp_v6_rehash(struct sock *sk)
109{
110 u16 new_hash = ipv6_portaddr_hash(sock_net(sk),
111 &sk->sk_v6_rcv_saddr,
112 inet_sk(sk)->inet_num);
113
114 udp_lib_rehash(sk, new_hash);
115}
116
117static 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,
120 int dif, int sdif)
121{
122 int bound_dev_if, score;
123 struct inet_sock *inet;
124 bool dev_match;
125
126 if (!net_eq(sock_net(sk), net) ||
127 udp_sk(sk)->udp_port_hash != hnum ||
128 sk->sk_family != PF_INET6)
129 return -1;
130
131 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
132 return -1;
133
134 score = 0;
135 inet = inet_sk(sk);
136
137 if (inet->inet_dport) {
138 if (inet->inet_dport != sport)
139 return -1;
140 score++;
141 }
142
143 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
144 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
145 return -1;
146 score++;
147 }
148
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);
151 if (!dev_match)
152 return -1;
153 if (bound_dev_if)
154 score++;
155
156 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
157 score++;
158
159 return score;
160}
161
162/* called with rcu_read_lock() */
163static 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,
167 struct sk_buff *skb)
168{
169 struct sock *sk, *result;
170 int score, badness;
171 bool need_rescore;
172
173 result = NULL;
174 badness = -1;
175 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
176 need_rescore = false;
177rescore:
178 score = compute_score(need_rescore ? result : sk, net, saddr,
179 sport, daddr, hnum, dif, sdif);
180 if (score > badness) {
181 badness = score;
182
183 if (need_rescore)
184 continue;
185
186 if (sk->sk_state == TCP_ESTABLISHED) {
187 result = sk;
188 continue;
189 }
190
191 result = inet6_lookup_reuseport(net, sk, skb, sizeof(struct udphdr),
192 saddr, sport, daddr, hnum, udp6_ehashfn);
193 if (!result) {
194 result = sk;
195 continue;
196 }
197
198 /* Fall back to scoring if group has connections */
199 if (!reuseport_has_conns(sk))
200 return result;
201
202 /* Reuseport logic returned an error, keep original score. */
203 if (IS_ERR(result))
204 continue;
205
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'.
211 */
212 need_rescore = true;
213 goto rescore;
214 }
215 }
216 return result;
217}
218
219/* rcu_read_lock() must be held */
220struct 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,
224 struct sk_buff *skb)
225{
226 unsigned short hnum = ntohs(dport);
227 unsigned int hash2, slot2;
228 struct udp_hslot *hslot2;
229 struct sock *result, *sk;
230
231 hash2 = ipv6_portaddr_hash(net, daddr, hnum);
232 slot2 = hash2 & udptable->mask;
233 hslot2 = &udptable->hash2[slot2];
234
235 /* Lookup connected or non-wildcard sockets */
236 result = udp6_lib_lookup2(net, saddr, sport,
237 daddr, hnum, dif, sdif,
238 hslot2, skb);
239 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
240 goto done;
241
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,
247 udp6_ehashfn);
248 if (sk) {
249 result = sk;
250 goto done;
251 }
252 }
253
254 /* Got non-wildcard socket or error on first lookup */
255 if (result)
256 goto done;
257
258 /* Lookup wildcard sockets */
259 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
260 slot2 = hash2 & udptable->mask;
261 hslot2 = &udptable->hash2[slot2];
262
263 result = udp6_lib_lookup2(net, saddr, sport,
264 &in6addr_any, hnum, dif, sdif,
265 hslot2, skb);
266done:
267 if (IS_ERR(result))
268 return NULL;
269 return result;
270}
271EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
272
273static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
274 __be16 sport, __be16 dport,
275 struct udp_table *udptable)
276{
277 const struct ipv6hdr *iph = ipv6_hdr(skb);
278
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);
282}
283
284struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb,
285 __be16 sport, __be16 dport)
286{
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);
290 int iif, sdif;
291
292 inet6_get_iif_sdif(skb, &iif, &sdif);
293
294 return __udp6_lib_lookup(net, &iph->saddr, sport,
295 &iph->daddr, dport, iif,
296 sdif, net->ipv4.udp_table, NULL);
297}
298
299/* Must be called under rcu_read_lock().
300 * Does increment socket refcount.
301 */
302#if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
303struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
304 const struct in6_addr *daddr, __be16 dport, int dif)
305{
306 struct sock *sk;
307
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))
311 sk = NULL;
312 return sk;
313}
314EXPORT_SYMBOL_GPL(udp6_lib_lookup);
315#endif
316
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
320 */
321static int udp6_skb_len(struct sk_buff *skb)
322{
323 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb);
324}
325
326/*
327 * This should be easy, if there is something there we
328 * return it, otherwise we block.
329 */
330
331int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
332 int flags, int *addr_len)
333{
334 struct ipv6_pinfo *np = inet6_sk(sk);
335 struct inet_sock *inet = inet_sk(sk);
336 struct sk_buff *skb;
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;
342 int is_udp4;
343
344 if (flags & MSG_ERRQUEUE)
345 return ipv6_recv_error(sk, msg, len, addr_len);
346
347 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
348 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
349
350try_again:
351 off = sk_peek_offset(sk, flags);
352 skb = __skb_recv_udp(sk, flags, &off, &err);
353 if (!skb)
354 return err;
355
356 ulen = udp6_skb_len(skb);
357 copied = len;
358 if (copied > ulen - off)
359 copied = ulen - off;
360 else if (copied < ulen)
361 msg->msg_flags |= MSG_TRUNC;
362
363 is_udp4 = (skb->protocol == htons(ETH_P_IP));
364 mib = __UDPX_MIB(sk, is_udp4);
365
366 /*
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.
370 */
371
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);
376 if (!checksum_valid)
377 goto csum_copy_err;
378 }
379
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);
383 else
384 err = skb_copy_datagram_msg(skb, off, msg, copied);
385 } else {
386 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
387 if (err == -EINVAL)
388 goto csum_copy_err;
389 }
390 if (unlikely(err)) {
391 if (!peeking) {
392 atomic_inc(&sk->sk_drops);
393 SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
394 }
395 kfree_skb(skb);
396 return err;
397 }
398 if (!peeking)
399 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS);
400
401 sock_recv_cmsgs(msg, sk, skb);
402
403 /* Copy the address. */
404 if (msg->msg_name) {
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;
409
410 if (is_udp4) {
411 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
412 &sin6->sin6_addr);
413 sin6->sin6_scope_id = 0;
414 } else {
415 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
416 sin6->sin6_scope_id =
417 ipv6_iface_scope_id(&sin6->sin6_addr,
418 inet6_iif(skb));
419 }
420 *addr_len = sizeof(*sin6);
421
422 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk,
423 (struct sockaddr *)sin6,
424 addr_len);
425 }
426
427 if (udp_test_bit(GRO_ENABLED, sk))
428 udp_cmsg_recv(msg, sk, skb);
429
430 if (np->rxopt.all)
431 ip6_datagram_recv_common_ctl(sk, msg, skb);
432
433 if (is_udp4) {
434 if (inet_cmsg_flags(inet))
435 ip_cmsg_recv_offset(msg, sk, skb,
436 sizeof(struct udphdr), off);
437 } else {
438 if (np->rxopt.all)
439 ip6_datagram_recv_specific_ctl(sk, msg, skb);
440 }
441
442 err = copied;
443 if (flags & MSG_TRUNC)
444 err = ulen;
445
446 skb_consume_udp(sk, skb, peeking ? -err : err);
447 return err;
448
449csum_copy_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);
454 }
455 kfree_skb(skb);
456
457 /* starting over for a new packet, but check if we need to yield */
458 cond_resched();
459 msg->msg_flags &= ~MSG_TRUNC;
460 goto try_again;
461}
462
463DECLARE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
464void udpv6_encap_enable(void)
465{
466 static_branch_inc(&udpv6_encap_needed_key);
467}
468EXPORT_SYMBOL(udpv6_encap_enable);
469
470/* Handler for tunnels with arbitrary destination ports: no socket lookup, go
471 * through error handlers in encapsulations looking for a match.
472 */
473static 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)
476{
477 int i;
478
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;
483
484 encap = rcu_dereference(ip6tun_encaps[i]);
485 if (!encap)
486 continue;
487 handler = encap->err_handler;
488 if (handler && !handler(skb, opt, type, code, offset, info))
489 return 0;
490 }
491
492 return -ENOENT;
493}
494
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.
501 *
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.
505 *
506 * Then ask the tunnel implementation to match the error against a valid
507 * association.
508 *
509 * Return an error if we can't find a match, the socket if we need further
510 * processing, zero otherwise.
511 */
512static struct sock *__udp6_lib_err_encap(struct net *net,
513 const struct ipv6hdr *hdr, int offset,
514 struct udphdr *uh,
515 struct udp_table *udptable,
516 struct sock *sk,
517 struct sk_buff *skb,
518 struct inet6_skb_parm *opt,
519 u8 type, u8 code, __be32 info)
520{
521 int (*lookup)(struct sock *sk, struct sk_buff *skb);
522 int network_offset, transport_offset;
523 struct udp_sock *up;
524
525 network_offset = skb_network_offset(skb);
526 transport_offset = skb_transport_offset(skb);
527
528 /* Network header needs to point to the outer IPv6 header inside ICMP */
529 skb_reset_network_header(skb);
530
531 /* Transport header needs to point to the UDP header */
532 skb_set_transport_header(skb, offset);
533
534 if (sk) {
535 up = udp_sk(sk);
536
537 lookup = READ_ONCE(up->encap_err_lookup);
538 if (lookup && lookup(sk, skb))
539 sk = NULL;
540
541 goto out;
542 }
543
544 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source,
545 &hdr->saddr, uh->dest,
546 inet6_iif(skb), 0, udptable, skb);
547 if (sk) {
548 up = udp_sk(sk);
549
550 lookup = READ_ONCE(up->encap_err_lookup);
551 if (!lookup || lookup(sk, skb))
552 sk = NULL;
553 }
554
555out:
556 if (!sk) {
557 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code,
558 offset, info));
559 }
560
561 skb_set_transport_header(skb, transport_offset);
562 skb_set_network_header(skb, network_offset);
563
564 return sk;
565}
566
567int __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)
570{
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);
576 bool tunnel = false;
577 struct sock *sk;
578 int harderr;
579 int err;
580 struct net *net = dev_net(skb->dev);
581
582 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
583 inet6_iif(skb), inet6_sdif(skb), udptable, NULL);
584
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,
589 udptable, sk, skb,
590 opt, type, code, info);
591 if (!sk)
592 return 0;
593 } else
594 sk = ERR_PTR(-ENOENT);
595
596 if (IS_ERR(sk)) {
597 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
598 ICMP6_MIB_INERRORS);
599 return PTR_ERR(sk);
600 }
601
602 tunnel = true;
603 }
604
605 harderr = icmpv6_err_convert(type, code, &err);
606 np = inet6_sk(sk);
607
608 if (type == ICMPV6_PKT_TOOBIG) {
609 if (!ip6_sk_accept_pmtu(sk))
610 goto out;
611 ip6_sk_update_pmtu(skb, sk, info);
612 if (READ_ONCE(np->pmtudisc) != IPV6_PMTUDISC_DONT)
613 harderr = 1;
614 }
615 if (type == NDISC_REDIRECT) {
616 if (tunnel) {
617 ip6_redirect(skb, sock_net(sk), inet6_iif(skb),
618 READ_ONCE(sk->sk_mark), sk->sk_uid);
619 } else {
620 ip6_sk_redirect(skb, sk);
621 }
622 goto out;
623 }
624
625 /* Tunnels don't have an application socket: don't pass errors back */
626 if (tunnel) {
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));
630 goto out;
631 }
632
633 if (!inet6_test_bit(RECVERR6, sk)) {
634 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
635 goto out;
636 } else {
637 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
638 }
639
640 sk->sk_err = err;
641 sk_error_report(sk);
642out:
643 return 0;
644}
645
646static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
647{
648 int rc;
649
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);
654 } else {
655 sk_mark_napi_id_once(sk, skb);
656 }
657
658 rc = __udp_enqueue_schedule_skb(sk, skb);
659 if (rc < 0) {
660 int is_udplite = IS_UDPLITE(sk);
661 enum skb_drop_reason drop_reason;
662
663 /* Note that an ENOMEM error is charged twice */
664 if (rc == -ENOMEM) {
665 UDP6_INC_STATS(sock_net(sk),
666 UDP_MIB_RCVBUFERRORS, is_udplite);
667 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
668 } else {
669 UDP6_INC_STATS(sock_net(sk),
670 UDP_MIB_MEMERRORS, is_udplite);
671 drop_reason = SKB_DROP_REASON_PROTO_MEM;
672 }
673 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
674 kfree_skb_reason(skb, drop_reason);
675 trace_udp_fail_queue_rcv_skb(rc, sk);
676 return -1;
677 }
678
679 return 0;
680}
681
682static __inline__ int udpv6_err(struct sk_buff *skb,
683 struct inet6_skb_parm *opt, u8 type,
684 u8 code, int offset, __be32 info)
685{
686 return __udp6_lib_err(skb, opt, type, code, offset, info,
687 dev_net(skb->dev)->ipv4.udp_table);
688}
689
690static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
691{
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);
695
696 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) {
697 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
698 goto drop;
699 }
700 nf_reset_ct(skb);
701
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);
705
706 /*
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
715 */
716
717 /* if we're overly short, let UDP handle it */
718 encap_rcv = READ_ONCE(up->encap_rcv);
719 if (encap_rcv) {
720 int ret;
721
722 /* Verify checksum before giving to encap */
723 if (udp_lib_checksum_complete(skb))
724 goto csum_error;
725
726 ret = encap_rcv(sk, skb);
727 if (ret <= 0) {
728 __UDP6_INC_STATS(sock_net(sk),
729 UDP_MIB_INDATAGRAMS,
730 is_udplite);
731 return -ret;
732 }
733 }
734
735 /* FALLTHROUGH -- it's a UDP Packet */
736 }
737
738 /*
739 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
740 */
741 if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) {
742 u16 pcrlen = READ_ONCE(up->pcrlen);
743
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);
747 goto drop;
748 }
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);
752 goto drop;
753 }
754 }
755
756 prefetch(&sk->sk_rmem_alloc);
757 if (rcu_access_pointer(sk->sk_filter) &&
758 udp_lib_checksum_complete(skb))
759 goto csum_error;
760
761 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
762 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
763 goto drop;
764 }
765
766 udp_csum_pull_header(skb);
767
768 skb_dst_drop(skb);
769
770 return __udpv6_queue_rcv_skb(sk, skb);
771
772csum_error:
773 drop_reason = SKB_DROP_REASON_UDP_CSUM;
774 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
775drop:
776 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
777 atomic_inc(&sk->sk_drops);
778 kfree_skb_reason(skb, drop_reason);
779 return -1;
780}
781
782static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
783{
784 struct sk_buff *next, *segs;
785 int ret;
786
787 if (likely(!udp_unexpected_gso(sk, skb)))
788 return udpv6_queue_rcv_one_skb(sk, skb);
789
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));
794
795 udp_post_segment_fix_csum(skb);
796 ret = udpv6_queue_rcv_one_skb(sk, skb);
797 if (ret > 0)
798 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret,
799 true);
800 }
801 return 0;
802}
803
804static 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)
808{
809 const struct inet_sock *inet = inet_sk(sk);
810
811 if (!net_eq(sock_net(sk), net))
812 return false;
813
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)))
822 return false;
823 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
824 return false;
825 return true;
826}
827
828static void udp6_csum_zero_error(struct sk_buff *skb)
829{
830 /* RFC 2460 section 8.1 says that we SHOULD log
831 * this error. Well, it is reasonable.
832 */
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));
836}
837
838/*
839 * Note: called only from the BH handler context,
840 * so we don't need to lock the hashes.
841 */
842static 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)
845{
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;
856
857 if (use_hash2) {
858 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) &
859 udptable->mask;
860 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask;
861start_lookup:
862 hslot = &udptable->hash2[hash2];
863 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
864 }
865
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,
869 hnum))
870 continue;
871 /* If zero checksum and no_check is not on for
872 * the socket then skip it.
873 */
874 if (!uh->check && !udp_get_no_check6_rx(sk))
875 continue;
876 if (!first) {
877 first = sk;
878 continue;
879 }
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,
884 IS_UDPLITE(sk));
885 __UDP6_INC_STATS(net, UDP_MIB_INERRORS,
886 IS_UDPLITE(sk));
887 continue;
888 }
889
890 if (udpv6_queue_rcv_skb(sk, nskb) > 0)
891 consume_skb(nskb);
892 }
893
894 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
895 if (use_hash2 && hash2 != hash2_any) {
896 hash2 = hash2_any;
897 goto start_lookup;
898 }
899
900 if (first) {
901 if (udpv6_queue_rcv_skb(first, skb) > 0)
902 consume_skb(skb);
903 } else {
904 kfree_skb(skb);
905 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
906 proto == IPPROTO_UDPLITE);
907 }
908 return 0;
909}
910
911static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
912{
913 if (udp_sk_rx_dst_set(sk, dst))
914 sk->sk_rx_dst_cookie = rt6_get_cookie(dst_rt6_info(dst));
915}
916
917/* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
918 * return code conversion for ip layer consumption
919 */
920static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
921 struct udphdr *uh)
922{
923 int ret;
924
925 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
926 skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo);
927
928 ret = udpv6_queue_rcv_skb(sk, skb);
929
930 /* a return value > 0 means to resubmit the input */
931 if (ret > 0)
932 return ret;
933 return 0;
934}
935
936int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
937 int proto)
938{
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 udphdr *uh;
943 struct sock *sk;
944 bool refcounted;
945 u32 ulen = 0;
946
947 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
948 goto discard;
949
950 saddr = &ipv6_hdr(skb)->saddr;
951 daddr = &ipv6_hdr(skb)->daddr;
952 uh = udp_hdr(skb);
953
954 ulen = ntohs(uh->len);
955 if (ulen > skb->len)
956 goto short_packet;
957
958 if (proto == IPPROTO_UDP) {
959 /* UDP validates ulen. */
960
961 /* Check for jumbo payload */
962 if (ulen == 0)
963 ulen = skb->len;
964
965 if (ulen < sizeof(*uh))
966 goto short_packet;
967
968 if (ulen < skb->len) {
969 if (pskb_trim_rcsum(skb, ulen))
970 goto short_packet;
971 saddr = &ipv6_hdr(skb)->saddr;
972 daddr = &ipv6_hdr(skb)->daddr;
973 uh = udp_hdr(skb);
974 }
975 }
976
977 if (udp6_csum_init(skb, uh, proto))
978 goto csum_error;
979
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);
983 if (IS_ERR(sk))
984 goto no_sk;
985
986 if (sk) {
987 struct dst_entry *dst = skb_dst(skb);
988 int ret;
989
990 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
991 udp6_sk_rx_dst_set(sk, dst);
992
993 if (!uh->check && !udp_get_no_check6_rx(sk)) {
994 if (refcounted)
995 sock_put(sk);
996 goto report_csum_error;
997 }
998
999 ret = udp6_unicast_rcv_skb(sk, skb, uh);
1000 if (refcounted)
1001 sock_put(sk);
1002 return ret;
1003 }
1004
1005 /*
1006 * Multicast receive code
1007 */
1008 if (ipv6_addr_is_multicast(daddr))
1009 return __udp6_lib_mcast_deliver(net, skb,
1010 saddr, daddr, udptable, proto);
1011
1012 /* Unicast */
1013 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1014 if (sk) {
1015 if (!uh->check && !udp_get_no_check6_rx(sk))
1016 goto report_csum_error;
1017 return udp6_unicast_rcv_skb(sk, skb, uh);
1018 }
1019no_sk:
1020 reason = SKB_DROP_REASON_NO_SOCKET;
1021
1022 if (!uh->check)
1023 goto report_csum_error;
1024
1025 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1026 goto discard;
1027 nf_reset_ct(skb);
1028
1029 if (udp_lib_checksum_complete(skb))
1030 goto csum_error;
1031
1032 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1033 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
1034
1035 kfree_skb_reason(skb, reason);
1036 return 0;
1037
1038short_packet:
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),
1044 ulen, skb->len,
1045 daddr, ntohs(uh->dest));
1046 goto discard;
1047
1048report_csum_error:
1049 udp6_csum_zero_error(skb);
1050csum_error:
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);
1054discard:
1055 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1056 kfree_skb_reason(skb, reason);
1057 return 0;
1058}
1059
1060
1061static 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,
1064 int dif, int sdif)
1065{
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;
1070 __portpair ports;
1071 struct sock *sk;
1072
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);
1077
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))
1081 return sk;
1082 /* Only check first socket in chain */
1083 break;
1084 }
1085 return NULL;
1086}
1087
1088void udp_v6_early_demux(struct sk_buff *skb)
1089{
1090 struct net *net = dev_net(skb->dev);
1091 const struct udphdr *uh;
1092 struct sock *sk;
1093 struct dst_entry *dst;
1094 int dif = skb->dev->ifindex;
1095 int sdif = inet6_sdif(skb);
1096
1097 if (!pskb_may_pull(skb, skb_transport_offset(skb) +
1098 sizeof(struct udphdr)))
1099 return;
1100
1101 uh = udp_hdr(skb);
1102
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,
1107 dif, sdif);
1108 else
1109 return;
1110
1111 if (!sk)
1112 return;
1113
1114 skb->sk = sk;
1115 DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk));
1116 skb->destructor = sock_pfree;
1117 dst = rcu_dereference(sk->sk_rx_dst);
1118
1119 if (dst)
1120 dst = dst_check(dst, sk->sk_rx_dst_cookie);
1121 if (dst) {
1122 /* set noref for now.
1123 * any place which wants to hold dst has to call
1124 * dst_hold_safe()
1125 */
1126 skb_dst_set_noref(skb, dst);
1127 }
1128}
1129
1130INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb)
1131{
1132 return __udp6_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
1133}
1134
1135/*
1136 * Throw away all pending data and cancel the corking. Socket is locked.
1137 */
1138static void udp_v6_flush_pending_frames(struct sock *sk)
1139{
1140 struct udp_sock *up = udp_sk(sk);
1141
1142 if (up->pending == AF_INET)
1143 udp_flush_pending_frames(sk);
1144 else if (up->pending) {
1145 up->len = 0;
1146 WRITE_ONCE(up->pending, 0);
1147 ip6_flush_pending_frames(sk);
1148 }
1149}
1150
1151static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
1152 int addr_len)
1153{
1154 if (addr_len < offsetofend(struct sockaddr, sa_family))
1155 return -EINVAL;
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.
1159 */
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);
1164 }
1165
1166 if (addr_len < SIN6_LEN_RFC2133)
1167 return -EINVAL;
1168
1169 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr, &addr_len);
1170}
1171
1172/**
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
1180 */
1181static 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)
1184{
1185 unsigned int offset;
1186 struct udphdr *uh = udp_hdr(skb);
1187 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
1188 __wsum csum = 0;
1189
1190 if (!frags) {
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);
1195 } else {
1196 /*
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
1200 */
1201 offset = skb_transport_offset(skb);
1202 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1203 csum = skb->csum;
1204
1205 skb->ip_summed = CHECKSUM_NONE;
1206
1207 do {
1208 csum = csum_add(csum, frags->csum);
1209 } while ((frags = frags->next));
1210
1211 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1212 csum);
1213 if (uh->check == 0)
1214 uh->check = CSUM_MANGLED_0;
1215 }
1216}
1217
1218/*
1219 * Sending
1220 */
1221
1222static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6,
1223 struct inet_cork *cork)
1224{
1225 struct sock *sk = skb->sk;
1226 struct udphdr *uh;
1227 int err = 0;
1228 int is_udplite = IS_UDPLITE(sk);
1229 __wsum csum = 0;
1230 int offset = skb_transport_offset(skb);
1231 int len = skb->len - offset;
1232 int datalen = len - sizeof(*uh);
1233
1234 /*
1235 * Create a UDP header
1236 */
1237 uh = udp_hdr(skb);
1238 uh->source = fl6->fl6_sport;
1239 uh->dest = fl6->fl6_dport;
1240 uh->len = htons(len);
1241 uh->check = 0;
1242
1243 if (cork->gso_size) {
1244 const int hlen = skb_network_header_len(skb) +
1245 sizeof(struct udphdr);
1246
1247 if (hlen + cork->gso_size > cork->fragsize) {
1248 kfree_skb(skb);
1249 return -EINVAL;
1250 }
1251 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
1252 kfree_skb(skb);
1253 return -EINVAL;
1254 }
1255 if (udp_get_no_check6_tx(sk)) {
1256 kfree_skb(skb);
1257 return -EINVAL;
1258 }
1259 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
1260 dst_xfrm(skb_dst(skb))) {
1261 kfree_skb(skb);
1262 return -EIO;
1263 }
1264
1265 if (datalen > cork->gso_size) {
1266 skb_shinfo(skb)->gso_size = cork->gso_size;
1267 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
1268 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
1269 cork->gso_size);
1270 }
1271 goto csum_partial;
1272 }
1273
1274 if (is_udplite)
1275 csum = udplite_csum(skb);
1276 else if (udp_get_no_check6_tx(sk)) { /* UDP csum disabled */
1277 skb->ip_summed = CHECKSUM_NONE;
1278 goto send;
1279 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1280csum_partial:
1281 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1282 goto send;
1283 } else
1284 csum = udp_csum(skb);
1285
1286 /* add protocol-dependent pseudo-header */
1287 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1288 len, fl6->flowi6_proto, csum);
1289 if (uh->check == 0)
1290 uh->check = CSUM_MANGLED_0;
1291
1292send:
1293 err = ip6_send_skb(skb);
1294 if (err) {
1295 if (err == -ENOBUFS && !inet6_test_bit(RECVERR6, sk)) {
1296 UDP6_INC_STATS(sock_net(sk),
1297 UDP_MIB_SNDBUFERRORS, is_udplite);
1298 err = 0;
1299 }
1300 } else {
1301 UDP6_INC_STATS(sock_net(sk),
1302 UDP_MIB_OUTDATAGRAMS, is_udplite);
1303 }
1304 return err;
1305}
1306
1307static int udp_v6_push_pending_frames(struct sock *sk)
1308{
1309 struct sk_buff *skb;
1310 struct udp_sock *up = udp_sk(sk);
1311 int err = 0;
1312
1313 if (up->pending == AF_INET)
1314 return udp_push_pending_frames(sk);
1315
1316 skb = ip6_finish_skb(sk);
1317 if (!skb)
1318 goto out;
1319
1320 err = udp_v6_send_skb(skb, &inet_sk(sk)->cork.fl.u.ip6,
1321 &inet_sk(sk)->cork.base);
1322out:
1323 up->len = 0;
1324 WRITE_ONCE(up->pending, 0);
1325 return err;
1326}
1327
1328int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1329{
1330 struct ipv6_txoptions opt_space;
1331 struct udp_sock *up = udp_sk(sk);
1332 struct inet_sock *inet = inet_sk(sk);
1333 struct ipv6_pinfo *np = inet6_sk(sk);
1334 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1335 struct in6_addr *daddr, *final_p, final;
1336 struct ipv6_txoptions *opt = NULL;
1337 struct ipv6_txoptions *opt_to_free = NULL;
1338 struct ip6_flowlabel *flowlabel = NULL;
1339 struct inet_cork_full cork;
1340 struct flowi6 *fl6 = &cork.fl.u.ip6;
1341 struct dst_entry *dst;
1342 struct ipcm6_cookie ipc6;
1343 int addr_len = msg->msg_namelen;
1344 bool connected = false;
1345 int ulen = len;
1346 int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE;
1347 int err;
1348 int is_udplite = IS_UDPLITE(sk);
1349 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1350
1351 ipcm6_init(&ipc6);
1352 ipc6.gso_size = READ_ONCE(up->gso_size);
1353 ipc6.sockc.tsflags = READ_ONCE(sk->sk_tsflags);
1354 ipc6.sockc.mark = READ_ONCE(sk->sk_mark);
1355
1356 /* destination address check */
1357 if (sin6) {
1358 if (addr_len < offsetof(struct sockaddr, sa_data))
1359 return -EINVAL;
1360
1361 switch (sin6->sin6_family) {
1362 case AF_INET6:
1363 if (addr_len < SIN6_LEN_RFC2133)
1364 return -EINVAL;
1365 daddr = &sin6->sin6_addr;
1366 if (ipv6_addr_any(daddr) &&
1367 ipv6_addr_v4mapped(&np->saddr))
1368 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1369 daddr);
1370 break;
1371 case AF_INET:
1372 goto do_udp_sendmsg;
1373 case AF_UNSPEC:
1374 msg->msg_name = sin6 = NULL;
1375 msg->msg_namelen = addr_len = 0;
1376 daddr = NULL;
1377 break;
1378 default:
1379 return -EINVAL;
1380 }
1381 } else if (!READ_ONCE(up->pending)) {
1382 if (sk->sk_state != TCP_ESTABLISHED)
1383 return -EDESTADDRREQ;
1384 daddr = &sk->sk_v6_daddr;
1385 } else
1386 daddr = NULL;
1387
1388 if (daddr) {
1389 if (ipv6_addr_v4mapped(daddr)) {
1390 struct sockaddr_in sin;
1391 sin.sin_family = AF_INET;
1392 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1393 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1394 msg->msg_name = &sin;
1395 msg->msg_namelen = sizeof(sin);
1396do_udp_sendmsg:
1397 err = ipv6_only_sock(sk) ?
1398 -ENETUNREACH : udp_sendmsg(sk, msg, len);
1399 msg->msg_name = sin6;
1400 msg->msg_namelen = addr_len;
1401 return err;
1402 }
1403 }
1404
1405 /* Rough check on arithmetic overflow,
1406 better check is made in ip6_append_data().
1407 */
1408 if (len > INT_MAX - sizeof(struct udphdr))
1409 return -EMSGSIZE;
1410
1411 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1412 if (READ_ONCE(up->pending)) {
1413 if (READ_ONCE(up->pending) == AF_INET)
1414 return udp_sendmsg(sk, msg, len);
1415 /*
1416 * There are pending frames.
1417 * The socket lock must be held while it's corked.
1418 */
1419 lock_sock(sk);
1420 if (likely(up->pending)) {
1421 if (unlikely(up->pending != AF_INET6)) {
1422 release_sock(sk);
1423 return -EAFNOSUPPORT;
1424 }
1425 dst = NULL;
1426 goto do_append_data;
1427 }
1428 release_sock(sk);
1429 }
1430 ulen += sizeof(struct udphdr);
1431
1432 memset(fl6, 0, sizeof(*fl6));
1433
1434 if (sin6) {
1435 if (sin6->sin6_port == 0)
1436 return -EINVAL;
1437
1438 fl6->fl6_dport = sin6->sin6_port;
1439 daddr = &sin6->sin6_addr;
1440
1441 if (inet6_test_bit(SNDFLOW, sk)) {
1442 fl6->flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1443 if (fl6->flowlabel & IPV6_FLOWLABEL_MASK) {
1444 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel);
1445 if (IS_ERR(flowlabel))
1446 return -EINVAL;
1447 }
1448 }
1449
1450 /*
1451 * Otherwise it will be difficult to maintain
1452 * sk->sk_dst_cache.
1453 */
1454 if (sk->sk_state == TCP_ESTABLISHED &&
1455 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1456 daddr = &sk->sk_v6_daddr;
1457
1458 if (addr_len >= sizeof(struct sockaddr_in6) &&
1459 sin6->sin6_scope_id &&
1460 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1461 fl6->flowi6_oif = sin6->sin6_scope_id;
1462 } else {
1463 if (sk->sk_state != TCP_ESTABLISHED)
1464 return -EDESTADDRREQ;
1465
1466 fl6->fl6_dport = inet->inet_dport;
1467 daddr = &sk->sk_v6_daddr;
1468 fl6->flowlabel = np->flow_label;
1469 connected = true;
1470 }
1471
1472 if (!fl6->flowi6_oif)
1473 fl6->flowi6_oif = READ_ONCE(sk->sk_bound_dev_if);
1474
1475 if (!fl6->flowi6_oif)
1476 fl6->flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1477
1478 fl6->flowi6_uid = sk->sk_uid;
1479
1480 if (msg->msg_controllen) {
1481 opt = &opt_space;
1482 memset(opt, 0, sizeof(struct ipv6_txoptions));
1483 opt->tot_len = sizeof(*opt);
1484 ipc6.opt = opt;
1485
1486 err = udp_cmsg_send(sk, msg, &ipc6.gso_size);
1487 if (err > 0) {
1488 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, fl6,
1489 &ipc6);
1490 connected = false;
1491 }
1492 if (err < 0) {
1493 fl6_sock_release(flowlabel);
1494 return err;
1495 }
1496 if ((fl6->flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1497 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel);
1498 if (IS_ERR(flowlabel))
1499 return -EINVAL;
1500 }
1501 if (!(opt->opt_nflen|opt->opt_flen))
1502 opt = NULL;
1503 }
1504 if (!opt) {
1505 opt = txopt_get(np);
1506 opt_to_free = opt;
1507 }
1508 if (flowlabel)
1509 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1510 opt = ipv6_fixup_options(&opt_space, opt);
1511 ipc6.opt = opt;
1512
1513 fl6->flowi6_proto = sk->sk_protocol;
1514 fl6->flowi6_mark = ipc6.sockc.mark;
1515 fl6->daddr = *daddr;
1516 if (ipv6_addr_any(&fl6->saddr) && !ipv6_addr_any(&np->saddr))
1517 fl6->saddr = np->saddr;
1518 fl6->fl6_sport = inet->inet_sport;
1519
1520 if (cgroup_bpf_enabled(CGROUP_UDP6_SENDMSG) && !connected) {
1521 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk,
1522 (struct sockaddr *)sin6,
1523 &addr_len,
1524 &fl6->saddr);
1525 if (err)
1526 goto out_no_dst;
1527 if (sin6) {
1528 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
1529 /* BPF program rewrote IPv6-only by IPv4-mapped
1530 * IPv6. It's currently unsupported.
1531 */
1532 err = -ENOTSUPP;
1533 goto out_no_dst;
1534 }
1535 if (sin6->sin6_port == 0) {
1536 /* BPF program set invalid port. Reject it. */
1537 err = -EINVAL;
1538 goto out_no_dst;
1539 }
1540 fl6->fl6_dport = sin6->sin6_port;
1541 fl6->daddr = sin6->sin6_addr;
1542 }
1543 }
1544
1545 if (ipv6_addr_any(&fl6->daddr))
1546 fl6->daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1547
1548 final_p = fl6_update_dst(fl6, opt, &final);
1549 if (final_p)
1550 connected = false;
1551
1552 if (!fl6->flowi6_oif && ipv6_addr_is_multicast(&fl6->daddr)) {
1553 fl6->flowi6_oif = READ_ONCE(np->mcast_oif);
1554 connected = false;
1555 } else if (!fl6->flowi6_oif)
1556 fl6->flowi6_oif = READ_ONCE(np->ucast_oif);
1557
1558 security_sk_classify_flow(sk, flowi6_to_flowi_common(fl6));
1559
1560 if (ipc6.tclass < 0)
1561 ipc6.tclass = np->tclass;
1562
1563 fl6->flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6->flowlabel);
1564
1565 dst = ip6_sk_dst_lookup_flow(sk, fl6, final_p, connected);
1566 if (IS_ERR(dst)) {
1567 err = PTR_ERR(dst);
1568 dst = NULL;
1569 goto out;
1570 }
1571
1572 if (ipc6.hlimit < 0)
1573 ipc6.hlimit = ip6_sk_dst_hoplimit(np, fl6, dst);
1574
1575 if (msg->msg_flags&MSG_CONFIRM)
1576 goto do_confirm;
1577back_from_confirm:
1578
1579 /* Lockless fast path for the non-corking case */
1580 if (!corkreq) {
1581 struct sk_buff *skb;
1582
1583 skb = ip6_make_skb(sk, getfrag, msg, ulen,
1584 sizeof(struct udphdr), &ipc6,
1585 dst_rt6_info(dst),
1586 msg->msg_flags, &cork);
1587 err = PTR_ERR(skb);
1588 if (!IS_ERR_OR_NULL(skb))
1589 err = udp_v6_send_skb(skb, fl6, &cork.base);
1590 /* ip6_make_skb steals dst reference */
1591 goto out_no_dst;
1592 }
1593
1594 lock_sock(sk);
1595 if (unlikely(up->pending)) {
1596 /* The socket is already corked while preparing it. */
1597 /* ... which is an evident application bug. --ANK */
1598 release_sock(sk);
1599
1600 net_dbg_ratelimited("udp cork app bug 2\n");
1601 err = -EINVAL;
1602 goto out;
1603 }
1604
1605 WRITE_ONCE(up->pending, AF_INET6);
1606
1607do_append_data:
1608 if (ipc6.dontfrag < 0)
1609 ipc6.dontfrag = inet6_test_bit(DONTFRAG, sk);
1610 up->len += ulen;
1611 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr),
1612 &ipc6, fl6, dst_rt6_info(dst),
1613 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1614 if (err)
1615 udp_v6_flush_pending_frames(sk);
1616 else if (!corkreq)
1617 err = udp_v6_push_pending_frames(sk);
1618 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1619 WRITE_ONCE(up->pending, 0);
1620
1621 if (err > 0)
1622 err = inet6_test_bit(RECVERR6, sk) ? net_xmit_errno(err) : 0;
1623 release_sock(sk);
1624
1625out:
1626 dst_release(dst);
1627out_no_dst:
1628 fl6_sock_release(flowlabel);
1629 txopt_put(opt_to_free);
1630 if (!err)
1631 return len;
1632 /*
1633 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1634 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1635 * we don't have a good statistic (IpOutDiscards but it can be too many
1636 * things). We could add another new stat but at least for now that
1637 * seems like overkill.
1638 */
1639 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1640 UDP6_INC_STATS(sock_net(sk),
1641 UDP_MIB_SNDBUFERRORS, is_udplite);
1642 }
1643 return err;
1644
1645do_confirm:
1646 if (msg->msg_flags & MSG_PROBE)
1647 dst_confirm_neigh(dst, &fl6->daddr);
1648 if (!(msg->msg_flags&MSG_PROBE) || len)
1649 goto back_from_confirm;
1650 err = 0;
1651 goto out;
1652}
1653EXPORT_SYMBOL(udpv6_sendmsg);
1654
1655static void udpv6_splice_eof(struct socket *sock)
1656{
1657 struct sock *sk = sock->sk;
1658 struct udp_sock *up = udp_sk(sk);
1659
1660 if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk))
1661 return;
1662
1663 lock_sock(sk);
1664 if (up->pending && !udp_test_bit(CORK, sk))
1665 udp_v6_push_pending_frames(sk);
1666 release_sock(sk);
1667}
1668
1669void udpv6_destroy_sock(struct sock *sk)
1670{
1671 struct udp_sock *up = udp_sk(sk);
1672 lock_sock(sk);
1673
1674 /* protects from races with udp_abort() */
1675 sock_set_flag(sk, SOCK_DEAD);
1676 udp_v6_flush_pending_frames(sk);
1677 release_sock(sk);
1678
1679 if (static_branch_unlikely(&udpv6_encap_needed_key)) {
1680 if (up->encap_type) {
1681 void (*encap_destroy)(struct sock *sk);
1682 encap_destroy = READ_ONCE(up->encap_destroy);
1683 if (encap_destroy)
1684 encap_destroy(sk);
1685 }
1686 if (udp_test_bit(ENCAP_ENABLED, sk)) {
1687 static_branch_dec(&udpv6_encap_needed_key);
1688 udp_encap_disable();
1689 }
1690 }
1691}
1692
1693/*
1694 * Socket option code for UDP
1695 */
1696int udpv6_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
1697 unsigned int optlen)
1698{
1699 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET)
1700 return udp_lib_setsockopt(sk, level, optname,
1701 optval, optlen,
1702 udp_v6_push_pending_frames);
1703 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1704}
1705
1706int udpv6_getsockopt(struct sock *sk, int level, int optname,
1707 char __user *optval, int __user *optlen)
1708{
1709 if (level == SOL_UDP || level == SOL_UDPLITE)
1710 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1711 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1712}
1713
1714
1715/* ------------------------------------------------------------------------ */
1716#ifdef CONFIG_PROC_FS
1717int udp6_seq_show(struct seq_file *seq, void *v)
1718{
1719 if (v == SEQ_START_TOKEN) {
1720 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1721 } else {
1722 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1723 const struct inet_sock *inet = inet_sk((const struct sock *)v);
1724 __u16 srcp = ntohs(inet->inet_sport);
1725 __u16 destp = ntohs(inet->inet_dport);
1726 __ip6_dgram_sock_seq_show(seq, v, srcp, destp,
1727 udp_rqueue_get(v), bucket);
1728 }
1729 return 0;
1730}
1731
1732const struct seq_operations udp6_seq_ops = {
1733 .start = udp_seq_start,
1734 .next = udp_seq_next,
1735 .stop = udp_seq_stop,
1736 .show = udp6_seq_show,
1737};
1738EXPORT_SYMBOL(udp6_seq_ops);
1739
1740static struct udp_seq_afinfo udp6_seq_afinfo = {
1741 .family = AF_INET6,
1742 .udp_table = NULL,
1743};
1744
1745int __net_init udp6_proc_init(struct net *net)
1746{
1747 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops,
1748 sizeof(struct udp_iter_state), &udp6_seq_afinfo))
1749 return -ENOMEM;
1750 return 0;
1751}
1752
1753void udp6_proc_exit(struct net *net)
1754{
1755 remove_proc_entry("udp6", net->proc_net);
1756}
1757#endif /* CONFIG_PROC_FS */
1758
1759/* ------------------------------------------------------------------------ */
1760
1761struct proto udpv6_prot = {
1762 .name = "UDPv6",
1763 .owner = THIS_MODULE,
1764 .close = udp_lib_close,
1765 .pre_connect = udpv6_pre_connect,
1766 .connect = ip6_datagram_connect,
1767 .disconnect = udp_disconnect,
1768 .ioctl = udp_ioctl,
1769 .init = udpv6_init_sock,
1770 .destroy = udpv6_destroy_sock,
1771 .setsockopt = udpv6_setsockopt,
1772 .getsockopt = udpv6_getsockopt,
1773 .sendmsg = udpv6_sendmsg,
1774 .recvmsg = udpv6_recvmsg,
1775 .splice_eof = udpv6_splice_eof,
1776 .release_cb = ip6_datagram_release_cb,
1777 .hash = udp_lib_hash,
1778 .unhash = udp_lib_unhash,
1779 .rehash = udp_v6_rehash,
1780 .get_port = udp_v6_get_port,
1781 .put_port = udp_lib_unhash,
1782#ifdef CONFIG_BPF_SYSCALL
1783 .psock_update_sk_prot = udp_bpf_update_proto,
1784#endif
1785
1786 .memory_allocated = &udp_memory_allocated,
1787 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
1788
1789 .sysctl_mem = sysctl_udp_mem,
1790 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
1791 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
1792 .obj_size = sizeof(struct udp6_sock),
1793 .ipv6_pinfo_offset = offsetof(struct udp6_sock, inet6),
1794 .h.udp_table = NULL,
1795 .diag_destroy = udp_abort,
1796};
1797
1798static struct inet_protosw udpv6_protosw = {
1799 .type = SOCK_DGRAM,
1800 .protocol = IPPROTO_UDP,
1801 .prot = &udpv6_prot,
1802 .ops = &inet6_dgram_ops,
1803 .flags = INET_PROTOSW_PERMANENT,
1804};
1805
1806int __init udpv6_init(void)
1807{
1808 int ret;
1809
1810 net_hotdata.udpv6_protocol = (struct inet6_protocol) {
1811 .handler = udpv6_rcv,
1812 .err_handler = udpv6_err,
1813 .flags = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL,
1814 };
1815 ret = inet6_add_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP);
1816 if (ret)
1817 goto out;
1818
1819 ret = inet6_register_protosw(&udpv6_protosw);
1820 if (ret)
1821 goto out_udpv6_protocol;
1822out:
1823 return ret;
1824
1825out_udpv6_protocol:
1826 inet6_del_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP);
1827 goto out;
1828}
1829
1830void udpv6_exit(void)
1831{
1832 inet6_unregister_protosw(&udpv6_protosw);
1833 inet6_del_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP);
1834}