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/errno.h>
21#include <linux/types.h>
22#include <linux/socket.h>
23#include <linux/sockios.h>
24#include <linux/net.h>
25#include <linux/in6.h>
26#include <linux/netdevice.h>
27#include <linux/if_arp.h>
28#include <linux/ipv6.h>
29#include <linux/icmpv6.h>
30#include <linux/init.h>
31#include <linux/module.h>
32#include <linux/skbuff.h>
33#include <linux/slab.h>
34#include <linux/uaccess.h>
35#include <linux/indirect_call_wrapper.h>
36
37#include <net/addrconf.h>
38#include <net/ndisc.h>
39#include <net/protocol.h>
40#include <net/transp_v6.h>
41#include <net/ip6_route.h>
42#include <net/raw.h>
43#include <net/tcp_states.h>
44#include <net/ip6_checksum.h>
45#include <net/ip6_tunnel.h>
46#include <net/xfrm.h>
47#include <net/inet_hashtables.h>
48#include <net/inet6_hashtables.h>
49#include <net/busy_poll.h>
50#include <net/sock_reuseport.h>
51
52#include <linux/proc_fs.h>
53#include <linux/seq_file.h>
54#include <trace/events/skb.h>
55#include "udp_impl.h"
56
57static u32 udp6_ehashfn(const struct net *net,
58 const struct in6_addr *laddr,
59 const u16 lport,
60 const struct in6_addr *faddr,
61 const __be16 fport)
62{
63 static u32 udp6_ehash_secret __read_mostly;
64 static u32 udp_ipv6_hash_secret __read_mostly;
65
66 u32 lhash, fhash;
67
68 net_get_random_once(&udp6_ehash_secret,
69 sizeof(udp6_ehash_secret));
70 net_get_random_once(&udp_ipv6_hash_secret,
71 sizeof(udp_ipv6_hash_secret));
72
73 lhash = (__force u32)laddr->s6_addr32[3];
74 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
75
76 return __inet6_ehashfn(lhash, lport, fhash, fport,
77 udp_ipv6_hash_secret + net_hash_mix(net));
78}
79
80int udp_v6_get_port(struct sock *sk, unsigned short snum)
81{
82 unsigned int hash2_nulladdr =
83 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
84 unsigned int hash2_partial =
85 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
86
87 /* precompute partial secondary hash */
88 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
89 return udp_lib_get_port(sk, snum, hash2_nulladdr);
90}
91
92void udp_v6_rehash(struct sock *sk)
93{
94 u16 new_hash = ipv6_portaddr_hash(sock_net(sk),
95 &sk->sk_v6_rcv_saddr,
96 inet_sk(sk)->inet_num);
97
98 udp_lib_rehash(sk, new_hash);
99}
100
101static int compute_score(struct sock *sk, struct net *net,
102 const struct in6_addr *saddr, __be16 sport,
103 const struct in6_addr *daddr, unsigned short hnum,
104 int dif, int sdif)
105{
106 int score;
107 struct inet_sock *inet;
108 bool dev_match;
109
110 if (!net_eq(sock_net(sk), net) ||
111 udp_sk(sk)->udp_port_hash != hnum ||
112 sk->sk_family != PF_INET6)
113 return -1;
114
115 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
116 return -1;
117
118 score = 0;
119 inet = inet_sk(sk);
120
121 if (inet->inet_dport) {
122 if (inet->inet_dport != sport)
123 return -1;
124 score++;
125 }
126
127 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
128 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
129 return -1;
130 score++;
131 }
132
133 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif);
134 if (!dev_match)
135 return -1;
136 score++;
137
138 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
139 score++;
140
141 return score;
142}
143
144static struct sock *lookup_reuseport(struct net *net, struct sock *sk,
145 struct sk_buff *skb,
146 const struct in6_addr *saddr,
147 __be16 sport,
148 const struct in6_addr *daddr,
149 unsigned int hnum)
150{
151 struct sock *reuse_sk = NULL;
152 u32 hash;
153
154 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) {
155 hash = udp6_ehashfn(net, daddr, hnum, saddr, sport);
156 reuse_sk = reuseport_select_sock(sk, hash, skb,
157 sizeof(struct udphdr));
158 }
159 return reuse_sk;
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
172 result = NULL;
173 badness = -1;
174 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
175 score = compute_score(sk, net, saddr, sport,
176 daddr, hnum, dif, sdif);
177 if (score > badness) {
178 result = lookup_reuseport(net, sk, skb,
179 saddr, sport, daddr, hnum);
180 /* Fall back to scoring if group has connections */
181 if (result && !reuseport_has_conns(sk, false))
182 return result;
183
184 result = result ? : sk;
185 badness = score;
186 }
187 }
188 return result;
189}
190
191static inline struct sock *udp6_lookup_run_bpf(struct net *net,
192 struct udp_table *udptable,
193 struct sk_buff *skb,
194 const struct in6_addr *saddr,
195 __be16 sport,
196 const struct in6_addr *daddr,
197 u16 hnum)
198{
199 struct sock *sk, *reuse_sk;
200 bool no_reuseport;
201
202 if (udptable != &udp_table)
203 return NULL; /* only UDP is supported */
204
205 no_reuseport = bpf_sk_lookup_run_v6(net, IPPROTO_UDP,
206 saddr, sport, daddr, hnum, &sk);
207 if (no_reuseport || IS_ERR_OR_NULL(sk))
208 return sk;
209
210 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum);
211 if (reuse_sk)
212 sk = reuse_sk;
213 return sk;
214}
215
216/* rcu_read_lock() must be held */
217struct sock *__udp6_lib_lookup(struct net *net,
218 const struct in6_addr *saddr, __be16 sport,
219 const struct in6_addr *daddr, __be16 dport,
220 int dif, int sdif, struct udp_table *udptable,
221 struct sk_buff *skb)
222{
223 unsigned short hnum = ntohs(dport);
224 unsigned int hash2, slot2;
225 struct udp_hslot *hslot2;
226 struct sock *result, *sk;
227
228 hash2 = ipv6_portaddr_hash(net, daddr, hnum);
229 slot2 = hash2 & udptable->mask;
230 hslot2 = &udptable->hash2[slot2];
231
232 /* Lookup connected or non-wildcard sockets */
233 result = udp6_lib_lookup2(net, saddr, sport,
234 daddr, hnum, dif, sdif,
235 hslot2, skb);
236 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
237 goto done;
238
239 /* Lookup redirect from BPF */
240 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
241 sk = udp6_lookup_run_bpf(net, udptable, skb,
242 saddr, sport, daddr, hnum);
243 if (sk) {
244 result = sk;
245 goto done;
246 }
247 }
248
249 /* Got non-wildcard socket or error on first lookup */
250 if (result)
251 goto done;
252
253 /* Lookup wildcard sockets */
254 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
255 slot2 = hash2 & udptable->mask;
256 hslot2 = &udptable->hash2[slot2];
257
258 result = udp6_lib_lookup2(net, saddr, sport,
259 &in6addr_any, hnum, dif, sdif,
260 hslot2, skb);
261done:
262 if (IS_ERR(result))
263 return NULL;
264 return result;
265}
266EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
267
268static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
269 __be16 sport, __be16 dport,
270 struct udp_table *udptable)
271{
272 const struct ipv6hdr *iph = ipv6_hdr(skb);
273
274 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
275 &iph->daddr, dport, inet6_iif(skb),
276 inet6_sdif(skb), udptable, skb);
277}
278
279struct sock *udp6_lib_lookup_skb(struct sk_buff *skb,
280 __be16 sport, __be16 dport)
281{
282 const struct ipv6hdr *iph = ipv6_hdr(skb);
283
284 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
285 &iph->daddr, dport, inet6_iif(skb),
286 inet6_sdif(skb), &udp_table, NULL);
287}
288EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb);
289
290/* Must be called under rcu_read_lock().
291 * Does increment socket refcount.
292 */
293#if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
294struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
295 const struct in6_addr *daddr, __be16 dport, int dif)
296{
297 struct sock *sk;
298
299 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport,
300 dif, 0, &udp_table, NULL);
301 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
302 sk = NULL;
303 return sk;
304}
305EXPORT_SYMBOL_GPL(udp6_lib_lookup);
306#endif
307
308/* do not use the scratch area len for jumbogram: their length execeeds the
309 * scratch area space; note that the IP6CB flags is still in the first
310 * cacheline, so checking for jumbograms is cheap
311 */
312static int udp6_skb_len(struct sk_buff *skb)
313{
314 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb);
315}
316
317/*
318 * This should be easy, if there is something there we
319 * return it, otherwise we block.
320 */
321
322int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
323 int noblock, int flags, int *addr_len)
324{
325 struct ipv6_pinfo *np = inet6_sk(sk);
326 struct inet_sock *inet = inet_sk(sk);
327 struct sk_buff *skb;
328 unsigned int ulen, copied;
329 int off, err, peeking = flags & MSG_PEEK;
330 int is_udplite = IS_UDPLITE(sk);
331 struct udp_mib __percpu *mib;
332 bool checksum_valid = false;
333 int is_udp4;
334
335 if (flags & MSG_ERRQUEUE)
336 return ipv6_recv_error(sk, msg, len, addr_len);
337
338 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
339 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
340
341try_again:
342 off = sk_peek_offset(sk, flags);
343 skb = __skb_recv_udp(sk, flags, noblock, &off, &err);
344 if (!skb)
345 return err;
346
347 ulen = udp6_skb_len(skb);
348 copied = len;
349 if (copied > ulen - off)
350 copied = ulen - off;
351 else if (copied < ulen)
352 msg->msg_flags |= MSG_TRUNC;
353
354 is_udp4 = (skb->protocol == htons(ETH_P_IP));
355 mib = __UDPX_MIB(sk, is_udp4);
356
357 /*
358 * If checksum is needed at all, try to do it while copying the
359 * data. If the data is truncated, or if we only want a partial
360 * coverage checksum (UDP-Lite), do it before the copy.
361 */
362
363 if (copied < ulen || peeking ||
364 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
365 checksum_valid = udp_skb_csum_unnecessary(skb) ||
366 !__udp_lib_checksum_complete(skb);
367 if (!checksum_valid)
368 goto csum_copy_err;
369 }
370
371 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
372 if (udp_skb_is_linear(skb))
373 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
374 else
375 err = skb_copy_datagram_msg(skb, off, msg, copied);
376 } else {
377 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
378 if (err == -EINVAL)
379 goto csum_copy_err;
380 }
381 if (unlikely(err)) {
382 if (!peeking) {
383 atomic_inc(&sk->sk_drops);
384 SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
385 }
386 kfree_skb(skb);
387 return err;
388 }
389 if (!peeking)
390 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS);
391
392 sock_recv_ts_and_drops(msg, sk, skb);
393
394 /* Copy the address. */
395 if (msg->msg_name) {
396 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
397 sin6->sin6_family = AF_INET6;
398 sin6->sin6_port = udp_hdr(skb)->source;
399 sin6->sin6_flowinfo = 0;
400
401 if (is_udp4) {
402 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
403 &sin6->sin6_addr);
404 sin6->sin6_scope_id = 0;
405 } else {
406 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
407 sin6->sin6_scope_id =
408 ipv6_iface_scope_id(&sin6->sin6_addr,
409 inet6_iif(skb));
410 }
411 *addr_len = sizeof(*sin6);
412
413 if (cgroup_bpf_enabled)
414 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk,
415 (struct sockaddr *)sin6);
416 }
417
418 if (udp_sk(sk)->gro_enabled)
419 udp_cmsg_recv(msg, sk, skb);
420
421 if (np->rxopt.all)
422 ip6_datagram_recv_common_ctl(sk, msg, skb);
423
424 if (is_udp4) {
425 if (inet->cmsg_flags)
426 ip_cmsg_recv_offset(msg, sk, skb,
427 sizeof(struct udphdr), off);
428 } else {
429 if (np->rxopt.all)
430 ip6_datagram_recv_specific_ctl(sk, msg, skb);
431 }
432
433 err = copied;
434 if (flags & MSG_TRUNC)
435 err = ulen;
436
437 skb_consume_udp(sk, skb, peeking ? -err : err);
438 return err;
439
440csum_copy_err:
441 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
442 udp_skb_destructor)) {
443 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS);
444 SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
445 }
446 kfree_skb(skb);
447
448 /* starting over for a new packet, but check if we need to yield */
449 cond_resched();
450 msg->msg_flags &= ~MSG_TRUNC;
451 goto try_again;
452}
453
454DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
455void udpv6_encap_enable(void)
456{
457 static_branch_inc(&udpv6_encap_needed_key);
458}
459EXPORT_SYMBOL(udpv6_encap_enable);
460
461/* Handler for tunnels with arbitrary destination ports: no socket lookup, go
462 * through error handlers in encapsulations looking for a match.
463 */
464static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb,
465 struct inet6_skb_parm *opt,
466 u8 type, u8 code, int offset, __be32 info)
467{
468 int i;
469
470 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
471 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt,
472 u8 type, u8 code, int offset, __be32 info);
473 const struct ip6_tnl_encap_ops *encap;
474
475 encap = rcu_dereference(ip6tun_encaps[i]);
476 if (!encap)
477 continue;
478 handler = encap->err_handler;
479 if (handler && !handler(skb, opt, type, code, offset, info))
480 return 0;
481 }
482
483 return -ENOENT;
484}
485
486/* Try to match ICMP errors to UDP tunnels by looking up a socket without
487 * reversing source and destination port: this will match tunnels that force the
488 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
489 * lwtunnels might actually break this assumption by being configured with
490 * different destination ports on endpoints, in this case we won't be able to
491 * trace ICMP messages back to them.
492 *
493 * If this doesn't match any socket, probe tunnels with arbitrary destination
494 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
495 * we've sent packets to won't necessarily match the local destination port.
496 *
497 * Then ask the tunnel implementation to match the error against a valid
498 * association.
499 *
500 * Return an error if we can't find a match, the socket if we need further
501 * processing, zero otherwise.
502 */
503static struct sock *__udp6_lib_err_encap(struct net *net,
504 const struct ipv6hdr *hdr, int offset,
505 struct udphdr *uh,
506 struct udp_table *udptable,
507 struct sk_buff *skb,
508 struct inet6_skb_parm *opt,
509 u8 type, u8 code, __be32 info)
510{
511 int network_offset, transport_offset;
512 struct sock *sk;
513
514 network_offset = skb_network_offset(skb);
515 transport_offset = skb_transport_offset(skb);
516
517 /* Network header needs to point to the outer IPv6 header inside ICMP */
518 skb_reset_network_header(skb);
519
520 /* Transport header needs to point to the UDP header */
521 skb_set_transport_header(skb, offset);
522
523 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source,
524 &hdr->saddr, uh->dest,
525 inet6_iif(skb), 0, udptable, skb);
526 if (sk) {
527 int (*lookup)(struct sock *sk, struct sk_buff *skb);
528 struct udp_sock *up = udp_sk(sk);
529
530 lookup = READ_ONCE(up->encap_err_lookup);
531 if (!lookup || lookup(sk, skb))
532 sk = NULL;
533 }
534
535 if (!sk) {
536 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code,
537 offset, info));
538 }
539
540 skb_set_transport_header(skb, transport_offset);
541 skb_set_network_header(skb, network_offset);
542
543 return sk;
544}
545
546int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
547 u8 type, u8 code, int offset, __be32 info,
548 struct udp_table *udptable)
549{
550 struct ipv6_pinfo *np;
551 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
552 const struct in6_addr *saddr = &hdr->saddr;
553 const struct in6_addr *daddr = &hdr->daddr;
554 struct udphdr *uh = (struct udphdr *)(skb->data+offset);
555 bool tunnel = false;
556 struct sock *sk;
557 int harderr;
558 int err;
559 struct net *net = dev_net(skb->dev);
560
561 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
562 inet6_iif(skb), inet6_sdif(skb), udptable, NULL);
563 if (!sk) {
564 /* No socket for error: try tunnels before discarding */
565 sk = ERR_PTR(-ENOENT);
566 if (static_branch_unlikely(&udpv6_encap_needed_key)) {
567 sk = __udp6_lib_err_encap(net, hdr, offset, uh,
568 udptable, skb,
569 opt, type, code, info);
570 if (!sk)
571 return 0;
572 }
573
574 if (IS_ERR(sk)) {
575 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
576 ICMP6_MIB_INERRORS);
577 return PTR_ERR(sk);
578 }
579
580 tunnel = true;
581 }
582
583 harderr = icmpv6_err_convert(type, code, &err);
584 np = inet6_sk(sk);
585
586 if (type == ICMPV6_PKT_TOOBIG) {
587 if (!ip6_sk_accept_pmtu(sk))
588 goto out;
589 ip6_sk_update_pmtu(skb, sk, info);
590 if (np->pmtudisc != IPV6_PMTUDISC_DONT)
591 harderr = 1;
592 }
593 if (type == NDISC_REDIRECT) {
594 if (tunnel) {
595 ip6_redirect(skb, sock_net(sk), inet6_iif(skb),
596 sk->sk_mark, sk->sk_uid);
597 } else {
598 ip6_sk_redirect(skb, sk);
599 }
600 goto out;
601 }
602
603 /* Tunnels don't have an application socket: don't pass errors back */
604 if (tunnel)
605 goto out;
606
607 if (!np->recverr) {
608 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
609 goto out;
610 } else {
611 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
612 }
613
614 sk->sk_err = err;
615 sk->sk_error_report(sk);
616out:
617 return 0;
618}
619
620static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
621{
622 int rc;
623
624 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
625 sock_rps_save_rxhash(sk, skb);
626 sk_mark_napi_id(sk, skb);
627 sk_incoming_cpu_update(sk);
628 } else {
629 sk_mark_napi_id_once(sk, skb);
630 }
631
632 rc = __udp_enqueue_schedule_skb(sk, skb);
633 if (rc < 0) {
634 int is_udplite = IS_UDPLITE(sk);
635
636 /* Note that an ENOMEM error is charged twice */
637 if (rc == -ENOMEM)
638 UDP6_INC_STATS(sock_net(sk),
639 UDP_MIB_RCVBUFERRORS, is_udplite);
640 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
641 kfree_skb(skb);
642 return -1;
643 }
644
645 return 0;
646}
647
648static __inline__ int udpv6_err(struct sk_buff *skb,
649 struct inet6_skb_parm *opt, u8 type,
650 u8 code, int offset, __be32 info)
651{
652 return __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
653}
654
655static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
656{
657 struct udp_sock *up = udp_sk(sk);
658 int is_udplite = IS_UDPLITE(sk);
659
660 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
661 goto drop;
662
663 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) {
664 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
665
666 /*
667 * This is an encapsulation socket so pass the skb to
668 * the socket's udp_encap_rcv() hook. Otherwise, just
669 * fall through and pass this up the UDP socket.
670 * up->encap_rcv() returns the following value:
671 * =0 if skb was successfully passed to the encap
672 * handler or was discarded by it.
673 * >0 if skb should be passed on to UDP.
674 * <0 if skb should be resubmitted as proto -N
675 */
676
677 /* if we're overly short, let UDP handle it */
678 encap_rcv = READ_ONCE(up->encap_rcv);
679 if (encap_rcv) {
680 int ret;
681
682 /* Verify checksum before giving to encap */
683 if (udp_lib_checksum_complete(skb))
684 goto csum_error;
685
686 ret = encap_rcv(sk, skb);
687 if (ret <= 0) {
688 __UDP_INC_STATS(sock_net(sk),
689 UDP_MIB_INDATAGRAMS,
690 is_udplite);
691 return -ret;
692 }
693 }
694
695 /* FALLTHROUGH -- it's a UDP Packet */
696 }
697
698 /*
699 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
700 */
701 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
702
703 if (up->pcrlen == 0) { /* full coverage was set */
704 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
705 UDP_SKB_CB(skb)->cscov, skb->len);
706 goto drop;
707 }
708 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
709 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
710 UDP_SKB_CB(skb)->cscov, up->pcrlen);
711 goto drop;
712 }
713 }
714
715 prefetch(&sk->sk_rmem_alloc);
716 if (rcu_access_pointer(sk->sk_filter) &&
717 udp_lib_checksum_complete(skb))
718 goto csum_error;
719
720 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
721 goto drop;
722
723 udp_csum_pull_header(skb);
724
725 skb_dst_drop(skb);
726
727 return __udpv6_queue_rcv_skb(sk, skb);
728
729csum_error:
730 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
731drop:
732 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
733 atomic_inc(&sk->sk_drops);
734 kfree_skb(skb);
735 return -1;
736}
737
738static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
739{
740 struct sk_buff *next, *segs;
741 int ret;
742
743 if (likely(!udp_unexpected_gso(sk, skb)))
744 return udpv6_queue_rcv_one_skb(sk, skb);
745
746 __skb_push(skb, -skb_mac_offset(skb));
747 segs = udp_rcv_segment(sk, skb, false);
748 skb_list_walk_safe(segs, skb, next) {
749 __skb_pull(skb, skb_transport_offset(skb));
750
751 ret = udpv6_queue_rcv_one_skb(sk, skb);
752 if (ret > 0)
753 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret,
754 true);
755 }
756 return 0;
757}
758
759static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
760 __be16 loc_port, const struct in6_addr *loc_addr,
761 __be16 rmt_port, const struct in6_addr *rmt_addr,
762 int dif, int sdif, unsigned short hnum)
763{
764 struct inet_sock *inet = inet_sk(sk);
765
766 if (!net_eq(sock_net(sk), net))
767 return false;
768
769 if (udp_sk(sk)->udp_port_hash != hnum ||
770 sk->sk_family != PF_INET6 ||
771 (inet->inet_dport && inet->inet_dport != rmt_port) ||
772 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
773 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
774 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif) ||
775 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
776 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
777 return false;
778 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
779 return false;
780 return true;
781}
782
783static void udp6_csum_zero_error(struct sk_buff *skb)
784{
785 /* RFC 2460 section 8.1 says that we SHOULD log
786 * this error. Well, it is reasonable.
787 */
788 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
789 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
790 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
791}
792
793/*
794 * Note: called only from the BH handler context,
795 * so we don't need to lock the hashes.
796 */
797static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
798 const struct in6_addr *saddr, const struct in6_addr *daddr,
799 struct udp_table *udptable, int proto)
800{
801 struct sock *sk, *first = NULL;
802 const struct udphdr *uh = udp_hdr(skb);
803 unsigned short hnum = ntohs(uh->dest);
804 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
805 unsigned int offset = offsetof(typeof(*sk), sk_node);
806 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
807 int dif = inet6_iif(skb);
808 int sdif = inet6_sdif(skb);
809 struct hlist_node *node;
810 struct sk_buff *nskb;
811
812 if (use_hash2) {
813 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) &
814 udptable->mask;
815 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask;
816start_lookup:
817 hslot = &udptable->hash2[hash2];
818 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
819 }
820
821 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
822 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
823 uh->source, saddr, dif, sdif,
824 hnum))
825 continue;
826 /* If zero checksum and no_check is not on for
827 * the socket then skip it.
828 */
829 if (!uh->check && !udp_sk(sk)->no_check6_rx)
830 continue;
831 if (!first) {
832 first = sk;
833 continue;
834 }
835 nskb = skb_clone(skb, GFP_ATOMIC);
836 if (unlikely(!nskb)) {
837 atomic_inc(&sk->sk_drops);
838 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
839 IS_UDPLITE(sk));
840 __UDP6_INC_STATS(net, UDP_MIB_INERRORS,
841 IS_UDPLITE(sk));
842 continue;
843 }
844
845 if (udpv6_queue_rcv_skb(sk, nskb) > 0)
846 consume_skb(nskb);
847 }
848
849 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
850 if (use_hash2 && hash2 != hash2_any) {
851 hash2 = hash2_any;
852 goto start_lookup;
853 }
854
855 if (first) {
856 if (udpv6_queue_rcv_skb(first, skb) > 0)
857 consume_skb(skb);
858 } else {
859 kfree_skb(skb);
860 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
861 proto == IPPROTO_UDPLITE);
862 }
863 return 0;
864}
865
866static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
867{
868 if (udp_sk_rx_dst_set(sk, dst)) {
869 const struct rt6_info *rt = (const struct rt6_info *)dst;
870
871 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
872 }
873}
874
875/* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
876 * return code conversion for ip layer consumption
877 */
878static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
879 struct udphdr *uh)
880{
881 int ret;
882
883 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
884 skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo);
885
886 ret = udpv6_queue_rcv_skb(sk, skb);
887
888 /* a return value > 0 means to resubmit the input */
889 if (ret > 0)
890 return ret;
891 return 0;
892}
893
894int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
895 int proto)
896{
897 const struct in6_addr *saddr, *daddr;
898 struct net *net = dev_net(skb->dev);
899 struct udphdr *uh;
900 struct sock *sk;
901 bool refcounted;
902 u32 ulen = 0;
903
904 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
905 goto discard;
906
907 saddr = &ipv6_hdr(skb)->saddr;
908 daddr = &ipv6_hdr(skb)->daddr;
909 uh = udp_hdr(skb);
910
911 ulen = ntohs(uh->len);
912 if (ulen > skb->len)
913 goto short_packet;
914
915 if (proto == IPPROTO_UDP) {
916 /* UDP validates ulen. */
917
918 /* Check for jumbo payload */
919 if (ulen == 0)
920 ulen = skb->len;
921
922 if (ulen < sizeof(*uh))
923 goto short_packet;
924
925 if (ulen < skb->len) {
926 if (pskb_trim_rcsum(skb, ulen))
927 goto short_packet;
928 saddr = &ipv6_hdr(skb)->saddr;
929 daddr = &ipv6_hdr(skb)->daddr;
930 uh = udp_hdr(skb);
931 }
932 }
933
934 if (udp6_csum_init(skb, uh, proto))
935 goto csum_error;
936
937 /* Check if the socket is already available, e.g. due to early demux */
938 sk = skb_steal_sock(skb, &refcounted);
939 if (sk) {
940 struct dst_entry *dst = skb_dst(skb);
941 int ret;
942
943 if (unlikely(sk->sk_rx_dst != dst))
944 udp6_sk_rx_dst_set(sk, dst);
945
946 if (!uh->check && !udp_sk(sk)->no_check6_rx) {
947 if (refcounted)
948 sock_put(sk);
949 goto report_csum_error;
950 }
951
952 ret = udp6_unicast_rcv_skb(sk, skb, uh);
953 if (refcounted)
954 sock_put(sk);
955 return ret;
956 }
957
958 /*
959 * Multicast receive code
960 */
961 if (ipv6_addr_is_multicast(daddr))
962 return __udp6_lib_mcast_deliver(net, skb,
963 saddr, daddr, udptable, proto);
964
965 /* Unicast */
966 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
967 if (sk) {
968 if (!uh->check && !udp_sk(sk)->no_check6_rx)
969 goto report_csum_error;
970 return udp6_unicast_rcv_skb(sk, skb, uh);
971 }
972
973 if (!uh->check)
974 goto report_csum_error;
975
976 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
977 goto discard;
978
979 if (udp_lib_checksum_complete(skb))
980 goto csum_error;
981
982 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
983 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
984
985 kfree_skb(skb);
986 return 0;
987
988short_packet:
989 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
990 proto == IPPROTO_UDPLITE ? "-Lite" : "",
991 saddr, ntohs(uh->source),
992 ulen, skb->len,
993 daddr, ntohs(uh->dest));
994 goto discard;
995
996report_csum_error:
997 udp6_csum_zero_error(skb);
998csum_error:
999 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1000discard:
1001 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1002 kfree_skb(skb);
1003 return 0;
1004}
1005
1006
1007static struct sock *__udp6_lib_demux_lookup(struct net *net,
1008 __be16 loc_port, const struct in6_addr *loc_addr,
1009 __be16 rmt_port, const struct in6_addr *rmt_addr,
1010 int dif, int sdif)
1011{
1012 unsigned short hnum = ntohs(loc_port);
1013 unsigned int hash2 = ipv6_portaddr_hash(net, loc_addr, hnum);
1014 unsigned int slot2 = hash2 & udp_table.mask;
1015 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
1016 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
1017 struct sock *sk;
1018
1019 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
1020 if (sk->sk_state == TCP_ESTABLISHED &&
1021 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif, sdif))
1022 return sk;
1023 /* Only check first socket in chain */
1024 break;
1025 }
1026 return NULL;
1027}
1028
1029INDIRECT_CALLABLE_SCOPE void udp_v6_early_demux(struct sk_buff *skb)
1030{
1031 struct net *net = dev_net(skb->dev);
1032 const struct udphdr *uh;
1033 struct sock *sk;
1034 struct dst_entry *dst;
1035 int dif = skb->dev->ifindex;
1036 int sdif = inet6_sdif(skb);
1037
1038 if (!pskb_may_pull(skb, skb_transport_offset(skb) +
1039 sizeof(struct udphdr)))
1040 return;
1041
1042 uh = udp_hdr(skb);
1043
1044 if (skb->pkt_type == PACKET_HOST)
1045 sk = __udp6_lib_demux_lookup(net, uh->dest,
1046 &ipv6_hdr(skb)->daddr,
1047 uh->source, &ipv6_hdr(skb)->saddr,
1048 dif, sdif);
1049 else
1050 return;
1051
1052 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
1053 return;
1054
1055 skb->sk = sk;
1056 skb->destructor = sock_efree;
1057 dst = READ_ONCE(sk->sk_rx_dst);
1058
1059 if (dst)
1060 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1061 if (dst) {
1062 /* set noref for now.
1063 * any place which wants to hold dst has to call
1064 * dst_hold_safe()
1065 */
1066 skb_dst_set_noref(skb, dst);
1067 }
1068}
1069
1070INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb)
1071{
1072 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1073}
1074
1075/*
1076 * Throw away all pending data and cancel the corking. Socket is locked.
1077 */
1078static void udp_v6_flush_pending_frames(struct sock *sk)
1079{
1080 struct udp_sock *up = udp_sk(sk);
1081
1082 if (up->pending == AF_INET)
1083 udp_flush_pending_frames(sk);
1084 else if (up->pending) {
1085 up->len = 0;
1086 up->pending = 0;
1087 ip6_flush_pending_frames(sk);
1088 }
1089}
1090
1091static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
1092 int addr_len)
1093{
1094 if (addr_len < offsetofend(struct sockaddr, sa_family))
1095 return -EINVAL;
1096 /* The following checks are replicated from __ip6_datagram_connect()
1097 * and intended to prevent BPF program called below from accessing
1098 * bytes that are out of the bound specified by user in addr_len.
1099 */
1100 if (uaddr->sa_family == AF_INET) {
1101 if (__ipv6_only_sock(sk))
1102 return -EAFNOSUPPORT;
1103 return udp_pre_connect(sk, uaddr, addr_len);
1104 }
1105
1106 if (addr_len < SIN6_LEN_RFC2133)
1107 return -EINVAL;
1108
1109 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr);
1110}
1111
1112/**
1113 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
1114 * @sk: socket we are sending on
1115 * @skb: sk_buff containing the filled-in UDP header
1116 * (checksum field must be zeroed out)
1117 * @saddr: source address
1118 * @daddr: destination address
1119 * @len: length of packet
1120 */
1121static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
1122 const struct in6_addr *saddr,
1123 const struct in6_addr *daddr, int len)
1124{
1125 unsigned int offset;
1126 struct udphdr *uh = udp_hdr(skb);
1127 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
1128 __wsum csum = 0;
1129
1130 if (!frags) {
1131 /* Only one fragment on the socket. */
1132 skb->csum_start = skb_transport_header(skb) - skb->head;
1133 skb->csum_offset = offsetof(struct udphdr, check);
1134 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1135 } else {
1136 /*
1137 * HW-checksum won't work as there are two or more
1138 * fragments on the socket so that all csums of sk_buffs
1139 * should be together
1140 */
1141 offset = skb_transport_offset(skb);
1142 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1143 csum = skb->csum;
1144
1145 skb->ip_summed = CHECKSUM_NONE;
1146
1147 do {
1148 csum = csum_add(csum, frags->csum);
1149 } while ((frags = frags->next));
1150
1151 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1152 csum);
1153 if (uh->check == 0)
1154 uh->check = CSUM_MANGLED_0;
1155 }
1156}
1157
1158/*
1159 * Sending
1160 */
1161
1162static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6,
1163 struct inet_cork *cork)
1164{
1165 struct sock *sk = skb->sk;
1166 struct udphdr *uh;
1167 int err = 0;
1168 int is_udplite = IS_UDPLITE(sk);
1169 __wsum csum = 0;
1170 int offset = skb_transport_offset(skb);
1171 int len = skb->len - offset;
1172 int datalen = len - sizeof(*uh);
1173
1174 /*
1175 * Create a UDP header
1176 */
1177 uh = udp_hdr(skb);
1178 uh->source = fl6->fl6_sport;
1179 uh->dest = fl6->fl6_dport;
1180 uh->len = htons(len);
1181 uh->check = 0;
1182
1183 if (cork->gso_size) {
1184 const int hlen = skb_network_header_len(skb) +
1185 sizeof(struct udphdr);
1186
1187 if (hlen + cork->gso_size > cork->fragsize) {
1188 kfree_skb(skb);
1189 return -EINVAL;
1190 }
1191 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) {
1192 kfree_skb(skb);
1193 return -EINVAL;
1194 }
1195 if (udp_sk(sk)->no_check6_tx) {
1196 kfree_skb(skb);
1197 return -EINVAL;
1198 }
1199 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
1200 dst_xfrm(skb_dst(skb))) {
1201 kfree_skb(skb);
1202 return -EIO;
1203 }
1204
1205 if (datalen > cork->gso_size) {
1206 skb_shinfo(skb)->gso_size = cork->gso_size;
1207 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
1208 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
1209 cork->gso_size);
1210 }
1211 goto csum_partial;
1212 }
1213
1214 if (is_udplite)
1215 csum = udplite_csum(skb);
1216 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */
1217 skb->ip_summed = CHECKSUM_NONE;
1218 goto send;
1219 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1220csum_partial:
1221 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1222 goto send;
1223 } else
1224 csum = udp_csum(skb);
1225
1226 /* add protocol-dependent pseudo-header */
1227 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1228 len, fl6->flowi6_proto, csum);
1229 if (uh->check == 0)
1230 uh->check = CSUM_MANGLED_0;
1231
1232send:
1233 err = ip6_send_skb(skb);
1234 if (err) {
1235 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1236 UDP6_INC_STATS(sock_net(sk),
1237 UDP_MIB_SNDBUFERRORS, is_udplite);
1238 err = 0;
1239 }
1240 } else {
1241 UDP6_INC_STATS(sock_net(sk),
1242 UDP_MIB_OUTDATAGRAMS, is_udplite);
1243 }
1244 return err;
1245}
1246
1247static int udp_v6_push_pending_frames(struct sock *sk)
1248{
1249 struct sk_buff *skb;
1250 struct udp_sock *up = udp_sk(sk);
1251 struct flowi6 fl6;
1252 int err = 0;
1253
1254 if (up->pending == AF_INET)
1255 return udp_push_pending_frames(sk);
1256
1257 /* ip6_finish_skb will release the cork, so make a copy of
1258 * fl6 here.
1259 */
1260 fl6 = inet_sk(sk)->cork.fl.u.ip6;
1261
1262 skb = ip6_finish_skb(sk);
1263 if (!skb)
1264 goto out;
1265
1266 err = udp_v6_send_skb(skb, &fl6, &inet_sk(sk)->cork.base);
1267
1268out:
1269 up->len = 0;
1270 up->pending = 0;
1271 return err;
1272}
1273
1274int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1275{
1276 struct ipv6_txoptions opt_space;
1277 struct udp_sock *up = udp_sk(sk);
1278 struct inet_sock *inet = inet_sk(sk);
1279 struct ipv6_pinfo *np = inet6_sk(sk);
1280 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1281 struct in6_addr *daddr, *final_p, final;
1282 struct ipv6_txoptions *opt = NULL;
1283 struct ipv6_txoptions *opt_to_free = NULL;
1284 struct ip6_flowlabel *flowlabel = NULL;
1285 struct flowi6 fl6;
1286 struct dst_entry *dst;
1287 struct ipcm6_cookie ipc6;
1288 int addr_len = msg->msg_namelen;
1289 bool connected = false;
1290 int ulen = len;
1291 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1292 int err;
1293 int is_udplite = IS_UDPLITE(sk);
1294 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1295
1296 ipcm6_init(&ipc6);
1297 ipc6.gso_size = up->gso_size;
1298 ipc6.sockc.tsflags = sk->sk_tsflags;
1299 ipc6.sockc.mark = sk->sk_mark;
1300
1301 /* destination address check */
1302 if (sin6) {
1303 if (addr_len < offsetof(struct sockaddr, sa_data))
1304 return -EINVAL;
1305
1306 switch (sin6->sin6_family) {
1307 case AF_INET6:
1308 if (addr_len < SIN6_LEN_RFC2133)
1309 return -EINVAL;
1310 daddr = &sin6->sin6_addr;
1311 if (ipv6_addr_any(daddr) &&
1312 ipv6_addr_v4mapped(&np->saddr))
1313 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1314 daddr);
1315 break;
1316 case AF_INET:
1317 goto do_udp_sendmsg;
1318 case AF_UNSPEC:
1319 msg->msg_name = sin6 = NULL;
1320 msg->msg_namelen = addr_len = 0;
1321 daddr = NULL;
1322 break;
1323 default:
1324 return -EINVAL;
1325 }
1326 } else if (!up->pending) {
1327 if (sk->sk_state != TCP_ESTABLISHED)
1328 return -EDESTADDRREQ;
1329 daddr = &sk->sk_v6_daddr;
1330 } else
1331 daddr = NULL;
1332
1333 if (daddr) {
1334 if (ipv6_addr_v4mapped(daddr)) {
1335 struct sockaddr_in sin;
1336 sin.sin_family = AF_INET;
1337 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1338 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1339 msg->msg_name = &sin;
1340 msg->msg_namelen = sizeof(sin);
1341do_udp_sendmsg:
1342 if (__ipv6_only_sock(sk))
1343 return -ENETUNREACH;
1344 return udp_sendmsg(sk, msg, len);
1345 }
1346 }
1347
1348 if (up->pending == AF_INET)
1349 return udp_sendmsg(sk, msg, len);
1350
1351 /* Rough check on arithmetic overflow,
1352 better check is made in ip6_append_data().
1353 */
1354 if (len > INT_MAX - sizeof(struct udphdr))
1355 return -EMSGSIZE;
1356
1357 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1358 if (up->pending) {
1359 /*
1360 * There are pending frames.
1361 * The socket lock must be held while it's corked.
1362 */
1363 lock_sock(sk);
1364 if (likely(up->pending)) {
1365 if (unlikely(up->pending != AF_INET6)) {
1366 release_sock(sk);
1367 return -EAFNOSUPPORT;
1368 }
1369 dst = NULL;
1370 goto do_append_data;
1371 }
1372 release_sock(sk);
1373 }
1374 ulen += sizeof(struct udphdr);
1375
1376 memset(&fl6, 0, sizeof(fl6));
1377
1378 if (sin6) {
1379 if (sin6->sin6_port == 0)
1380 return -EINVAL;
1381
1382 fl6.fl6_dport = sin6->sin6_port;
1383 daddr = &sin6->sin6_addr;
1384
1385 if (np->sndflow) {
1386 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1387 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1388 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1389 if (IS_ERR(flowlabel))
1390 return -EINVAL;
1391 }
1392 }
1393
1394 /*
1395 * Otherwise it will be difficult to maintain
1396 * sk->sk_dst_cache.
1397 */
1398 if (sk->sk_state == TCP_ESTABLISHED &&
1399 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1400 daddr = &sk->sk_v6_daddr;
1401
1402 if (addr_len >= sizeof(struct sockaddr_in6) &&
1403 sin6->sin6_scope_id &&
1404 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1405 fl6.flowi6_oif = sin6->sin6_scope_id;
1406 } else {
1407 if (sk->sk_state != TCP_ESTABLISHED)
1408 return -EDESTADDRREQ;
1409
1410 fl6.fl6_dport = inet->inet_dport;
1411 daddr = &sk->sk_v6_daddr;
1412 fl6.flowlabel = np->flow_label;
1413 connected = true;
1414 }
1415
1416 if (!fl6.flowi6_oif)
1417 fl6.flowi6_oif = sk->sk_bound_dev_if;
1418
1419 if (!fl6.flowi6_oif)
1420 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1421
1422 fl6.flowi6_mark = ipc6.sockc.mark;
1423 fl6.flowi6_uid = sk->sk_uid;
1424
1425 if (msg->msg_controllen) {
1426 opt = &opt_space;
1427 memset(opt, 0, sizeof(struct ipv6_txoptions));
1428 opt->tot_len = sizeof(*opt);
1429 ipc6.opt = opt;
1430
1431 err = udp_cmsg_send(sk, msg, &ipc6.gso_size);
1432 if (err > 0)
1433 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6,
1434 &ipc6);
1435 if (err < 0) {
1436 fl6_sock_release(flowlabel);
1437 return err;
1438 }
1439 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1440 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1441 if (IS_ERR(flowlabel))
1442 return -EINVAL;
1443 }
1444 if (!(opt->opt_nflen|opt->opt_flen))
1445 opt = NULL;
1446 connected = false;
1447 }
1448 if (!opt) {
1449 opt = txopt_get(np);
1450 opt_to_free = opt;
1451 }
1452 if (flowlabel)
1453 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1454 opt = ipv6_fixup_options(&opt_space, opt);
1455 ipc6.opt = opt;
1456
1457 fl6.flowi6_proto = sk->sk_protocol;
1458 fl6.daddr = *daddr;
1459 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1460 fl6.saddr = np->saddr;
1461 fl6.fl6_sport = inet->inet_sport;
1462
1463 if (cgroup_bpf_enabled && !connected) {
1464 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk,
1465 (struct sockaddr *)sin6, &fl6.saddr);
1466 if (err)
1467 goto out_no_dst;
1468 if (sin6) {
1469 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
1470 /* BPF program rewrote IPv6-only by IPv4-mapped
1471 * IPv6. It's currently unsupported.
1472 */
1473 err = -ENOTSUPP;
1474 goto out_no_dst;
1475 }
1476 if (sin6->sin6_port == 0) {
1477 /* BPF program set invalid port. Reject it. */
1478 err = -EINVAL;
1479 goto out_no_dst;
1480 }
1481 fl6.fl6_dport = sin6->sin6_port;
1482 fl6.daddr = sin6->sin6_addr;
1483 }
1484 }
1485
1486 if (ipv6_addr_any(&fl6.daddr))
1487 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1488
1489 final_p = fl6_update_dst(&fl6, opt, &final);
1490 if (final_p)
1491 connected = false;
1492
1493 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1494 fl6.flowi6_oif = np->mcast_oif;
1495 connected = false;
1496 } else if (!fl6.flowi6_oif)
1497 fl6.flowi6_oif = np->ucast_oif;
1498
1499 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1500
1501 if (ipc6.tclass < 0)
1502 ipc6.tclass = np->tclass;
1503
1504 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel);
1505
1506 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, connected);
1507 if (IS_ERR(dst)) {
1508 err = PTR_ERR(dst);
1509 dst = NULL;
1510 goto out;
1511 }
1512
1513 if (ipc6.hlimit < 0)
1514 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1515
1516 if (msg->msg_flags&MSG_CONFIRM)
1517 goto do_confirm;
1518back_from_confirm:
1519
1520 /* Lockless fast path for the non-corking case */
1521 if (!corkreq) {
1522 struct inet_cork_full cork;
1523 struct sk_buff *skb;
1524
1525 skb = ip6_make_skb(sk, getfrag, msg, ulen,
1526 sizeof(struct udphdr), &ipc6,
1527 &fl6, (struct rt6_info *)dst,
1528 msg->msg_flags, &cork);
1529 err = PTR_ERR(skb);
1530 if (!IS_ERR_OR_NULL(skb))
1531 err = udp_v6_send_skb(skb, &fl6, &cork.base);
1532 goto out;
1533 }
1534
1535 lock_sock(sk);
1536 if (unlikely(up->pending)) {
1537 /* The socket is already corked while preparing it. */
1538 /* ... which is an evident application bug. --ANK */
1539 release_sock(sk);
1540
1541 net_dbg_ratelimited("udp cork app bug 2\n");
1542 err = -EINVAL;
1543 goto out;
1544 }
1545
1546 up->pending = AF_INET6;
1547
1548do_append_data:
1549 if (ipc6.dontfrag < 0)
1550 ipc6.dontfrag = np->dontfrag;
1551 up->len += ulen;
1552 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr),
1553 &ipc6, &fl6, (struct rt6_info *)dst,
1554 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1555 if (err)
1556 udp_v6_flush_pending_frames(sk);
1557 else if (!corkreq)
1558 err = udp_v6_push_pending_frames(sk);
1559 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1560 up->pending = 0;
1561
1562 if (err > 0)
1563 err = np->recverr ? net_xmit_errno(err) : 0;
1564 release_sock(sk);
1565
1566out:
1567 dst_release(dst);
1568out_no_dst:
1569 fl6_sock_release(flowlabel);
1570 txopt_put(opt_to_free);
1571 if (!err)
1572 return len;
1573 /*
1574 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1575 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1576 * we don't have a good statistic (IpOutDiscards but it can be too many
1577 * things). We could add another new stat but at least for now that
1578 * seems like overkill.
1579 */
1580 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1581 UDP6_INC_STATS(sock_net(sk),
1582 UDP_MIB_SNDBUFERRORS, is_udplite);
1583 }
1584 return err;
1585
1586do_confirm:
1587 if (msg->msg_flags & MSG_PROBE)
1588 dst_confirm_neigh(dst, &fl6.daddr);
1589 if (!(msg->msg_flags&MSG_PROBE) || len)
1590 goto back_from_confirm;
1591 err = 0;
1592 goto out;
1593}
1594
1595void udpv6_destroy_sock(struct sock *sk)
1596{
1597 struct udp_sock *up = udp_sk(sk);
1598 lock_sock(sk);
1599 udp_v6_flush_pending_frames(sk);
1600 release_sock(sk);
1601
1602 if (static_branch_unlikely(&udpv6_encap_needed_key)) {
1603 if (up->encap_type) {
1604 void (*encap_destroy)(struct sock *sk);
1605 encap_destroy = READ_ONCE(up->encap_destroy);
1606 if (encap_destroy)
1607 encap_destroy(sk);
1608 }
1609 if (up->encap_enabled)
1610 static_branch_dec(&udpv6_encap_needed_key);
1611 }
1612
1613 inet6_destroy_sock(sk);
1614}
1615
1616/*
1617 * Socket option code for UDP
1618 */
1619int udpv6_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
1620 unsigned int optlen)
1621{
1622 if (level == SOL_UDP || level == SOL_UDPLITE)
1623 return udp_lib_setsockopt(sk, level, optname,
1624 optval, optlen,
1625 udp_v6_push_pending_frames);
1626 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1627}
1628
1629int udpv6_getsockopt(struct sock *sk, int level, int optname,
1630 char __user *optval, int __user *optlen)
1631{
1632 if (level == SOL_UDP || level == SOL_UDPLITE)
1633 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1634 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1635}
1636
1637/* thinking of making this const? Don't.
1638 * early_demux can change based on sysctl.
1639 */
1640static struct inet6_protocol udpv6_protocol = {
1641 .early_demux = udp_v6_early_demux,
1642 .early_demux_handler = udp_v6_early_demux,
1643 .handler = udpv6_rcv,
1644 .err_handler = udpv6_err,
1645 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1646};
1647
1648/* ------------------------------------------------------------------------ */
1649#ifdef CONFIG_PROC_FS
1650int udp6_seq_show(struct seq_file *seq, void *v)
1651{
1652 if (v == SEQ_START_TOKEN) {
1653 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1654 } else {
1655 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1656 struct inet_sock *inet = inet_sk(v);
1657 __u16 srcp = ntohs(inet->inet_sport);
1658 __u16 destp = ntohs(inet->inet_dport);
1659 __ip6_dgram_sock_seq_show(seq, v, srcp, destp,
1660 udp_rqueue_get(v), bucket);
1661 }
1662 return 0;
1663}
1664
1665const struct seq_operations udp6_seq_ops = {
1666 .start = udp_seq_start,
1667 .next = udp_seq_next,
1668 .stop = udp_seq_stop,
1669 .show = udp6_seq_show,
1670};
1671EXPORT_SYMBOL(udp6_seq_ops);
1672
1673static struct udp_seq_afinfo udp6_seq_afinfo = {
1674 .family = AF_INET6,
1675 .udp_table = &udp_table,
1676};
1677
1678int __net_init udp6_proc_init(struct net *net)
1679{
1680 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops,
1681 sizeof(struct udp_iter_state), &udp6_seq_afinfo))
1682 return -ENOMEM;
1683 return 0;
1684}
1685
1686void udp6_proc_exit(struct net *net)
1687{
1688 remove_proc_entry("udp6", net->proc_net);
1689}
1690#endif /* CONFIG_PROC_FS */
1691
1692/* ------------------------------------------------------------------------ */
1693
1694struct proto udpv6_prot = {
1695 .name = "UDPv6",
1696 .owner = THIS_MODULE,
1697 .close = udp_lib_close,
1698 .pre_connect = udpv6_pre_connect,
1699 .connect = ip6_datagram_connect,
1700 .disconnect = udp_disconnect,
1701 .ioctl = udp_ioctl,
1702 .init = udp_init_sock,
1703 .destroy = udpv6_destroy_sock,
1704 .setsockopt = udpv6_setsockopt,
1705 .getsockopt = udpv6_getsockopt,
1706 .sendmsg = udpv6_sendmsg,
1707 .recvmsg = udpv6_recvmsg,
1708 .release_cb = ip6_datagram_release_cb,
1709 .hash = udp_lib_hash,
1710 .unhash = udp_lib_unhash,
1711 .rehash = udp_v6_rehash,
1712 .get_port = udp_v6_get_port,
1713 .memory_allocated = &udp_memory_allocated,
1714 .sysctl_mem = sysctl_udp_mem,
1715 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
1716 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
1717 .obj_size = sizeof(struct udp6_sock),
1718 .h.udp_table = &udp_table,
1719 .diag_destroy = udp_abort,
1720};
1721
1722static struct inet_protosw udpv6_protosw = {
1723 .type = SOCK_DGRAM,
1724 .protocol = IPPROTO_UDP,
1725 .prot = &udpv6_prot,
1726 .ops = &inet6_dgram_ops,
1727 .flags = INET_PROTOSW_PERMANENT,
1728};
1729
1730int __init udpv6_init(void)
1731{
1732 int ret;
1733
1734 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1735 if (ret)
1736 goto out;
1737
1738 ret = inet6_register_protosw(&udpv6_protosw);
1739 if (ret)
1740 goto out_udpv6_protocol;
1741out:
1742 return ret;
1743
1744out_udpv6_protocol:
1745 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1746 goto out;
1747}
1748
1749void udpv6_exit(void)
1750{
1751 inet6_unregister_protosw(&udpv6_protosw);
1752 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1753}
1/*
2 * UDP over IPv6
3 * Linux INET6 implementation
4 *
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
7 *
8 * Based on linux/ipv4/udp.c
9 *
10 * Fixes:
11 * Hideaki YOSHIFUJI : sin6_scope_id support
12 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
13 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
14 * a single port at the same time.
15 * Kazunori MIYAZAWA @USAGI: change process style to use ip6_append_data
16 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/udp6 to seq_file.
17 *
18 * This program is free software; you can redistribute it and/or
19 * modify it under the terms of the GNU General Public License
20 * as published by the Free Software Foundation; either version
21 * 2 of the License, or (at your option) any later version.
22 */
23
24#include <linux/errno.h>
25#include <linux/types.h>
26#include <linux/socket.h>
27#include <linux/sockios.h>
28#include <linux/net.h>
29#include <linux/in6.h>
30#include <linux/netdevice.h>
31#include <linux/if_arp.h>
32#include <linux/ipv6.h>
33#include <linux/icmpv6.h>
34#include <linux/init.h>
35#include <linux/module.h>
36#include <linux/skbuff.h>
37#include <linux/slab.h>
38#include <asm/uaccess.h>
39
40#include <net/ndisc.h>
41#include <net/protocol.h>
42#include <net/transp_v6.h>
43#include <net/ip6_route.h>
44#include <net/raw.h>
45#include <net/tcp_states.h>
46#include <net/ip6_checksum.h>
47#include <net/xfrm.h>
48#include <net/inet6_hashtables.h>
49#include <net/busy_poll.h>
50
51#include <linux/proc_fs.h>
52#include <linux/seq_file.h>
53#include <trace/events/skb.h>
54#include "udp_impl.h"
55
56static unsigned int udp6_ehashfn(struct net *net,
57 const struct in6_addr *laddr,
58 const u16 lport,
59 const struct in6_addr *faddr,
60 const __be16 fport)
61{
62 static u32 udp6_ehash_secret __read_mostly;
63 static u32 udp_ipv6_hash_secret __read_mostly;
64
65 u32 lhash, fhash;
66
67 net_get_random_once(&udp6_ehash_secret,
68 sizeof(udp6_ehash_secret));
69 net_get_random_once(&udp_ipv6_hash_secret,
70 sizeof(udp_ipv6_hash_secret));
71
72 lhash = (__force u32)laddr->s6_addr32[3];
73 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
74
75 return __inet6_ehashfn(lhash, lport, fhash, fport,
76 udp_ipv6_hash_secret + net_hash_mix(net));
77}
78
79int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2)
80{
81 const struct in6_addr *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2);
82 int sk_ipv6only = ipv6_only_sock(sk);
83 int sk2_ipv6only = inet_v6_ipv6only(sk2);
84 int addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
85 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
86
87 /* if both are mapped, treat as IPv4 */
88 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED)
89 return (!sk2_ipv6only &&
90 (!sk->sk_rcv_saddr || !sk2->sk_rcv_saddr ||
91 sk->sk_rcv_saddr == sk2->sk_rcv_saddr));
92
93 if (addr_type2 == IPV6_ADDR_ANY &&
94 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
95 return 1;
96
97 if (addr_type == IPV6_ADDR_ANY &&
98 !(sk_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
99 return 1;
100
101 if (sk2_rcv_saddr6 &&
102 ipv6_addr_equal(&sk->sk_v6_rcv_saddr, sk2_rcv_saddr6))
103 return 1;
104
105 return 0;
106}
107
108static unsigned int udp6_portaddr_hash(struct net *net,
109 const struct in6_addr *addr6,
110 unsigned int port)
111{
112 unsigned int hash, mix = net_hash_mix(net);
113
114 if (ipv6_addr_any(addr6))
115 hash = jhash_1word(0, mix);
116 else if (ipv6_addr_v4mapped(addr6))
117 hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix);
118 else
119 hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix);
120
121 return hash ^ port;
122}
123
124
125int udp_v6_get_port(struct sock *sk, unsigned short snum)
126{
127 unsigned int hash2_nulladdr =
128 udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
129 unsigned int hash2_partial =
130 udp6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
131
132 /* precompute partial secondary hash */
133 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
134 return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal, hash2_nulladdr);
135}
136
137static void udp_v6_rehash(struct sock *sk)
138{
139 u16 new_hash = udp6_portaddr_hash(sock_net(sk),
140 &sk->sk_v6_rcv_saddr,
141 inet_sk(sk)->inet_num);
142
143 udp_lib_rehash(sk, new_hash);
144}
145
146static inline int compute_score(struct sock *sk, struct net *net,
147 unsigned short hnum,
148 const struct in6_addr *saddr, __be16 sport,
149 const struct in6_addr *daddr, __be16 dport,
150 int dif)
151{
152 int score = -1;
153
154 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
155 sk->sk_family == PF_INET6) {
156 struct inet_sock *inet = inet_sk(sk);
157
158 score = 0;
159 if (inet->inet_dport) {
160 if (inet->inet_dport != sport)
161 return -1;
162 score++;
163 }
164 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
165 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
166 return -1;
167 score++;
168 }
169 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
170 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
171 return -1;
172 score++;
173 }
174 if (sk->sk_bound_dev_if) {
175 if (sk->sk_bound_dev_if != dif)
176 return -1;
177 score++;
178 }
179 }
180 return score;
181}
182
183#define SCORE2_MAX (1 + 1 + 1)
184static inline int compute_score2(struct sock *sk, struct net *net,
185 const struct in6_addr *saddr, __be16 sport,
186 const struct in6_addr *daddr, unsigned short hnum,
187 int dif)
188{
189 int score = -1;
190
191 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
192 sk->sk_family == PF_INET6) {
193 struct inet_sock *inet = inet_sk(sk);
194
195 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
196 return -1;
197 score = 0;
198 if (inet->inet_dport) {
199 if (inet->inet_dport != sport)
200 return -1;
201 score++;
202 }
203 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
204 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
205 return -1;
206 score++;
207 }
208 if (sk->sk_bound_dev_if) {
209 if (sk->sk_bound_dev_if != dif)
210 return -1;
211 score++;
212 }
213 }
214 return score;
215}
216
217
218/* called with read_rcu_lock() */
219static struct sock *udp6_lib_lookup2(struct net *net,
220 const struct in6_addr *saddr, __be16 sport,
221 const struct in6_addr *daddr, unsigned int hnum, int dif,
222 struct udp_hslot *hslot2, unsigned int slot2)
223{
224 struct sock *sk, *result;
225 struct hlist_nulls_node *node;
226 int score, badness, matches = 0, reuseport = 0;
227 u32 hash = 0;
228
229begin:
230 result = NULL;
231 badness = -1;
232 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
233 score = compute_score2(sk, net, saddr, sport,
234 daddr, hnum, dif);
235 if (score > badness) {
236 result = sk;
237 badness = score;
238 reuseport = sk->sk_reuseport;
239 if (reuseport) {
240 hash = udp6_ehashfn(net, daddr, hnum,
241 saddr, sport);
242 matches = 1;
243 } else if (score == SCORE2_MAX)
244 goto exact_match;
245 } else if (score == badness && reuseport) {
246 matches++;
247 if (((u64)hash * matches) >> 32 == 0)
248 result = sk;
249 hash = next_pseudo_random32(hash);
250 }
251 }
252 /*
253 * if the nulls value we got at the end of this lookup is
254 * not the expected one, we must restart lookup.
255 * We probably met an item that was moved to another chain.
256 */
257 if (get_nulls_value(node) != slot2)
258 goto begin;
259
260 if (result) {
261exact_match:
262 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
263 result = NULL;
264 else if (unlikely(compute_score2(result, net, saddr, sport,
265 daddr, hnum, dif) < badness)) {
266 sock_put(result);
267 goto begin;
268 }
269 }
270 return result;
271}
272
273struct sock *__udp6_lib_lookup(struct net *net,
274 const struct in6_addr *saddr, __be16 sport,
275 const struct in6_addr *daddr, __be16 dport,
276 int dif, struct udp_table *udptable)
277{
278 struct sock *sk, *result;
279 struct hlist_nulls_node *node;
280 unsigned short hnum = ntohs(dport);
281 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
282 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
283 int score, badness, matches = 0, reuseport = 0;
284 u32 hash = 0;
285
286 rcu_read_lock();
287 if (hslot->count > 10) {
288 hash2 = udp6_portaddr_hash(net, daddr, hnum);
289 slot2 = hash2 & udptable->mask;
290 hslot2 = &udptable->hash2[slot2];
291 if (hslot->count < hslot2->count)
292 goto begin;
293
294 result = udp6_lib_lookup2(net, saddr, sport,
295 daddr, hnum, dif,
296 hslot2, slot2);
297 if (!result) {
298 hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
299 slot2 = hash2 & udptable->mask;
300 hslot2 = &udptable->hash2[slot2];
301 if (hslot->count < hslot2->count)
302 goto begin;
303
304 result = udp6_lib_lookup2(net, saddr, sport,
305 &in6addr_any, hnum, dif,
306 hslot2, slot2);
307 }
308 rcu_read_unlock();
309 return result;
310 }
311begin:
312 result = NULL;
313 badness = -1;
314 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
315 score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif);
316 if (score > badness) {
317 result = sk;
318 badness = score;
319 reuseport = sk->sk_reuseport;
320 if (reuseport) {
321 hash = udp6_ehashfn(net, daddr, hnum,
322 saddr, sport);
323 matches = 1;
324 }
325 } else if (score == badness && reuseport) {
326 matches++;
327 if (((u64)hash * matches) >> 32 == 0)
328 result = sk;
329 hash = next_pseudo_random32(hash);
330 }
331 }
332 /*
333 * if the nulls value we got at the end of this lookup is
334 * not the expected one, we must restart lookup.
335 * We probably met an item that was moved to another chain.
336 */
337 if (get_nulls_value(node) != slot)
338 goto begin;
339
340 if (result) {
341 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
342 result = NULL;
343 else if (unlikely(compute_score(result, net, hnum, saddr, sport,
344 daddr, dport, dif) < badness)) {
345 sock_put(result);
346 goto begin;
347 }
348 }
349 rcu_read_unlock();
350 return result;
351}
352EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
353
354static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
355 __be16 sport, __be16 dport,
356 struct udp_table *udptable)
357{
358 struct sock *sk;
359 const struct ipv6hdr *iph = ipv6_hdr(skb);
360
361 if (unlikely(sk = skb_steal_sock(skb)))
362 return sk;
363 return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport,
364 &iph->daddr, dport, inet6_iif(skb),
365 udptable);
366}
367
368struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
369 const struct in6_addr *daddr, __be16 dport, int dif)
370{
371 return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
372}
373EXPORT_SYMBOL_GPL(udp6_lib_lookup);
374
375
376/*
377 * This should be easy, if there is something there we
378 * return it, otherwise we block.
379 */
380
381int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk,
382 struct msghdr *msg, size_t len,
383 int noblock, int flags, int *addr_len)
384{
385 struct ipv6_pinfo *np = inet6_sk(sk);
386 struct inet_sock *inet = inet_sk(sk);
387 struct sk_buff *skb;
388 unsigned int ulen, copied;
389 int peeked, off = 0;
390 int err;
391 int is_udplite = IS_UDPLITE(sk);
392 int is_udp4;
393 bool slow;
394
395 if (flags & MSG_ERRQUEUE)
396 return ipv6_recv_error(sk, msg, len, addr_len);
397
398 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
399 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
400
401try_again:
402 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
403 &peeked, &off, &err);
404 if (!skb)
405 goto out;
406
407 ulen = skb->len - sizeof(struct udphdr);
408 copied = len;
409 if (copied > ulen)
410 copied = ulen;
411 else if (copied < ulen)
412 msg->msg_flags |= MSG_TRUNC;
413
414 is_udp4 = (skb->protocol == htons(ETH_P_IP));
415
416 /*
417 * If checksum is needed at all, try to do it while copying the
418 * data. If the data is truncated, or if we only want a partial
419 * coverage checksum (UDP-Lite), do it before the copy.
420 */
421
422 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
423 if (udp_lib_checksum_complete(skb))
424 goto csum_copy_err;
425 }
426
427 if (skb_csum_unnecessary(skb))
428 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
429 msg->msg_iov, copied);
430 else {
431 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
432 if (err == -EINVAL)
433 goto csum_copy_err;
434 }
435 if (unlikely(err)) {
436 trace_kfree_skb(skb, udpv6_recvmsg);
437 if (!peeked) {
438 atomic_inc(&sk->sk_drops);
439 if (is_udp4)
440 UDP_INC_STATS_USER(sock_net(sk),
441 UDP_MIB_INERRORS,
442 is_udplite);
443 else
444 UDP6_INC_STATS_USER(sock_net(sk),
445 UDP_MIB_INERRORS,
446 is_udplite);
447 }
448 goto out_free;
449 }
450 if (!peeked) {
451 if (is_udp4)
452 UDP_INC_STATS_USER(sock_net(sk),
453 UDP_MIB_INDATAGRAMS, is_udplite);
454 else
455 UDP6_INC_STATS_USER(sock_net(sk),
456 UDP_MIB_INDATAGRAMS, is_udplite);
457 }
458
459 sock_recv_ts_and_drops(msg, sk, skb);
460
461 /* Copy the address. */
462 if (msg->msg_name) {
463 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
464 sin6->sin6_family = AF_INET6;
465 sin6->sin6_port = udp_hdr(skb)->source;
466 sin6->sin6_flowinfo = 0;
467
468 if (is_udp4) {
469 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
470 &sin6->sin6_addr);
471 sin6->sin6_scope_id = 0;
472 } else {
473 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
474 sin6->sin6_scope_id =
475 ipv6_iface_scope_id(&sin6->sin6_addr,
476 IP6CB(skb)->iif);
477 }
478 *addr_len = sizeof(*sin6);
479 }
480
481 if (np->rxopt.all)
482 ip6_datagram_recv_common_ctl(sk, msg, skb);
483
484 if (is_udp4) {
485 if (inet->cmsg_flags)
486 ip_cmsg_recv(msg, skb);
487 } else {
488 if (np->rxopt.all)
489 ip6_datagram_recv_specific_ctl(sk, msg, skb);
490 }
491
492 err = copied;
493 if (flags & MSG_TRUNC)
494 err = ulen;
495
496out_free:
497 skb_free_datagram_locked(sk, skb);
498out:
499 return err;
500
501csum_copy_err:
502 slow = lock_sock_fast(sk);
503 if (!skb_kill_datagram(sk, skb, flags)) {
504 if (is_udp4) {
505 UDP_INC_STATS_USER(sock_net(sk),
506 UDP_MIB_CSUMERRORS, is_udplite);
507 UDP_INC_STATS_USER(sock_net(sk),
508 UDP_MIB_INERRORS, is_udplite);
509 } else {
510 UDP6_INC_STATS_USER(sock_net(sk),
511 UDP_MIB_CSUMERRORS, is_udplite);
512 UDP6_INC_STATS_USER(sock_net(sk),
513 UDP_MIB_INERRORS, is_udplite);
514 }
515 }
516 unlock_sock_fast(sk, slow);
517
518 if (noblock)
519 return -EAGAIN;
520
521 /* starting over for a new packet */
522 msg->msg_flags &= ~MSG_TRUNC;
523 goto try_again;
524}
525
526void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
527 u8 type, u8 code, int offset, __be32 info,
528 struct udp_table *udptable)
529{
530 struct ipv6_pinfo *np;
531 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
532 const struct in6_addr *saddr = &hdr->saddr;
533 const struct in6_addr *daddr = &hdr->daddr;
534 struct udphdr *uh = (struct udphdr*)(skb->data+offset);
535 struct sock *sk;
536 int err;
537
538 sk = __udp6_lib_lookup(dev_net(skb->dev), daddr, uh->dest,
539 saddr, uh->source, inet6_iif(skb), udptable);
540 if (sk == NULL)
541 return;
542
543 if (type == ICMPV6_PKT_TOOBIG) {
544 if (!ip6_sk_accept_pmtu(sk))
545 goto out;
546 ip6_sk_update_pmtu(skb, sk, info);
547 }
548 if (type == NDISC_REDIRECT) {
549 ip6_sk_redirect(skb, sk);
550 goto out;
551 }
552
553 np = inet6_sk(sk);
554
555 if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
556 goto out;
557
558 if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
559 goto out;
560
561 if (np->recverr)
562 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
563
564 sk->sk_err = err;
565 sk->sk_error_report(sk);
566out:
567 sock_put(sk);
568}
569
570static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
571{
572 int rc;
573
574 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
575 sock_rps_save_rxhash(sk, skb);
576 sk_mark_napi_id(sk, skb);
577 }
578
579 rc = sock_queue_rcv_skb(sk, skb);
580 if (rc < 0) {
581 int is_udplite = IS_UDPLITE(sk);
582
583 /* Note that an ENOMEM error is charged twice */
584 if (rc == -ENOMEM)
585 UDP6_INC_STATS_BH(sock_net(sk),
586 UDP_MIB_RCVBUFERRORS, is_udplite);
587 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
588 kfree_skb(skb);
589 return -1;
590 }
591 return 0;
592}
593
594static __inline__ void udpv6_err(struct sk_buff *skb,
595 struct inet6_skb_parm *opt, u8 type,
596 u8 code, int offset, __be32 info )
597{
598 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
599}
600
601static struct static_key udpv6_encap_needed __read_mostly;
602void udpv6_encap_enable(void)
603{
604 if (!static_key_enabled(&udpv6_encap_needed))
605 static_key_slow_inc(&udpv6_encap_needed);
606}
607EXPORT_SYMBOL(udpv6_encap_enable);
608
609int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
610{
611 struct udp_sock *up = udp_sk(sk);
612 int rc;
613 int is_udplite = IS_UDPLITE(sk);
614
615 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
616 goto drop;
617
618 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
619 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
620
621 /*
622 * This is an encapsulation socket so pass the skb to
623 * the socket's udp_encap_rcv() hook. Otherwise, just
624 * fall through and pass this up the UDP socket.
625 * up->encap_rcv() returns the following value:
626 * =0 if skb was successfully passed to the encap
627 * handler or was discarded by it.
628 * >0 if skb should be passed on to UDP.
629 * <0 if skb should be resubmitted as proto -N
630 */
631
632 /* if we're overly short, let UDP handle it */
633 encap_rcv = ACCESS_ONCE(up->encap_rcv);
634 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
635 int ret;
636
637 ret = encap_rcv(sk, skb);
638 if (ret <= 0) {
639 UDP_INC_STATS_BH(sock_net(sk),
640 UDP_MIB_INDATAGRAMS,
641 is_udplite);
642 return -ret;
643 }
644 }
645
646 /* FALLTHROUGH -- it's a UDP Packet */
647 }
648
649 /*
650 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
651 */
652 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
653
654 if (up->pcrlen == 0) { /* full coverage was set */
655 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage"
656 " %d while full coverage %d requested\n",
657 UDP_SKB_CB(skb)->cscov, skb->len);
658 goto drop;
659 }
660 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
661 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d "
662 "too small, need min %d\n",
663 UDP_SKB_CB(skb)->cscov, up->pcrlen);
664 goto drop;
665 }
666 }
667
668 if (rcu_access_pointer(sk->sk_filter)) {
669 if (udp_lib_checksum_complete(skb))
670 goto csum_error;
671 }
672
673 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf))
674 goto drop;
675
676 skb_dst_drop(skb);
677
678 bh_lock_sock(sk);
679 rc = 0;
680 if (!sock_owned_by_user(sk))
681 rc = __udpv6_queue_rcv_skb(sk, skb);
682 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
683 bh_unlock_sock(sk);
684 goto drop;
685 }
686 bh_unlock_sock(sk);
687
688 return rc;
689csum_error:
690 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
691drop:
692 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
693 atomic_inc(&sk->sk_drops);
694 kfree_skb(skb);
695 return -1;
696}
697
698static struct sock *udp_v6_mcast_next(struct net *net, struct sock *sk,
699 __be16 loc_port, const struct in6_addr *loc_addr,
700 __be16 rmt_port, const struct in6_addr *rmt_addr,
701 int dif)
702{
703 struct hlist_nulls_node *node;
704 struct sock *s = sk;
705 unsigned short num = ntohs(loc_port);
706
707 sk_nulls_for_each_from(s, node) {
708 struct inet_sock *inet = inet_sk(s);
709
710 if (!net_eq(sock_net(s), net))
711 continue;
712
713 if (udp_sk(s)->udp_port_hash == num &&
714 s->sk_family == PF_INET6) {
715 if (inet->inet_dport) {
716 if (inet->inet_dport != rmt_port)
717 continue;
718 }
719 if (!ipv6_addr_any(&sk->sk_v6_daddr) &&
720 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr))
721 continue;
722
723 if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)
724 continue;
725
726 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
727 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))
728 continue;
729 }
730 if (!inet6_mc_check(s, loc_addr, rmt_addr))
731 continue;
732 return s;
733 }
734 }
735 return NULL;
736}
737
738static void flush_stack(struct sock **stack, unsigned int count,
739 struct sk_buff *skb, unsigned int final)
740{
741 struct sk_buff *skb1 = NULL;
742 struct sock *sk;
743 unsigned int i;
744
745 for (i = 0; i < count; i++) {
746 sk = stack[i];
747 if (likely(skb1 == NULL))
748 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
749 if (!skb1) {
750 atomic_inc(&sk->sk_drops);
751 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
752 IS_UDPLITE(sk));
753 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
754 IS_UDPLITE(sk));
755 }
756
757 if (skb1 && udpv6_queue_rcv_skb(sk, skb1) <= 0)
758 skb1 = NULL;
759 }
760 if (unlikely(skb1))
761 kfree_skb(skb1);
762}
763/*
764 * Note: called only from the BH handler context,
765 * so we don't need to lock the hashes.
766 */
767static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
768 const struct in6_addr *saddr, const struct in6_addr *daddr,
769 struct udp_table *udptable)
770{
771 struct sock *sk, *stack[256 / sizeof(struct sock *)];
772 const struct udphdr *uh = udp_hdr(skb);
773 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
774 int dif;
775 unsigned int i, count = 0;
776
777 spin_lock(&hslot->lock);
778 sk = sk_nulls_head(&hslot->head);
779 dif = inet6_iif(skb);
780 sk = udp_v6_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
781 while (sk) {
782 stack[count++] = sk;
783 sk = udp_v6_mcast_next(net, sk_nulls_next(sk), uh->dest, daddr,
784 uh->source, saddr, dif);
785 if (unlikely(count == ARRAY_SIZE(stack))) {
786 if (!sk)
787 break;
788 flush_stack(stack, count, skb, ~0);
789 count = 0;
790 }
791 }
792 /*
793 * before releasing the lock, we must take reference on sockets
794 */
795 for (i = 0; i < count; i++)
796 sock_hold(stack[i]);
797
798 spin_unlock(&hslot->lock);
799
800 if (count) {
801 flush_stack(stack, count, skb, count - 1);
802
803 for (i = 0; i < count; i++)
804 sock_put(stack[i]);
805 } else {
806 kfree_skb(skb);
807 }
808 return 0;
809}
810
811int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
812 int proto)
813{
814 struct net *net = dev_net(skb->dev);
815 struct sock *sk;
816 struct udphdr *uh;
817 const struct in6_addr *saddr, *daddr;
818 u32 ulen = 0;
819
820 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
821 goto discard;
822
823 saddr = &ipv6_hdr(skb)->saddr;
824 daddr = &ipv6_hdr(skb)->daddr;
825 uh = udp_hdr(skb);
826
827 ulen = ntohs(uh->len);
828 if (ulen > skb->len)
829 goto short_packet;
830
831 if (proto == IPPROTO_UDP) {
832 /* UDP validates ulen. */
833
834 /* Check for jumbo payload */
835 if (ulen == 0)
836 ulen = skb->len;
837
838 if (ulen < sizeof(*uh))
839 goto short_packet;
840
841 if (ulen < skb->len) {
842 if (pskb_trim_rcsum(skb, ulen))
843 goto short_packet;
844 saddr = &ipv6_hdr(skb)->saddr;
845 daddr = &ipv6_hdr(skb)->daddr;
846 uh = udp_hdr(skb);
847 }
848 }
849
850 if (udp6_csum_init(skb, uh, proto))
851 goto csum_error;
852
853 /*
854 * Multicast receive code
855 */
856 if (ipv6_addr_is_multicast(daddr))
857 return __udp6_lib_mcast_deliver(net, skb,
858 saddr, daddr, udptable);
859
860 /* Unicast */
861
862 /*
863 * check socket cache ... must talk to Alan about his plans
864 * for sock caches... i'll skip this for now.
865 */
866 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
867 if (sk != NULL) {
868 int ret;
869
870 ret = udpv6_queue_rcv_skb(sk, skb);
871 sock_put(sk);
872
873 /* a return value > 0 means to resubmit the input, but
874 * it wants the return to be -protocol, or 0
875 */
876 if (ret > 0)
877 return -ret;
878
879 return 0;
880 }
881
882 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
883 goto discard;
884
885 if (udp_lib_checksum_complete(skb))
886 goto csum_error;
887
888 UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
889 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
890
891 kfree_skb(skb);
892 return 0;
893
894short_packet:
895 LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
896 proto == IPPROTO_UDPLITE ? "-Lite" : "",
897 saddr,
898 ntohs(uh->source),
899 ulen,
900 skb->len,
901 daddr,
902 ntohs(uh->dest));
903 goto discard;
904csum_error:
905 UDP6_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
906discard:
907 UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
908 kfree_skb(skb);
909 return 0;
910}
911
912static __inline__ int udpv6_rcv(struct sk_buff *skb)
913{
914 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
915}
916
917/*
918 * Throw away all pending data and cancel the corking. Socket is locked.
919 */
920static void udp_v6_flush_pending_frames(struct sock *sk)
921{
922 struct udp_sock *up = udp_sk(sk);
923
924 if (up->pending == AF_INET)
925 udp_flush_pending_frames(sk);
926 else if (up->pending) {
927 up->len = 0;
928 up->pending = 0;
929 ip6_flush_pending_frames(sk);
930 }
931}
932
933/**
934 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
935 * @sk: socket we are sending on
936 * @skb: sk_buff containing the filled-in UDP header
937 * (checksum field must be zeroed out)
938 */
939static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
940 const struct in6_addr *saddr,
941 const struct in6_addr *daddr, int len)
942{
943 unsigned int offset;
944 struct udphdr *uh = udp_hdr(skb);
945 __wsum csum = 0;
946
947 if (skb_queue_len(&sk->sk_write_queue) == 1) {
948 /* Only one fragment on the socket. */
949 skb->csum_start = skb_transport_header(skb) - skb->head;
950 skb->csum_offset = offsetof(struct udphdr, check);
951 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
952 } else {
953 /*
954 * HW-checksum won't work as there are two or more
955 * fragments on the socket so that all csums of sk_buffs
956 * should be together
957 */
958 offset = skb_transport_offset(skb);
959 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
960
961 skb->ip_summed = CHECKSUM_NONE;
962
963 skb_queue_walk(&sk->sk_write_queue, skb) {
964 csum = csum_add(csum, skb->csum);
965 }
966
967 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
968 csum);
969 if (uh->check == 0)
970 uh->check = CSUM_MANGLED_0;
971 }
972}
973
974/*
975 * Sending
976 */
977
978static int udp_v6_push_pending_frames(struct sock *sk)
979{
980 struct sk_buff *skb;
981 struct udphdr *uh;
982 struct udp_sock *up = udp_sk(sk);
983 struct inet_sock *inet = inet_sk(sk);
984 struct flowi6 *fl6;
985 int err = 0;
986 int is_udplite = IS_UDPLITE(sk);
987 __wsum csum = 0;
988
989 if (up->pending == AF_INET)
990 return udp_push_pending_frames(sk);
991
992 fl6 = &inet->cork.fl.u.ip6;
993
994 /* Grab the skbuff where UDP header space exists. */
995 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
996 goto out;
997
998 /*
999 * Create a UDP header
1000 */
1001 uh = udp_hdr(skb);
1002 uh->source = fl6->fl6_sport;
1003 uh->dest = fl6->fl6_dport;
1004 uh->len = htons(up->len);
1005 uh->check = 0;
1006
1007 if (is_udplite)
1008 csum = udplite_csum_outgoing(sk, skb);
1009 else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1010 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr,
1011 up->len);
1012 goto send;
1013 } else
1014 csum = udp_csum_outgoing(sk, skb);
1015
1016 /* add protocol-dependent pseudo-header */
1017 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1018 up->len, fl6->flowi6_proto, csum);
1019 if (uh->check == 0)
1020 uh->check = CSUM_MANGLED_0;
1021
1022send:
1023 err = ip6_push_pending_frames(sk);
1024 if (err) {
1025 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1026 UDP6_INC_STATS_USER(sock_net(sk),
1027 UDP_MIB_SNDBUFERRORS, is_udplite);
1028 err = 0;
1029 }
1030 } else
1031 UDP6_INC_STATS_USER(sock_net(sk),
1032 UDP_MIB_OUTDATAGRAMS, is_udplite);
1033out:
1034 up->len = 0;
1035 up->pending = 0;
1036 return err;
1037}
1038
1039int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
1040 struct msghdr *msg, size_t len)
1041{
1042 struct ipv6_txoptions opt_space;
1043 struct udp_sock *up = udp_sk(sk);
1044 struct inet_sock *inet = inet_sk(sk);
1045 struct ipv6_pinfo *np = inet6_sk(sk);
1046 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1047 struct in6_addr *daddr, *final_p, final;
1048 struct ipv6_txoptions *opt = NULL;
1049 struct ip6_flowlabel *flowlabel = NULL;
1050 struct flowi6 fl6;
1051 struct dst_entry *dst;
1052 int addr_len = msg->msg_namelen;
1053 int ulen = len;
1054 int hlimit = -1;
1055 int tclass = -1;
1056 int dontfrag = -1;
1057 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1058 int err;
1059 int connected = 0;
1060 int is_udplite = IS_UDPLITE(sk);
1061 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1062
1063 /* destination address check */
1064 if (sin6) {
1065 if (addr_len < offsetof(struct sockaddr, sa_data))
1066 return -EINVAL;
1067
1068 switch (sin6->sin6_family) {
1069 case AF_INET6:
1070 if (addr_len < SIN6_LEN_RFC2133)
1071 return -EINVAL;
1072 daddr = &sin6->sin6_addr;
1073 break;
1074 case AF_INET:
1075 goto do_udp_sendmsg;
1076 case AF_UNSPEC:
1077 msg->msg_name = sin6 = NULL;
1078 msg->msg_namelen = addr_len = 0;
1079 daddr = NULL;
1080 break;
1081 default:
1082 return -EINVAL;
1083 }
1084 } else if (!up->pending) {
1085 if (sk->sk_state != TCP_ESTABLISHED)
1086 return -EDESTADDRREQ;
1087 daddr = &sk->sk_v6_daddr;
1088 } else
1089 daddr = NULL;
1090
1091 if (daddr) {
1092 if (ipv6_addr_v4mapped(daddr)) {
1093 struct sockaddr_in sin;
1094 sin.sin_family = AF_INET;
1095 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1096 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1097 msg->msg_name = &sin;
1098 msg->msg_namelen = sizeof(sin);
1099do_udp_sendmsg:
1100 if (__ipv6_only_sock(sk))
1101 return -ENETUNREACH;
1102 return udp_sendmsg(iocb, sk, msg, len);
1103 }
1104 }
1105
1106 if (up->pending == AF_INET)
1107 return udp_sendmsg(iocb, sk, msg, len);
1108
1109 /* Rough check on arithmetic overflow,
1110 better check is made in ip6_append_data().
1111 */
1112 if (len > INT_MAX - sizeof(struct udphdr))
1113 return -EMSGSIZE;
1114
1115 if (up->pending) {
1116 /*
1117 * There are pending frames.
1118 * The socket lock must be held while it's corked.
1119 */
1120 lock_sock(sk);
1121 if (likely(up->pending)) {
1122 if (unlikely(up->pending != AF_INET6)) {
1123 release_sock(sk);
1124 return -EAFNOSUPPORT;
1125 }
1126 dst = NULL;
1127 goto do_append_data;
1128 }
1129 release_sock(sk);
1130 }
1131 ulen += sizeof(struct udphdr);
1132
1133 memset(&fl6, 0, sizeof(fl6));
1134
1135 if (sin6) {
1136 if (sin6->sin6_port == 0)
1137 return -EINVAL;
1138
1139 fl6.fl6_dport = sin6->sin6_port;
1140 daddr = &sin6->sin6_addr;
1141
1142 if (np->sndflow) {
1143 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1144 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1145 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1146 if (flowlabel == NULL)
1147 return -EINVAL;
1148 }
1149 }
1150
1151 /*
1152 * Otherwise it will be difficult to maintain
1153 * sk->sk_dst_cache.
1154 */
1155 if (sk->sk_state == TCP_ESTABLISHED &&
1156 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1157 daddr = &sk->sk_v6_daddr;
1158
1159 if (addr_len >= sizeof(struct sockaddr_in6) &&
1160 sin6->sin6_scope_id &&
1161 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1162 fl6.flowi6_oif = sin6->sin6_scope_id;
1163 } else {
1164 if (sk->sk_state != TCP_ESTABLISHED)
1165 return -EDESTADDRREQ;
1166
1167 fl6.fl6_dport = inet->inet_dport;
1168 daddr = &sk->sk_v6_daddr;
1169 fl6.flowlabel = np->flow_label;
1170 connected = 1;
1171 }
1172
1173 if (!fl6.flowi6_oif)
1174 fl6.flowi6_oif = sk->sk_bound_dev_if;
1175
1176 if (!fl6.flowi6_oif)
1177 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1178
1179 fl6.flowi6_mark = sk->sk_mark;
1180
1181 if (msg->msg_controllen) {
1182 opt = &opt_space;
1183 memset(opt, 0, sizeof(struct ipv6_txoptions));
1184 opt->tot_len = sizeof(*opt);
1185
1186 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt,
1187 &hlimit, &tclass, &dontfrag);
1188 if (err < 0) {
1189 fl6_sock_release(flowlabel);
1190 return err;
1191 }
1192 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1193 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1194 if (flowlabel == NULL)
1195 return -EINVAL;
1196 }
1197 if (!(opt->opt_nflen|opt->opt_flen))
1198 opt = NULL;
1199 connected = 0;
1200 }
1201 if (opt == NULL)
1202 opt = np->opt;
1203 if (flowlabel)
1204 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1205 opt = ipv6_fixup_options(&opt_space, opt);
1206
1207 fl6.flowi6_proto = sk->sk_protocol;
1208 if (!ipv6_addr_any(daddr))
1209 fl6.daddr = *daddr;
1210 else
1211 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1212 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1213 fl6.saddr = np->saddr;
1214 fl6.fl6_sport = inet->inet_sport;
1215
1216 final_p = fl6_update_dst(&fl6, opt, &final);
1217 if (final_p)
1218 connected = 0;
1219
1220 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1221 fl6.flowi6_oif = np->mcast_oif;
1222 connected = 0;
1223 } else if (!fl6.flowi6_oif)
1224 fl6.flowi6_oif = np->ucast_oif;
1225
1226 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1227
1228 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p);
1229 if (IS_ERR(dst)) {
1230 err = PTR_ERR(dst);
1231 dst = NULL;
1232 goto out;
1233 }
1234
1235 if (hlimit < 0) {
1236 if (ipv6_addr_is_multicast(&fl6.daddr))
1237 hlimit = np->mcast_hops;
1238 else
1239 hlimit = np->hop_limit;
1240 if (hlimit < 0)
1241 hlimit = ip6_dst_hoplimit(dst);
1242 }
1243
1244 if (tclass < 0)
1245 tclass = np->tclass;
1246
1247 if (msg->msg_flags&MSG_CONFIRM)
1248 goto do_confirm;
1249back_from_confirm:
1250
1251 lock_sock(sk);
1252 if (unlikely(up->pending)) {
1253 /* The socket is already corked while preparing it. */
1254 /* ... which is an evident application bug. --ANK */
1255 release_sock(sk);
1256
1257 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
1258 err = -EINVAL;
1259 goto out;
1260 }
1261
1262 up->pending = AF_INET6;
1263
1264do_append_data:
1265 if (dontfrag < 0)
1266 dontfrag = np->dontfrag;
1267 up->len += ulen;
1268 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1269 err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen,
1270 sizeof(struct udphdr), hlimit, tclass, opt, &fl6,
1271 (struct rt6_info*)dst,
1272 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag);
1273 if (err)
1274 udp_v6_flush_pending_frames(sk);
1275 else if (!corkreq)
1276 err = udp_v6_push_pending_frames(sk);
1277 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1278 up->pending = 0;
1279
1280 if (dst) {
1281 if (connected) {
1282 ip6_dst_store(sk, dst,
1283 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ?
1284 &sk->sk_v6_daddr : NULL,
1285#ifdef CONFIG_IPV6_SUBTREES
1286 ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1287 &np->saddr :
1288#endif
1289 NULL);
1290 } else {
1291 dst_release(dst);
1292 }
1293 dst = NULL;
1294 }
1295
1296 if (err > 0)
1297 err = np->recverr ? net_xmit_errno(err) : 0;
1298 release_sock(sk);
1299out:
1300 dst_release(dst);
1301 fl6_sock_release(flowlabel);
1302 if (!err)
1303 return len;
1304 /*
1305 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1306 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1307 * we don't have a good statistic (IpOutDiscards but it can be too many
1308 * things). We could add another new stat but at least for now that
1309 * seems like overkill.
1310 */
1311 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1312 UDP6_INC_STATS_USER(sock_net(sk),
1313 UDP_MIB_SNDBUFERRORS, is_udplite);
1314 }
1315 return err;
1316
1317do_confirm:
1318 dst_confirm(dst);
1319 if (!(msg->msg_flags&MSG_PROBE) || len)
1320 goto back_from_confirm;
1321 err = 0;
1322 goto out;
1323}
1324
1325void udpv6_destroy_sock(struct sock *sk)
1326{
1327 struct udp_sock *up = udp_sk(sk);
1328 lock_sock(sk);
1329 udp_v6_flush_pending_frames(sk);
1330 release_sock(sk);
1331
1332 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
1333 void (*encap_destroy)(struct sock *sk);
1334 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1335 if (encap_destroy)
1336 encap_destroy(sk);
1337 }
1338
1339 inet6_destroy_sock(sk);
1340}
1341
1342/*
1343 * Socket option code for UDP
1344 */
1345int udpv6_setsockopt(struct sock *sk, int level, int optname,
1346 char __user *optval, unsigned int optlen)
1347{
1348 if (level == SOL_UDP || level == SOL_UDPLITE)
1349 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1350 udp_v6_push_pending_frames);
1351 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1352}
1353
1354#ifdef CONFIG_COMPAT
1355int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1356 char __user *optval, unsigned int optlen)
1357{
1358 if (level == SOL_UDP || level == SOL_UDPLITE)
1359 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1360 udp_v6_push_pending_frames);
1361 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1362}
1363#endif
1364
1365int udpv6_getsockopt(struct sock *sk, int level, int optname,
1366 char __user *optval, int __user *optlen)
1367{
1368 if (level == SOL_UDP || level == SOL_UDPLITE)
1369 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1370 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1371}
1372
1373#ifdef CONFIG_COMPAT
1374int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1375 char __user *optval, int __user *optlen)
1376{
1377 if (level == SOL_UDP || level == SOL_UDPLITE)
1378 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1379 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1380}
1381#endif
1382
1383static const struct inet6_protocol udpv6_protocol = {
1384 .handler = udpv6_rcv,
1385 .err_handler = udpv6_err,
1386 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1387};
1388
1389/* ------------------------------------------------------------------------ */
1390#ifdef CONFIG_PROC_FS
1391int udp6_seq_show(struct seq_file *seq, void *v)
1392{
1393 if (v == SEQ_START_TOKEN) {
1394 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1395 } else {
1396 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1397 struct inet_sock *inet = inet_sk(v);
1398 __u16 srcp = ntohs(inet->inet_sport);
1399 __u16 destp = ntohs(inet->inet_dport);
1400 ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket);
1401 }
1402 return 0;
1403}
1404
1405static const struct file_operations udp6_afinfo_seq_fops = {
1406 .owner = THIS_MODULE,
1407 .open = udp_seq_open,
1408 .read = seq_read,
1409 .llseek = seq_lseek,
1410 .release = seq_release_net
1411};
1412
1413static struct udp_seq_afinfo udp6_seq_afinfo = {
1414 .name = "udp6",
1415 .family = AF_INET6,
1416 .udp_table = &udp_table,
1417 .seq_fops = &udp6_afinfo_seq_fops,
1418 .seq_ops = {
1419 .show = udp6_seq_show,
1420 },
1421};
1422
1423int __net_init udp6_proc_init(struct net *net)
1424{
1425 return udp_proc_register(net, &udp6_seq_afinfo);
1426}
1427
1428void udp6_proc_exit(struct net *net) {
1429 udp_proc_unregister(net, &udp6_seq_afinfo);
1430}
1431#endif /* CONFIG_PROC_FS */
1432
1433void udp_v6_clear_sk(struct sock *sk, int size)
1434{
1435 struct inet_sock *inet = inet_sk(sk);
1436
1437 /* we do not want to clear pinet6 field, because of RCU lookups */
1438 sk_prot_clear_portaddr_nulls(sk, offsetof(struct inet_sock, pinet6));
1439
1440 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1441 memset(&inet->pinet6 + 1, 0, size);
1442}
1443
1444/* ------------------------------------------------------------------------ */
1445
1446struct proto udpv6_prot = {
1447 .name = "UDPv6",
1448 .owner = THIS_MODULE,
1449 .close = udp_lib_close,
1450 .connect = ip6_datagram_connect,
1451 .disconnect = udp_disconnect,
1452 .ioctl = udp_ioctl,
1453 .destroy = udpv6_destroy_sock,
1454 .setsockopt = udpv6_setsockopt,
1455 .getsockopt = udpv6_getsockopt,
1456 .sendmsg = udpv6_sendmsg,
1457 .recvmsg = udpv6_recvmsg,
1458 .backlog_rcv = __udpv6_queue_rcv_skb,
1459 .hash = udp_lib_hash,
1460 .unhash = udp_lib_unhash,
1461 .rehash = udp_v6_rehash,
1462 .get_port = udp_v6_get_port,
1463 .memory_allocated = &udp_memory_allocated,
1464 .sysctl_mem = sysctl_udp_mem,
1465 .sysctl_wmem = &sysctl_udp_wmem_min,
1466 .sysctl_rmem = &sysctl_udp_rmem_min,
1467 .obj_size = sizeof(struct udp6_sock),
1468 .slab_flags = SLAB_DESTROY_BY_RCU,
1469 .h.udp_table = &udp_table,
1470#ifdef CONFIG_COMPAT
1471 .compat_setsockopt = compat_udpv6_setsockopt,
1472 .compat_getsockopt = compat_udpv6_getsockopt,
1473#endif
1474 .clear_sk = udp_v6_clear_sk,
1475};
1476
1477static struct inet_protosw udpv6_protosw = {
1478 .type = SOCK_DGRAM,
1479 .protocol = IPPROTO_UDP,
1480 .prot = &udpv6_prot,
1481 .ops = &inet6_dgram_ops,
1482 .no_check = UDP_CSUM_DEFAULT,
1483 .flags = INET_PROTOSW_PERMANENT,
1484};
1485
1486
1487int __init udpv6_init(void)
1488{
1489 int ret;
1490
1491 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1492 if (ret)
1493 goto out;
1494
1495 ret = inet6_register_protosw(&udpv6_protosw);
1496 if (ret)
1497 goto out_udpv6_protocol;
1498out:
1499 return ret;
1500
1501out_udpv6_protocol:
1502 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1503 goto out;
1504}
1505
1506void udpv6_exit(void)
1507{
1508 inet6_unregister_protosw(&udpv6_protosw);
1509 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1510}