Loading...
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/addrconf.h>
41#include <net/ndisc.h>
42#include <net/protocol.h>
43#include <net/transp_v6.h>
44#include <net/ip6_route.h>
45#include <net/raw.h>
46#include <net/tcp_states.h>
47#include <net/ip6_checksum.h>
48#include <net/xfrm.h>
49#include <net/inet6_hashtables.h>
50#include <net/busy_poll.h>
51#include <net/sock_reuseport.h>
52
53#include <linux/proc_fs.h>
54#include <linux/seq_file.h>
55#include <trace/events/skb.h>
56#include "udp_impl.h"
57
58static u32 udp6_ehashfn(const struct net *net,
59 const struct in6_addr *laddr,
60 const u16 lport,
61 const struct in6_addr *faddr,
62 const __be16 fport)
63{
64 static u32 udp6_ehash_secret __read_mostly;
65 static u32 udp_ipv6_hash_secret __read_mostly;
66
67 u32 lhash, fhash;
68
69 net_get_random_once(&udp6_ehash_secret,
70 sizeof(udp6_ehash_secret));
71 net_get_random_once(&udp_ipv6_hash_secret,
72 sizeof(udp_ipv6_hash_secret));
73
74 lhash = (__force u32)laddr->s6_addr32[3];
75 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
76
77 return __inet6_ehashfn(lhash, lport, fhash, fport,
78 udp_ipv6_hash_secret + net_hash_mix(net));
79}
80
81static u32 udp6_portaddr_hash(const struct net *net,
82 const struct in6_addr *addr6,
83 unsigned int port)
84{
85 unsigned int hash, mix = net_hash_mix(net);
86
87 if (ipv6_addr_any(addr6))
88 hash = jhash_1word(0, mix);
89 else if (ipv6_addr_v4mapped(addr6))
90 hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix);
91 else
92 hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix);
93
94 return hash ^ port;
95}
96
97int udp_v6_get_port(struct sock *sk, unsigned short snum)
98{
99 unsigned int hash2_nulladdr =
100 udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
101 unsigned int hash2_partial =
102 udp6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
103
104 /* precompute partial secondary hash */
105 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
106 return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal, hash2_nulladdr);
107}
108
109static void udp_v6_rehash(struct sock *sk)
110{
111 u16 new_hash = udp6_portaddr_hash(sock_net(sk),
112 &sk->sk_v6_rcv_saddr,
113 inet_sk(sk)->inet_num);
114
115 udp_lib_rehash(sk, new_hash);
116}
117
118static inline int compute_score(struct sock *sk, struct net *net,
119 unsigned short hnum,
120 const struct in6_addr *saddr, __be16 sport,
121 const struct in6_addr *daddr, __be16 dport,
122 int dif)
123{
124 int score;
125 struct inet_sock *inet;
126
127 if (!net_eq(sock_net(sk), net) ||
128 udp_sk(sk)->udp_port_hash != hnum ||
129 sk->sk_family != PF_INET6)
130 return -1;
131
132 score = 0;
133 inet = inet_sk(sk);
134
135 if (inet->inet_dport) {
136 if (inet->inet_dport != sport)
137 return -1;
138 score++;
139 }
140
141 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
142 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
143 return -1;
144 score++;
145 }
146
147 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
148 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
149 return -1;
150 score++;
151 }
152
153 if (sk->sk_bound_dev_if) {
154 if (sk->sk_bound_dev_if != dif)
155 return -1;
156 score++;
157 }
158
159 if (sk->sk_incoming_cpu == raw_smp_processor_id())
160 score++;
161
162 return score;
163}
164
165static inline int compute_score2(struct sock *sk, struct net *net,
166 const struct in6_addr *saddr, __be16 sport,
167 const struct in6_addr *daddr,
168 unsigned short hnum, int dif)
169{
170 int score;
171 struct inet_sock *inet;
172
173 if (!net_eq(sock_net(sk), net) ||
174 udp_sk(sk)->udp_port_hash != hnum ||
175 sk->sk_family != PF_INET6)
176 return -1;
177
178 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
179 return -1;
180
181 score = 0;
182 inet = inet_sk(sk);
183
184 if (inet->inet_dport) {
185 if (inet->inet_dport != sport)
186 return -1;
187 score++;
188 }
189
190 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
191 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
192 return -1;
193 score++;
194 }
195
196 if (sk->sk_bound_dev_if) {
197 if (sk->sk_bound_dev_if != dif)
198 return -1;
199 score++;
200 }
201
202 if (sk->sk_incoming_cpu == raw_smp_processor_id())
203 score++;
204
205 return score;
206}
207
208/* called with read_rcu_lock() */
209static struct sock *udp6_lib_lookup2(struct net *net,
210 const struct in6_addr *saddr, __be16 sport,
211 const struct in6_addr *daddr, unsigned int hnum, int dif,
212 struct udp_hslot *hslot2, unsigned int slot2,
213 struct sk_buff *skb)
214{
215 struct sock *sk, *result;
216 struct hlist_nulls_node *node;
217 int score, badness, matches = 0, reuseport = 0;
218 bool select_ok = true;
219 u32 hash = 0;
220
221begin:
222 result = NULL;
223 badness = -1;
224 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
225 score = compute_score2(sk, net, saddr, sport,
226 daddr, hnum, dif);
227 if (score > badness) {
228 result = sk;
229 badness = score;
230 reuseport = sk->sk_reuseport;
231 if (reuseport) {
232 hash = udp6_ehashfn(net, daddr, hnum,
233 saddr, sport);
234 if (select_ok) {
235 struct sock *sk2;
236
237 sk2 = reuseport_select_sock(sk, hash, skb,
238 sizeof(struct udphdr));
239 if (sk2) {
240 result = sk2;
241 select_ok = false;
242 goto found;
243 }
244 }
245 matches = 1;
246 }
247 } else if (score == badness && reuseport) {
248 matches++;
249 if (reciprocal_scale(hash, matches) == 0)
250 result = sk;
251 hash = next_pseudo_random32(hash);
252 }
253 }
254 /*
255 * if the nulls value we got at the end of this lookup is
256 * not the expected one, we must restart lookup.
257 * We probably met an item that was moved to another chain.
258 */
259 if (get_nulls_value(node) != slot2)
260 goto begin;
261
262 if (result) {
263found:
264 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
265 result = NULL;
266 else if (unlikely(compute_score2(result, net, saddr, sport,
267 daddr, hnum, dif) < badness)) {
268 sock_put(result);
269 goto begin;
270 }
271 }
272 return result;
273}
274
275struct sock *__udp6_lib_lookup(struct net *net,
276 const struct in6_addr *saddr, __be16 sport,
277 const struct in6_addr *daddr, __be16 dport,
278 int dif, struct udp_table *udptable,
279 struct sk_buff *skb)
280{
281 struct sock *sk, *result;
282 struct hlist_nulls_node *node;
283 unsigned short hnum = ntohs(dport);
284 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
285 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
286 int score, badness, matches = 0, reuseport = 0;
287 bool select_ok = true;
288 u32 hash = 0;
289
290 rcu_read_lock();
291 if (hslot->count > 10) {
292 hash2 = udp6_portaddr_hash(net, daddr, hnum);
293 slot2 = hash2 & udptable->mask;
294 hslot2 = &udptable->hash2[slot2];
295 if (hslot->count < hslot2->count)
296 goto begin;
297
298 result = udp6_lib_lookup2(net, saddr, sport,
299 daddr, hnum, dif,
300 hslot2, slot2, skb);
301 if (!result) {
302 hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
303 slot2 = hash2 & udptable->mask;
304 hslot2 = &udptable->hash2[slot2];
305 if (hslot->count < hslot2->count)
306 goto begin;
307
308 result = udp6_lib_lookup2(net, saddr, sport,
309 &in6addr_any, hnum, dif,
310 hslot2, slot2, skb);
311 }
312 rcu_read_unlock();
313 return result;
314 }
315begin:
316 result = NULL;
317 badness = -1;
318 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
319 score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif);
320 if (score > badness) {
321 result = sk;
322 badness = score;
323 reuseport = sk->sk_reuseport;
324 if (reuseport) {
325 hash = udp6_ehashfn(net, daddr, hnum,
326 saddr, sport);
327 if (select_ok) {
328 struct sock *sk2;
329
330 sk2 = reuseport_select_sock(sk, hash, skb,
331 sizeof(struct udphdr));
332 if (sk2) {
333 result = sk2;
334 select_ok = false;
335 goto found;
336 }
337 }
338 matches = 1;
339 }
340 } else if (score == badness && reuseport) {
341 matches++;
342 if (reciprocal_scale(hash, matches) == 0)
343 result = sk;
344 hash = next_pseudo_random32(hash);
345 }
346 }
347 /*
348 * if the nulls value we got at the end of this lookup is
349 * not the expected one, we must restart lookup.
350 * We probably met an item that was moved to another chain.
351 */
352 if (get_nulls_value(node) != slot)
353 goto begin;
354
355 if (result) {
356found:
357 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
358 result = NULL;
359 else if (unlikely(compute_score(result, net, hnum, saddr, sport,
360 daddr, dport, dif) < badness)) {
361 sock_put(result);
362 goto begin;
363 }
364 }
365 rcu_read_unlock();
366 return result;
367}
368EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
369
370static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
371 __be16 sport, __be16 dport,
372 struct udp_table *udptable)
373{
374 struct sock *sk;
375 const struct ipv6hdr *iph = ipv6_hdr(skb);
376
377 sk = skb_steal_sock(skb);
378 if (unlikely(sk))
379 return sk;
380 return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport,
381 &iph->daddr, dport, inet6_iif(skb),
382 udptable, skb);
383}
384
385struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
386 const struct in6_addr *daddr, __be16 dport, int dif)
387{
388 return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table, NULL);
389}
390EXPORT_SYMBOL_GPL(udp6_lib_lookup);
391
392/*
393 * This should be easy, if there is something there we
394 * return it, otherwise we block.
395 */
396
397int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
398 int noblock, int flags, int *addr_len)
399{
400 struct ipv6_pinfo *np = inet6_sk(sk);
401 struct inet_sock *inet = inet_sk(sk);
402 struct sk_buff *skb;
403 unsigned int ulen, copied;
404 int peeked, off = 0;
405 int err;
406 int is_udplite = IS_UDPLITE(sk);
407 bool checksum_valid = false;
408 int is_udp4;
409 bool slow;
410
411 if (flags & MSG_ERRQUEUE)
412 return ipv6_recv_error(sk, msg, len, addr_len);
413
414 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
415 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
416
417try_again:
418 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
419 &peeked, &off, &err);
420 if (!skb)
421 goto out;
422
423 ulen = skb->len - sizeof(struct udphdr);
424 copied = len;
425 if (copied > ulen)
426 copied = ulen;
427 else if (copied < ulen)
428 msg->msg_flags |= MSG_TRUNC;
429
430 is_udp4 = (skb->protocol == htons(ETH_P_IP));
431
432 /*
433 * If checksum is needed at all, try to do it while copying the
434 * data. If the data is truncated, or if we only want a partial
435 * coverage checksum (UDP-Lite), do it before the copy.
436 */
437
438 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
439 checksum_valid = !udp_lib_checksum_complete(skb);
440 if (!checksum_valid)
441 goto csum_copy_err;
442 }
443
444 if (checksum_valid || skb_csum_unnecessary(skb))
445 err = skb_copy_datagram_msg(skb, sizeof(struct udphdr),
446 msg, copied);
447 else {
448 err = skb_copy_and_csum_datagram_msg(skb, sizeof(struct udphdr), msg);
449 if (err == -EINVAL)
450 goto csum_copy_err;
451 }
452 if (unlikely(err)) {
453 trace_kfree_skb(skb, udpv6_recvmsg);
454 if (!peeked) {
455 atomic_inc(&sk->sk_drops);
456 if (is_udp4)
457 UDP_INC_STATS_USER(sock_net(sk),
458 UDP_MIB_INERRORS,
459 is_udplite);
460 else
461 UDP6_INC_STATS_USER(sock_net(sk),
462 UDP_MIB_INERRORS,
463 is_udplite);
464 }
465 goto out_free;
466 }
467 if (!peeked) {
468 if (is_udp4)
469 UDP_INC_STATS_USER(sock_net(sk),
470 UDP_MIB_INDATAGRAMS, is_udplite);
471 else
472 UDP6_INC_STATS_USER(sock_net(sk),
473 UDP_MIB_INDATAGRAMS, is_udplite);
474 }
475
476 sock_recv_ts_and_drops(msg, sk, skb);
477
478 /* Copy the address. */
479 if (msg->msg_name) {
480 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
481 sin6->sin6_family = AF_INET6;
482 sin6->sin6_port = udp_hdr(skb)->source;
483 sin6->sin6_flowinfo = 0;
484
485 if (is_udp4) {
486 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
487 &sin6->sin6_addr);
488 sin6->sin6_scope_id = 0;
489 } else {
490 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
491 sin6->sin6_scope_id =
492 ipv6_iface_scope_id(&sin6->sin6_addr,
493 inet6_iif(skb));
494 }
495 *addr_len = sizeof(*sin6);
496 }
497
498 if (np->rxopt.all)
499 ip6_datagram_recv_common_ctl(sk, msg, skb);
500
501 if (is_udp4) {
502 if (inet->cmsg_flags)
503 ip_cmsg_recv(msg, skb);
504 } else {
505 if (np->rxopt.all)
506 ip6_datagram_recv_specific_ctl(sk, msg, skb);
507 }
508
509 err = copied;
510 if (flags & MSG_TRUNC)
511 err = ulen;
512
513out_free:
514 skb_free_datagram_locked(sk, skb);
515out:
516 return err;
517
518csum_copy_err:
519 slow = lock_sock_fast(sk);
520 if (!skb_kill_datagram(sk, skb, flags)) {
521 if (is_udp4) {
522 UDP_INC_STATS_USER(sock_net(sk),
523 UDP_MIB_CSUMERRORS, is_udplite);
524 UDP_INC_STATS_USER(sock_net(sk),
525 UDP_MIB_INERRORS, is_udplite);
526 } else {
527 UDP6_INC_STATS_USER(sock_net(sk),
528 UDP_MIB_CSUMERRORS, is_udplite);
529 UDP6_INC_STATS_USER(sock_net(sk),
530 UDP_MIB_INERRORS, is_udplite);
531 }
532 }
533 unlock_sock_fast(sk, slow);
534
535 /* starting over for a new packet, but check if we need to yield */
536 cond_resched();
537 msg->msg_flags &= ~MSG_TRUNC;
538 goto try_again;
539}
540
541void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
542 u8 type, u8 code, int offset, __be32 info,
543 struct udp_table *udptable)
544{
545 struct ipv6_pinfo *np;
546 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
547 const struct in6_addr *saddr = &hdr->saddr;
548 const struct in6_addr *daddr = &hdr->daddr;
549 struct udphdr *uh = (struct udphdr *)(skb->data+offset);
550 struct sock *sk;
551 int harderr;
552 int err;
553 struct net *net = dev_net(skb->dev);
554
555 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
556 inet6_iif(skb), udptable, skb);
557 if (!sk) {
558 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
559 ICMP6_MIB_INERRORS);
560 return;
561 }
562
563 harderr = icmpv6_err_convert(type, code, &err);
564 np = inet6_sk(sk);
565
566 if (type == ICMPV6_PKT_TOOBIG) {
567 if (!ip6_sk_accept_pmtu(sk))
568 goto out;
569 ip6_sk_update_pmtu(skb, sk, info);
570 if (np->pmtudisc != IPV6_PMTUDISC_DONT)
571 harderr = 1;
572 }
573 if (type == NDISC_REDIRECT) {
574 ip6_sk_redirect(skb, sk);
575 goto out;
576 }
577
578 if (!np->recverr) {
579 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
580 goto out;
581 } else {
582 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
583 }
584
585 sk->sk_err = err;
586 sk->sk_error_report(sk);
587out:
588 sock_put(sk);
589}
590
591static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
592{
593 int rc;
594
595 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
596 sock_rps_save_rxhash(sk, skb);
597 sk_mark_napi_id(sk, skb);
598 sk_incoming_cpu_update(sk);
599 }
600
601 rc = sock_queue_rcv_skb(sk, skb);
602 if (rc < 0) {
603 int is_udplite = IS_UDPLITE(sk);
604
605 /* Note that an ENOMEM error is charged twice */
606 if (rc == -ENOMEM)
607 UDP6_INC_STATS_BH(sock_net(sk),
608 UDP_MIB_RCVBUFERRORS, is_udplite);
609 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
610 kfree_skb(skb);
611 return -1;
612 }
613 return 0;
614}
615
616static __inline__ void udpv6_err(struct sk_buff *skb,
617 struct inet6_skb_parm *opt, u8 type,
618 u8 code, int offset, __be32 info)
619{
620 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
621}
622
623static struct static_key udpv6_encap_needed __read_mostly;
624void udpv6_encap_enable(void)
625{
626 if (!static_key_enabled(&udpv6_encap_needed))
627 static_key_slow_inc(&udpv6_encap_needed);
628}
629EXPORT_SYMBOL(udpv6_encap_enable);
630
631int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
632{
633 struct udp_sock *up = udp_sk(sk);
634 int rc;
635 int is_udplite = IS_UDPLITE(sk);
636
637 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
638 goto drop;
639
640 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
641 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
642
643 /*
644 * This is an encapsulation socket so pass the skb to
645 * the socket's udp_encap_rcv() hook. Otherwise, just
646 * fall through and pass this up the UDP socket.
647 * up->encap_rcv() returns the following value:
648 * =0 if skb was successfully passed to the encap
649 * handler or was discarded by it.
650 * >0 if skb should be passed on to UDP.
651 * <0 if skb should be resubmitted as proto -N
652 */
653
654 /* if we're overly short, let UDP handle it */
655 encap_rcv = ACCESS_ONCE(up->encap_rcv);
656 if (skb->len > sizeof(struct udphdr) && encap_rcv) {
657 int ret;
658
659 /* Verify checksum before giving to encap */
660 if (udp_lib_checksum_complete(skb))
661 goto csum_error;
662
663 ret = encap_rcv(sk, skb);
664 if (ret <= 0) {
665 UDP_INC_STATS_BH(sock_net(sk),
666 UDP_MIB_INDATAGRAMS,
667 is_udplite);
668 return -ret;
669 }
670 }
671
672 /* FALLTHROUGH -- it's a UDP Packet */
673 }
674
675 /*
676 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
677 */
678 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
679
680 if (up->pcrlen == 0) { /* full coverage was set */
681 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
682 UDP_SKB_CB(skb)->cscov, skb->len);
683 goto drop;
684 }
685 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
686 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
687 UDP_SKB_CB(skb)->cscov, up->pcrlen);
688 goto drop;
689 }
690 }
691
692 if (rcu_access_pointer(sk->sk_filter)) {
693 if (udp_lib_checksum_complete(skb))
694 goto csum_error;
695 }
696
697 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
698 UDP6_INC_STATS_BH(sock_net(sk),
699 UDP_MIB_RCVBUFERRORS, is_udplite);
700 goto drop;
701 }
702
703 skb_dst_drop(skb);
704
705 bh_lock_sock(sk);
706 rc = 0;
707 if (!sock_owned_by_user(sk))
708 rc = __udpv6_queue_rcv_skb(sk, skb);
709 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
710 bh_unlock_sock(sk);
711 goto drop;
712 }
713 bh_unlock_sock(sk);
714
715 return rc;
716
717csum_error:
718 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
719drop:
720 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
721 atomic_inc(&sk->sk_drops);
722 kfree_skb(skb);
723 return -1;
724}
725
726static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
727 __be16 loc_port, const struct in6_addr *loc_addr,
728 __be16 rmt_port, const struct in6_addr *rmt_addr,
729 int dif, unsigned short hnum)
730{
731 struct inet_sock *inet = inet_sk(sk);
732
733 if (!net_eq(sock_net(sk), net))
734 return false;
735
736 if (udp_sk(sk)->udp_port_hash != hnum ||
737 sk->sk_family != PF_INET6 ||
738 (inet->inet_dport && inet->inet_dport != rmt_port) ||
739 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
740 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
741 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif) ||
742 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
743 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
744 return false;
745 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
746 return false;
747 return true;
748}
749
750static void flush_stack(struct sock **stack, unsigned int count,
751 struct sk_buff *skb, unsigned int final)
752{
753 struct sk_buff *skb1 = NULL;
754 struct sock *sk;
755 unsigned int i;
756
757 for (i = 0; i < count; i++) {
758 sk = stack[i];
759 if (likely(!skb1))
760 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
761 if (!skb1) {
762 atomic_inc(&sk->sk_drops);
763 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
764 IS_UDPLITE(sk));
765 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
766 IS_UDPLITE(sk));
767 }
768
769 if (skb1 && udpv6_queue_rcv_skb(sk, skb1) <= 0)
770 skb1 = NULL;
771 sock_put(sk);
772 }
773 if (unlikely(skb1))
774 kfree_skb(skb1);
775}
776
777static void udp6_csum_zero_error(struct sk_buff *skb)
778{
779 /* RFC 2460 section 8.1 says that we SHOULD log
780 * this error. Well, it is reasonable.
781 */
782 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
783 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
784 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
785}
786
787/*
788 * Note: called only from the BH handler context,
789 * so we don't need to lock the hashes.
790 */
791static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
792 const struct in6_addr *saddr, const struct in6_addr *daddr,
793 struct udp_table *udptable, int proto)
794{
795 struct sock *sk, *stack[256 / sizeof(struct sock *)];
796 const struct udphdr *uh = udp_hdr(skb);
797 struct hlist_nulls_node *node;
798 unsigned short hnum = ntohs(uh->dest);
799 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
800 int dif = inet6_iif(skb);
801 unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node);
802 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
803 bool inner_flushed = false;
804
805 if (use_hash2) {
806 hash2_any = udp6_portaddr_hash(net, &in6addr_any, hnum) &
807 udp_table.mask;
808 hash2 = udp6_portaddr_hash(net, daddr, hnum) & udp_table.mask;
809start_lookup:
810 hslot = &udp_table.hash2[hash2];
811 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
812 }
813
814 spin_lock(&hslot->lock);
815 sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) {
816 if (__udp_v6_is_mcast_sock(net, sk,
817 uh->dest, daddr,
818 uh->source, saddr,
819 dif, hnum) &&
820 /* If zero checksum and no_check is not on for
821 * the socket then skip it.
822 */
823 (uh->check || udp_sk(sk)->no_check6_rx)) {
824 if (unlikely(count == ARRAY_SIZE(stack))) {
825 flush_stack(stack, count, skb, ~0);
826 inner_flushed = true;
827 count = 0;
828 }
829 stack[count++] = sk;
830 sock_hold(sk);
831 }
832 }
833
834 spin_unlock(&hslot->lock);
835
836 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
837 if (use_hash2 && hash2 != hash2_any) {
838 hash2 = hash2_any;
839 goto start_lookup;
840 }
841
842 if (count) {
843 flush_stack(stack, count, skb, count - 1);
844 } else {
845 if (!inner_flushed)
846 UDP6_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI,
847 proto == IPPROTO_UDPLITE);
848 consume_skb(skb);
849 }
850 return 0;
851}
852
853int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
854 int proto)
855{
856 struct net *net = dev_net(skb->dev);
857 struct sock *sk;
858 struct udphdr *uh;
859 const struct in6_addr *saddr, *daddr;
860 u32 ulen = 0;
861
862 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
863 goto discard;
864
865 saddr = &ipv6_hdr(skb)->saddr;
866 daddr = &ipv6_hdr(skb)->daddr;
867 uh = udp_hdr(skb);
868
869 ulen = ntohs(uh->len);
870 if (ulen > skb->len)
871 goto short_packet;
872
873 if (proto == IPPROTO_UDP) {
874 /* UDP validates ulen. */
875
876 /* Check for jumbo payload */
877 if (ulen == 0)
878 ulen = skb->len;
879
880 if (ulen < sizeof(*uh))
881 goto short_packet;
882
883 if (ulen < skb->len) {
884 if (pskb_trim_rcsum(skb, ulen))
885 goto short_packet;
886 saddr = &ipv6_hdr(skb)->saddr;
887 daddr = &ipv6_hdr(skb)->daddr;
888 uh = udp_hdr(skb);
889 }
890 }
891
892 if (udp6_csum_init(skb, uh, proto))
893 goto csum_error;
894
895 /*
896 * Multicast receive code
897 */
898 if (ipv6_addr_is_multicast(daddr))
899 return __udp6_lib_mcast_deliver(net, skb,
900 saddr, daddr, udptable, proto);
901
902 /* Unicast */
903
904 /*
905 * check socket cache ... must talk to Alan about his plans
906 * for sock caches... i'll skip this for now.
907 */
908 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
909 if (sk) {
910 int ret;
911
912 if (!uh->check && !udp_sk(sk)->no_check6_rx) {
913 sock_put(sk);
914 udp6_csum_zero_error(skb);
915 goto csum_error;
916 }
917
918 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
919 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
920 ip6_compute_pseudo);
921
922 ret = udpv6_queue_rcv_skb(sk, skb);
923 sock_put(sk);
924
925 /* a return value > 0 means to resubmit the input */
926 if (ret > 0)
927 return ret;
928
929 return 0;
930 }
931
932 if (!uh->check) {
933 udp6_csum_zero_error(skb);
934 goto csum_error;
935 }
936
937 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
938 goto discard;
939
940 if (udp_lib_checksum_complete(skb))
941 goto csum_error;
942
943 UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
944 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
945
946 kfree_skb(skb);
947 return 0;
948
949short_packet:
950 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
951 proto == IPPROTO_UDPLITE ? "-Lite" : "",
952 saddr, ntohs(uh->source),
953 ulen, skb->len,
954 daddr, ntohs(uh->dest));
955 goto discard;
956csum_error:
957 UDP6_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
958discard:
959 UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
960 kfree_skb(skb);
961 return 0;
962}
963
964static __inline__ int udpv6_rcv(struct sk_buff *skb)
965{
966 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
967}
968
969/*
970 * Throw away all pending data and cancel the corking. Socket is locked.
971 */
972static void udp_v6_flush_pending_frames(struct sock *sk)
973{
974 struct udp_sock *up = udp_sk(sk);
975
976 if (up->pending == AF_INET)
977 udp_flush_pending_frames(sk);
978 else if (up->pending) {
979 up->len = 0;
980 up->pending = 0;
981 ip6_flush_pending_frames(sk);
982 }
983}
984
985/**
986 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
987 * @sk: socket we are sending on
988 * @skb: sk_buff containing the filled-in UDP header
989 * (checksum field must be zeroed out)
990 */
991static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
992 const struct in6_addr *saddr,
993 const struct in6_addr *daddr, int len)
994{
995 unsigned int offset;
996 struct udphdr *uh = udp_hdr(skb);
997 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
998 __wsum csum = 0;
999
1000 if (!frags) {
1001 /* Only one fragment on the socket. */
1002 skb->csum_start = skb_transport_header(skb) - skb->head;
1003 skb->csum_offset = offsetof(struct udphdr, check);
1004 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1005 } else {
1006 /*
1007 * HW-checksum won't work as there are two or more
1008 * fragments on the socket so that all csums of sk_buffs
1009 * should be together
1010 */
1011 offset = skb_transport_offset(skb);
1012 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1013
1014 skb->ip_summed = CHECKSUM_NONE;
1015
1016 do {
1017 csum = csum_add(csum, frags->csum);
1018 } while ((frags = frags->next));
1019
1020 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1021 csum);
1022 if (uh->check == 0)
1023 uh->check = CSUM_MANGLED_0;
1024 }
1025}
1026
1027/*
1028 * Sending
1029 */
1030
1031static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6)
1032{
1033 struct sock *sk = skb->sk;
1034 struct udphdr *uh;
1035 int err = 0;
1036 int is_udplite = IS_UDPLITE(sk);
1037 __wsum csum = 0;
1038 int offset = skb_transport_offset(skb);
1039 int len = skb->len - offset;
1040
1041 /*
1042 * Create a UDP header
1043 */
1044 uh = udp_hdr(skb);
1045 uh->source = fl6->fl6_sport;
1046 uh->dest = fl6->fl6_dport;
1047 uh->len = htons(len);
1048 uh->check = 0;
1049
1050 if (is_udplite)
1051 csum = udplite_csum(skb);
1052 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */
1053 skb->ip_summed = CHECKSUM_NONE;
1054 goto send;
1055 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1056 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1057 goto send;
1058 } else
1059 csum = udp_csum(skb);
1060
1061 /* add protocol-dependent pseudo-header */
1062 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1063 len, fl6->flowi6_proto, csum);
1064 if (uh->check == 0)
1065 uh->check = CSUM_MANGLED_0;
1066
1067send:
1068 err = ip6_send_skb(skb);
1069 if (err) {
1070 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1071 UDP6_INC_STATS_USER(sock_net(sk),
1072 UDP_MIB_SNDBUFERRORS, is_udplite);
1073 err = 0;
1074 }
1075 } else
1076 UDP6_INC_STATS_USER(sock_net(sk),
1077 UDP_MIB_OUTDATAGRAMS, is_udplite);
1078 return err;
1079}
1080
1081static int udp_v6_push_pending_frames(struct sock *sk)
1082{
1083 struct sk_buff *skb;
1084 struct udp_sock *up = udp_sk(sk);
1085 struct flowi6 fl6;
1086 int err = 0;
1087
1088 if (up->pending == AF_INET)
1089 return udp_push_pending_frames(sk);
1090
1091 /* ip6_finish_skb will release the cork, so make a copy of
1092 * fl6 here.
1093 */
1094 fl6 = inet_sk(sk)->cork.fl.u.ip6;
1095
1096 skb = ip6_finish_skb(sk);
1097 if (!skb)
1098 goto out;
1099
1100 err = udp_v6_send_skb(skb, &fl6);
1101
1102out:
1103 up->len = 0;
1104 up->pending = 0;
1105 return err;
1106}
1107
1108int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1109{
1110 struct ipv6_txoptions opt_space;
1111 struct udp_sock *up = udp_sk(sk);
1112 struct inet_sock *inet = inet_sk(sk);
1113 struct ipv6_pinfo *np = inet6_sk(sk);
1114 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1115 struct in6_addr *daddr, *final_p, final;
1116 struct ipv6_txoptions *opt = NULL;
1117 struct ipv6_txoptions *opt_to_free = NULL;
1118 struct ip6_flowlabel *flowlabel = NULL;
1119 struct flowi6 fl6;
1120 struct dst_entry *dst;
1121 int addr_len = msg->msg_namelen;
1122 int ulen = len;
1123 int hlimit = -1;
1124 int tclass = -1;
1125 int dontfrag = -1;
1126 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1127 int err;
1128 int connected = 0;
1129 int is_udplite = IS_UDPLITE(sk);
1130 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1131
1132 /* destination address check */
1133 if (sin6) {
1134 if (addr_len < offsetof(struct sockaddr, sa_data))
1135 return -EINVAL;
1136
1137 switch (sin6->sin6_family) {
1138 case AF_INET6:
1139 if (addr_len < SIN6_LEN_RFC2133)
1140 return -EINVAL;
1141 daddr = &sin6->sin6_addr;
1142 break;
1143 case AF_INET:
1144 goto do_udp_sendmsg;
1145 case AF_UNSPEC:
1146 msg->msg_name = sin6 = NULL;
1147 msg->msg_namelen = addr_len = 0;
1148 daddr = NULL;
1149 break;
1150 default:
1151 return -EINVAL;
1152 }
1153 } else if (!up->pending) {
1154 if (sk->sk_state != TCP_ESTABLISHED)
1155 return -EDESTADDRREQ;
1156 daddr = &sk->sk_v6_daddr;
1157 } else
1158 daddr = NULL;
1159
1160 if (daddr) {
1161 if (ipv6_addr_v4mapped(daddr)) {
1162 struct sockaddr_in sin;
1163 sin.sin_family = AF_INET;
1164 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1165 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1166 msg->msg_name = &sin;
1167 msg->msg_namelen = sizeof(sin);
1168do_udp_sendmsg:
1169 if (__ipv6_only_sock(sk))
1170 return -ENETUNREACH;
1171 return udp_sendmsg(sk, msg, len);
1172 }
1173 }
1174
1175 if (up->pending == AF_INET)
1176 return udp_sendmsg(sk, msg, len);
1177
1178 /* Rough check on arithmetic overflow,
1179 better check is made in ip6_append_data().
1180 */
1181 if (len > INT_MAX - sizeof(struct udphdr))
1182 return -EMSGSIZE;
1183
1184 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1185 if (up->pending) {
1186 /*
1187 * There are pending frames.
1188 * The socket lock must be held while it's corked.
1189 */
1190 lock_sock(sk);
1191 if (likely(up->pending)) {
1192 if (unlikely(up->pending != AF_INET6)) {
1193 release_sock(sk);
1194 return -EAFNOSUPPORT;
1195 }
1196 dst = NULL;
1197 goto do_append_data;
1198 }
1199 release_sock(sk);
1200 }
1201 ulen += sizeof(struct udphdr);
1202
1203 memset(&fl6, 0, sizeof(fl6));
1204
1205 if (sin6) {
1206 if (sin6->sin6_port == 0)
1207 return -EINVAL;
1208
1209 fl6.fl6_dport = sin6->sin6_port;
1210 daddr = &sin6->sin6_addr;
1211
1212 if (np->sndflow) {
1213 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1214 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1215 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1216 if (!flowlabel)
1217 return -EINVAL;
1218 }
1219 }
1220
1221 /*
1222 * Otherwise it will be difficult to maintain
1223 * sk->sk_dst_cache.
1224 */
1225 if (sk->sk_state == TCP_ESTABLISHED &&
1226 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1227 daddr = &sk->sk_v6_daddr;
1228
1229 if (addr_len >= sizeof(struct sockaddr_in6) &&
1230 sin6->sin6_scope_id &&
1231 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1232 fl6.flowi6_oif = sin6->sin6_scope_id;
1233 } else {
1234 if (sk->sk_state != TCP_ESTABLISHED)
1235 return -EDESTADDRREQ;
1236
1237 fl6.fl6_dport = inet->inet_dport;
1238 daddr = &sk->sk_v6_daddr;
1239 fl6.flowlabel = np->flow_label;
1240 connected = 1;
1241 }
1242
1243 if (!fl6.flowi6_oif)
1244 fl6.flowi6_oif = sk->sk_bound_dev_if;
1245
1246 if (!fl6.flowi6_oif)
1247 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1248
1249 fl6.flowi6_mark = sk->sk_mark;
1250
1251 if (msg->msg_controllen) {
1252 opt = &opt_space;
1253 memset(opt, 0, sizeof(struct ipv6_txoptions));
1254 opt->tot_len = sizeof(*opt);
1255
1256 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt,
1257 &hlimit, &tclass, &dontfrag);
1258 if (err < 0) {
1259 fl6_sock_release(flowlabel);
1260 return err;
1261 }
1262 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1263 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1264 if (!flowlabel)
1265 return -EINVAL;
1266 }
1267 if (!(opt->opt_nflen|opt->opt_flen))
1268 opt = NULL;
1269 connected = 0;
1270 }
1271 if (!opt) {
1272 opt = txopt_get(np);
1273 opt_to_free = opt;
1274 }
1275 if (flowlabel)
1276 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1277 opt = ipv6_fixup_options(&opt_space, opt);
1278
1279 fl6.flowi6_proto = sk->sk_protocol;
1280 if (!ipv6_addr_any(daddr))
1281 fl6.daddr = *daddr;
1282 else
1283 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1284 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1285 fl6.saddr = np->saddr;
1286 fl6.fl6_sport = inet->inet_sport;
1287
1288 final_p = fl6_update_dst(&fl6, opt, &final);
1289 if (final_p)
1290 connected = 0;
1291
1292 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1293 fl6.flowi6_oif = np->mcast_oif;
1294 connected = 0;
1295 } else if (!fl6.flowi6_oif)
1296 fl6.flowi6_oif = np->ucast_oif;
1297
1298 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1299
1300 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p);
1301 if (IS_ERR(dst)) {
1302 err = PTR_ERR(dst);
1303 dst = NULL;
1304 goto out;
1305 }
1306
1307 if (hlimit < 0)
1308 hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1309
1310 if (tclass < 0)
1311 tclass = np->tclass;
1312
1313 if (msg->msg_flags&MSG_CONFIRM)
1314 goto do_confirm;
1315back_from_confirm:
1316
1317 /* Lockless fast path for the non-corking case */
1318 if (!corkreq) {
1319 struct sk_buff *skb;
1320
1321 skb = ip6_make_skb(sk, getfrag, msg, ulen,
1322 sizeof(struct udphdr), hlimit, tclass, opt,
1323 &fl6, (struct rt6_info *)dst,
1324 msg->msg_flags, dontfrag);
1325 err = PTR_ERR(skb);
1326 if (!IS_ERR_OR_NULL(skb))
1327 err = udp_v6_send_skb(skb, &fl6);
1328 goto release_dst;
1329 }
1330
1331 lock_sock(sk);
1332 if (unlikely(up->pending)) {
1333 /* The socket is already corked while preparing it. */
1334 /* ... which is an evident application bug. --ANK */
1335 release_sock(sk);
1336
1337 net_dbg_ratelimited("udp cork app bug 2\n");
1338 err = -EINVAL;
1339 goto out;
1340 }
1341
1342 up->pending = AF_INET6;
1343
1344do_append_data:
1345 if (dontfrag < 0)
1346 dontfrag = np->dontfrag;
1347 up->len += ulen;
1348 err = ip6_append_data(sk, getfrag, msg, ulen,
1349 sizeof(struct udphdr), hlimit, tclass, opt, &fl6,
1350 (struct rt6_info *)dst,
1351 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag);
1352 if (err)
1353 udp_v6_flush_pending_frames(sk);
1354 else if (!corkreq)
1355 err = udp_v6_push_pending_frames(sk);
1356 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1357 up->pending = 0;
1358
1359 if (err > 0)
1360 err = np->recverr ? net_xmit_errno(err) : 0;
1361 release_sock(sk);
1362
1363release_dst:
1364 if (dst) {
1365 if (connected) {
1366 ip6_dst_store(sk, dst,
1367 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ?
1368 &sk->sk_v6_daddr : NULL,
1369#ifdef CONFIG_IPV6_SUBTREES
1370 ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1371 &np->saddr :
1372#endif
1373 NULL);
1374 } else {
1375 dst_release(dst);
1376 }
1377 dst = NULL;
1378 }
1379
1380out:
1381 dst_release(dst);
1382 fl6_sock_release(flowlabel);
1383 txopt_put(opt_to_free);
1384 if (!err)
1385 return len;
1386 /*
1387 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1388 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1389 * we don't have a good statistic (IpOutDiscards but it can be too many
1390 * things). We could add another new stat but at least for now that
1391 * seems like overkill.
1392 */
1393 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1394 UDP6_INC_STATS_USER(sock_net(sk),
1395 UDP_MIB_SNDBUFERRORS, is_udplite);
1396 }
1397 return err;
1398
1399do_confirm:
1400 dst_confirm(dst);
1401 if (!(msg->msg_flags&MSG_PROBE) || len)
1402 goto back_from_confirm;
1403 err = 0;
1404 goto out;
1405}
1406
1407void udpv6_destroy_sock(struct sock *sk)
1408{
1409 struct udp_sock *up = udp_sk(sk);
1410 lock_sock(sk);
1411 udp_v6_flush_pending_frames(sk);
1412 release_sock(sk);
1413
1414 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
1415 void (*encap_destroy)(struct sock *sk);
1416 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1417 if (encap_destroy)
1418 encap_destroy(sk);
1419 }
1420
1421 inet6_destroy_sock(sk);
1422}
1423
1424/*
1425 * Socket option code for UDP
1426 */
1427int udpv6_setsockopt(struct sock *sk, int level, int optname,
1428 char __user *optval, unsigned int optlen)
1429{
1430 if (level == SOL_UDP || level == SOL_UDPLITE)
1431 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1432 udp_v6_push_pending_frames);
1433 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1434}
1435
1436#ifdef CONFIG_COMPAT
1437int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1438 char __user *optval, unsigned int optlen)
1439{
1440 if (level == SOL_UDP || level == SOL_UDPLITE)
1441 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1442 udp_v6_push_pending_frames);
1443 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1444}
1445#endif
1446
1447int udpv6_getsockopt(struct sock *sk, int level, int optname,
1448 char __user *optval, int __user *optlen)
1449{
1450 if (level == SOL_UDP || level == SOL_UDPLITE)
1451 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1452 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1453}
1454
1455#ifdef CONFIG_COMPAT
1456int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1457 char __user *optval, int __user *optlen)
1458{
1459 if (level == SOL_UDP || level == SOL_UDPLITE)
1460 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1461 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1462}
1463#endif
1464
1465static const struct inet6_protocol udpv6_protocol = {
1466 .handler = udpv6_rcv,
1467 .err_handler = udpv6_err,
1468 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1469};
1470
1471/* ------------------------------------------------------------------------ */
1472#ifdef CONFIG_PROC_FS
1473int udp6_seq_show(struct seq_file *seq, void *v)
1474{
1475 if (v == SEQ_START_TOKEN) {
1476 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1477 } else {
1478 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1479 struct inet_sock *inet = inet_sk(v);
1480 __u16 srcp = ntohs(inet->inet_sport);
1481 __u16 destp = ntohs(inet->inet_dport);
1482 ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket);
1483 }
1484 return 0;
1485}
1486
1487static const struct file_operations udp6_afinfo_seq_fops = {
1488 .owner = THIS_MODULE,
1489 .open = udp_seq_open,
1490 .read = seq_read,
1491 .llseek = seq_lseek,
1492 .release = seq_release_net
1493};
1494
1495static struct udp_seq_afinfo udp6_seq_afinfo = {
1496 .name = "udp6",
1497 .family = AF_INET6,
1498 .udp_table = &udp_table,
1499 .seq_fops = &udp6_afinfo_seq_fops,
1500 .seq_ops = {
1501 .show = udp6_seq_show,
1502 },
1503};
1504
1505int __net_init udp6_proc_init(struct net *net)
1506{
1507 return udp_proc_register(net, &udp6_seq_afinfo);
1508}
1509
1510void udp6_proc_exit(struct net *net)
1511{
1512 udp_proc_unregister(net, &udp6_seq_afinfo);
1513}
1514#endif /* CONFIG_PROC_FS */
1515
1516void udp_v6_clear_sk(struct sock *sk, int size)
1517{
1518 struct inet_sock *inet = inet_sk(sk);
1519
1520 /* we do not want to clear pinet6 field, because of RCU lookups */
1521 sk_prot_clear_portaddr_nulls(sk, offsetof(struct inet_sock, pinet6));
1522
1523 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1524 memset(&inet->pinet6 + 1, 0, size);
1525}
1526
1527/* ------------------------------------------------------------------------ */
1528
1529struct proto udpv6_prot = {
1530 .name = "UDPv6",
1531 .owner = THIS_MODULE,
1532 .close = udp_lib_close,
1533 .connect = ip6_datagram_connect,
1534 .disconnect = udp_disconnect,
1535 .ioctl = udp_ioctl,
1536 .destroy = udpv6_destroy_sock,
1537 .setsockopt = udpv6_setsockopt,
1538 .getsockopt = udpv6_getsockopt,
1539 .sendmsg = udpv6_sendmsg,
1540 .recvmsg = udpv6_recvmsg,
1541 .backlog_rcv = __udpv6_queue_rcv_skb,
1542 .release_cb = ip6_datagram_release_cb,
1543 .hash = udp_lib_hash,
1544 .unhash = udp_lib_unhash,
1545 .rehash = udp_v6_rehash,
1546 .get_port = udp_v6_get_port,
1547 .memory_allocated = &udp_memory_allocated,
1548 .sysctl_mem = sysctl_udp_mem,
1549 .sysctl_wmem = &sysctl_udp_wmem_min,
1550 .sysctl_rmem = &sysctl_udp_rmem_min,
1551 .obj_size = sizeof(struct udp6_sock),
1552 .slab_flags = SLAB_DESTROY_BY_RCU,
1553 .h.udp_table = &udp_table,
1554#ifdef CONFIG_COMPAT
1555 .compat_setsockopt = compat_udpv6_setsockopt,
1556 .compat_getsockopt = compat_udpv6_getsockopt,
1557#endif
1558 .clear_sk = udp_v6_clear_sk,
1559};
1560
1561static struct inet_protosw udpv6_protosw = {
1562 .type = SOCK_DGRAM,
1563 .protocol = IPPROTO_UDP,
1564 .prot = &udpv6_prot,
1565 .ops = &inet6_dgram_ops,
1566 .flags = INET_PROTOSW_PERMANENT,
1567};
1568
1569int __init udpv6_init(void)
1570{
1571 int ret;
1572
1573 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1574 if (ret)
1575 goto out;
1576
1577 ret = inet6_register_protosw(&udpv6_protosw);
1578 if (ret)
1579 goto out_udpv6_protocol;
1580out:
1581 return ret;
1582
1583out_udpv6_protocol:
1584 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1585 goto out;
1586}
1587
1588void udpv6_exit(void)
1589{
1590 inet6_unregister_protosw(&udpv6_protosw);
1591 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1592}
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 <linux/uaccess.h>
39
40#include <net/addrconf.h>
41#include <net/ndisc.h>
42#include <net/protocol.h>
43#include <net/transp_v6.h>
44#include <net/ip6_route.h>
45#include <net/raw.h>
46#include <net/tcp_states.h>
47#include <net/ip6_checksum.h>
48#include <net/xfrm.h>
49#include <net/inet_hashtables.h>
50#include <net/inet6_hashtables.h>
51#include <net/busy_poll.h>
52#include <net/sock_reuseport.h>
53
54#include <linux/proc_fs.h>
55#include <linux/seq_file.h>
56#include <trace/events/skb.h>
57#include "udp_impl.h"
58
59static bool udp6_lib_exact_dif_match(struct net *net, struct sk_buff *skb)
60{
61#if defined(CONFIG_NET_L3_MASTER_DEV)
62 if (!net->ipv4.sysctl_udp_l3mdev_accept &&
63 skb && ipv6_l3mdev_skb(IP6CB(skb)->flags))
64 return true;
65#endif
66 return false;
67}
68
69static u32 udp6_ehashfn(const struct net *net,
70 const struct in6_addr *laddr,
71 const u16 lport,
72 const struct in6_addr *faddr,
73 const __be16 fport)
74{
75 static u32 udp6_ehash_secret __read_mostly;
76 static u32 udp_ipv6_hash_secret __read_mostly;
77
78 u32 lhash, fhash;
79
80 net_get_random_once(&udp6_ehash_secret,
81 sizeof(udp6_ehash_secret));
82 net_get_random_once(&udp_ipv6_hash_secret,
83 sizeof(udp_ipv6_hash_secret));
84
85 lhash = (__force u32)laddr->s6_addr32[3];
86 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
87
88 return __inet6_ehashfn(lhash, lport, fhash, fport,
89 udp_ipv6_hash_secret + net_hash_mix(net));
90}
91
92int udp_v6_get_port(struct sock *sk, unsigned short snum)
93{
94 unsigned int hash2_nulladdr =
95 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
96 unsigned int hash2_partial =
97 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
98
99 /* precompute partial secondary hash */
100 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
101 return udp_lib_get_port(sk, snum, hash2_nulladdr);
102}
103
104static void udp_v6_rehash(struct sock *sk)
105{
106 u16 new_hash = ipv6_portaddr_hash(sock_net(sk),
107 &sk->sk_v6_rcv_saddr,
108 inet_sk(sk)->inet_num);
109
110 udp_lib_rehash(sk, new_hash);
111}
112
113static int compute_score(struct sock *sk, struct net *net,
114 const struct in6_addr *saddr, __be16 sport,
115 const struct in6_addr *daddr, unsigned short hnum,
116 int dif, int sdif, bool exact_dif)
117{
118 int score;
119 struct inet_sock *inet;
120
121 if (!net_eq(sock_net(sk), net) ||
122 udp_sk(sk)->udp_port_hash != hnum ||
123 sk->sk_family != PF_INET6)
124 return -1;
125
126 score = 0;
127 inet = inet_sk(sk);
128
129 if (inet->inet_dport) {
130 if (inet->inet_dport != sport)
131 return -1;
132 score++;
133 }
134
135 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
136 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
137 return -1;
138 score++;
139 }
140
141 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
142 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
143 return -1;
144 score++;
145 }
146
147 if (sk->sk_bound_dev_if || exact_dif) {
148 bool dev_match = (sk->sk_bound_dev_if == dif ||
149 sk->sk_bound_dev_if == sdif);
150
151 if (!dev_match)
152 return -1;
153 if (sk->sk_bound_dev_if)
154 score++;
155 }
156
157 if (sk->sk_incoming_cpu == raw_smp_processor_id())
158 score++;
159
160 return score;
161}
162
163/* called with rcu_read_lock() */
164static struct sock *udp6_lib_lookup2(struct net *net,
165 const struct in6_addr *saddr, __be16 sport,
166 const struct in6_addr *daddr, unsigned int hnum,
167 int dif, int sdif, bool exact_dif,
168 struct udp_hslot *hslot2, struct sk_buff *skb)
169{
170 struct sock *sk, *result;
171 int score, badness;
172 u32 hash = 0;
173
174 result = NULL;
175 badness = -1;
176 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
177 score = compute_score(sk, net, saddr, sport,
178 daddr, hnum, dif, sdif, exact_dif);
179 if (score > badness) {
180 if (sk->sk_reuseport) {
181 hash = udp6_ehashfn(net, daddr, hnum,
182 saddr, sport);
183
184 result = reuseport_select_sock(sk, hash, skb,
185 sizeof(struct udphdr));
186 if (result)
187 return result;
188 }
189 result = sk;
190 badness = score;
191 }
192 }
193 return result;
194}
195
196/* rcu_read_lock() must be held */
197struct sock *__udp6_lib_lookup(struct net *net,
198 const struct in6_addr *saddr, __be16 sport,
199 const struct in6_addr *daddr, __be16 dport,
200 int dif, int sdif, struct udp_table *udptable,
201 struct sk_buff *skb)
202{
203 struct sock *sk, *result;
204 unsigned short hnum = ntohs(dport);
205 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
206 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
207 bool exact_dif = udp6_lib_exact_dif_match(net, skb);
208 int score, badness;
209 u32 hash = 0;
210
211 if (hslot->count > 10) {
212 hash2 = ipv6_portaddr_hash(net, daddr, hnum);
213 slot2 = hash2 & udptable->mask;
214 hslot2 = &udptable->hash2[slot2];
215 if (hslot->count < hslot2->count)
216 goto begin;
217
218 result = udp6_lib_lookup2(net, saddr, sport,
219 daddr, hnum, dif, sdif, exact_dif,
220 hslot2, skb);
221 if (!result) {
222 unsigned int old_slot2 = slot2;
223 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
224 slot2 = hash2 & udptable->mask;
225 /* avoid searching the same slot again. */
226 if (unlikely(slot2 == old_slot2))
227 return result;
228
229 hslot2 = &udptable->hash2[slot2];
230 if (hslot->count < hslot2->count)
231 goto begin;
232
233 result = udp6_lib_lookup2(net, saddr, sport,
234 daddr, hnum, dif, sdif,
235 exact_dif, hslot2,
236 skb);
237 }
238 return result;
239 }
240begin:
241 result = NULL;
242 badness = -1;
243 sk_for_each_rcu(sk, &hslot->head) {
244 score = compute_score(sk, net, saddr, sport, daddr, hnum, dif,
245 sdif, exact_dif);
246 if (score > badness) {
247 if (sk->sk_reuseport) {
248 hash = udp6_ehashfn(net, daddr, hnum,
249 saddr, sport);
250 result = reuseport_select_sock(sk, hash, skb,
251 sizeof(struct udphdr));
252 if (result)
253 return result;
254 }
255 result = sk;
256 badness = score;
257 }
258 }
259 return result;
260}
261EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
262
263static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
264 __be16 sport, __be16 dport,
265 struct udp_table *udptable)
266{
267 const struct ipv6hdr *iph = ipv6_hdr(skb);
268
269 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
270 &iph->daddr, dport, inet6_iif(skb),
271 inet6_sdif(skb), udptable, skb);
272}
273
274struct sock *udp6_lib_lookup_skb(struct sk_buff *skb,
275 __be16 sport, __be16 dport)
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), &udp_table, skb);
282}
283EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb);
284
285/* Must be called under rcu_read_lock().
286 * Does increment socket refcount.
287 */
288#if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
289 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \
290 IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
291struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
292 const struct in6_addr *daddr, __be16 dport, int dif)
293{
294 struct sock *sk;
295
296 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport,
297 dif, 0, &udp_table, NULL);
298 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
299 sk = NULL;
300 return sk;
301}
302EXPORT_SYMBOL_GPL(udp6_lib_lookup);
303#endif
304
305/* do not use the scratch area len for jumbogram: their length execeeds the
306 * scratch area space; note that the IP6CB flags is still in the first
307 * cacheline, so checking for jumbograms is cheap
308 */
309static int udp6_skb_len(struct sk_buff *skb)
310{
311 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb);
312}
313
314/*
315 * This should be easy, if there is something there we
316 * return it, otherwise we block.
317 */
318
319int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
320 int noblock, int flags, int *addr_len)
321{
322 struct ipv6_pinfo *np = inet6_sk(sk);
323 struct inet_sock *inet = inet_sk(sk);
324 struct sk_buff *skb;
325 unsigned int ulen, copied;
326 int peeked, peeking, off;
327 int err;
328 int is_udplite = IS_UDPLITE(sk);
329 bool checksum_valid = false;
330 int is_udp4;
331
332 if (flags & MSG_ERRQUEUE)
333 return ipv6_recv_error(sk, msg, len, addr_len);
334
335 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
336 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
337
338try_again:
339 peeking = flags & MSG_PEEK;
340 off = sk_peek_offset(sk, flags);
341 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
342 if (!skb)
343 return err;
344
345 ulen = udp6_skb_len(skb);
346 copied = len;
347 if (copied > ulen - off)
348 copied = ulen - off;
349 else if (copied < ulen)
350 msg->msg_flags |= MSG_TRUNC;
351
352 is_udp4 = (skb->protocol == htons(ETH_P_IP));
353
354 /*
355 * If checksum is needed at all, try to do it while copying the
356 * data. If the data is truncated, or if we only want a partial
357 * coverage checksum (UDP-Lite), do it before the copy.
358 */
359
360 if (copied < ulen || peeking ||
361 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
362 checksum_valid = udp_skb_csum_unnecessary(skb) ||
363 !__udp_lib_checksum_complete(skb);
364 if (!checksum_valid)
365 goto csum_copy_err;
366 }
367
368 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
369 if (udp_skb_is_linear(skb))
370 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
371 else
372 err = skb_copy_datagram_msg(skb, off, msg, copied);
373 } else {
374 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
375 if (err == -EINVAL)
376 goto csum_copy_err;
377 }
378 if (unlikely(err)) {
379 if (!peeked) {
380 atomic_inc(&sk->sk_drops);
381 if (is_udp4)
382 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
383 is_udplite);
384 else
385 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
386 is_udplite);
387 }
388 kfree_skb(skb);
389 return err;
390 }
391 if (!peeked) {
392 if (is_udp4)
393 UDP_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS,
394 is_udplite);
395 else
396 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS,
397 is_udplite);
398 }
399
400 sock_recv_ts_and_drops(msg, sk, skb);
401
402 /* Copy the address. */
403 if (msg->msg_name) {
404 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
405 sin6->sin6_family = AF_INET6;
406 sin6->sin6_port = udp_hdr(skb)->source;
407 sin6->sin6_flowinfo = 0;
408
409 if (is_udp4) {
410 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
411 &sin6->sin6_addr);
412 sin6->sin6_scope_id = 0;
413 } else {
414 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
415 sin6->sin6_scope_id =
416 ipv6_iface_scope_id(&sin6->sin6_addr,
417 inet6_iif(skb));
418 }
419 *addr_len = sizeof(*sin6);
420 }
421
422 if (np->rxopt.all)
423 ip6_datagram_recv_common_ctl(sk, msg, skb);
424
425 if (is_udp4) {
426 if (inet->cmsg_flags)
427 ip_cmsg_recv_offset(msg, sk, skb,
428 sizeof(struct udphdr), off);
429 } else {
430 if (np->rxopt.all)
431 ip6_datagram_recv_specific_ctl(sk, msg, skb);
432 }
433
434 err = copied;
435 if (flags & MSG_TRUNC)
436 err = ulen;
437
438 skb_consume_udp(sk, skb, peeking ? -err : err);
439 return err;
440
441csum_copy_err:
442 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
443 udp_skb_destructor)) {
444 if (is_udp4) {
445 UDP_INC_STATS(sock_net(sk),
446 UDP_MIB_CSUMERRORS, is_udplite);
447 UDP_INC_STATS(sock_net(sk),
448 UDP_MIB_INERRORS, is_udplite);
449 } else {
450 UDP6_INC_STATS(sock_net(sk),
451 UDP_MIB_CSUMERRORS, is_udplite);
452 UDP6_INC_STATS(sock_net(sk),
453 UDP_MIB_INERRORS, is_udplite);
454 }
455 }
456 kfree_skb(skb);
457
458 /* starting over for a new packet, but check if we need to yield */
459 cond_resched();
460 msg->msg_flags &= ~MSG_TRUNC;
461 goto try_again;
462}
463
464void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
465 u8 type, u8 code, int offset, __be32 info,
466 struct udp_table *udptable)
467{
468 struct ipv6_pinfo *np;
469 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
470 const struct in6_addr *saddr = &hdr->saddr;
471 const struct in6_addr *daddr = &hdr->daddr;
472 struct udphdr *uh = (struct udphdr *)(skb->data+offset);
473 struct sock *sk;
474 int harderr;
475 int err;
476 struct net *net = dev_net(skb->dev);
477
478 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
479 inet6_iif(skb), 0, udptable, skb);
480 if (!sk) {
481 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
482 ICMP6_MIB_INERRORS);
483 return;
484 }
485
486 harderr = icmpv6_err_convert(type, code, &err);
487 np = inet6_sk(sk);
488
489 if (type == ICMPV6_PKT_TOOBIG) {
490 if (!ip6_sk_accept_pmtu(sk))
491 goto out;
492 ip6_sk_update_pmtu(skb, sk, info);
493 if (np->pmtudisc != IPV6_PMTUDISC_DONT)
494 harderr = 1;
495 }
496 if (type == NDISC_REDIRECT) {
497 ip6_sk_redirect(skb, sk);
498 goto out;
499 }
500
501 if (!np->recverr) {
502 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
503 goto out;
504 } else {
505 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
506 }
507
508 sk->sk_err = err;
509 sk->sk_error_report(sk);
510out:
511 return;
512}
513
514static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
515{
516 int rc;
517
518 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
519 sock_rps_save_rxhash(sk, skb);
520 sk_mark_napi_id(sk, skb);
521 sk_incoming_cpu_update(sk);
522 } else {
523 sk_mark_napi_id_once(sk, skb);
524 }
525
526 rc = __udp_enqueue_schedule_skb(sk, skb);
527 if (rc < 0) {
528 int is_udplite = IS_UDPLITE(sk);
529
530 /* Note that an ENOMEM error is charged twice */
531 if (rc == -ENOMEM)
532 UDP6_INC_STATS(sock_net(sk),
533 UDP_MIB_RCVBUFERRORS, is_udplite);
534 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
535 kfree_skb(skb);
536 return -1;
537 }
538
539 return 0;
540}
541
542static __inline__ void udpv6_err(struct sk_buff *skb,
543 struct inet6_skb_parm *opt, u8 type,
544 u8 code, int offset, __be32 info)
545{
546 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
547}
548
549static struct static_key udpv6_encap_needed __read_mostly;
550void udpv6_encap_enable(void)
551{
552 static_key_enable(&udpv6_encap_needed);
553}
554EXPORT_SYMBOL(udpv6_encap_enable);
555
556static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
557{
558 struct udp_sock *up = udp_sk(sk);
559 int is_udplite = IS_UDPLITE(sk);
560
561 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
562 goto drop;
563
564 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
565 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
566
567 /*
568 * This is an encapsulation socket so pass the skb to
569 * the socket's udp_encap_rcv() hook. Otherwise, just
570 * fall through and pass this up the UDP socket.
571 * up->encap_rcv() returns the following value:
572 * =0 if skb was successfully passed to the encap
573 * handler or was discarded by it.
574 * >0 if skb should be passed on to UDP.
575 * <0 if skb should be resubmitted as proto -N
576 */
577
578 /* if we're overly short, let UDP handle it */
579 encap_rcv = READ_ONCE(up->encap_rcv);
580 if (encap_rcv) {
581 int ret;
582
583 /* Verify checksum before giving to encap */
584 if (udp_lib_checksum_complete(skb))
585 goto csum_error;
586
587 ret = encap_rcv(sk, skb);
588 if (ret <= 0) {
589 __UDP_INC_STATS(sock_net(sk),
590 UDP_MIB_INDATAGRAMS,
591 is_udplite);
592 return -ret;
593 }
594 }
595
596 /* FALLTHROUGH -- it's a UDP Packet */
597 }
598
599 /*
600 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
601 */
602 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
603
604 if (up->pcrlen == 0) { /* full coverage was set */
605 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
606 UDP_SKB_CB(skb)->cscov, skb->len);
607 goto drop;
608 }
609 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
610 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
611 UDP_SKB_CB(skb)->cscov, up->pcrlen);
612 goto drop;
613 }
614 }
615
616 prefetch(&sk->sk_rmem_alloc);
617 if (rcu_access_pointer(sk->sk_filter) &&
618 udp_lib_checksum_complete(skb))
619 goto csum_error;
620
621 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
622 goto drop;
623
624 udp_csum_pull_header(skb);
625
626 skb_dst_drop(skb);
627
628 return __udpv6_queue_rcv_skb(sk, skb);
629
630csum_error:
631 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
632drop:
633 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
634 atomic_inc(&sk->sk_drops);
635 kfree_skb(skb);
636 return -1;
637}
638
639static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
640 __be16 loc_port, const struct in6_addr *loc_addr,
641 __be16 rmt_port, const struct in6_addr *rmt_addr,
642 int dif, unsigned short hnum)
643{
644 struct inet_sock *inet = inet_sk(sk);
645
646 if (!net_eq(sock_net(sk), net))
647 return false;
648
649 if (udp_sk(sk)->udp_port_hash != hnum ||
650 sk->sk_family != PF_INET6 ||
651 (inet->inet_dport && inet->inet_dport != rmt_port) ||
652 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
653 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
654 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif) ||
655 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
656 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
657 return false;
658 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
659 return false;
660 return true;
661}
662
663static void udp6_csum_zero_error(struct sk_buff *skb)
664{
665 /* RFC 2460 section 8.1 says that we SHOULD log
666 * this error. Well, it is reasonable.
667 */
668 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
669 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
670 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
671}
672
673/*
674 * Note: called only from the BH handler context,
675 * so we don't need to lock the hashes.
676 */
677static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
678 const struct in6_addr *saddr, const struct in6_addr *daddr,
679 struct udp_table *udptable, int proto)
680{
681 struct sock *sk, *first = NULL;
682 const struct udphdr *uh = udp_hdr(skb);
683 unsigned short hnum = ntohs(uh->dest);
684 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
685 unsigned int offset = offsetof(typeof(*sk), sk_node);
686 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
687 int dif = inet6_iif(skb);
688 struct hlist_node *node;
689 struct sk_buff *nskb;
690
691 if (use_hash2) {
692 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) &
693 udptable->mask;
694 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask;
695start_lookup:
696 hslot = &udptable->hash2[hash2];
697 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
698 }
699
700 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
701 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
702 uh->source, saddr, dif, hnum))
703 continue;
704 /* If zero checksum and no_check is not on for
705 * the socket then skip it.
706 */
707 if (!uh->check && !udp_sk(sk)->no_check6_rx)
708 continue;
709 if (!first) {
710 first = sk;
711 continue;
712 }
713 nskb = skb_clone(skb, GFP_ATOMIC);
714 if (unlikely(!nskb)) {
715 atomic_inc(&sk->sk_drops);
716 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
717 IS_UDPLITE(sk));
718 __UDP6_INC_STATS(net, UDP_MIB_INERRORS,
719 IS_UDPLITE(sk));
720 continue;
721 }
722
723 if (udpv6_queue_rcv_skb(sk, nskb) > 0)
724 consume_skb(nskb);
725 }
726
727 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
728 if (use_hash2 && hash2 != hash2_any) {
729 hash2 = hash2_any;
730 goto start_lookup;
731 }
732
733 if (first) {
734 if (udpv6_queue_rcv_skb(first, skb) > 0)
735 consume_skb(skb);
736 } else {
737 kfree_skb(skb);
738 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
739 proto == IPPROTO_UDPLITE);
740 }
741 return 0;
742}
743
744static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
745{
746 if (udp_sk_rx_dst_set(sk, dst)) {
747 const struct rt6_info *rt = (const struct rt6_info *)dst;
748
749 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
750 }
751}
752
753int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
754 int proto)
755{
756 const struct in6_addr *saddr, *daddr;
757 struct net *net = dev_net(skb->dev);
758 struct udphdr *uh;
759 struct sock *sk;
760 u32 ulen = 0;
761
762 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
763 goto discard;
764
765 saddr = &ipv6_hdr(skb)->saddr;
766 daddr = &ipv6_hdr(skb)->daddr;
767 uh = udp_hdr(skb);
768
769 ulen = ntohs(uh->len);
770 if (ulen > skb->len)
771 goto short_packet;
772
773 if (proto == IPPROTO_UDP) {
774 /* UDP validates ulen. */
775
776 /* Check for jumbo payload */
777 if (ulen == 0)
778 ulen = skb->len;
779
780 if (ulen < sizeof(*uh))
781 goto short_packet;
782
783 if (ulen < skb->len) {
784 if (pskb_trim_rcsum(skb, ulen))
785 goto short_packet;
786 saddr = &ipv6_hdr(skb)->saddr;
787 daddr = &ipv6_hdr(skb)->daddr;
788 uh = udp_hdr(skb);
789 }
790 }
791
792 if (udp6_csum_init(skb, uh, proto))
793 goto csum_error;
794
795 /* Check if the socket is already available, e.g. due to early demux */
796 sk = skb_steal_sock(skb);
797 if (sk) {
798 struct dst_entry *dst = skb_dst(skb);
799 int ret;
800
801 if (unlikely(sk->sk_rx_dst != dst))
802 udp6_sk_rx_dst_set(sk, dst);
803
804 ret = udpv6_queue_rcv_skb(sk, skb);
805 sock_put(sk);
806
807 /* a return value > 0 means to resubmit the input */
808 if (ret > 0)
809 return ret;
810 return 0;
811 }
812
813 /*
814 * Multicast receive code
815 */
816 if (ipv6_addr_is_multicast(daddr))
817 return __udp6_lib_mcast_deliver(net, skb,
818 saddr, daddr, udptable, proto);
819
820 /* Unicast */
821 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
822 if (sk) {
823 int ret;
824
825 if (!uh->check && !udp_sk(sk)->no_check6_rx) {
826 udp6_csum_zero_error(skb);
827 goto csum_error;
828 }
829
830 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
831 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
832 ip6_compute_pseudo);
833
834 ret = udpv6_queue_rcv_skb(sk, skb);
835
836 /* a return value > 0 means to resubmit the input */
837 if (ret > 0)
838 return ret;
839
840 return 0;
841 }
842
843 if (!uh->check) {
844 udp6_csum_zero_error(skb);
845 goto csum_error;
846 }
847
848 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
849 goto discard;
850
851 if (udp_lib_checksum_complete(skb))
852 goto csum_error;
853
854 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
855 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
856
857 kfree_skb(skb);
858 return 0;
859
860short_packet:
861 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
862 proto == IPPROTO_UDPLITE ? "-Lite" : "",
863 saddr, ntohs(uh->source),
864 ulen, skb->len,
865 daddr, ntohs(uh->dest));
866 goto discard;
867csum_error:
868 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
869discard:
870 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
871 kfree_skb(skb);
872 return 0;
873}
874
875
876static struct sock *__udp6_lib_demux_lookup(struct net *net,
877 __be16 loc_port, const struct in6_addr *loc_addr,
878 __be16 rmt_port, const struct in6_addr *rmt_addr,
879 int dif, int sdif)
880{
881 unsigned short hnum = ntohs(loc_port);
882 unsigned int hash2 = ipv6_portaddr_hash(net, loc_addr, hnum);
883 unsigned int slot2 = hash2 & udp_table.mask;
884 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
885 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
886 struct sock *sk;
887
888 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
889 if (sk->sk_state == TCP_ESTABLISHED &&
890 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif, sdif))
891 return sk;
892 /* Only check first socket in chain */
893 break;
894 }
895 return NULL;
896}
897
898static void udp_v6_early_demux(struct sk_buff *skb)
899{
900 struct net *net = dev_net(skb->dev);
901 const struct udphdr *uh;
902 struct sock *sk;
903 struct dst_entry *dst;
904 int dif = skb->dev->ifindex;
905 int sdif = inet6_sdif(skb);
906
907 if (!pskb_may_pull(skb, skb_transport_offset(skb) +
908 sizeof(struct udphdr)))
909 return;
910
911 uh = udp_hdr(skb);
912
913 if (skb->pkt_type == PACKET_HOST)
914 sk = __udp6_lib_demux_lookup(net, uh->dest,
915 &ipv6_hdr(skb)->daddr,
916 uh->source, &ipv6_hdr(skb)->saddr,
917 dif, sdif);
918 else
919 return;
920
921 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
922 return;
923
924 skb->sk = sk;
925 skb->destructor = sock_efree;
926 dst = READ_ONCE(sk->sk_rx_dst);
927
928 if (dst)
929 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
930 if (dst) {
931 /* set noref for now.
932 * any place which wants to hold dst has to call
933 * dst_hold_safe()
934 */
935 skb_dst_set_noref(skb, dst);
936 }
937}
938
939static __inline__ int udpv6_rcv(struct sk_buff *skb)
940{
941 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
942}
943
944/*
945 * Throw away all pending data and cancel the corking. Socket is locked.
946 */
947static void udp_v6_flush_pending_frames(struct sock *sk)
948{
949 struct udp_sock *up = udp_sk(sk);
950
951 if (up->pending == AF_INET)
952 udp_flush_pending_frames(sk);
953 else if (up->pending) {
954 up->len = 0;
955 up->pending = 0;
956 ip6_flush_pending_frames(sk);
957 }
958}
959
960static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
961 int addr_len)
962{
963 /* The following checks are replicated from __ip6_datagram_connect()
964 * and intended to prevent BPF program called below from accessing
965 * bytes that are out of the bound specified by user in addr_len.
966 */
967 if (uaddr->sa_family == AF_INET) {
968 if (__ipv6_only_sock(sk))
969 return -EAFNOSUPPORT;
970 return udp_pre_connect(sk, uaddr, addr_len);
971 }
972
973 if (addr_len < SIN6_LEN_RFC2133)
974 return -EINVAL;
975
976 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr);
977}
978
979/**
980 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
981 * @sk: socket we are sending on
982 * @skb: sk_buff containing the filled-in UDP header
983 * (checksum field must be zeroed out)
984 */
985static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
986 const struct in6_addr *saddr,
987 const struct in6_addr *daddr, int len)
988{
989 unsigned int offset;
990 struct udphdr *uh = udp_hdr(skb);
991 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
992 __wsum csum = 0;
993
994 if (!frags) {
995 /* Only one fragment on the socket. */
996 skb->csum_start = skb_transport_header(skb) - skb->head;
997 skb->csum_offset = offsetof(struct udphdr, check);
998 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
999 } else {
1000 /*
1001 * HW-checksum won't work as there are two or more
1002 * fragments on the socket so that all csums of sk_buffs
1003 * should be together
1004 */
1005 offset = skb_transport_offset(skb);
1006 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1007 csum = skb->csum;
1008
1009 skb->ip_summed = CHECKSUM_NONE;
1010
1011 do {
1012 csum = csum_add(csum, frags->csum);
1013 } while ((frags = frags->next));
1014
1015 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1016 csum);
1017 if (uh->check == 0)
1018 uh->check = CSUM_MANGLED_0;
1019 }
1020}
1021
1022/*
1023 * Sending
1024 */
1025
1026static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6)
1027{
1028 struct sock *sk = skb->sk;
1029 struct udphdr *uh;
1030 int err = 0;
1031 int is_udplite = IS_UDPLITE(sk);
1032 __wsum csum = 0;
1033 int offset = skb_transport_offset(skb);
1034 int len = skb->len - offset;
1035
1036 /*
1037 * Create a UDP header
1038 */
1039 uh = udp_hdr(skb);
1040 uh->source = fl6->fl6_sport;
1041 uh->dest = fl6->fl6_dport;
1042 uh->len = htons(len);
1043 uh->check = 0;
1044
1045 if (is_udplite)
1046 csum = udplite_csum(skb);
1047 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */
1048 skb->ip_summed = CHECKSUM_NONE;
1049 goto send;
1050 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1051 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1052 goto send;
1053 } else
1054 csum = udp_csum(skb);
1055
1056 /* add protocol-dependent pseudo-header */
1057 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1058 len, fl6->flowi6_proto, csum);
1059 if (uh->check == 0)
1060 uh->check = CSUM_MANGLED_0;
1061
1062send:
1063 err = ip6_send_skb(skb);
1064 if (err) {
1065 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1066 UDP6_INC_STATS(sock_net(sk),
1067 UDP_MIB_SNDBUFERRORS, is_udplite);
1068 err = 0;
1069 }
1070 } else {
1071 UDP6_INC_STATS(sock_net(sk),
1072 UDP_MIB_OUTDATAGRAMS, is_udplite);
1073 }
1074 return err;
1075}
1076
1077static int udp_v6_push_pending_frames(struct sock *sk)
1078{
1079 struct sk_buff *skb;
1080 struct udp_sock *up = udp_sk(sk);
1081 struct flowi6 fl6;
1082 int err = 0;
1083
1084 if (up->pending == AF_INET)
1085 return udp_push_pending_frames(sk);
1086
1087 /* ip6_finish_skb will release the cork, so make a copy of
1088 * fl6 here.
1089 */
1090 fl6 = inet_sk(sk)->cork.fl.u.ip6;
1091
1092 skb = ip6_finish_skb(sk);
1093 if (!skb)
1094 goto out;
1095
1096 err = udp_v6_send_skb(skb, &fl6);
1097
1098out:
1099 up->len = 0;
1100 up->pending = 0;
1101 return err;
1102}
1103
1104int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1105{
1106 struct ipv6_txoptions opt_space;
1107 struct udp_sock *up = udp_sk(sk);
1108 struct inet_sock *inet = inet_sk(sk);
1109 struct ipv6_pinfo *np = inet6_sk(sk);
1110 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1111 struct in6_addr *daddr, *final_p, final;
1112 struct ipv6_txoptions *opt = NULL;
1113 struct ipv6_txoptions *opt_to_free = NULL;
1114 struct ip6_flowlabel *flowlabel = NULL;
1115 struct flowi6 fl6;
1116 struct dst_entry *dst;
1117 struct ipcm6_cookie ipc6;
1118 int addr_len = msg->msg_namelen;
1119 bool connected = false;
1120 int ulen = len;
1121 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1122 int err;
1123 int is_udplite = IS_UDPLITE(sk);
1124 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1125 struct sockcm_cookie sockc;
1126
1127 ipc6.hlimit = -1;
1128 ipc6.tclass = -1;
1129 ipc6.dontfrag = -1;
1130 sockc.tsflags = sk->sk_tsflags;
1131
1132 /* destination address check */
1133 if (sin6) {
1134 if (addr_len < offsetof(struct sockaddr, sa_data))
1135 return -EINVAL;
1136
1137 switch (sin6->sin6_family) {
1138 case AF_INET6:
1139 if (addr_len < SIN6_LEN_RFC2133)
1140 return -EINVAL;
1141 daddr = &sin6->sin6_addr;
1142 if (ipv6_addr_any(daddr) &&
1143 ipv6_addr_v4mapped(&np->saddr))
1144 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1145 daddr);
1146 break;
1147 case AF_INET:
1148 goto do_udp_sendmsg;
1149 case AF_UNSPEC:
1150 msg->msg_name = sin6 = NULL;
1151 msg->msg_namelen = addr_len = 0;
1152 daddr = NULL;
1153 break;
1154 default:
1155 return -EINVAL;
1156 }
1157 } else if (!up->pending) {
1158 if (sk->sk_state != TCP_ESTABLISHED)
1159 return -EDESTADDRREQ;
1160 daddr = &sk->sk_v6_daddr;
1161 } else
1162 daddr = NULL;
1163
1164 if (daddr) {
1165 if (ipv6_addr_v4mapped(daddr)) {
1166 struct sockaddr_in sin;
1167 sin.sin_family = AF_INET;
1168 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1169 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1170 msg->msg_name = &sin;
1171 msg->msg_namelen = sizeof(sin);
1172do_udp_sendmsg:
1173 if (__ipv6_only_sock(sk))
1174 return -ENETUNREACH;
1175 return udp_sendmsg(sk, msg, len);
1176 }
1177 }
1178
1179 if (up->pending == AF_INET)
1180 return udp_sendmsg(sk, msg, len);
1181
1182 /* Rough check on arithmetic overflow,
1183 better check is made in ip6_append_data().
1184 */
1185 if (len > INT_MAX - sizeof(struct udphdr))
1186 return -EMSGSIZE;
1187
1188 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1189 if (up->pending) {
1190 /*
1191 * There are pending frames.
1192 * The socket lock must be held while it's corked.
1193 */
1194 lock_sock(sk);
1195 if (likely(up->pending)) {
1196 if (unlikely(up->pending != AF_INET6)) {
1197 release_sock(sk);
1198 return -EAFNOSUPPORT;
1199 }
1200 dst = NULL;
1201 goto do_append_data;
1202 }
1203 release_sock(sk);
1204 }
1205 ulen += sizeof(struct udphdr);
1206
1207 memset(&fl6, 0, sizeof(fl6));
1208
1209 if (sin6) {
1210 if (sin6->sin6_port == 0)
1211 return -EINVAL;
1212
1213 fl6.fl6_dport = sin6->sin6_port;
1214 daddr = &sin6->sin6_addr;
1215
1216 if (np->sndflow) {
1217 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1218 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1219 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1220 if (!flowlabel)
1221 return -EINVAL;
1222 }
1223 }
1224
1225 /*
1226 * Otherwise it will be difficult to maintain
1227 * sk->sk_dst_cache.
1228 */
1229 if (sk->sk_state == TCP_ESTABLISHED &&
1230 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1231 daddr = &sk->sk_v6_daddr;
1232
1233 if (addr_len >= sizeof(struct sockaddr_in6) &&
1234 sin6->sin6_scope_id &&
1235 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1236 fl6.flowi6_oif = sin6->sin6_scope_id;
1237 } else {
1238 if (sk->sk_state != TCP_ESTABLISHED)
1239 return -EDESTADDRREQ;
1240
1241 fl6.fl6_dport = inet->inet_dport;
1242 daddr = &sk->sk_v6_daddr;
1243 fl6.flowlabel = np->flow_label;
1244 connected = true;
1245 }
1246
1247 if (!fl6.flowi6_oif)
1248 fl6.flowi6_oif = sk->sk_bound_dev_if;
1249
1250 if (!fl6.flowi6_oif)
1251 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1252
1253 fl6.flowi6_mark = sk->sk_mark;
1254 fl6.flowi6_uid = sk->sk_uid;
1255
1256 if (msg->msg_controllen) {
1257 opt = &opt_space;
1258 memset(opt, 0, sizeof(struct ipv6_txoptions));
1259 opt->tot_len = sizeof(*opt);
1260 ipc6.opt = opt;
1261
1262 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, &ipc6, &sockc);
1263 if (err < 0) {
1264 fl6_sock_release(flowlabel);
1265 return err;
1266 }
1267 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1268 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1269 if (!flowlabel)
1270 return -EINVAL;
1271 }
1272 if (!(opt->opt_nflen|opt->opt_flen))
1273 opt = NULL;
1274 connected = false;
1275 }
1276 if (!opt) {
1277 opt = txopt_get(np);
1278 opt_to_free = opt;
1279 }
1280 if (flowlabel)
1281 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1282 opt = ipv6_fixup_options(&opt_space, opt);
1283 ipc6.opt = opt;
1284
1285 fl6.flowi6_proto = sk->sk_protocol;
1286 if (!ipv6_addr_any(daddr))
1287 fl6.daddr = *daddr;
1288 else
1289 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1290 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1291 fl6.saddr = np->saddr;
1292 fl6.fl6_sport = inet->inet_sport;
1293
1294 final_p = fl6_update_dst(&fl6, opt, &final);
1295 if (final_p)
1296 connected = false;
1297
1298 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1299 fl6.flowi6_oif = np->mcast_oif;
1300 connected = false;
1301 } else if (!fl6.flowi6_oif)
1302 fl6.flowi6_oif = np->ucast_oif;
1303
1304 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1305
1306 if (ipc6.tclass < 0)
1307 ipc6.tclass = np->tclass;
1308
1309 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel);
1310
1311 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, connected);
1312 if (IS_ERR(dst)) {
1313 err = PTR_ERR(dst);
1314 dst = NULL;
1315 goto out;
1316 }
1317
1318 if (ipc6.hlimit < 0)
1319 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1320
1321 if (msg->msg_flags&MSG_CONFIRM)
1322 goto do_confirm;
1323back_from_confirm:
1324
1325 /* Lockless fast path for the non-corking case */
1326 if (!corkreq) {
1327 struct sk_buff *skb;
1328
1329 skb = ip6_make_skb(sk, getfrag, msg, ulen,
1330 sizeof(struct udphdr), &ipc6,
1331 &fl6, (struct rt6_info *)dst,
1332 msg->msg_flags, &sockc);
1333 err = PTR_ERR(skb);
1334 if (!IS_ERR_OR_NULL(skb))
1335 err = udp_v6_send_skb(skb, &fl6);
1336 goto out;
1337 }
1338
1339 lock_sock(sk);
1340 if (unlikely(up->pending)) {
1341 /* The socket is already corked while preparing it. */
1342 /* ... which is an evident application bug. --ANK */
1343 release_sock(sk);
1344
1345 net_dbg_ratelimited("udp cork app bug 2\n");
1346 err = -EINVAL;
1347 goto out;
1348 }
1349
1350 up->pending = AF_INET6;
1351
1352do_append_data:
1353 if (ipc6.dontfrag < 0)
1354 ipc6.dontfrag = np->dontfrag;
1355 up->len += ulen;
1356 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr),
1357 &ipc6, &fl6, (struct rt6_info *)dst,
1358 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, &sockc);
1359 if (err)
1360 udp_v6_flush_pending_frames(sk);
1361 else if (!corkreq)
1362 err = udp_v6_push_pending_frames(sk);
1363 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1364 up->pending = 0;
1365
1366 if (err > 0)
1367 err = np->recverr ? net_xmit_errno(err) : 0;
1368 release_sock(sk);
1369
1370out:
1371 dst_release(dst);
1372 fl6_sock_release(flowlabel);
1373 txopt_put(opt_to_free);
1374 if (!err)
1375 return len;
1376 /*
1377 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1378 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1379 * we don't have a good statistic (IpOutDiscards but it can be too many
1380 * things). We could add another new stat but at least for now that
1381 * seems like overkill.
1382 */
1383 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1384 UDP6_INC_STATS(sock_net(sk),
1385 UDP_MIB_SNDBUFERRORS, is_udplite);
1386 }
1387 return err;
1388
1389do_confirm:
1390 if (msg->msg_flags & MSG_PROBE)
1391 dst_confirm_neigh(dst, &fl6.daddr);
1392 if (!(msg->msg_flags&MSG_PROBE) || len)
1393 goto back_from_confirm;
1394 err = 0;
1395 goto out;
1396}
1397
1398void udpv6_destroy_sock(struct sock *sk)
1399{
1400 struct udp_sock *up = udp_sk(sk);
1401 lock_sock(sk);
1402 udp_v6_flush_pending_frames(sk);
1403 release_sock(sk);
1404
1405 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
1406 void (*encap_destroy)(struct sock *sk);
1407 encap_destroy = READ_ONCE(up->encap_destroy);
1408 if (encap_destroy)
1409 encap_destroy(sk);
1410 }
1411
1412 inet6_destroy_sock(sk);
1413}
1414
1415/*
1416 * Socket option code for UDP
1417 */
1418int udpv6_setsockopt(struct sock *sk, int level, int optname,
1419 char __user *optval, unsigned int optlen)
1420{
1421 if (level == SOL_UDP || level == SOL_UDPLITE)
1422 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1423 udp_v6_push_pending_frames);
1424 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1425}
1426
1427#ifdef CONFIG_COMPAT
1428int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1429 char __user *optval, unsigned int optlen)
1430{
1431 if (level == SOL_UDP || level == SOL_UDPLITE)
1432 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1433 udp_v6_push_pending_frames);
1434 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1435}
1436#endif
1437
1438int udpv6_getsockopt(struct sock *sk, int level, int optname,
1439 char __user *optval, int __user *optlen)
1440{
1441 if (level == SOL_UDP || level == SOL_UDPLITE)
1442 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1443 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1444}
1445
1446#ifdef CONFIG_COMPAT
1447int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1448 char __user *optval, int __user *optlen)
1449{
1450 if (level == SOL_UDP || level == SOL_UDPLITE)
1451 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1452 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1453}
1454#endif
1455
1456/* thinking of making this const? Don't.
1457 * early_demux can change based on sysctl.
1458 */
1459static struct inet6_protocol udpv6_protocol = {
1460 .early_demux = udp_v6_early_demux,
1461 .early_demux_handler = udp_v6_early_demux,
1462 .handler = udpv6_rcv,
1463 .err_handler = udpv6_err,
1464 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1465};
1466
1467/* ------------------------------------------------------------------------ */
1468#ifdef CONFIG_PROC_FS
1469int udp6_seq_show(struct seq_file *seq, void *v)
1470{
1471 if (v == SEQ_START_TOKEN) {
1472 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1473 } else {
1474 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1475 struct inet_sock *inet = inet_sk(v);
1476 __u16 srcp = ntohs(inet->inet_sport);
1477 __u16 destp = ntohs(inet->inet_dport);
1478 ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket);
1479 }
1480 return 0;
1481}
1482
1483static const struct file_operations udp6_afinfo_seq_fops = {
1484 .open = udp_seq_open,
1485 .read = seq_read,
1486 .llseek = seq_lseek,
1487 .release = seq_release_net
1488};
1489
1490static struct udp_seq_afinfo udp6_seq_afinfo = {
1491 .name = "udp6",
1492 .family = AF_INET6,
1493 .udp_table = &udp_table,
1494 .seq_fops = &udp6_afinfo_seq_fops,
1495 .seq_ops = {
1496 .show = udp6_seq_show,
1497 },
1498};
1499
1500int __net_init udp6_proc_init(struct net *net)
1501{
1502 return udp_proc_register(net, &udp6_seq_afinfo);
1503}
1504
1505void udp6_proc_exit(struct net *net)
1506{
1507 udp_proc_unregister(net, &udp6_seq_afinfo);
1508}
1509#endif /* CONFIG_PROC_FS */
1510
1511/* ------------------------------------------------------------------------ */
1512
1513struct proto udpv6_prot = {
1514 .name = "UDPv6",
1515 .owner = THIS_MODULE,
1516 .close = udp_lib_close,
1517 .pre_connect = udpv6_pre_connect,
1518 .connect = ip6_datagram_connect,
1519 .disconnect = udp_disconnect,
1520 .ioctl = udp_ioctl,
1521 .init = udp_init_sock,
1522 .destroy = udpv6_destroy_sock,
1523 .setsockopt = udpv6_setsockopt,
1524 .getsockopt = udpv6_getsockopt,
1525 .sendmsg = udpv6_sendmsg,
1526 .recvmsg = udpv6_recvmsg,
1527 .release_cb = ip6_datagram_release_cb,
1528 .hash = udp_lib_hash,
1529 .unhash = udp_lib_unhash,
1530 .rehash = udp_v6_rehash,
1531 .get_port = udp_v6_get_port,
1532 .memory_allocated = &udp_memory_allocated,
1533 .sysctl_mem = sysctl_udp_mem,
1534 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
1535 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
1536 .obj_size = sizeof(struct udp6_sock),
1537 .h.udp_table = &udp_table,
1538#ifdef CONFIG_COMPAT
1539 .compat_setsockopt = compat_udpv6_setsockopt,
1540 .compat_getsockopt = compat_udpv6_getsockopt,
1541#endif
1542 .diag_destroy = udp_abort,
1543};
1544
1545static struct inet_protosw udpv6_protosw = {
1546 .type = SOCK_DGRAM,
1547 .protocol = IPPROTO_UDP,
1548 .prot = &udpv6_prot,
1549 .ops = &inet6_dgram_ops,
1550 .flags = INET_PROTOSW_PERMANENT,
1551};
1552
1553int __init udpv6_init(void)
1554{
1555 int ret;
1556
1557 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1558 if (ret)
1559 goto out;
1560
1561 ret = inet6_register_protosw(&udpv6_protosw);
1562 if (ret)
1563 goto out_udpv6_protocol;
1564out:
1565 return ret;
1566
1567out_udpv6_protocol:
1568 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1569 goto out;
1570}
1571
1572void udpv6_exit(void)
1573{
1574 inet6_unregister_protosw(&udpv6_protosw);
1575 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1576}