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 <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
49#include <linux/proc_fs.h>
50#include <linux/seq_file.h>
51#include "udp_impl.h"
52
53int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2)
54{
55 const struct in6_addr *sk_rcv_saddr6 = &inet6_sk(sk)->rcv_saddr;
56 const struct in6_addr *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2);
57 __be32 sk1_rcv_saddr = sk_rcv_saddr(sk);
58 __be32 sk2_rcv_saddr = sk_rcv_saddr(sk2);
59 int sk_ipv6only = ipv6_only_sock(sk);
60 int sk2_ipv6only = inet_v6_ipv6only(sk2);
61 int addr_type = ipv6_addr_type(sk_rcv_saddr6);
62 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
63
64 /* if both are mapped, treat as IPv4 */
65 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED)
66 return (!sk2_ipv6only &&
67 (!sk1_rcv_saddr || !sk2_rcv_saddr ||
68 sk1_rcv_saddr == sk2_rcv_saddr));
69
70 if (addr_type2 == IPV6_ADDR_ANY &&
71 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
72 return 1;
73
74 if (addr_type == IPV6_ADDR_ANY &&
75 !(sk_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
76 return 1;
77
78 if (sk2_rcv_saddr6 &&
79 ipv6_addr_equal(sk_rcv_saddr6, sk2_rcv_saddr6))
80 return 1;
81
82 return 0;
83}
84
85static unsigned int udp6_portaddr_hash(struct net *net,
86 const struct in6_addr *addr6,
87 unsigned int port)
88{
89 unsigned int hash, mix = net_hash_mix(net);
90
91 if (ipv6_addr_any(addr6))
92 hash = jhash_1word(0, mix);
93 else if (ipv6_addr_v4mapped(addr6))
94 hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix);
95 else
96 hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix);
97
98 return hash ^ port;
99}
100
101
102int udp_v6_get_port(struct sock *sk, unsigned short snum)
103{
104 unsigned int hash2_nulladdr =
105 udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
106 unsigned int hash2_partial =
107 udp6_portaddr_hash(sock_net(sk), &inet6_sk(sk)->rcv_saddr, 0);
108
109 /* precompute partial secondary hash */
110 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
111 return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal, hash2_nulladdr);
112}
113
114static void udp_v6_rehash(struct sock *sk)
115{
116 u16 new_hash = udp6_portaddr_hash(sock_net(sk),
117 &inet6_sk(sk)->rcv_saddr,
118 inet_sk(sk)->inet_num);
119
120 udp_lib_rehash(sk, new_hash);
121}
122
123static inline int compute_score(struct sock *sk, struct net *net,
124 unsigned short hnum,
125 const struct in6_addr *saddr, __be16 sport,
126 const struct in6_addr *daddr, __be16 dport,
127 int dif)
128{
129 int score = -1;
130
131 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
132 sk->sk_family == PF_INET6) {
133 struct ipv6_pinfo *np = inet6_sk(sk);
134 struct inet_sock *inet = inet_sk(sk);
135
136 score = 0;
137 if (inet->inet_dport) {
138 if (inet->inet_dport != sport)
139 return -1;
140 score++;
141 }
142 if (!ipv6_addr_any(&np->rcv_saddr)) {
143 if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
144 return -1;
145 score++;
146 }
147 if (!ipv6_addr_any(&np->daddr)) {
148 if (!ipv6_addr_equal(&np->daddr, saddr))
149 return -1;
150 score++;
151 }
152 if (sk->sk_bound_dev_if) {
153 if (sk->sk_bound_dev_if != dif)
154 return -1;
155 score++;
156 }
157 }
158 return score;
159}
160
161#define SCORE2_MAX (1 + 1 + 1)
162static inline int compute_score2(struct sock *sk, struct net *net,
163 const struct in6_addr *saddr, __be16 sport,
164 const struct in6_addr *daddr, unsigned short hnum,
165 int dif)
166{
167 int score = -1;
168
169 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
170 sk->sk_family == PF_INET6) {
171 struct ipv6_pinfo *np = inet6_sk(sk);
172 struct inet_sock *inet = inet_sk(sk);
173
174 if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
175 return -1;
176 score = 0;
177 if (inet->inet_dport) {
178 if (inet->inet_dport != sport)
179 return -1;
180 score++;
181 }
182 if (!ipv6_addr_any(&np->daddr)) {
183 if (!ipv6_addr_equal(&np->daddr, saddr))
184 return -1;
185 score++;
186 }
187 if (sk->sk_bound_dev_if) {
188 if (sk->sk_bound_dev_if != dif)
189 return -1;
190 score++;
191 }
192 }
193 return score;
194}
195
196
197/* called with read_rcu_lock() */
198static struct sock *udp6_lib_lookup2(struct net *net,
199 const struct in6_addr *saddr, __be16 sport,
200 const struct in6_addr *daddr, unsigned int hnum, int dif,
201 struct udp_hslot *hslot2, unsigned int slot2)
202{
203 struct sock *sk, *result;
204 struct hlist_nulls_node *node;
205 int score, badness;
206
207begin:
208 result = NULL;
209 badness = -1;
210 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
211 score = compute_score2(sk, net, saddr, sport,
212 daddr, hnum, dif);
213 if (score > badness) {
214 result = sk;
215 badness = score;
216 if (score == SCORE2_MAX)
217 goto exact_match;
218 }
219 }
220 /*
221 * if the nulls value we got at the end of this lookup is
222 * not the expected one, we must restart lookup.
223 * We probably met an item that was moved to another chain.
224 */
225 if (get_nulls_value(node) != slot2)
226 goto begin;
227
228 if (result) {
229exact_match:
230 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
231 result = NULL;
232 else if (unlikely(compute_score2(result, net, saddr, sport,
233 daddr, hnum, dif) < badness)) {
234 sock_put(result);
235 goto begin;
236 }
237 }
238 return result;
239}
240
241struct sock *__udp6_lib_lookup(struct net *net,
242 const struct in6_addr *saddr, __be16 sport,
243 const struct in6_addr *daddr, __be16 dport,
244 int dif, struct udp_table *udptable)
245{
246 struct sock *sk, *result;
247 struct hlist_nulls_node *node;
248 unsigned short hnum = ntohs(dport);
249 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
250 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
251 int score, badness;
252
253 rcu_read_lock();
254 if (hslot->count > 10) {
255 hash2 = udp6_portaddr_hash(net, daddr, hnum);
256 slot2 = hash2 & udptable->mask;
257 hslot2 = &udptable->hash2[slot2];
258 if (hslot->count < hslot2->count)
259 goto begin;
260
261 result = udp6_lib_lookup2(net, saddr, sport,
262 daddr, hnum, dif,
263 hslot2, slot2);
264 if (!result) {
265 hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
266 slot2 = hash2 & udptable->mask;
267 hslot2 = &udptable->hash2[slot2];
268 if (hslot->count < hslot2->count)
269 goto begin;
270
271 result = udp6_lib_lookup2(net, saddr, sport,
272 &in6addr_any, hnum, dif,
273 hslot2, slot2);
274 }
275 rcu_read_unlock();
276 return result;
277 }
278begin:
279 result = NULL;
280 badness = -1;
281 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
282 score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif);
283 if (score > badness) {
284 result = sk;
285 badness = score;
286 }
287 }
288 /*
289 * if the nulls value we got at the end of this lookup is
290 * not the expected one, we must restart lookup.
291 * We probably met an item that was moved to another chain.
292 */
293 if (get_nulls_value(node) != slot)
294 goto begin;
295
296 if (result) {
297 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
298 result = NULL;
299 else if (unlikely(compute_score(result, net, hnum, saddr, sport,
300 daddr, dport, dif) < badness)) {
301 sock_put(result);
302 goto begin;
303 }
304 }
305 rcu_read_unlock();
306 return result;
307}
308EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
309
310static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
311 __be16 sport, __be16 dport,
312 struct udp_table *udptable)
313{
314 struct sock *sk;
315 const struct ipv6hdr *iph = ipv6_hdr(skb);
316
317 if (unlikely(sk = skb_steal_sock(skb)))
318 return sk;
319 return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport,
320 &iph->daddr, dport, inet6_iif(skb),
321 udptable);
322}
323
324struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
325 const struct in6_addr *daddr, __be16 dport, int dif)
326{
327 return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
328}
329EXPORT_SYMBOL_GPL(udp6_lib_lookup);
330
331
332/*
333 * This should be easy, if there is something there we
334 * return it, otherwise we block.
335 */
336
337int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk,
338 struct msghdr *msg, size_t len,
339 int noblock, int flags, int *addr_len)
340{
341 struct ipv6_pinfo *np = inet6_sk(sk);
342 struct inet_sock *inet = inet_sk(sk);
343 struct sk_buff *skb;
344 unsigned int ulen, copied;
345 int peeked, off = 0;
346 int err;
347 int is_udplite = IS_UDPLITE(sk);
348 int is_udp4;
349 bool slow;
350
351 if (addr_len)
352 *addr_len = sizeof(struct sockaddr_in6);
353
354 if (flags & MSG_ERRQUEUE)
355 return ipv6_recv_error(sk, msg, len);
356
357 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
358 return ipv6_recv_rxpmtu(sk, msg, len);
359
360try_again:
361 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
362 &peeked, &off, &err);
363 if (!skb)
364 goto out;
365
366 ulen = skb->len - sizeof(struct udphdr);
367 copied = len;
368 if (copied > ulen)
369 copied = ulen;
370 else if (copied < ulen)
371 msg->msg_flags |= MSG_TRUNC;
372
373 is_udp4 = (skb->protocol == htons(ETH_P_IP));
374
375 /*
376 * If checksum is needed at all, try to do it while copying the
377 * data. If the data is truncated, or if we only want a partial
378 * coverage checksum (UDP-Lite), do it before the copy.
379 */
380
381 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
382 if (udp_lib_checksum_complete(skb))
383 goto csum_copy_err;
384 }
385
386 if (skb_csum_unnecessary(skb))
387 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
388 msg->msg_iov, copied );
389 else {
390 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
391 if (err == -EINVAL)
392 goto csum_copy_err;
393 }
394 if (err)
395 goto out_free;
396
397 if (!peeked) {
398 if (is_udp4)
399 UDP_INC_STATS_USER(sock_net(sk),
400 UDP_MIB_INDATAGRAMS, is_udplite);
401 else
402 UDP6_INC_STATS_USER(sock_net(sk),
403 UDP_MIB_INDATAGRAMS, is_udplite);
404 }
405
406 sock_recv_ts_and_drops(msg, sk, skb);
407
408 /* Copy the address. */
409 if (msg->msg_name) {
410 struct sockaddr_in6 *sin6;
411
412 sin6 = (struct sockaddr_in6 *) msg->msg_name;
413 sin6->sin6_family = AF_INET6;
414 sin6->sin6_port = udp_hdr(skb)->source;
415 sin6->sin6_flowinfo = 0;
416 sin6->sin6_scope_id = 0;
417
418 if (is_udp4)
419 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
420 &sin6->sin6_addr);
421 else {
422 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
423 if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
424 sin6->sin6_scope_id = IP6CB(skb)->iif;
425 }
426
427 }
428 if (is_udp4) {
429 if (inet->cmsg_flags)
430 ip_cmsg_recv(msg, skb);
431 } else {
432 if (np->rxopt.all)
433 datagram_recv_ctl(sk, msg, skb);
434 }
435
436 err = copied;
437 if (flags & MSG_TRUNC)
438 err = ulen;
439
440out_free:
441 skb_free_datagram_locked(sk, skb);
442out:
443 return err;
444
445csum_copy_err:
446 slow = lock_sock_fast(sk);
447 if (!skb_kill_datagram(sk, skb, flags)) {
448 if (is_udp4)
449 UDP_INC_STATS_USER(sock_net(sk),
450 UDP_MIB_INERRORS, is_udplite);
451 else
452 UDP6_INC_STATS_USER(sock_net(sk),
453 UDP_MIB_INERRORS, is_udplite);
454 }
455 unlock_sock_fast(sk, slow);
456
457 if (noblock)
458 return -EAGAIN;
459
460 /* starting over for a new packet */
461 msg->msg_flags &= ~MSG_TRUNC;
462 goto try_again;
463}
464
465void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
466 u8 type, u8 code, int offset, __be32 info,
467 struct udp_table *udptable)
468{
469 struct ipv6_pinfo *np;
470 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
471 const struct in6_addr *saddr = &hdr->saddr;
472 const struct in6_addr *daddr = &hdr->daddr;
473 struct udphdr *uh = (struct udphdr*)(skb->data+offset);
474 struct sock *sk;
475 int err;
476
477 sk = __udp6_lib_lookup(dev_net(skb->dev), daddr, uh->dest,
478 saddr, uh->source, inet6_iif(skb), udptable);
479 if (sk == NULL)
480 return;
481
482 np = inet6_sk(sk);
483
484 if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
485 goto out;
486
487 if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
488 goto out;
489
490 if (np->recverr)
491 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
492
493 sk->sk_err = err;
494 sk->sk_error_report(sk);
495out:
496 sock_put(sk);
497}
498
499static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
500{
501 int rc;
502
503 if (!ipv6_addr_any(&inet6_sk(sk)->daddr))
504 sock_rps_save_rxhash(sk, skb);
505
506 rc = sock_queue_rcv_skb(sk, skb);
507 if (rc < 0) {
508 int is_udplite = IS_UDPLITE(sk);
509
510 /* Note that an ENOMEM error is charged twice */
511 if (rc == -ENOMEM)
512 UDP6_INC_STATS_BH(sock_net(sk),
513 UDP_MIB_RCVBUFERRORS, is_udplite);
514 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
515 kfree_skb(skb);
516 return -1;
517 }
518 return 0;
519}
520
521static __inline__ void udpv6_err(struct sk_buff *skb,
522 struct inet6_skb_parm *opt, u8 type,
523 u8 code, int offset, __be32 info )
524{
525 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
526}
527
528static struct static_key udpv6_encap_needed __read_mostly;
529void udpv6_encap_enable(void)
530{
531 if (!static_key_enabled(&udpv6_encap_needed))
532 static_key_slow_inc(&udpv6_encap_needed);
533}
534EXPORT_SYMBOL(udpv6_encap_enable);
535
536int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
537{
538 struct udp_sock *up = udp_sk(sk);
539 int rc;
540 int is_udplite = IS_UDPLITE(sk);
541
542 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
543 goto drop;
544
545 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
546 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
547
548 /*
549 * This is an encapsulation socket so pass the skb to
550 * the socket's udp_encap_rcv() hook. Otherwise, just
551 * fall through and pass this up the UDP socket.
552 * up->encap_rcv() returns the following value:
553 * =0 if skb was successfully passed to the encap
554 * handler or was discarded by it.
555 * >0 if skb should be passed on to UDP.
556 * <0 if skb should be resubmitted as proto -N
557 */
558
559 /* if we're overly short, let UDP handle it */
560 encap_rcv = ACCESS_ONCE(up->encap_rcv);
561 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
562 int ret;
563
564 ret = encap_rcv(sk, skb);
565 if (ret <= 0) {
566 UDP_INC_STATS_BH(sock_net(sk),
567 UDP_MIB_INDATAGRAMS,
568 is_udplite);
569 return -ret;
570 }
571 }
572
573 /* FALLTHROUGH -- it's a UDP Packet */
574 }
575
576 /*
577 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
578 */
579 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
580
581 if (up->pcrlen == 0) { /* full coverage was set */
582 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage"
583 " %d while full coverage %d requested\n",
584 UDP_SKB_CB(skb)->cscov, skb->len);
585 goto drop;
586 }
587 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
588 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d "
589 "too small, need min %d\n",
590 UDP_SKB_CB(skb)->cscov, up->pcrlen);
591 goto drop;
592 }
593 }
594
595 if (rcu_access_pointer(sk->sk_filter)) {
596 if (udp_lib_checksum_complete(skb))
597 goto drop;
598 }
599
600 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf))
601 goto drop;
602
603 skb_dst_drop(skb);
604
605 bh_lock_sock(sk);
606 rc = 0;
607 if (!sock_owned_by_user(sk))
608 rc = __udpv6_queue_rcv_skb(sk, skb);
609 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
610 bh_unlock_sock(sk);
611 goto drop;
612 }
613 bh_unlock_sock(sk);
614
615 return rc;
616drop:
617 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
618 atomic_inc(&sk->sk_drops);
619 kfree_skb(skb);
620 return -1;
621}
622
623static struct sock *udp_v6_mcast_next(struct net *net, struct sock *sk,
624 __be16 loc_port, const struct in6_addr *loc_addr,
625 __be16 rmt_port, const struct in6_addr *rmt_addr,
626 int dif)
627{
628 struct hlist_nulls_node *node;
629 struct sock *s = sk;
630 unsigned short num = ntohs(loc_port);
631
632 sk_nulls_for_each_from(s, node) {
633 struct inet_sock *inet = inet_sk(s);
634
635 if (!net_eq(sock_net(s), net))
636 continue;
637
638 if (udp_sk(s)->udp_port_hash == num &&
639 s->sk_family == PF_INET6) {
640 struct ipv6_pinfo *np = inet6_sk(s);
641 if (inet->inet_dport) {
642 if (inet->inet_dport != rmt_port)
643 continue;
644 }
645 if (!ipv6_addr_any(&np->daddr) &&
646 !ipv6_addr_equal(&np->daddr, rmt_addr))
647 continue;
648
649 if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)
650 continue;
651
652 if (!ipv6_addr_any(&np->rcv_saddr)) {
653 if (!ipv6_addr_equal(&np->rcv_saddr, loc_addr))
654 continue;
655 }
656 if (!inet6_mc_check(s, loc_addr, rmt_addr))
657 continue;
658 return s;
659 }
660 }
661 return NULL;
662}
663
664static void flush_stack(struct sock **stack, unsigned int count,
665 struct sk_buff *skb, unsigned int final)
666{
667 struct sk_buff *skb1 = NULL;
668 struct sock *sk;
669 unsigned int i;
670
671 for (i = 0; i < count; i++) {
672 sk = stack[i];
673 if (likely(skb1 == NULL))
674 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
675 if (!skb1) {
676 atomic_inc(&sk->sk_drops);
677 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
678 IS_UDPLITE(sk));
679 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
680 IS_UDPLITE(sk));
681 }
682
683 if (skb1 && udpv6_queue_rcv_skb(sk, skb1) <= 0)
684 skb1 = NULL;
685 }
686 if (unlikely(skb1))
687 kfree_skb(skb1);
688}
689/*
690 * Note: called only from the BH handler context,
691 * so we don't need to lock the hashes.
692 */
693static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
694 const struct in6_addr *saddr, const struct in6_addr *daddr,
695 struct udp_table *udptable)
696{
697 struct sock *sk, *stack[256 / sizeof(struct sock *)];
698 const struct udphdr *uh = udp_hdr(skb);
699 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
700 int dif;
701 unsigned int i, count = 0;
702
703 spin_lock(&hslot->lock);
704 sk = sk_nulls_head(&hslot->head);
705 dif = inet6_iif(skb);
706 sk = udp_v6_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
707 while (sk) {
708 stack[count++] = sk;
709 sk = udp_v6_mcast_next(net, sk_nulls_next(sk), uh->dest, daddr,
710 uh->source, saddr, dif);
711 if (unlikely(count == ARRAY_SIZE(stack))) {
712 if (!sk)
713 break;
714 flush_stack(stack, count, skb, ~0);
715 count = 0;
716 }
717 }
718 /*
719 * before releasing the lock, we must take reference on sockets
720 */
721 for (i = 0; i < count; i++)
722 sock_hold(stack[i]);
723
724 spin_unlock(&hslot->lock);
725
726 if (count) {
727 flush_stack(stack, count, skb, count - 1);
728
729 for (i = 0; i < count; i++)
730 sock_put(stack[i]);
731 } else {
732 kfree_skb(skb);
733 }
734 return 0;
735}
736
737static inline int udp6_csum_init(struct sk_buff *skb, struct udphdr *uh,
738 int proto)
739{
740 int err;
741
742 UDP_SKB_CB(skb)->partial_cov = 0;
743 UDP_SKB_CB(skb)->cscov = skb->len;
744
745 if (proto == IPPROTO_UDPLITE) {
746 err = udplite_checksum_init(skb, uh);
747 if (err)
748 return err;
749 }
750
751 if (uh->check == 0) {
752 /* RFC 2460 section 8.1 says that we SHOULD log
753 this error. Well, it is reasonable.
754 */
755 LIMIT_NETDEBUG(KERN_INFO "IPv6: udp checksum is 0\n");
756 return 1;
757 }
758 if (skb->ip_summed == CHECKSUM_COMPLETE &&
759 !csum_ipv6_magic(&ipv6_hdr(skb)->saddr, &ipv6_hdr(skb)->daddr,
760 skb->len, proto, skb->csum))
761 skb->ip_summed = CHECKSUM_UNNECESSARY;
762
763 if (!skb_csum_unnecessary(skb))
764 skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
765 &ipv6_hdr(skb)->daddr,
766 skb->len, proto, 0));
767
768 return 0;
769}
770
771int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
772 int proto)
773{
774 struct net *net = dev_net(skb->dev);
775 struct sock *sk;
776 struct udphdr *uh;
777 const struct in6_addr *saddr, *daddr;
778 u32 ulen = 0;
779
780 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
781 goto discard;
782
783 saddr = &ipv6_hdr(skb)->saddr;
784 daddr = &ipv6_hdr(skb)->daddr;
785 uh = udp_hdr(skb);
786
787 ulen = ntohs(uh->len);
788 if (ulen > skb->len)
789 goto short_packet;
790
791 if (proto == IPPROTO_UDP) {
792 /* UDP validates ulen. */
793
794 /* Check for jumbo payload */
795 if (ulen == 0)
796 ulen = skb->len;
797
798 if (ulen < sizeof(*uh))
799 goto short_packet;
800
801 if (ulen < skb->len) {
802 if (pskb_trim_rcsum(skb, ulen))
803 goto short_packet;
804 saddr = &ipv6_hdr(skb)->saddr;
805 daddr = &ipv6_hdr(skb)->daddr;
806 uh = udp_hdr(skb);
807 }
808 }
809
810 if (udp6_csum_init(skb, uh, proto))
811 goto discard;
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);
819
820 /* Unicast */
821
822 /*
823 * check socket cache ... must talk to Alan about his plans
824 * for sock caches... i'll skip this for now.
825 */
826 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
827 if (sk != NULL) {
828 int ret = udpv6_queue_rcv_skb(sk, skb);
829 sock_put(sk);
830
831 /* a return value > 0 means to resubmit the input, but
832 * it wants the return to be -protocol, or 0
833 */
834 if (ret > 0)
835 return -ret;
836
837 return 0;
838 }
839
840 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
841 goto discard;
842
843 if (udp_lib_checksum_complete(skb))
844 goto discard;
845
846 UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
847 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
848
849 kfree_skb(skb);
850 return 0;
851
852short_packet:
853 LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
854 proto == IPPROTO_UDPLITE ? "-Lite" : "",
855 saddr,
856 ntohs(uh->source),
857 ulen,
858 skb->len,
859 daddr,
860 ntohs(uh->dest));
861
862discard:
863 UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
864 kfree_skb(skb);
865 return 0;
866}
867
868static __inline__ int udpv6_rcv(struct sk_buff *skb)
869{
870 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
871}
872
873/*
874 * Throw away all pending data and cancel the corking. Socket is locked.
875 */
876static void udp_v6_flush_pending_frames(struct sock *sk)
877{
878 struct udp_sock *up = udp_sk(sk);
879
880 if (up->pending == AF_INET)
881 udp_flush_pending_frames(sk);
882 else if (up->pending) {
883 up->len = 0;
884 up->pending = 0;
885 ip6_flush_pending_frames(sk);
886 }
887}
888
889/**
890 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
891 * @sk: socket we are sending on
892 * @skb: sk_buff containing the filled-in UDP header
893 * (checksum field must be zeroed out)
894 */
895static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
896 const struct in6_addr *saddr,
897 const struct in6_addr *daddr, int len)
898{
899 unsigned int offset;
900 struct udphdr *uh = udp_hdr(skb);
901 __wsum csum = 0;
902
903 if (skb_queue_len(&sk->sk_write_queue) == 1) {
904 /* Only one fragment on the socket. */
905 skb->csum_start = skb_transport_header(skb) - skb->head;
906 skb->csum_offset = offsetof(struct udphdr, check);
907 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
908 } else {
909 /*
910 * HW-checksum won't work as there are two or more
911 * fragments on the socket so that all csums of sk_buffs
912 * should be together
913 */
914 offset = skb_transport_offset(skb);
915 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
916
917 skb->ip_summed = CHECKSUM_NONE;
918
919 skb_queue_walk(&sk->sk_write_queue, skb) {
920 csum = csum_add(csum, skb->csum);
921 }
922
923 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
924 csum);
925 if (uh->check == 0)
926 uh->check = CSUM_MANGLED_0;
927 }
928}
929
930/*
931 * Sending
932 */
933
934static int udp_v6_push_pending_frames(struct sock *sk)
935{
936 struct sk_buff *skb;
937 struct udphdr *uh;
938 struct udp_sock *up = udp_sk(sk);
939 struct inet_sock *inet = inet_sk(sk);
940 struct flowi6 *fl6 = &inet->cork.fl.u.ip6;
941 int err = 0;
942 int is_udplite = IS_UDPLITE(sk);
943 __wsum csum = 0;
944
945 /* Grab the skbuff where UDP header space exists. */
946 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
947 goto out;
948
949 /*
950 * Create a UDP header
951 */
952 uh = udp_hdr(skb);
953 uh->source = fl6->fl6_sport;
954 uh->dest = fl6->fl6_dport;
955 uh->len = htons(up->len);
956 uh->check = 0;
957
958 if (is_udplite)
959 csum = udplite_csum_outgoing(sk, skb);
960 else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
961 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr,
962 up->len);
963 goto send;
964 } else
965 csum = udp_csum_outgoing(sk, skb);
966
967 /* add protocol-dependent pseudo-header */
968 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
969 up->len, fl6->flowi6_proto, csum);
970 if (uh->check == 0)
971 uh->check = CSUM_MANGLED_0;
972
973send:
974 err = ip6_push_pending_frames(sk);
975 if (err) {
976 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
977 UDP6_INC_STATS_USER(sock_net(sk),
978 UDP_MIB_SNDBUFERRORS, is_udplite);
979 err = 0;
980 }
981 } else
982 UDP6_INC_STATS_USER(sock_net(sk),
983 UDP_MIB_OUTDATAGRAMS, is_udplite);
984out:
985 up->len = 0;
986 up->pending = 0;
987 return err;
988}
989
990int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
991 struct msghdr *msg, size_t len)
992{
993 struct ipv6_txoptions opt_space;
994 struct udp_sock *up = udp_sk(sk);
995 struct inet_sock *inet = inet_sk(sk);
996 struct ipv6_pinfo *np = inet6_sk(sk);
997 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) msg->msg_name;
998 struct in6_addr *daddr, *final_p, final;
999 struct ipv6_txoptions *opt = NULL;
1000 struct ip6_flowlabel *flowlabel = NULL;
1001 struct flowi6 fl6;
1002 struct dst_entry *dst;
1003 int addr_len = msg->msg_namelen;
1004 int ulen = len;
1005 int hlimit = -1;
1006 int tclass = -1;
1007 int dontfrag = -1;
1008 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1009 int err;
1010 int connected = 0;
1011 int is_udplite = IS_UDPLITE(sk);
1012 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1013
1014 /* destination address check */
1015 if (sin6) {
1016 if (addr_len < offsetof(struct sockaddr, sa_data))
1017 return -EINVAL;
1018
1019 switch (sin6->sin6_family) {
1020 case AF_INET6:
1021 if (addr_len < SIN6_LEN_RFC2133)
1022 return -EINVAL;
1023 daddr = &sin6->sin6_addr;
1024 break;
1025 case AF_INET:
1026 goto do_udp_sendmsg;
1027 case AF_UNSPEC:
1028 msg->msg_name = sin6 = NULL;
1029 msg->msg_namelen = addr_len = 0;
1030 daddr = NULL;
1031 break;
1032 default:
1033 return -EINVAL;
1034 }
1035 } else if (!up->pending) {
1036 if (sk->sk_state != TCP_ESTABLISHED)
1037 return -EDESTADDRREQ;
1038 daddr = &np->daddr;
1039 } else
1040 daddr = NULL;
1041
1042 if (daddr) {
1043 if (ipv6_addr_v4mapped(daddr)) {
1044 struct sockaddr_in sin;
1045 sin.sin_family = AF_INET;
1046 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1047 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1048 msg->msg_name = &sin;
1049 msg->msg_namelen = sizeof(sin);
1050do_udp_sendmsg:
1051 if (__ipv6_only_sock(sk))
1052 return -ENETUNREACH;
1053 return udp_sendmsg(iocb, sk, msg, len);
1054 }
1055 }
1056
1057 if (up->pending == AF_INET)
1058 return udp_sendmsg(iocb, sk, msg, len);
1059
1060 /* Rough check on arithmetic overflow,
1061 better check is made in ip6_append_data().
1062 */
1063 if (len > INT_MAX - sizeof(struct udphdr))
1064 return -EMSGSIZE;
1065
1066 if (up->pending) {
1067 /*
1068 * There are pending frames.
1069 * The socket lock must be held while it's corked.
1070 */
1071 lock_sock(sk);
1072 if (likely(up->pending)) {
1073 if (unlikely(up->pending != AF_INET6)) {
1074 release_sock(sk);
1075 return -EAFNOSUPPORT;
1076 }
1077 dst = NULL;
1078 goto do_append_data;
1079 }
1080 release_sock(sk);
1081 }
1082 ulen += sizeof(struct udphdr);
1083
1084 memset(&fl6, 0, sizeof(fl6));
1085
1086 if (sin6) {
1087 if (sin6->sin6_port == 0)
1088 return -EINVAL;
1089
1090 fl6.fl6_dport = sin6->sin6_port;
1091 daddr = &sin6->sin6_addr;
1092
1093 if (np->sndflow) {
1094 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1095 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1096 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1097 if (flowlabel == NULL)
1098 return -EINVAL;
1099 daddr = &flowlabel->dst;
1100 }
1101 }
1102
1103 /*
1104 * Otherwise it will be difficult to maintain
1105 * sk->sk_dst_cache.
1106 */
1107 if (sk->sk_state == TCP_ESTABLISHED &&
1108 ipv6_addr_equal(daddr, &np->daddr))
1109 daddr = &np->daddr;
1110
1111 if (addr_len >= sizeof(struct sockaddr_in6) &&
1112 sin6->sin6_scope_id &&
1113 ipv6_addr_type(daddr)&IPV6_ADDR_LINKLOCAL)
1114 fl6.flowi6_oif = sin6->sin6_scope_id;
1115 } else {
1116 if (sk->sk_state != TCP_ESTABLISHED)
1117 return -EDESTADDRREQ;
1118
1119 fl6.fl6_dport = inet->inet_dport;
1120 daddr = &np->daddr;
1121 fl6.flowlabel = np->flow_label;
1122 connected = 1;
1123 }
1124
1125 if (!fl6.flowi6_oif)
1126 fl6.flowi6_oif = sk->sk_bound_dev_if;
1127
1128 if (!fl6.flowi6_oif)
1129 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1130
1131 fl6.flowi6_mark = sk->sk_mark;
1132
1133 if (msg->msg_controllen) {
1134 opt = &opt_space;
1135 memset(opt, 0, sizeof(struct ipv6_txoptions));
1136 opt->tot_len = sizeof(*opt);
1137
1138 err = datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt,
1139 &hlimit, &tclass, &dontfrag);
1140 if (err < 0) {
1141 fl6_sock_release(flowlabel);
1142 return err;
1143 }
1144 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1145 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1146 if (flowlabel == NULL)
1147 return -EINVAL;
1148 }
1149 if (!(opt->opt_nflen|opt->opt_flen))
1150 opt = NULL;
1151 connected = 0;
1152 }
1153 if (opt == NULL)
1154 opt = np->opt;
1155 if (flowlabel)
1156 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1157 opt = ipv6_fixup_options(&opt_space, opt);
1158
1159 fl6.flowi6_proto = sk->sk_protocol;
1160 if (!ipv6_addr_any(daddr))
1161 fl6.daddr = *daddr;
1162 else
1163 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1164 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1165 fl6.saddr = np->saddr;
1166 fl6.fl6_sport = inet->inet_sport;
1167
1168 final_p = fl6_update_dst(&fl6, opt, &final);
1169 if (final_p)
1170 connected = 0;
1171
1172 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1173 fl6.flowi6_oif = np->mcast_oif;
1174 connected = 0;
1175 } else if (!fl6.flowi6_oif)
1176 fl6.flowi6_oif = np->ucast_oif;
1177
1178 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1179
1180 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, true);
1181 if (IS_ERR(dst)) {
1182 err = PTR_ERR(dst);
1183 dst = NULL;
1184 goto out;
1185 }
1186
1187 if (hlimit < 0) {
1188 if (ipv6_addr_is_multicast(&fl6.daddr))
1189 hlimit = np->mcast_hops;
1190 else
1191 hlimit = np->hop_limit;
1192 if (hlimit < 0)
1193 hlimit = ip6_dst_hoplimit(dst);
1194 }
1195
1196 if (tclass < 0)
1197 tclass = np->tclass;
1198
1199 if (dontfrag < 0)
1200 dontfrag = np->dontfrag;
1201
1202 if (msg->msg_flags&MSG_CONFIRM)
1203 goto do_confirm;
1204back_from_confirm:
1205
1206 lock_sock(sk);
1207 if (unlikely(up->pending)) {
1208 /* The socket is already corked while preparing it. */
1209 /* ... which is an evident application bug. --ANK */
1210 release_sock(sk);
1211
1212 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
1213 err = -EINVAL;
1214 goto out;
1215 }
1216
1217 up->pending = AF_INET6;
1218
1219do_append_data:
1220 up->len += ulen;
1221 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1222 err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen,
1223 sizeof(struct udphdr), hlimit, tclass, opt, &fl6,
1224 (struct rt6_info*)dst,
1225 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag);
1226 if (err)
1227 udp_v6_flush_pending_frames(sk);
1228 else if (!corkreq)
1229 err = udp_v6_push_pending_frames(sk);
1230 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1231 up->pending = 0;
1232
1233 if (dst) {
1234 if (connected) {
1235 ip6_dst_store(sk, dst,
1236 ipv6_addr_equal(&fl6.daddr, &np->daddr) ?
1237 &np->daddr : NULL,
1238#ifdef CONFIG_IPV6_SUBTREES
1239 ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1240 &np->saddr :
1241#endif
1242 NULL);
1243 } else {
1244 dst_release(dst);
1245 }
1246 dst = NULL;
1247 }
1248
1249 if (err > 0)
1250 err = np->recverr ? net_xmit_errno(err) : 0;
1251 release_sock(sk);
1252out:
1253 dst_release(dst);
1254 fl6_sock_release(flowlabel);
1255 if (!err)
1256 return len;
1257 /*
1258 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1259 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1260 * we don't have a good statistic (IpOutDiscards but it can be too many
1261 * things). We could add another new stat but at least for now that
1262 * seems like overkill.
1263 */
1264 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1265 UDP6_INC_STATS_USER(sock_net(sk),
1266 UDP_MIB_SNDBUFERRORS, is_udplite);
1267 }
1268 return err;
1269
1270do_confirm:
1271 dst_confirm(dst);
1272 if (!(msg->msg_flags&MSG_PROBE) || len)
1273 goto back_from_confirm;
1274 err = 0;
1275 goto out;
1276}
1277
1278void udpv6_destroy_sock(struct sock *sk)
1279{
1280 lock_sock(sk);
1281 udp_v6_flush_pending_frames(sk);
1282 release_sock(sk);
1283
1284 inet6_destroy_sock(sk);
1285}
1286
1287/*
1288 * Socket option code for UDP
1289 */
1290int udpv6_setsockopt(struct sock *sk, int level, int optname,
1291 char __user *optval, unsigned int optlen)
1292{
1293 if (level == SOL_UDP || level == SOL_UDPLITE)
1294 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1295 udp_v6_push_pending_frames);
1296 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1297}
1298
1299#ifdef CONFIG_COMPAT
1300int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1301 char __user *optval, unsigned int optlen)
1302{
1303 if (level == SOL_UDP || level == SOL_UDPLITE)
1304 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1305 udp_v6_push_pending_frames);
1306 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1307}
1308#endif
1309
1310int udpv6_getsockopt(struct sock *sk, int level, int optname,
1311 char __user *optval, int __user *optlen)
1312{
1313 if (level == SOL_UDP || level == SOL_UDPLITE)
1314 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1315 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1316}
1317
1318#ifdef CONFIG_COMPAT
1319int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1320 char __user *optval, int __user *optlen)
1321{
1322 if (level == SOL_UDP || level == SOL_UDPLITE)
1323 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1324 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1325}
1326#endif
1327
1328static int udp6_ufo_send_check(struct sk_buff *skb)
1329{
1330 const struct ipv6hdr *ipv6h;
1331 struct udphdr *uh;
1332
1333 if (!pskb_may_pull(skb, sizeof(*uh)))
1334 return -EINVAL;
1335
1336 ipv6h = ipv6_hdr(skb);
1337 uh = udp_hdr(skb);
1338
1339 uh->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, skb->len,
1340 IPPROTO_UDP, 0);
1341 skb->csum_start = skb_transport_header(skb) - skb->head;
1342 skb->csum_offset = offsetof(struct udphdr, check);
1343 skb->ip_summed = CHECKSUM_PARTIAL;
1344 return 0;
1345}
1346
1347static struct sk_buff *udp6_ufo_fragment(struct sk_buff *skb,
1348 netdev_features_t features)
1349{
1350 struct sk_buff *segs = ERR_PTR(-EINVAL);
1351 unsigned int mss;
1352 unsigned int unfrag_ip6hlen, unfrag_len;
1353 struct frag_hdr *fptr;
1354 u8 *mac_start, *prevhdr;
1355 u8 nexthdr;
1356 u8 frag_hdr_sz = sizeof(struct frag_hdr);
1357 int offset;
1358 __wsum csum;
1359
1360 mss = skb_shinfo(skb)->gso_size;
1361 if (unlikely(skb->len <= mss))
1362 goto out;
1363
1364 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
1365 /* Packet is from an untrusted source, reset gso_segs. */
1366 int type = skb_shinfo(skb)->gso_type;
1367
1368 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
1369 !(type & (SKB_GSO_UDP))))
1370 goto out;
1371
1372 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
1373
1374 segs = NULL;
1375 goto out;
1376 }
1377
1378 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
1379 * do checksum of UDP packets sent as multiple IP fragments.
1380 */
1381 offset = skb_checksum_start_offset(skb);
1382 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1383 offset += skb->csum_offset;
1384 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1385 skb->ip_summed = CHECKSUM_NONE;
1386
1387 /* Check if there is enough headroom to insert fragment header. */
1388 if ((skb_mac_header(skb) < skb->head + frag_hdr_sz) &&
1389 pskb_expand_head(skb, frag_hdr_sz, 0, GFP_ATOMIC))
1390 goto out;
1391
1392 /* Find the unfragmentable header and shift it left by frag_hdr_sz
1393 * bytes to insert fragment header.
1394 */
1395 unfrag_ip6hlen = ip6_find_1stfragopt(skb, &prevhdr);
1396 nexthdr = *prevhdr;
1397 *prevhdr = NEXTHDR_FRAGMENT;
1398 unfrag_len = skb_network_header(skb) - skb_mac_header(skb) +
1399 unfrag_ip6hlen;
1400 mac_start = skb_mac_header(skb);
1401 memmove(mac_start-frag_hdr_sz, mac_start, unfrag_len);
1402
1403 skb->mac_header -= frag_hdr_sz;
1404 skb->network_header -= frag_hdr_sz;
1405
1406 fptr = (struct frag_hdr *)(skb_network_header(skb) + unfrag_ip6hlen);
1407 fptr->nexthdr = nexthdr;
1408 fptr->reserved = 0;
1409 ipv6_select_ident(fptr, (struct rt6_info *)skb_dst(skb));
1410
1411 /* Fragment the skb. ipv6 header and the remaining fields of the
1412 * fragment header are updated in ipv6_gso_segment()
1413 */
1414 segs = skb_segment(skb, features);
1415
1416out:
1417 return segs;
1418}
1419
1420static const struct inet6_protocol udpv6_protocol = {
1421 .handler = udpv6_rcv,
1422 .err_handler = udpv6_err,
1423 .gso_send_check = udp6_ufo_send_check,
1424 .gso_segment = udp6_ufo_fragment,
1425 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1426};
1427
1428/* ------------------------------------------------------------------------ */
1429#ifdef CONFIG_PROC_FS
1430
1431static void udp6_sock_seq_show(struct seq_file *seq, struct sock *sp, int bucket)
1432{
1433 struct inet_sock *inet = inet_sk(sp);
1434 struct ipv6_pinfo *np = inet6_sk(sp);
1435 const struct in6_addr *dest, *src;
1436 __u16 destp, srcp;
1437
1438 dest = &np->daddr;
1439 src = &np->rcv_saddr;
1440 destp = ntohs(inet->inet_dport);
1441 srcp = ntohs(inet->inet_sport);
1442 seq_printf(seq,
1443 "%5d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1444 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d\n",
1445 bucket,
1446 src->s6_addr32[0], src->s6_addr32[1],
1447 src->s6_addr32[2], src->s6_addr32[3], srcp,
1448 dest->s6_addr32[0], dest->s6_addr32[1],
1449 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1450 sp->sk_state,
1451 sk_wmem_alloc_get(sp),
1452 sk_rmem_alloc_get(sp),
1453 0, 0L, 0,
1454 sock_i_uid(sp), 0,
1455 sock_i_ino(sp),
1456 atomic_read(&sp->sk_refcnt), sp,
1457 atomic_read(&sp->sk_drops));
1458}
1459
1460int udp6_seq_show(struct seq_file *seq, void *v)
1461{
1462 if (v == SEQ_START_TOKEN)
1463 seq_printf(seq,
1464 " sl "
1465 "local_address "
1466 "remote_address "
1467 "st tx_queue rx_queue tr tm->when retrnsmt"
1468 " uid timeout inode ref pointer drops\n");
1469 else
1470 udp6_sock_seq_show(seq, v, ((struct udp_iter_state *)seq->private)->bucket);
1471 return 0;
1472}
1473
1474static const struct file_operations udp6_afinfo_seq_fops = {
1475 .owner = THIS_MODULE,
1476 .open = udp_seq_open,
1477 .read = seq_read,
1478 .llseek = seq_lseek,
1479 .release = seq_release_net
1480};
1481
1482static struct udp_seq_afinfo udp6_seq_afinfo = {
1483 .name = "udp6",
1484 .family = AF_INET6,
1485 .udp_table = &udp_table,
1486 .seq_fops = &udp6_afinfo_seq_fops,
1487 .seq_ops = {
1488 .show = udp6_seq_show,
1489 },
1490};
1491
1492int __net_init udp6_proc_init(struct net *net)
1493{
1494 return udp_proc_register(net, &udp6_seq_afinfo);
1495}
1496
1497void udp6_proc_exit(struct net *net) {
1498 udp_proc_unregister(net, &udp6_seq_afinfo);
1499}
1500#endif /* CONFIG_PROC_FS */
1501
1502/* ------------------------------------------------------------------------ */
1503
1504struct proto udpv6_prot = {
1505 .name = "UDPv6",
1506 .owner = THIS_MODULE,
1507 .close = udp_lib_close,
1508 .connect = ip6_datagram_connect,
1509 .disconnect = udp_disconnect,
1510 .ioctl = udp_ioctl,
1511 .destroy = udpv6_destroy_sock,
1512 .setsockopt = udpv6_setsockopt,
1513 .getsockopt = udpv6_getsockopt,
1514 .sendmsg = udpv6_sendmsg,
1515 .recvmsg = udpv6_recvmsg,
1516 .backlog_rcv = __udpv6_queue_rcv_skb,
1517 .hash = udp_lib_hash,
1518 .unhash = udp_lib_unhash,
1519 .rehash = udp_v6_rehash,
1520 .get_port = udp_v6_get_port,
1521 .memory_allocated = &udp_memory_allocated,
1522 .sysctl_mem = sysctl_udp_mem,
1523 .sysctl_wmem = &sysctl_udp_wmem_min,
1524 .sysctl_rmem = &sysctl_udp_rmem_min,
1525 .obj_size = sizeof(struct udp6_sock),
1526 .slab_flags = SLAB_DESTROY_BY_RCU,
1527 .h.udp_table = &udp_table,
1528#ifdef CONFIG_COMPAT
1529 .compat_setsockopt = compat_udpv6_setsockopt,
1530 .compat_getsockopt = compat_udpv6_getsockopt,
1531#endif
1532 .clear_sk = sk_prot_clear_portaddr_nulls,
1533};
1534
1535static struct inet_protosw udpv6_protosw = {
1536 .type = SOCK_DGRAM,
1537 .protocol = IPPROTO_UDP,
1538 .prot = &udpv6_prot,
1539 .ops = &inet6_dgram_ops,
1540 .no_check = UDP_CSUM_DEFAULT,
1541 .flags = INET_PROTOSW_PERMANENT,
1542};
1543
1544
1545int __init udpv6_init(void)
1546{
1547 int ret;
1548
1549 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1550 if (ret)
1551 goto out;
1552
1553 ret = inet6_register_protosw(&udpv6_protosw);
1554 if (ret)
1555 goto out_udpv6_protocol;
1556out:
1557 return ret;
1558
1559out_udpv6_protocol:
1560 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1561 goto out;
1562}
1563
1564void udpv6_exit(void)
1565{
1566 inet6_unregister_protosw(&udpv6_protosw);
1567 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1568}