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/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
241static struct 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}
308
309static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
310 __be16 sport, __be16 dport,
311 struct udp_table *udptable)
312{
313 struct sock *sk;
314 const struct ipv6hdr *iph = ipv6_hdr(skb);
315
316 if (unlikely(sk = skb_steal_sock(skb)))
317 return sk;
318 return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport,
319 &iph->daddr, dport, inet6_iif(skb),
320 udptable);
321}
322
323struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
324 const struct in6_addr *daddr, __be16 dport, int dif)
325{
326 return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
327}
328EXPORT_SYMBOL_GPL(udp6_lib_lookup);
329
330
331/*
332 * This should be easy, if there is something there we
333 * return it, otherwise we block.
334 */
335
336int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk,
337 struct msghdr *msg, size_t len,
338 int noblock, int flags, int *addr_len)
339{
340 struct ipv6_pinfo *np = inet6_sk(sk);
341 struct inet_sock *inet = inet_sk(sk);
342 struct sk_buff *skb;
343 unsigned int ulen;
344 int peeked;
345 int err;
346 int is_udplite = IS_UDPLITE(sk);
347 int is_udp4;
348 bool slow;
349
350 if (addr_len)
351 *addr_len=sizeof(struct sockaddr_in6);
352
353 if (flags & MSG_ERRQUEUE)
354 return ipv6_recv_error(sk, msg, len);
355
356 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
357 return ipv6_recv_rxpmtu(sk, msg, len);
358
359try_again:
360 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
361 &peeked, &err);
362 if (!skb)
363 goto out;
364
365 ulen = skb->len - sizeof(struct udphdr);
366 if (len > ulen)
367 len = ulen;
368 else if (len < ulen)
369 msg->msg_flags |= MSG_TRUNC;
370
371 is_udp4 = (skb->protocol == htons(ETH_P_IP));
372
373 /*
374 * If checksum is needed at all, try to do it while copying the
375 * data. If the data is truncated, or if we only want a partial
376 * coverage checksum (UDP-Lite), do it before the copy.
377 */
378
379 if (len < ulen || UDP_SKB_CB(skb)->partial_cov) {
380 if (udp_lib_checksum_complete(skb))
381 goto csum_copy_err;
382 }
383
384 if (skb_csum_unnecessary(skb))
385 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
386 msg->msg_iov,len);
387 else {
388 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
389 if (err == -EINVAL)
390 goto csum_copy_err;
391 }
392 if (err)
393 goto out_free;
394
395 if (!peeked) {
396 if (is_udp4)
397 UDP_INC_STATS_USER(sock_net(sk),
398 UDP_MIB_INDATAGRAMS, is_udplite);
399 else
400 UDP6_INC_STATS_USER(sock_net(sk),
401 UDP_MIB_INDATAGRAMS, is_udplite);
402 }
403
404 sock_recv_ts_and_drops(msg, sk, skb);
405
406 /* Copy the address. */
407 if (msg->msg_name) {
408 struct sockaddr_in6 *sin6;
409
410 sin6 = (struct sockaddr_in6 *) msg->msg_name;
411 sin6->sin6_family = AF_INET6;
412 sin6->sin6_port = udp_hdr(skb)->source;
413 sin6->sin6_flowinfo = 0;
414 sin6->sin6_scope_id = 0;
415
416 if (is_udp4)
417 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
418 &sin6->sin6_addr);
419 else {
420 ipv6_addr_copy(&sin6->sin6_addr,
421 &ipv6_hdr(skb)->saddr);
422 if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
423 sin6->sin6_scope_id = IP6CB(skb)->iif;
424 }
425
426 }
427 if (is_udp4) {
428 if (inet->cmsg_flags)
429 ip_cmsg_recv(msg, skb);
430 } else {
431 if (np->rxopt.all)
432 datagram_recv_ctl(sk, msg, skb);
433 }
434
435 err = len;
436 if (flags & MSG_TRUNC)
437 err = ulen;
438
439out_free:
440 skb_free_datagram_locked(sk, skb);
441out:
442 return err;
443
444csum_copy_err:
445 slow = lock_sock_fast(sk);
446 if (!skb_kill_datagram(sk, skb, flags)) {
447 if (is_udp4)
448 UDP_INC_STATS_USER(sock_net(sk),
449 UDP_MIB_INERRORS, is_udplite);
450 else
451 UDP6_INC_STATS_USER(sock_net(sk),
452 UDP_MIB_INERRORS, is_udplite);
453 }
454 unlock_sock_fast(sk, slow);
455
456 if (noblock)
457 return -EAGAIN;
458
459 /* starting over for a new packet */
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 err;
475
476 sk = __udp6_lib_lookup(dev_net(skb->dev), daddr, uh->dest,
477 saddr, uh->source, inet6_iif(skb), udptable);
478 if (sk == NULL)
479 return;
480
481 np = inet6_sk(sk);
482
483 if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
484 goto out;
485
486 if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
487 goto out;
488
489 if (np->recverr)
490 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
491
492 sk->sk_err = err;
493 sk->sk_error_report(sk);
494out:
495 sock_put(sk);
496}
497
498static __inline__ void udpv6_err(struct sk_buff *skb,
499 struct inet6_skb_parm *opt, u8 type,
500 u8 code, int offset, __be32 info )
501{
502 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
503}
504
505int udpv6_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
506{
507 struct udp_sock *up = udp_sk(sk);
508 int rc;
509 int is_udplite = IS_UDPLITE(sk);
510
511 if (!ipv6_addr_any(&inet6_sk(sk)->daddr))
512 sock_rps_save_rxhash(sk, skb->rxhash);
513
514 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
515 goto drop;
516
517 /*
518 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
519 */
520 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
521
522 if (up->pcrlen == 0) { /* full coverage was set */
523 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage"
524 " %d while full coverage %d requested\n",
525 UDP_SKB_CB(skb)->cscov, skb->len);
526 goto drop;
527 }
528 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
529 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d "
530 "too small, need min %d\n",
531 UDP_SKB_CB(skb)->cscov, up->pcrlen);
532 goto drop;
533 }
534 }
535
536 if (rcu_dereference_raw(sk->sk_filter)) {
537 if (udp_lib_checksum_complete(skb))
538 goto drop;
539 }
540
541 if ((rc = ip_queue_rcv_skb(sk, skb)) < 0) {
542 /* Note that an ENOMEM error is charged twice */
543 if (rc == -ENOMEM)
544 UDP6_INC_STATS_BH(sock_net(sk),
545 UDP_MIB_RCVBUFERRORS, is_udplite);
546 goto drop_no_sk_drops_inc;
547 }
548
549 return 0;
550drop:
551 atomic_inc(&sk->sk_drops);
552drop_no_sk_drops_inc:
553 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
554 kfree_skb(skb);
555 return -1;
556}
557
558static struct sock *udp_v6_mcast_next(struct net *net, struct sock *sk,
559 __be16 loc_port, const struct in6_addr *loc_addr,
560 __be16 rmt_port, const struct in6_addr *rmt_addr,
561 int dif)
562{
563 struct hlist_nulls_node *node;
564 struct sock *s = sk;
565 unsigned short num = ntohs(loc_port);
566
567 sk_nulls_for_each_from(s, node) {
568 struct inet_sock *inet = inet_sk(s);
569
570 if (!net_eq(sock_net(s), net))
571 continue;
572
573 if (udp_sk(s)->udp_port_hash == num &&
574 s->sk_family == PF_INET6) {
575 struct ipv6_pinfo *np = inet6_sk(s);
576 if (inet->inet_dport) {
577 if (inet->inet_dport != rmt_port)
578 continue;
579 }
580 if (!ipv6_addr_any(&np->daddr) &&
581 !ipv6_addr_equal(&np->daddr, rmt_addr))
582 continue;
583
584 if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)
585 continue;
586
587 if (!ipv6_addr_any(&np->rcv_saddr)) {
588 if (!ipv6_addr_equal(&np->rcv_saddr, loc_addr))
589 continue;
590 }
591 if (!inet6_mc_check(s, loc_addr, rmt_addr))
592 continue;
593 return s;
594 }
595 }
596 return NULL;
597}
598
599static void flush_stack(struct sock **stack, unsigned int count,
600 struct sk_buff *skb, unsigned int final)
601{
602 unsigned int i;
603 struct sock *sk;
604 struct sk_buff *skb1;
605
606 for (i = 0; i < count; i++) {
607 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
608
609 sk = stack[i];
610 if (skb1) {
611 if (sk_rcvqueues_full(sk, skb1)) {
612 kfree_skb(skb1);
613 goto drop;
614 }
615 bh_lock_sock(sk);
616 if (!sock_owned_by_user(sk))
617 udpv6_queue_rcv_skb(sk, skb1);
618 else if (sk_add_backlog(sk, skb1)) {
619 kfree_skb(skb1);
620 bh_unlock_sock(sk);
621 goto drop;
622 }
623 bh_unlock_sock(sk);
624 continue;
625 }
626drop:
627 atomic_inc(&sk->sk_drops);
628 UDP6_INC_STATS_BH(sock_net(sk),
629 UDP_MIB_RCVBUFERRORS, IS_UDPLITE(sk));
630 UDP6_INC_STATS_BH(sock_net(sk),
631 UDP_MIB_INERRORS, IS_UDPLITE(sk));
632 }
633}
634/*
635 * Note: called only from the BH handler context,
636 * so we don't need to lock the hashes.
637 */
638static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
639 const struct in6_addr *saddr, const struct in6_addr *daddr,
640 struct udp_table *udptable)
641{
642 struct sock *sk, *stack[256 / sizeof(struct sock *)];
643 const struct udphdr *uh = udp_hdr(skb);
644 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
645 int dif;
646 unsigned int i, count = 0;
647
648 spin_lock(&hslot->lock);
649 sk = sk_nulls_head(&hslot->head);
650 dif = inet6_iif(skb);
651 sk = udp_v6_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
652 while (sk) {
653 stack[count++] = sk;
654 sk = udp_v6_mcast_next(net, sk_nulls_next(sk), uh->dest, daddr,
655 uh->source, saddr, dif);
656 if (unlikely(count == ARRAY_SIZE(stack))) {
657 if (!sk)
658 break;
659 flush_stack(stack, count, skb, ~0);
660 count = 0;
661 }
662 }
663 /*
664 * before releasing the lock, we must take reference on sockets
665 */
666 for (i = 0; i < count; i++)
667 sock_hold(stack[i]);
668
669 spin_unlock(&hslot->lock);
670
671 if (count) {
672 flush_stack(stack, count, skb, count - 1);
673
674 for (i = 0; i < count; i++)
675 sock_put(stack[i]);
676 } else {
677 kfree_skb(skb);
678 }
679 return 0;
680}
681
682static inline int udp6_csum_init(struct sk_buff *skb, struct udphdr *uh,
683 int proto)
684{
685 int err;
686
687 UDP_SKB_CB(skb)->partial_cov = 0;
688 UDP_SKB_CB(skb)->cscov = skb->len;
689
690 if (proto == IPPROTO_UDPLITE) {
691 err = udplite_checksum_init(skb, uh);
692 if (err)
693 return err;
694 }
695
696 if (uh->check == 0) {
697 /* RFC 2460 section 8.1 says that we SHOULD log
698 this error. Well, it is reasonable.
699 */
700 LIMIT_NETDEBUG(KERN_INFO "IPv6: udp checksum is 0\n");
701 return 1;
702 }
703 if (skb->ip_summed == CHECKSUM_COMPLETE &&
704 !csum_ipv6_magic(&ipv6_hdr(skb)->saddr, &ipv6_hdr(skb)->daddr,
705 skb->len, proto, skb->csum))
706 skb->ip_summed = CHECKSUM_UNNECESSARY;
707
708 if (!skb_csum_unnecessary(skb))
709 skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
710 &ipv6_hdr(skb)->daddr,
711 skb->len, proto, 0));
712
713 return 0;
714}
715
716int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
717 int proto)
718{
719 struct net *net = dev_net(skb->dev);
720 struct sock *sk;
721 struct udphdr *uh;
722 const struct in6_addr *saddr, *daddr;
723 u32 ulen = 0;
724
725 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
726 goto discard;
727
728 saddr = &ipv6_hdr(skb)->saddr;
729 daddr = &ipv6_hdr(skb)->daddr;
730 uh = udp_hdr(skb);
731
732 ulen = ntohs(uh->len);
733 if (ulen > skb->len)
734 goto short_packet;
735
736 if (proto == IPPROTO_UDP) {
737 /* UDP validates ulen. */
738
739 /* Check for jumbo payload */
740 if (ulen == 0)
741 ulen = skb->len;
742
743 if (ulen < sizeof(*uh))
744 goto short_packet;
745
746 if (ulen < skb->len) {
747 if (pskb_trim_rcsum(skb, ulen))
748 goto short_packet;
749 saddr = &ipv6_hdr(skb)->saddr;
750 daddr = &ipv6_hdr(skb)->daddr;
751 uh = udp_hdr(skb);
752 }
753 }
754
755 if (udp6_csum_init(skb, uh, proto))
756 goto discard;
757
758 /*
759 * Multicast receive code
760 */
761 if (ipv6_addr_is_multicast(daddr))
762 return __udp6_lib_mcast_deliver(net, skb,
763 saddr, daddr, udptable);
764
765 /* Unicast */
766
767 /*
768 * check socket cache ... must talk to Alan about his plans
769 * for sock caches... i'll skip this for now.
770 */
771 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
772
773 if (sk == NULL) {
774 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
775 goto discard;
776
777 if (udp_lib_checksum_complete(skb))
778 goto discard;
779 UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS,
780 proto == IPPROTO_UDPLITE);
781
782 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
783
784 kfree_skb(skb);
785 return 0;
786 }
787
788 /* deliver */
789
790 if (sk_rcvqueues_full(sk, skb)) {
791 sock_put(sk);
792 goto discard;
793 }
794 bh_lock_sock(sk);
795 if (!sock_owned_by_user(sk))
796 udpv6_queue_rcv_skb(sk, skb);
797 else if (sk_add_backlog(sk, skb)) {
798 atomic_inc(&sk->sk_drops);
799 bh_unlock_sock(sk);
800 sock_put(sk);
801 goto discard;
802 }
803 bh_unlock_sock(sk);
804 sock_put(sk);
805 return 0;
806
807short_packet:
808 LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
809 proto == IPPROTO_UDPLITE ? "-Lite" : "",
810 saddr,
811 ntohs(uh->source),
812 ulen,
813 skb->len,
814 daddr,
815 ntohs(uh->dest));
816
817discard:
818 UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
819 kfree_skb(skb);
820 return 0;
821}
822
823static __inline__ int udpv6_rcv(struct sk_buff *skb)
824{
825 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
826}
827
828/*
829 * Throw away all pending data and cancel the corking. Socket is locked.
830 */
831static void udp_v6_flush_pending_frames(struct sock *sk)
832{
833 struct udp_sock *up = udp_sk(sk);
834
835 if (up->pending == AF_INET)
836 udp_flush_pending_frames(sk);
837 else if (up->pending) {
838 up->len = 0;
839 up->pending = 0;
840 ip6_flush_pending_frames(sk);
841 }
842}
843
844/**
845 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
846 * @sk: socket we are sending on
847 * @skb: sk_buff containing the filled-in UDP header
848 * (checksum field must be zeroed out)
849 */
850static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
851 const struct in6_addr *saddr,
852 const struct in6_addr *daddr, int len)
853{
854 unsigned int offset;
855 struct udphdr *uh = udp_hdr(skb);
856 __wsum csum = 0;
857
858 if (skb_queue_len(&sk->sk_write_queue) == 1) {
859 /* Only one fragment on the socket. */
860 skb->csum_start = skb_transport_header(skb) - skb->head;
861 skb->csum_offset = offsetof(struct udphdr, check);
862 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
863 } else {
864 /*
865 * HW-checksum won't work as there are two or more
866 * fragments on the socket so that all csums of sk_buffs
867 * should be together
868 */
869 offset = skb_transport_offset(skb);
870 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
871
872 skb->ip_summed = CHECKSUM_NONE;
873
874 skb_queue_walk(&sk->sk_write_queue, skb) {
875 csum = csum_add(csum, skb->csum);
876 }
877
878 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
879 csum);
880 if (uh->check == 0)
881 uh->check = CSUM_MANGLED_0;
882 }
883}
884
885/*
886 * Sending
887 */
888
889static int udp_v6_push_pending_frames(struct sock *sk)
890{
891 struct sk_buff *skb;
892 struct udphdr *uh;
893 struct udp_sock *up = udp_sk(sk);
894 struct inet_sock *inet = inet_sk(sk);
895 struct flowi6 *fl6 = &inet->cork.fl.u.ip6;
896 int err = 0;
897 int is_udplite = IS_UDPLITE(sk);
898 __wsum csum = 0;
899
900 /* Grab the skbuff where UDP header space exists. */
901 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
902 goto out;
903
904 /*
905 * Create a UDP header
906 */
907 uh = udp_hdr(skb);
908 uh->source = fl6->fl6_sport;
909 uh->dest = fl6->fl6_dport;
910 uh->len = htons(up->len);
911 uh->check = 0;
912
913 if (is_udplite)
914 csum = udplite_csum_outgoing(sk, skb);
915 else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
916 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr,
917 up->len);
918 goto send;
919 } else
920 csum = udp_csum_outgoing(sk, skb);
921
922 /* add protocol-dependent pseudo-header */
923 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
924 up->len, fl6->flowi6_proto, csum);
925 if (uh->check == 0)
926 uh->check = CSUM_MANGLED_0;
927
928send:
929 err = ip6_push_pending_frames(sk);
930 if (err) {
931 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
932 UDP6_INC_STATS_USER(sock_net(sk),
933 UDP_MIB_SNDBUFERRORS, is_udplite);
934 err = 0;
935 }
936 } else
937 UDP6_INC_STATS_USER(sock_net(sk),
938 UDP_MIB_OUTDATAGRAMS, is_udplite);
939out:
940 up->len = 0;
941 up->pending = 0;
942 return err;
943}
944
945int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
946 struct msghdr *msg, size_t len)
947{
948 struct ipv6_txoptions opt_space;
949 struct udp_sock *up = udp_sk(sk);
950 struct inet_sock *inet = inet_sk(sk);
951 struct ipv6_pinfo *np = inet6_sk(sk);
952 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) msg->msg_name;
953 struct in6_addr *daddr, *final_p, final;
954 struct ipv6_txoptions *opt = NULL;
955 struct ip6_flowlabel *flowlabel = NULL;
956 struct flowi6 fl6;
957 struct dst_entry *dst;
958 int addr_len = msg->msg_namelen;
959 int ulen = len;
960 int hlimit = -1;
961 int tclass = -1;
962 int dontfrag = -1;
963 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
964 int err;
965 int connected = 0;
966 int is_udplite = IS_UDPLITE(sk);
967 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
968
969 /* destination address check */
970 if (sin6) {
971 if (addr_len < offsetof(struct sockaddr, sa_data))
972 return -EINVAL;
973
974 switch (sin6->sin6_family) {
975 case AF_INET6:
976 if (addr_len < SIN6_LEN_RFC2133)
977 return -EINVAL;
978 daddr = &sin6->sin6_addr;
979 break;
980 case AF_INET:
981 goto do_udp_sendmsg;
982 case AF_UNSPEC:
983 msg->msg_name = sin6 = NULL;
984 msg->msg_namelen = addr_len = 0;
985 daddr = NULL;
986 break;
987 default:
988 return -EINVAL;
989 }
990 } else if (!up->pending) {
991 if (sk->sk_state != TCP_ESTABLISHED)
992 return -EDESTADDRREQ;
993 daddr = &np->daddr;
994 } else
995 daddr = NULL;
996
997 if (daddr) {
998 if (ipv6_addr_v4mapped(daddr)) {
999 struct sockaddr_in sin;
1000 sin.sin_family = AF_INET;
1001 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1002 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1003 msg->msg_name = &sin;
1004 msg->msg_namelen = sizeof(sin);
1005do_udp_sendmsg:
1006 if (__ipv6_only_sock(sk))
1007 return -ENETUNREACH;
1008 return udp_sendmsg(iocb, sk, msg, len);
1009 }
1010 }
1011
1012 if (up->pending == AF_INET)
1013 return udp_sendmsg(iocb, sk, msg, len);
1014
1015 /* Rough check on arithmetic overflow,
1016 better check is made in ip6_append_data().
1017 */
1018 if (len > INT_MAX - sizeof(struct udphdr))
1019 return -EMSGSIZE;
1020
1021 if (up->pending) {
1022 /*
1023 * There are pending frames.
1024 * The socket lock must be held while it's corked.
1025 */
1026 lock_sock(sk);
1027 if (likely(up->pending)) {
1028 if (unlikely(up->pending != AF_INET6)) {
1029 release_sock(sk);
1030 return -EAFNOSUPPORT;
1031 }
1032 dst = NULL;
1033 goto do_append_data;
1034 }
1035 release_sock(sk);
1036 }
1037 ulen += sizeof(struct udphdr);
1038
1039 memset(&fl6, 0, sizeof(fl6));
1040
1041 if (sin6) {
1042 if (sin6->sin6_port == 0)
1043 return -EINVAL;
1044
1045 fl6.fl6_dport = sin6->sin6_port;
1046 daddr = &sin6->sin6_addr;
1047
1048 if (np->sndflow) {
1049 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1050 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1051 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1052 if (flowlabel == NULL)
1053 return -EINVAL;
1054 daddr = &flowlabel->dst;
1055 }
1056 }
1057
1058 /*
1059 * Otherwise it will be difficult to maintain
1060 * sk->sk_dst_cache.
1061 */
1062 if (sk->sk_state == TCP_ESTABLISHED &&
1063 ipv6_addr_equal(daddr, &np->daddr))
1064 daddr = &np->daddr;
1065
1066 if (addr_len >= sizeof(struct sockaddr_in6) &&
1067 sin6->sin6_scope_id &&
1068 ipv6_addr_type(daddr)&IPV6_ADDR_LINKLOCAL)
1069 fl6.flowi6_oif = sin6->sin6_scope_id;
1070 } else {
1071 if (sk->sk_state != TCP_ESTABLISHED)
1072 return -EDESTADDRREQ;
1073
1074 fl6.fl6_dport = inet->inet_dport;
1075 daddr = &np->daddr;
1076 fl6.flowlabel = np->flow_label;
1077 connected = 1;
1078 }
1079
1080 if (!fl6.flowi6_oif)
1081 fl6.flowi6_oif = sk->sk_bound_dev_if;
1082
1083 if (!fl6.flowi6_oif)
1084 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1085
1086 fl6.flowi6_mark = sk->sk_mark;
1087
1088 if (msg->msg_controllen) {
1089 opt = &opt_space;
1090 memset(opt, 0, sizeof(struct ipv6_txoptions));
1091 opt->tot_len = sizeof(*opt);
1092
1093 err = datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt,
1094 &hlimit, &tclass, &dontfrag);
1095 if (err < 0) {
1096 fl6_sock_release(flowlabel);
1097 return err;
1098 }
1099 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1100 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1101 if (flowlabel == NULL)
1102 return -EINVAL;
1103 }
1104 if (!(opt->opt_nflen|opt->opt_flen))
1105 opt = NULL;
1106 connected = 0;
1107 }
1108 if (opt == NULL)
1109 opt = np->opt;
1110 if (flowlabel)
1111 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1112 opt = ipv6_fixup_options(&opt_space, opt);
1113
1114 fl6.flowi6_proto = sk->sk_protocol;
1115 if (!ipv6_addr_any(daddr))
1116 ipv6_addr_copy(&fl6.daddr, daddr);
1117 else
1118 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1119 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1120 ipv6_addr_copy(&fl6.saddr, &np->saddr);
1121 fl6.fl6_sport = inet->inet_sport;
1122
1123 final_p = fl6_update_dst(&fl6, opt, &final);
1124 if (final_p)
1125 connected = 0;
1126
1127 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1128 fl6.flowi6_oif = np->mcast_oif;
1129 connected = 0;
1130 }
1131
1132 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1133
1134 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, true);
1135 if (IS_ERR(dst)) {
1136 err = PTR_ERR(dst);
1137 dst = NULL;
1138 goto out;
1139 }
1140
1141 if (hlimit < 0) {
1142 if (ipv6_addr_is_multicast(&fl6.daddr))
1143 hlimit = np->mcast_hops;
1144 else
1145 hlimit = np->hop_limit;
1146 if (hlimit < 0)
1147 hlimit = ip6_dst_hoplimit(dst);
1148 }
1149
1150 if (tclass < 0)
1151 tclass = np->tclass;
1152
1153 if (dontfrag < 0)
1154 dontfrag = np->dontfrag;
1155
1156 if (msg->msg_flags&MSG_CONFIRM)
1157 goto do_confirm;
1158back_from_confirm:
1159
1160 lock_sock(sk);
1161 if (unlikely(up->pending)) {
1162 /* The socket is already corked while preparing it. */
1163 /* ... which is an evident application bug. --ANK */
1164 release_sock(sk);
1165
1166 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
1167 err = -EINVAL;
1168 goto out;
1169 }
1170
1171 up->pending = AF_INET6;
1172
1173do_append_data:
1174 up->len += ulen;
1175 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1176 err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen,
1177 sizeof(struct udphdr), hlimit, tclass, opt, &fl6,
1178 (struct rt6_info*)dst,
1179 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag);
1180 if (err)
1181 udp_v6_flush_pending_frames(sk);
1182 else if (!corkreq)
1183 err = udp_v6_push_pending_frames(sk);
1184 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1185 up->pending = 0;
1186
1187 if (dst) {
1188 if (connected) {
1189 ip6_dst_store(sk, dst,
1190 ipv6_addr_equal(&fl6.daddr, &np->daddr) ?
1191 &np->daddr : NULL,
1192#ifdef CONFIG_IPV6_SUBTREES
1193 ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1194 &np->saddr :
1195#endif
1196 NULL);
1197 } else {
1198 dst_release(dst);
1199 }
1200 dst = NULL;
1201 }
1202
1203 if (err > 0)
1204 err = np->recverr ? net_xmit_errno(err) : 0;
1205 release_sock(sk);
1206out:
1207 dst_release(dst);
1208 fl6_sock_release(flowlabel);
1209 if (!err)
1210 return len;
1211 /*
1212 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1213 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1214 * we don't have a good statistic (IpOutDiscards but it can be too many
1215 * things). We could add another new stat but at least for now that
1216 * seems like overkill.
1217 */
1218 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1219 UDP6_INC_STATS_USER(sock_net(sk),
1220 UDP_MIB_SNDBUFERRORS, is_udplite);
1221 }
1222 return err;
1223
1224do_confirm:
1225 dst_confirm(dst);
1226 if (!(msg->msg_flags&MSG_PROBE) || len)
1227 goto back_from_confirm;
1228 err = 0;
1229 goto out;
1230}
1231
1232void udpv6_destroy_sock(struct sock *sk)
1233{
1234 lock_sock(sk);
1235 udp_v6_flush_pending_frames(sk);
1236 release_sock(sk);
1237
1238 inet6_destroy_sock(sk);
1239}
1240
1241/*
1242 * Socket option code for UDP
1243 */
1244int udpv6_setsockopt(struct sock *sk, int level, int optname,
1245 char __user *optval, unsigned int optlen)
1246{
1247 if (level == SOL_UDP || level == SOL_UDPLITE)
1248 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1249 udp_v6_push_pending_frames);
1250 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1251}
1252
1253#ifdef CONFIG_COMPAT
1254int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1255 char __user *optval, unsigned int optlen)
1256{
1257 if (level == SOL_UDP || level == SOL_UDPLITE)
1258 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1259 udp_v6_push_pending_frames);
1260 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1261}
1262#endif
1263
1264int udpv6_getsockopt(struct sock *sk, int level, int optname,
1265 char __user *optval, int __user *optlen)
1266{
1267 if (level == SOL_UDP || level == SOL_UDPLITE)
1268 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1269 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1270}
1271
1272#ifdef CONFIG_COMPAT
1273int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1274 char __user *optval, int __user *optlen)
1275{
1276 if (level == SOL_UDP || level == SOL_UDPLITE)
1277 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1278 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1279}
1280#endif
1281
1282static int udp6_ufo_send_check(struct sk_buff *skb)
1283{
1284 const struct ipv6hdr *ipv6h;
1285 struct udphdr *uh;
1286
1287 if (!pskb_may_pull(skb, sizeof(*uh)))
1288 return -EINVAL;
1289
1290 ipv6h = ipv6_hdr(skb);
1291 uh = udp_hdr(skb);
1292
1293 uh->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, skb->len,
1294 IPPROTO_UDP, 0);
1295 skb->csum_start = skb_transport_header(skb) - skb->head;
1296 skb->csum_offset = offsetof(struct udphdr, check);
1297 skb->ip_summed = CHECKSUM_PARTIAL;
1298 return 0;
1299}
1300
1301static struct sk_buff *udp6_ufo_fragment(struct sk_buff *skb, u32 features)
1302{
1303 struct sk_buff *segs = ERR_PTR(-EINVAL);
1304 unsigned int mss;
1305 unsigned int unfrag_ip6hlen, unfrag_len;
1306 struct frag_hdr *fptr;
1307 u8 *mac_start, *prevhdr;
1308 u8 nexthdr;
1309 u8 frag_hdr_sz = sizeof(struct frag_hdr);
1310 int offset;
1311 __wsum csum;
1312
1313 mss = skb_shinfo(skb)->gso_size;
1314 if (unlikely(skb->len <= mss))
1315 goto out;
1316
1317 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
1318 /* Packet is from an untrusted source, reset gso_segs. */
1319 int type = skb_shinfo(skb)->gso_type;
1320
1321 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
1322 !(type & (SKB_GSO_UDP))))
1323 goto out;
1324
1325 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
1326
1327 segs = NULL;
1328 goto out;
1329 }
1330
1331 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
1332 * do checksum of UDP packets sent as multiple IP fragments.
1333 */
1334 offset = skb_checksum_start_offset(skb);
1335 csum = skb_checksum(skb, offset, skb->len- offset, 0);
1336 offset += skb->csum_offset;
1337 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1338 skb->ip_summed = CHECKSUM_NONE;
1339
1340 /* Check if there is enough headroom to insert fragment header. */
1341 if ((skb_mac_header(skb) < skb->head + frag_hdr_sz) &&
1342 pskb_expand_head(skb, frag_hdr_sz, 0, GFP_ATOMIC))
1343 goto out;
1344
1345 /* Find the unfragmentable header and shift it left by frag_hdr_sz
1346 * bytes to insert fragment header.
1347 */
1348 unfrag_ip6hlen = ip6_find_1stfragopt(skb, &prevhdr);
1349 nexthdr = *prevhdr;
1350 *prevhdr = NEXTHDR_FRAGMENT;
1351 unfrag_len = skb_network_header(skb) - skb_mac_header(skb) +
1352 unfrag_ip6hlen;
1353 mac_start = skb_mac_header(skb);
1354 memmove(mac_start-frag_hdr_sz, mac_start, unfrag_len);
1355
1356 skb->mac_header -= frag_hdr_sz;
1357 skb->network_header -= frag_hdr_sz;
1358
1359 fptr = (struct frag_hdr *)(skb_network_header(skb) + unfrag_ip6hlen);
1360 fptr->nexthdr = nexthdr;
1361 fptr->reserved = 0;
1362 ipv6_select_ident(fptr, (struct rt6_info *)skb_dst(skb));
1363
1364 /* Fragment the skb. ipv6 header and the remaining fields of the
1365 * fragment header are updated in ipv6_gso_segment()
1366 */
1367 segs = skb_segment(skb, features);
1368
1369out:
1370 return segs;
1371}
1372
1373static const struct inet6_protocol udpv6_protocol = {
1374 .handler = udpv6_rcv,
1375 .err_handler = udpv6_err,
1376 .gso_send_check = udp6_ufo_send_check,
1377 .gso_segment = udp6_ufo_fragment,
1378 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1379};
1380
1381/* ------------------------------------------------------------------------ */
1382#ifdef CONFIG_PROC_FS
1383
1384static void udp6_sock_seq_show(struct seq_file *seq, struct sock *sp, int bucket)
1385{
1386 struct inet_sock *inet = inet_sk(sp);
1387 struct ipv6_pinfo *np = inet6_sk(sp);
1388 const struct in6_addr *dest, *src;
1389 __u16 destp, srcp;
1390
1391 dest = &np->daddr;
1392 src = &np->rcv_saddr;
1393 destp = ntohs(inet->inet_dport);
1394 srcp = ntohs(inet->inet_sport);
1395 seq_printf(seq,
1396 "%5d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1397 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d\n",
1398 bucket,
1399 src->s6_addr32[0], src->s6_addr32[1],
1400 src->s6_addr32[2], src->s6_addr32[3], srcp,
1401 dest->s6_addr32[0], dest->s6_addr32[1],
1402 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1403 sp->sk_state,
1404 sk_wmem_alloc_get(sp),
1405 sk_rmem_alloc_get(sp),
1406 0, 0L, 0,
1407 sock_i_uid(sp), 0,
1408 sock_i_ino(sp),
1409 atomic_read(&sp->sk_refcnt), sp,
1410 atomic_read(&sp->sk_drops));
1411}
1412
1413int udp6_seq_show(struct seq_file *seq, void *v)
1414{
1415 if (v == SEQ_START_TOKEN)
1416 seq_printf(seq,
1417 " sl "
1418 "local_address "
1419 "remote_address "
1420 "st tx_queue rx_queue tr tm->when retrnsmt"
1421 " uid timeout inode ref pointer drops\n");
1422 else
1423 udp6_sock_seq_show(seq, v, ((struct udp_iter_state *)seq->private)->bucket);
1424 return 0;
1425}
1426
1427static struct udp_seq_afinfo udp6_seq_afinfo = {
1428 .name = "udp6",
1429 .family = AF_INET6,
1430 .udp_table = &udp_table,
1431 .seq_fops = {
1432 .owner = THIS_MODULE,
1433 },
1434 .seq_ops = {
1435 .show = udp6_seq_show,
1436 },
1437};
1438
1439int __net_init udp6_proc_init(struct net *net)
1440{
1441 return udp_proc_register(net, &udp6_seq_afinfo);
1442}
1443
1444void udp6_proc_exit(struct net *net) {
1445 udp_proc_unregister(net, &udp6_seq_afinfo);
1446}
1447#endif /* CONFIG_PROC_FS */
1448
1449/* ------------------------------------------------------------------------ */
1450
1451struct proto udpv6_prot = {
1452 .name = "UDPv6",
1453 .owner = THIS_MODULE,
1454 .close = udp_lib_close,
1455 .connect = ip6_datagram_connect,
1456 .disconnect = udp_disconnect,
1457 .ioctl = udp_ioctl,
1458 .destroy = udpv6_destroy_sock,
1459 .setsockopt = udpv6_setsockopt,
1460 .getsockopt = udpv6_getsockopt,
1461 .sendmsg = udpv6_sendmsg,
1462 .recvmsg = udpv6_recvmsg,
1463 .backlog_rcv = udpv6_queue_rcv_skb,
1464 .hash = udp_lib_hash,
1465 .unhash = udp_lib_unhash,
1466 .rehash = udp_v6_rehash,
1467 .get_port = udp_v6_get_port,
1468 .memory_allocated = &udp_memory_allocated,
1469 .sysctl_mem = sysctl_udp_mem,
1470 .sysctl_wmem = &sysctl_udp_wmem_min,
1471 .sysctl_rmem = &sysctl_udp_rmem_min,
1472 .obj_size = sizeof(struct udp6_sock),
1473 .slab_flags = SLAB_DESTROY_BY_RCU,
1474 .h.udp_table = &udp_table,
1475#ifdef CONFIG_COMPAT
1476 .compat_setsockopt = compat_udpv6_setsockopt,
1477 .compat_getsockopt = compat_udpv6_getsockopt,
1478#endif
1479 .clear_sk = sk_prot_clear_portaddr_nulls,
1480};
1481
1482static struct inet_protosw udpv6_protosw = {
1483 .type = SOCK_DGRAM,
1484 .protocol = IPPROTO_UDP,
1485 .prot = &udpv6_prot,
1486 .ops = &inet6_dgram_ops,
1487 .no_check = UDP_CSUM_DEFAULT,
1488 .flags = INET_PROTOSW_PERMANENT,
1489};
1490
1491
1492int __init udpv6_init(void)
1493{
1494 int ret;
1495
1496 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1497 if (ret)
1498 goto out;
1499
1500 ret = inet6_register_protosw(&udpv6_protosw);
1501 if (ret)
1502 goto out_udpv6_protocol;
1503out:
1504 return ret;
1505
1506out_udpv6_protocol:
1507 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1508 goto out;
1509}
1510
1511void udpv6_exit(void)
1512{
1513 inet6_unregister_protosw(&udpv6_protosw);
1514 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1515}
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * UDP over IPv6
4 * Linux INET6 implementation
5 *
6 * Authors:
7 * Pedro Roque <roque@di.fc.ul.pt>
8 *
9 * Based on linux/ipv4/udp.c
10 *
11 * Fixes:
12 * Hideaki YOSHIFUJI : sin6_scope_id support
13 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
14 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
15 * a single port at the same time.
16 * Kazunori MIYAZAWA @USAGI: change process style to use ip6_append_data
17 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/udp6 to seq_file.
18 */
19
20#include <linux/errno.h>
21#include <linux/types.h>
22#include <linux/socket.h>
23#include <linux/sockios.h>
24#include <linux/net.h>
25#include <linux/in6.h>
26#include <linux/netdevice.h>
27#include <linux/if_arp.h>
28#include <linux/ipv6.h>
29#include <linux/icmpv6.h>
30#include <linux/init.h>
31#include <linux/module.h>
32#include <linux/skbuff.h>
33#include <linux/slab.h>
34#include <linux/uaccess.h>
35#include <linux/indirect_call_wrapper.h>
36
37#include <net/addrconf.h>
38#include <net/ndisc.h>
39#include <net/protocol.h>
40#include <net/transp_v6.h>
41#include <net/ip6_route.h>
42#include <net/raw.h>
43#include <net/tcp_states.h>
44#include <net/ip6_checksum.h>
45#include <net/ip6_tunnel.h>
46#include <net/xfrm.h>
47#include <net/inet_hashtables.h>
48#include <net/inet6_hashtables.h>
49#include <net/busy_poll.h>
50#include <net/sock_reuseport.h>
51
52#include <linux/proc_fs.h>
53#include <linux/seq_file.h>
54#include <trace/events/skb.h>
55#include "udp_impl.h"
56
57static u32 udp6_ehashfn(const struct net *net,
58 const struct in6_addr *laddr,
59 const u16 lport,
60 const struct in6_addr *faddr,
61 const __be16 fport)
62{
63 static u32 udp6_ehash_secret __read_mostly;
64 static u32 udp_ipv6_hash_secret __read_mostly;
65
66 u32 lhash, fhash;
67
68 net_get_random_once(&udp6_ehash_secret,
69 sizeof(udp6_ehash_secret));
70 net_get_random_once(&udp_ipv6_hash_secret,
71 sizeof(udp_ipv6_hash_secret));
72
73 lhash = (__force u32)laddr->s6_addr32[3];
74 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
75
76 return __inet6_ehashfn(lhash, lport, fhash, fport,
77 udp_ipv6_hash_secret + net_hash_mix(net));
78}
79
80int udp_v6_get_port(struct sock *sk, unsigned short snum)
81{
82 unsigned int hash2_nulladdr =
83 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
84 unsigned int hash2_partial =
85 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
86
87 /* precompute partial secondary hash */
88 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
89 return udp_lib_get_port(sk, snum, hash2_nulladdr);
90}
91
92void udp_v6_rehash(struct sock *sk)
93{
94 u16 new_hash = ipv6_portaddr_hash(sock_net(sk),
95 &sk->sk_v6_rcv_saddr,
96 inet_sk(sk)->inet_num);
97
98 udp_lib_rehash(sk, new_hash);
99}
100
101static int compute_score(struct sock *sk, struct net *net,
102 const struct in6_addr *saddr, __be16 sport,
103 const struct in6_addr *daddr, unsigned short hnum,
104 int dif, int sdif)
105{
106 int score;
107 struct inet_sock *inet;
108 bool dev_match;
109
110 if (!net_eq(sock_net(sk), net) ||
111 udp_sk(sk)->udp_port_hash != hnum ||
112 sk->sk_family != PF_INET6)
113 return -1;
114
115 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
116 return -1;
117
118 score = 0;
119 inet = inet_sk(sk);
120
121 if (inet->inet_dport) {
122 if (inet->inet_dport != sport)
123 return -1;
124 score++;
125 }
126
127 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
128 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
129 return -1;
130 score++;
131 }
132
133 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif);
134 if (!dev_match)
135 return -1;
136 score++;
137
138 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
139 score++;
140
141 return score;
142}
143
144static struct sock *lookup_reuseport(struct net *net, struct sock *sk,
145 struct sk_buff *skb,
146 const struct in6_addr *saddr,
147 __be16 sport,
148 const struct in6_addr *daddr,
149 unsigned int hnum)
150{
151 struct sock *reuse_sk = NULL;
152 u32 hash;
153
154 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) {
155 hash = udp6_ehashfn(net, daddr, hnum, saddr, sport);
156 reuse_sk = reuseport_select_sock(sk, hash, skb,
157 sizeof(struct udphdr));
158 }
159 return reuse_sk;
160}
161
162/* called with rcu_read_lock() */
163static struct sock *udp6_lib_lookup2(struct net *net,
164 const struct in6_addr *saddr, __be16 sport,
165 const struct in6_addr *daddr, unsigned int hnum,
166 int dif, int sdif, struct udp_hslot *hslot2,
167 struct sk_buff *skb)
168{
169 struct sock *sk, *result;
170 int score, badness;
171
172 result = NULL;
173 badness = -1;
174 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
175 score = compute_score(sk, net, saddr, sport,
176 daddr, hnum, dif, sdif);
177 if (score > badness) {
178 result = lookup_reuseport(net, sk, skb,
179 saddr, sport, daddr, hnum);
180 /* Fall back to scoring if group has connections */
181 if (result && !reuseport_has_conns(sk, false))
182 return result;
183
184 result = result ? : sk;
185 badness = score;
186 }
187 }
188 return result;
189}
190
191static inline struct sock *udp6_lookup_run_bpf(struct net *net,
192 struct udp_table *udptable,
193 struct sk_buff *skb,
194 const struct in6_addr *saddr,
195 __be16 sport,
196 const struct in6_addr *daddr,
197 u16 hnum)
198{
199 struct sock *sk, *reuse_sk;
200 bool no_reuseport;
201
202 if (udptable != &udp_table)
203 return NULL; /* only UDP is supported */
204
205 no_reuseport = bpf_sk_lookup_run_v6(net, IPPROTO_UDP,
206 saddr, sport, daddr, hnum, &sk);
207 if (no_reuseport || IS_ERR_OR_NULL(sk))
208 return sk;
209
210 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum);
211 if (reuse_sk)
212 sk = reuse_sk;
213 return sk;
214}
215
216/* rcu_read_lock() must be held */
217struct sock *__udp6_lib_lookup(struct net *net,
218 const struct in6_addr *saddr, __be16 sport,
219 const struct in6_addr *daddr, __be16 dport,
220 int dif, int sdif, struct udp_table *udptable,
221 struct sk_buff *skb)
222{
223 unsigned short hnum = ntohs(dport);
224 unsigned int hash2, slot2;
225 struct udp_hslot *hslot2;
226 struct sock *result, *sk;
227
228 hash2 = ipv6_portaddr_hash(net, daddr, hnum);
229 slot2 = hash2 & udptable->mask;
230 hslot2 = &udptable->hash2[slot2];
231
232 /* Lookup connected or non-wildcard sockets */
233 result = udp6_lib_lookup2(net, saddr, sport,
234 daddr, hnum, dif, sdif,
235 hslot2, skb);
236 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
237 goto done;
238
239 /* Lookup redirect from BPF */
240 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
241 sk = udp6_lookup_run_bpf(net, udptable, skb,
242 saddr, sport, daddr, hnum);
243 if (sk) {
244 result = sk;
245 goto done;
246 }
247 }
248
249 /* Got non-wildcard socket or error on first lookup */
250 if (result)
251 goto done;
252
253 /* Lookup wildcard sockets */
254 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
255 slot2 = hash2 & udptable->mask;
256 hslot2 = &udptable->hash2[slot2];
257
258 result = udp6_lib_lookup2(net, saddr, sport,
259 &in6addr_any, hnum, dif, sdif,
260 hslot2, skb);
261done:
262 if (IS_ERR(result))
263 return NULL;
264 return result;
265}
266EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
267
268static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
269 __be16 sport, __be16 dport,
270 struct udp_table *udptable)
271{
272 const struct ipv6hdr *iph = ipv6_hdr(skb);
273
274 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
275 &iph->daddr, dport, inet6_iif(skb),
276 inet6_sdif(skb), udptable, skb);
277}
278
279struct sock *udp6_lib_lookup_skb(struct sk_buff *skb,
280 __be16 sport, __be16 dport)
281{
282 const struct ipv6hdr *iph = ipv6_hdr(skb);
283
284 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
285 &iph->daddr, dport, inet6_iif(skb),
286 inet6_sdif(skb), &udp_table, NULL);
287}
288EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb);
289
290/* Must be called under rcu_read_lock().
291 * Does increment socket refcount.
292 */
293#if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
294struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
295 const struct in6_addr *daddr, __be16 dport, int dif)
296{
297 struct sock *sk;
298
299 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport,
300 dif, 0, &udp_table, NULL);
301 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
302 sk = NULL;
303 return sk;
304}
305EXPORT_SYMBOL_GPL(udp6_lib_lookup);
306#endif
307
308/* do not use the scratch area len for jumbogram: their length execeeds the
309 * scratch area space; note that the IP6CB flags is still in the first
310 * cacheline, so checking for jumbograms is cheap
311 */
312static int udp6_skb_len(struct sk_buff *skb)
313{
314 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb);
315}
316
317/*
318 * This should be easy, if there is something there we
319 * return it, otherwise we block.
320 */
321
322int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
323 int noblock, int flags, int *addr_len)
324{
325 struct ipv6_pinfo *np = inet6_sk(sk);
326 struct inet_sock *inet = inet_sk(sk);
327 struct sk_buff *skb;
328 unsigned int ulen, copied;
329 int off, err, peeking = flags & MSG_PEEK;
330 int is_udplite = IS_UDPLITE(sk);
331 struct udp_mib __percpu *mib;
332 bool checksum_valid = false;
333 int is_udp4;
334
335 if (flags & MSG_ERRQUEUE)
336 return ipv6_recv_error(sk, msg, len, addr_len);
337
338 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
339 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
340
341try_again:
342 off = sk_peek_offset(sk, flags);
343 skb = __skb_recv_udp(sk, flags, noblock, &off, &err);
344 if (!skb)
345 return err;
346
347 ulen = udp6_skb_len(skb);
348 copied = len;
349 if (copied > ulen - off)
350 copied = ulen - off;
351 else if (copied < ulen)
352 msg->msg_flags |= MSG_TRUNC;
353
354 is_udp4 = (skb->protocol == htons(ETH_P_IP));
355 mib = __UDPX_MIB(sk, is_udp4);
356
357 /*
358 * If checksum is needed at all, try to do it while copying the
359 * data. If the data is truncated, or if we only want a partial
360 * coverage checksum (UDP-Lite), do it before the copy.
361 */
362
363 if (copied < ulen || peeking ||
364 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
365 checksum_valid = udp_skb_csum_unnecessary(skb) ||
366 !__udp_lib_checksum_complete(skb);
367 if (!checksum_valid)
368 goto csum_copy_err;
369 }
370
371 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
372 if (udp_skb_is_linear(skb))
373 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
374 else
375 err = skb_copy_datagram_msg(skb, off, msg, copied);
376 } else {
377 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
378 if (err == -EINVAL)
379 goto csum_copy_err;
380 }
381 if (unlikely(err)) {
382 if (!peeking) {
383 atomic_inc(&sk->sk_drops);
384 SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
385 }
386 kfree_skb(skb);
387 return err;
388 }
389 if (!peeking)
390 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS);
391
392 sock_recv_ts_and_drops(msg, sk, skb);
393
394 /* Copy the address. */
395 if (msg->msg_name) {
396 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
397 sin6->sin6_family = AF_INET6;
398 sin6->sin6_port = udp_hdr(skb)->source;
399 sin6->sin6_flowinfo = 0;
400
401 if (is_udp4) {
402 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
403 &sin6->sin6_addr);
404 sin6->sin6_scope_id = 0;
405 } else {
406 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
407 sin6->sin6_scope_id =
408 ipv6_iface_scope_id(&sin6->sin6_addr,
409 inet6_iif(skb));
410 }
411 *addr_len = sizeof(*sin6);
412
413 if (cgroup_bpf_enabled)
414 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk,
415 (struct sockaddr *)sin6);
416 }
417
418 if (udp_sk(sk)->gro_enabled)
419 udp_cmsg_recv(msg, sk, skb);
420
421 if (np->rxopt.all)
422 ip6_datagram_recv_common_ctl(sk, msg, skb);
423
424 if (is_udp4) {
425 if (inet->cmsg_flags)
426 ip_cmsg_recv_offset(msg, sk, skb,
427 sizeof(struct udphdr), off);
428 } else {
429 if (np->rxopt.all)
430 ip6_datagram_recv_specific_ctl(sk, msg, skb);
431 }
432
433 err = copied;
434 if (flags & MSG_TRUNC)
435 err = ulen;
436
437 skb_consume_udp(sk, skb, peeking ? -err : err);
438 return err;
439
440csum_copy_err:
441 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
442 udp_skb_destructor)) {
443 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS);
444 SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
445 }
446 kfree_skb(skb);
447
448 /* starting over for a new packet, but check if we need to yield */
449 cond_resched();
450 msg->msg_flags &= ~MSG_TRUNC;
451 goto try_again;
452}
453
454DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
455void udpv6_encap_enable(void)
456{
457 static_branch_inc(&udpv6_encap_needed_key);
458}
459EXPORT_SYMBOL(udpv6_encap_enable);
460
461/* Handler for tunnels with arbitrary destination ports: no socket lookup, go
462 * through error handlers in encapsulations looking for a match.
463 */
464static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb,
465 struct inet6_skb_parm *opt,
466 u8 type, u8 code, int offset, __be32 info)
467{
468 int i;
469
470 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
471 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt,
472 u8 type, u8 code, int offset, __be32 info);
473 const struct ip6_tnl_encap_ops *encap;
474
475 encap = rcu_dereference(ip6tun_encaps[i]);
476 if (!encap)
477 continue;
478 handler = encap->err_handler;
479 if (handler && !handler(skb, opt, type, code, offset, info))
480 return 0;
481 }
482
483 return -ENOENT;
484}
485
486/* Try to match ICMP errors to UDP tunnels by looking up a socket without
487 * reversing source and destination port: this will match tunnels that force the
488 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
489 * lwtunnels might actually break this assumption by being configured with
490 * different destination ports on endpoints, in this case we won't be able to
491 * trace ICMP messages back to them.
492 *
493 * If this doesn't match any socket, probe tunnels with arbitrary destination
494 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
495 * we've sent packets to won't necessarily match the local destination port.
496 *
497 * Then ask the tunnel implementation to match the error against a valid
498 * association.
499 *
500 * Return an error if we can't find a match, the socket if we need further
501 * processing, zero otherwise.
502 */
503static struct sock *__udp6_lib_err_encap(struct net *net,
504 const struct ipv6hdr *hdr, int offset,
505 struct udphdr *uh,
506 struct udp_table *udptable,
507 struct sk_buff *skb,
508 struct inet6_skb_parm *opt,
509 u8 type, u8 code, __be32 info)
510{
511 int network_offset, transport_offset;
512 struct sock *sk;
513
514 network_offset = skb_network_offset(skb);
515 transport_offset = skb_transport_offset(skb);
516
517 /* Network header needs to point to the outer IPv6 header inside ICMP */
518 skb_reset_network_header(skb);
519
520 /* Transport header needs to point to the UDP header */
521 skb_set_transport_header(skb, offset);
522
523 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source,
524 &hdr->saddr, uh->dest,
525 inet6_iif(skb), 0, udptable, skb);
526 if (sk) {
527 int (*lookup)(struct sock *sk, struct sk_buff *skb);
528 struct udp_sock *up = udp_sk(sk);
529
530 lookup = READ_ONCE(up->encap_err_lookup);
531 if (!lookup || lookup(sk, skb))
532 sk = NULL;
533 }
534
535 if (!sk) {
536 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code,
537 offset, info));
538 }
539
540 skb_set_transport_header(skb, transport_offset);
541 skb_set_network_header(skb, network_offset);
542
543 return sk;
544}
545
546int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
547 u8 type, u8 code, int offset, __be32 info,
548 struct udp_table *udptable)
549{
550 struct ipv6_pinfo *np;
551 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
552 const struct in6_addr *saddr = &hdr->saddr;
553 const struct in6_addr *daddr = &hdr->daddr;
554 struct udphdr *uh = (struct udphdr *)(skb->data+offset);
555 bool tunnel = false;
556 struct sock *sk;
557 int harderr;
558 int err;
559 struct net *net = dev_net(skb->dev);
560
561 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
562 inet6_iif(skb), inet6_sdif(skb), udptable, NULL);
563 if (!sk) {
564 /* No socket for error: try tunnels before discarding */
565 sk = ERR_PTR(-ENOENT);
566 if (static_branch_unlikely(&udpv6_encap_needed_key)) {
567 sk = __udp6_lib_err_encap(net, hdr, offset, uh,
568 udptable, skb,
569 opt, type, code, info);
570 if (!sk)
571 return 0;
572 }
573
574 if (IS_ERR(sk)) {
575 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
576 ICMP6_MIB_INERRORS);
577 return PTR_ERR(sk);
578 }
579
580 tunnel = true;
581 }
582
583 harderr = icmpv6_err_convert(type, code, &err);
584 np = inet6_sk(sk);
585
586 if (type == ICMPV6_PKT_TOOBIG) {
587 if (!ip6_sk_accept_pmtu(sk))
588 goto out;
589 ip6_sk_update_pmtu(skb, sk, info);
590 if (np->pmtudisc != IPV6_PMTUDISC_DONT)
591 harderr = 1;
592 }
593 if (type == NDISC_REDIRECT) {
594 if (tunnel) {
595 ip6_redirect(skb, sock_net(sk), inet6_iif(skb),
596 sk->sk_mark, sk->sk_uid);
597 } else {
598 ip6_sk_redirect(skb, sk);
599 }
600 goto out;
601 }
602
603 /* Tunnels don't have an application socket: don't pass errors back */
604 if (tunnel)
605 goto out;
606
607 if (!np->recverr) {
608 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
609 goto out;
610 } else {
611 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
612 }
613
614 sk->sk_err = err;
615 sk->sk_error_report(sk);
616out:
617 return 0;
618}
619
620static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
621{
622 int rc;
623
624 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
625 sock_rps_save_rxhash(sk, skb);
626 sk_mark_napi_id(sk, skb);
627 sk_incoming_cpu_update(sk);
628 } else {
629 sk_mark_napi_id_once(sk, skb);
630 }
631
632 rc = __udp_enqueue_schedule_skb(sk, skb);
633 if (rc < 0) {
634 int is_udplite = IS_UDPLITE(sk);
635
636 /* Note that an ENOMEM error is charged twice */
637 if (rc == -ENOMEM)
638 UDP6_INC_STATS(sock_net(sk),
639 UDP_MIB_RCVBUFERRORS, is_udplite);
640 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
641 kfree_skb(skb);
642 return -1;
643 }
644
645 return 0;
646}
647
648static __inline__ int udpv6_err(struct sk_buff *skb,
649 struct inet6_skb_parm *opt, u8 type,
650 u8 code, int offset, __be32 info)
651{
652 return __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
653}
654
655static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
656{
657 struct udp_sock *up = udp_sk(sk);
658 int is_udplite = IS_UDPLITE(sk);
659
660 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
661 goto drop;
662
663 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) {
664 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
665
666 /*
667 * This is an encapsulation socket so pass the skb to
668 * the socket's udp_encap_rcv() hook. Otherwise, just
669 * fall through and pass this up the UDP socket.
670 * up->encap_rcv() returns the following value:
671 * =0 if skb was successfully passed to the encap
672 * handler or was discarded by it.
673 * >0 if skb should be passed on to UDP.
674 * <0 if skb should be resubmitted as proto -N
675 */
676
677 /* if we're overly short, let UDP handle it */
678 encap_rcv = READ_ONCE(up->encap_rcv);
679 if (encap_rcv) {
680 int ret;
681
682 /* Verify checksum before giving to encap */
683 if (udp_lib_checksum_complete(skb))
684 goto csum_error;
685
686 ret = encap_rcv(sk, skb);
687 if (ret <= 0) {
688 __UDP_INC_STATS(sock_net(sk),
689 UDP_MIB_INDATAGRAMS,
690 is_udplite);
691 return -ret;
692 }
693 }
694
695 /* FALLTHROUGH -- it's a UDP Packet */
696 }
697
698 /*
699 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
700 */
701 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
702
703 if (up->pcrlen == 0) { /* full coverage was set */
704 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
705 UDP_SKB_CB(skb)->cscov, skb->len);
706 goto drop;
707 }
708 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
709 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
710 UDP_SKB_CB(skb)->cscov, up->pcrlen);
711 goto drop;
712 }
713 }
714
715 prefetch(&sk->sk_rmem_alloc);
716 if (rcu_access_pointer(sk->sk_filter) &&
717 udp_lib_checksum_complete(skb))
718 goto csum_error;
719
720 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
721 goto drop;
722
723 udp_csum_pull_header(skb);
724
725 skb_dst_drop(skb);
726
727 return __udpv6_queue_rcv_skb(sk, skb);
728
729csum_error:
730 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
731drop:
732 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
733 atomic_inc(&sk->sk_drops);
734 kfree_skb(skb);
735 return -1;
736}
737
738static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
739{
740 struct sk_buff *next, *segs;
741 int ret;
742
743 if (likely(!udp_unexpected_gso(sk, skb)))
744 return udpv6_queue_rcv_one_skb(sk, skb);
745
746 __skb_push(skb, -skb_mac_offset(skb));
747 segs = udp_rcv_segment(sk, skb, false);
748 skb_list_walk_safe(segs, skb, next) {
749 __skb_pull(skb, skb_transport_offset(skb));
750
751 ret = udpv6_queue_rcv_one_skb(sk, skb);
752 if (ret > 0)
753 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret,
754 true);
755 }
756 return 0;
757}
758
759static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
760 __be16 loc_port, const struct in6_addr *loc_addr,
761 __be16 rmt_port, const struct in6_addr *rmt_addr,
762 int dif, int sdif, unsigned short hnum)
763{
764 struct inet_sock *inet = inet_sk(sk);
765
766 if (!net_eq(sock_net(sk), net))
767 return false;
768
769 if (udp_sk(sk)->udp_port_hash != hnum ||
770 sk->sk_family != PF_INET6 ||
771 (inet->inet_dport && inet->inet_dport != rmt_port) ||
772 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
773 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
774 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif) ||
775 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
776 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
777 return false;
778 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
779 return false;
780 return true;
781}
782
783static void udp6_csum_zero_error(struct sk_buff *skb)
784{
785 /* RFC 2460 section 8.1 says that we SHOULD log
786 * this error. Well, it is reasonable.
787 */
788 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
789 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
790 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
791}
792
793/*
794 * Note: called only from the BH handler context,
795 * so we don't need to lock the hashes.
796 */
797static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
798 const struct in6_addr *saddr, const struct in6_addr *daddr,
799 struct udp_table *udptable, int proto)
800{
801 struct sock *sk, *first = NULL;
802 const struct udphdr *uh = udp_hdr(skb);
803 unsigned short hnum = ntohs(uh->dest);
804 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
805 unsigned int offset = offsetof(typeof(*sk), sk_node);
806 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
807 int dif = inet6_iif(skb);
808 int sdif = inet6_sdif(skb);
809 struct hlist_node *node;
810 struct sk_buff *nskb;
811
812 if (use_hash2) {
813 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) &
814 udptable->mask;
815 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask;
816start_lookup:
817 hslot = &udptable->hash2[hash2];
818 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
819 }
820
821 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
822 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
823 uh->source, saddr, dif, sdif,
824 hnum))
825 continue;
826 /* If zero checksum and no_check is not on for
827 * the socket then skip it.
828 */
829 if (!uh->check && !udp_sk(sk)->no_check6_rx)
830 continue;
831 if (!first) {
832 first = sk;
833 continue;
834 }
835 nskb = skb_clone(skb, GFP_ATOMIC);
836 if (unlikely(!nskb)) {
837 atomic_inc(&sk->sk_drops);
838 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
839 IS_UDPLITE(sk));
840 __UDP6_INC_STATS(net, UDP_MIB_INERRORS,
841 IS_UDPLITE(sk));
842 continue;
843 }
844
845 if (udpv6_queue_rcv_skb(sk, nskb) > 0)
846 consume_skb(nskb);
847 }
848
849 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
850 if (use_hash2 && hash2 != hash2_any) {
851 hash2 = hash2_any;
852 goto start_lookup;
853 }
854
855 if (first) {
856 if (udpv6_queue_rcv_skb(first, skb) > 0)
857 consume_skb(skb);
858 } else {
859 kfree_skb(skb);
860 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
861 proto == IPPROTO_UDPLITE);
862 }
863 return 0;
864}
865
866static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
867{
868 if (udp_sk_rx_dst_set(sk, dst)) {
869 const struct rt6_info *rt = (const struct rt6_info *)dst;
870
871 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
872 }
873}
874
875/* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
876 * return code conversion for ip layer consumption
877 */
878static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
879 struct udphdr *uh)
880{
881 int ret;
882
883 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
884 skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo);
885
886 ret = udpv6_queue_rcv_skb(sk, skb);
887
888 /* a return value > 0 means to resubmit the input */
889 if (ret > 0)
890 return ret;
891 return 0;
892}
893
894int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
895 int proto)
896{
897 const struct in6_addr *saddr, *daddr;
898 struct net *net = dev_net(skb->dev);
899 struct udphdr *uh;
900 struct sock *sk;
901 bool refcounted;
902 u32 ulen = 0;
903
904 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
905 goto discard;
906
907 saddr = &ipv6_hdr(skb)->saddr;
908 daddr = &ipv6_hdr(skb)->daddr;
909 uh = udp_hdr(skb);
910
911 ulen = ntohs(uh->len);
912 if (ulen > skb->len)
913 goto short_packet;
914
915 if (proto == IPPROTO_UDP) {
916 /* UDP validates ulen. */
917
918 /* Check for jumbo payload */
919 if (ulen == 0)
920 ulen = skb->len;
921
922 if (ulen < sizeof(*uh))
923 goto short_packet;
924
925 if (ulen < skb->len) {
926 if (pskb_trim_rcsum(skb, ulen))
927 goto short_packet;
928 saddr = &ipv6_hdr(skb)->saddr;
929 daddr = &ipv6_hdr(skb)->daddr;
930 uh = udp_hdr(skb);
931 }
932 }
933
934 if (udp6_csum_init(skb, uh, proto))
935 goto csum_error;
936
937 /* Check if the socket is already available, e.g. due to early demux */
938 sk = skb_steal_sock(skb, &refcounted);
939 if (sk) {
940 struct dst_entry *dst = skb_dst(skb);
941 int ret;
942
943 if (unlikely(sk->sk_rx_dst != dst))
944 udp6_sk_rx_dst_set(sk, dst);
945
946 if (!uh->check && !udp_sk(sk)->no_check6_rx) {
947 if (refcounted)
948 sock_put(sk);
949 goto report_csum_error;
950 }
951
952 ret = udp6_unicast_rcv_skb(sk, skb, uh);
953 if (refcounted)
954 sock_put(sk);
955 return ret;
956 }
957
958 /*
959 * Multicast receive code
960 */
961 if (ipv6_addr_is_multicast(daddr))
962 return __udp6_lib_mcast_deliver(net, skb,
963 saddr, daddr, udptable, proto);
964
965 /* Unicast */
966 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
967 if (sk) {
968 if (!uh->check && !udp_sk(sk)->no_check6_rx)
969 goto report_csum_error;
970 return udp6_unicast_rcv_skb(sk, skb, uh);
971 }
972
973 if (!uh->check)
974 goto report_csum_error;
975
976 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
977 goto discard;
978
979 if (udp_lib_checksum_complete(skb))
980 goto csum_error;
981
982 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
983 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
984
985 kfree_skb(skb);
986 return 0;
987
988short_packet:
989 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
990 proto == IPPROTO_UDPLITE ? "-Lite" : "",
991 saddr, ntohs(uh->source),
992 ulen, skb->len,
993 daddr, ntohs(uh->dest));
994 goto discard;
995
996report_csum_error:
997 udp6_csum_zero_error(skb);
998csum_error:
999 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1000discard:
1001 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1002 kfree_skb(skb);
1003 return 0;
1004}
1005
1006
1007static struct sock *__udp6_lib_demux_lookup(struct net *net,
1008 __be16 loc_port, const struct in6_addr *loc_addr,
1009 __be16 rmt_port, const struct in6_addr *rmt_addr,
1010 int dif, int sdif)
1011{
1012 unsigned short hnum = ntohs(loc_port);
1013 unsigned int hash2 = ipv6_portaddr_hash(net, loc_addr, hnum);
1014 unsigned int slot2 = hash2 & udp_table.mask;
1015 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
1016 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
1017 struct sock *sk;
1018
1019 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
1020 if (sk->sk_state == TCP_ESTABLISHED &&
1021 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif, sdif))
1022 return sk;
1023 /* Only check first socket in chain */
1024 break;
1025 }
1026 return NULL;
1027}
1028
1029INDIRECT_CALLABLE_SCOPE void udp_v6_early_demux(struct sk_buff *skb)
1030{
1031 struct net *net = dev_net(skb->dev);
1032 const struct udphdr *uh;
1033 struct sock *sk;
1034 struct dst_entry *dst;
1035 int dif = skb->dev->ifindex;
1036 int sdif = inet6_sdif(skb);
1037
1038 if (!pskb_may_pull(skb, skb_transport_offset(skb) +
1039 sizeof(struct udphdr)))
1040 return;
1041
1042 uh = udp_hdr(skb);
1043
1044 if (skb->pkt_type == PACKET_HOST)
1045 sk = __udp6_lib_demux_lookup(net, uh->dest,
1046 &ipv6_hdr(skb)->daddr,
1047 uh->source, &ipv6_hdr(skb)->saddr,
1048 dif, sdif);
1049 else
1050 return;
1051
1052 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
1053 return;
1054
1055 skb->sk = sk;
1056 skb->destructor = sock_efree;
1057 dst = READ_ONCE(sk->sk_rx_dst);
1058
1059 if (dst)
1060 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1061 if (dst) {
1062 /* set noref for now.
1063 * any place which wants to hold dst has to call
1064 * dst_hold_safe()
1065 */
1066 skb_dst_set_noref(skb, dst);
1067 }
1068}
1069
1070INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb)
1071{
1072 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1073}
1074
1075/*
1076 * Throw away all pending data and cancel the corking. Socket is locked.
1077 */
1078static void udp_v6_flush_pending_frames(struct sock *sk)
1079{
1080 struct udp_sock *up = udp_sk(sk);
1081
1082 if (up->pending == AF_INET)
1083 udp_flush_pending_frames(sk);
1084 else if (up->pending) {
1085 up->len = 0;
1086 up->pending = 0;
1087 ip6_flush_pending_frames(sk);
1088 }
1089}
1090
1091static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
1092 int addr_len)
1093{
1094 if (addr_len < offsetofend(struct sockaddr, sa_family))
1095 return -EINVAL;
1096 /* The following checks are replicated from __ip6_datagram_connect()
1097 * and intended to prevent BPF program called below from accessing
1098 * bytes that are out of the bound specified by user in addr_len.
1099 */
1100 if (uaddr->sa_family == AF_INET) {
1101 if (__ipv6_only_sock(sk))
1102 return -EAFNOSUPPORT;
1103 return udp_pre_connect(sk, uaddr, addr_len);
1104 }
1105
1106 if (addr_len < SIN6_LEN_RFC2133)
1107 return -EINVAL;
1108
1109 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr);
1110}
1111
1112/**
1113 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
1114 * @sk: socket we are sending on
1115 * @skb: sk_buff containing the filled-in UDP header
1116 * (checksum field must be zeroed out)
1117 * @saddr: source address
1118 * @daddr: destination address
1119 * @len: length of packet
1120 */
1121static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
1122 const struct in6_addr *saddr,
1123 const struct in6_addr *daddr, int len)
1124{
1125 unsigned int offset;
1126 struct udphdr *uh = udp_hdr(skb);
1127 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
1128 __wsum csum = 0;
1129
1130 if (!frags) {
1131 /* Only one fragment on the socket. */
1132 skb->csum_start = skb_transport_header(skb) - skb->head;
1133 skb->csum_offset = offsetof(struct udphdr, check);
1134 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1135 } else {
1136 /*
1137 * HW-checksum won't work as there are two or more
1138 * fragments on the socket so that all csums of sk_buffs
1139 * should be together
1140 */
1141 offset = skb_transport_offset(skb);
1142 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1143 csum = skb->csum;
1144
1145 skb->ip_summed = CHECKSUM_NONE;
1146
1147 do {
1148 csum = csum_add(csum, frags->csum);
1149 } while ((frags = frags->next));
1150
1151 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1152 csum);
1153 if (uh->check == 0)
1154 uh->check = CSUM_MANGLED_0;
1155 }
1156}
1157
1158/*
1159 * Sending
1160 */
1161
1162static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6,
1163 struct inet_cork *cork)
1164{
1165 struct sock *sk = skb->sk;
1166 struct udphdr *uh;
1167 int err = 0;
1168 int is_udplite = IS_UDPLITE(sk);
1169 __wsum csum = 0;
1170 int offset = skb_transport_offset(skb);
1171 int len = skb->len - offset;
1172 int datalen = len - sizeof(*uh);
1173
1174 /*
1175 * Create a UDP header
1176 */
1177 uh = udp_hdr(skb);
1178 uh->source = fl6->fl6_sport;
1179 uh->dest = fl6->fl6_dport;
1180 uh->len = htons(len);
1181 uh->check = 0;
1182
1183 if (cork->gso_size) {
1184 const int hlen = skb_network_header_len(skb) +
1185 sizeof(struct udphdr);
1186
1187 if (hlen + cork->gso_size > cork->fragsize) {
1188 kfree_skb(skb);
1189 return -EINVAL;
1190 }
1191 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) {
1192 kfree_skb(skb);
1193 return -EINVAL;
1194 }
1195 if (udp_sk(sk)->no_check6_tx) {
1196 kfree_skb(skb);
1197 return -EINVAL;
1198 }
1199 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
1200 dst_xfrm(skb_dst(skb))) {
1201 kfree_skb(skb);
1202 return -EIO;
1203 }
1204
1205 if (datalen > cork->gso_size) {
1206 skb_shinfo(skb)->gso_size = cork->gso_size;
1207 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
1208 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
1209 cork->gso_size);
1210 }
1211 goto csum_partial;
1212 }
1213
1214 if (is_udplite)
1215 csum = udplite_csum(skb);
1216 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */
1217 skb->ip_summed = CHECKSUM_NONE;
1218 goto send;
1219 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1220csum_partial:
1221 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1222 goto send;
1223 } else
1224 csum = udp_csum(skb);
1225
1226 /* add protocol-dependent pseudo-header */
1227 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1228 len, fl6->flowi6_proto, csum);
1229 if (uh->check == 0)
1230 uh->check = CSUM_MANGLED_0;
1231
1232send:
1233 err = ip6_send_skb(skb);
1234 if (err) {
1235 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1236 UDP6_INC_STATS(sock_net(sk),
1237 UDP_MIB_SNDBUFERRORS, is_udplite);
1238 err = 0;
1239 }
1240 } else {
1241 UDP6_INC_STATS(sock_net(sk),
1242 UDP_MIB_OUTDATAGRAMS, is_udplite);
1243 }
1244 return err;
1245}
1246
1247static int udp_v6_push_pending_frames(struct sock *sk)
1248{
1249 struct sk_buff *skb;
1250 struct udp_sock *up = udp_sk(sk);
1251 struct flowi6 fl6;
1252 int err = 0;
1253
1254 if (up->pending == AF_INET)
1255 return udp_push_pending_frames(sk);
1256
1257 /* ip6_finish_skb will release the cork, so make a copy of
1258 * fl6 here.
1259 */
1260 fl6 = inet_sk(sk)->cork.fl.u.ip6;
1261
1262 skb = ip6_finish_skb(sk);
1263 if (!skb)
1264 goto out;
1265
1266 err = udp_v6_send_skb(skb, &fl6, &inet_sk(sk)->cork.base);
1267
1268out:
1269 up->len = 0;
1270 up->pending = 0;
1271 return err;
1272}
1273
1274int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1275{
1276 struct ipv6_txoptions opt_space;
1277 struct udp_sock *up = udp_sk(sk);
1278 struct inet_sock *inet = inet_sk(sk);
1279 struct ipv6_pinfo *np = inet6_sk(sk);
1280 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1281 struct in6_addr *daddr, *final_p, final;
1282 struct ipv6_txoptions *opt = NULL;
1283 struct ipv6_txoptions *opt_to_free = NULL;
1284 struct ip6_flowlabel *flowlabel = NULL;
1285 struct flowi6 fl6;
1286 struct dst_entry *dst;
1287 struct ipcm6_cookie ipc6;
1288 int addr_len = msg->msg_namelen;
1289 bool connected = false;
1290 int ulen = len;
1291 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1292 int err;
1293 int is_udplite = IS_UDPLITE(sk);
1294 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1295
1296 ipcm6_init(&ipc6);
1297 ipc6.gso_size = up->gso_size;
1298 ipc6.sockc.tsflags = sk->sk_tsflags;
1299 ipc6.sockc.mark = sk->sk_mark;
1300
1301 /* destination address check */
1302 if (sin6) {
1303 if (addr_len < offsetof(struct sockaddr, sa_data))
1304 return -EINVAL;
1305
1306 switch (sin6->sin6_family) {
1307 case AF_INET6:
1308 if (addr_len < SIN6_LEN_RFC2133)
1309 return -EINVAL;
1310 daddr = &sin6->sin6_addr;
1311 if (ipv6_addr_any(daddr) &&
1312 ipv6_addr_v4mapped(&np->saddr))
1313 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1314 daddr);
1315 break;
1316 case AF_INET:
1317 goto do_udp_sendmsg;
1318 case AF_UNSPEC:
1319 msg->msg_name = sin6 = NULL;
1320 msg->msg_namelen = addr_len = 0;
1321 daddr = NULL;
1322 break;
1323 default:
1324 return -EINVAL;
1325 }
1326 } else if (!up->pending) {
1327 if (sk->sk_state != TCP_ESTABLISHED)
1328 return -EDESTADDRREQ;
1329 daddr = &sk->sk_v6_daddr;
1330 } else
1331 daddr = NULL;
1332
1333 if (daddr) {
1334 if (ipv6_addr_v4mapped(daddr)) {
1335 struct sockaddr_in sin;
1336 sin.sin_family = AF_INET;
1337 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1338 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1339 msg->msg_name = &sin;
1340 msg->msg_namelen = sizeof(sin);
1341do_udp_sendmsg:
1342 if (__ipv6_only_sock(sk))
1343 return -ENETUNREACH;
1344 return udp_sendmsg(sk, msg, len);
1345 }
1346 }
1347
1348 if (up->pending == AF_INET)
1349 return udp_sendmsg(sk, msg, len);
1350
1351 /* Rough check on arithmetic overflow,
1352 better check is made in ip6_append_data().
1353 */
1354 if (len > INT_MAX - sizeof(struct udphdr))
1355 return -EMSGSIZE;
1356
1357 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1358 if (up->pending) {
1359 /*
1360 * There are pending frames.
1361 * The socket lock must be held while it's corked.
1362 */
1363 lock_sock(sk);
1364 if (likely(up->pending)) {
1365 if (unlikely(up->pending != AF_INET6)) {
1366 release_sock(sk);
1367 return -EAFNOSUPPORT;
1368 }
1369 dst = NULL;
1370 goto do_append_data;
1371 }
1372 release_sock(sk);
1373 }
1374 ulen += sizeof(struct udphdr);
1375
1376 memset(&fl6, 0, sizeof(fl6));
1377
1378 if (sin6) {
1379 if (sin6->sin6_port == 0)
1380 return -EINVAL;
1381
1382 fl6.fl6_dport = sin6->sin6_port;
1383 daddr = &sin6->sin6_addr;
1384
1385 if (np->sndflow) {
1386 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1387 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1388 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1389 if (IS_ERR(flowlabel))
1390 return -EINVAL;
1391 }
1392 }
1393
1394 /*
1395 * Otherwise it will be difficult to maintain
1396 * sk->sk_dst_cache.
1397 */
1398 if (sk->sk_state == TCP_ESTABLISHED &&
1399 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1400 daddr = &sk->sk_v6_daddr;
1401
1402 if (addr_len >= sizeof(struct sockaddr_in6) &&
1403 sin6->sin6_scope_id &&
1404 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1405 fl6.flowi6_oif = sin6->sin6_scope_id;
1406 } else {
1407 if (sk->sk_state != TCP_ESTABLISHED)
1408 return -EDESTADDRREQ;
1409
1410 fl6.fl6_dport = inet->inet_dport;
1411 daddr = &sk->sk_v6_daddr;
1412 fl6.flowlabel = np->flow_label;
1413 connected = true;
1414 }
1415
1416 if (!fl6.flowi6_oif)
1417 fl6.flowi6_oif = sk->sk_bound_dev_if;
1418
1419 if (!fl6.flowi6_oif)
1420 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1421
1422 fl6.flowi6_mark = ipc6.sockc.mark;
1423 fl6.flowi6_uid = sk->sk_uid;
1424
1425 if (msg->msg_controllen) {
1426 opt = &opt_space;
1427 memset(opt, 0, sizeof(struct ipv6_txoptions));
1428 opt->tot_len = sizeof(*opt);
1429 ipc6.opt = opt;
1430
1431 err = udp_cmsg_send(sk, msg, &ipc6.gso_size);
1432 if (err > 0)
1433 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6,
1434 &ipc6);
1435 if (err < 0) {
1436 fl6_sock_release(flowlabel);
1437 return err;
1438 }
1439 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1440 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1441 if (IS_ERR(flowlabel))
1442 return -EINVAL;
1443 }
1444 if (!(opt->opt_nflen|opt->opt_flen))
1445 opt = NULL;
1446 connected = false;
1447 }
1448 if (!opt) {
1449 opt = txopt_get(np);
1450 opt_to_free = opt;
1451 }
1452 if (flowlabel)
1453 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1454 opt = ipv6_fixup_options(&opt_space, opt);
1455 ipc6.opt = opt;
1456
1457 fl6.flowi6_proto = sk->sk_protocol;
1458 fl6.daddr = *daddr;
1459 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1460 fl6.saddr = np->saddr;
1461 fl6.fl6_sport = inet->inet_sport;
1462
1463 if (cgroup_bpf_enabled && !connected) {
1464 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk,
1465 (struct sockaddr *)sin6, &fl6.saddr);
1466 if (err)
1467 goto out_no_dst;
1468 if (sin6) {
1469 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
1470 /* BPF program rewrote IPv6-only by IPv4-mapped
1471 * IPv6. It's currently unsupported.
1472 */
1473 err = -ENOTSUPP;
1474 goto out_no_dst;
1475 }
1476 if (sin6->sin6_port == 0) {
1477 /* BPF program set invalid port. Reject it. */
1478 err = -EINVAL;
1479 goto out_no_dst;
1480 }
1481 fl6.fl6_dport = sin6->sin6_port;
1482 fl6.daddr = sin6->sin6_addr;
1483 }
1484 }
1485
1486 if (ipv6_addr_any(&fl6.daddr))
1487 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1488
1489 final_p = fl6_update_dst(&fl6, opt, &final);
1490 if (final_p)
1491 connected = false;
1492
1493 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1494 fl6.flowi6_oif = np->mcast_oif;
1495 connected = false;
1496 } else if (!fl6.flowi6_oif)
1497 fl6.flowi6_oif = np->ucast_oif;
1498
1499 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1500
1501 if (ipc6.tclass < 0)
1502 ipc6.tclass = np->tclass;
1503
1504 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel);
1505
1506 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, connected);
1507 if (IS_ERR(dst)) {
1508 err = PTR_ERR(dst);
1509 dst = NULL;
1510 goto out;
1511 }
1512
1513 if (ipc6.hlimit < 0)
1514 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1515
1516 if (msg->msg_flags&MSG_CONFIRM)
1517 goto do_confirm;
1518back_from_confirm:
1519
1520 /* Lockless fast path for the non-corking case */
1521 if (!corkreq) {
1522 struct inet_cork_full cork;
1523 struct sk_buff *skb;
1524
1525 skb = ip6_make_skb(sk, getfrag, msg, ulen,
1526 sizeof(struct udphdr), &ipc6,
1527 &fl6, (struct rt6_info *)dst,
1528 msg->msg_flags, &cork);
1529 err = PTR_ERR(skb);
1530 if (!IS_ERR_OR_NULL(skb))
1531 err = udp_v6_send_skb(skb, &fl6, &cork.base);
1532 goto out;
1533 }
1534
1535 lock_sock(sk);
1536 if (unlikely(up->pending)) {
1537 /* The socket is already corked while preparing it. */
1538 /* ... which is an evident application bug. --ANK */
1539 release_sock(sk);
1540
1541 net_dbg_ratelimited("udp cork app bug 2\n");
1542 err = -EINVAL;
1543 goto out;
1544 }
1545
1546 up->pending = AF_INET6;
1547
1548do_append_data:
1549 if (ipc6.dontfrag < 0)
1550 ipc6.dontfrag = np->dontfrag;
1551 up->len += ulen;
1552 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr),
1553 &ipc6, &fl6, (struct rt6_info *)dst,
1554 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1555 if (err)
1556 udp_v6_flush_pending_frames(sk);
1557 else if (!corkreq)
1558 err = udp_v6_push_pending_frames(sk);
1559 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1560 up->pending = 0;
1561
1562 if (err > 0)
1563 err = np->recverr ? net_xmit_errno(err) : 0;
1564 release_sock(sk);
1565
1566out:
1567 dst_release(dst);
1568out_no_dst:
1569 fl6_sock_release(flowlabel);
1570 txopt_put(opt_to_free);
1571 if (!err)
1572 return len;
1573 /*
1574 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1575 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1576 * we don't have a good statistic (IpOutDiscards but it can be too many
1577 * things). We could add another new stat but at least for now that
1578 * seems like overkill.
1579 */
1580 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1581 UDP6_INC_STATS(sock_net(sk),
1582 UDP_MIB_SNDBUFERRORS, is_udplite);
1583 }
1584 return err;
1585
1586do_confirm:
1587 if (msg->msg_flags & MSG_PROBE)
1588 dst_confirm_neigh(dst, &fl6.daddr);
1589 if (!(msg->msg_flags&MSG_PROBE) || len)
1590 goto back_from_confirm;
1591 err = 0;
1592 goto out;
1593}
1594
1595void udpv6_destroy_sock(struct sock *sk)
1596{
1597 struct udp_sock *up = udp_sk(sk);
1598 lock_sock(sk);
1599 udp_v6_flush_pending_frames(sk);
1600 release_sock(sk);
1601
1602 if (static_branch_unlikely(&udpv6_encap_needed_key)) {
1603 if (up->encap_type) {
1604 void (*encap_destroy)(struct sock *sk);
1605 encap_destroy = READ_ONCE(up->encap_destroy);
1606 if (encap_destroy)
1607 encap_destroy(sk);
1608 }
1609 if (up->encap_enabled)
1610 static_branch_dec(&udpv6_encap_needed_key);
1611 }
1612
1613 inet6_destroy_sock(sk);
1614}
1615
1616/*
1617 * Socket option code for UDP
1618 */
1619int udpv6_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
1620 unsigned int optlen)
1621{
1622 if (level == SOL_UDP || level == SOL_UDPLITE)
1623 return udp_lib_setsockopt(sk, level, optname,
1624 optval, optlen,
1625 udp_v6_push_pending_frames);
1626 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1627}
1628
1629int udpv6_getsockopt(struct sock *sk, int level, int optname,
1630 char __user *optval, int __user *optlen)
1631{
1632 if (level == SOL_UDP || level == SOL_UDPLITE)
1633 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1634 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1635}
1636
1637/* thinking of making this const? Don't.
1638 * early_demux can change based on sysctl.
1639 */
1640static struct inet6_protocol udpv6_protocol = {
1641 .early_demux = udp_v6_early_demux,
1642 .early_demux_handler = udp_v6_early_demux,
1643 .handler = udpv6_rcv,
1644 .err_handler = udpv6_err,
1645 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1646};
1647
1648/* ------------------------------------------------------------------------ */
1649#ifdef CONFIG_PROC_FS
1650int udp6_seq_show(struct seq_file *seq, void *v)
1651{
1652 if (v == SEQ_START_TOKEN) {
1653 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1654 } else {
1655 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1656 struct inet_sock *inet = inet_sk(v);
1657 __u16 srcp = ntohs(inet->inet_sport);
1658 __u16 destp = ntohs(inet->inet_dport);
1659 __ip6_dgram_sock_seq_show(seq, v, srcp, destp,
1660 udp_rqueue_get(v), bucket);
1661 }
1662 return 0;
1663}
1664
1665const struct seq_operations udp6_seq_ops = {
1666 .start = udp_seq_start,
1667 .next = udp_seq_next,
1668 .stop = udp_seq_stop,
1669 .show = udp6_seq_show,
1670};
1671EXPORT_SYMBOL(udp6_seq_ops);
1672
1673static struct udp_seq_afinfo udp6_seq_afinfo = {
1674 .family = AF_INET6,
1675 .udp_table = &udp_table,
1676};
1677
1678int __net_init udp6_proc_init(struct net *net)
1679{
1680 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops,
1681 sizeof(struct udp_iter_state), &udp6_seq_afinfo))
1682 return -ENOMEM;
1683 return 0;
1684}
1685
1686void udp6_proc_exit(struct net *net)
1687{
1688 remove_proc_entry("udp6", net->proc_net);
1689}
1690#endif /* CONFIG_PROC_FS */
1691
1692/* ------------------------------------------------------------------------ */
1693
1694struct proto udpv6_prot = {
1695 .name = "UDPv6",
1696 .owner = THIS_MODULE,
1697 .close = udp_lib_close,
1698 .pre_connect = udpv6_pre_connect,
1699 .connect = ip6_datagram_connect,
1700 .disconnect = udp_disconnect,
1701 .ioctl = udp_ioctl,
1702 .init = udp_init_sock,
1703 .destroy = udpv6_destroy_sock,
1704 .setsockopt = udpv6_setsockopt,
1705 .getsockopt = udpv6_getsockopt,
1706 .sendmsg = udpv6_sendmsg,
1707 .recvmsg = udpv6_recvmsg,
1708 .release_cb = ip6_datagram_release_cb,
1709 .hash = udp_lib_hash,
1710 .unhash = udp_lib_unhash,
1711 .rehash = udp_v6_rehash,
1712 .get_port = udp_v6_get_port,
1713 .memory_allocated = &udp_memory_allocated,
1714 .sysctl_mem = sysctl_udp_mem,
1715 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
1716 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
1717 .obj_size = sizeof(struct udp6_sock),
1718 .h.udp_table = &udp_table,
1719 .diag_destroy = udp_abort,
1720};
1721
1722static struct inet_protosw udpv6_protosw = {
1723 .type = SOCK_DGRAM,
1724 .protocol = IPPROTO_UDP,
1725 .prot = &udpv6_prot,
1726 .ops = &inet6_dgram_ops,
1727 .flags = INET_PROTOSW_PERMANENT,
1728};
1729
1730int __init udpv6_init(void)
1731{
1732 int ret;
1733
1734 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1735 if (ret)
1736 goto out;
1737
1738 ret = inet6_register_protosw(&udpv6_protosw);
1739 if (ret)
1740 goto out_udpv6_protocol;
1741out:
1742 return ret;
1743
1744out_udpv6_protocol:
1745 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1746 goto out;
1747}
1748
1749void udpv6_exit(void)
1750{
1751 inet6_unregister_protosw(&udpv6_protosw);
1752 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1753}