1 /*
2 * UDP over IPv6
3 * Linux INET6 implementation
4 *
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
7 *
8 * Based on linux/ipv4/udp.c
9 *
10 * Fixes:
11 * Hideaki YOSHIFUJI : sin6_scope_id support
12 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
13 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
14 * a single port at the same time.
15 * Kazunori MIYAZAWA @USAGI: change process style to use ip6_append_data
16 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/udp6 to seq_file.
17 *
18 * This program is free software; you can redistribute it and/or
19 * modify it under the terms of the GNU General Public License
20 * as published by the Free Software Foundation; either version
21 * 2 of the License, or (at your option) any later version.
22 */
23
24 #include <linux/errno.h>
25 #include <linux/types.h>
26 #include <linux/socket.h>
27 #include <linux/sockios.h>
28 #include <linux/net.h>
29 #include <linux/in6.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_arp.h>
32 #include <linux/ipv6.h>
33 #include <linux/icmpv6.h>
34 #include <linux/init.h>
35 #include <linux/module.h>
36 #include <linux/skbuff.h>
37 #include <linux/slab.h>
38 #include <linux/uaccess.h>
39
40 #include <net/addrconf.h>
41 #include <net/ndisc.h>
42 #include <net/protocol.h>
43 #include <net/transp_v6.h>
44 #include <net/ip6_route.h>
45 #include <net/raw.h>
46 #include <net/tcp_states.h>
47 #include <net/ip6_checksum.h>
48 #include <net/xfrm.h>
49 #include <net/inet_hashtables.h>
50 #include <net/inet6_hashtables.h>
51 #include <net/busy_poll.h>
52 #include <net/sock_reuseport.h>
53
54 #include <linux/proc_fs.h>
55 #include <linux/seq_file.h>
56 #include <trace/events/skb.h>
57 #include "udp_impl.h"
58
59 static bool udp6_lib_exact_dif_match(struct net *net, struct sk_buff *skb)
60 {
61 #if defined(CONFIG_NET_L3_MASTER_DEV)
62 if (!net->ipv4.sysctl_udp_l3mdev_accept &&
63 skb && ipv6_l3mdev_skb(IP6CB(skb)->flags))
64 return true;
65 #endif
66 return false;
67 }
68
69 static u32 udp6_ehashfn(const struct net *net,
70 const struct in6_addr *laddr,
71 const u16 lport,
72 const struct in6_addr *faddr,
73 const __be16 fport)
74 {
75 static u32 udp6_ehash_secret __read_mostly;
76 static u32 udp_ipv6_hash_secret __read_mostly;
77
78 u32 lhash, fhash;
79
80 net_get_random_once(&udp6_ehash_secret,
81 sizeof(udp6_ehash_secret));
82 net_get_random_once(&udp_ipv6_hash_secret,
83 sizeof(udp_ipv6_hash_secret));
84
85 lhash = (__force u32)laddr->s6_addr32[3];
86 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
87
88 return __inet6_ehashfn(lhash, lport, fhash, fport,
89 udp_ipv6_hash_secret + net_hash_mix(net));
90 }
91
92 static u32 udp6_portaddr_hash(const struct net *net,
93 const struct in6_addr *addr6,
94 unsigned int port)
95 {
96 unsigned int hash, mix = net_hash_mix(net);
97
98 if (ipv6_addr_any(addr6))
99 hash = jhash_1word(0, mix);
100 else if (ipv6_addr_v4mapped(addr6))
101 hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix);
102 else
103 hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix);
104
105 return hash ^ port;
106 }
107
108 int udp_v6_get_port(struct sock *sk, unsigned short snum)
109 {
110 unsigned int hash2_nulladdr =
111 udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
112 unsigned int hash2_partial =
113 udp6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
114
115 /* precompute partial secondary hash */
116 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
117 return udp_lib_get_port(sk, snum, hash2_nulladdr);
118 }
119
120 static void udp_v6_rehash(struct sock *sk)
121 {
122 u16 new_hash = udp6_portaddr_hash(sock_net(sk),
123 &sk->sk_v6_rcv_saddr,
124 inet_sk(sk)->inet_num);
125
126 udp_lib_rehash(sk, new_hash);
127 }
128
129 static int compute_score(struct sock *sk, struct net *net,
130 const struct in6_addr *saddr, __be16 sport,
131 const struct in6_addr *daddr, unsigned short hnum,
132 int dif, int sdif, bool exact_dif)
133 {
134 int score;
135 struct inet_sock *inet;
136
137 if (!net_eq(sock_net(sk), net) ||
138 udp_sk(sk)->udp_port_hash != hnum ||
139 sk->sk_family != PF_INET6)
140 return -1;
141
142 score = 0;
143 inet = inet_sk(sk);
144
145 if (inet->inet_dport) {
146 if (inet->inet_dport != sport)
147 return -1;
148 score++;
149 }
150
151 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
152 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
153 return -1;
154 score++;
155 }
156
157 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
158 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
159 return -1;
160 score++;
161 }
162
163 if (sk->sk_bound_dev_if || exact_dif) {
164 bool dev_match = (sk->sk_bound_dev_if == dif ||
165 sk->sk_bound_dev_if == sdif);
166
167 if (exact_dif && !dev_match)
168 return -1;
169 if (sk->sk_bound_dev_if && dev_match)
170 score++;
171 }
172
173 if (sk->sk_incoming_cpu == raw_smp_processor_id())
174 score++;
175
176 return score;
177 }
178
179 /* called with rcu_read_lock() */
180 static struct sock *udp6_lib_lookup2(struct net *net,
181 const struct in6_addr *saddr, __be16 sport,
182 const struct in6_addr *daddr, unsigned int hnum,
183 int dif, int sdif, bool exact_dif,
184 struct udp_hslot *hslot2, struct sk_buff *skb)
185 {
186 struct sock *sk, *result;
187 int score, badness, matches = 0, reuseport = 0;
188 u32 hash = 0;
189
190 result = NULL;
191 badness = -1;
192 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
193 score = compute_score(sk, net, saddr, sport,
194 daddr, hnum, dif, sdif, exact_dif);
195 if (score > badness) {
196 reuseport = sk->sk_reuseport;
197 if (reuseport) {
198 hash = udp6_ehashfn(net, daddr, hnum,
199 saddr, sport);
200
201 result = reuseport_select_sock(sk, hash, skb,
202 sizeof(struct udphdr));
203 if (result)
204 return result;
205 matches = 1;
206 }
207 result = sk;
208 badness = score;
209 } else if (score == badness && reuseport) {
210 matches++;
211 if (reciprocal_scale(hash, matches) == 0)
212 result = sk;
213 hash = next_pseudo_random32(hash);
214 }
215 }
216 return result;
217 }
218
219 /* rcu_read_lock() must be held */
220 struct sock *__udp6_lib_lookup(struct net *net,
221 const struct in6_addr *saddr, __be16 sport,
222 const struct in6_addr *daddr, __be16 dport,
223 int dif, int sdif, struct udp_table *udptable,
224 struct sk_buff *skb)
225 {
226 struct sock *sk, *result;
227 unsigned short hnum = ntohs(dport);
228 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
229 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
230 bool exact_dif = udp6_lib_exact_dif_match(net, skb);
231 int score, badness, matches = 0, reuseport = 0;
232 u32 hash = 0;
233
234 if (hslot->count > 10) {
235 hash2 = udp6_portaddr_hash(net, daddr, hnum);
236 slot2 = hash2 & udptable->mask;
237 hslot2 = &udptable->hash2[slot2];
238 if (hslot->count < hslot2->count)
239 goto begin;
240
241 result = udp6_lib_lookup2(net, saddr, sport,
242 daddr, hnum, dif, sdif, exact_dif,
243 hslot2, skb);
244 if (!result) {
245 unsigned int old_slot2 = slot2;
246 hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
247 slot2 = hash2 & udptable->mask;
248 /* avoid searching the same slot again. */
249 if (unlikely(slot2 == old_slot2))
250 return result;
251
252 hslot2 = &udptable->hash2[slot2];
253 if (hslot->count < hslot2->count)
254 goto begin;
255
256 result = udp6_lib_lookup2(net, saddr, sport,
257 daddr, hnum, dif, sdif,
258 exact_dif, hslot2,
259 skb);
260 }
261 return result;
262 }
263 begin:
264 result = NULL;
265 badness = -1;
266 sk_for_each_rcu(sk, &hslot->head) {
267 score = compute_score(sk, net, saddr, sport, daddr, hnum, dif,
268 sdif, exact_dif);
269 if (score > badness) {
270 reuseport = sk->sk_reuseport;
271 if (reuseport) {
272 hash = udp6_ehashfn(net, daddr, hnum,
273 saddr, sport);
274 result = reuseport_select_sock(sk, hash, skb,
275 sizeof(struct udphdr));
276 if (result)
277 return result;
278 matches = 1;
279 }
280 result = sk;
281 badness = score;
282 } else if (score == badness && reuseport) {
283 matches++;
284 if (reciprocal_scale(hash, matches) == 0)
285 result = sk;
286 hash = next_pseudo_random32(hash);
287 }
288 }
289 return result;
290 }
291 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
292
293 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
294 __be16 sport, __be16 dport,
295 struct udp_table *udptable)
296 {
297 const struct ipv6hdr *iph = ipv6_hdr(skb);
298
299 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
300 &iph->daddr, dport, inet6_iif(skb),
301 inet6_sdif(skb), udptable, skb);
302 }
303
304 struct sock *udp6_lib_lookup_skb(struct sk_buff *skb,
305 __be16 sport, __be16 dport)
306 {
307 const struct ipv6hdr *iph = ipv6_hdr(skb);
308
309 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
310 &iph->daddr, dport, inet6_iif(skb),
311 inet6_sdif(skb), &udp_table, skb);
312 }
313 EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb);
314
315 /* Must be called under rcu_read_lock().
316 * Does increment socket refcount.
317 */
318 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
319 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \
320 IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
321 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
322 const struct in6_addr *daddr, __be16 dport, int dif)
323 {
324 struct sock *sk;
325
326 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport,
327 dif, 0, &udp_table, NULL);
328 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
329 sk = NULL;
330 return sk;
331 }
332 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
333 #endif
334
335 /* do not use the scratch area len for jumbogram: their length execeeds the
336 * scratch area space; note that the IP6CB flags is still in the first
337 * cacheline, so checking for jumbograms is cheap
338 */
339 static int udp6_skb_len(struct sk_buff *skb)
340 {
341 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb);
342 }
343
344 /*
345 * This should be easy, if there is something there we
346 * return it, otherwise we block.
347 */
348
349 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
350 int noblock, int flags, int *addr_len)
351 {
352 struct ipv6_pinfo *np = inet6_sk(sk);
353 struct inet_sock *inet = inet_sk(sk);
354 struct sk_buff *skb;
355 unsigned int ulen, copied;
356 int peeked, peeking, off;
357 int err;
358 int is_udplite = IS_UDPLITE(sk);
359 bool checksum_valid = false;
360 int is_udp4;
361
362 if (flags & MSG_ERRQUEUE)
363 return ipv6_recv_error(sk, msg, len, addr_len);
364
365 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
366 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
367
368 try_again:
369 peeking = flags & MSG_PEEK;
370 off = sk_peek_offset(sk, flags);
371 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
372 if (!skb)
373 return err;
374 if (ccs_socket_post_recvmsg_permission(sk, skb, flags))
375 return -EAGAIN; /* Hope less harmful than -EPERM. */
376
377 ulen = udp6_skb_len(skb);
378 copied = len;
379 if (copied > ulen - off)
380 copied = ulen - off;
381 else if (copied < ulen)
382 msg->msg_flags |= MSG_TRUNC;
383
384 is_udp4 = (skb->protocol == htons(ETH_P_IP));
385
386 /*
387 * If checksum is needed at all, try to do it while copying the
388 * data. If the data is truncated, or if we only want a partial
389 * coverage checksum (UDP-Lite), do it before the copy.
390 */
391
392 if (copied < ulen || peeking ||
393 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
394 checksum_valid = udp_skb_csum_unnecessary(skb) ||
395 !__udp_lib_checksum_complete(skb);
396 if (!checksum_valid)
397 goto csum_copy_err;
398 }
399
400 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
401 if (udp_skb_is_linear(skb))
402 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
403 else
404 err = skb_copy_datagram_msg(skb, off, msg, copied);
405 } else {
406 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
407 if (err == -EINVAL)
408 goto csum_copy_err;
409 }
410 if (unlikely(err)) {
411 if (!peeked) {
412 atomic_inc(&sk->sk_drops);
413 if (is_udp4)
414 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
415 is_udplite);
416 else
417 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
418 is_udplite);
419 }
420 kfree_skb(skb);
421 return err;
422 }
423 if (!peeked) {
424 if (is_udp4)
425 UDP_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS,
426 is_udplite);
427 else
428 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS,
429 is_udplite);
430 }
431
432 sock_recv_ts_and_drops(msg, sk, skb);
433
434 /* Copy the address. */
435 if (msg->msg_name) {
436 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
437 sin6->sin6_family = AF_INET6;
438 sin6->sin6_port = udp_hdr(skb)->source;
439 sin6->sin6_flowinfo = 0;
440
441 if (is_udp4) {
442 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
443 &sin6->sin6_addr);
444 sin6->sin6_scope_id = 0;
445 } else {
446 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
447 sin6->sin6_scope_id =
448 ipv6_iface_scope_id(&sin6->sin6_addr,
449 inet6_iif(skb));
450 }
451 *addr_len = sizeof(*sin6);
452 }
453
454 if (np->rxopt.all)
455 ip6_datagram_recv_common_ctl(sk, msg, skb);
456
457 if (is_udp4) {
458 if (inet->cmsg_flags)
459 ip_cmsg_recv_offset(msg, sk, skb,
460 sizeof(struct udphdr), off);
461 } else {
462 if (np->rxopt.all)
463 ip6_datagram_recv_specific_ctl(sk, msg, skb);
464 }
465
466 err = copied;
467 if (flags & MSG_TRUNC)
468 err = ulen;
469
470 skb_consume_udp(sk, skb, peeking ? -err : err);
471 return err;
472
473 csum_copy_err:
474 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
475 udp_skb_destructor)) {
476 if (is_udp4) {
477 UDP_INC_STATS(sock_net(sk),
478 UDP_MIB_CSUMERRORS, is_udplite);
479 UDP_INC_STATS(sock_net(sk),
480 UDP_MIB_INERRORS, is_udplite);
481 } else {
482 UDP6_INC_STATS(sock_net(sk),
483 UDP_MIB_CSUMERRORS, is_udplite);
484 UDP6_INC_STATS(sock_net(sk),
485 UDP_MIB_INERRORS, is_udplite);
486 }
487 }
488 kfree_skb(skb);
489
490 /* starting over for a new packet, but check if we need to yield */
491 cond_resched();
492 msg->msg_flags &= ~MSG_TRUNC;
493 goto try_again;
494 }
495
496 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
497 u8 type, u8 code, int offset, __be32 info,
498 struct udp_table *udptable)
499 {
500 struct ipv6_pinfo *np;
501 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
502 const struct in6_addr *saddr = &hdr->saddr;
503 const struct in6_addr *daddr = &hdr->daddr;
504 struct udphdr *uh = (struct udphdr *)(skb->data+offset);
505 struct sock *sk;
506 int harderr;
507 int err;
508 struct net *net = dev_net(skb->dev);
509
510 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
511 inet6_iif(skb), 0, udptable, skb);
512 if (!sk) {
513 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
514 ICMP6_MIB_INERRORS);
515 return;
516 }
517
518 harderr = icmpv6_err_convert(type, code, &err);
519 np = inet6_sk(sk);
520
521 if (type == ICMPV6_PKT_TOOBIG) {
522 if (!ip6_sk_accept_pmtu(sk))
523 goto out;
524 ip6_sk_update_pmtu(skb, sk, info);
525 if (np->pmtudisc != IPV6_PMTUDISC_DONT)
526 harderr = 1;
527 }
528 if (type == NDISC_REDIRECT) {
529 ip6_sk_redirect(skb, sk);
530 goto out;
531 }
532
533 if (!np->recverr) {
534 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
535 goto out;
536 } else {
537 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
538 }
539
540 sk->sk_err = err;
541 sk->sk_error_report(sk);
542 out:
543 return;
544 }
545
546 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
547 {
548 int rc;
549
550 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
551 sock_rps_save_rxhash(sk, skb);
552 sk_mark_napi_id(sk, skb);
553 sk_incoming_cpu_update(sk);
554 } else {
555 sk_mark_napi_id_once(sk, skb);
556 }
557
558 rc = __udp_enqueue_schedule_skb(sk, skb);
559 if (rc < 0) {
560 int is_udplite = IS_UDPLITE(sk);
561
562 /* Note that an ENOMEM error is charged twice */
563 if (rc == -ENOMEM)
564 UDP6_INC_STATS(sock_net(sk),
565 UDP_MIB_RCVBUFERRORS, is_udplite);
566 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
567 kfree_skb(skb);
568 return -1;
569 }
570
571 return 0;
572 }
573
574 static __inline__ void udpv6_err(struct sk_buff *skb,
575 struct inet6_skb_parm *opt, u8 type,
576 u8 code, int offset, __be32 info)
577 {
578 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
579 }
580
581 static struct static_key udpv6_encap_needed __read_mostly;
582 void udpv6_encap_enable(void)
583 {
584 static_key_enable(&udpv6_encap_needed);
585 }
586 EXPORT_SYMBOL(udpv6_encap_enable);
587
588 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
589 {
590 struct udp_sock *up = udp_sk(sk);
591 int is_udplite = IS_UDPLITE(sk);
592
593 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
594 goto drop;
595
596 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
597 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
598
599 /*
600 * This is an encapsulation socket so pass the skb to
601 * the socket's udp_encap_rcv() hook. Otherwise, just
602 * fall through and pass this up the UDP socket.
603 * up->encap_rcv() returns the following value:
604 * =0 if skb was successfully passed to the encap
605 * handler or was discarded by it.
606 * >0 if skb should be passed on to UDP.
607 * <0 if skb should be resubmitted as proto -N
608 */
609
610 /* if we're overly short, let UDP handle it */
611 encap_rcv = READ_ONCE(up->encap_rcv);
612 if (encap_rcv) {
613 int ret;
614
615 /* Verify checksum before giving to encap */
616 if (udp_lib_checksum_complete(skb))
617 goto csum_error;
618
619 ret = encap_rcv(sk, skb);
620 if (ret <= 0) {
621 __UDP_INC_STATS(sock_net(sk),
622 UDP_MIB_INDATAGRAMS,
623 is_udplite);
624 return -ret;
625 }
626 }
627
628 /* FALLTHROUGH -- it's a UDP Packet */
629 }
630
631 /*
632 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
633 */
634 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
635
636 if (up->pcrlen == 0) { /* full coverage was set */
637 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
638 UDP_SKB_CB(skb)->cscov, skb->len);
639 goto drop;
640 }
641 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
642 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
643 UDP_SKB_CB(skb)->cscov, up->pcrlen);
644 goto drop;
645 }
646 }
647
648 prefetch(&sk->sk_rmem_alloc);
649 if (rcu_access_pointer(sk->sk_filter) &&
650 udp_lib_checksum_complete(skb))
651 goto csum_error;
652
653 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
654 goto drop;
655
656 udp_csum_pull_header(skb);
657
658 skb_dst_drop(skb);
659
660 return __udpv6_queue_rcv_skb(sk, skb);
661
662 csum_error:
663 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
664 drop:
665 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
666 atomic_inc(&sk->sk_drops);
667 kfree_skb(skb);
668 return -1;
669 }
670
671 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
672 __be16 loc_port, const struct in6_addr *loc_addr,
673 __be16 rmt_port, const struct in6_addr *rmt_addr,
674 int dif, unsigned short hnum)
675 {
676 struct inet_sock *inet = inet_sk(sk);
677
678 if (!net_eq(sock_net(sk), net))
679 return false;
680
681 if (udp_sk(sk)->udp_port_hash != hnum ||
682 sk->sk_family != PF_INET6 ||
683 (inet->inet_dport && inet->inet_dport != rmt_port) ||
684 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
685 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
686 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif) ||
687 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
688 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
689 return false;
690 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
691 return false;
692 return true;
693 }
694
695 static void udp6_csum_zero_error(struct sk_buff *skb)
696 {
697 /* RFC 2460 section 8.1 says that we SHOULD log
698 * this error. Well, it is reasonable.
699 */
700 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
701 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
702 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
703 }
704
705 /*
706 * Note: called only from the BH handler context,
707 * so we don't need to lock the hashes.
708 */
709 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
710 const struct in6_addr *saddr, const struct in6_addr *daddr,
711 struct udp_table *udptable, int proto)
712 {
713 struct sock *sk, *first = NULL;
714 const struct udphdr *uh = udp_hdr(skb);
715 unsigned short hnum = ntohs(uh->dest);
716 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
717 unsigned int offset = offsetof(typeof(*sk), sk_node);
718 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
719 int dif = inet6_iif(skb);
720 struct hlist_node *node;
721 struct sk_buff *nskb;
722
723 if (use_hash2) {
724 hash2_any = udp6_portaddr_hash(net, &in6addr_any, hnum) &
725 udptable->mask;
726 hash2 = udp6_portaddr_hash(net, daddr, hnum) & udptable->mask;
727 start_lookup:
728 hslot = &udptable->hash2[hash2];
729 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
730 }
731
732 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
733 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
734 uh->source, saddr, dif, hnum))
735 continue;
736 /* If zero checksum and no_check is not on for
737 * the socket then skip it.
738 */
739 if (!uh->check && !udp_sk(sk)->no_check6_rx)
740 continue;
741 if (!first) {
742 first = sk;
743 continue;
744 }
745 nskb = skb_clone(skb, GFP_ATOMIC);
746 if (unlikely(!nskb)) {
747 atomic_inc(&sk->sk_drops);
748 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
749 IS_UDPLITE(sk));
750 __UDP6_INC_STATS(net, UDP_MIB_INERRORS,
751 IS_UDPLITE(sk));
752 continue;
753 }
754
755 if (udpv6_queue_rcv_skb(sk, nskb) > 0)
756 consume_skb(nskb);
757 }
758
759 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
760 if (use_hash2 && hash2 != hash2_any) {
761 hash2 = hash2_any;
762 goto start_lookup;
763 }
764
765 if (first) {
766 if (udpv6_queue_rcv_skb(first, skb) > 0)
767 consume_skb(skb);
768 } else {
769 kfree_skb(skb);
770 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
771 proto == IPPROTO_UDPLITE);
772 }
773 return 0;
774 }
775
776 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
777 {
778 if (udp_sk_rx_dst_set(sk, dst)) {
779 const struct rt6_info *rt = (const struct rt6_info *)dst;
780
781 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
782 }
783 }
784
785 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
786 int proto)
787 {
788 const struct in6_addr *saddr, *daddr;
789 struct net *net = dev_net(skb->dev);
790 struct udphdr *uh;
791 struct sock *sk;
792 u32 ulen = 0;
793
794 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
795 goto discard;
796
797 saddr = &ipv6_hdr(skb)->saddr;
798 daddr = &ipv6_hdr(skb)->daddr;
799 uh = udp_hdr(skb);
800
801 ulen = ntohs(uh->len);
802 if (ulen > skb->len)
803 goto short_packet;
804
805 if (proto == IPPROTO_UDP) {
806 /* UDP validates ulen. */
807
808 /* Check for jumbo payload */
809 if (ulen == 0)
810 ulen = skb->len;
811
812 if (ulen < sizeof(*uh))
813 goto short_packet;
814
815 if (ulen < skb->len) {
816 if (pskb_trim_rcsum(skb, ulen))
817 goto short_packet;
818 saddr = &ipv6_hdr(skb)->saddr;
819 daddr = &ipv6_hdr(skb)->daddr;
820 uh = udp_hdr(skb);
821 }
822 }
823
824 if (udp6_csum_init(skb, uh, proto))
825 goto csum_error;
826
827 /* Check if the socket is already available, e.g. due to early demux */
828 sk = skb_steal_sock(skb);
829 if (sk) {
830 struct dst_entry *dst = skb_dst(skb);
831 int ret;
832
833 if (unlikely(sk->sk_rx_dst != dst))
834 udp6_sk_rx_dst_set(sk, dst);
835
836 ret = udpv6_queue_rcv_skb(sk, skb);
837 sock_put(sk);
838
839 /* a return value > 0 means to resubmit the input */
840 if (ret > 0)
841 return ret;
842 return 0;
843 }
844
845 /*
846 * Multicast receive code
847 */
848 if (ipv6_addr_is_multicast(daddr))
849 return __udp6_lib_mcast_deliver(net, skb,
850 saddr, daddr, udptable, proto);
851
852 /* Unicast */
853 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
854 if (sk) {
855 int ret;
856
857 if (!uh->check && !udp_sk(sk)->no_check6_rx) {
858 udp6_csum_zero_error(skb);
859 goto csum_error;
860 }
861
862 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
863 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
864 ip6_compute_pseudo);
865
866 ret = udpv6_queue_rcv_skb(sk, skb);
867
868 /* a return value > 0 means to resubmit the input */
869 if (ret > 0)
870 return ret;
871
872 return 0;
873 }
874
875 if (!uh->check) {
876 udp6_csum_zero_error(skb);
877 goto csum_error;
878 }
879
880 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
881 goto discard;
882
883 if (udp_lib_checksum_complete(skb))
884 goto csum_error;
885
886 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
887 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
888
889 kfree_skb(skb);
890 return 0;
891
892 short_packet:
893 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
894 proto == IPPROTO_UDPLITE ? "-Lite" : "",
895 saddr, ntohs(uh->source),
896 ulen, skb->len,
897 daddr, ntohs(uh->dest));
898 goto discard;
899 csum_error:
900 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
901 discard:
902 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
903 kfree_skb(skb);
904 return 0;
905 }
906
907
908 static struct sock *__udp6_lib_demux_lookup(struct net *net,
909 __be16 loc_port, const struct in6_addr *loc_addr,
910 __be16 rmt_port, const struct in6_addr *rmt_addr,
911 int dif, int sdif)
912 {
913 unsigned short hnum = ntohs(loc_port);
914 unsigned int hash2 = udp6_portaddr_hash(net, loc_addr, hnum);
915 unsigned int slot2 = hash2 & udp_table.mask;
916 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
917 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
918 struct sock *sk;
919
920 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
921 if (sk->sk_state == TCP_ESTABLISHED &&
922 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif, sdif))
923 return sk;
924 /* Only check first socket in chain */
925 break;
926 }
927 return NULL;
928 }
929
930 static void udp_v6_early_demux(struct sk_buff *skb)
931 {
932 struct net *net = dev_net(skb->dev);
933 const struct udphdr *uh;
934 struct sock *sk;
935 struct dst_entry *dst;
936 int dif = skb->dev->ifindex;
937 int sdif = inet6_sdif(skb);
938
939 if (!pskb_may_pull(skb, skb_transport_offset(skb) +
940 sizeof(struct udphdr)))
941 return;
942
943 uh = udp_hdr(skb);
944
945 if (skb->pkt_type == PACKET_HOST)
946 sk = __udp6_lib_demux_lookup(net, uh->dest,
947 &ipv6_hdr(skb)->daddr,
948 uh->source, &ipv6_hdr(skb)->saddr,
949 dif, sdif);
950 else
951 return;
952
953 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
954 return;
955
956 skb->sk = sk;
957 skb->destructor = sock_efree;
958 dst = READ_ONCE(sk->sk_rx_dst);
959
960 if (dst)
961 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
962 if (dst) {
963 /* set noref for now.
964 * any place which wants to hold dst has to call
965 * dst_hold_safe()
966 */
967 skb_dst_set_noref(skb, dst);
968 }
969 }
970
971 static __inline__ int udpv6_rcv(struct sk_buff *skb)
972 {
973 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
974 }
975
976 /*
977 * Throw away all pending data and cancel the corking. Socket is locked.
978 */
979 static void udp_v6_flush_pending_frames(struct sock *sk)
980 {
981 struct udp_sock *up = udp_sk(sk);
982
983 if (up->pending == AF_INET)
984 udp_flush_pending_frames(sk);
985 else if (up->pending) {
986 up->len = 0;
987 up->pending = 0;
988 ip6_flush_pending_frames(sk);
989 }
990 }
991
992 /**
993 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
994 * @sk: socket we are sending on
995 * @skb: sk_buff containing the filled-in UDP header
996 * (checksum field must be zeroed out)
997 */
998 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
999 const struct in6_addr *saddr,
1000 const struct in6_addr *daddr, int len)
1001 {
1002 unsigned int offset;
1003 struct udphdr *uh = udp_hdr(skb);
1004 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
1005 __wsum csum = 0;
1006
1007 if (!frags) {
1008 /* Only one fragment on the socket. */
1009 skb->csum_start = skb_transport_header(skb) - skb->head;
1010 skb->csum_offset = offsetof(struct udphdr, check);
1011 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1012 } else {
1013 /*
1014 * HW-checksum won't work as there are two or more
1015 * fragments on the socket so that all csums of sk_buffs
1016 * should be together
1017 */
1018 offset = skb_transport_offset(skb);
1019 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1020 csum = skb->csum;
1021
1022 skb->ip_summed = CHECKSUM_NONE;
1023
1024 do {
1025 csum = csum_add(csum, frags->csum);
1026 } while ((frags = frags->next));
1027
1028 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1029 csum);
1030 if (uh->check == 0)
1031 uh->check = CSUM_MANGLED_0;
1032 }
1033 }
1034
1035 /*
1036 * Sending
1037 */
1038
1039 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6)
1040 {
1041 struct sock *sk = skb->sk;
1042 struct udphdr *uh;
1043 int err = 0;
1044 int is_udplite = IS_UDPLITE(sk);
1045 __wsum csum = 0;
1046 int offset = skb_transport_offset(skb);
1047 int len = skb->len - offset;
1048
1049 /*
1050 * Create a UDP header
1051 */
1052 uh = udp_hdr(skb);
1053 uh->source = fl6->fl6_sport;
1054 uh->dest = fl6->fl6_dport;
1055 uh->len = htons(len);
1056 uh->check = 0;
1057
1058 if (is_udplite)
1059 csum = udplite_csum(skb);
1060 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */
1061 skb->ip_summed = CHECKSUM_NONE;
1062 goto send;
1063 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1064 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1065 goto send;
1066 } else
1067 csum = udp_csum(skb);
1068
1069 /* add protocol-dependent pseudo-header */
1070 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1071 len, fl6->flowi6_proto, csum);
1072 if (uh->check == 0)
1073 uh->check = CSUM_MANGLED_0;
1074
1075 send:
1076 err = ip6_send_skb(skb);
1077 if (err) {
1078 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1079 UDP6_INC_STATS(sock_net(sk),
1080 UDP_MIB_SNDBUFERRORS, is_udplite);
1081 err = 0;
1082 }
1083 } else {
1084 UDP6_INC_STATS(sock_net(sk),
1085 UDP_MIB_OUTDATAGRAMS, is_udplite);
1086 }
1087 return err;
1088 }
1089
1090 static int udp_v6_push_pending_frames(struct sock *sk)
1091 {
1092 struct sk_buff *skb;
1093 struct udp_sock *up = udp_sk(sk);
1094 struct flowi6 fl6;
1095 int err = 0;
1096
1097 if (up->pending == AF_INET)
1098 return udp_push_pending_frames(sk);
1099
1100 /* ip6_finish_skb will release the cork, so make a copy of
1101 * fl6 here.
1102 */
1103 fl6 = inet_sk(sk)->cork.fl.u.ip6;
1104
1105 skb = ip6_finish_skb(sk);
1106 if (!skb)
1107 goto out;
1108
1109 err = udp_v6_send_skb(skb, &fl6);
1110
1111 out:
1112 up->len = 0;
1113 up->pending = 0;
1114 return err;
1115 }
1116
1117 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1118 {
1119 struct ipv6_txoptions opt_space;
1120 struct udp_sock *up = udp_sk(sk);
1121 struct inet_sock *inet = inet_sk(sk);
1122 struct ipv6_pinfo *np = inet6_sk(sk);
1123 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1124 struct in6_addr *daddr, *final_p, final;
1125 struct ipv6_txoptions *opt = NULL;
1126 struct ipv6_txoptions *opt_to_free = NULL;
1127 struct ip6_flowlabel *flowlabel = NULL;
1128 struct flowi6 fl6;
1129 struct dst_entry *dst;
1130 struct ipcm6_cookie ipc6;
1131 int addr_len = msg->msg_namelen;
1132 int ulen = len;
1133 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1134 int err;
1135 int connected = 0;
1136 int is_udplite = IS_UDPLITE(sk);
1137 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1138 struct sockcm_cookie sockc;
1139
1140 ipc6.hlimit = -1;
1141 ipc6.tclass = -1;
1142 ipc6.dontfrag = -1;
1143 sockc.tsflags = sk->sk_tsflags;
1144
1145 /* destination address check */
1146 if (sin6) {
1147 if (addr_len < offsetof(struct sockaddr, sa_data))
1148 return -EINVAL;
1149
1150 switch (sin6->sin6_family) {
1151 case AF_INET6:
1152 if (addr_len < SIN6_LEN_RFC2133)
1153 return -EINVAL;
1154 daddr = &sin6->sin6_addr;
1155 if (ipv6_addr_any(daddr) &&
1156 ipv6_addr_v4mapped(&np->saddr))
1157 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1158 daddr);
1159 break;
1160 case AF_INET:
1161 goto do_udp_sendmsg;
1162 case AF_UNSPEC:
1163 msg->msg_name = sin6 = NULL;
1164 msg->msg_namelen = addr_len = 0;
1165 daddr = NULL;
1166 break;
1167 default:
1168 return -EINVAL;
1169 }
1170 } else if (!up->pending) {
1171 if (sk->sk_state != TCP_ESTABLISHED)
1172 return -EDESTADDRREQ;
1173 daddr = &sk->sk_v6_daddr;
1174 } else
1175 daddr = NULL;
1176
1177 if (daddr) {
1178 if (ipv6_addr_v4mapped(daddr)) {
1179 struct sockaddr_in sin;
1180 sin.sin_family = AF_INET;
1181 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1182 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1183 msg->msg_name = &sin;
1184 msg->msg_namelen = sizeof(sin);
1185 do_udp_sendmsg:
1186 if (__ipv6_only_sock(sk))
1187 return -ENETUNREACH;
1188 return udp_sendmsg(sk, msg, len);
1189 }
1190 }
1191
1192 if (up->pending == AF_INET)
1193 return udp_sendmsg(sk, msg, len);
1194
1195 /* Rough check on arithmetic overflow,
1196 better check is made in ip6_append_data().
1197 */
1198 if (len > INT_MAX - sizeof(struct udphdr))
1199 return -EMSGSIZE;
1200
1201 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1202 if (up->pending) {
1203 /*
1204 * There are pending frames.
1205 * The socket lock must be held while it's corked.
1206 */
1207 lock_sock(sk);
1208 if (likely(up->pending)) {
1209 if (unlikely(up->pending != AF_INET6)) {
1210 release_sock(sk);
1211 return -EAFNOSUPPORT;
1212 }
1213 dst = NULL;
1214 goto do_append_data;
1215 }
1216 release_sock(sk);
1217 }
1218 ulen += sizeof(struct udphdr);
1219
1220 memset(&fl6, 0, sizeof(fl6));
1221
1222 if (sin6) {
1223 if (sin6->sin6_port == 0)
1224 return -EINVAL;
1225
1226 fl6.fl6_dport = sin6->sin6_port;
1227 daddr = &sin6->sin6_addr;
1228
1229 if (np->sndflow) {
1230 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1231 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1232 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1233 if (!flowlabel)
1234 return -EINVAL;
1235 }
1236 }
1237
1238 /*
1239 * Otherwise it will be difficult to maintain
1240 * sk->sk_dst_cache.
1241 */
1242 if (sk->sk_state == TCP_ESTABLISHED &&
1243 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1244 daddr = &sk->sk_v6_daddr;
1245
1246 if (addr_len >= sizeof(struct sockaddr_in6) &&
1247 sin6->sin6_scope_id &&
1248 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1249 fl6.flowi6_oif = sin6->sin6_scope_id;
1250 } else {
1251 if (sk->sk_state != TCP_ESTABLISHED)
1252 return -EDESTADDRREQ;
1253
1254 fl6.fl6_dport = inet->inet_dport;
1255 daddr = &sk->sk_v6_daddr;
1256 fl6.flowlabel = np->flow_label;
1257 connected = 1;
1258 }
1259
1260 if (!fl6.flowi6_oif)
1261 fl6.flowi6_oif = sk->sk_bound_dev_if;
1262
1263 if (!fl6.flowi6_oif)
1264 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1265
1266 fl6.flowi6_mark = sk->sk_mark;
1267 fl6.flowi6_uid = sk->sk_uid;
1268
1269 if (msg->msg_controllen) {
1270 opt = &opt_space;
1271 memset(opt, 0, sizeof(struct ipv6_txoptions));
1272 opt->tot_len = sizeof(*opt);
1273 ipc6.opt = opt;
1274
1275 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, &ipc6, &sockc);
1276 if (err < 0) {
1277 fl6_sock_release(flowlabel);
1278 return err;
1279 }
1280 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1281 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1282 if (!flowlabel)
1283 return -EINVAL;
1284 }
1285 if (!(opt->opt_nflen|opt->opt_flen))
1286 opt = NULL;
1287 connected = 0;
1288 }
1289 if (!opt) {
1290 opt = txopt_get(np);
1291 opt_to_free = opt;
1292 }
1293 if (flowlabel)
1294 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1295 opt = ipv6_fixup_options(&opt_space, opt);
1296 ipc6.opt = opt;
1297
1298 fl6.flowi6_proto = sk->sk_protocol;
1299 if (!ipv6_addr_any(daddr))
1300 fl6.daddr = *daddr;
1301 else
1302 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1303 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1304 fl6.saddr = np->saddr;
1305 fl6.fl6_sport = inet->inet_sport;
1306
1307 final_p = fl6_update_dst(&fl6, opt, &final);
1308 if (final_p)
1309 connected = 0;
1310
1311 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1312 fl6.flowi6_oif = np->mcast_oif;
1313 connected = 0;
1314 } else if (!fl6.flowi6_oif)
1315 fl6.flowi6_oif = np->ucast_oif;
1316
1317 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1318
1319 if (ipc6.tclass < 0)
1320 ipc6.tclass = np->tclass;
1321
1322 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel);
1323
1324 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p);
1325 if (IS_ERR(dst)) {
1326 err = PTR_ERR(dst);
1327 dst = NULL;
1328 goto out;
1329 }
1330
1331 if (ipc6.hlimit < 0)
1332 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1333
1334 if (msg->msg_flags&MSG_CONFIRM)
1335 goto do_confirm;
1336 back_from_confirm:
1337
1338 /* Lockless fast path for the non-corking case */
1339 if (!corkreq) {
1340 struct sk_buff *skb;
1341
1342 skb = ip6_make_skb(sk, getfrag, msg, ulen,
1343 sizeof(struct udphdr), &ipc6,
1344 &fl6, (struct rt6_info *)dst,
1345 msg->msg_flags, &sockc);
1346 err = PTR_ERR(skb);
1347 if (!IS_ERR_OR_NULL(skb))
1348 err = udp_v6_send_skb(skb, &fl6);
1349 goto release_dst;
1350 }
1351
1352 lock_sock(sk);
1353 if (unlikely(up->pending)) {
1354 /* The socket is already corked while preparing it. */
1355 /* ... which is an evident application bug. --ANK */
1356 release_sock(sk);
1357
1358 net_dbg_ratelimited("udp cork app bug 2\n");
1359 err = -EINVAL;
1360 goto out;
1361 }
1362
1363 up->pending = AF_INET6;
1364
1365 do_append_data:
1366 if (ipc6.dontfrag < 0)
1367 ipc6.dontfrag = np->dontfrag;
1368 up->len += ulen;
1369 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr),
1370 &ipc6, &fl6, (struct rt6_info *)dst,
1371 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, &sockc);
1372 if (err)
1373 udp_v6_flush_pending_frames(sk);
1374 else if (!corkreq)
1375 err = udp_v6_push_pending_frames(sk);
1376 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1377 up->pending = 0;
1378
1379 if (err > 0)
1380 err = np->recverr ? net_xmit_errno(err) : 0;
1381 release_sock(sk);
1382
1383 release_dst:
1384 if (dst) {
1385 if (connected) {
1386 ip6_dst_store(sk, dst,
1387 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ?
1388 &sk->sk_v6_daddr : NULL,
1389 #ifdef CONFIG_IPV6_SUBTREES
1390 ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1391 &np->saddr :
1392 #endif
1393 NULL);
1394 } else {
1395 dst_release(dst);
1396 }
1397 dst = NULL;
1398 }
1399
1400 out:
1401 dst_release(dst);
1402 fl6_sock_release(flowlabel);
1403 txopt_put(opt_to_free);
1404 if (!err)
1405 return len;
1406 /*
1407 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1408 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1409 * we don't have a good statistic (IpOutDiscards but it can be too many
1410 * things). We could add another new stat but at least for now that
1411 * seems like overkill.
1412 */
1413 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1414 UDP6_INC_STATS(sock_net(sk),
1415 UDP_MIB_SNDBUFERRORS, is_udplite);
1416 }
1417 return err;
1418
1419 do_confirm:
1420 if (msg->msg_flags & MSG_PROBE)
1421 dst_confirm_neigh(dst, &fl6.daddr);
1422 if (!(msg->msg_flags&MSG_PROBE) || len)
1423 goto back_from_confirm;
1424 err = 0;
1425 goto out;
1426 }
1427
1428 void udpv6_destroy_sock(struct sock *sk)
1429 {
1430 struct udp_sock *up = udp_sk(sk);
1431 lock_sock(sk);
1432 udp_v6_flush_pending_frames(sk);
1433 release_sock(sk);
1434
1435 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
1436 void (*encap_destroy)(struct sock *sk);
1437 encap_destroy = READ_ONCE(up->encap_destroy);
1438 if (encap_destroy)
1439 encap_destroy(sk);
1440 }
1441
1442 inet6_destroy_sock(sk);
1443 }
1444
1445 /*
1446 * Socket option code for UDP
1447 */
1448 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1449 char __user *optval, unsigned int optlen)
1450 {
1451 if (level == SOL_UDP || level == SOL_UDPLITE)
1452 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1453 udp_v6_push_pending_frames);
1454 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1455 }
1456
1457 #ifdef CONFIG_COMPAT
1458 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1459 char __user *optval, unsigned int optlen)
1460 {
1461 if (level == SOL_UDP || level == SOL_UDPLITE)
1462 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1463 udp_v6_push_pending_frames);
1464 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1465 }
1466 #endif
1467
1468 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1469 char __user *optval, int __user *optlen)
1470 {
1471 if (level == SOL_UDP || level == SOL_UDPLITE)
1472 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1473 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1474 }
1475
1476 #ifdef CONFIG_COMPAT
1477 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1478 char __user *optval, int __user *optlen)
1479 {
1480 if (level == SOL_UDP || level == SOL_UDPLITE)
1481 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1482 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1483 }
1484 #endif
1485
1486 /* thinking of making this const? Don't.
1487 * early_demux can change based on sysctl.
1488 */
1489 static struct inet6_protocol udpv6_protocol = {
1490 .early_demux = udp_v6_early_demux,
1491 .early_demux_handler = udp_v6_early_demux,
1492 .handler = udpv6_rcv,
1493 .err_handler = udpv6_err,
1494 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1495 };
1496
1497 /* ------------------------------------------------------------------------ */
1498 #ifdef CONFIG_PROC_FS
1499 int udp6_seq_show(struct seq_file *seq, void *v)
1500 {
1501 if (v == SEQ_START_TOKEN) {
1502 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1503 } else {
1504 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1505 struct inet_sock *inet = inet_sk(v);
1506 __u16 srcp = ntohs(inet->inet_sport);
1507 __u16 destp = ntohs(inet->inet_dport);
1508 ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket);
1509 }
1510 return 0;
1511 }
1512
1513 static const struct file_operations udp6_afinfo_seq_fops = {
1514 .owner = THIS_MODULE,
1515 .open = udp_seq_open,
1516 .read = seq_read,
1517 .llseek = seq_lseek,
1518 .release = seq_release_net
1519 };
1520
1521 static struct udp_seq_afinfo udp6_seq_afinfo = {
1522 .name = "udp6",
1523 .family = AF_INET6,
1524 .udp_table = &udp_table,
1525 .seq_fops = &udp6_afinfo_seq_fops,
1526 .seq_ops = {
1527 .show = udp6_seq_show,
1528 },
1529 };
1530
1531 int __net_init udp6_proc_init(struct net *net)
1532 {
1533 return udp_proc_register(net, &udp6_seq_afinfo);
1534 }
1535
1536 void udp6_proc_exit(struct net *net)
1537 {
1538 udp_proc_unregister(net, &udp6_seq_afinfo);
1539 }
1540 #endif /* CONFIG_PROC_FS */
1541
1542 /* ------------------------------------------------------------------------ */
1543
1544 struct proto udpv6_prot = {
1545 .name = "UDPv6",
1546 .owner = THIS_MODULE,
1547 .close = udp_lib_close,
1548 .connect = ip6_datagram_connect,
1549 .disconnect = udp_disconnect,
1550 .ioctl = udp_ioctl,
1551 .init = udp_init_sock,
1552 .destroy = udpv6_destroy_sock,
1553 .setsockopt = udpv6_setsockopt,
1554 .getsockopt = udpv6_getsockopt,
1555 .sendmsg = udpv6_sendmsg,
1556 .recvmsg = udpv6_recvmsg,
1557 .release_cb = ip6_datagram_release_cb,
1558 .hash = udp_lib_hash,
1559 .unhash = udp_lib_unhash,
1560 .rehash = udp_v6_rehash,
1561 .get_port = udp_v6_get_port,
1562 .memory_allocated = &udp_memory_allocated,
1563 .sysctl_mem = sysctl_udp_mem,
1564 .sysctl_wmem = &sysctl_udp_wmem_min,
1565 .sysctl_rmem = &sysctl_udp_rmem_min,
1566 .obj_size = sizeof(struct udp6_sock),
1567 .h.udp_table = &udp_table,
1568 #ifdef CONFIG_COMPAT
1569 .compat_setsockopt = compat_udpv6_setsockopt,
1570 .compat_getsockopt = compat_udpv6_getsockopt,
1571 #endif
1572 .diag_destroy = udp_abort,
1573 };
1574
1575 static struct inet_protosw udpv6_protosw = {
1576 .type = SOCK_DGRAM,
1577 .protocol = IPPROTO_UDP,
1578 .prot = &udpv6_prot,
1579 .ops = &inet6_dgram_ops,
1580 .flags = INET_PROTOSW_PERMANENT,
1581 };
1582
1583 int __init udpv6_init(void)
1584 {
1585 int ret;
1586
1587 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1588 if (ret)
1589 goto out;
1590
1591 ret = inet6_register_protosw(&udpv6_protosw);
1592 if (ret)
1593 goto out_udpv6_protocol;
1594 out:
1595 return ret;
1596
1597 out_udpv6_protocol:
1598 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1599 goto out;
1600 }
1601
1602 void udpv6_exit(void)
1603 {
1604 inet6_unregister_protosw(&udpv6_protosw);
1605 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1606 }
1607
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.