?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
xtensa: support DMA buffers in high memory
[cris-mirror.git] / net / ipv6 / udp.c
blob52e3ea0e6f507702251b3dade8ae43181331d975
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 int udp_v6_get_port(struct sock *sk, unsigned short snum)
93 {
94 unsigned int hash2_nulladdr =
95 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
96 unsigned int hash2_partial =
97 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
98
99 /* precompute partial secondary hash */
100 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
101 return udp_lib_get_port(sk, snum, hash2_nulladdr);
102 }
103
104 static void udp_v6_rehash(struct sock *sk)
105 {
106 u16 new_hash = ipv6_portaddr_hash(sock_net(sk),
107 &sk->sk_v6_rcv_saddr,
108 inet_sk(sk)->inet_num);
109
110 udp_lib_rehash(sk, new_hash);
111 }
112
113 static int compute_score(struct sock *sk, struct net *net,
114 const struct in6_addr *saddr, __be16 sport,
115 const struct in6_addr *daddr, unsigned short hnum,
116 int dif, int sdif, bool exact_dif)
117 {
118 int score;
119 struct inet_sock *inet;
120
121 if (!net_eq(sock_net(sk), net) ||
122 udp_sk(sk)->udp_port_hash != hnum ||
123 sk->sk_family != PF_INET6)
124 return -1;
125
126 score = 0;
127 inet = inet_sk(sk);
128
129 if (inet->inet_dport) {
130 if (inet->inet_dport != sport)
131 return -1;
132 score++;
133 }
134
135 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
136 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
137 return -1;
138 score++;
139 }
140
141 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
142 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
143 return -1;
144 score++;
145 }
146
147 if (sk->sk_bound_dev_if || exact_dif) {
148 bool dev_match = (sk->sk_bound_dev_if == dif ||
149 sk->sk_bound_dev_if == sdif);
150
151 if (exact_dif && !dev_match)
152 return -1;
153 if (sk->sk_bound_dev_if && dev_match)
154 score++;
155 }
156
157 if (sk->sk_incoming_cpu == raw_smp_processor_id())
158 score++;
159
160 return score;
161 }
162
163 /* called with rcu_read_lock() */
164 static struct sock *udp6_lib_lookup2(struct net *net,
165 const struct in6_addr *saddr, __be16 sport,
166 const struct in6_addr *daddr, unsigned int hnum,
167 int dif, int sdif, bool exact_dif,
168 struct udp_hslot *hslot2, struct sk_buff *skb)
169 {
170 struct sock *sk, *result;
171 int score, badness;
172 u32 hash = 0;
173
174 result = NULL;
175 badness = -1;
176 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
177 score = compute_score(sk, net, saddr, sport,
178 daddr, hnum, dif, sdif, exact_dif);
179 if (score > badness) {
180 if (sk->sk_reuseport) {
181 hash = udp6_ehashfn(net, daddr, hnum,
182 saddr, sport);
183
184 result = reuseport_select_sock(sk, hash, skb,
185 sizeof(struct udphdr));
186 if (result)
187 return result;
188 }
189 result = sk;
190 badness = score;
191 }
192 }
193 return result;
194 }
195
196 /* rcu_read_lock() must be held */
197 struct sock *__udp6_lib_lookup(struct net *net,
198 const struct in6_addr *saddr, __be16 sport,
199 const struct in6_addr *daddr, __be16 dport,
200 int dif, int sdif, struct udp_table *udptable,
201 struct sk_buff *skb)
202 {
203 struct sock *sk, *result;
204 unsigned short hnum = ntohs(dport);
205 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
206 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
207 bool exact_dif = udp6_lib_exact_dif_match(net, skb);
208 int score, badness;
209 u32 hash = 0;
210
211 if (hslot->count > 10) {
212 hash2 = ipv6_portaddr_hash(net, daddr, hnum);
213 slot2 = hash2 & udptable->mask;
214 hslot2 = &udptable->hash2[slot2];
215 if (hslot->count < hslot2->count)
216 goto begin;
217
218 result = udp6_lib_lookup2(net, saddr, sport,
219 daddr, hnum, dif, sdif, exact_dif,
220 hslot2, skb);
221 if (!result) {
222 unsigned int old_slot2 = slot2;
223 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
224 slot2 = hash2 & udptable->mask;
225 /* avoid searching the same slot again. */
226 if (unlikely(slot2 == old_slot2))
227 return result;
228
229 hslot2 = &udptable->hash2[slot2];
230 if (hslot->count < hslot2->count)
231 goto begin;
232
233 result = udp6_lib_lookup2(net, saddr, sport,
234 daddr, hnum, dif, sdif,
235 exact_dif, hslot2,
236 skb);
237 }
238 return result;
239 }
240 begin:
241 result = NULL;
242 badness = -1;
243 sk_for_each_rcu(sk, &hslot->head) {
244 score = compute_score(sk, net, saddr, sport, daddr, hnum, dif,
245 sdif, exact_dif);
246 if (score > badness) {
247 if (sk->sk_reuseport) {
248 hash = udp6_ehashfn(net, daddr, hnum,
249 saddr, sport);
250 result = reuseport_select_sock(sk, hash, skb,
251 sizeof(struct udphdr));
252 if (result)
253 return result;
254 }
255 result = sk;
256 badness = score;
257 }
258 }
259 return result;
260 }
261 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
262
263 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
264 __be16 sport, __be16 dport,
265 struct udp_table *udptable)
266 {
267 const struct ipv6hdr *iph = ipv6_hdr(skb);
268
269 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
270 &iph->daddr, dport, inet6_iif(skb),
271 inet6_sdif(skb), udptable, skb);
272 }
273
274 struct sock *udp6_lib_lookup_skb(struct sk_buff *skb,
275 __be16 sport, __be16 dport)
276 {
277 const struct ipv6hdr *iph = ipv6_hdr(skb);
278
279 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
280 &iph->daddr, dport, inet6_iif(skb),
281 inet6_sdif(skb), &udp_table, skb);
282 }
283 EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb);
284
285 /* Must be called under rcu_read_lock().
286 * Does increment socket refcount.
287 */
288 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
289 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \
290 IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
291 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
292 const struct in6_addr *daddr, __be16 dport, int dif)
293 {
294 struct sock *sk;
295
296 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport,
297 dif, 0, &udp_table, NULL);
298 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
299 sk = NULL;
300 return sk;
301 }
302 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
303 #endif
304
305 /* do not use the scratch area len for jumbogram: their length execeeds the
306 * scratch area space; note that the IP6CB flags is still in the first
307 * cacheline, so checking for jumbograms is cheap
308 */
309 static int udp6_skb_len(struct sk_buff *skb)
310 {
311 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb);
312 }
313
314 /*
315 * This should be easy, if there is something there we
316 * return it, otherwise we block.
317 */
318
319 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
320 int noblock, int flags, int *addr_len)
321 {
322 struct ipv6_pinfo *np = inet6_sk(sk);
323 struct inet_sock *inet = inet_sk(sk);
324 struct sk_buff *skb;
325 unsigned int ulen, copied;
326 int peeked, peeking, off;
327 int err;
328 int is_udplite = IS_UDPLITE(sk);
329 bool checksum_valid = false;
330 int is_udp4;
331
332 if (flags & MSG_ERRQUEUE)
333 return ipv6_recv_error(sk, msg, len, addr_len);
334
335 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
336 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
337
338 try_again:
339 peeking = flags & MSG_PEEK;
340 off = sk_peek_offset(sk, flags);
341 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
342 if (!skb)
343 return err;
344
345 ulen = udp6_skb_len(skb);
346 copied = len;
347 if (copied > ulen - off)
348 copied = ulen - off;
349 else if (copied < ulen)
350 msg->msg_flags |= MSG_TRUNC;
351
352 is_udp4 = (skb->protocol == htons(ETH_P_IP));
353
354 /*
355 * If checksum is needed at all, try to do it while copying the
356 * data. If the data is truncated, or if we only want a partial
357 * coverage checksum (UDP-Lite), do it before the copy.
358 */
359
360 if (copied < ulen || peeking ||
361 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
362 checksum_valid = udp_skb_csum_unnecessary(skb) ||
363 !__udp_lib_checksum_complete(skb);
364 if (!checksum_valid)
365 goto csum_copy_err;
366 }
367
368 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
369 if (udp_skb_is_linear(skb))
370 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
371 else
372 err = skb_copy_datagram_msg(skb, off, msg, copied);
373 } else {
374 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
375 if (err == -EINVAL)
376 goto csum_copy_err;
377 }
378 if (unlikely(err)) {
379 if (!peeked) {
380 atomic_inc(&sk->sk_drops);
381 if (is_udp4)
382 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
383 is_udplite);
384 else
385 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
386 is_udplite);
387 }
388 kfree_skb(skb);
389 return err;
390 }
391 if (!peeked) {
392 if (is_udp4)
393 UDP_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS,
394 is_udplite);
395 else
396 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS,
397 is_udplite);
398 }
399
400 sock_recv_ts_and_drops(msg, sk, skb);
401
402 /* Copy the address. */
403 if (msg->msg_name) {
404 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
405 sin6->sin6_family = AF_INET6;
406 sin6->sin6_port = udp_hdr(skb)->source;
407 sin6->sin6_flowinfo = 0;
408
409 if (is_udp4) {
410 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
411 &sin6->sin6_addr);
412 sin6->sin6_scope_id = 0;
413 } else {
414 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
415 sin6->sin6_scope_id =
416 ipv6_iface_scope_id(&sin6->sin6_addr,
417 inet6_iif(skb));
418 }
419 *addr_len = sizeof(*sin6);
420 }
421
422 if (np->rxopt.all)
423 ip6_datagram_recv_common_ctl(sk, msg, skb);
424
425 if (is_udp4) {
426 if (inet->cmsg_flags)
427 ip_cmsg_recv_offset(msg, sk, skb,
428 sizeof(struct udphdr), off);
429 } else {
430 if (np->rxopt.all)
431 ip6_datagram_recv_specific_ctl(sk, msg, skb);
432 }
433
434 err = copied;
435 if (flags & MSG_TRUNC)
436 err = ulen;
437
438 skb_consume_udp(sk, skb, peeking ? -err : err);
439 return err;
440
441 csum_copy_err:
442 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
443 udp_skb_destructor)) {
444 if (is_udp4) {
445 UDP_INC_STATS(sock_net(sk),
446 UDP_MIB_CSUMERRORS, is_udplite);
447 UDP_INC_STATS(sock_net(sk),
448 UDP_MIB_INERRORS, is_udplite);
449 } else {
450 UDP6_INC_STATS(sock_net(sk),
451 UDP_MIB_CSUMERRORS, is_udplite);
452 UDP6_INC_STATS(sock_net(sk),
453 UDP_MIB_INERRORS, is_udplite);
454 }
455 }
456 kfree_skb(skb);
457
458 /* starting over for a new packet, but check if we need to yield */
459 cond_resched();
460 msg->msg_flags &= ~MSG_TRUNC;
461 goto try_again;
462 }
463
464 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
465 u8 type, u8 code, int offset, __be32 info,
466 struct udp_table *udptable)
467 {
468 struct ipv6_pinfo *np;
469 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
470 const struct in6_addr *saddr = &hdr->saddr;
471 const struct in6_addr *daddr = &hdr->daddr;
472 struct udphdr *uh = (struct udphdr *)(skb->data+offset);
473 struct sock *sk;
474 int harderr;
475 int err;
476 struct net *net = dev_net(skb->dev);
477
478 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
479 inet6_iif(skb), 0, udptable, skb);
480 if (!sk) {
481 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
482 ICMP6_MIB_INERRORS);
483 return;
484 }
485
486 harderr = icmpv6_err_convert(type, code, &err);
487 np = inet6_sk(sk);
488
489 if (type == ICMPV6_PKT_TOOBIG) {
490 if (!ip6_sk_accept_pmtu(sk))
491 goto out;
492 ip6_sk_update_pmtu(skb, sk, info);
493 if (np->pmtudisc != IPV6_PMTUDISC_DONT)
494 harderr = 1;
495 }
496 if (type == NDISC_REDIRECT) {
497 ip6_sk_redirect(skb, sk);
498 goto out;
499 }
500
501 if (!np->recverr) {
502 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
503 goto out;
504 } else {
505 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
506 }
507
508 sk->sk_err = err;
509 sk->sk_error_report(sk);
510 out:
511 return;
512 }
513
514 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
515 {
516 int rc;
517
518 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
519 sock_rps_save_rxhash(sk, skb);
520 sk_mark_napi_id(sk, skb);
521 sk_incoming_cpu_update(sk);
522 } else {
523 sk_mark_napi_id_once(sk, skb);
524 }
525
526 rc = __udp_enqueue_schedule_skb(sk, skb);
527 if (rc < 0) {
528 int is_udplite = IS_UDPLITE(sk);
529
530 /* Note that an ENOMEM error is charged twice */
531 if (rc == -ENOMEM)
532 UDP6_INC_STATS(sock_net(sk),
533 UDP_MIB_RCVBUFERRORS, is_udplite);
534 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
535 kfree_skb(skb);
536 return -1;
537 }
538
539 return 0;
540 }
541
542 static __inline__ void udpv6_err(struct sk_buff *skb,
543 struct inet6_skb_parm *opt, u8 type,
544 u8 code, int offset, __be32 info)
545 {
546 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
547 }
548
549 static struct static_key udpv6_encap_needed __read_mostly;
550 void udpv6_encap_enable(void)
551 {
552 static_key_enable(&udpv6_encap_needed);
553 }
554 EXPORT_SYMBOL(udpv6_encap_enable);
555
556 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
557 {
558 struct udp_sock *up = udp_sk(sk);
559 int is_udplite = IS_UDPLITE(sk);
560
561 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
562 goto drop;
563
564 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
565 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
566
567 /*
568 * This is an encapsulation socket so pass the skb to
569 * the socket's udp_encap_rcv() hook. Otherwise, just
570 * fall through and pass this up the UDP socket.
571 * up->encap_rcv() returns the following value:
572 * =0 if skb was successfully passed to the encap
573 * handler or was discarded by it.
574 * >0 if skb should be passed on to UDP.
575 * <0 if skb should be resubmitted as proto -N
576 */
577
578 /* if we're overly short, let UDP handle it */
579 encap_rcv = READ_ONCE(up->encap_rcv);
580 if (encap_rcv) {
581 int ret;
582
583 /* Verify checksum before giving to encap */
584 if (udp_lib_checksum_complete(skb))
585 goto csum_error;
586
587 ret = encap_rcv(sk, skb);
588 if (ret <= 0) {
589 __UDP_INC_STATS(sock_net(sk),
590 UDP_MIB_INDATAGRAMS,
591 is_udplite);
592 return -ret;
593 }
594 }
595
596 /* FALLTHROUGH -- it's a UDP Packet */
597 }
598
599 /*
600 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
601 */
602 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
603
604 if (up->pcrlen == 0) { /* full coverage was set */
605 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
606 UDP_SKB_CB(skb)->cscov, skb->len);
607 goto drop;
608 }
609 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
610 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
611 UDP_SKB_CB(skb)->cscov, up->pcrlen);
612 goto drop;
613 }
614 }
615
616 prefetch(&sk->sk_rmem_alloc);
617 if (rcu_access_pointer(sk->sk_filter) &&
618 udp_lib_checksum_complete(skb))
619 goto csum_error;
620
621 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
622 goto drop;
623
624 udp_csum_pull_header(skb);
625
626 skb_dst_drop(skb);
627
628 return __udpv6_queue_rcv_skb(sk, skb);
629
630 csum_error:
631 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
632 drop:
633 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
634 atomic_inc(&sk->sk_drops);
635 kfree_skb(skb);
636 return -1;
637 }
638
639 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
640 __be16 loc_port, const struct in6_addr *loc_addr,
641 __be16 rmt_port, const struct in6_addr *rmt_addr,
642 int dif, unsigned short hnum)
643 {
644 struct inet_sock *inet = inet_sk(sk);
645
646 if (!net_eq(sock_net(sk), net))
647 return false;
648
649 if (udp_sk(sk)->udp_port_hash != hnum ||
650 sk->sk_family != PF_INET6 ||
651 (inet->inet_dport && inet->inet_dport != rmt_port) ||
652 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
653 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
654 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif) ||
655 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
656 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
657 return false;
658 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
659 return false;
660 return true;
661 }
662
663 static void udp6_csum_zero_error(struct sk_buff *skb)
664 {
665 /* RFC 2460 section 8.1 says that we SHOULD log
666 * this error. Well, it is reasonable.
667 */
668 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
669 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
670 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
671 }
672
673 /*
674 * Note: called only from the BH handler context,
675 * so we don't need to lock the hashes.
676 */
677 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
678 const struct in6_addr *saddr, const struct in6_addr *daddr,
679 struct udp_table *udptable, int proto)
680 {
681 struct sock *sk, *first = NULL;
682 const struct udphdr *uh = udp_hdr(skb);
683 unsigned short hnum = ntohs(uh->dest);
684 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
685 unsigned int offset = offsetof(typeof(*sk), sk_node);
686 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
687 int dif = inet6_iif(skb);
688 struct hlist_node *node;
689 struct sk_buff *nskb;
690
691 if (use_hash2) {
692 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) &
693 udptable->mask;
694 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask;
695 start_lookup:
696 hslot = &udptable->hash2[hash2];
697 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
698 }
699
700 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
701 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
702 uh->source, saddr, dif, hnum))
703 continue;
704 /* If zero checksum and no_check is not on for
705 * the socket then skip it.
706 */
707 if (!uh->check && !udp_sk(sk)->no_check6_rx)
708 continue;
709 if (!first) {
710 first = sk;
711 continue;
712 }
713 nskb = skb_clone(skb, GFP_ATOMIC);
714 if (unlikely(!nskb)) {
715 atomic_inc(&sk->sk_drops);
716 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
717 IS_UDPLITE(sk));
718 __UDP6_INC_STATS(net, UDP_MIB_INERRORS,
719 IS_UDPLITE(sk));
720 continue;
721 }
722
723 if (udpv6_queue_rcv_skb(sk, nskb) > 0)
724 consume_skb(nskb);
725 }
726
727 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
728 if (use_hash2 && hash2 != hash2_any) {
729 hash2 = hash2_any;
730 goto start_lookup;
731 }
732
733 if (first) {
734 if (udpv6_queue_rcv_skb(first, skb) > 0)
735 consume_skb(skb);
736 } else {
737 kfree_skb(skb);
738 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
739 proto == IPPROTO_UDPLITE);
740 }
741 return 0;
742 }
743
744 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
745 {
746 if (udp_sk_rx_dst_set(sk, dst)) {
747 const struct rt6_info *rt = (const struct rt6_info *)dst;
748
749 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
750 }
751 }
752
753 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
754 int proto)
755 {
756 const struct in6_addr *saddr, *daddr;
757 struct net *net = dev_net(skb->dev);
758 struct udphdr *uh;
759 struct sock *sk;
760 u32 ulen = 0;
761
762 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
763 goto discard;
764
765 saddr = &ipv6_hdr(skb)->saddr;
766 daddr = &ipv6_hdr(skb)->daddr;
767 uh = udp_hdr(skb);
768
769 ulen = ntohs(uh->len);
770 if (ulen > skb->len)
771 goto short_packet;
772
773 if (proto == IPPROTO_UDP) {
774 /* UDP validates ulen. */
775
776 /* Check for jumbo payload */
777 if (ulen == 0)
778 ulen = skb->len;
779
780 if (ulen < sizeof(*uh))
781 goto short_packet;
782
783 if (ulen < skb->len) {
784 if (pskb_trim_rcsum(skb, ulen))
785 goto short_packet;
786 saddr = &ipv6_hdr(skb)->saddr;
787 daddr = &ipv6_hdr(skb)->daddr;
788 uh = udp_hdr(skb);
789 }
790 }
791
792 if (udp6_csum_init(skb, uh, proto))
793 goto csum_error;
794
795 /* Check if the socket is already available, e.g. due to early demux */
796 sk = skb_steal_sock(skb);
797 if (sk) {
798 struct dst_entry *dst = skb_dst(skb);
799 int ret;
800
801 if (unlikely(sk->sk_rx_dst != dst))
802 udp6_sk_rx_dst_set(sk, dst);
803
804 ret = udpv6_queue_rcv_skb(sk, skb);
805 sock_put(sk);
806
807 /* a return value > 0 means to resubmit the input */
808 if (ret > 0)
809 return ret;
810 return 0;
811 }
812
813 /*
814 * Multicast receive code
815 */
816 if (ipv6_addr_is_multicast(daddr))
817 return __udp6_lib_mcast_deliver(net, skb,
818 saddr, daddr, udptable, proto);
819
820 /* Unicast */
821 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
822 if (sk) {
823 int ret;
824
825 if (!uh->check && !udp_sk(sk)->no_check6_rx) {
826 udp6_csum_zero_error(skb);
827 goto csum_error;
828 }
829
830 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
831 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
832 ip6_compute_pseudo);
833
834 ret = udpv6_queue_rcv_skb(sk, skb);
835
836 /* a return value > 0 means to resubmit the input */
837 if (ret > 0)
838 return ret;
839
840 return 0;
841 }
842
843 if (!uh->check) {
844 udp6_csum_zero_error(skb);
845 goto csum_error;
846 }
847
848 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
849 goto discard;
850
851 if (udp_lib_checksum_complete(skb))
852 goto csum_error;
853
854 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
855 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
856
857 kfree_skb(skb);
858 return 0;
859
860 short_packet:
861 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
862 proto == IPPROTO_UDPLITE ? "-Lite" : "",
863 saddr, ntohs(uh->source),
864 ulen, skb->len,
865 daddr, ntohs(uh->dest));
866 goto discard;
867 csum_error:
868 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
869 discard:
870 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
871 kfree_skb(skb);
872 return 0;
873 }
874
875
876 static struct sock *__udp6_lib_demux_lookup(struct net *net,
877 __be16 loc_port, const struct in6_addr *loc_addr,
878 __be16 rmt_port, const struct in6_addr *rmt_addr,
879 int dif, int sdif)
880 {
881 unsigned short hnum = ntohs(loc_port);
882 unsigned int hash2 = ipv6_portaddr_hash(net, loc_addr, hnum);
883 unsigned int slot2 = hash2 & udp_table.mask;
884 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
885 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
886 struct sock *sk;
887
888 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
889 if (sk->sk_state == TCP_ESTABLISHED &&
890 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif, sdif))
891 return sk;
892 /* Only check first socket in chain */
893 break;
894 }
895 return NULL;
896 }
897
898 static void udp_v6_early_demux(struct sk_buff *skb)
899 {
900 struct net *net = dev_net(skb->dev);
901 const struct udphdr *uh;
902 struct sock *sk;
903 struct dst_entry *dst;
904 int dif = skb->dev->ifindex;
905 int sdif = inet6_sdif(skb);
906
907 if (!pskb_may_pull(skb, skb_transport_offset(skb) +
908 sizeof(struct udphdr)))
909 return;
910
911 uh = udp_hdr(skb);
912
913 if (skb->pkt_type == PACKET_HOST)
914 sk = __udp6_lib_demux_lookup(net, uh->dest,
915 &ipv6_hdr(skb)->daddr,
916 uh->source, &ipv6_hdr(skb)->saddr,
917 dif, sdif);
918 else
919 return;
920
921 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
922 return;
923
924 skb->sk = sk;
925 skb->destructor = sock_efree;
926 dst = READ_ONCE(sk->sk_rx_dst);
927
928 if (dst)
929 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
930 if (dst) {
931 /* set noref for now.
932 * any place which wants to hold dst has to call
933 * dst_hold_safe()
934 */
935 skb_dst_set_noref(skb, dst);
936 }
937 }
938
939 static __inline__ int udpv6_rcv(struct sk_buff *skb)
940 {
941 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
942 }
943
944 /*
945 * Throw away all pending data and cancel the corking. Socket is locked.
946 */
947 static void udp_v6_flush_pending_frames(struct sock *sk)
948 {
949 struct udp_sock *up = udp_sk(sk);
950
951 if (up->pending == AF_INET)
952 udp_flush_pending_frames(sk);
953 else if (up->pending) {
954 up->len = 0;
955 up->pending = 0;
956 ip6_flush_pending_frames(sk);
957 }
958 }
959
960 /**
961 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
962 * @sk: socket we are sending on
963 * @skb: sk_buff containing the filled-in UDP header
964 * (checksum field must be zeroed out)
965 */
966 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
967 const struct in6_addr *saddr,
968 const struct in6_addr *daddr, int len)
969 {
970 unsigned int offset;
971 struct udphdr *uh = udp_hdr(skb);
972 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
973 __wsum csum = 0;
974
975 if (!frags) {
976 /* Only one fragment on the socket. */
977 skb->csum_start = skb_transport_header(skb) - skb->head;
978 skb->csum_offset = offsetof(struct udphdr, check);
979 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
980 } else {
981 /*
982 * HW-checksum won't work as there are two or more
983 * fragments on the socket so that all csums of sk_buffs
984 * should be together
985 */
986 offset = skb_transport_offset(skb);
987 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
988 csum = skb->csum;
989
990 skb->ip_summed = CHECKSUM_NONE;
991
992 do {
993 csum = csum_add(csum, frags->csum);
994 } while ((frags = frags->next));
995
996 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
997 csum);
998 if (uh->check == 0)
999 uh->check = CSUM_MANGLED_0;
1000 }
1001 }
1002
1003 /*
1004 * Sending
1005 */
1006
1007 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6)
1008 {
1009 struct sock *sk = skb->sk;
1010 struct udphdr *uh;
1011 int err = 0;
1012 int is_udplite = IS_UDPLITE(sk);
1013 __wsum csum = 0;
1014 int offset = skb_transport_offset(skb);
1015 int len = skb->len - offset;
1016
1017 /*
1018 * Create a UDP header
1019 */
1020 uh = udp_hdr(skb);
1021 uh->source = fl6->fl6_sport;
1022 uh->dest = fl6->fl6_dport;
1023 uh->len = htons(len);
1024 uh->check = 0;
1025
1026 if (is_udplite)
1027 csum = udplite_csum(skb);
1028 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */
1029 skb->ip_summed = CHECKSUM_NONE;
1030 goto send;
1031 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1032 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1033 goto send;
1034 } else
1035 csum = udp_csum(skb);
1036
1037 /* add protocol-dependent pseudo-header */
1038 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1039 len, fl6->flowi6_proto, csum);
1040 if (uh->check == 0)
1041 uh->check = CSUM_MANGLED_0;
1042
1043 send:
1044 err = ip6_send_skb(skb);
1045 if (err) {
1046 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1047 UDP6_INC_STATS(sock_net(sk),
1048 UDP_MIB_SNDBUFERRORS, is_udplite);
1049 err = 0;
1050 }
1051 } else {
1052 UDP6_INC_STATS(sock_net(sk),
1053 UDP_MIB_OUTDATAGRAMS, is_udplite);
1054 }
1055 return err;
1056 }
1057
1058 static int udp_v6_push_pending_frames(struct sock *sk)
1059 {
1060 struct sk_buff *skb;
1061 struct udp_sock *up = udp_sk(sk);
1062 struct flowi6 fl6;
1063 int err = 0;
1064
1065 if (up->pending == AF_INET)
1066 return udp_push_pending_frames(sk);
1067
1068 /* ip6_finish_skb will release the cork, so make a copy of
1069 * fl6 here.
1070 */
1071 fl6 = inet_sk(sk)->cork.fl.u.ip6;
1072
1073 skb = ip6_finish_skb(sk);
1074 if (!skb)
1075 goto out;
1076
1077 err = udp_v6_send_skb(skb, &fl6);
1078
1079 out:
1080 up->len = 0;
1081 up->pending = 0;
1082 return err;
1083 }
1084
1085 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1086 {
1087 struct ipv6_txoptions opt_space;
1088 struct udp_sock *up = udp_sk(sk);
1089 struct inet_sock *inet = inet_sk(sk);
1090 struct ipv6_pinfo *np = inet6_sk(sk);
1091 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1092 struct in6_addr *daddr, *final_p, final;
1093 struct ipv6_txoptions *opt = NULL;
1094 struct ipv6_txoptions *opt_to_free = NULL;
1095 struct ip6_flowlabel *flowlabel = NULL;
1096 struct flowi6 fl6;
1097 struct dst_entry *dst;
1098 struct ipcm6_cookie ipc6;
1099 int addr_len = msg->msg_namelen;
1100 int ulen = len;
1101 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1102 int err;
1103 int connected = 0;
1104 int is_udplite = IS_UDPLITE(sk);
1105 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1106 struct sockcm_cookie sockc;
1107
1108 ipc6.hlimit = -1;
1109 ipc6.tclass = -1;
1110 ipc6.dontfrag = -1;
1111 sockc.tsflags = sk->sk_tsflags;
1112
1113 /* destination address check */
1114 if (sin6) {
1115 if (addr_len < offsetof(struct sockaddr, sa_data))
1116 return -EINVAL;
1117
1118 switch (sin6->sin6_family) {
1119 case AF_INET6:
1120 if (addr_len < SIN6_LEN_RFC2133)
1121 return -EINVAL;
1122 daddr = &sin6->sin6_addr;
1123 if (ipv6_addr_any(daddr) &&
1124 ipv6_addr_v4mapped(&np->saddr))
1125 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1126 daddr);
1127 break;
1128 case AF_INET:
1129 goto do_udp_sendmsg;
1130 case AF_UNSPEC:
1131 msg->msg_name = sin6 = NULL;
1132 msg->msg_namelen = addr_len = 0;
1133 daddr = NULL;
1134 break;
1135 default:
1136 return -EINVAL;
1137 }
1138 } else if (!up->pending) {
1139 if (sk->sk_state != TCP_ESTABLISHED)
1140 return -EDESTADDRREQ;
1141 daddr = &sk->sk_v6_daddr;
1142 } else
1143 daddr = NULL;
1144
1145 if (daddr) {
1146 if (ipv6_addr_v4mapped(daddr)) {
1147 struct sockaddr_in sin;
1148 sin.sin_family = AF_INET;
1149 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1150 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1151 msg->msg_name = &sin;
1152 msg->msg_namelen = sizeof(sin);
1153 do_udp_sendmsg:
1154 if (__ipv6_only_sock(sk))
1155 return -ENETUNREACH;
1156 return udp_sendmsg(sk, msg, len);
1157 }
1158 }
1159
1160 if (up->pending == AF_INET)
1161 return udp_sendmsg(sk, msg, len);
1162
1163 /* Rough check on arithmetic overflow,
1164 better check is made in ip6_append_data().
1165 */
1166 if (len > INT_MAX - sizeof(struct udphdr))
1167 return -EMSGSIZE;
1168
1169 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1170 if (up->pending) {
1171 /*
1172 * There are pending frames.
1173 * The socket lock must be held while it's corked.
1174 */
1175 lock_sock(sk);
1176 if (likely(up->pending)) {
1177 if (unlikely(up->pending != AF_INET6)) {
1178 release_sock(sk);
1179 return -EAFNOSUPPORT;
1180 }
1181 dst = NULL;
1182 goto do_append_data;
1183 }
1184 release_sock(sk);
1185 }
1186 ulen += sizeof(struct udphdr);
1187
1188 memset(&fl6, 0, sizeof(fl6));
1189
1190 if (sin6) {
1191 if (sin6->sin6_port == 0)
1192 return -EINVAL;
1193
1194 fl6.fl6_dport = sin6->sin6_port;
1195 daddr = &sin6->sin6_addr;
1196
1197 if (np->sndflow) {
1198 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1199 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1200 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1201 if (!flowlabel)
1202 return -EINVAL;
1203 }
1204 }
1205
1206 /*
1207 * Otherwise it will be difficult to maintain
1208 * sk->sk_dst_cache.
1209 */
1210 if (sk->sk_state == TCP_ESTABLISHED &&
1211 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1212 daddr = &sk->sk_v6_daddr;
1213
1214 if (addr_len >= sizeof(struct sockaddr_in6) &&
1215 sin6->sin6_scope_id &&
1216 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1217 fl6.flowi6_oif = sin6->sin6_scope_id;
1218 } else {
1219 if (sk->sk_state != TCP_ESTABLISHED)
1220 return -EDESTADDRREQ;
1221
1222 fl6.fl6_dport = inet->inet_dport;
1223 daddr = &sk->sk_v6_daddr;
1224 fl6.flowlabel = np->flow_label;
1225 connected = 1;
1226 }
1227
1228 if (!fl6.flowi6_oif)
1229 fl6.flowi6_oif = sk->sk_bound_dev_if;
1230
1231 if (!fl6.flowi6_oif)
1232 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1233
1234 fl6.flowi6_mark = sk->sk_mark;
1235 fl6.flowi6_uid = sk->sk_uid;
1236
1237 if (msg->msg_controllen) {
1238 opt = &opt_space;
1239 memset(opt, 0, sizeof(struct ipv6_txoptions));
1240 opt->tot_len = sizeof(*opt);
1241 ipc6.opt = opt;
1242
1243 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, &ipc6, &sockc);
1244 if (err < 0) {
1245 fl6_sock_release(flowlabel);
1246 return err;
1247 }
1248 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1249 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1250 if (!flowlabel)
1251 return -EINVAL;
1252 }
1253 if (!(opt->opt_nflen|opt->opt_flen))
1254 opt = NULL;
1255 connected = 0;
1256 }
1257 if (!opt) {
1258 opt = txopt_get(np);
1259 opt_to_free = opt;
1260 }
1261 if (flowlabel)
1262 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1263 opt = ipv6_fixup_options(&opt_space, opt);
1264 ipc6.opt = opt;
1265
1266 fl6.flowi6_proto = sk->sk_protocol;
1267 if (!ipv6_addr_any(daddr))
1268 fl6.daddr = *daddr;
1269 else
1270 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1271 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1272 fl6.saddr = np->saddr;
1273 fl6.fl6_sport = inet->inet_sport;
1274
1275 final_p = fl6_update_dst(&fl6, opt, &final);
1276 if (final_p)
1277 connected = 0;
1278
1279 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1280 fl6.flowi6_oif = np->mcast_oif;
1281 connected = 0;
1282 } else if (!fl6.flowi6_oif)
1283 fl6.flowi6_oif = np->ucast_oif;
1284
1285 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1286
1287 if (ipc6.tclass < 0)
1288 ipc6.tclass = np->tclass;
1289
1290 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel);
1291
1292 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p);
1293 if (IS_ERR(dst)) {
1294 err = PTR_ERR(dst);
1295 dst = NULL;
1296 goto out;
1297 }
1298
1299 if (ipc6.hlimit < 0)
1300 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1301
1302 if (msg->msg_flags&MSG_CONFIRM)
1303 goto do_confirm;
1304 back_from_confirm:
1305
1306 /* Lockless fast path for the non-corking case */
1307 if (!corkreq) {
1308 struct sk_buff *skb;
1309
1310 skb = ip6_make_skb(sk, getfrag, msg, ulen,
1311 sizeof(struct udphdr), &ipc6,
1312 &fl6, (struct rt6_info *)dst,
1313 msg->msg_flags, &sockc);
1314 err = PTR_ERR(skb);
1315 if (!IS_ERR_OR_NULL(skb))
1316 err = udp_v6_send_skb(skb, &fl6);
1317 goto release_dst;
1318 }
1319
1320 lock_sock(sk);
1321 if (unlikely(up->pending)) {
1322 /* The socket is already corked while preparing it. */
1323 /* ... which is an evident application bug. --ANK */
1324 release_sock(sk);
1325
1326 net_dbg_ratelimited("udp cork app bug 2\n");
1327 err = -EINVAL;
1328 goto out;
1329 }
1330
1331 up->pending = AF_INET6;
1332
1333 do_append_data:
1334 if (ipc6.dontfrag < 0)
1335 ipc6.dontfrag = np->dontfrag;
1336 up->len += ulen;
1337 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr),
1338 &ipc6, &fl6, (struct rt6_info *)dst,
1339 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, &sockc);
1340 if (err)
1341 udp_v6_flush_pending_frames(sk);
1342 else if (!corkreq)
1343 err = udp_v6_push_pending_frames(sk);
1344 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1345 up->pending = 0;
1346
1347 if (err > 0)
1348 err = np->recverr ? net_xmit_errno(err) : 0;
1349 release_sock(sk);
1350
1351 release_dst:
1352 if (dst) {
1353 if (connected) {
1354 ip6_dst_store(sk, dst,
1355 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ?
1356 &sk->sk_v6_daddr : NULL,
1357 #ifdef CONFIG_IPV6_SUBTREES
1358 ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1359 &np->saddr :
1360 #endif
1361 NULL);
1362 } else {
1363 dst_release(dst);
1364 }
1365 dst = NULL;
1366 }
1367
1368 out:
1369 dst_release(dst);
1370 fl6_sock_release(flowlabel);
1371 txopt_put(opt_to_free);
1372 if (!err)
1373 return len;
1374 /*
1375 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1376 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1377 * we don't have a good statistic (IpOutDiscards but it can be too many
1378 * things). We could add another new stat but at least for now that
1379 * seems like overkill.
1380 */
1381 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1382 UDP6_INC_STATS(sock_net(sk),
1383 UDP_MIB_SNDBUFERRORS, is_udplite);
1384 }
1385 return err;
1386
1387 do_confirm:
1388 if (msg->msg_flags & MSG_PROBE)
1389 dst_confirm_neigh(dst, &fl6.daddr);
1390 if (!(msg->msg_flags&MSG_PROBE) || len)
1391 goto back_from_confirm;
1392 err = 0;
1393 goto out;
1394 }
1395
1396 void udpv6_destroy_sock(struct sock *sk)
1397 {
1398 struct udp_sock *up = udp_sk(sk);
1399 lock_sock(sk);
1400 udp_v6_flush_pending_frames(sk);
1401 release_sock(sk);
1402
1403 if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
1404 void (*encap_destroy)(struct sock *sk);
1405 encap_destroy = READ_ONCE(up->encap_destroy);
1406 if (encap_destroy)
1407 encap_destroy(sk);
1408 }
1409
1410 inet6_destroy_sock(sk);
1411 }
1412
1413 /*
1414 * Socket option code for UDP
1415 */
1416 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1417 char __user *optval, unsigned int optlen)
1418 {
1419 if (level == SOL_UDP || level == SOL_UDPLITE)
1420 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1421 udp_v6_push_pending_frames);
1422 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1423 }
1424
1425 #ifdef CONFIG_COMPAT
1426 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1427 char __user *optval, unsigned int optlen)
1428 {
1429 if (level == SOL_UDP || level == SOL_UDPLITE)
1430 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1431 udp_v6_push_pending_frames);
1432 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1433 }
1434 #endif
1435
1436 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1437 char __user *optval, int __user *optlen)
1438 {
1439 if (level == SOL_UDP || level == SOL_UDPLITE)
1440 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1441 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1442 }
1443
1444 #ifdef CONFIG_COMPAT
1445 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1446 char __user *optval, int __user *optlen)
1447 {
1448 if (level == SOL_UDP || level == SOL_UDPLITE)
1449 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1450 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1451 }
1452 #endif
1453
1454 /* thinking of making this const? Don't.
1455 * early_demux can change based on sysctl.
1456 */
1457 static struct inet6_protocol udpv6_protocol = {
1458 .early_demux = udp_v6_early_demux,
1459 .early_demux_handler = udp_v6_early_demux,
1460 .handler = udpv6_rcv,
1461 .err_handler = udpv6_err,
1462 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1463 };
1464
1465 /* ------------------------------------------------------------------------ */
1466 #ifdef CONFIG_PROC_FS
1467 int udp6_seq_show(struct seq_file *seq, void *v)
1468 {
1469 if (v == SEQ_START_TOKEN) {
1470 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1471 } else {
1472 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1473 struct inet_sock *inet = inet_sk(v);
1474 __u16 srcp = ntohs(inet->inet_sport);
1475 __u16 destp = ntohs(inet->inet_dport);
1476 ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket);
1477 }
1478 return 0;
1479 }
1480
1481 static const struct file_operations udp6_afinfo_seq_fops = {
1482 .open = udp_seq_open,
1483 .read = seq_read,
1484 .llseek = seq_lseek,
1485 .release = seq_release_net
1486 };
1487
1488 static struct udp_seq_afinfo udp6_seq_afinfo = {
1489 .name = "udp6",
1490 .family = AF_INET6,
1491 .udp_table = &udp_table,
1492 .seq_fops = &udp6_afinfo_seq_fops,
1493 .seq_ops = {
1494 .show = udp6_seq_show,
1495 },
1496 };
1497
1498 int __net_init udp6_proc_init(struct net *net)
1499 {
1500 return udp_proc_register(net, &udp6_seq_afinfo);
1501 }
1502
1503 void udp6_proc_exit(struct net *net)
1504 {
1505 udp_proc_unregister(net, &udp6_seq_afinfo);
1506 }
1507 #endif /* CONFIG_PROC_FS */
1508
1509 /* ------------------------------------------------------------------------ */
1510
1511 struct proto udpv6_prot = {
1512 .name = "UDPv6",
1513 .owner = THIS_MODULE,
1514 .close = udp_lib_close,
1515 .connect = ip6_datagram_connect,
1516 .disconnect = udp_disconnect,
1517 .ioctl = udp_ioctl,
1518 .init = udp_init_sock,
1519 .destroy = udpv6_destroy_sock,
1520 .setsockopt = udpv6_setsockopt,
1521 .getsockopt = udpv6_getsockopt,
1522 .sendmsg = udpv6_sendmsg,
1523 .recvmsg = udpv6_recvmsg,
1524 .release_cb = ip6_datagram_release_cb,
1525 .hash = udp_lib_hash,
1526 .unhash = udp_lib_unhash,
1527 .rehash = udp_v6_rehash,
1528 .get_port = udp_v6_get_port,
1529 .memory_allocated = &udp_memory_allocated,
1530 .sysctl_mem = sysctl_udp_mem,
1531 .sysctl_wmem = &sysctl_udp_wmem_min,
1532 .sysctl_rmem = &sysctl_udp_rmem_min,
1533 .obj_size = sizeof(struct udp6_sock),
1534 .h.udp_table = &udp_table,
1535 #ifdef CONFIG_COMPAT
1536 .compat_setsockopt = compat_udpv6_setsockopt,
1537 .compat_getsockopt = compat_udpv6_getsockopt,
1538 #endif
1539 .diag_destroy = udp_abort,
1540 };
1541
1542 static struct inet_protosw udpv6_protosw = {
1543 .type = SOCK_DGRAM,
1544 .protocol = IPPROTO_UDP,
1545 .prot = &udpv6_prot,
1546 .ops = &inet6_dgram_ops,
1547 .flags = INET_PROTOSW_PERMANENT,
1548 };
1549
1550 int __init udpv6_init(void)
1551 {
1552 int ret;
1553
1554 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1555 if (ret)
1556 goto out;
1557
1558 ret = inet6_register_protosw(&udpv6_protosw);
1559 if (ret)
1560 goto out_udpv6_protocol;
1561 out:
1562 return ret;
1563
1564 out_udpv6_protocol:
1565 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1566 goto out;
1567 }
1568
1569 void udpv6_exit(void)
1570 {
1571 inet6_unregister_protosw(&udpv6_protosw);
1572 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1573 }
CRIS linux kernel tree