projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
ext3: Flush disk caches on fsync when needed
[deliverable/linux.git] / net / ipv6 / udp.c
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 <asm/uaccess.h>
38
39 #include <net/ndisc.h>
40 #include <net/protocol.h>
41 #include <net/transp_v6.h>
42 #include <net/ip6_route.h>
43 #include <net/raw.h>
44 #include <net/tcp_states.h>
45 #include <net/ip6_checksum.h>
46 #include <net/xfrm.h>
47
48 #include <linux/proc_fs.h>
49 #include <linux/seq_file.h>
50 #include "udp_impl.h"
51
52 int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2)
53 {
54 const struct in6_addr *sk_rcv_saddr6 = &inet6_sk(sk)->rcv_saddr;
55 const struct in6_addr *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2);
56 __be32 sk_rcv_saddr = inet_sk(sk)->rcv_saddr;
57 __be32 sk2_rcv_saddr = inet_rcv_saddr(sk2);
58 int sk_ipv6only = ipv6_only_sock(sk);
59 int sk2_ipv6only = inet_v6_ipv6only(sk2);
60 int addr_type = ipv6_addr_type(sk_rcv_saddr6);
61 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
62
63 /* if both are mapped, treat as IPv4 */
64 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED)
65 return (!sk2_ipv6only &&
66 (!sk_rcv_saddr || !sk2_rcv_saddr ||
67 sk_rcv_saddr == sk2_rcv_saddr));
68
69 if (addr_type2 == IPV6_ADDR_ANY &&
70 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
71 return 1;
72
73 if (addr_type == IPV6_ADDR_ANY &&
74 !(sk_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
75 return 1;
76
77 if (sk2_rcv_saddr6 &&
78 ipv6_addr_equal(sk_rcv_saddr6, sk2_rcv_saddr6))
79 return 1;
80
81 return 0;
82 }
83
84 int udp_v6_get_port(struct sock *sk, unsigned short snum)
85 {
86 return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal);
87 }
88
89 static inline int compute_score(struct sock *sk, struct net *net,
90 unsigned short hnum,
91 struct in6_addr *saddr, __be16 sport,
92 struct in6_addr *daddr, __be16 dport,
93 int dif)
94 {
95 int score = -1;
96
97 if (net_eq(sock_net(sk), net) && sk->sk_hash == hnum &&
98 sk->sk_family == PF_INET6) {
99 struct ipv6_pinfo *np = inet6_sk(sk);
100 struct inet_sock *inet = inet_sk(sk);
101
102 score = 0;
103 if (inet->dport) {
104 if (inet->dport != sport)
105 return -1;
106 score++;
107 }
108 if (!ipv6_addr_any(&np->rcv_saddr)) {
109 if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
110 return -1;
111 score++;
112 }
113 if (!ipv6_addr_any(&np->daddr)) {
114 if (!ipv6_addr_equal(&np->daddr, saddr))
115 return -1;
116 score++;
117 }
118 if (sk->sk_bound_dev_if) {
119 if (sk->sk_bound_dev_if != dif)
120 return -1;
121 score++;
122 }
123 }
124 return score;
125 }
126
127 static struct sock *__udp6_lib_lookup(struct net *net,
128 struct in6_addr *saddr, __be16 sport,
129 struct in6_addr *daddr, __be16 dport,
130 int dif, struct udp_table *udptable)
131 {
132 struct sock *sk, *result;
133 struct hlist_nulls_node *node;
134 unsigned short hnum = ntohs(dport);
135 unsigned int hash = udp_hashfn(net, hnum);
136 struct udp_hslot *hslot = &udptable->hash[hash];
137 int score, badness;
138
139 rcu_read_lock();
140 begin:
141 result = NULL;
142 badness = -1;
143 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
144 score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif);
145 if (score > badness) {
146 result = sk;
147 badness = score;
148 }
149 }
150 /*
151 * if the nulls value we got at the end of this lookup is
152 * not the expected one, we must restart lookup.
153 * We probably met an item that was moved to another chain.
154 */
155 if (get_nulls_value(node) != hash)
156 goto begin;
157
158 if (result) {
159 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
160 result = NULL;
161 else if (unlikely(compute_score(result, net, hnum, saddr, sport,
162 daddr, dport, dif) < badness)) {
163 sock_put(result);
164 goto begin;
165 }
166 }
167 rcu_read_unlock();
168 return result;
169 }
170
171 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
172 __be16 sport, __be16 dport,
173 struct udp_table *udptable)
174 {
175 struct sock *sk;
176 struct ipv6hdr *iph = ipv6_hdr(skb);
177
178 if (unlikely(sk = skb_steal_sock(skb)))
179 return sk;
180 return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport,
181 &iph->daddr, dport, inet6_iif(skb),
182 udptable);
183 }
184
185 /*
186 * This should be easy, if there is something there we
187 * return it, otherwise we block.
188 */
189
190 int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk,
191 struct msghdr *msg, size_t len,
192 int noblock, int flags, int *addr_len)
193 {
194 struct ipv6_pinfo *np = inet6_sk(sk);
195 struct inet_sock *inet = inet_sk(sk);
196 struct sk_buff *skb;
197 unsigned int ulen, copied;
198 int peeked;
199 int err;
200 int is_udplite = IS_UDPLITE(sk);
201 int is_udp4;
202
203 if (addr_len)
204 *addr_len=sizeof(struct sockaddr_in6);
205
206 if (flags & MSG_ERRQUEUE)
207 return ipv6_recv_error(sk, msg, len);
208
209 try_again:
210 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
211 &peeked, &err);
212 if (!skb)
213 goto out;
214
215 ulen = skb->len - sizeof(struct udphdr);
216 copied = len;
217 if (copied > ulen)
218 copied = ulen;
219 else if (copied < ulen)
220 msg->msg_flags |= MSG_TRUNC;
221
222 is_udp4 = (skb->protocol == htons(ETH_P_IP));
223
224 /*
225 * If checksum is needed at all, try to do it while copying the
226 * data. If the data is truncated, or if we only want a partial
227 * coverage checksum (UDP-Lite), do it before the copy.
228 */
229
230 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
231 if (udp_lib_checksum_complete(skb))
232 goto csum_copy_err;
233 }
234
235 if (skb_csum_unnecessary(skb))
236 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
237 msg->msg_iov, copied );
238 else {
239 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
240 if (err == -EINVAL)
241 goto csum_copy_err;
242 }
243 if (err)
244 goto out_free;
245
246 if (!peeked) {
247 if (is_udp4)
248 UDP_INC_STATS_USER(sock_net(sk),
249 UDP_MIB_INDATAGRAMS, is_udplite);
250 else
251 UDP6_INC_STATS_USER(sock_net(sk),
252 UDP_MIB_INDATAGRAMS, is_udplite);
253 }
254
255 sock_recv_timestamp(msg, sk, skb);
256
257 /* Copy the address. */
258 if (msg->msg_name) {
259 struct sockaddr_in6 *sin6;
260
261 sin6 = (struct sockaddr_in6 *) msg->msg_name;
262 sin6->sin6_family = AF_INET6;
263 sin6->sin6_port = udp_hdr(skb)->source;
264 sin6->sin6_flowinfo = 0;
265 sin6->sin6_scope_id = 0;
266
267 if (is_udp4)
268 ipv6_addr_set(&sin6->sin6_addr, 0, 0,
269 htonl(0xffff), ip_hdr(skb)->saddr);
270 else {
271 ipv6_addr_copy(&sin6->sin6_addr,
272 &ipv6_hdr(skb)->saddr);
273 if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
274 sin6->sin6_scope_id = IP6CB(skb)->iif;
275 }
276
277 }
278 if (is_udp4) {
279 if (inet->cmsg_flags)
280 ip_cmsg_recv(msg, skb);
281 } else {
282 if (np->rxopt.all)
283 datagram_recv_ctl(sk, msg, skb);
284 }
285
286 err = copied;
287 if (flags & MSG_TRUNC)
288 err = ulen;
289
290 out_free:
291 lock_sock(sk);
292 skb_free_datagram(sk, skb);
293 release_sock(sk);
294 out:
295 return err;
296
297 csum_copy_err:
298 lock_sock(sk);
299 if (!skb_kill_datagram(sk, skb, flags)) {
300 if (is_udp4)
301 UDP_INC_STATS_USER(sock_net(sk),
302 UDP_MIB_INERRORS, is_udplite);
303 else
304 UDP6_INC_STATS_USER(sock_net(sk),
305 UDP_MIB_INERRORS, is_udplite);
306 }
307 release_sock(sk);
308
309 if (flags & MSG_DONTWAIT)
310 return -EAGAIN;
311 goto try_again;
312 }
313
314 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
315 u8 type, u8 code, int offset, __be32 info,
316 struct udp_table *udptable)
317 {
318 struct ipv6_pinfo *np;
319 struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
320 struct in6_addr *saddr = &hdr->saddr;
321 struct in6_addr *daddr = &hdr->daddr;
322 struct udphdr *uh = (struct udphdr*)(skb->data+offset);
323 struct sock *sk;
324 int err;
325
326 sk = __udp6_lib_lookup(dev_net(skb->dev), daddr, uh->dest,
327 saddr, uh->source, inet6_iif(skb), udptable);
328 if (sk == NULL)
329 return;
330
331 np = inet6_sk(sk);
332
333 if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
334 goto out;
335
336 if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
337 goto out;
338
339 if (np->recverr)
340 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
341
342 sk->sk_err = err;
343 sk->sk_error_report(sk);
344 out:
345 sock_put(sk);
346 }
347
348 static __inline__ void udpv6_err(struct sk_buff *skb,
349 struct inet6_skb_parm *opt, u8 type,
350 u8 code, int offset, __be32 info )
351 {
352 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
353 }
354
355 int udpv6_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
356 {
357 struct udp_sock *up = udp_sk(sk);
358 int rc;
359 int is_udplite = IS_UDPLITE(sk);
360
361 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
362 goto drop;
363
364 /*
365 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
366 */
367 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
368
369 if (up->pcrlen == 0) { /* full coverage was set */
370 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage"
371 " %d while full coverage %d requested\n",
372 UDP_SKB_CB(skb)->cscov, skb->len);
373 goto drop;
374 }
375 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
376 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d "
377 "too small, need min %d\n",
378 UDP_SKB_CB(skb)->cscov, up->pcrlen);
379 goto drop;
380 }
381 }
382
383 if (sk->sk_filter) {
384 if (udp_lib_checksum_complete(skb))
385 goto drop;
386 }
387
388 if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) {
389 /* Note that an ENOMEM error is charged twice */
390 if (rc == -ENOMEM) {
391 UDP6_INC_STATS_BH(sock_net(sk),
392 UDP_MIB_RCVBUFERRORS, is_udplite);
393 atomic_inc(&sk->sk_drops);
394 }
395 goto drop;
396 }
397
398 return 0;
399 drop:
400 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
401 kfree_skb(skb);
402 return -1;
403 }
404
405 static struct sock *udp_v6_mcast_next(struct net *net, struct sock *sk,
406 __be16 loc_port, struct in6_addr *loc_addr,
407 __be16 rmt_port, struct in6_addr *rmt_addr,
408 int dif)
409 {
410 struct hlist_nulls_node *node;
411 struct sock *s = sk;
412 unsigned short num = ntohs(loc_port);
413
414 sk_nulls_for_each_from(s, node) {
415 struct inet_sock *inet = inet_sk(s);
416
417 if (!net_eq(sock_net(s), net))
418 continue;
419
420 if (s->sk_hash == num && s->sk_family == PF_INET6) {
421 struct ipv6_pinfo *np = inet6_sk(s);
422 if (inet->dport) {
423 if (inet->dport != rmt_port)
424 continue;
425 }
426 if (!ipv6_addr_any(&np->daddr) &&
427 !ipv6_addr_equal(&np->daddr, rmt_addr))
428 continue;
429
430 if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)
431 continue;
432
433 if (!ipv6_addr_any(&np->rcv_saddr)) {
434 if (!ipv6_addr_equal(&np->rcv_saddr, loc_addr))
435 continue;
436 }
437 if (!inet6_mc_check(s, loc_addr, rmt_addr))
438 continue;
439 return s;
440 }
441 }
442 return NULL;
443 }
444
445 /*
446 * Note: called only from the BH handler context,
447 * so we don't need to lock the hashes.
448 */
449 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
450 struct in6_addr *saddr, struct in6_addr *daddr,
451 struct udp_table *udptable)
452 {
453 struct sock *sk, *sk2;
454 const struct udphdr *uh = udp_hdr(skb);
455 struct udp_hslot *hslot = &udptable->hash[udp_hashfn(net, ntohs(uh->dest))];
456 int dif;
457
458 spin_lock(&hslot->lock);
459 sk = sk_nulls_head(&hslot->head);
460 dif = inet6_iif(skb);
461 sk = udp_v6_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
462 if (!sk) {
463 kfree_skb(skb);
464 goto out;
465 }
466
467 sk2 = sk;
468 while ((sk2 = udp_v6_mcast_next(net, sk_nulls_next(sk2), uh->dest, daddr,
469 uh->source, saddr, dif))) {
470 struct sk_buff *buff = skb_clone(skb, GFP_ATOMIC);
471 if (buff) {
472 bh_lock_sock(sk2);
473 if (!sock_owned_by_user(sk2))
474 udpv6_queue_rcv_skb(sk2, buff);
475 else
476 sk_add_backlog(sk2, buff);
477 bh_unlock_sock(sk2);
478 }
479 }
480 bh_lock_sock(sk);
481 if (!sock_owned_by_user(sk))
482 udpv6_queue_rcv_skb(sk, skb);
483 else
484 sk_add_backlog(sk, skb);
485 bh_unlock_sock(sk);
486 out:
487 spin_unlock(&hslot->lock);
488 return 0;
489 }
490
491 static inline int udp6_csum_init(struct sk_buff *skb, struct udphdr *uh,
492 int proto)
493 {
494 int err;
495
496 UDP_SKB_CB(skb)->partial_cov = 0;
497 UDP_SKB_CB(skb)->cscov = skb->len;
498
499 if (proto == IPPROTO_UDPLITE) {
500 err = udplite_checksum_init(skb, uh);
501 if (err)
502 return err;
503 }
504
505 if (uh->check == 0) {
506 /* RFC 2460 section 8.1 says that we SHOULD log
507 this error. Well, it is reasonable.
508 */
509 LIMIT_NETDEBUG(KERN_INFO "IPv6: udp checksum is 0\n");
510 return 1;
511 }
512 if (skb->ip_summed == CHECKSUM_COMPLETE &&
513 !csum_ipv6_magic(&ipv6_hdr(skb)->saddr, &ipv6_hdr(skb)->daddr,
514 skb->len, proto, skb->csum))
515 skb->ip_summed = CHECKSUM_UNNECESSARY;
516
517 if (!skb_csum_unnecessary(skb))
518 skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
519 &ipv6_hdr(skb)->daddr,
520 skb->len, proto, 0));
521
522 return 0;
523 }
524
525 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
526 int proto)
527 {
528 struct sock *sk;
529 struct udphdr *uh;
530 struct net_device *dev = skb->dev;
531 struct in6_addr *saddr, *daddr;
532 u32 ulen = 0;
533 struct net *net = dev_net(skb->dev);
534
535 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
536 goto short_packet;
537
538 saddr = &ipv6_hdr(skb)->saddr;
539 daddr = &ipv6_hdr(skb)->daddr;
540 uh = udp_hdr(skb);
541
542 ulen = ntohs(uh->len);
543 if (ulen > skb->len)
544 goto short_packet;
545
546 if (proto == IPPROTO_UDP) {
547 /* UDP validates ulen. */
548
549 /* Check for jumbo payload */
550 if (ulen == 0)
551 ulen = skb->len;
552
553 if (ulen < sizeof(*uh))
554 goto short_packet;
555
556 if (ulen < skb->len) {
557 if (pskb_trim_rcsum(skb, ulen))
558 goto short_packet;
559 saddr = &ipv6_hdr(skb)->saddr;
560 daddr = &ipv6_hdr(skb)->daddr;
561 uh = udp_hdr(skb);
562 }
563 }
564
565 if (udp6_csum_init(skb, uh, proto))
566 goto discard;
567
568 /*
569 * Multicast receive code
570 */
571 if (ipv6_addr_is_multicast(daddr))
572 return __udp6_lib_mcast_deliver(net, skb,
573 saddr, daddr, udptable);
574
575 /* Unicast */
576
577 /*
578 * check socket cache ... must talk to Alan about his plans
579 * for sock caches... i'll skip this for now.
580 */
581 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
582
583 if (sk == NULL) {
584 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
585 goto discard;
586
587 if (udp_lib_checksum_complete(skb))
588 goto discard;
589 UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS,
590 proto == IPPROTO_UDPLITE);
591
592 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0, dev);
593
594 kfree_skb(skb);
595 return 0;
596 }
597
598 /* deliver */
599
600 bh_lock_sock(sk);
601 if (!sock_owned_by_user(sk))
602 udpv6_queue_rcv_skb(sk, skb);
603 else
604 sk_add_backlog(sk, skb);
605 bh_unlock_sock(sk);
606 sock_put(sk);
607 return 0;
608
609 short_packet:
610 LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: %d/%u\n",
611 proto == IPPROTO_UDPLITE ? "-Lite" : "",
612 ulen, skb->len);
613
614 discard:
615 UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
616 kfree_skb(skb);
617 return 0;
618 }
619
620 static __inline__ int udpv6_rcv(struct sk_buff *skb)
621 {
622 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
623 }
624
625 /*
626 * Throw away all pending data and cancel the corking. Socket is locked.
627 */
628 static void udp_v6_flush_pending_frames(struct sock *sk)
629 {
630 struct udp_sock *up = udp_sk(sk);
631
632 if (up->pending == AF_INET)
633 udp_flush_pending_frames(sk);
634 else if (up->pending) {
635 up->len = 0;
636 up->pending = 0;
637 ip6_flush_pending_frames(sk);
638 }
639 }
640
641 /**
642 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
643 * @sk: socket we are sending on
644 * @skb: sk_buff containing the filled-in UDP header
645 * (checksum field must be zeroed out)
646 */
647 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
648 const struct in6_addr *saddr,
649 const struct in6_addr *daddr, int len)
650 {
651 unsigned int offset;
652 struct udphdr *uh = udp_hdr(skb);
653 __wsum csum = 0;
654
655 if (skb_queue_len(&sk->sk_write_queue) == 1) {
656 /* Only one fragment on the socket. */
657 skb->csum_start = skb_transport_header(skb) - skb->head;
658 skb->csum_offset = offsetof(struct udphdr, check);
659 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
660 } else {
661 /*
662 * HW-checksum won't work as there are two or more
663 * fragments on the socket so that all csums of sk_buffs
664 * should be together
665 */
666 offset = skb_transport_offset(skb);
667 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
668
669 skb->ip_summed = CHECKSUM_NONE;
670
671 skb_queue_walk(&sk->sk_write_queue, skb) {
672 csum = csum_add(csum, skb->csum);
673 }
674
675 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
676 csum);
677 if (uh->check == 0)
678 uh->check = CSUM_MANGLED_0;
679 }
680 }
681
682 /*
683 * Sending
684 */
685
686 static int udp_v6_push_pending_frames(struct sock *sk)
687 {
688 struct sk_buff *skb;
689 struct udphdr *uh;
690 struct udp_sock *up = udp_sk(sk);
691 struct inet_sock *inet = inet_sk(sk);
692 struct flowi *fl = &inet->cork.fl;
693 int err = 0;
694 int is_udplite = IS_UDPLITE(sk);
695 __wsum csum = 0;
696
697 /* Grab the skbuff where UDP header space exists. */
698 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
699 goto out;
700
701 /*
702 * Create a UDP header
703 */
704 uh = udp_hdr(skb);
705 uh->source = fl->fl_ip_sport;
706 uh->dest = fl->fl_ip_dport;
707 uh->len = htons(up->len);
708 uh->check = 0;
709
710 if (is_udplite)
711 csum = udplite_csum_outgoing(sk, skb);
712 else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
713 udp6_hwcsum_outgoing(sk, skb, &fl->fl6_src, &fl->fl6_dst,
714 up->len);
715 goto send;
716 } else
717 csum = udp_csum_outgoing(sk, skb);
718
719 /* add protocol-dependent pseudo-header */
720 uh->check = csum_ipv6_magic(&fl->fl6_src, &fl->fl6_dst,
721 up->len, fl->proto, csum );
722 if (uh->check == 0)
723 uh->check = CSUM_MANGLED_0;
724
725 send:
726 err = ip6_push_pending_frames(sk);
727 if (err) {
728 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
729 UDP6_INC_STATS_USER(sock_net(sk),
730 UDP_MIB_SNDBUFERRORS, is_udplite);
731 err = 0;
732 }
733 } else
734 UDP6_INC_STATS_USER(sock_net(sk),
735 UDP_MIB_OUTDATAGRAMS, is_udplite);
736 out:
737 up->len = 0;
738 up->pending = 0;
739 return err;
740 }
741
742 int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
743 struct msghdr *msg, size_t len)
744 {
745 struct ipv6_txoptions opt_space;
746 struct udp_sock *up = udp_sk(sk);
747 struct inet_sock *inet = inet_sk(sk);
748 struct ipv6_pinfo *np = inet6_sk(sk);
749 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) msg->msg_name;
750 struct in6_addr *daddr, *final_p = NULL, final;
751 struct ipv6_txoptions *opt = NULL;
752 struct ip6_flowlabel *flowlabel = NULL;
753 struct flowi fl;
754 struct dst_entry *dst;
755 int addr_len = msg->msg_namelen;
756 int ulen = len;
757 int hlimit = -1;
758 int tclass = -1;
759 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
760 int err;
761 int connected = 0;
762 int is_udplite = IS_UDPLITE(sk);
763 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
764
765 /* destination address check */
766 if (sin6) {
767 if (addr_len < offsetof(struct sockaddr, sa_data))
768 return -EINVAL;
769
770 switch (sin6->sin6_family) {
771 case AF_INET6:
772 if (addr_len < SIN6_LEN_RFC2133)
773 return -EINVAL;
774 daddr = &sin6->sin6_addr;
775 break;
776 case AF_INET:
777 goto do_udp_sendmsg;
778 case AF_UNSPEC:
779 msg->msg_name = sin6 = NULL;
780 msg->msg_namelen = addr_len = 0;
781 daddr = NULL;
782 break;
783 default:
784 return -EINVAL;
785 }
786 } else if (!up->pending) {
787 if (sk->sk_state != TCP_ESTABLISHED)
788 return -EDESTADDRREQ;
789 daddr = &np->daddr;
790 } else
791 daddr = NULL;
792
793 if (daddr) {
794 if (ipv6_addr_v4mapped(daddr)) {
795 struct sockaddr_in sin;
796 sin.sin_family = AF_INET;
797 sin.sin_port = sin6 ? sin6->sin6_port : inet->dport;
798 sin.sin_addr.s_addr = daddr->s6_addr32[3];
799 msg->msg_name = &sin;
800 msg->msg_namelen = sizeof(sin);
801 do_udp_sendmsg:
802 if (__ipv6_only_sock(sk))
803 return -ENETUNREACH;
804 return udp_sendmsg(iocb, sk, msg, len);
805 }
806 }
807
808 if (up->pending == AF_INET)
809 return udp_sendmsg(iocb, sk, msg, len);
810
811 /* Rough check on arithmetic overflow,
812 better check is made in ip6_append_data().
813 */
814 if (len > INT_MAX - sizeof(struct udphdr))
815 return -EMSGSIZE;
816
817 if (up->pending) {
818 /*
819 * There are pending frames.
820 * The socket lock must be held while it's corked.
821 */
822 lock_sock(sk);
823 if (likely(up->pending)) {
824 if (unlikely(up->pending != AF_INET6)) {
825 release_sock(sk);
826 return -EAFNOSUPPORT;
827 }
828 dst = NULL;
829 goto do_append_data;
830 }
831 release_sock(sk);
832 }
833 ulen += sizeof(struct udphdr);
834
835 memset(&fl, 0, sizeof(fl));
836
837 if (sin6) {
838 if (sin6->sin6_port == 0)
839 return -EINVAL;
840
841 fl.fl_ip_dport = sin6->sin6_port;
842 daddr = &sin6->sin6_addr;
843
844 if (np->sndflow) {
845 fl.fl6_flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
846 if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
847 flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
848 if (flowlabel == NULL)
849 return -EINVAL;
850 daddr = &flowlabel->dst;
851 }
852 }
853
854 /*
855 * Otherwise it will be difficult to maintain
856 * sk->sk_dst_cache.
857 */
858 if (sk->sk_state == TCP_ESTABLISHED &&
859 ipv6_addr_equal(daddr, &np->daddr))
860 daddr = &np->daddr;
861
862 if (addr_len >= sizeof(struct sockaddr_in6) &&
863 sin6->sin6_scope_id &&
864 ipv6_addr_type(daddr)&IPV6_ADDR_LINKLOCAL)
865 fl.oif = sin6->sin6_scope_id;
866 } else {
867 if (sk->sk_state != TCP_ESTABLISHED)
868 return -EDESTADDRREQ;
869
870 fl.fl_ip_dport = inet->dport;
871 daddr = &np->daddr;
872 fl.fl6_flowlabel = np->flow_label;
873 connected = 1;
874 }
875
876 if (!fl.oif)
877 fl.oif = sk->sk_bound_dev_if;
878
879 if (!fl.oif)
880 fl.oif = np->sticky_pktinfo.ipi6_ifindex;
881
882 if (msg->msg_controllen) {
883 opt = &opt_space;
884 memset(opt, 0, sizeof(struct ipv6_txoptions));
885 opt->tot_len = sizeof(*opt);
886
887 err = datagram_send_ctl(sock_net(sk), msg, &fl, opt, &hlimit, &tclass);
888 if (err < 0) {
889 fl6_sock_release(flowlabel);
890 return err;
891 }
892 if ((fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
893 flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
894 if (flowlabel == NULL)
895 return -EINVAL;
896 }
897 if (!(opt->opt_nflen|opt->opt_flen))
898 opt = NULL;
899 connected = 0;
900 }
901 if (opt == NULL)
902 opt = np->opt;
903 if (flowlabel)
904 opt = fl6_merge_options(&opt_space, flowlabel, opt);
905 opt = ipv6_fixup_options(&opt_space, opt);
906
907 fl.proto = sk->sk_protocol;
908 if (!ipv6_addr_any(daddr))
909 ipv6_addr_copy(&fl.fl6_dst, daddr);
910 else
911 fl.fl6_dst.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
912 if (ipv6_addr_any(&fl.fl6_src) && !ipv6_addr_any(&np->saddr))
913 ipv6_addr_copy(&fl.fl6_src, &np->saddr);
914 fl.fl_ip_sport = inet->sport;
915
916 /* merge ip6_build_xmit from ip6_output */
917 if (opt && opt->srcrt) {
918 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
919 ipv6_addr_copy(&final, &fl.fl6_dst);
920 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
921 final_p = &final;
922 connected = 0;
923 }
924
925 if (!fl.oif && ipv6_addr_is_multicast(&fl.fl6_dst)) {
926 fl.oif = np->mcast_oif;
927 connected = 0;
928 }
929
930 security_sk_classify_flow(sk, &fl);
931
932 err = ip6_sk_dst_lookup(sk, &dst, &fl);
933 if (err)
934 goto out;
935 if (final_p)
936 ipv6_addr_copy(&fl.fl6_dst, final_p);
937
938 err = __xfrm_lookup(sock_net(sk), &dst, &fl, sk, XFRM_LOOKUP_WAIT);
939 if (err < 0) {
940 if (err == -EREMOTE)
941 err = ip6_dst_blackhole(sk, &dst, &fl);
942 if (err < 0)
943 goto out;
944 }
945
946 if (hlimit < 0) {
947 if (ipv6_addr_is_multicast(&fl.fl6_dst))
948 hlimit = np->mcast_hops;
949 else
950 hlimit = np->hop_limit;
951 if (hlimit < 0)
952 hlimit = ip6_dst_hoplimit(dst);
953 }
954
955 if (tclass < 0)
956 tclass = np->tclass;
957
958 if (msg->msg_flags&MSG_CONFIRM)
959 goto do_confirm;
960 back_from_confirm:
961
962 lock_sock(sk);
963 if (unlikely(up->pending)) {
964 /* The socket is already corked while preparing it. */
965 /* ... which is an evident application bug. --ANK */
966 release_sock(sk);
967
968 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
969 err = -EINVAL;
970 goto out;
971 }
972
973 up->pending = AF_INET6;
974
975 do_append_data:
976 up->len += ulen;
977 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
978 err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen,
979 sizeof(struct udphdr), hlimit, tclass, opt, &fl,
980 (struct rt6_info*)dst,
981 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
982 if (err)
983 udp_v6_flush_pending_frames(sk);
984 else if (!corkreq)
985 err = udp_v6_push_pending_frames(sk);
986 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
987 up->pending = 0;
988
989 if (dst) {
990 if (connected) {
991 ip6_dst_store(sk, dst,
992 ipv6_addr_equal(&fl.fl6_dst, &np->daddr) ?
993 &np->daddr : NULL,
994 #ifdef CONFIG_IPV6_SUBTREES
995 ipv6_addr_equal(&fl.fl6_src, &np->saddr) ?
996 &np->saddr :
997 #endif
998 NULL);
999 } else {
1000 dst_release(dst);
1001 }
1002 dst = NULL;
1003 }
1004
1005 if (err > 0)
1006 err = np->recverr ? net_xmit_errno(err) : 0;
1007 release_sock(sk);
1008 out:
1009 dst_release(dst);
1010 fl6_sock_release(flowlabel);
1011 if (!err)
1012 return len;
1013 /*
1014 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1015 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1016 * we don't have a good statistic (IpOutDiscards but it can be too many
1017 * things). We could add another new stat but at least for now that
1018 * seems like overkill.
1019 */
1020 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1021 UDP6_INC_STATS_USER(sock_net(sk),
1022 UDP_MIB_SNDBUFERRORS, is_udplite);
1023 }
1024 return err;
1025
1026 do_confirm:
1027 dst_confirm(dst);
1028 if (!(msg->msg_flags&MSG_PROBE) || len)
1029 goto back_from_confirm;
1030 err = 0;
1031 goto out;
1032 }
1033
1034 void udpv6_destroy_sock(struct sock *sk)
1035 {
1036 lock_sock(sk);
1037 udp_v6_flush_pending_frames(sk);
1038 release_sock(sk);
1039
1040 inet6_destroy_sock(sk);
1041 }
1042
1043 /*
1044 * Socket option code for UDP
1045 */
1046 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1047 char __user *optval, int optlen)
1048 {
1049 if (level == SOL_UDP || level == SOL_UDPLITE)
1050 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1051 udp_v6_push_pending_frames);
1052 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1053 }
1054
1055 #ifdef CONFIG_COMPAT
1056 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1057 char __user *optval, int optlen)
1058 {
1059 if (level == SOL_UDP || level == SOL_UDPLITE)
1060 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1061 udp_v6_push_pending_frames);
1062 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1063 }
1064 #endif
1065
1066 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1067 char __user *optval, int __user *optlen)
1068 {
1069 if (level == SOL_UDP || level == SOL_UDPLITE)
1070 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1071 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1072 }
1073
1074 #ifdef CONFIG_COMPAT
1075 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1076 char __user *optval, int __user *optlen)
1077 {
1078 if (level == SOL_UDP || level == SOL_UDPLITE)
1079 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1080 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1081 }
1082 #endif
1083
1084 static int udp6_ufo_send_check(struct sk_buff *skb)
1085 {
1086 struct ipv6hdr *ipv6h;
1087 struct udphdr *uh;
1088
1089 if (!pskb_may_pull(skb, sizeof(*uh)))
1090 return -EINVAL;
1091
1092 ipv6h = ipv6_hdr(skb);
1093 uh = udp_hdr(skb);
1094
1095 uh->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, skb->len,
1096 IPPROTO_UDP, 0);
1097 skb->csum_start = skb_transport_header(skb) - skb->head;
1098 skb->csum_offset = offsetof(struct udphdr, check);
1099 skb->ip_summed = CHECKSUM_PARTIAL;
1100 return 0;
1101 }
1102
1103 static struct sk_buff *udp6_ufo_fragment(struct sk_buff *skb, int features)
1104 {
1105 struct sk_buff *segs = ERR_PTR(-EINVAL);
1106 unsigned int mss;
1107 unsigned int unfrag_ip6hlen, unfrag_len;
1108 struct frag_hdr *fptr;
1109 u8 *mac_start, *prevhdr;
1110 u8 nexthdr;
1111 u8 frag_hdr_sz = sizeof(struct frag_hdr);
1112 int offset;
1113 __wsum csum;
1114
1115 mss = skb_shinfo(skb)->gso_size;
1116 if (unlikely(skb->len <= mss))
1117 goto out;
1118
1119 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
1120 /* Packet is from an untrusted source, reset gso_segs. */
1121 int type = skb_shinfo(skb)->gso_type;
1122
1123 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
1124 !(type & (SKB_GSO_UDP))))
1125 goto out;
1126
1127 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
1128
1129 segs = NULL;
1130 goto out;
1131 }
1132
1133 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
1134 * do checksum of UDP packets sent as multiple IP fragments.
1135 */
1136 offset = skb->csum_start - skb_headroom(skb);
1137 csum = skb_checksum(skb, offset, skb->len- offset, 0);
1138 offset += skb->csum_offset;
1139 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1140 skb->ip_summed = CHECKSUM_NONE;
1141
1142 /* Check if there is enough headroom to insert fragment header. */
1143 if ((skb_headroom(skb) < frag_hdr_sz) &&
1144 pskb_expand_head(skb, frag_hdr_sz, 0, GFP_ATOMIC))
1145 goto out;
1146
1147 /* Find the unfragmentable header and shift it left by frag_hdr_sz
1148 * bytes to insert fragment header.
1149 */
1150 unfrag_ip6hlen = ip6_find_1stfragopt(skb, &prevhdr);
1151 nexthdr = *prevhdr;
1152 *prevhdr = NEXTHDR_FRAGMENT;
1153 unfrag_len = skb_network_header(skb) - skb_mac_header(skb) +
1154 unfrag_ip6hlen;
1155 mac_start = skb_mac_header(skb);
1156 memmove(mac_start-frag_hdr_sz, mac_start, unfrag_len);
1157
1158 skb->mac_header -= frag_hdr_sz;
1159 skb->network_header -= frag_hdr_sz;
1160
1161 fptr = (struct frag_hdr *)(skb_network_header(skb) + unfrag_ip6hlen);
1162 fptr->nexthdr = nexthdr;
1163 fptr->reserved = 0;
1164 ipv6_select_ident(fptr);
1165
1166 /* Fragment the skb. ipv6 header and the remaining fields of the
1167 * fragment header are updated in ipv6_gso_segment()
1168 */
1169 segs = skb_segment(skb, features);
1170
1171 out:
1172 return segs;
1173 }
1174
1175 static struct inet6_protocol udpv6_protocol = {
1176 .handler = udpv6_rcv,
1177 .err_handler = udpv6_err,
1178 .gso_send_check = udp6_ufo_send_check,
1179 .gso_segment = udp6_ufo_fragment,
1180 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1181 };
1182
1183 /* ------------------------------------------------------------------------ */
1184 #ifdef CONFIG_PROC_FS
1185
1186 static void udp6_sock_seq_show(struct seq_file *seq, struct sock *sp, int bucket)
1187 {
1188 struct inet_sock *inet = inet_sk(sp);
1189 struct ipv6_pinfo *np = inet6_sk(sp);
1190 struct in6_addr *dest, *src;
1191 __u16 destp, srcp;
1192
1193 dest = &np->daddr;
1194 src = &np->rcv_saddr;
1195 destp = ntohs(inet->dport);
1196 srcp = ntohs(inet->sport);
1197 seq_printf(seq,
1198 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1199 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d\n",
1200 bucket,
1201 src->s6_addr32[0], src->s6_addr32[1],
1202 src->s6_addr32[2], src->s6_addr32[3], srcp,
1203 dest->s6_addr32[0], dest->s6_addr32[1],
1204 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1205 sp->sk_state,
1206 sk_wmem_alloc_get(sp),
1207 sk_rmem_alloc_get(sp),
1208 0, 0L, 0,
1209 sock_i_uid(sp), 0,
1210 sock_i_ino(sp),
1211 atomic_read(&sp->sk_refcnt), sp,
1212 atomic_read(&sp->sk_drops));
1213 }
1214
1215 int udp6_seq_show(struct seq_file *seq, void *v)
1216 {
1217 if (v == SEQ_START_TOKEN)
1218 seq_printf(seq,
1219 " sl "
1220 "local_address "
1221 "remote_address "
1222 "st tx_queue rx_queue tr tm->when retrnsmt"
1223 " uid timeout inode ref pointer drops\n");
1224 else
1225 udp6_sock_seq_show(seq, v, ((struct udp_iter_state *)seq->private)->bucket);
1226 return 0;
1227 }
1228
1229 static struct udp_seq_afinfo udp6_seq_afinfo = {
1230 .name = "udp6",
1231 .family = AF_INET6,
1232 .udp_table = &udp_table,
1233 .seq_fops = {
1234 .owner = THIS_MODULE,
1235 },
1236 .seq_ops = {
1237 .show = udp6_seq_show,
1238 },
1239 };
1240
1241 int udp6_proc_init(struct net *net)
1242 {
1243 return udp_proc_register(net, &udp6_seq_afinfo);
1244 }
1245
1246 void udp6_proc_exit(struct net *net) {
1247 udp_proc_unregister(net, &udp6_seq_afinfo);
1248 }
1249 #endif /* CONFIG_PROC_FS */
1250
1251 /* ------------------------------------------------------------------------ */
1252
1253 struct proto udpv6_prot = {
1254 .name = "UDPv6",
1255 .owner = THIS_MODULE,
1256 .close = udp_lib_close,
1257 .connect = ip6_datagram_connect,
1258 .disconnect = udp_disconnect,
1259 .ioctl = udp_ioctl,
1260 .destroy = udpv6_destroy_sock,
1261 .setsockopt = udpv6_setsockopt,
1262 .getsockopt = udpv6_getsockopt,
1263 .sendmsg = udpv6_sendmsg,
1264 .recvmsg = udpv6_recvmsg,
1265 .backlog_rcv = udpv6_queue_rcv_skb,
1266 .hash = udp_lib_hash,
1267 .unhash = udp_lib_unhash,
1268 .get_port = udp_v6_get_port,
1269 .memory_allocated = &udp_memory_allocated,
1270 .sysctl_mem = sysctl_udp_mem,
1271 .sysctl_wmem = &sysctl_udp_wmem_min,
1272 .sysctl_rmem = &sysctl_udp_rmem_min,
1273 .obj_size = sizeof(struct udp6_sock),
1274 .slab_flags = SLAB_DESTROY_BY_RCU,
1275 .h.udp_table = &udp_table,
1276 #ifdef CONFIG_COMPAT
1277 .compat_setsockopt = compat_udpv6_setsockopt,
1278 .compat_getsockopt = compat_udpv6_getsockopt,
1279 #endif
1280 };
1281
1282 static struct inet_protosw udpv6_protosw = {
1283 .type = SOCK_DGRAM,
1284 .protocol = IPPROTO_UDP,
1285 .prot = &udpv6_prot,
1286 .ops = &inet6_dgram_ops,
1287 .capability =-1,
1288 .no_check = UDP_CSUM_DEFAULT,
1289 .flags = INET_PROTOSW_PERMANENT,
1290 };
1291
1292
1293 int __init udpv6_init(void)
1294 {
1295 int ret;
1296
1297 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1298 if (ret)
1299 goto out;
1300
1301 ret = inet6_register_protosw(&udpv6_protosw);
1302 if (ret)
1303 goto out_udpv6_protocol;
1304 out:
1305 return ret;
1306
1307 out_udpv6_protocol:
1308 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1309 goto out;
1310 }
1311
1312 void udpv6_exit(void)
1313 {
1314 inet6_unregister_protosw(&udpv6_protosw);
1315 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1316 }
Linux kernel - Deliverables
This page took 0.076203 seconds and 5 git commands to generate.