?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
x86: fix 32-bit FRAME_POINTER chasing code
[linux-2.6/zen-sources.git] / net / ipv6 / udp.c
blobbd4b9df8f614018322c6b9176f136d30db3dafac
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 * $Id: udp.c,v 1.65 2002/02/01 22:01:04 davem Exp $
11 *
12 * Fixes:
13 * Hideaki YOSHIFUJI : sin6_scope_id support
14 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
15 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
16 * a single port at the same time.
17 * Kazunori MIYAZAWA @USAGI: change process style to use ip6_append_data
18 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/udp6 to seq_file.
19 *
20 * This program is free software; you can redistribute it and/or
21 * modify it under the terms of the GNU General Public License
22 * as published by the Free Software Foundation; either version
23 * 2 of the License, or (at your option) any later version.
24 */
25
26 #include <linux/errno.h>
27 #include <linux/types.h>
28 #include <linux/socket.h>
29 #include <linux/sockios.h>
30 #include <linux/net.h>
31 #include <linux/in6.h>
32 #include <linux/netdevice.h>
33 #include <linux/if_arp.h>
34 #include <linux/ipv6.h>
35 #include <linux/icmpv6.h>
36 #include <linux/init.h>
37 #include <linux/module.h>
38 #include <linux/skbuff.h>
39 #include <asm/uaccess.h>
40
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
50 #include <linux/proc_fs.h>
51 #include <linux/seq_file.h>
52 #include "udp_impl.h"
53
54 static inline int udp_v6_get_port(struct sock *sk, unsigned short snum)
55 {
56 return udp_get_port(sk, snum, ipv6_rcv_saddr_equal);
57 }
58
59 static struct sock *__udp6_lib_lookup(struct in6_addr *saddr, __be16 sport,
60 struct in6_addr *daddr, __be16 dport,
61 int dif, struct hlist_head udptable[])
62 {
63 struct sock *sk, *result = NULL;
64 struct hlist_node *node;
65 unsigned short hnum = ntohs(dport);
66 int badness = -1;
67
68 read_lock(&udp_hash_lock);
69 sk_for_each(sk, node, &udptable[hnum & (UDP_HTABLE_SIZE - 1)]) {
70 struct inet_sock *inet = inet_sk(sk);
71
72 if (sk->sk_hash == hnum && sk->sk_family == PF_INET6) {
73 struct ipv6_pinfo *np = inet6_sk(sk);
74 int score = 0;
75 if (inet->dport) {
76 if (inet->dport != sport)
77 continue;
78 score++;
79 }
80 if (!ipv6_addr_any(&np->rcv_saddr)) {
81 if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
82 continue;
83 score++;
84 }
85 if (!ipv6_addr_any(&np->daddr)) {
86 if (!ipv6_addr_equal(&np->daddr, saddr))
87 continue;
88 score++;
89 }
90 if (sk->sk_bound_dev_if) {
91 if (sk->sk_bound_dev_if != dif)
92 continue;
93 score++;
94 }
95 if (score == 4) {
96 result = sk;
97 break;
98 } else if (score > badness) {
99 result = sk;
100 badness = score;
101 }
102 }
103 }
104 if (result)
105 sock_hold(result);
106 read_unlock(&udp_hash_lock);
107 return result;
108 }
109
110 /*
111 * This should be easy, if there is something there we
112 * return it, otherwise we block.
113 */
114
115 int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk,
116 struct msghdr *msg, size_t len,
117 int noblock, int flags, int *addr_len)
118 {
119 struct ipv6_pinfo *np = inet6_sk(sk);
120 struct inet_sock *inet = inet_sk(sk);
121 struct sk_buff *skb;
122 unsigned int ulen, copied;
123 int peeked;
124 int err;
125 int is_udplite = IS_UDPLITE(sk);
126
127 if (addr_len)
128 *addr_len=sizeof(struct sockaddr_in6);
129
130 if (flags & MSG_ERRQUEUE)
131 return ipv6_recv_error(sk, msg, len);
132
133 try_again:
134 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
135 &peeked, &err);
136 if (!skb)
137 goto out;
138
139 ulen = skb->len - sizeof(struct udphdr);
140 copied = len;
141 if (copied > ulen)
142 copied = ulen;
143 else if (copied < ulen)
144 msg->msg_flags |= MSG_TRUNC;
145
146 /*
147 * If checksum is needed at all, try to do it while copying the
148 * data. If the data is truncated, or if we only want a partial
149 * coverage checksum (UDP-Lite), do it before the copy.
150 */
151
152 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
153 if (udp_lib_checksum_complete(skb))
154 goto csum_copy_err;
155 }
156
157 if (skb_csum_unnecessary(skb))
158 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
159 msg->msg_iov, copied );
160 else {
161 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
162 if (err == -EINVAL)
163 goto csum_copy_err;
164 }
165 if (err)
166 goto out_free;
167
168 if (!peeked)
169 UDP6_INC_STATS_USER(UDP_MIB_INDATAGRAMS, is_udplite);
170
171 sock_recv_timestamp(msg, sk, skb);
172
173 /* Copy the address. */
174 if (msg->msg_name) {
175 struct sockaddr_in6 *sin6;
176
177 sin6 = (struct sockaddr_in6 *) msg->msg_name;
178 sin6->sin6_family = AF_INET6;
179 sin6->sin6_port = udp_hdr(skb)->source;
180 sin6->sin6_flowinfo = 0;
181 sin6->sin6_scope_id = 0;
182
183 if (skb->protocol == htons(ETH_P_IP))
184 ipv6_addr_set(&sin6->sin6_addr, 0, 0,
185 htonl(0xffff), ip_hdr(skb)->saddr);
186 else {
187 ipv6_addr_copy(&sin6->sin6_addr,
188 &ipv6_hdr(skb)->saddr);
189 if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
190 sin6->sin6_scope_id = IP6CB(skb)->iif;
191 }
192
193 }
194 if (skb->protocol == htons(ETH_P_IP)) {
195 if (inet->cmsg_flags)
196 ip_cmsg_recv(msg, skb);
197 } else {
198 if (np->rxopt.all)
199 datagram_recv_ctl(sk, msg, skb);
200 }
201
202 err = copied;
203 if (flags & MSG_TRUNC)
204 err = ulen;
205
206 out_free:
207 lock_sock(sk);
208 skb_free_datagram(sk, skb);
209 release_sock(sk);
210 out:
211 return err;
212
213 csum_copy_err:
214 lock_sock(sk);
215 if (!skb_kill_datagram(sk, skb, flags))
216 UDP6_INC_STATS_USER(UDP_MIB_INERRORS, is_udplite);
217 release_sock(sk);
218
219 if (flags & MSG_DONTWAIT)
220 return -EAGAIN;
221 goto try_again;
222 }
223
224 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
225 int type, int code, int offset, __be32 info,
226 struct hlist_head udptable[] )
227 {
228 struct ipv6_pinfo *np;
229 struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
230 struct in6_addr *saddr = &hdr->saddr;
231 struct in6_addr *daddr = &hdr->daddr;
232 struct udphdr *uh = (struct udphdr*)(skb->data+offset);
233 struct sock *sk;
234 int err;
235
236 sk = __udp6_lib_lookup(daddr, uh->dest,
237 saddr, uh->source, inet6_iif(skb), udptable);
238 if (sk == NULL)
239 return;
240
241 np = inet6_sk(sk);
242
243 if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
244 goto out;
245
246 if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
247 goto out;
248
249 if (np->recverr)
250 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
251
252 sk->sk_err = err;
253 sk->sk_error_report(sk);
254 out:
255 sock_put(sk);
256 }
257
258 static __inline__ void udpv6_err(struct sk_buff *skb,
259 struct inet6_skb_parm *opt, int type,
260 int code, int offset, __be32 info )
261 {
262 __udp6_lib_err(skb, opt, type, code, offset, info, udp_hash);
263 }
264
265 int udpv6_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
266 {
267 struct udp_sock *up = udp_sk(sk);
268 int rc;
269 int is_udplite = IS_UDPLITE(sk);
270
271 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
272 goto drop;
273
274 /*
275 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
276 */
277 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
278
279 if (up->pcrlen == 0) { /* full coverage was set */
280 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage"
281 " %d while full coverage %d requested\n",
282 UDP_SKB_CB(skb)->cscov, skb->len);
283 goto drop;
284 }
285 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
286 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d "
287 "too small, need min %d\n",
288 UDP_SKB_CB(skb)->cscov, up->pcrlen);
289 goto drop;
290 }
291 }
292
293 if (sk->sk_filter) {
294 if (udp_lib_checksum_complete(skb))
295 goto drop;
296 }
297
298 if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) {
299 /* Note that an ENOMEM error is charged twice */
300 if (rc == -ENOMEM)
301 UDP6_INC_STATS_BH(UDP_MIB_RCVBUFERRORS, is_udplite);
302 goto drop;
303 }
304
305 return 0;
306 drop:
307 UDP6_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite);
308 kfree_skb(skb);
309 return -1;
310 }
311
312 static struct sock *udp_v6_mcast_next(struct sock *sk,
313 __be16 loc_port, struct in6_addr *loc_addr,
314 __be16 rmt_port, struct in6_addr *rmt_addr,
315 int dif)
316 {
317 struct hlist_node *node;
318 struct sock *s = sk;
319 unsigned short num = ntohs(loc_port);
320
321 sk_for_each_from(s, node) {
322 struct inet_sock *inet = inet_sk(s);
323
324 if (s->sk_hash == num && s->sk_family == PF_INET6) {
325 struct ipv6_pinfo *np = inet6_sk(s);
326 if (inet->dport) {
327 if (inet->dport != rmt_port)
328 continue;
329 }
330 if (!ipv6_addr_any(&np->daddr) &&
331 !ipv6_addr_equal(&np->daddr, rmt_addr))
332 continue;
333
334 if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)
335 continue;
336
337 if (!ipv6_addr_any(&np->rcv_saddr)) {
338 if (!ipv6_addr_equal(&np->rcv_saddr, loc_addr))
339 continue;
340 }
341 if (!inet6_mc_check(s, loc_addr, rmt_addr))
342 continue;
343 return s;
344 }
345 }
346 return NULL;
347 }
348
349 /*
350 * Note: called only from the BH handler context,
351 * so we don't need to lock the hashes.
352 */
353 static int __udp6_lib_mcast_deliver(struct sk_buff *skb, struct in6_addr *saddr,
354 struct in6_addr *daddr, struct hlist_head udptable[])
355 {
356 struct sock *sk, *sk2;
357 const struct udphdr *uh = udp_hdr(skb);
358 int dif;
359
360 read_lock(&udp_hash_lock);
361 sk = sk_head(&udptable[ntohs(uh->dest) & (UDP_HTABLE_SIZE - 1)]);
362 dif = inet6_iif(skb);
363 sk = udp_v6_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
364 if (!sk) {
365 kfree_skb(skb);
366 goto out;
367 }
368
369 sk2 = sk;
370 while ((sk2 = udp_v6_mcast_next(sk_next(sk2), uh->dest, daddr,
371 uh->source, saddr, dif))) {
372 struct sk_buff *buff = skb_clone(skb, GFP_ATOMIC);
373 if (buff) {
374 bh_lock_sock_nested(sk2);
375 if (!sock_owned_by_user(sk2))
376 udpv6_queue_rcv_skb(sk2, buff);
377 else
378 sk_add_backlog(sk2, buff);
379 bh_unlock_sock(sk2);
380 }
381 }
382 bh_lock_sock_nested(sk);
383 if (!sock_owned_by_user(sk))
384 udpv6_queue_rcv_skb(sk, skb);
385 else
386 sk_add_backlog(sk, skb);
387 bh_unlock_sock(sk);
388 out:
389 read_unlock(&udp_hash_lock);
390 return 0;
391 }
392
393 static inline int udp6_csum_init(struct sk_buff *skb, struct udphdr *uh,
394 int proto)
395 {
396 int err;
397
398 UDP_SKB_CB(skb)->partial_cov = 0;
399 UDP_SKB_CB(skb)->cscov = skb->len;
400
401 if (proto == IPPROTO_UDPLITE) {
402 err = udplite_checksum_init(skb, uh);
403 if (err)
404 return err;
405 }
406
407 if (uh->check == 0) {
408 /* RFC 2460 section 8.1 says that we SHOULD log
409 this error. Well, it is reasonable.
410 */
411 LIMIT_NETDEBUG(KERN_INFO "IPv6: udp checksum is 0\n");
412 return 1;
413 }
414 if (skb->ip_summed == CHECKSUM_COMPLETE &&
415 !csum_ipv6_magic(&ipv6_hdr(skb)->saddr, &ipv6_hdr(skb)->daddr,
416 skb->len, proto, skb->csum))
417 skb->ip_summed = CHECKSUM_UNNECESSARY;
418
419 if (!skb_csum_unnecessary(skb))
420 skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
421 &ipv6_hdr(skb)->daddr,
422 skb->len, proto, 0));
423
424 return 0;
425 }
426
427 int __udp6_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
428 int proto)
429 {
430 struct sock *sk;
431 struct udphdr *uh;
432 struct net_device *dev = skb->dev;
433 struct in6_addr *saddr, *daddr;
434 u32 ulen = 0;
435
436 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
437 goto short_packet;
438
439 saddr = &ipv6_hdr(skb)->saddr;
440 daddr = &ipv6_hdr(skb)->daddr;
441 uh = udp_hdr(skb);
442
443 ulen = ntohs(uh->len);
444 if (ulen > skb->len)
445 goto short_packet;
446
447 if (proto == IPPROTO_UDP) {
448 /* UDP validates ulen. */
449
450 /* Check for jumbo payload */
451 if (ulen == 0)
452 ulen = skb->len;
453
454 if (ulen < sizeof(*uh))
455 goto short_packet;
456
457 if (ulen < skb->len) {
458 if (pskb_trim_rcsum(skb, ulen))
459 goto short_packet;
460 saddr = &ipv6_hdr(skb)->saddr;
461 daddr = &ipv6_hdr(skb)->daddr;
462 uh = udp_hdr(skb);
463 }
464 }
465
466 if (udp6_csum_init(skb, uh, proto))
467 goto discard;
468
469 /*
470 * Multicast receive code
471 */
472 if (ipv6_addr_is_multicast(daddr))
473 return __udp6_lib_mcast_deliver(skb, saddr, daddr, udptable);
474
475 /* Unicast */
476
477 /*
478 * check socket cache ... must talk to Alan about his plans
479 * for sock caches... i'll skip this for now.
480 */
481 sk = __udp6_lib_lookup(saddr, uh->source,
482 daddr, uh->dest, inet6_iif(skb), udptable);
483
484 if (sk == NULL) {
485 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
486 goto discard;
487
488 if (udp_lib_checksum_complete(skb))
489 goto discard;
490 UDP6_INC_STATS_BH(UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
491
492 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0, dev);
493
494 kfree_skb(skb);
495 return 0;
496 }
497
498 /* deliver */
499
500 bh_lock_sock_nested(sk);
501 if (!sock_owned_by_user(sk))
502 udpv6_queue_rcv_skb(sk, skb);
503 else
504 sk_add_backlog(sk, skb);
505 bh_unlock_sock(sk);
506 sock_put(sk);
507 return 0;
508
509 short_packet:
510 LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: %d/%u\n",
511 proto == IPPROTO_UDPLITE ? "-Lite" : "",
512 ulen, skb->len);
513
514 discard:
515 UDP6_INC_STATS_BH(UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
516 kfree_skb(skb);
517 return 0;
518 }
519
520 static __inline__ int udpv6_rcv(struct sk_buff *skb)
521 {
522 return __udp6_lib_rcv(skb, udp_hash, IPPROTO_UDP);
523 }
524
525 /*
526 * Throw away all pending data and cancel the corking. Socket is locked.
527 */
528 static void udp_v6_flush_pending_frames(struct sock *sk)
529 {
530 struct udp_sock *up = udp_sk(sk);
531
532 if (up->pending) {
533 up->len = 0;
534 up->pending = 0;
535 ip6_flush_pending_frames(sk);
536 }
537 }
538
539 /*
540 * Sending
541 */
542
543 static int udp_v6_push_pending_frames(struct sock *sk)
544 {
545 struct sk_buff *skb;
546 struct udphdr *uh;
547 struct udp_sock *up = udp_sk(sk);
548 struct inet_sock *inet = inet_sk(sk);
549 struct flowi *fl = &inet->cork.fl;
550 int err = 0;
551 int is_udplite = IS_UDPLITE(sk);
552 __wsum csum = 0;
553
554 /* Grab the skbuff where UDP header space exists. */
555 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
556 goto out;
557
558 /*
559 * Create a UDP header
560 */
561 uh = udp_hdr(skb);
562 uh->source = fl->fl_ip_sport;
563 uh->dest = fl->fl_ip_dport;
564 uh->len = htons(up->len);
565 uh->check = 0;
566
567 if (is_udplite)
568 csum = udplite_csum_outgoing(sk, skb);
569 else
570 csum = udp_csum_outgoing(sk, skb);
571
572 /* add protocol-dependent pseudo-header */
573 uh->check = csum_ipv6_magic(&fl->fl6_src, &fl->fl6_dst,
574 up->len, fl->proto, csum );
575 if (uh->check == 0)
576 uh->check = CSUM_MANGLED_0;
577
578 err = ip6_push_pending_frames(sk);
579 out:
580 up->len = 0;
581 up->pending = 0;
582 if (!err)
583 UDP6_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS, is_udplite);
584 return err;
585 }
586
587 int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
588 struct msghdr *msg, size_t len)
589 {
590 struct ipv6_txoptions opt_space;
591 struct udp_sock *up = udp_sk(sk);
592 struct inet_sock *inet = inet_sk(sk);
593 struct ipv6_pinfo *np = inet6_sk(sk);
594 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) msg->msg_name;
595 struct in6_addr *daddr, *final_p = NULL, final;
596 struct ipv6_txoptions *opt = NULL;
597 struct ip6_flowlabel *flowlabel = NULL;
598 struct flowi fl;
599 struct dst_entry *dst;
600 int addr_len = msg->msg_namelen;
601 int ulen = len;
602 int hlimit = -1;
603 int tclass = -1;
604 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
605 int err;
606 int connected = 0;
607 int is_udplite = IS_UDPLITE(sk);
608 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
609
610 /* destination address check */
611 if (sin6) {
612 if (addr_len < offsetof(struct sockaddr, sa_data))
613 return -EINVAL;
614
615 switch (sin6->sin6_family) {
616 case AF_INET6:
617 if (addr_len < SIN6_LEN_RFC2133)
618 return -EINVAL;
619 daddr = &sin6->sin6_addr;
620 break;
621 case AF_INET:
622 goto do_udp_sendmsg;
623 case AF_UNSPEC:
624 msg->msg_name = sin6 = NULL;
625 msg->msg_namelen = addr_len = 0;
626 daddr = NULL;
627 break;
628 default:
629 return -EINVAL;
630 }
631 } else if (!up->pending) {
632 if (sk->sk_state != TCP_ESTABLISHED)
633 return -EDESTADDRREQ;
634 daddr = &np->daddr;
635 } else
636 daddr = NULL;
637
638 if (daddr) {
639 if (ipv6_addr_v4mapped(daddr)) {
640 struct sockaddr_in sin;
641 sin.sin_family = AF_INET;
642 sin.sin_port = sin6 ? sin6->sin6_port : inet->dport;
643 sin.sin_addr.s_addr = daddr->s6_addr32[3];
644 msg->msg_name = &sin;
645 msg->msg_namelen = sizeof(sin);
646 do_udp_sendmsg:
647 if (__ipv6_only_sock(sk))
648 return -ENETUNREACH;
649 return udp_sendmsg(iocb, sk, msg, len);
650 }
651 }
652
653 if (up->pending == AF_INET)
654 return udp_sendmsg(iocb, sk, msg, len);
655
656 /* Rough check on arithmetic overflow,
657 better check is made in ip6_append_data().
658 */
659 if (len > INT_MAX - sizeof(struct udphdr))
660 return -EMSGSIZE;
661
662 if (up->pending) {
663 /*
664 * There are pending frames.
665 * The socket lock must be held while it's corked.
666 */
667 lock_sock(sk);
668 if (likely(up->pending)) {
669 if (unlikely(up->pending != AF_INET6)) {
670 release_sock(sk);
671 return -EAFNOSUPPORT;
672 }
673 dst = NULL;
674 goto do_append_data;
675 }
676 release_sock(sk);
677 }
678 ulen += sizeof(struct udphdr);
679
680 memset(&fl, 0, sizeof(fl));
681
682 if (sin6) {
683 if (sin6->sin6_port == 0)
684 return -EINVAL;
685
686 fl.fl_ip_dport = sin6->sin6_port;
687 daddr = &sin6->sin6_addr;
688
689 if (np->sndflow) {
690 fl.fl6_flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
691 if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
692 flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
693 if (flowlabel == NULL)
694 return -EINVAL;
695 daddr = &flowlabel->dst;
696 }
697 }
698
699 /*
700 * Otherwise it will be difficult to maintain
701 * sk->sk_dst_cache.
702 */
703 if (sk->sk_state == TCP_ESTABLISHED &&
704 ipv6_addr_equal(daddr, &np->daddr))
705 daddr = &np->daddr;
706
707 if (addr_len >= sizeof(struct sockaddr_in6) &&
708 sin6->sin6_scope_id &&
709 ipv6_addr_type(daddr)&IPV6_ADDR_LINKLOCAL)
710 fl.oif = sin6->sin6_scope_id;
711 } else {
712 if (sk->sk_state != TCP_ESTABLISHED)
713 return -EDESTADDRREQ;
714
715 fl.fl_ip_dport = inet->dport;
716 daddr = &np->daddr;
717 fl.fl6_flowlabel = np->flow_label;
718 connected = 1;
719 }
720
721 if (!fl.oif)
722 fl.oif = sk->sk_bound_dev_if;
723
724 if (msg->msg_controllen) {
725 opt = &opt_space;
726 memset(opt, 0, sizeof(struct ipv6_txoptions));
727 opt->tot_len = sizeof(*opt);
728
729 err = datagram_send_ctl(msg, &fl, opt, &hlimit, &tclass);
730 if (err < 0) {
731 fl6_sock_release(flowlabel);
732 return err;
733 }
734 if ((fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
735 flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
736 if (flowlabel == NULL)
737 return -EINVAL;
738 }
739 if (!(opt->opt_nflen|opt->opt_flen))
740 opt = NULL;
741 connected = 0;
742 }
743 if (opt == NULL)
744 opt = np->opt;
745 if (flowlabel)
746 opt = fl6_merge_options(&opt_space, flowlabel, opt);
747 opt = ipv6_fixup_options(&opt_space, opt);
748
749 fl.proto = sk->sk_protocol;
750 ipv6_addr_copy(&fl.fl6_dst, daddr);
751 if (ipv6_addr_any(&fl.fl6_src) && !ipv6_addr_any(&np->saddr))
752 ipv6_addr_copy(&fl.fl6_src, &np->saddr);
753 fl.fl_ip_sport = inet->sport;
754
755 /* merge ip6_build_xmit from ip6_output */
756 if (opt && opt->srcrt) {
757 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
758 ipv6_addr_copy(&final, &fl.fl6_dst);
759 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
760 final_p = &final;
761 connected = 0;
762 }
763
764 if (!fl.oif && ipv6_addr_is_multicast(&fl.fl6_dst)) {
765 fl.oif = np->mcast_oif;
766 connected = 0;
767 }
768
769 security_sk_classify_flow(sk, &fl);
770
771 err = ip6_sk_dst_lookup(sk, &dst, &fl);
772 if (err)
773 goto out;
774 if (final_p)
775 ipv6_addr_copy(&fl.fl6_dst, final_p);
776
777 if ((err = __xfrm_lookup(&dst, &fl, sk, XFRM_LOOKUP_WAIT)) < 0) {
778 if (err == -EREMOTE)
779 err = ip6_dst_blackhole(sk, &dst, &fl);
780 if (err < 0)
781 goto out;
782 }
783
784 if (hlimit < 0) {
785 if (ipv6_addr_is_multicast(&fl.fl6_dst))
786 hlimit = np->mcast_hops;
787 else
788 hlimit = np->hop_limit;
789 if (hlimit < 0)
790 hlimit = dst_metric(dst, RTAX_HOPLIMIT);
791 if (hlimit < 0)
792 hlimit = ipv6_get_hoplimit(dst->dev);
793 }
794
795 if (tclass < 0) {
796 tclass = np->tclass;
797 if (tclass < 0)
798 tclass = 0;
799 }
800
801 if (msg->msg_flags&MSG_CONFIRM)
802 goto do_confirm;
803 back_from_confirm:
804
805 lock_sock(sk);
806 if (unlikely(up->pending)) {
807 /* The socket is already corked while preparing it. */
808 /* ... which is an evident application bug. --ANK */
809 release_sock(sk);
810
811 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
812 err = -EINVAL;
813 goto out;
814 }
815
816 up->pending = AF_INET6;
817
818 do_append_data:
819 up->len += ulen;
820 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
821 err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen,
822 sizeof(struct udphdr), hlimit, tclass, opt, &fl,
823 (struct rt6_info*)dst,
824 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
825 if (err)
826 udp_v6_flush_pending_frames(sk);
827 else if (!corkreq)
828 err = udp_v6_push_pending_frames(sk);
829 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
830 up->pending = 0;
831
832 if (dst) {
833 if (connected) {
834 ip6_dst_store(sk, dst,
835 ipv6_addr_equal(&fl.fl6_dst, &np->daddr) ?
836 &np->daddr : NULL,
837 #ifdef CONFIG_IPV6_SUBTREES
838 ipv6_addr_equal(&fl.fl6_src, &np->saddr) ?
839 &np->saddr :
840 #endif
841 NULL);
842 } else {
843 dst_release(dst);
844 }
845 }
846
847 if (err > 0)
848 err = np->recverr ? net_xmit_errno(err) : 0;
849 release_sock(sk);
850 out:
851 fl6_sock_release(flowlabel);
852 if (!err)
853 return len;
854 /*
855 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
856 * ENOBUFS might not be good (it's not tunable per se), but otherwise
857 * we don't have a good statistic (IpOutDiscards but it can be too many
858 * things). We could add another new stat but at least for now that
859 * seems like overkill.
860 */
861 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
862 UDP6_INC_STATS_USER(UDP_MIB_SNDBUFERRORS, is_udplite);
863 }
864 return err;
865
866 do_confirm:
867 dst_confirm(dst);
868 if (!(msg->msg_flags&MSG_PROBE) || len)
869 goto back_from_confirm;
870 err = 0;
871 goto out;
872 }
873
874 int udpv6_destroy_sock(struct sock *sk)
875 {
876 lock_sock(sk);
877 udp_v6_flush_pending_frames(sk);
878 release_sock(sk);
879
880 inet6_destroy_sock(sk);
881
882 return 0;
883 }
884
885 /*
886 * Socket option code for UDP
887 */
888 int udpv6_setsockopt(struct sock *sk, int level, int optname,
889 char __user *optval, int optlen)
890 {
891 if (level == SOL_UDP || level == SOL_UDPLITE)
892 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
893 udp_v6_push_pending_frames);
894 return ipv6_setsockopt(sk, level, optname, optval, optlen);
895 }
896
897 #ifdef CONFIG_COMPAT
898 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
899 char __user *optval, int optlen)
900 {
901 if (level == SOL_UDP || level == SOL_UDPLITE)
902 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
903 udp_v6_push_pending_frames);
904 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
905 }
906 #endif
907
908 int udpv6_getsockopt(struct sock *sk, int level, int optname,
909 char __user *optval, int __user *optlen)
910 {
911 if (level == SOL_UDP || level == SOL_UDPLITE)
912 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
913 return ipv6_getsockopt(sk, level, optname, optval, optlen);
914 }
915
916 #ifdef CONFIG_COMPAT
917 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
918 char __user *optval, int __user *optlen)
919 {
920 if (level == SOL_UDP || level == SOL_UDPLITE)
921 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
922 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
923 }
924 #endif
925
926 static struct inet6_protocol udpv6_protocol = {
927 .handler = udpv6_rcv,
928 .err_handler = udpv6_err,
929 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
930 };
931
932 /* ------------------------------------------------------------------------ */
933 #ifdef CONFIG_PROC_FS
934
935 static void udp6_sock_seq_show(struct seq_file *seq, struct sock *sp, int bucket)
936 {
937 struct inet_sock *inet = inet_sk(sp);
938 struct ipv6_pinfo *np = inet6_sk(sp);
939 struct in6_addr *dest, *src;
940 __u16 destp, srcp;
941
942 dest = &np->daddr;
943 src = &np->rcv_saddr;
944 destp = ntohs(inet->dport);
945 srcp = ntohs(inet->sport);
946 seq_printf(seq,
947 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
948 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p\n",
949 bucket,
950 src->s6_addr32[0], src->s6_addr32[1],
951 src->s6_addr32[2], src->s6_addr32[3], srcp,
952 dest->s6_addr32[0], dest->s6_addr32[1],
953 dest->s6_addr32[2], dest->s6_addr32[3], destp,
954 sp->sk_state,
955 atomic_read(&sp->sk_wmem_alloc),
956 atomic_read(&sp->sk_rmem_alloc),
957 0, 0L, 0,
958 sock_i_uid(sp), 0,
959 sock_i_ino(sp),
960 atomic_read(&sp->sk_refcnt), sp);
961 }
962
963 int udp6_seq_show(struct seq_file *seq, void *v)
964 {
965 if (v == SEQ_START_TOKEN)
966 seq_printf(seq,
967 " sl "
968 "local_address "
969 "remote_address "
970 "st tx_queue rx_queue tr tm->when retrnsmt"
971 " uid timeout inode\n");
972 else
973 udp6_sock_seq_show(seq, v, ((struct udp_iter_state *)seq->private)->bucket);
974 return 0;
975 }
976
977 static struct file_operations udp6_seq_fops;
978 static struct udp_seq_afinfo udp6_seq_afinfo = {
979 .owner = THIS_MODULE,
980 .name = "udp6",
981 .family = AF_INET6,
982 .hashtable = udp_hash,
983 .seq_show = udp6_seq_show,
984 .seq_fops = &udp6_seq_fops,
985 };
986
987 int __init udp6_proc_init(void)
988 {
989 return udp_proc_register(&udp6_seq_afinfo);
990 }
991
992 void udp6_proc_exit(void) {
993 udp_proc_unregister(&udp6_seq_afinfo);
994 }
995 #endif /* CONFIG_PROC_FS */
996
997 /* ------------------------------------------------------------------------ */
998
999 DEFINE_PROTO_INUSE(udpv6)
1000
1001 struct proto udpv6_prot = {
1002 .name = "UDPv6",
1003 .owner = THIS_MODULE,
1004 .close = udp_lib_close,
1005 .connect = ip6_datagram_connect,
1006 .disconnect = udp_disconnect,
1007 .ioctl = udp_ioctl,
1008 .destroy = udpv6_destroy_sock,
1009 .setsockopt = udpv6_setsockopt,
1010 .getsockopt = udpv6_getsockopt,
1011 .sendmsg = udpv6_sendmsg,
1012 .recvmsg = udpv6_recvmsg,
1013 .backlog_rcv = udpv6_queue_rcv_skb,
1014 .hash = udp_lib_hash,
1015 .unhash = udp_lib_unhash,
1016 .get_port = udp_v6_get_port,
1017 .memory_allocated = &udp_memory_allocated,
1018 .sysctl_mem = sysctl_udp_mem,
1019 .sysctl_wmem = &sysctl_udp_wmem_min,
1020 .sysctl_rmem = &sysctl_udp_rmem_min,
1021 .obj_size = sizeof(struct udp6_sock),
1022 #ifdef CONFIG_COMPAT
1023 .compat_setsockopt = compat_udpv6_setsockopt,
1024 .compat_getsockopt = compat_udpv6_getsockopt,
1025 #endif
1026 REF_PROTO_INUSE(udpv6)
1027 };
1028
1029 static struct inet_protosw udpv6_protosw = {
1030 .type = SOCK_DGRAM,
1031 .protocol = IPPROTO_UDP,
1032 .prot = &udpv6_prot,
1033 .ops = &inet6_dgram_ops,
1034 .capability =-1,
1035 .no_check = UDP_CSUM_DEFAULT,
1036 .flags = INET_PROTOSW_PERMANENT,
1037 };
1038
1039
1040 int __init udpv6_init(void)
1041 {
1042 int ret;
1043
1044 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1045 if (ret)
1046 goto out;
1047
1048 ret = inet6_register_protosw(&udpv6_protosw);
1049 if (ret)
1050 goto out_udpv6_protocol;
1051 out:
1052 return ret;
1053
1054 out_udpv6_protocol:
1055 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1056 goto out;
1057 }
1058
1059 void udpv6_exit(void)
1060 {
1061 inet6_unregister_protosw(&udpv6_protosw);
1062 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1063 }
Stable & featurefull patchset based on Linus's git branch