?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
IB/ehca: Make scaling code work without CPU hotplug
[linux-2.6/libata-dev.git] / net / ipv4 / udp.c
blobfc620a7c1db4d18027ade93bc04c1e9de750f4c7
1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * The User Datagram Protocol (UDP).
7 *
8 * Version: $Id: udp.c,v 1.102 2002/02/01 22:01:04 davem Exp $
9 *
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
13 * Alan Cox, <Alan.Cox@linux.org>
14 * Hirokazu Takahashi, <taka@valinux.co.jp>
15 *
16 * Fixes:
17 * Alan Cox : verify_area() calls
18 * Alan Cox : stopped close while in use off icmp
19 * messages. Not a fix but a botch that
20 * for udp at least is 'valid'.
21 * Alan Cox : Fixed icmp handling properly
22 * Alan Cox : Correct error for oversized datagrams
23 * Alan Cox : Tidied select() semantics.
24 * Alan Cox : udp_err() fixed properly, also now
25 * select and read wake correctly on errors
26 * Alan Cox : udp_send verify_area moved to avoid mem leak
27 * Alan Cox : UDP can count its memory
28 * Alan Cox : send to an unknown connection causes
29 * an ECONNREFUSED off the icmp, but
30 * does NOT close.
31 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
32 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
33 * bug no longer crashes it.
34 * Fred Van Kempen : Net2e support for sk->broadcast.
35 * Alan Cox : Uses skb_free_datagram
36 * Alan Cox : Added get/set sockopt support.
37 * Alan Cox : Broadcasting without option set returns EACCES.
38 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
39 * Alan Cox : Use ip_tos and ip_ttl
40 * Alan Cox : SNMP Mibs
41 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
42 * Matt Dillon : UDP length checks.
43 * Alan Cox : Smarter af_inet used properly.
44 * Alan Cox : Use new kernel side addressing.
45 * Alan Cox : Incorrect return on truncated datagram receive.
46 * Arnt Gulbrandsen : New udp_send and stuff
47 * Alan Cox : Cache last socket
48 * Alan Cox : Route cache
49 * Jon Peatfield : Minor efficiency fix to sendto().
50 * Mike Shaver : RFC1122 checks.
51 * Alan Cox : Nonblocking error fix.
52 * Willy Konynenberg : Transparent proxying support.
53 * Mike McLagan : Routing by source
54 * David S. Miller : New socket lookup architecture.
55 * Last socket cache retained as it
56 * does have a high hit rate.
57 * Olaf Kirch : Don't linearise iovec on sendmsg.
58 * Andi Kleen : Some cleanups, cache destination entry
59 * for connect.
60 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
61 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
62 * return ENOTCONN for unconnected sockets (POSIX)
63 * Janos Farkas : don't deliver multi/broadcasts to a different
64 * bound-to-device socket
65 * Hirokazu Takahashi : HW checksumming for outgoing UDP
66 * datagrams.
67 * Hirokazu Takahashi : sendfile() on UDP works now.
68 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
69 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
70 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
71 * a single port at the same time.
72 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
73 *
74 *
75 * This program is free software; you can redistribute it and/or
76 * modify it under the terms of the GNU General Public License
77 * as published by the Free Software Foundation; either version
78 * 2 of the License, or (at your option) any later version.
79 */
80
81 #include <asm/system.h>
82 #include <asm/uaccess.h>
83 #include <asm/ioctls.h>
84 #include <linux/types.h>
85 #include <linux/fcntl.h>
86 #include <linux/module.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/igmp.h>
90 #include <linux/in.h>
91 #include <linux/errno.h>
92 #include <linux/timer.h>
93 #include <linux/mm.h>
94 #include <linux/inet.h>
95 #include <linux/netdevice.h>
96 #include <net/tcp_states.h>
97 #include <linux/skbuff.h>
98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h>
100 #include <net/icmp.h>
101 #include <net/route.h>
102 #include <net/checksum.h>
103 #include <net/xfrm.h>
104 #include "udp_impl.h"
105
106 /*
107 * Snmp MIB for the UDP layer
108 */
109
110 DEFINE_SNMP_STAT(struct udp_mib, udp_statistics) __read_mostly;
111
112 struct hlist_head udp_hash[UDP_HTABLE_SIZE];
113 DEFINE_RWLOCK(udp_hash_lock);
114
115 static int udp_port_rover;
116
117 static inline int __udp_lib_lport_inuse(__u16 num, struct hlist_head udptable[])
118 {
119 struct sock *sk;
120 struct hlist_node *node;
121
122 sk_for_each(sk, node, &udptable[num & (UDP_HTABLE_SIZE - 1)])
123 if (sk->sk_hash == num)
124 return 1;
125 return 0;
126 }
127
128 /**
129 * __udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
130 *
131 * @sk: socket struct in question
132 * @snum: port number to look up
133 * @udptable: hash list table, must be of UDP_HTABLE_SIZE
134 * @port_rover: pointer to record of last unallocated port
135 * @saddr_comp: AF-dependent comparison of bound local IP addresses
136 */
137 int __udp_lib_get_port(struct sock *sk, unsigned short snum,
138 struct hlist_head udptable[], int *port_rover,
139 int (*saddr_comp)(const struct sock *sk1,
140 const struct sock *sk2 ) )
141 {
142 struct hlist_node *node;
143 struct hlist_head *head;
144 struct sock *sk2;
145 int error = 1;
146
147 write_lock_bh(&udp_hash_lock);
148 if (snum == 0) {
149 int best_size_so_far, best, result, i;
150
151 if (*port_rover > sysctl_local_port_range[1] ||
152 *port_rover < sysctl_local_port_range[0])
153 *port_rover = sysctl_local_port_range[0];
154 best_size_so_far = 32767;
155 best = result = *port_rover;
156 for (i = 0; i < UDP_HTABLE_SIZE; i++, result++) {
157 int size;
158
159 head = &udptable[result & (UDP_HTABLE_SIZE - 1)];
160 if (hlist_empty(head)) {
161 if (result > sysctl_local_port_range[1])
162 result = sysctl_local_port_range[0] +
163 ((result - sysctl_local_port_range[0]) &
164 (UDP_HTABLE_SIZE - 1));
165 goto gotit;
166 }
167 size = 0;
168 sk_for_each(sk2, node, head) {
169 if (++size >= best_size_so_far)
170 goto next;
171 }
172 best_size_so_far = size;
173 best = result;
174 next:
175 ;
176 }
177 result = best;
178 for(i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; i++, result += UDP_HTABLE_SIZE) {
179 if (result > sysctl_local_port_range[1])
180 result = sysctl_local_port_range[0]
181 + ((result - sysctl_local_port_range[0]) &
182 (UDP_HTABLE_SIZE - 1));
183 if (! __udp_lib_lport_inuse(result, udptable))
184 break;
185 }
186 if (i >= (1 << 16) / UDP_HTABLE_SIZE)
187 goto fail;
188 gotit:
189 *port_rover = snum = result;
190 } else {
191 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
192
193 sk_for_each(sk2, node, head)
194 if (sk2->sk_hash == snum &&
195 sk2 != sk &&
196 (!sk2->sk_reuse || !sk->sk_reuse) &&
197 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if
198 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
199 (*saddr_comp)(sk, sk2) )
200 goto fail;
201 }
202 inet_sk(sk)->num = snum;
203 sk->sk_hash = snum;
204 if (sk_unhashed(sk)) {
205 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
206 sk_add_node(sk, head);
207 sock_prot_inc_use(sk->sk_prot);
208 }
209 error = 0;
210 fail:
211 write_unlock_bh(&udp_hash_lock);
212 return error;
213 }
214
215 __inline__ int udp_get_port(struct sock *sk, unsigned short snum,
216 int (*scmp)(const struct sock *, const struct sock *))
217 {
218 return __udp_lib_get_port(sk, snum, udp_hash, &udp_port_rover, scmp);
219 }
220
221 inline int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
222 {
223 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
224
225 return ( !ipv6_only_sock(sk2) &&
226 (!inet1->rcv_saddr || !inet2->rcv_saddr ||
227 inet1->rcv_saddr == inet2->rcv_saddr ));
228 }
229
230 static inline int udp_v4_get_port(struct sock *sk, unsigned short snum)
231 {
232 return udp_get_port(sk, snum, ipv4_rcv_saddr_equal);
233 }
234
235 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
236 * harder than this. -DaveM
237 */
238 static struct sock *__udp4_lib_lookup(__be32 saddr, __be16 sport,
239 __be32 daddr, __be16 dport,
240 int dif, struct hlist_head udptable[])
241 {
242 struct sock *sk, *result = NULL;
243 struct hlist_node *node;
244 unsigned short hnum = ntohs(dport);
245 int badness = -1;
246
247 read_lock(&udp_hash_lock);
248 sk_for_each(sk, node, &udptable[hnum & (UDP_HTABLE_SIZE - 1)]) {
249 struct inet_sock *inet = inet_sk(sk);
250
251 if (sk->sk_hash == hnum && !ipv6_only_sock(sk)) {
252 int score = (sk->sk_family == PF_INET ? 1 : 0);
253 if (inet->rcv_saddr) {
254 if (inet->rcv_saddr != daddr)
255 continue;
256 score+=2;
257 }
258 if (inet->daddr) {
259 if (inet->daddr != saddr)
260 continue;
261 score+=2;
262 }
263 if (inet->dport) {
264 if (inet->dport != sport)
265 continue;
266 score+=2;
267 }
268 if (sk->sk_bound_dev_if) {
269 if (sk->sk_bound_dev_if != dif)
270 continue;
271 score+=2;
272 }
273 if(score == 9) {
274 result = sk;
275 break;
276 } else if(score > badness) {
277 result = sk;
278 badness = score;
279 }
280 }
281 }
282 if (result)
283 sock_hold(result);
284 read_unlock(&udp_hash_lock);
285 return result;
286 }
287
288 static inline struct sock *udp_v4_mcast_next(struct sock *sk,
289 __be16 loc_port, __be32 loc_addr,
290 __be16 rmt_port, __be32 rmt_addr,
291 int dif)
292 {
293 struct hlist_node *node;
294 struct sock *s = sk;
295 unsigned short hnum = ntohs(loc_port);
296
297 sk_for_each_from(s, node) {
298 struct inet_sock *inet = inet_sk(s);
299
300 if (s->sk_hash != hnum ||
301 (inet->daddr && inet->daddr != rmt_addr) ||
302 (inet->dport != rmt_port && inet->dport) ||
303 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) ||
304 ipv6_only_sock(s) ||
305 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
306 continue;
307 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
308 continue;
309 goto found;
310 }
311 s = NULL;
312 found:
313 return s;
314 }
315
316 /*
317 * This routine is called by the ICMP module when it gets some
318 * sort of error condition. If err < 0 then the socket should
319 * be closed and the error returned to the user. If err > 0
320 * it's just the icmp type << 8 | icmp code.
321 * Header points to the ip header of the error packet. We move
322 * on past this. Then (as it used to claim before adjustment)
323 * header points to the first 8 bytes of the udp header. We need
324 * to find the appropriate port.
325 */
326
327 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[])
328 {
329 struct inet_sock *inet;
330 struct iphdr *iph = (struct iphdr*)skb->data;
331 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2));
332 int type = skb->h.icmph->type;
333 int code = skb->h.icmph->code;
334 struct sock *sk;
335 int harderr;
336 int err;
337
338 sk = __udp4_lib_lookup(iph->daddr, uh->dest, iph->saddr, uh->source,
339 skb->dev->ifindex, udptable );
340 if (sk == NULL) {
341 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
342 return; /* No socket for error */
343 }
344
345 err = 0;
346 harderr = 0;
347 inet = inet_sk(sk);
348
349 switch (type) {
350 default:
351 case ICMP_TIME_EXCEEDED:
352 err = EHOSTUNREACH;
353 break;
354 case ICMP_SOURCE_QUENCH:
355 goto out;
356 case ICMP_PARAMETERPROB:
357 err = EPROTO;
358 harderr = 1;
359 break;
360 case ICMP_DEST_UNREACH:
361 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
362 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
363 err = EMSGSIZE;
364 harderr = 1;
365 break;
366 }
367 goto out;
368 }
369 err = EHOSTUNREACH;
370 if (code <= NR_ICMP_UNREACH) {
371 harderr = icmp_err_convert[code].fatal;
372 err = icmp_err_convert[code].errno;
373 }
374 break;
375 }
376
377 /*
378 * RFC1122: OK. Passes ICMP errors back to application, as per
379 * 4.1.3.3.
380 */
381 if (!inet->recverr) {
382 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
383 goto out;
384 } else {
385 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1));
386 }
387 sk->sk_err = err;
388 sk->sk_error_report(sk);
389 out:
390 sock_put(sk);
391 }
392
393 __inline__ void udp_err(struct sk_buff *skb, u32 info)
394 {
395 return __udp4_lib_err(skb, info, udp_hash);
396 }
397
398 /*
399 * Throw away all pending data and cancel the corking. Socket is locked.
400 */
401 static void udp_flush_pending_frames(struct sock *sk)
402 {
403 struct udp_sock *up = udp_sk(sk);
404
405 if (up->pending) {
406 up->len = 0;
407 up->pending = 0;
408 ip_flush_pending_frames(sk);
409 }
410 }
411
412 /**
413 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
414 * @sk: socket we are sending on
415 * @skb: sk_buff containing the filled-in UDP header
416 * (checksum field must be zeroed out)
417 */
418 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
419 __be32 src, __be32 dst, int len )
420 {
421 unsigned int offset;
422 struct udphdr *uh = skb->h.uh;
423 __wsum csum = 0;
424
425 if (skb_queue_len(&sk->sk_write_queue) == 1) {
426 /*
427 * Only one fragment on the socket.
428 */
429 skb->csum_offset = offsetof(struct udphdr, check);
430 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
431 } else {
432 /*
433 * HW-checksum won't work as there are two or more
434 * fragments on the socket so that all csums of sk_buffs
435 * should be together
436 */
437 offset = skb->h.raw - skb->data;
438 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
439
440 skb->ip_summed = CHECKSUM_NONE;
441
442 skb_queue_walk(&sk->sk_write_queue, skb) {
443 csum = csum_add(csum, skb->csum);
444 }
445
446 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
447 if (uh->check == 0)
448 uh->check = CSUM_MANGLED_0;
449 }
450 }
451
452 /*
453 * Push out all pending data as one UDP datagram. Socket is locked.
454 */
455 static int udp_push_pending_frames(struct sock *sk)
456 {
457 struct udp_sock *up = udp_sk(sk);
458 struct inet_sock *inet = inet_sk(sk);
459 struct flowi *fl = &inet->cork.fl;
460 struct sk_buff *skb;
461 struct udphdr *uh;
462 int err = 0;
463 __wsum csum = 0;
464
465 /* Grab the skbuff where UDP header space exists. */
466 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
467 goto out;
468
469 /*
470 * Create a UDP header
471 */
472 uh = skb->h.uh;
473 uh->source = fl->fl_ip_sport;
474 uh->dest = fl->fl_ip_dport;
475 uh->len = htons(up->len);
476 uh->check = 0;
477
478 if (up->pcflag) /* UDP-Lite */
479 csum = udplite_csum_outgoing(sk, skb);
480
481 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
482
483 skb->ip_summed = CHECKSUM_NONE;
484 goto send;
485
486 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
487
488 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len);
489 goto send;
490
491 } else /* `normal' UDP */
492 csum = udp_csum_outgoing(sk, skb);
493
494 /* add protocol-dependent pseudo-header */
495 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
496 sk->sk_protocol, csum );
497 if (uh->check == 0)
498 uh->check = CSUM_MANGLED_0;
499
500 send:
501 err = ip_push_pending_frames(sk);
502 out:
503 up->len = 0;
504 up->pending = 0;
505 return err;
506 }
507
508 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
509 size_t len)
510 {
511 struct inet_sock *inet = inet_sk(sk);
512 struct udp_sock *up = udp_sk(sk);
513 int ulen = len;
514 struct ipcm_cookie ipc;
515 struct rtable *rt = NULL;
516 int free = 0;
517 int connected = 0;
518 __be32 daddr, faddr, saddr;
519 __be16 dport;
520 u8 tos;
521 int err, is_udplite = up->pcflag;
522 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
523 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
524
525 if (len > 0xFFFF)
526 return -EMSGSIZE;
527
528 /*
529 * Check the flags.
530 */
531
532 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */
533 return -EOPNOTSUPP;
534
535 ipc.opt = NULL;
536
537 if (up->pending) {
538 /*
539 * There are pending frames.
540 * The socket lock must be held while it's corked.
541 */
542 lock_sock(sk);
543 if (likely(up->pending)) {
544 if (unlikely(up->pending != AF_INET)) {
545 release_sock(sk);
546 return -EINVAL;
547 }
548 goto do_append_data;
549 }
550 release_sock(sk);
551 }
552 ulen += sizeof(struct udphdr);
553
554 /*
555 * Get and verify the address.
556 */
557 if (msg->msg_name) {
558 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name;
559 if (msg->msg_namelen < sizeof(*usin))
560 return -EINVAL;
561 if (usin->sin_family != AF_INET) {
562 if (usin->sin_family != AF_UNSPEC)
563 return -EAFNOSUPPORT;
564 }
565
566 daddr = usin->sin_addr.s_addr;
567 dport = usin->sin_port;
568 if (dport == 0)
569 return -EINVAL;
570 } else {
571 if (sk->sk_state != TCP_ESTABLISHED)
572 return -EDESTADDRREQ;
573 daddr = inet->daddr;
574 dport = inet->dport;
575 /* Open fast path for connected socket.
576 Route will not be used, if at least one option is set.
577 */
578 connected = 1;
579 }
580 ipc.addr = inet->saddr;
581
582 ipc.oif = sk->sk_bound_dev_if;
583 if (msg->msg_controllen) {
584 err = ip_cmsg_send(msg, &ipc);
585 if (err)
586 return err;
587 if (ipc.opt)
588 free = 1;
589 connected = 0;
590 }
591 if (!ipc.opt)
592 ipc.opt = inet->opt;
593
594 saddr = ipc.addr;
595 ipc.addr = faddr = daddr;
596
597 if (ipc.opt && ipc.opt->srr) {
598 if (!daddr)
599 return -EINVAL;
600 faddr = ipc.opt->faddr;
601 connected = 0;
602 }
603 tos = RT_TOS(inet->tos);
604 if (sock_flag(sk, SOCK_LOCALROUTE) ||
605 (msg->msg_flags & MSG_DONTROUTE) ||
606 (ipc.opt && ipc.opt->is_strictroute)) {
607 tos |= RTO_ONLINK;
608 connected = 0;
609 }
610
611 if (MULTICAST(daddr)) {
612 if (!ipc.oif)
613 ipc.oif = inet->mc_index;
614 if (!saddr)
615 saddr = inet->mc_addr;
616 connected = 0;
617 }
618
619 if (connected)
620 rt = (struct rtable*)sk_dst_check(sk, 0);
621
622 if (rt == NULL) {
623 struct flowi fl = { .oif = ipc.oif,
624 .nl_u = { .ip4_u =
625 { .daddr = faddr,
626 .saddr = saddr,
627 .tos = tos } },
628 .proto = sk->sk_protocol,
629 .uli_u = { .ports =
630 { .sport = inet->sport,
631 .dport = dport } } };
632 security_sk_classify_flow(sk, &fl);
633 err = ip_route_output_flow(&rt, &fl, sk, 1);
634 if (err)
635 goto out;
636
637 err = -EACCES;
638 if ((rt->rt_flags & RTCF_BROADCAST) &&
639 !sock_flag(sk, SOCK_BROADCAST))
640 goto out;
641 if (connected)
642 sk_dst_set(sk, dst_clone(&rt->u.dst));
643 }
644
645 if (msg->msg_flags&MSG_CONFIRM)
646 goto do_confirm;
647 back_from_confirm:
648
649 saddr = rt->rt_src;
650 if (!ipc.addr)
651 daddr = ipc.addr = rt->rt_dst;
652
653 lock_sock(sk);
654 if (unlikely(up->pending)) {
655 /* The socket is already corked while preparing it. */
656 /* ... which is an evident application bug. --ANK */
657 release_sock(sk);
658
659 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
660 err = -EINVAL;
661 goto out;
662 }
663 /*
664 * Now cork the socket to pend data.
665 */
666 inet->cork.fl.fl4_dst = daddr;
667 inet->cork.fl.fl_ip_dport = dport;
668 inet->cork.fl.fl4_src = saddr;
669 inet->cork.fl.fl_ip_sport = inet->sport;
670 up->pending = AF_INET;
671
672 do_append_data:
673 up->len += ulen;
674 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
675 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
676 sizeof(struct udphdr), &ipc, rt,
677 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
678 if (err)
679 udp_flush_pending_frames(sk);
680 else if (!corkreq)
681 err = udp_push_pending_frames(sk);
682 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
683 up->pending = 0;
684 release_sock(sk);
685
686 out:
687 ip_rt_put(rt);
688 if (free)
689 kfree(ipc.opt);
690 if (!err) {
691 UDP_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS, is_udplite);
692 return len;
693 }
694 /*
695 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
696 * ENOBUFS might not be good (it's not tunable per se), but otherwise
697 * we don't have a good statistic (IpOutDiscards but it can be too many
698 * things). We could add another new stat but at least for now that
699 * seems like overkill.
700 */
701 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
702 UDP_INC_STATS_USER(UDP_MIB_SNDBUFERRORS, is_udplite);
703 }
704 return err;
705
706 do_confirm:
707 dst_confirm(&rt->u.dst);
708 if (!(msg->msg_flags&MSG_PROBE) || len)
709 goto back_from_confirm;
710 err = 0;
711 goto out;
712 }
713
714 int udp_sendpage(struct sock *sk, struct page *page, int offset,
715 size_t size, int flags)
716 {
717 struct udp_sock *up = udp_sk(sk);
718 int ret;
719
720 if (!up->pending) {
721 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
722
723 /* Call udp_sendmsg to specify destination address which
724 * sendpage interface can't pass.
725 * This will succeed only when the socket is connected.
726 */
727 ret = udp_sendmsg(NULL, sk, &msg, 0);
728 if (ret < 0)
729 return ret;
730 }
731
732 lock_sock(sk);
733
734 if (unlikely(!up->pending)) {
735 release_sock(sk);
736
737 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
738 return -EINVAL;
739 }
740
741 ret = ip_append_page(sk, page, offset, size, flags);
742 if (ret == -EOPNOTSUPP) {
743 release_sock(sk);
744 return sock_no_sendpage(sk->sk_socket, page, offset,
745 size, flags);
746 }
747 if (ret < 0) {
748 udp_flush_pending_frames(sk);
749 goto out;
750 }
751
752 up->len += size;
753 if (!(up->corkflag || (flags&MSG_MORE)))
754 ret = udp_push_pending_frames(sk);
755 if (!ret)
756 ret = size;
757 out:
758 release_sock(sk);
759 return ret;
760 }
761
762 /*
763 * IOCTL requests applicable to the UDP protocol
764 */
765
766 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
767 {
768 switch(cmd)
769 {
770 case SIOCOUTQ:
771 {
772 int amount = atomic_read(&sk->sk_wmem_alloc);
773 return put_user(amount, (int __user *)arg);
774 }
775
776 case SIOCINQ:
777 {
778 struct sk_buff *skb;
779 unsigned long amount;
780
781 amount = 0;
782 spin_lock_bh(&sk->sk_receive_queue.lock);
783 skb = skb_peek(&sk->sk_receive_queue);
784 if (skb != NULL) {
785 /*
786 * We will only return the amount
787 * of this packet since that is all
788 * that will be read.
789 */
790 amount = skb->len - sizeof(struct udphdr);
791 }
792 spin_unlock_bh(&sk->sk_receive_queue.lock);
793 return put_user(amount, (int __user *)arg);
794 }
795
796 default:
797 return -ENOIOCTLCMD;
798 }
799 return(0);
800 }
801
802 /*
803 * This should be easy, if there is something there we
804 * return it, otherwise we block.
805 */
806
807 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
808 size_t len, int noblock, int flags, int *addr_len)
809 {
810 struct inet_sock *inet = inet_sk(sk);
811 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
812 struct sk_buff *skb;
813 int copied, err, copy_only, is_udplite = IS_UDPLITE(sk);
814
815 /*
816 * Check any passed addresses
817 */
818 if (addr_len)
819 *addr_len=sizeof(*sin);
820
821 if (flags & MSG_ERRQUEUE)
822 return ip_recv_error(sk, msg, len);
823
824 try_again:
825 skb = skb_recv_datagram(sk, flags, noblock, &err);
826 if (!skb)
827 goto out;
828
829 copied = skb->len - sizeof(struct udphdr);
830 if (copied > len) {
831 copied = len;
832 msg->msg_flags |= MSG_TRUNC;
833 }
834
835 /*
836 * Decide whether to checksum and/or copy data.
837 *
838 * UDP: checksum may have been computed in HW,
839 * (re-)compute it if message is truncated.
840 * UDP-Lite: always needs to checksum, no HW support.
841 */
842 copy_only = (skb->ip_summed==CHECKSUM_UNNECESSARY);
843
844 if (is_udplite || (!copy_only && msg->msg_flags&MSG_TRUNC)) {
845 if (__udp_lib_checksum_complete(skb))
846 goto csum_copy_err;
847 copy_only = 1;
848 }
849
850 if (copy_only)
851 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
852 msg->msg_iov, copied );
853 else {
854 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
855
856 if (err == -EINVAL)
857 goto csum_copy_err;
858 }
859
860 if (err)
861 goto out_free;
862
863 sock_recv_timestamp(msg, sk, skb);
864
865 /* Copy the address. */
866 if (sin)
867 {
868 sin->sin_family = AF_INET;
869 sin->sin_port = skb->h.uh->source;
870 sin->sin_addr.s_addr = skb->nh.iph->saddr;
871 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
872 }
873 if (inet->cmsg_flags)
874 ip_cmsg_recv(msg, skb);
875
876 err = copied;
877 if (flags & MSG_TRUNC)
878 err = skb->len - sizeof(struct udphdr);
879
880 out_free:
881 skb_free_datagram(sk, skb);
882 out:
883 return err;
884
885 csum_copy_err:
886 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite);
887
888 skb_kill_datagram(sk, skb, flags);
889
890 if (noblock)
891 return -EAGAIN;
892 goto try_again;
893 }
894
895
896 int udp_disconnect(struct sock *sk, int flags)
897 {
898 struct inet_sock *inet = inet_sk(sk);
899 /*
900 * 1003.1g - break association.
901 */
902
903 sk->sk_state = TCP_CLOSE;
904 inet->daddr = 0;
905 inet->dport = 0;
906 sk->sk_bound_dev_if = 0;
907 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
908 inet_reset_saddr(sk);
909
910 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
911 sk->sk_prot->unhash(sk);
912 inet->sport = 0;
913 }
914 sk_dst_reset(sk);
915 return 0;
916 }
917
918 /* return:
919 * 1 if the the UDP system should process it
920 * 0 if we should drop this packet
921 * -1 if it should get processed by xfrm4_rcv_encap
922 */
923 static int udp_encap_rcv(struct sock * sk, struct sk_buff *skb)
924 {
925 #ifndef CONFIG_XFRM
926 return 1;
927 #else
928 struct udp_sock *up = udp_sk(sk);
929 struct udphdr *uh;
930 struct iphdr *iph;
931 int iphlen, len;
932
933 __u8 *udpdata;
934 __be32 *udpdata32;
935 __u16 encap_type = up->encap_type;
936
937 /* if we're overly short, let UDP handle it */
938 len = skb->len - sizeof(struct udphdr);
939 if (len <= 0)
940 return 1;
941
942 /* if this is not encapsulated socket, then just return now */
943 if (!encap_type)
944 return 1;
945
946 /* If this is a paged skb, make sure we pull up
947 * whatever data we need to look at. */
948 if (!pskb_may_pull(skb, sizeof(struct udphdr) + min(len, 8)))
949 return 1;
950
951 /* Now we can get the pointers */
952 uh = skb->h.uh;
953 udpdata = (__u8 *)uh + sizeof(struct udphdr);
954 udpdata32 = (__be32 *)udpdata;
955
956 switch (encap_type) {
957 default:
958 case UDP_ENCAP_ESPINUDP:
959 /* Check if this is a keepalive packet. If so, eat it. */
960 if (len == 1 && udpdata[0] == 0xff) {
961 return 0;
962 } else if (len > sizeof(struct ip_esp_hdr) && udpdata32[0] != 0 ) {
963 /* ESP Packet without Non-ESP header */
964 len = sizeof(struct udphdr);
965 } else
966 /* Must be an IKE packet.. pass it through */
967 return 1;
968 break;
969 case UDP_ENCAP_ESPINUDP_NON_IKE:
970 /* Check if this is a keepalive packet. If so, eat it. */
971 if (len == 1 && udpdata[0] == 0xff) {
972 return 0;
973 } else if (len > 2 * sizeof(u32) + sizeof(struct ip_esp_hdr) &&
974 udpdata32[0] == 0 && udpdata32[1] == 0) {
975
976 /* ESP Packet with Non-IKE marker */
977 len = sizeof(struct udphdr) + 2 * sizeof(u32);
978 } else
979 /* Must be an IKE packet.. pass it through */
980 return 1;
981 break;
982 }
983
984 /* At this point we are sure that this is an ESPinUDP packet,
985 * so we need to remove 'len' bytes from the packet (the UDP
986 * header and optional ESP marker bytes) and then modify the
987 * protocol to ESP, and then call into the transform receiver.
988 */
989 if (skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
990 return 0;
991
992 /* Now we can update and verify the packet length... */
993 iph = skb->nh.iph;
994 iphlen = iph->ihl << 2;
995 iph->tot_len = htons(ntohs(iph->tot_len) - len);
996 if (skb->len < iphlen + len) {
997 /* packet is too small!?! */
998 return 0;
999 }
1000
1001 /* pull the data buffer up to the ESP header and set the
1002 * transport header to point to ESP. Keep UDP on the stack
1003 * for later.
1004 */
1005 skb->h.raw = skb_pull(skb, len);
1006
1007 /* modify the protocol (it's ESP!) */
1008 iph->protocol = IPPROTO_ESP;
1009
1010 /* and let the caller know to send this into the ESP processor... */
1011 return -1;
1012 #endif
1013 }
1014
1015 /* returns:
1016 * -1: error
1017 * 0: success
1018 * >0: "udp encap" protocol resubmission
1019 *
1020 * Note that in the success and error cases, the skb is assumed to
1021 * have either been requeued or freed.
1022 */
1023 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
1024 {
1025 struct udp_sock *up = udp_sk(sk);
1026 int rc;
1027
1028 /*
1029 * Charge it to the socket, dropping if the queue is full.
1030 */
1031 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1032 goto drop;
1033 nf_reset(skb);
1034
1035 if (up->encap_type) {
1036 /*
1037 * This is an encapsulation socket, so let's see if this is
1038 * an encapsulated packet.
1039 * If it's a keepalive packet, then just eat it.
1040 * If it's an encapsulateed packet, then pass it to the
1041 * IPsec xfrm input and return the response
1042 * appropriately. Otherwise, just fall through and
1043 * pass this up the UDP socket.
1044 */
1045 int ret;
1046
1047 ret = udp_encap_rcv(sk, skb);
1048 if (ret == 0) {
1049 /* Eat the packet .. */
1050 kfree_skb(skb);
1051 return 0;
1052 }
1053 if (ret < 0) {
1054 /* process the ESP packet */
1055 ret = xfrm4_rcv_encap(skb, up->encap_type);
1056 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS, up->pcflag);
1057 return -ret;
1058 }
1059 /* FALLTHROUGH -- it's a UDP Packet */
1060 }
1061
1062 /*
1063 * UDP-Lite specific tests, ignored on UDP sockets
1064 */
1065 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1066
1067 /*
1068 * MIB statistics other than incrementing the error count are
1069 * disabled for the following two types of errors: these depend
1070 * on the application settings, not on the functioning of the
1071 * protocol stack as such.
1072 *
1073 * RFC 3828 here recommends (sec 3.3): "There should also be a
1074 * way ... to ... at least let the receiving application block
1075 * delivery of packets with coverage values less than a value
1076 * provided by the application."
1077 */
1078 if (up->pcrlen == 0) { /* full coverage was set */
1079 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1080 "%d while full coverage %d requested\n",
1081 UDP_SKB_CB(skb)->cscov, skb->len);
1082 goto drop;
1083 }
1084 /* The next case involves violating the min. coverage requested
1085 * by the receiver. This is subtle: if receiver wants x and x is
1086 * greater than the buffersize/MTU then receiver will complain
1087 * that it wants x while sender emits packets of smaller size y.
1088 * Therefore the above ...()->partial_cov statement is essential.
1089 */
1090 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1091 LIMIT_NETDEBUG(KERN_WARNING
1092 "UDPLITE: coverage %d too small, need min %d\n",
1093 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1094 goto drop;
1095 }
1096 }
1097
1098 if (sk->sk_filter && skb->ip_summed != CHECKSUM_UNNECESSARY) {
1099 if (__udp_lib_checksum_complete(skb))
1100 goto drop;
1101 skb->ip_summed = CHECKSUM_UNNECESSARY;
1102 }
1103
1104 if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) {
1105 /* Note that an ENOMEM error is charged twice */
1106 if (rc == -ENOMEM)
1107 UDP_INC_STATS_BH(UDP_MIB_RCVBUFERRORS, up->pcflag);
1108 goto drop;
1109 }
1110
1111 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS, up->pcflag);
1112 return 0;
1113
1114 drop:
1115 UDP_INC_STATS_BH(UDP_MIB_INERRORS, up->pcflag);
1116 kfree_skb(skb);
1117 return -1;
1118 }
1119
1120 /*
1121 * Multicasts and broadcasts go to each listener.
1122 *
1123 * Note: called only from the BH handler context,
1124 * so we don't need to lock the hashes.
1125 */
1126 static int __udp4_lib_mcast_deliver(struct sk_buff *skb,
1127 struct udphdr *uh,
1128 __be32 saddr, __be32 daddr,
1129 struct hlist_head udptable[])
1130 {
1131 struct sock *sk;
1132 int dif;
1133
1134 read_lock(&udp_hash_lock);
1135 sk = sk_head(&udptable[ntohs(uh->dest) & (UDP_HTABLE_SIZE - 1)]);
1136 dif = skb->dev->ifindex;
1137 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
1138 if (sk) {
1139 struct sock *sknext = NULL;
1140
1141 do {
1142 struct sk_buff *skb1 = skb;
1143
1144 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr,
1145 uh->source, saddr, dif);
1146 if(sknext)
1147 skb1 = skb_clone(skb, GFP_ATOMIC);
1148
1149 if(skb1) {
1150 int ret = udp_queue_rcv_skb(sk, skb1);
1151 if (ret > 0)
1152 /* we should probably re-process instead
1153 * of dropping packets here. */
1154 kfree_skb(skb1);
1155 }
1156 sk = sknext;
1157 } while(sknext);
1158 } else
1159 kfree_skb(skb);
1160 read_unlock(&udp_hash_lock);
1161 return 0;
1162 }
1163
1164 /* Initialize UDP checksum. If exited with zero value (success),
1165 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1166 * Otherwise, csum completion requires chacksumming packet body,
1167 * including udp header and folding it to skb->csum.
1168 */
1169 static inline void udp4_csum_init(struct sk_buff *skb, struct udphdr *uh)
1170 {
1171 if (uh->check == 0) {
1172 skb->ip_summed = CHECKSUM_UNNECESSARY;
1173 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1174 if (!csum_tcpudp_magic(skb->nh.iph->saddr, skb->nh.iph->daddr,
1175 skb->len, IPPROTO_UDP, skb->csum ))
1176 skb->ip_summed = CHECKSUM_UNNECESSARY;
1177 }
1178 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
1179 skb->csum = csum_tcpudp_nofold(skb->nh.iph->saddr,
1180 skb->nh.iph->daddr,
1181 skb->len, IPPROTO_UDP, 0);
1182 /* Probably, we should checksum udp header (it should be in cache
1183 * in any case) and data in tiny packets (< rx copybreak).
1184 */
1185
1186 /* UDP = UDP-Lite with a non-partial checksum coverage */
1187 UDP_SKB_CB(skb)->partial_cov = 0;
1188 }
1189
1190 /*
1191 * All we need to do is get the socket, and then do a checksum.
1192 */
1193
1194 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
1195 int is_udplite)
1196 {
1197 struct sock *sk;
1198 struct udphdr *uh = skb->h.uh;
1199 unsigned short ulen;
1200 struct rtable *rt = (struct rtable*)skb->dst;
1201 __be32 saddr = skb->nh.iph->saddr;
1202 __be32 daddr = skb->nh.iph->daddr;
1203
1204 /*
1205 * Validate the packet.
1206 */
1207 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1208 goto drop; /* No space for header. */
1209
1210 ulen = ntohs(uh->len);
1211 if (ulen > skb->len)
1212 goto short_packet;
1213
1214 if(! is_udplite ) { /* UDP validates ulen. */
1215
1216 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1217 goto short_packet;
1218 uh = skb->h.uh;
1219
1220 udp4_csum_init(skb, uh);
1221
1222 } else { /* UDP-Lite validates cscov. */
1223 if (udplite4_csum_init(skb, uh))
1224 goto csum_error;
1225 }
1226
1227 if(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1228 return __udp4_lib_mcast_deliver(skb, uh, saddr, daddr, udptable);
1229
1230 sk = __udp4_lib_lookup(saddr, uh->source, daddr, uh->dest,
1231 skb->dev->ifindex, udptable );
1232
1233 if (sk != NULL) {
1234 int ret = udp_queue_rcv_skb(sk, skb);
1235 sock_put(sk);
1236
1237 /* a return value > 0 means to resubmit the input, but
1238 * it wants the return to be -protocol, or 0
1239 */
1240 if (ret > 0)
1241 return -ret;
1242 return 0;
1243 }
1244
1245 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1246 goto drop;
1247 nf_reset(skb);
1248
1249 /* No socket. Drop packet silently, if checksum is wrong */
1250 if (udp_lib_checksum_complete(skb))
1251 goto csum_error;
1252
1253 UDP_INC_STATS_BH(UDP_MIB_NOPORTS, is_udplite);
1254 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1255
1256 /*
1257 * Hmm. We got an UDP packet to a port to which we
1258 * don't wanna listen. Ignore it.
1259 */
1260 kfree_skb(skb);
1261 return(0);
1262
1263 short_packet:
1264 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %u.%u.%u.%u:%u %d/%d to %u.%u.%u.%u:%u\n",
1265 is_udplite? "-Lite" : "",
1266 NIPQUAD(saddr),
1267 ntohs(uh->source),
1268 ulen,
1269 skb->len,
1270 NIPQUAD(daddr),
1271 ntohs(uh->dest));
1272 goto drop;
1273
1274 csum_error:
1275 /*
1276 * RFC1122: OK. Discards the bad packet silently (as far as
1277 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1278 */
1279 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %d.%d.%d.%d:%d to %d.%d.%d.%d:%d ulen %d\n",
1280 is_udplite? "-Lite" : "",
1281 NIPQUAD(saddr),
1282 ntohs(uh->source),
1283 NIPQUAD(daddr),
1284 ntohs(uh->dest),
1285 ulen);
1286 drop:
1287 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite);
1288 kfree_skb(skb);
1289 return(0);
1290 }
1291
1292 __inline__ int udp_rcv(struct sk_buff *skb)
1293 {
1294 return __udp4_lib_rcv(skb, udp_hash, 0);
1295 }
1296
1297 int udp_destroy_sock(struct sock *sk)
1298 {
1299 lock_sock(sk);
1300 udp_flush_pending_frames(sk);
1301 release_sock(sk);
1302 return 0;
1303 }
1304
1305 /*
1306 * Socket option code for UDP
1307 */
1308 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1309 char __user *optval, int optlen,
1310 int (*push_pending_frames)(struct sock *))
1311 {
1312 struct udp_sock *up = udp_sk(sk);
1313 int val;
1314 int err = 0;
1315
1316 if(optlen<sizeof(int))
1317 return -EINVAL;
1318
1319 if (get_user(val, (int __user *)optval))
1320 return -EFAULT;
1321
1322 switch(optname) {
1323 case UDP_CORK:
1324 if (val != 0) {
1325 up->corkflag = 1;
1326 } else {
1327 up->corkflag = 0;
1328 lock_sock(sk);
1329 (*push_pending_frames)(sk);
1330 release_sock(sk);
1331 }
1332 break;
1333
1334 case UDP_ENCAP:
1335 switch (val) {
1336 case 0:
1337 case UDP_ENCAP_ESPINUDP:
1338 case UDP_ENCAP_ESPINUDP_NON_IKE:
1339 up->encap_type = val;
1340 break;
1341 default:
1342 err = -ENOPROTOOPT;
1343 break;
1344 }
1345 break;
1346
1347 /*
1348 * UDP-Lite's partial checksum coverage (RFC 3828).
1349 */
1350 /* The sender sets actual checksum coverage length via this option.
1351 * The case coverage > packet length is handled by send module. */
1352 case UDPLITE_SEND_CSCOV:
1353 if (!up->pcflag) /* Disable the option on UDP sockets */
1354 return -ENOPROTOOPT;
1355 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1356 val = 8;
1357 up->pcslen = val;
1358 up->pcflag |= UDPLITE_SEND_CC;
1359 break;
1360
1361 /* The receiver specifies a minimum checksum coverage value. To make
1362 * sense, this should be set to at least 8 (as done below). If zero is
1363 * used, this again means full checksum coverage. */
1364 case UDPLITE_RECV_CSCOV:
1365 if (!up->pcflag) /* Disable the option on UDP sockets */
1366 return -ENOPROTOOPT;
1367 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1368 val = 8;
1369 up->pcrlen = val;
1370 up->pcflag |= UDPLITE_RECV_CC;
1371 break;
1372
1373 default:
1374 err = -ENOPROTOOPT;
1375 break;
1376 };
1377
1378 return err;
1379 }
1380
1381 int udp_setsockopt(struct sock *sk, int level, int optname,
1382 char __user *optval, int optlen)
1383 {
1384 if (level == SOL_UDP || level == SOL_UDPLITE)
1385 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1386 udp_push_pending_frames);
1387 return ip_setsockopt(sk, level, optname, optval, optlen);
1388 }
1389
1390 #ifdef CONFIG_COMPAT
1391 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1392 char __user *optval, int optlen)
1393 {
1394 if (level == SOL_UDP || level == SOL_UDPLITE)
1395 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1396 udp_push_pending_frames);
1397 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1398 }
1399 #endif
1400
1401 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1402 char __user *optval, int __user *optlen)
1403 {
1404 struct udp_sock *up = udp_sk(sk);
1405 int val, len;
1406
1407 if(get_user(len,optlen))
1408 return -EFAULT;
1409
1410 len = min_t(unsigned int, len, sizeof(int));
1411
1412 if(len < 0)
1413 return -EINVAL;
1414
1415 switch(optname) {
1416 case UDP_CORK:
1417 val = up->corkflag;
1418 break;
1419
1420 case UDP_ENCAP:
1421 val = up->encap_type;
1422 break;
1423
1424 /* The following two cannot be changed on UDP sockets, the return is
1425 * always 0 (which corresponds to the full checksum coverage of UDP). */
1426 case UDPLITE_SEND_CSCOV:
1427 val = up->pcslen;
1428 break;
1429
1430 case UDPLITE_RECV_CSCOV:
1431 val = up->pcrlen;
1432 break;
1433
1434 default:
1435 return -ENOPROTOOPT;
1436 };
1437
1438 if(put_user(len, optlen))
1439 return -EFAULT;
1440 if(copy_to_user(optval, &val,len))
1441 return -EFAULT;
1442 return 0;
1443 }
1444
1445 int udp_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 ip_getsockopt(sk, level, optname, optval, optlen);
1451 }
1452
1453 #ifdef CONFIG_COMPAT
1454 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1455 char __user *optval, int __user *optlen)
1456 {
1457 if (level == SOL_UDP || level == SOL_UDPLITE)
1458 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1459 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1460 }
1461 #endif
1462 /**
1463 * udp_poll - wait for a UDP event.
1464 * @file - file struct
1465 * @sock - socket
1466 * @wait - poll table
1467 *
1468 * This is same as datagram poll, except for the special case of
1469 * blocking sockets. If application is using a blocking fd
1470 * and a packet with checksum error is in the queue;
1471 * then it could get return from select indicating data available
1472 * but then block when reading it. Add special case code
1473 * to work around these arguably broken applications.
1474 */
1475 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1476 {
1477 unsigned int mask = datagram_poll(file, sock, wait);
1478 struct sock *sk = sock->sk;
1479 int is_lite = IS_UDPLITE(sk);
1480
1481 /* Check for false positives due to checksum errors */
1482 if ( (mask & POLLRDNORM) &&
1483 !(file->f_flags & O_NONBLOCK) &&
1484 !(sk->sk_shutdown & RCV_SHUTDOWN)){
1485 struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1486 struct sk_buff *skb;
1487
1488 spin_lock_bh(&rcvq->lock);
1489 while ((skb = skb_peek(rcvq)) != NULL) {
1490 if (udp_lib_checksum_complete(skb)) {
1491 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_lite);
1492 __skb_unlink(skb, rcvq);
1493 kfree_skb(skb);
1494 } else {
1495 skb->ip_summed = CHECKSUM_UNNECESSARY;
1496 break;
1497 }
1498 }
1499 spin_unlock_bh(&rcvq->lock);
1500
1501 /* nothing to see, move along */
1502 if (skb == NULL)
1503 mask &= ~(POLLIN | POLLRDNORM);
1504 }
1505
1506 return mask;
1507
1508 }
1509
1510 struct proto udp_prot = {
1511 .name = "UDP",
1512 .owner = THIS_MODULE,
1513 .close = udp_lib_close,
1514 .connect = ip4_datagram_connect,
1515 .disconnect = udp_disconnect,
1516 .ioctl = udp_ioctl,
1517 .destroy = udp_destroy_sock,
1518 .setsockopt = udp_setsockopt,
1519 .getsockopt = udp_getsockopt,
1520 .sendmsg = udp_sendmsg,
1521 .recvmsg = udp_recvmsg,
1522 .sendpage = udp_sendpage,
1523 .backlog_rcv = udp_queue_rcv_skb,
1524 .hash = udp_lib_hash,
1525 .unhash = udp_lib_unhash,
1526 .get_port = udp_v4_get_port,
1527 .obj_size = sizeof(struct udp_sock),
1528 #ifdef CONFIG_COMPAT
1529 .compat_setsockopt = compat_udp_setsockopt,
1530 .compat_getsockopt = compat_udp_getsockopt,
1531 #endif
1532 };
1533
1534 /* ------------------------------------------------------------------------ */
1535 #ifdef CONFIG_PROC_FS
1536
1537 static struct sock *udp_get_first(struct seq_file *seq)
1538 {
1539 struct sock *sk;
1540 struct udp_iter_state *state = seq->private;
1541
1542 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) {
1543 struct hlist_node *node;
1544 sk_for_each(sk, node, state->hashtable + state->bucket) {
1545 if (sk->sk_family == state->family)
1546 goto found;
1547 }
1548 }
1549 sk = NULL;
1550 found:
1551 return sk;
1552 }
1553
1554 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1555 {
1556 struct udp_iter_state *state = seq->private;
1557
1558 do {
1559 sk = sk_next(sk);
1560 try_again:
1561 ;
1562 } while (sk && sk->sk_family != state->family);
1563
1564 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) {
1565 sk = sk_head(state->hashtable + state->bucket);
1566 goto try_again;
1567 }
1568 return sk;
1569 }
1570
1571 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1572 {
1573 struct sock *sk = udp_get_first(seq);
1574
1575 if (sk)
1576 while(pos && (sk = udp_get_next(seq, sk)) != NULL)
1577 --pos;
1578 return pos ? NULL : sk;
1579 }
1580
1581 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1582 {
1583 read_lock(&udp_hash_lock);
1584 return *pos ? udp_get_idx(seq, *pos-1) : (void *)1;
1585 }
1586
1587 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1588 {
1589 struct sock *sk;
1590
1591 if (v == (void *)1)
1592 sk = udp_get_idx(seq, 0);
1593 else
1594 sk = udp_get_next(seq, v);
1595
1596 ++*pos;
1597 return sk;
1598 }
1599
1600 static void udp_seq_stop(struct seq_file *seq, void *v)
1601 {
1602 read_unlock(&udp_hash_lock);
1603 }
1604
1605 static int udp_seq_open(struct inode *inode, struct file *file)
1606 {
1607 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1608 struct seq_file *seq;
1609 int rc = -ENOMEM;
1610 struct udp_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL);
1611
1612 if (!s)
1613 goto out;
1614 s->family = afinfo->family;
1615 s->hashtable = afinfo->hashtable;
1616 s->seq_ops.start = udp_seq_start;
1617 s->seq_ops.next = udp_seq_next;
1618 s->seq_ops.show = afinfo->seq_show;
1619 s->seq_ops.stop = udp_seq_stop;
1620
1621 rc = seq_open(file, &s->seq_ops);
1622 if (rc)
1623 goto out_kfree;
1624
1625 seq = file->private_data;
1626 seq->private = s;
1627 out:
1628 return rc;
1629 out_kfree:
1630 kfree(s);
1631 goto out;
1632 }
1633
1634 /* ------------------------------------------------------------------------ */
1635 int udp_proc_register(struct udp_seq_afinfo *afinfo)
1636 {
1637 struct proc_dir_entry *p;
1638 int rc = 0;
1639
1640 if (!afinfo)
1641 return -EINVAL;
1642 afinfo->seq_fops->owner = afinfo->owner;
1643 afinfo->seq_fops->open = udp_seq_open;
1644 afinfo->seq_fops->read = seq_read;
1645 afinfo->seq_fops->llseek = seq_lseek;
1646 afinfo->seq_fops->release = seq_release_private;
1647
1648 p = proc_net_fops_create(afinfo->name, S_IRUGO, afinfo->seq_fops);
1649 if (p)
1650 p->data = afinfo;
1651 else
1652 rc = -ENOMEM;
1653 return rc;
1654 }
1655
1656 void udp_proc_unregister(struct udp_seq_afinfo *afinfo)
1657 {
1658 if (!afinfo)
1659 return;
1660 proc_net_remove(afinfo->name);
1661 memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops));
1662 }
1663
1664 /* ------------------------------------------------------------------------ */
1665 static void udp4_format_sock(struct sock *sp, char *tmpbuf, int bucket)
1666 {
1667 struct inet_sock *inet = inet_sk(sp);
1668 __be32 dest = inet->daddr;
1669 __be32 src = inet->rcv_saddr;
1670 __u16 destp = ntohs(inet->dport);
1671 __u16 srcp = ntohs(inet->sport);
1672
1673 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
1674 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p",
1675 bucket, src, srcp, dest, destp, sp->sk_state,
1676 atomic_read(&sp->sk_wmem_alloc),
1677 atomic_read(&sp->sk_rmem_alloc),
1678 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1679 atomic_read(&sp->sk_refcnt), sp);
1680 }
1681
1682 int udp4_seq_show(struct seq_file *seq, void *v)
1683 {
1684 if (v == SEQ_START_TOKEN)
1685 seq_printf(seq, "%-127s\n",
1686 " sl local_address rem_address st tx_queue "
1687 "rx_queue tr tm->when retrnsmt uid timeout "
1688 "inode");
1689 else {
1690 char tmpbuf[129];
1691 struct udp_iter_state *state = seq->private;
1692
1693 udp4_format_sock(v, tmpbuf, state->bucket);
1694 seq_printf(seq, "%-127s\n", tmpbuf);
1695 }
1696 return 0;
1697 }
1698
1699 /* ------------------------------------------------------------------------ */
1700 static struct file_operations udp4_seq_fops;
1701 static struct udp_seq_afinfo udp4_seq_afinfo = {
1702 .owner = THIS_MODULE,
1703 .name = "udp",
1704 .family = AF_INET,
1705 .hashtable = udp_hash,
1706 .seq_show = udp4_seq_show,
1707 .seq_fops = &udp4_seq_fops,
1708 };
1709
1710 int __init udp4_proc_init(void)
1711 {
1712 return udp_proc_register(&udp4_seq_afinfo);
1713 }
1714
1715 void udp4_proc_exit(void)
1716 {
1717 udp_proc_unregister(&udp4_seq_afinfo);
1718 }
1719 #endif /* CONFIG_PROC_FS */
1720
1721 EXPORT_SYMBOL(udp_disconnect);
1722 EXPORT_SYMBOL(udp_hash);
1723 EXPORT_SYMBOL(udp_hash_lock);
1724 EXPORT_SYMBOL(udp_ioctl);
1725 EXPORT_SYMBOL(udp_get_port);
1726 EXPORT_SYMBOL(udp_prot);
1727 EXPORT_SYMBOL(udp_sendmsg);
1728 EXPORT_SYMBOL(udp_lib_getsockopt);
1729 EXPORT_SYMBOL(udp_lib_setsockopt);
1730 EXPORT_SYMBOL(udp_poll);
1731
1732 #ifdef CONFIG_PROC_FS
1733 EXPORT_SYMBOL(udp_proc_register);
1734 EXPORT_SYMBOL(udp_proc_unregister);
1735 #endif
Linux 2.6 libata-dev (mirror)