?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[NETFILTER]: nf_conntrack: round up hashsize to next multiple of PAGE_SIZE
[linux-2.6/x86.git] / net / ipv4 / udp.c
blob4ec4a25a8d0cb9b0f6ff4c1dd1d9371566d32a6c
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 * James Chapman : Add L2TP encapsulation type.
74 *
75 *
76 * This program is free software; you can redistribute it and/or
77 * modify it under the terms of the GNU General Public License
78 * as published by the Free Software Foundation; either version
79 * 2 of the License, or (at your option) any later version.
80 */
81
82 #include <asm/system.h>
83 #include <asm/uaccess.h>
84 #include <asm/ioctls.h>
85 #include <linux/types.h>
86 #include <linux/fcntl.h>
87 #include <linux/module.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/igmp.h>
91 #include <linux/in.h>
92 #include <linux/errno.h>
93 #include <linux/timer.h>
94 #include <linux/mm.h>
95 #include <linux/inet.h>
96 #include <linux/netdevice.h>
97 #include <net/tcp_states.h>
98 #include <linux/skbuff.h>
99 #include <linux/proc_fs.h>
100 #include <linux/seq_file.h>
101 #include <net/icmp.h>
102 #include <net/route.h>
103 #include <net/checksum.h>
104 #include <net/xfrm.h>
105 #include "udp_impl.h"
106
107 /*
108 * Snmp MIB for the UDP layer
109 */
110
111 DEFINE_SNMP_STAT(struct udp_mib, udp_statistics) __read_mostly;
112
113 struct hlist_head udp_hash[UDP_HTABLE_SIZE];
114 DEFINE_RWLOCK(udp_hash_lock);
115
116 static int udp_port_rover;
117
118 static inline int __udp_lib_lport_inuse(__u16 num, struct hlist_head udptable[])
119 {
120 struct sock *sk;
121 struct hlist_node *node;
122
123 sk_for_each(sk, node, &udptable[num & (UDP_HTABLE_SIZE - 1)])
124 if (sk->sk_hash == num)
125 return 1;
126 return 0;
127 }
128
129 /**
130 * __udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
131 *
132 * @sk: socket struct in question
133 * @snum: port number to look up
134 * @udptable: hash list table, must be of UDP_HTABLE_SIZE
135 * @port_rover: pointer to record of last unallocated port
136 * @saddr_comp: AF-dependent comparison of bound local IP addresses
137 */
138 int __udp_lib_get_port(struct sock *sk, unsigned short snum,
139 struct hlist_head udptable[], int *port_rover,
140 int (*saddr_comp)(const struct sock *sk1,
141 const struct sock *sk2 ) )
142 {
143 struct hlist_node *node;
144 struct hlist_head *head;
145 struct sock *sk2;
146 int error = 1;
147
148 write_lock_bh(&udp_hash_lock);
149 if (snum == 0) {
150 int best_size_so_far, best, result, i;
151
152 if (*port_rover > sysctl_local_port_range[1] ||
153 *port_rover < sysctl_local_port_range[0])
154 *port_rover = sysctl_local_port_range[0];
155 best_size_so_far = 32767;
156 best = result = *port_rover;
157 for (i = 0; i < UDP_HTABLE_SIZE; i++, result++) {
158 int size;
159
160 head = &udptable[result & (UDP_HTABLE_SIZE - 1)];
161 if (hlist_empty(head)) {
162 if (result > sysctl_local_port_range[1])
163 result = sysctl_local_port_range[0] +
164 ((result - sysctl_local_port_range[0]) &
165 (UDP_HTABLE_SIZE - 1));
166 goto gotit;
167 }
168 size = 0;
169 sk_for_each(sk2, node, head) {
170 if (++size >= best_size_so_far)
171 goto next;
172 }
173 best_size_so_far = size;
174 best = result;
175 next:
176 ;
177 }
178 result = best;
179 for (i = 0; i < (1 << 16) / UDP_HTABLE_SIZE;
180 i++, result += UDP_HTABLE_SIZE) {
181 if (result > sysctl_local_port_range[1])
182 result = sysctl_local_port_range[0]
183 + ((result - sysctl_local_port_range[0]) &
184 (UDP_HTABLE_SIZE - 1));
185 if (! __udp_lib_lport_inuse(result, udptable))
186 break;
187 }
188 if (i >= (1 << 16) / UDP_HTABLE_SIZE)
189 goto fail;
190 gotit:
191 *port_rover = snum = result;
192 } else {
193 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
194
195 sk_for_each(sk2, node, head)
196 if (sk2->sk_hash == snum &&
197 sk2 != sk &&
198 (!sk2->sk_reuse || !sk->sk_reuse) &&
199 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if
200 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
201 (*saddr_comp)(sk, sk2) )
202 goto fail;
203 }
204 inet_sk(sk)->num = snum;
205 sk->sk_hash = snum;
206 if (sk_unhashed(sk)) {
207 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
208 sk_add_node(sk, head);
209 sock_prot_inc_use(sk->sk_prot);
210 }
211 error = 0;
212 fail:
213 write_unlock_bh(&udp_hash_lock);
214 return error;
215 }
216
217 int udp_get_port(struct sock *sk, unsigned short snum,
218 int (*scmp)(const struct sock *, const struct sock *))
219 {
220 return __udp_lib_get_port(sk, snum, udp_hash, &udp_port_rover, scmp);
221 }
222
223 int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
224 {
225 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
226
227 return ( !ipv6_only_sock(sk2) &&
228 (!inet1->rcv_saddr || !inet2->rcv_saddr ||
229 inet1->rcv_saddr == inet2->rcv_saddr ));
230 }
231
232 static inline int udp_v4_get_port(struct sock *sk, unsigned short snum)
233 {
234 return udp_get_port(sk, snum, ipv4_rcv_saddr_equal);
235 }
236
237 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
238 * harder than this. -DaveM
239 */
240 static struct sock *__udp4_lib_lookup(__be32 saddr, __be16 sport,
241 __be32 daddr, __be16 dport,
242 int dif, struct hlist_head udptable[])
243 {
244 struct sock *sk, *result = NULL;
245 struct hlist_node *node;
246 unsigned short hnum = ntohs(dport);
247 int badness = -1;
248
249 read_lock(&udp_hash_lock);
250 sk_for_each(sk, node, &udptable[hnum & (UDP_HTABLE_SIZE - 1)]) {
251 struct inet_sock *inet = inet_sk(sk);
252
253 if (sk->sk_hash == hnum && !ipv6_only_sock(sk)) {
254 int score = (sk->sk_family == PF_INET ? 1 : 0);
255 if (inet->rcv_saddr) {
256 if (inet->rcv_saddr != daddr)
257 continue;
258 score+=2;
259 }
260 if (inet->daddr) {
261 if (inet->daddr != saddr)
262 continue;
263 score+=2;
264 }
265 if (inet->dport) {
266 if (inet->dport != sport)
267 continue;
268 score+=2;
269 }
270 if (sk->sk_bound_dev_if) {
271 if (sk->sk_bound_dev_if != dif)
272 continue;
273 score+=2;
274 }
275 if (score == 9) {
276 result = sk;
277 break;
278 } else if (score > badness) {
279 result = sk;
280 badness = score;
281 }
282 }
283 }
284 if (result)
285 sock_hold(result);
286 read_unlock(&udp_hash_lock);
287 return result;
288 }
289
290 static inline struct sock *udp_v4_mcast_next(struct sock *sk,
291 __be16 loc_port, __be32 loc_addr,
292 __be16 rmt_port, __be32 rmt_addr,
293 int dif)
294 {
295 struct hlist_node *node;
296 struct sock *s = sk;
297 unsigned short hnum = ntohs(loc_port);
298
299 sk_for_each_from(s, node) {
300 struct inet_sock *inet = inet_sk(s);
301
302 if (s->sk_hash != hnum ||
303 (inet->daddr && inet->daddr != rmt_addr) ||
304 (inet->dport != rmt_port && inet->dport) ||
305 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) ||
306 ipv6_only_sock(s) ||
307 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
308 continue;
309 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
310 continue;
311 goto found;
312 }
313 s = NULL;
314 found:
315 return s;
316 }
317
318 /*
319 * This routine is called by the ICMP module when it gets some
320 * sort of error condition. If err < 0 then the socket should
321 * be closed and the error returned to the user. If err > 0
322 * it's just the icmp type << 8 | icmp code.
323 * Header points to the ip header of the error packet. We move
324 * on past this. Then (as it used to claim before adjustment)
325 * header points to the first 8 bytes of the udp header. We need
326 * to find the appropriate port.
327 */
328
329 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[])
330 {
331 struct inet_sock *inet;
332 struct iphdr *iph = (struct iphdr*)skb->data;
333 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2));
334 const int type = icmp_hdr(skb)->type;
335 const int code = icmp_hdr(skb)->code;
336 struct sock *sk;
337 int harderr;
338 int err;
339
340 sk = __udp4_lib_lookup(iph->daddr, uh->dest, iph->saddr, uh->source,
341 skb->dev->ifindex, udptable );
342 if (sk == NULL) {
343 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
344 return; /* No socket for error */
345 }
346
347 err = 0;
348 harderr = 0;
349 inet = inet_sk(sk);
350
351 switch (type) {
352 default:
353 case ICMP_TIME_EXCEEDED:
354 err = EHOSTUNREACH;
355 break;
356 case ICMP_SOURCE_QUENCH:
357 goto out;
358 case ICMP_PARAMETERPROB:
359 err = EPROTO;
360 harderr = 1;
361 break;
362 case ICMP_DEST_UNREACH:
363 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
364 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
365 err = EMSGSIZE;
366 harderr = 1;
367 break;
368 }
369 goto out;
370 }
371 err = EHOSTUNREACH;
372 if (code <= NR_ICMP_UNREACH) {
373 harderr = icmp_err_convert[code].fatal;
374 err = icmp_err_convert[code].errno;
375 }
376 break;
377 }
378
379 /*
380 * RFC1122: OK. Passes ICMP errors back to application, as per
381 * 4.1.3.3.
382 */
383 if (!inet->recverr) {
384 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
385 goto out;
386 } else {
387 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1));
388 }
389 sk->sk_err = err;
390 sk->sk_error_report(sk);
391 out:
392 sock_put(sk);
393 }
394
395 void udp_err(struct sk_buff *skb, u32 info)
396 {
397 return __udp4_lib_err(skb, info, udp_hash);
398 }
399
400 /*
401 * Throw away all pending data and cancel the corking. Socket is locked.
402 */
403 static void udp_flush_pending_frames(struct sock *sk)
404 {
405 struct udp_sock *up = udp_sk(sk);
406
407 if (up->pending) {
408 up->len = 0;
409 up->pending = 0;
410 ip_flush_pending_frames(sk);
411 }
412 }
413
414 /**
415 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
416 * @sk: socket we are sending on
417 * @skb: sk_buff containing the filled-in UDP header
418 * (checksum field must be zeroed out)
419 */
420 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
421 __be32 src, __be32 dst, int len )
422 {
423 unsigned int offset;
424 struct udphdr *uh = udp_hdr(skb);
425 __wsum csum = 0;
426
427 if (skb_queue_len(&sk->sk_write_queue) == 1) {
428 /*
429 * Only one fragment on the socket.
430 */
431 skb->csum_start = skb_transport_header(skb) - skb->head;
432 skb->csum_offset = offsetof(struct udphdr, check);
433 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
434 } else {
435 /*
436 * HW-checksum won't work as there are two or more
437 * fragments on the socket so that all csums of sk_buffs
438 * should be together
439 */
440 offset = skb_transport_offset(skb);
441 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
442
443 skb->ip_summed = CHECKSUM_NONE;
444
445 skb_queue_walk(&sk->sk_write_queue, skb) {
446 csum = csum_add(csum, skb->csum);
447 }
448
449 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
450 if (uh->check == 0)
451 uh->check = CSUM_MANGLED_0;
452 }
453 }
454
455 /*
456 * Push out all pending data as one UDP datagram. Socket is locked.
457 */
458 static int udp_push_pending_frames(struct sock *sk)
459 {
460 struct udp_sock *up = udp_sk(sk);
461 struct inet_sock *inet = inet_sk(sk);
462 struct flowi *fl = &inet->cork.fl;
463 struct sk_buff *skb;
464 struct udphdr *uh;
465 int err = 0;
466 __wsum csum = 0;
467
468 /* Grab the skbuff where UDP header space exists. */
469 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
470 goto out;
471
472 /*
473 * Create a UDP header
474 */
475 uh = udp_hdr(skb);
476 uh->source = fl->fl_ip_sport;
477 uh->dest = fl->fl_ip_dport;
478 uh->len = htons(up->len);
479 uh->check = 0;
480
481 if (up->pcflag) /* UDP-Lite */
482 csum = udplite_csum_outgoing(sk, skb);
483
484 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
485
486 skb->ip_summed = CHECKSUM_NONE;
487 goto send;
488
489 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
490
491 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len);
492 goto send;
493
494 } else /* `normal' UDP */
495 csum = udp_csum_outgoing(sk, skb);
496
497 /* add protocol-dependent pseudo-header */
498 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
499 sk->sk_protocol, csum );
500 if (uh->check == 0)
501 uh->check = CSUM_MANGLED_0;
502
503 send:
504 err = ip_push_pending_frames(sk);
505 out:
506 up->len = 0;
507 up->pending = 0;
508 return err;
509 }
510
511 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
512 size_t len)
513 {
514 struct inet_sock *inet = inet_sk(sk);
515 struct udp_sock *up = udp_sk(sk);
516 int ulen = len;
517 struct ipcm_cookie ipc;
518 struct rtable *rt = NULL;
519 int free = 0;
520 int connected = 0;
521 __be32 daddr, faddr, saddr;
522 __be16 dport;
523 u8 tos;
524 int err, is_udplite = up->pcflag;
525 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
526 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
527
528 if (len > 0xFFFF)
529 return -EMSGSIZE;
530
531 /*
532 * Check the flags.
533 */
534
535 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */
536 return -EOPNOTSUPP;
537
538 ipc.opt = NULL;
539
540 if (up->pending) {
541 /*
542 * There are pending frames.
543 * The socket lock must be held while it's corked.
544 */
545 lock_sock(sk);
546 if (likely(up->pending)) {
547 if (unlikely(up->pending != AF_INET)) {
548 release_sock(sk);
549 return -EINVAL;
550 }
551 goto do_append_data;
552 }
553 release_sock(sk);
554 }
555 ulen += sizeof(struct udphdr);
556
557 /*
558 * Get and verify the address.
559 */
560 if (msg->msg_name) {
561 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name;
562 if (msg->msg_namelen < sizeof(*usin))
563 return -EINVAL;
564 if (usin->sin_family != AF_INET) {
565 if (usin->sin_family != AF_UNSPEC)
566 return -EAFNOSUPPORT;
567 }
568
569 daddr = usin->sin_addr.s_addr;
570 dport = usin->sin_port;
571 if (dport == 0)
572 return -EINVAL;
573 } else {
574 if (sk->sk_state != TCP_ESTABLISHED)
575 return -EDESTADDRREQ;
576 daddr = inet->daddr;
577 dport = inet->dport;
578 /* Open fast path for connected socket.
579 Route will not be used, if at least one option is set.
580 */
581 connected = 1;
582 }
583 ipc.addr = inet->saddr;
584
585 ipc.oif = sk->sk_bound_dev_if;
586 if (msg->msg_controllen) {
587 err = ip_cmsg_send(msg, &ipc);
588 if (err)
589 return err;
590 if (ipc.opt)
591 free = 1;
592 connected = 0;
593 }
594 if (!ipc.opt)
595 ipc.opt = inet->opt;
596
597 saddr = ipc.addr;
598 ipc.addr = faddr = daddr;
599
600 if (ipc.opt && ipc.opt->srr) {
601 if (!daddr)
602 return -EINVAL;
603 faddr = ipc.opt->faddr;
604 connected = 0;
605 }
606 tos = RT_TOS(inet->tos);
607 if (sock_flag(sk, SOCK_LOCALROUTE) ||
608 (msg->msg_flags & MSG_DONTROUTE) ||
609 (ipc.opt && ipc.opt->is_strictroute)) {
610 tos |= RTO_ONLINK;
611 connected = 0;
612 }
613
614 if (MULTICAST(daddr)) {
615 if (!ipc.oif)
616 ipc.oif = inet->mc_index;
617 if (!saddr)
618 saddr = inet->mc_addr;
619 connected = 0;
620 }
621
622 if (connected)
623 rt = (struct rtable*)sk_dst_check(sk, 0);
624
625 if (rt == NULL) {
626 struct flowi fl = { .oif = ipc.oif,
627 .nl_u = { .ip4_u =
628 { .daddr = faddr,
629 .saddr = saddr,
630 .tos = tos } },
631 .proto = sk->sk_protocol,
632 .uli_u = { .ports =
633 { .sport = inet->sport,
634 .dport = dport } } };
635 security_sk_classify_flow(sk, &fl);
636 err = ip_route_output_flow(&rt, &fl, sk, 1);
637 if (err) {
638 if (err == -ENETUNREACH)
639 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
640 goto out;
641 }
642
643 err = -EACCES;
644 if ((rt->rt_flags & RTCF_BROADCAST) &&
645 !sock_flag(sk, SOCK_BROADCAST))
646 goto out;
647 if (connected)
648 sk_dst_set(sk, dst_clone(&rt->u.dst));
649 }
650
651 if (msg->msg_flags&MSG_CONFIRM)
652 goto do_confirm;
653 back_from_confirm:
654
655 saddr = rt->rt_src;
656 if (!ipc.addr)
657 daddr = ipc.addr = rt->rt_dst;
658
659 lock_sock(sk);
660 if (unlikely(up->pending)) {
661 /* The socket is already corked while preparing it. */
662 /* ... which is an evident application bug. --ANK */
663 release_sock(sk);
664
665 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
666 err = -EINVAL;
667 goto out;
668 }
669 /*
670 * Now cork the socket to pend data.
671 */
672 inet->cork.fl.fl4_dst = daddr;
673 inet->cork.fl.fl_ip_dport = dport;
674 inet->cork.fl.fl4_src = saddr;
675 inet->cork.fl.fl_ip_sport = inet->sport;
676 up->pending = AF_INET;
677
678 do_append_data:
679 up->len += ulen;
680 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
681 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
682 sizeof(struct udphdr), &ipc, rt,
683 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
684 if (err)
685 udp_flush_pending_frames(sk);
686 else if (!corkreq)
687 err = udp_push_pending_frames(sk);
688 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
689 up->pending = 0;
690 release_sock(sk);
691
692 out:
693 ip_rt_put(rt);
694 if (free)
695 kfree(ipc.opt);
696 if (!err) {
697 UDP_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS, is_udplite);
698 return len;
699 }
700 /*
701 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
702 * ENOBUFS might not be good (it's not tunable per se), but otherwise
703 * we don't have a good statistic (IpOutDiscards but it can be too many
704 * things). We could add another new stat but at least for now that
705 * seems like overkill.
706 */
707 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
708 UDP_INC_STATS_USER(UDP_MIB_SNDBUFERRORS, is_udplite);
709 }
710 return err;
711
712 do_confirm:
713 dst_confirm(&rt->u.dst);
714 if (!(msg->msg_flags&MSG_PROBE) || len)
715 goto back_from_confirm;
716 err = 0;
717 goto out;
718 }
719
720 int udp_sendpage(struct sock *sk, struct page *page, int offset,
721 size_t size, int flags)
722 {
723 struct udp_sock *up = udp_sk(sk);
724 int ret;
725
726 if (!up->pending) {
727 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
728
729 /* Call udp_sendmsg to specify destination address which
730 * sendpage interface can't pass.
731 * This will succeed only when the socket is connected.
732 */
733 ret = udp_sendmsg(NULL, sk, &msg, 0);
734 if (ret < 0)
735 return ret;
736 }
737
738 lock_sock(sk);
739
740 if (unlikely(!up->pending)) {
741 release_sock(sk);
742
743 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
744 return -EINVAL;
745 }
746
747 ret = ip_append_page(sk, page, offset, size, flags);
748 if (ret == -EOPNOTSUPP) {
749 release_sock(sk);
750 return sock_no_sendpage(sk->sk_socket, page, offset,
751 size, flags);
752 }
753 if (ret < 0) {
754 udp_flush_pending_frames(sk);
755 goto out;
756 }
757
758 up->len += size;
759 if (!(up->corkflag || (flags&MSG_MORE)))
760 ret = udp_push_pending_frames(sk);
761 if (!ret)
762 ret = size;
763 out:
764 release_sock(sk);
765 return ret;
766 }
767
768 /*
769 * IOCTL requests applicable to the UDP protocol
770 */
771
772 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
773 {
774 switch (cmd) {
775 case SIOCOUTQ:
776 {
777 int amount = atomic_read(&sk->sk_wmem_alloc);
778 return put_user(amount, (int __user *)arg);
779 }
780
781 case SIOCINQ:
782 {
783 struct sk_buff *skb;
784 unsigned long amount;
785
786 amount = 0;
787 spin_lock_bh(&sk->sk_receive_queue.lock);
788 skb = skb_peek(&sk->sk_receive_queue);
789 if (skb != NULL) {
790 /*
791 * We will only return the amount
792 * of this packet since that is all
793 * that will be read.
794 */
795 amount = skb->len - sizeof(struct udphdr);
796 }
797 spin_unlock_bh(&sk->sk_receive_queue.lock);
798 return put_user(amount, (int __user *)arg);
799 }
800
801 default:
802 return -ENOIOCTLCMD;
803 }
804
805 return 0;
806 }
807
808 /*
809 * This should be easy, if there is something there we
810 * return it, otherwise we block.
811 */
812
813 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
814 size_t len, int noblock, int flags, int *addr_len)
815 {
816 struct inet_sock *inet = inet_sk(sk);
817 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
818 struct sk_buff *skb;
819 unsigned int ulen, copied;
820 int err;
821 int is_udplite = IS_UDPLITE(sk);
822
823 /*
824 * Check any passed addresses
825 */
826 if (addr_len)
827 *addr_len=sizeof(*sin);
828
829 if (flags & MSG_ERRQUEUE)
830 return ip_recv_error(sk, msg, len);
831
832 try_again:
833 skb = skb_recv_datagram(sk, flags, noblock, &err);
834 if (!skb)
835 goto out;
836
837 ulen = skb->len - sizeof(struct udphdr);
838 copied = len;
839 if (copied > ulen)
840 copied = ulen;
841 else if (copied < ulen)
842 msg->msg_flags |= MSG_TRUNC;
843
844 /*
845 * If checksum is needed at all, try to do it while copying the
846 * data. If the data is truncated, or if we only want a partial
847 * coverage checksum (UDP-Lite), do it before the copy.
848 */
849
850 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
851 if (udp_lib_checksum_complete(skb))
852 goto csum_copy_err;
853 }
854
855 if (skb_csum_unnecessary(skb))
856 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
857 msg->msg_iov, copied );
858 else {
859 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
860
861 if (err == -EINVAL)
862 goto csum_copy_err;
863 }
864
865 if (err)
866 goto out_free;
867
868 sock_recv_timestamp(msg, sk, skb);
869
870 /* Copy the address. */
871 if (sin)
872 {
873 sin->sin_family = AF_INET;
874 sin->sin_port = udp_hdr(skb)->source;
875 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
876 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
877 }
878 if (inet->cmsg_flags)
879 ip_cmsg_recv(msg, skb);
880
881 err = copied;
882 if (flags & MSG_TRUNC)
883 err = ulen;
884
885 out_free:
886 skb_free_datagram(sk, skb);
887 out:
888 return err;
889
890 csum_copy_err:
891 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite);
892
893 skb_kill_datagram(sk, skb, flags);
894
895 if (noblock)
896 return -EAGAIN;
897 goto try_again;
898 }
899
900
901 int udp_disconnect(struct sock *sk, int flags)
902 {
903 struct inet_sock *inet = inet_sk(sk);
904 /*
905 * 1003.1g - break association.
906 */
907
908 sk->sk_state = TCP_CLOSE;
909 inet->daddr = 0;
910 inet->dport = 0;
911 sk->sk_bound_dev_if = 0;
912 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
913 inet_reset_saddr(sk);
914
915 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
916 sk->sk_prot->unhash(sk);
917 inet->sport = 0;
918 }
919 sk_dst_reset(sk);
920 return 0;
921 }
922
923 /* returns:
924 * -1: error
925 * 0: success
926 * >0: "udp encap" protocol resubmission
927 *
928 * Note that in the success and error cases, the skb is assumed to
929 * have either been requeued or freed.
930 */
931 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
932 {
933 struct udp_sock *up = udp_sk(sk);
934 int rc;
935
936 /*
937 * Charge it to the socket, dropping if the queue is full.
938 */
939 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
940 goto drop;
941 nf_reset(skb);
942
943 if (up->encap_type) {
944 /*
945 * This is an encapsulation socket so pass the skb to
946 * the socket's udp_encap_rcv() hook. Otherwise, just
947 * fall through and pass this up the UDP socket.
948 * up->encap_rcv() returns the following value:
949 * =0 if skb was successfully passed to the encap
950 * handler or was discarded by it.
951 * >0 if skb should be passed on to UDP.
952 * <0 if skb should be resubmitted as proto -N
953 */
954 unsigned int len;
955
956 /* if we're overly short, let UDP handle it */
957 len = skb->len - sizeof(struct udphdr);
958 if (len <= 0)
959 goto udp;
960
961 if (up->encap_rcv != NULL) {
962 int ret;
963
964 ret = (*up->encap_rcv)(sk, skb);
965 if (ret <= 0) {
966 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS, up->pcflag);
967 return -ret;
968 }
969 }
970
971 /* FALLTHROUGH -- it's a UDP Packet */
972 }
973
974 udp:
975 /*
976 * UDP-Lite specific tests, ignored on UDP sockets
977 */
978 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
979
980 /*
981 * MIB statistics other than incrementing the error count are
982 * disabled for the following two types of errors: these depend
983 * on the application settings, not on the functioning of the
984 * protocol stack as such.
985 *
986 * RFC 3828 here recommends (sec 3.3): "There should also be a
987 * way ... to ... at least let the receiving application block
988 * delivery of packets with coverage values less than a value
989 * provided by the application."
990 */
991 if (up->pcrlen == 0) { /* full coverage was set */
992 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
993 "%d while full coverage %d requested\n",
994 UDP_SKB_CB(skb)->cscov, skb->len);
995 goto drop;
996 }
997 /* The next case involves violating the min. coverage requested
998 * by the receiver. This is subtle: if receiver wants x and x is
999 * greater than the buffersize/MTU then receiver will complain
1000 * that it wants x while sender emits packets of smaller size y.
1001 * Therefore the above ...()->partial_cov statement is essential.
1002 */
1003 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1004 LIMIT_NETDEBUG(KERN_WARNING
1005 "UDPLITE: coverage %d too small, need min %d\n",
1006 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1007 goto drop;
1008 }
1009 }
1010
1011 if (sk->sk_filter) {
1012 if (udp_lib_checksum_complete(skb))
1013 goto drop;
1014 }
1015
1016 if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) {
1017 /* Note that an ENOMEM error is charged twice */
1018 if (rc == -ENOMEM)
1019 UDP_INC_STATS_BH(UDP_MIB_RCVBUFERRORS, up->pcflag);
1020 goto drop;
1021 }
1022
1023 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS, up->pcflag);
1024 return 0;
1025
1026 drop:
1027 UDP_INC_STATS_BH(UDP_MIB_INERRORS, up->pcflag);
1028 kfree_skb(skb);
1029 return -1;
1030 }
1031
1032 /*
1033 * Multicasts and broadcasts go to each listener.
1034 *
1035 * Note: called only from the BH handler context,
1036 * so we don't need to lock the hashes.
1037 */
1038 static int __udp4_lib_mcast_deliver(struct sk_buff *skb,
1039 struct udphdr *uh,
1040 __be32 saddr, __be32 daddr,
1041 struct hlist_head udptable[])
1042 {
1043 struct sock *sk;
1044 int dif;
1045
1046 read_lock(&udp_hash_lock);
1047 sk = sk_head(&udptable[ntohs(uh->dest) & (UDP_HTABLE_SIZE - 1)]);
1048 dif = skb->dev->ifindex;
1049 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
1050 if (sk) {
1051 struct sock *sknext = NULL;
1052
1053 do {
1054 struct sk_buff *skb1 = skb;
1055
1056 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr,
1057 uh->source, saddr, dif);
1058 if (sknext)
1059 skb1 = skb_clone(skb, GFP_ATOMIC);
1060
1061 if (skb1) {
1062 int ret = udp_queue_rcv_skb(sk, skb1);
1063 if (ret > 0)
1064 /* we should probably re-process instead
1065 * of dropping packets here. */
1066 kfree_skb(skb1);
1067 }
1068 sk = sknext;
1069 } while (sknext);
1070 } else
1071 kfree_skb(skb);
1072 read_unlock(&udp_hash_lock);
1073 return 0;
1074 }
1075
1076 /* Initialize UDP checksum. If exited with zero value (success),
1077 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1078 * Otherwise, csum completion requires chacksumming packet body,
1079 * including udp header and folding it to skb->csum.
1080 */
1081 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1082 int proto)
1083 {
1084 const struct iphdr *iph;
1085 int err;
1086
1087 UDP_SKB_CB(skb)->partial_cov = 0;
1088 UDP_SKB_CB(skb)->cscov = skb->len;
1089
1090 if (proto == IPPROTO_UDPLITE) {
1091 err = udplite_checksum_init(skb, uh);
1092 if (err)
1093 return err;
1094 }
1095
1096 iph = ip_hdr(skb);
1097 if (uh->check == 0) {
1098 skb->ip_summed = CHECKSUM_UNNECESSARY;
1099 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1100 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1101 proto, skb->csum))
1102 skb->ip_summed = CHECKSUM_UNNECESSARY;
1103 }
1104 if (!skb_csum_unnecessary(skb))
1105 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1106 skb->len, proto, 0);
1107 /* Probably, we should checksum udp header (it should be in cache
1108 * in any case) and data in tiny packets (< rx copybreak).
1109 */
1110
1111 return 0;
1112 }
1113
1114 /*
1115 * All we need to do is get the socket, and then do a checksum.
1116 */
1117
1118 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
1119 int proto)
1120 {
1121 struct sock *sk;
1122 struct udphdr *uh = udp_hdr(skb);
1123 unsigned short ulen;
1124 struct rtable *rt = (struct rtable*)skb->dst;
1125 __be32 saddr = ip_hdr(skb)->saddr;
1126 __be32 daddr = ip_hdr(skb)->daddr;
1127
1128 /*
1129 * Validate the packet.
1130 */
1131 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1132 goto drop; /* No space for header. */
1133
1134 ulen = ntohs(uh->len);
1135 if (ulen > skb->len)
1136 goto short_packet;
1137
1138 if (proto == IPPROTO_UDP) {
1139 /* UDP validates ulen. */
1140 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1141 goto short_packet;
1142 uh = udp_hdr(skb);
1143 }
1144
1145 if (udp4_csum_init(skb, uh, proto))
1146 goto csum_error;
1147
1148 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1149 return __udp4_lib_mcast_deliver(skb, uh, saddr, daddr, udptable);
1150
1151 sk = __udp4_lib_lookup(saddr, uh->source, daddr, uh->dest,
1152 skb->dev->ifindex, udptable );
1153
1154 if (sk != NULL) {
1155 int ret = udp_queue_rcv_skb(sk, skb);
1156 sock_put(sk);
1157
1158 /* a return value > 0 means to resubmit the input, but
1159 * it wants the return to be -protocol, or 0
1160 */
1161 if (ret > 0)
1162 return -ret;
1163 return 0;
1164 }
1165
1166 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1167 goto drop;
1168 nf_reset(skb);
1169
1170 /* No socket. Drop packet silently, if checksum is wrong */
1171 if (udp_lib_checksum_complete(skb))
1172 goto csum_error;
1173
1174 UDP_INC_STATS_BH(UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1175 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1176
1177 /*
1178 * Hmm. We got an UDP packet to a port to which we
1179 * don't wanna listen. Ignore it.
1180 */
1181 kfree_skb(skb);
1182 return 0;
1183
1184 short_packet:
1185 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %u.%u.%u.%u:%u %d/%d to %u.%u.%u.%u:%u\n",
1186 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1187 NIPQUAD(saddr),
1188 ntohs(uh->source),
1189 ulen,
1190 skb->len,
1191 NIPQUAD(daddr),
1192 ntohs(uh->dest));
1193 goto drop;
1194
1195 csum_error:
1196 /*
1197 * RFC1122: OK. Discards the bad packet silently (as far as
1198 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1199 */
1200 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %d.%d.%d.%d:%d to %d.%d.%d.%d:%d ulen %d\n",
1201 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1202 NIPQUAD(saddr),
1203 ntohs(uh->source),
1204 NIPQUAD(daddr),
1205 ntohs(uh->dest),
1206 ulen);
1207 drop:
1208 UDP_INC_STATS_BH(UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1209 kfree_skb(skb);
1210 return 0;
1211 }
1212
1213 int udp_rcv(struct sk_buff *skb)
1214 {
1215 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP);
1216 }
1217
1218 int udp_destroy_sock(struct sock *sk)
1219 {
1220 lock_sock(sk);
1221 udp_flush_pending_frames(sk);
1222 release_sock(sk);
1223 return 0;
1224 }
1225
1226 /*
1227 * Socket option code for UDP
1228 */
1229 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1230 char __user *optval, int optlen,
1231 int (*push_pending_frames)(struct sock *))
1232 {
1233 struct udp_sock *up = udp_sk(sk);
1234 int val;
1235 int err = 0;
1236
1237 if (optlen<sizeof(int))
1238 return -EINVAL;
1239
1240 if (get_user(val, (int __user *)optval))
1241 return -EFAULT;
1242
1243 switch (optname) {
1244 case UDP_CORK:
1245 if (val != 0) {
1246 up->corkflag = 1;
1247 } else {
1248 up->corkflag = 0;
1249 lock_sock(sk);
1250 (*push_pending_frames)(sk);
1251 release_sock(sk);
1252 }
1253 break;
1254
1255 case UDP_ENCAP:
1256 switch (val) {
1257 case 0:
1258 case UDP_ENCAP_ESPINUDP:
1259 case UDP_ENCAP_ESPINUDP_NON_IKE:
1260 up->encap_rcv = xfrm4_udp_encap_rcv;
1261 /* FALLTHROUGH */
1262 case UDP_ENCAP_L2TPINUDP:
1263 up->encap_type = val;
1264 break;
1265 default:
1266 err = -ENOPROTOOPT;
1267 break;
1268 }
1269 break;
1270
1271 /*
1272 * UDP-Lite's partial checksum coverage (RFC 3828).
1273 */
1274 /* The sender sets actual checksum coverage length via this option.
1275 * The case coverage > packet length is handled by send module. */
1276 case UDPLITE_SEND_CSCOV:
1277 if (!up->pcflag) /* Disable the option on UDP sockets */
1278 return -ENOPROTOOPT;
1279 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1280 val = 8;
1281 up->pcslen = val;
1282 up->pcflag |= UDPLITE_SEND_CC;
1283 break;
1284
1285 /* The receiver specifies a minimum checksum coverage value. To make
1286 * sense, this should be set to at least 8 (as done below). If zero is
1287 * used, this again means full checksum coverage. */
1288 case UDPLITE_RECV_CSCOV:
1289 if (!up->pcflag) /* Disable the option on UDP sockets */
1290 return -ENOPROTOOPT;
1291 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1292 val = 8;
1293 up->pcrlen = val;
1294 up->pcflag |= UDPLITE_RECV_CC;
1295 break;
1296
1297 default:
1298 err = -ENOPROTOOPT;
1299 break;
1300 }
1301
1302 return err;
1303 }
1304
1305 int udp_setsockopt(struct sock *sk, int level, int optname,
1306 char __user *optval, int optlen)
1307 {
1308 if (level == SOL_UDP || level == SOL_UDPLITE)
1309 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1310 udp_push_pending_frames);
1311 return ip_setsockopt(sk, level, optname, optval, optlen);
1312 }
1313
1314 #ifdef CONFIG_COMPAT
1315 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1316 char __user *optval, int optlen)
1317 {
1318 if (level == SOL_UDP || level == SOL_UDPLITE)
1319 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1320 udp_push_pending_frames);
1321 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1322 }
1323 #endif
1324
1325 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1326 char __user *optval, int __user *optlen)
1327 {
1328 struct udp_sock *up = udp_sk(sk);
1329 int val, len;
1330
1331 if (get_user(len,optlen))
1332 return -EFAULT;
1333
1334 len = min_t(unsigned int, len, sizeof(int));
1335
1336 if (len < 0)
1337 return -EINVAL;
1338
1339 switch (optname) {
1340 case UDP_CORK:
1341 val = up->corkflag;
1342 break;
1343
1344 case UDP_ENCAP:
1345 val = up->encap_type;
1346 break;
1347
1348 /* The following two cannot be changed on UDP sockets, the return is
1349 * always 0 (which corresponds to the full checksum coverage of UDP). */
1350 case UDPLITE_SEND_CSCOV:
1351 val = up->pcslen;
1352 break;
1353
1354 case UDPLITE_RECV_CSCOV:
1355 val = up->pcrlen;
1356 break;
1357
1358 default:
1359 return -ENOPROTOOPT;
1360 }
1361
1362 if (put_user(len, optlen))
1363 return -EFAULT;
1364 if (copy_to_user(optval, &val,len))
1365 return -EFAULT;
1366 return 0;
1367 }
1368
1369 int udp_getsockopt(struct sock *sk, int level, int optname,
1370 char __user *optval, int __user *optlen)
1371 {
1372 if (level == SOL_UDP || level == SOL_UDPLITE)
1373 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1374 return ip_getsockopt(sk, level, optname, optval, optlen);
1375 }
1376
1377 #ifdef CONFIG_COMPAT
1378 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1379 char __user *optval, int __user *optlen)
1380 {
1381 if (level == SOL_UDP || level == SOL_UDPLITE)
1382 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1383 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1384 }
1385 #endif
1386 /**
1387 * udp_poll - wait for a UDP event.
1388 * @file - file struct
1389 * @sock - socket
1390 * @wait - poll table
1391 *
1392 * This is same as datagram poll, except for the special case of
1393 * blocking sockets. If application is using a blocking fd
1394 * and a packet with checksum error is in the queue;
1395 * then it could get return from select indicating data available
1396 * but then block when reading it. Add special case code
1397 * to work around these arguably broken applications.
1398 */
1399 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1400 {
1401 unsigned int mask = datagram_poll(file, sock, wait);
1402 struct sock *sk = sock->sk;
1403 int is_lite = IS_UDPLITE(sk);
1404
1405 /* Check for false positives due to checksum errors */
1406 if ( (mask & POLLRDNORM) &&
1407 !(file->f_flags & O_NONBLOCK) &&
1408 !(sk->sk_shutdown & RCV_SHUTDOWN)){
1409 struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1410 struct sk_buff *skb;
1411
1412 spin_lock_bh(&rcvq->lock);
1413 while ((skb = skb_peek(rcvq)) != NULL &&
1414 udp_lib_checksum_complete(skb)) {
1415 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_lite);
1416 __skb_unlink(skb, rcvq);
1417 kfree_skb(skb);
1418 }
1419 spin_unlock_bh(&rcvq->lock);
1420
1421 /* nothing to see, move along */
1422 if (skb == NULL)
1423 mask &= ~(POLLIN | POLLRDNORM);
1424 }
1425
1426 return mask;
1427
1428 }
1429
1430 struct proto udp_prot = {
1431 .name = "UDP",
1432 .owner = THIS_MODULE,
1433 .close = udp_lib_close,
1434 .connect = ip4_datagram_connect,
1435 .disconnect = udp_disconnect,
1436 .ioctl = udp_ioctl,
1437 .destroy = udp_destroy_sock,
1438 .setsockopt = udp_setsockopt,
1439 .getsockopt = udp_getsockopt,
1440 .sendmsg = udp_sendmsg,
1441 .recvmsg = udp_recvmsg,
1442 .sendpage = udp_sendpage,
1443 .backlog_rcv = udp_queue_rcv_skb,
1444 .hash = udp_lib_hash,
1445 .unhash = udp_lib_unhash,
1446 .get_port = udp_v4_get_port,
1447 .obj_size = sizeof(struct udp_sock),
1448 #ifdef CONFIG_COMPAT
1449 .compat_setsockopt = compat_udp_setsockopt,
1450 .compat_getsockopt = compat_udp_getsockopt,
1451 #endif
1452 };
1453
1454 /* ------------------------------------------------------------------------ */
1455 #ifdef CONFIG_PROC_FS
1456
1457 static struct sock *udp_get_first(struct seq_file *seq)
1458 {
1459 struct sock *sk;
1460 struct udp_iter_state *state = seq->private;
1461
1462 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) {
1463 struct hlist_node *node;
1464 sk_for_each(sk, node, state->hashtable + state->bucket) {
1465 if (sk->sk_family == state->family)
1466 goto found;
1467 }
1468 }
1469 sk = NULL;
1470 found:
1471 return sk;
1472 }
1473
1474 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1475 {
1476 struct udp_iter_state *state = seq->private;
1477
1478 do {
1479 sk = sk_next(sk);
1480 try_again:
1481 ;
1482 } while (sk && sk->sk_family != state->family);
1483
1484 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) {
1485 sk = sk_head(state->hashtable + state->bucket);
1486 goto try_again;
1487 }
1488 return sk;
1489 }
1490
1491 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1492 {
1493 struct sock *sk = udp_get_first(seq);
1494
1495 if (sk)
1496 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1497 --pos;
1498 return pos ? NULL : sk;
1499 }
1500
1501 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1502 {
1503 read_lock(&udp_hash_lock);
1504 return *pos ? udp_get_idx(seq, *pos-1) : (void *)1;
1505 }
1506
1507 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1508 {
1509 struct sock *sk;
1510
1511 if (v == (void *)1)
1512 sk = udp_get_idx(seq, 0);
1513 else
1514 sk = udp_get_next(seq, v);
1515
1516 ++*pos;
1517 return sk;
1518 }
1519
1520 static void udp_seq_stop(struct seq_file *seq, void *v)
1521 {
1522 read_unlock(&udp_hash_lock);
1523 }
1524
1525 static int udp_seq_open(struct inode *inode, struct file *file)
1526 {
1527 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1528 struct seq_file *seq;
1529 int rc = -ENOMEM;
1530 struct udp_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL);
1531
1532 if (!s)
1533 goto out;
1534 s->family = afinfo->family;
1535 s->hashtable = afinfo->hashtable;
1536 s->seq_ops.start = udp_seq_start;
1537 s->seq_ops.next = udp_seq_next;
1538 s->seq_ops.show = afinfo->seq_show;
1539 s->seq_ops.stop = udp_seq_stop;
1540
1541 rc = seq_open(file, &s->seq_ops);
1542 if (rc)
1543 goto out_kfree;
1544
1545 seq = file->private_data;
1546 seq->private = s;
1547 out:
1548 return rc;
1549 out_kfree:
1550 kfree(s);
1551 goto out;
1552 }
1553
1554 /* ------------------------------------------------------------------------ */
1555 int udp_proc_register(struct udp_seq_afinfo *afinfo)
1556 {
1557 struct proc_dir_entry *p;
1558 int rc = 0;
1559
1560 if (!afinfo)
1561 return -EINVAL;
1562 afinfo->seq_fops->owner = afinfo->owner;
1563 afinfo->seq_fops->open = udp_seq_open;
1564 afinfo->seq_fops->read = seq_read;
1565 afinfo->seq_fops->llseek = seq_lseek;
1566 afinfo->seq_fops->release = seq_release_private;
1567
1568 p = proc_net_fops_create(afinfo->name, S_IRUGO, afinfo->seq_fops);
1569 if (p)
1570 p->data = afinfo;
1571 else
1572 rc = -ENOMEM;
1573 return rc;
1574 }
1575
1576 void udp_proc_unregister(struct udp_seq_afinfo *afinfo)
1577 {
1578 if (!afinfo)
1579 return;
1580 proc_net_remove(afinfo->name);
1581 memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops));
1582 }
1583
1584 /* ------------------------------------------------------------------------ */
1585 static void udp4_format_sock(struct sock *sp, char *tmpbuf, int bucket)
1586 {
1587 struct inet_sock *inet = inet_sk(sp);
1588 __be32 dest = inet->daddr;
1589 __be32 src = inet->rcv_saddr;
1590 __u16 destp = ntohs(inet->dport);
1591 __u16 srcp = ntohs(inet->sport);
1592
1593 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
1594 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p",
1595 bucket, src, srcp, dest, destp, sp->sk_state,
1596 atomic_read(&sp->sk_wmem_alloc),
1597 atomic_read(&sp->sk_rmem_alloc),
1598 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1599 atomic_read(&sp->sk_refcnt), sp);
1600 }
1601
1602 int udp4_seq_show(struct seq_file *seq, void *v)
1603 {
1604 if (v == SEQ_START_TOKEN)
1605 seq_printf(seq, "%-127s\n",
1606 " sl local_address rem_address st tx_queue "
1607 "rx_queue tr tm->when retrnsmt uid timeout "
1608 "inode");
1609 else {
1610 char tmpbuf[129];
1611 struct udp_iter_state *state = seq->private;
1612
1613 udp4_format_sock(v, tmpbuf, state->bucket);
1614 seq_printf(seq, "%-127s\n", tmpbuf);
1615 }
1616 return 0;
1617 }
1618
1619 /* ------------------------------------------------------------------------ */
1620 static struct file_operations udp4_seq_fops;
1621 static struct udp_seq_afinfo udp4_seq_afinfo = {
1622 .owner = THIS_MODULE,
1623 .name = "udp",
1624 .family = AF_INET,
1625 .hashtable = udp_hash,
1626 .seq_show = udp4_seq_show,
1627 .seq_fops = &udp4_seq_fops,
1628 };
1629
1630 int __init udp4_proc_init(void)
1631 {
1632 return udp_proc_register(&udp4_seq_afinfo);
1633 }
1634
1635 void udp4_proc_exit(void)
1636 {
1637 udp_proc_unregister(&udp4_seq_afinfo);
1638 }
1639 #endif /* CONFIG_PROC_FS */
1640
1641 EXPORT_SYMBOL(udp_disconnect);
1642 EXPORT_SYMBOL(udp_hash);
1643 EXPORT_SYMBOL(udp_hash_lock);
1644 EXPORT_SYMBOL(udp_ioctl);
1645 EXPORT_SYMBOL(udp_get_port);
1646 EXPORT_SYMBOL(udp_prot);
1647 EXPORT_SYMBOL(udp_sendmsg);
1648 EXPORT_SYMBOL(udp_lib_getsockopt);
1649 EXPORT_SYMBOL(udp_lib_setsockopt);
1650 EXPORT_SYMBOL(udp_poll);
1651
1652 #ifdef CONFIG_PROC_FS
1653 EXPORT_SYMBOL(udp_proc_register);
1654 EXPORT_SYMBOL(udp_proc_unregister);
1655 #endif
x86 "subsystem" repository