projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
[UDP]: Only increment counter on first peek/recv
[deliverable/linux.git] / net / ipv4 / udp.c
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/net_namespace.h>
102 #include <net/icmp.h>
103 #include <net/route.h>
104 #include <net/checksum.h>
105 #include <net/xfrm.h>
106 #include "udp_impl.h"
107
108 /*
109 * Snmp MIB for the UDP layer
110 */
111
112 DEFINE_SNMP_STAT(struct udp_mib, udp_statistics) __read_mostly;
113 EXPORT_SYMBOL(udp_statistics);
114
115 struct hlist_head udp_hash[UDP_HTABLE_SIZE];
116 DEFINE_RWLOCK(udp_hash_lock);
117
118 static inline int __udp_lib_lport_inuse(__u16 num,
119 const struct hlist_head udptable[])
120 {
121 struct sock *sk;
122 struct hlist_node *node;
123
124 sk_for_each(sk, node, &udptable[num & (UDP_HTABLE_SIZE - 1)])
125 if (sk->sk_hash == num)
126 return 1;
127 return 0;
128 }
129
130 /**
131 * __udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
132 *
133 * @sk: socket struct in question
134 * @snum: port number to look up
135 * @udptable: hash list table, must be of UDP_HTABLE_SIZE
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[],
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
150 if (!snum) {
151 int i, low, high, remaining;
152 unsigned rover, best, best_size_so_far;
153
154 inet_get_local_port_range(&low, &high);
155 remaining = (high - low) + 1;
156
157 best_size_so_far = UINT_MAX;
158 best = rover = net_random() % remaining + low;
159
160 /* 1st pass: look for empty (or shortest) hash chain */
161 for (i = 0; i < UDP_HTABLE_SIZE; i++) {
162 int size = 0;
163
164 head = &udptable[rover & (UDP_HTABLE_SIZE - 1)];
165 if (hlist_empty(head))
166 goto gotit;
167
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 = rover;
174 next:
175 /* fold back if end of range */
176 if (++rover > high)
177 rover = low + ((rover - low)
178 & (UDP_HTABLE_SIZE - 1));
179
180
181 }
182
183 /* 2nd pass: find hole in shortest hash chain */
184 rover = best;
185 for (i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; i++) {
186 if (! __udp_lib_lport_inuse(rover, udptable))
187 goto gotit;
188 rover += UDP_HTABLE_SIZE;
189 if (rover > high)
190 rover = low + ((rover - low)
191 & (UDP_HTABLE_SIZE - 1));
192 }
193
194
195 /* All ports in use! */
196 goto fail;
197
198 gotit:
199 snum = rover;
200 } else {
201 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
202
203 sk_for_each(sk2, node, head)
204 if (sk2->sk_hash == snum &&
205 sk2 != sk &&
206 (!sk2->sk_reuse || !sk->sk_reuse) &&
207 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if
208 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
209 (*saddr_comp)(sk, sk2) )
210 goto fail;
211 }
212
213 inet_sk(sk)->num = snum;
214 sk->sk_hash = snum;
215 if (sk_unhashed(sk)) {
216 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
217 sk_add_node(sk, head);
218 sock_prot_inc_use(sk->sk_prot);
219 }
220 error = 0;
221 fail:
222 write_unlock_bh(&udp_hash_lock);
223 return error;
224 }
225
226 int udp_get_port(struct sock *sk, unsigned short snum,
227 int (*scmp)(const struct sock *, const struct sock *))
228 {
229 return __udp_lib_get_port(sk, snum, udp_hash, scmp);
230 }
231
232 int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
233 {
234 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
235
236 return ( !ipv6_only_sock(sk2) &&
237 (!inet1->rcv_saddr || !inet2->rcv_saddr ||
238 inet1->rcv_saddr == inet2->rcv_saddr ));
239 }
240
241 static inline int udp_v4_get_port(struct sock *sk, unsigned short snum)
242 {
243 return udp_get_port(sk, snum, ipv4_rcv_saddr_equal);
244 }
245
246 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
247 * harder than this. -DaveM
248 */
249 static struct sock *__udp4_lib_lookup(__be32 saddr, __be16 sport,
250 __be32 daddr, __be16 dport,
251 int dif, struct hlist_head udptable[])
252 {
253 struct sock *sk, *result = NULL;
254 struct hlist_node *node;
255 unsigned short hnum = ntohs(dport);
256 int badness = -1;
257
258 read_lock(&udp_hash_lock);
259 sk_for_each(sk, node, &udptable[hnum & (UDP_HTABLE_SIZE - 1)]) {
260 struct inet_sock *inet = inet_sk(sk);
261
262 if (sk->sk_hash == hnum && !ipv6_only_sock(sk)) {
263 int score = (sk->sk_family == PF_INET ? 1 : 0);
264 if (inet->rcv_saddr) {
265 if (inet->rcv_saddr != daddr)
266 continue;
267 score+=2;
268 }
269 if (inet->daddr) {
270 if (inet->daddr != saddr)
271 continue;
272 score+=2;
273 }
274 if (inet->dport) {
275 if (inet->dport != sport)
276 continue;
277 score+=2;
278 }
279 if (sk->sk_bound_dev_if) {
280 if (sk->sk_bound_dev_if != dif)
281 continue;
282 score+=2;
283 }
284 if (score == 9) {
285 result = sk;
286 break;
287 } else if (score > badness) {
288 result = sk;
289 badness = score;
290 }
291 }
292 }
293 if (result)
294 sock_hold(result);
295 read_unlock(&udp_hash_lock);
296 return result;
297 }
298
299 static inline struct sock *udp_v4_mcast_next(struct sock *sk,
300 __be16 loc_port, __be32 loc_addr,
301 __be16 rmt_port, __be32 rmt_addr,
302 int dif)
303 {
304 struct hlist_node *node;
305 struct sock *s = sk;
306 unsigned short hnum = ntohs(loc_port);
307
308 sk_for_each_from(s, node) {
309 struct inet_sock *inet = inet_sk(s);
310
311 if (s->sk_hash != hnum ||
312 (inet->daddr && inet->daddr != rmt_addr) ||
313 (inet->dport != rmt_port && inet->dport) ||
314 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) ||
315 ipv6_only_sock(s) ||
316 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
317 continue;
318 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
319 continue;
320 goto found;
321 }
322 s = NULL;
323 found:
324 return s;
325 }
326
327 /*
328 * This routine is called by the ICMP module when it gets some
329 * sort of error condition. If err < 0 then the socket should
330 * be closed and the error returned to the user. If err > 0
331 * it's just the icmp type << 8 | icmp code.
332 * Header points to the ip header of the error packet. We move
333 * on past this. Then (as it used to claim before adjustment)
334 * header points to the first 8 bytes of the udp header. We need
335 * to find the appropriate port.
336 */
337
338 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[])
339 {
340 struct inet_sock *inet;
341 struct iphdr *iph = (struct iphdr*)skb->data;
342 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2));
343 const int type = icmp_hdr(skb)->type;
344 const int code = icmp_hdr(skb)->code;
345 struct sock *sk;
346 int harderr;
347 int err;
348
349 sk = __udp4_lib_lookup(iph->daddr, uh->dest, iph->saddr, uh->source,
350 skb->dev->ifindex, udptable );
351 if (sk == NULL) {
352 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
353 return; /* No socket for error */
354 }
355
356 err = 0;
357 harderr = 0;
358 inet = inet_sk(sk);
359
360 switch (type) {
361 default:
362 case ICMP_TIME_EXCEEDED:
363 err = EHOSTUNREACH;
364 break;
365 case ICMP_SOURCE_QUENCH:
366 goto out;
367 case ICMP_PARAMETERPROB:
368 err = EPROTO;
369 harderr = 1;
370 break;
371 case ICMP_DEST_UNREACH:
372 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
373 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
374 err = EMSGSIZE;
375 harderr = 1;
376 break;
377 }
378 goto out;
379 }
380 err = EHOSTUNREACH;
381 if (code <= NR_ICMP_UNREACH) {
382 harderr = icmp_err_convert[code].fatal;
383 err = icmp_err_convert[code].errno;
384 }
385 break;
386 }
387
388 /*
389 * RFC1122: OK. Passes ICMP errors back to application, as per
390 * 4.1.3.3.
391 */
392 if (!inet->recverr) {
393 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
394 goto out;
395 } else {
396 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1));
397 }
398 sk->sk_err = err;
399 sk->sk_error_report(sk);
400 out:
401 sock_put(sk);
402 }
403
404 void udp_err(struct sk_buff *skb, u32 info)
405 {
406 return __udp4_lib_err(skb, info, udp_hash);
407 }
408
409 /*
410 * Throw away all pending data and cancel the corking. Socket is locked.
411 */
412 static void udp_flush_pending_frames(struct sock *sk)
413 {
414 struct udp_sock *up = udp_sk(sk);
415
416 if (up->pending) {
417 up->len = 0;
418 up->pending = 0;
419 ip_flush_pending_frames(sk);
420 }
421 }
422
423 /**
424 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
425 * @sk: socket we are sending on
426 * @skb: sk_buff containing the filled-in UDP header
427 * (checksum field must be zeroed out)
428 */
429 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
430 __be32 src, __be32 dst, int len )
431 {
432 unsigned int offset;
433 struct udphdr *uh = udp_hdr(skb);
434 __wsum csum = 0;
435
436 if (skb_queue_len(&sk->sk_write_queue) == 1) {
437 /*
438 * Only one fragment on the socket.
439 */
440 skb->csum_start = skb_transport_header(skb) - skb->head;
441 skb->csum_offset = offsetof(struct udphdr, check);
442 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
443 } else {
444 /*
445 * HW-checksum won't work as there are two or more
446 * fragments on the socket so that all csums of sk_buffs
447 * should be together
448 */
449 offset = skb_transport_offset(skb);
450 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
451
452 skb->ip_summed = CHECKSUM_NONE;
453
454 skb_queue_walk(&sk->sk_write_queue, skb) {
455 csum = csum_add(csum, skb->csum);
456 }
457
458 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
459 if (uh->check == 0)
460 uh->check = CSUM_MANGLED_0;
461 }
462 }
463
464 /*
465 * Push out all pending data as one UDP datagram. Socket is locked.
466 */
467 static int udp_push_pending_frames(struct sock *sk)
468 {
469 struct udp_sock *up = udp_sk(sk);
470 struct inet_sock *inet = inet_sk(sk);
471 struct flowi *fl = &inet->cork.fl;
472 struct sk_buff *skb;
473 struct udphdr *uh;
474 int err = 0;
475 int is_udplite = IS_UDPLITE(sk);
476 __wsum csum = 0;
477
478 /* Grab the skbuff where UDP header space exists. */
479 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
480 goto out;
481
482 /*
483 * Create a UDP header
484 */
485 uh = udp_hdr(skb);
486 uh->source = fl->fl_ip_sport;
487 uh->dest = fl->fl_ip_dport;
488 uh->len = htons(up->len);
489 uh->check = 0;
490
491 if (is_udplite) /* UDP-Lite */
492 csum = udplite_csum_outgoing(sk, skb);
493
494 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
495
496 skb->ip_summed = CHECKSUM_NONE;
497 goto send;
498
499 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
500
501 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len);
502 goto send;
503
504 } else /* `normal' UDP */
505 csum = udp_csum_outgoing(sk, skb);
506
507 /* add protocol-dependent pseudo-header */
508 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
509 sk->sk_protocol, csum );
510 if (uh->check == 0)
511 uh->check = CSUM_MANGLED_0;
512
513 send:
514 err = ip_push_pending_frames(sk);
515 out:
516 up->len = 0;
517 up->pending = 0;
518 if (!err)
519 UDP_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS, is_udplite);
520 return err;
521 }
522
523 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
524 size_t len)
525 {
526 struct inet_sock *inet = inet_sk(sk);
527 struct udp_sock *up = udp_sk(sk);
528 int ulen = len;
529 struct ipcm_cookie ipc;
530 struct rtable *rt = NULL;
531 int free = 0;
532 int connected = 0;
533 __be32 daddr, faddr, saddr;
534 __be16 dport;
535 u8 tos;
536 int err, is_udplite = IS_UDPLITE(sk);
537 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
538 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
539
540 if (len > 0xFFFF)
541 return -EMSGSIZE;
542
543 /*
544 * Check the flags.
545 */
546
547 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */
548 return -EOPNOTSUPP;
549
550 ipc.opt = NULL;
551
552 if (up->pending) {
553 /*
554 * There are pending frames.
555 * The socket lock must be held while it's corked.
556 */
557 lock_sock(sk);
558 if (likely(up->pending)) {
559 if (unlikely(up->pending != AF_INET)) {
560 release_sock(sk);
561 return -EINVAL;
562 }
563 goto do_append_data;
564 }
565 release_sock(sk);
566 }
567 ulen += sizeof(struct udphdr);
568
569 /*
570 * Get and verify the address.
571 */
572 if (msg->msg_name) {
573 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name;
574 if (msg->msg_namelen < sizeof(*usin))
575 return -EINVAL;
576 if (usin->sin_family != AF_INET) {
577 if (usin->sin_family != AF_UNSPEC)
578 return -EAFNOSUPPORT;
579 }
580
581 daddr = usin->sin_addr.s_addr;
582 dport = usin->sin_port;
583 if (dport == 0)
584 return -EINVAL;
585 } else {
586 if (sk->sk_state != TCP_ESTABLISHED)
587 return -EDESTADDRREQ;
588 daddr = inet->daddr;
589 dport = inet->dport;
590 /* Open fast path for connected socket.
591 Route will not be used, if at least one option is set.
592 */
593 connected = 1;
594 }
595 ipc.addr = inet->saddr;
596
597 ipc.oif = sk->sk_bound_dev_if;
598 if (msg->msg_controllen) {
599 err = ip_cmsg_send(msg, &ipc);
600 if (err)
601 return err;
602 if (ipc.opt)
603 free = 1;
604 connected = 0;
605 }
606 if (!ipc.opt)
607 ipc.opt = inet->opt;
608
609 saddr = ipc.addr;
610 ipc.addr = faddr = daddr;
611
612 if (ipc.opt && ipc.opt->srr) {
613 if (!daddr)
614 return -EINVAL;
615 faddr = ipc.opt->faddr;
616 connected = 0;
617 }
618 tos = RT_TOS(inet->tos);
619 if (sock_flag(sk, SOCK_LOCALROUTE) ||
620 (msg->msg_flags & MSG_DONTROUTE) ||
621 (ipc.opt && ipc.opt->is_strictroute)) {
622 tos |= RTO_ONLINK;
623 connected = 0;
624 }
625
626 if (MULTICAST(daddr)) {
627 if (!ipc.oif)
628 ipc.oif = inet->mc_index;
629 if (!saddr)
630 saddr = inet->mc_addr;
631 connected = 0;
632 }
633
634 if (connected)
635 rt = (struct rtable*)sk_dst_check(sk, 0);
636
637 if (rt == NULL) {
638 struct flowi fl = { .oif = ipc.oif,
639 .nl_u = { .ip4_u =
640 { .daddr = faddr,
641 .saddr = saddr,
642 .tos = tos } },
643 .proto = sk->sk_protocol,
644 .uli_u = { .ports =
645 { .sport = inet->sport,
646 .dport = dport } } };
647 security_sk_classify_flow(sk, &fl);
648 err = ip_route_output_flow(&rt, &fl, sk, 1);
649 if (err) {
650 if (err == -ENETUNREACH)
651 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
652 goto out;
653 }
654
655 err = -EACCES;
656 if ((rt->rt_flags & RTCF_BROADCAST) &&
657 !sock_flag(sk, SOCK_BROADCAST))
658 goto out;
659 if (connected)
660 sk_dst_set(sk, dst_clone(&rt->u.dst));
661 }
662
663 if (msg->msg_flags&MSG_CONFIRM)
664 goto do_confirm;
665 back_from_confirm:
666
667 saddr = rt->rt_src;
668 if (!ipc.addr)
669 daddr = ipc.addr = rt->rt_dst;
670
671 lock_sock(sk);
672 if (unlikely(up->pending)) {
673 /* The socket is already corked while preparing it. */
674 /* ... which is an evident application bug. --ANK */
675 release_sock(sk);
676
677 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
678 err = -EINVAL;
679 goto out;
680 }
681 /*
682 * Now cork the socket to pend data.
683 */
684 inet->cork.fl.fl4_dst = daddr;
685 inet->cork.fl.fl_ip_dport = dport;
686 inet->cork.fl.fl4_src = saddr;
687 inet->cork.fl.fl_ip_sport = inet->sport;
688 up->pending = AF_INET;
689
690 do_append_data:
691 up->len += ulen;
692 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
693 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
694 sizeof(struct udphdr), &ipc, rt,
695 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
696 if (err)
697 udp_flush_pending_frames(sk);
698 else if (!corkreq)
699 err = udp_push_pending_frames(sk);
700 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
701 up->pending = 0;
702 release_sock(sk);
703
704 out:
705 ip_rt_put(rt);
706 if (free)
707 kfree(ipc.opt);
708 if (!err)
709 return len;
710 /*
711 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
712 * ENOBUFS might not be good (it's not tunable per se), but otherwise
713 * we don't have a good statistic (IpOutDiscards but it can be too many
714 * things). We could add another new stat but at least for now that
715 * seems like overkill.
716 */
717 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
718 UDP_INC_STATS_USER(UDP_MIB_SNDBUFERRORS, is_udplite);
719 }
720 return err;
721
722 do_confirm:
723 dst_confirm(&rt->u.dst);
724 if (!(msg->msg_flags&MSG_PROBE) || len)
725 goto back_from_confirm;
726 err = 0;
727 goto out;
728 }
729
730 int udp_sendpage(struct sock *sk, struct page *page, int offset,
731 size_t size, int flags)
732 {
733 struct udp_sock *up = udp_sk(sk);
734 int ret;
735
736 if (!up->pending) {
737 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
738
739 /* Call udp_sendmsg to specify destination address which
740 * sendpage interface can't pass.
741 * This will succeed only when the socket is connected.
742 */
743 ret = udp_sendmsg(NULL, sk, &msg, 0);
744 if (ret < 0)
745 return ret;
746 }
747
748 lock_sock(sk);
749
750 if (unlikely(!up->pending)) {
751 release_sock(sk);
752
753 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
754 return -EINVAL;
755 }
756
757 ret = ip_append_page(sk, page, offset, size, flags);
758 if (ret == -EOPNOTSUPP) {
759 release_sock(sk);
760 return sock_no_sendpage(sk->sk_socket, page, offset,
761 size, flags);
762 }
763 if (ret < 0) {
764 udp_flush_pending_frames(sk);
765 goto out;
766 }
767
768 up->len += size;
769 if (!(up->corkflag || (flags&MSG_MORE)))
770 ret = udp_push_pending_frames(sk);
771 if (!ret)
772 ret = size;
773 out:
774 release_sock(sk);
775 return ret;
776 }
777
778 /*
779 * IOCTL requests applicable to the UDP protocol
780 */
781
782 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
783 {
784 switch (cmd) {
785 case SIOCOUTQ:
786 {
787 int amount = atomic_read(&sk->sk_wmem_alloc);
788 return put_user(amount, (int __user *)arg);
789 }
790
791 case SIOCINQ:
792 {
793 struct sk_buff *skb;
794 unsigned long amount;
795
796 amount = 0;
797 spin_lock_bh(&sk->sk_receive_queue.lock);
798 skb = skb_peek(&sk->sk_receive_queue);
799 if (skb != NULL) {
800 /*
801 * We will only return the amount
802 * of this packet since that is all
803 * that will be read.
804 */
805 amount = skb->len - sizeof(struct udphdr);
806 }
807 spin_unlock_bh(&sk->sk_receive_queue.lock);
808 return put_user(amount, (int __user *)arg);
809 }
810
811 default:
812 return -ENOIOCTLCMD;
813 }
814
815 return 0;
816 }
817
818 /*
819 * This should be easy, if there is something there we
820 * return it, otherwise we block.
821 */
822
823 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
824 size_t len, int noblock, int flags, int *addr_len)
825 {
826 struct inet_sock *inet = inet_sk(sk);
827 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
828 struct sk_buff *skb;
829 unsigned int ulen, copied;
830 int peeked;
831 int err;
832 int is_udplite = IS_UDPLITE(sk);
833
834 /*
835 * Check any passed addresses
836 */
837 if (addr_len)
838 *addr_len=sizeof(*sin);
839
840 if (flags & MSG_ERRQUEUE)
841 return ip_recv_error(sk, msg, len);
842
843 try_again:
844 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
845 &peeked, &err);
846 if (!skb)
847 goto out;
848
849 ulen = skb->len - sizeof(struct udphdr);
850 copied = len;
851 if (copied > ulen)
852 copied = ulen;
853 else if (copied < ulen)
854 msg->msg_flags |= MSG_TRUNC;
855
856 /*
857 * If checksum is needed at all, try to do it while copying the
858 * data. If the data is truncated, or if we only want a partial
859 * coverage checksum (UDP-Lite), do it before the copy.
860 */
861
862 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
863 if (udp_lib_checksum_complete(skb))
864 goto csum_copy_err;
865 }
866
867 if (skb_csum_unnecessary(skb))
868 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
869 msg->msg_iov, copied );
870 else {
871 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
872
873 if (err == -EINVAL)
874 goto csum_copy_err;
875 }
876
877 if (err)
878 goto out_free;
879
880 if (!peeked)
881 UDP_INC_STATS_USER(UDP_MIB_INDATAGRAMS, is_udplite);
882
883 sock_recv_timestamp(msg, sk, skb);
884
885 /* Copy the address. */
886 if (sin)
887 {
888 sin->sin_family = AF_INET;
889 sin->sin_port = udp_hdr(skb)->source;
890 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
891 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
892 }
893 if (inet->cmsg_flags)
894 ip_cmsg_recv(msg, skb);
895
896 err = copied;
897 if (flags & MSG_TRUNC)
898 err = ulen;
899
900 out_free:
901 skb_free_datagram(sk, skb);
902 out:
903 return err;
904
905 csum_copy_err:
906 if (!skb_kill_datagram(sk, skb, flags))
907 UDP_INC_STATS_USER(UDP_MIB_INERRORS, is_udplite);
908
909 if (noblock)
910 return -EAGAIN;
911 goto try_again;
912 }
913
914
915 int udp_disconnect(struct sock *sk, int flags)
916 {
917 struct inet_sock *inet = inet_sk(sk);
918 /*
919 * 1003.1g - break association.
920 */
921
922 sk->sk_state = TCP_CLOSE;
923 inet->daddr = 0;
924 inet->dport = 0;
925 sk->sk_bound_dev_if = 0;
926 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
927 inet_reset_saddr(sk);
928
929 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
930 sk->sk_prot->unhash(sk);
931 inet->sport = 0;
932 }
933 sk_dst_reset(sk);
934 return 0;
935 }
936
937 /* returns:
938 * -1: error
939 * 0: success
940 * >0: "udp encap" protocol resubmission
941 *
942 * Note that in the success and error cases, the skb is assumed to
943 * have either been requeued or freed.
944 */
945 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
946 {
947 struct udp_sock *up = udp_sk(sk);
948 int rc;
949 int is_udplite = IS_UDPLITE(sk);
950
951 /*
952 * Charge it to the socket, dropping if the queue is full.
953 */
954 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
955 goto drop;
956 nf_reset(skb);
957
958 if (up->encap_type) {
959 /*
960 * This is an encapsulation socket so pass the skb to
961 * the socket's udp_encap_rcv() hook. Otherwise, just
962 * fall through and pass this up the UDP socket.
963 * up->encap_rcv() returns the following value:
964 * =0 if skb was successfully passed to the encap
965 * handler or was discarded by it.
966 * >0 if skb should be passed on to UDP.
967 * <0 if skb should be resubmitted as proto -N
968 */
969
970 /* if we're overly short, let UDP handle it */
971 if (skb->len > sizeof(struct udphdr) &&
972 up->encap_rcv != NULL) {
973 int ret;
974
975 ret = (*up->encap_rcv)(sk, skb);
976 if (ret <= 0) {
977 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS,
978 is_udplite);
979 return -ret;
980 }
981 }
982
983 /* FALLTHROUGH -- it's a UDP Packet */
984 }
985
986 /*
987 * UDP-Lite specific tests, ignored on UDP sockets
988 */
989 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
990
991 /*
992 * MIB statistics other than incrementing the error count are
993 * disabled for the following two types of errors: these depend
994 * on the application settings, not on the functioning of the
995 * protocol stack as such.
996 *
997 * RFC 3828 here recommends (sec 3.3): "There should also be a
998 * way ... to ... at least let the receiving application block
999 * delivery of packets with coverage values less than a value
1000 * provided by the application."
1001 */
1002 if (up->pcrlen == 0) { /* full coverage was set */
1003 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1004 "%d while full coverage %d requested\n",
1005 UDP_SKB_CB(skb)->cscov, skb->len);
1006 goto drop;
1007 }
1008 /* The next case involves violating the min. coverage requested
1009 * by the receiver. This is subtle: if receiver wants x and x is
1010 * greater than the buffersize/MTU then receiver will complain
1011 * that it wants x while sender emits packets of smaller size y.
1012 * Therefore the above ...()->partial_cov statement is essential.
1013 */
1014 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1015 LIMIT_NETDEBUG(KERN_WARNING
1016 "UDPLITE: coverage %d too small, need min %d\n",
1017 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1018 goto drop;
1019 }
1020 }
1021
1022 if (sk->sk_filter) {
1023 if (udp_lib_checksum_complete(skb))
1024 goto drop;
1025 }
1026
1027 if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) {
1028 /* Note that an ENOMEM error is charged twice */
1029 if (rc == -ENOMEM)
1030 UDP_INC_STATS_BH(UDP_MIB_RCVBUFERRORS, is_udplite);
1031 goto drop;
1032 }
1033
1034 return 0;
1035
1036 drop:
1037 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite);
1038 kfree_skb(skb);
1039 return -1;
1040 }
1041
1042 /*
1043 * Multicasts and broadcasts go to each listener.
1044 *
1045 * Note: called only from the BH handler context,
1046 * so we don't need to lock the hashes.
1047 */
1048 static int __udp4_lib_mcast_deliver(struct sk_buff *skb,
1049 struct udphdr *uh,
1050 __be32 saddr, __be32 daddr,
1051 struct hlist_head udptable[])
1052 {
1053 struct sock *sk;
1054 int dif;
1055
1056 read_lock(&udp_hash_lock);
1057 sk = sk_head(&udptable[ntohs(uh->dest) & (UDP_HTABLE_SIZE - 1)]);
1058 dif = skb->dev->ifindex;
1059 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
1060 if (sk) {
1061 struct sock *sknext = NULL;
1062
1063 do {
1064 struct sk_buff *skb1 = skb;
1065
1066 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr,
1067 uh->source, saddr, dif);
1068 if (sknext)
1069 skb1 = skb_clone(skb, GFP_ATOMIC);
1070
1071 if (skb1) {
1072 int ret = udp_queue_rcv_skb(sk, skb1);
1073 if (ret > 0)
1074 /* we should probably re-process instead
1075 * of dropping packets here. */
1076 kfree_skb(skb1);
1077 }
1078 sk = sknext;
1079 } while (sknext);
1080 } else
1081 kfree_skb(skb);
1082 read_unlock(&udp_hash_lock);
1083 return 0;
1084 }
1085
1086 /* Initialize UDP checksum. If exited with zero value (success),
1087 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1088 * Otherwise, csum completion requires chacksumming packet body,
1089 * including udp header and folding it to skb->csum.
1090 */
1091 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1092 int proto)
1093 {
1094 const struct iphdr *iph;
1095 int err;
1096
1097 UDP_SKB_CB(skb)->partial_cov = 0;
1098 UDP_SKB_CB(skb)->cscov = skb->len;
1099
1100 if (proto == IPPROTO_UDPLITE) {
1101 err = udplite_checksum_init(skb, uh);
1102 if (err)
1103 return err;
1104 }
1105
1106 iph = ip_hdr(skb);
1107 if (uh->check == 0) {
1108 skb->ip_summed = CHECKSUM_UNNECESSARY;
1109 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1110 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1111 proto, skb->csum))
1112 skb->ip_summed = CHECKSUM_UNNECESSARY;
1113 }
1114 if (!skb_csum_unnecessary(skb))
1115 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1116 skb->len, proto, 0);
1117 /* Probably, we should checksum udp header (it should be in cache
1118 * in any case) and data in tiny packets (< rx copybreak).
1119 */
1120
1121 return 0;
1122 }
1123
1124 /*
1125 * All we need to do is get the socket, and then do a checksum.
1126 */
1127
1128 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
1129 int proto)
1130 {
1131 struct sock *sk;
1132 struct udphdr *uh = udp_hdr(skb);
1133 unsigned short ulen;
1134 struct rtable *rt = (struct rtable*)skb->dst;
1135 __be32 saddr = ip_hdr(skb)->saddr;
1136 __be32 daddr = ip_hdr(skb)->daddr;
1137
1138 /*
1139 * Validate the packet.
1140 */
1141 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1142 goto drop; /* No space for header. */
1143
1144 ulen = ntohs(uh->len);
1145 if (ulen > skb->len)
1146 goto short_packet;
1147
1148 if (proto == IPPROTO_UDP) {
1149 /* UDP validates ulen. */
1150 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1151 goto short_packet;
1152 uh = udp_hdr(skb);
1153 }
1154
1155 if (udp4_csum_init(skb, uh, proto))
1156 goto csum_error;
1157
1158 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1159 return __udp4_lib_mcast_deliver(skb, uh, saddr, daddr, udptable);
1160
1161 sk = __udp4_lib_lookup(saddr, uh->source, daddr, uh->dest,
1162 inet_iif(skb), udptable);
1163
1164 if (sk != NULL) {
1165 int ret = udp_queue_rcv_skb(sk, skb);
1166 sock_put(sk);
1167
1168 /* a return value > 0 means to resubmit the input, but
1169 * it wants the return to be -protocol, or 0
1170 */
1171 if (ret > 0)
1172 return -ret;
1173 return 0;
1174 }
1175
1176 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1177 goto drop;
1178 nf_reset(skb);
1179
1180 /* No socket. Drop packet silently, if checksum is wrong */
1181 if (udp_lib_checksum_complete(skb))
1182 goto csum_error;
1183
1184 UDP_INC_STATS_BH(UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1185 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1186
1187 /*
1188 * Hmm. We got an UDP packet to a port to which we
1189 * don't wanna listen. Ignore it.
1190 */
1191 kfree_skb(skb);
1192 return 0;
1193
1194 short_packet:
1195 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %u.%u.%u.%u:%u %d/%d to %u.%u.%u.%u:%u\n",
1196 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1197 NIPQUAD(saddr),
1198 ntohs(uh->source),
1199 ulen,
1200 skb->len,
1201 NIPQUAD(daddr),
1202 ntohs(uh->dest));
1203 goto drop;
1204
1205 csum_error:
1206 /*
1207 * RFC1122: OK. Discards the bad packet silently (as far as
1208 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1209 */
1210 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %d.%d.%d.%d:%d to %d.%d.%d.%d:%d ulen %d\n",
1211 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1212 NIPQUAD(saddr),
1213 ntohs(uh->source),
1214 NIPQUAD(daddr),
1215 ntohs(uh->dest),
1216 ulen);
1217 drop:
1218 UDP_INC_STATS_BH(UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1219 kfree_skb(skb);
1220 return 0;
1221 }
1222
1223 int udp_rcv(struct sk_buff *skb)
1224 {
1225 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP);
1226 }
1227
1228 int udp_destroy_sock(struct sock *sk)
1229 {
1230 lock_sock(sk);
1231 udp_flush_pending_frames(sk);
1232 release_sock(sk);
1233 return 0;
1234 }
1235
1236 /*
1237 * Socket option code for UDP
1238 */
1239 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1240 char __user *optval, int optlen,
1241 int (*push_pending_frames)(struct sock *))
1242 {
1243 struct udp_sock *up = udp_sk(sk);
1244 int val;
1245 int err = 0;
1246 int is_udplite = IS_UDPLITE(sk);
1247
1248 if (optlen<sizeof(int))
1249 return -EINVAL;
1250
1251 if (get_user(val, (int __user *)optval))
1252 return -EFAULT;
1253
1254 switch (optname) {
1255 case UDP_CORK:
1256 if (val != 0) {
1257 up->corkflag = 1;
1258 } else {
1259 up->corkflag = 0;
1260 lock_sock(sk);
1261 (*push_pending_frames)(sk);
1262 release_sock(sk);
1263 }
1264 break;
1265
1266 case UDP_ENCAP:
1267 switch (val) {
1268 case 0:
1269 case UDP_ENCAP_ESPINUDP:
1270 case UDP_ENCAP_ESPINUDP_NON_IKE:
1271 up->encap_rcv = xfrm4_udp_encap_rcv;
1272 /* FALLTHROUGH */
1273 case UDP_ENCAP_L2TPINUDP:
1274 up->encap_type = val;
1275 break;
1276 default:
1277 err = -ENOPROTOOPT;
1278 break;
1279 }
1280 break;
1281
1282 /*
1283 * UDP-Lite's partial checksum coverage (RFC 3828).
1284 */
1285 /* The sender sets actual checksum coverage length via this option.
1286 * The case coverage > packet length is handled by send module. */
1287 case UDPLITE_SEND_CSCOV:
1288 if (!is_udplite) /* Disable the option on UDP sockets */
1289 return -ENOPROTOOPT;
1290 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1291 val = 8;
1292 up->pcslen = val;
1293 up->pcflag |= UDPLITE_SEND_CC;
1294 break;
1295
1296 /* The receiver specifies a minimum checksum coverage value. To make
1297 * sense, this should be set to at least 8 (as done below). If zero is
1298 * used, this again means full checksum coverage. */
1299 case UDPLITE_RECV_CSCOV:
1300 if (!is_udplite) /* Disable the option on UDP sockets */
1301 return -ENOPROTOOPT;
1302 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1303 val = 8;
1304 up->pcrlen = val;
1305 up->pcflag |= UDPLITE_RECV_CC;
1306 break;
1307
1308 default:
1309 err = -ENOPROTOOPT;
1310 break;
1311 }
1312
1313 return err;
1314 }
1315
1316 int udp_setsockopt(struct sock *sk, int level, int optname,
1317 char __user *optval, int optlen)
1318 {
1319 if (level == SOL_UDP || level == SOL_UDPLITE)
1320 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1321 udp_push_pending_frames);
1322 return ip_setsockopt(sk, level, optname, optval, optlen);
1323 }
1324
1325 #ifdef CONFIG_COMPAT
1326 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1327 char __user *optval, int optlen)
1328 {
1329 if (level == SOL_UDP || level == SOL_UDPLITE)
1330 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1331 udp_push_pending_frames);
1332 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1333 }
1334 #endif
1335
1336 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1337 char __user *optval, int __user *optlen)
1338 {
1339 struct udp_sock *up = udp_sk(sk);
1340 int val, len;
1341
1342 if (get_user(len,optlen))
1343 return -EFAULT;
1344
1345 len = min_t(unsigned int, len, sizeof(int));
1346
1347 if (len < 0)
1348 return -EINVAL;
1349
1350 switch (optname) {
1351 case UDP_CORK:
1352 val = up->corkflag;
1353 break;
1354
1355 case UDP_ENCAP:
1356 val = up->encap_type;
1357 break;
1358
1359 /* The following two cannot be changed on UDP sockets, the return is
1360 * always 0 (which corresponds to the full checksum coverage of UDP). */
1361 case UDPLITE_SEND_CSCOV:
1362 val = up->pcslen;
1363 break;
1364
1365 case UDPLITE_RECV_CSCOV:
1366 val = up->pcrlen;
1367 break;
1368
1369 default:
1370 return -ENOPROTOOPT;
1371 }
1372
1373 if (put_user(len, optlen))
1374 return -EFAULT;
1375 if (copy_to_user(optval, &val,len))
1376 return -EFAULT;
1377 return 0;
1378 }
1379
1380 int udp_getsockopt(struct sock *sk, int level, int optname,
1381 char __user *optval, int __user *optlen)
1382 {
1383 if (level == SOL_UDP || level == SOL_UDPLITE)
1384 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1385 return ip_getsockopt(sk, level, optname, optval, optlen);
1386 }
1387
1388 #ifdef CONFIG_COMPAT
1389 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1390 char __user *optval, int __user *optlen)
1391 {
1392 if (level == SOL_UDP || level == SOL_UDPLITE)
1393 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1394 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1395 }
1396 #endif
1397 /**
1398 * udp_poll - wait for a UDP event.
1399 * @file - file struct
1400 * @sock - socket
1401 * @wait - poll table
1402 *
1403 * This is same as datagram poll, except for the special case of
1404 * blocking sockets. If application is using a blocking fd
1405 * and a packet with checksum error is in the queue;
1406 * then it could get return from select indicating data available
1407 * but then block when reading it. Add special case code
1408 * to work around these arguably broken applications.
1409 */
1410 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1411 {
1412 unsigned int mask = datagram_poll(file, sock, wait);
1413 struct sock *sk = sock->sk;
1414 int is_lite = IS_UDPLITE(sk);
1415
1416 /* Check for false positives due to checksum errors */
1417 if ( (mask & POLLRDNORM) &&
1418 !(file->f_flags & O_NONBLOCK) &&
1419 !(sk->sk_shutdown & RCV_SHUTDOWN)){
1420 struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1421 struct sk_buff *skb;
1422
1423 spin_lock_bh(&rcvq->lock);
1424 while ((skb = skb_peek(rcvq)) != NULL &&
1425 udp_lib_checksum_complete(skb)) {
1426 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_lite);
1427 __skb_unlink(skb, rcvq);
1428 kfree_skb(skb);
1429 }
1430 spin_unlock_bh(&rcvq->lock);
1431
1432 /* nothing to see, move along */
1433 if (skb == NULL)
1434 mask &= ~(POLLIN | POLLRDNORM);
1435 }
1436
1437 return mask;
1438
1439 }
1440
1441 DEFINE_PROTO_INUSE(udp)
1442
1443 struct proto udp_prot = {
1444 .name = "UDP",
1445 .owner = THIS_MODULE,
1446 .close = udp_lib_close,
1447 .connect = ip4_datagram_connect,
1448 .disconnect = udp_disconnect,
1449 .ioctl = udp_ioctl,
1450 .destroy = udp_destroy_sock,
1451 .setsockopt = udp_setsockopt,
1452 .getsockopt = udp_getsockopt,
1453 .sendmsg = udp_sendmsg,
1454 .recvmsg = udp_recvmsg,
1455 .sendpage = udp_sendpage,
1456 .backlog_rcv = udp_queue_rcv_skb,
1457 .hash = udp_lib_hash,
1458 .unhash = udp_lib_unhash,
1459 .get_port = udp_v4_get_port,
1460 .obj_size = sizeof(struct udp_sock),
1461 #ifdef CONFIG_COMPAT
1462 .compat_setsockopt = compat_udp_setsockopt,
1463 .compat_getsockopt = compat_udp_getsockopt,
1464 #endif
1465 REF_PROTO_INUSE(udp)
1466 };
1467
1468 /* ------------------------------------------------------------------------ */
1469 #ifdef CONFIG_PROC_FS
1470
1471 static struct sock *udp_get_first(struct seq_file *seq)
1472 {
1473 struct sock *sk;
1474 struct udp_iter_state *state = seq->private;
1475
1476 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) {
1477 struct hlist_node *node;
1478 sk_for_each(sk, node, state->hashtable + state->bucket) {
1479 if (sk->sk_family == state->family)
1480 goto found;
1481 }
1482 }
1483 sk = NULL;
1484 found:
1485 return sk;
1486 }
1487
1488 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1489 {
1490 struct udp_iter_state *state = seq->private;
1491
1492 do {
1493 sk = sk_next(sk);
1494 try_again:
1495 ;
1496 } while (sk && sk->sk_family != state->family);
1497
1498 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) {
1499 sk = sk_head(state->hashtable + state->bucket);
1500 goto try_again;
1501 }
1502 return sk;
1503 }
1504
1505 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1506 {
1507 struct sock *sk = udp_get_first(seq);
1508
1509 if (sk)
1510 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1511 --pos;
1512 return pos ? NULL : sk;
1513 }
1514
1515 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1516 {
1517 read_lock(&udp_hash_lock);
1518 return *pos ? udp_get_idx(seq, *pos-1) : (void *)1;
1519 }
1520
1521 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1522 {
1523 struct sock *sk;
1524
1525 if (v == (void *)1)
1526 sk = udp_get_idx(seq, 0);
1527 else
1528 sk = udp_get_next(seq, v);
1529
1530 ++*pos;
1531 return sk;
1532 }
1533
1534 static void udp_seq_stop(struct seq_file *seq, void *v)
1535 {
1536 read_unlock(&udp_hash_lock);
1537 }
1538
1539 static int udp_seq_open(struct inode *inode, struct file *file)
1540 {
1541 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1542 struct seq_file *seq;
1543 int rc = -ENOMEM;
1544 struct udp_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL);
1545
1546 if (!s)
1547 goto out;
1548 s->family = afinfo->family;
1549 s->hashtable = afinfo->hashtable;
1550 s->seq_ops.start = udp_seq_start;
1551 s->seq_ops.next = udp_seq_next;
1552 s->seq_ops.show = afinfo->seq_show;
1553 s->seq_ops.stop = udp_seq_stop;
1554
1555 rc = seq_open(file, &s->seq_ops);
1556 if (rc)
1557 goto out_kfree;
1558
1559 seq = file->private_data;
1560 seq->private = s;
1561 out:
1562 return rc;
1563 out_kfree:
1564 kfree(s);
1565 goto out;
1566 }
1567
1568 /* ------------------------------------------------------------------------ */
1569 int udp_proc_register(struct udp_seq_afinfo *afinfo)
1570 {
1571 struct proc_dir_entry *p;
1572 int rc = 0;
1573
1574 if (!afinfo)
1575 return -EINVAL;
1576 afinfo->seq_fops->owner = afinfo->owner;
1577 afinfo->seq_fops->open = udp_seq_open;
1578 afinfo->seq_fops->read = seq_read;
1579 afinfo->seq_fops->llseek = seq_lseek;
1580 afinfo->seq_fops->release = seq_release_private;
1581
1582 p = proc_net_fops_create(&init_net, afinfo->name, S_IRUGO, afinfo->seq_fops);
1583 if (p)
1584 p->data = afinfo;
1585 else
1586 rc = -ENOMEM;
1587 return rc;
1588 }
1589
1590 void udp_proc_unregister(struct udp_seq_afinfo *afinfo)
1591 {
1592 if (!afinfo)
1593 return;
1594 proc_net_remove(&init_net, afinfo->name);
1595 memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops));
1596 }
1597
1598 /* ------------------------------------------------------------------------ */
1599 static void udp4_format_sock(struct sock *sp, char *tmpbuf, int bucket)
1600 {
1601 struct inet_sock *inet = inet_sk(sp);
1602 __be32 dest = inet->daddr;
1603 __be32 src = inet->rcv_saddr;
1604 __u16 destp = ntohs(inet->dport);
1605 __u16 srcp = ntohs(inet->sport);
1606
1607 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
1608 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p",
1609 bucket, src, srcp, dest, destp, sp->sk_state,
1610 atomic_read(&sp->sk_wmem_alloc),
1611 atomic_read(&sp->sk_rmem_alloc),
1612 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1613 atomic_read(&sp->sk_refcnt), sp);
1614 }
1615
1616 int udp4_seq_show(struct seq_file *seq, void *v)
1617 {
1618 if (v == SEQ_START_TOKEN)
1619 seq_printf(seq, "%-127s\n",
1620 " sl local_address rem_address st tx_queue "
1621 "rx_queue tr tm->when retrnsmt uid timeout "
1622 "inode");
1623 else {
1624 char tmpbuf[129];
1625 struct udp_iter_state *state = seq->private;
1626
1627 udp4_format_sock(v, tmpbuf, state->bucket);
1628 seq_printf(seq, "%-127s\n", tmpbuf);
1629 }
1630 return 0;
1631 }
1632
1633 /* ------------------------------------------------------------------------ */
1634 static struct file_operations udp4_seq_fops;
1635 static struct udp_seq_afinfo udp4_seq_afinfo = {
1636 .owner = THIS_MODULE,
1637 .name = "udp",
1638 .family = AF_INET,
1639 .hashtable = udp_hash,
1640 .seq_show = udp4_seq_show,
1641 .seq_fops = &udp4_seq_fops,
1642 };
1643
1644 int __init udp4_proc_init(void)
1645 {
1646 return udp_proc_register(&udp4_seq_afinfo);
1647 }
1648
1649 void udp4_proc_exit(void)
1650 {
1651 udp_proc_unregister(&udp4_seq_afinfo);
1652 }
1653 #endif /* CONFIG_PROC_FS */
1654
1655 EXPORT_SYMBOL(udp_disconnect);
1656 EXPORT_SYMBOL(udp_hash);
1657 EXPORT_SYMBOL(udp_hash_lock);
1658 EXPORT_SYMBOL(udp_ioctl);
1659 EXPORT_SYMBOL(udp_get_port);
1660 EXPORT_SYMBOL(udp_prot);
1661 EXPORT_SYMBOL(udp_sendmsg);
1662 EXPORT_SYMBOL(udp_lib_getsockopt);
1663 EXPORT_SYMBOL(udp_lib_setsockopt);
1664 EXPORT_SYMBOL(udp_poll);
1665
1666 #ifdef CONFIG_PROC_FS
1667 EXPORT_SYMBOL(udp_proc_register);
1668 EXPORT_SYMBOL(udp_proc_unregister);
1669 #endif
Linux kernel - Deliverables
This page took 0.063269 seconds and 6 git commands to generate.