projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
[UDP]: Counter increment should be in USER mode for recvmsg
[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
114 struct hlist_head udp_hash[UDP_HTABLE_SIZE];
115 DEFINE_RWLOCK(udp_hash_lock);
116
117 static inline int __udp_lib_lport_inuse(__u16 num,
118 const 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 * @saddr_comp: AF-dependent comparison of bound local IP addresses
136 */
137 int __udp_lib_get_port(struct sock *sk, unsigned short snum,
138 struct hlist_head udptable[],
139 int (*saddr_comp)(const struct sock *sk1,
140 const struct sock *sk2 ) )
141 {
142 struct hlist_node *node;
143 struct hlist_head *head;
144 struct sock *sk2;
145 int error = 1;
146
147 write_lock_bh(&udp_hash_lock);
148
149 if (!snum) {
150 int i, low, high, remaining;
151 unsigned rover, best, best_size_so_far;
152
153 inet_get_local_port_range(&low, &high);
154 remaining = (high - low) + 1;
155
156 best_size_so_far = UINT_MAX;
157 best = rover = net_random() % remaining + low;
158
159 /* 1st pass: look for empty (or shortest) hash chain */
160 for (i = 0; i < UDP_HTABLE_SIZE; i++) {
161 int size = 0;
162
163 head = &udptable[rover & (UDP_HTABLE_SIZE - 1)];
164 if (hlist_empty(head))
165 goto gotit;
166
167 sk_for_each(sk2, node, head) {
168 if (++size >= best_size_so_far)
169 goto next;
170 }
171 best_size_so_far = size;
172 best = rover;
173 next:
174 /* fold back if end of range */
175 if (++rover > high)
176 rover = low + ((rover - low)
177 & (UDP_HTABLE_SIZE - 1));
178
179
180 }
181
182 /* 2nd pass: find hole in shortest hash chain */
183 rover = best;
184 for (i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; i++) {
185 if (! __udp_lib_lport_inuse(rover, udptable))
186 goto gotit;
187 rover += UDP_HTABLE_SIZE;
188 if (rover > high)
189 rover = low + ((rover - low)
190 & (UDP_HTABLE_SIZE - 1));
191 }
192
193
194 /* All ports in use! */
195 goto fail;
196
197 gotit:
198 snum = rover;
199 } else {
200 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
201
202 sk_for_each(sk2, node, head)
203 if (sk2->sk_hash == snum &&
204 sk2 != sk &&
205 (!sk2->sk_reuse || !sk->sk_reuse) &&
206 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if
207 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
208 (*saddr_comp)(sk, sk2) )
209 goto fail;
210 }
211
212 inet_sk(sk)->num = snum;
213 sk->sk_hash = snum;
214 if (sk_unhashed(sk)) {
215 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
216 sk_add_node(sk, head);
217 sock_prot_inc_use(sk->sk_prot);
218 }
219 error = 0;
220 fail:
221 write_unlock_bh(&udp_hash_lock);
222 return error;
223 }
224
225 int udp_get_port(struct sock *sk, unsigned short snum,
226 int (*scmp)(const struct sock *, const struct sock *))
227 {
228 return __udp_lib_get_port(sk, snum, udp_hash, scmp);
229 }
230
231 int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
232 {
233 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
234
235 return ( !ipv6_only_sock(sk2) &&
236 (!inet1->rcv_saddr || !inet2->rcv_saddr ||
237 inet1->rcv_saddr == inet2->rcv_saddr ));
238 }
239
240 static inline int udp_v4_get_port(struct sock *sk, unsigned short snum)
241 {
242 return udp_get_port(sk, snum, ipv4_rcv_saddr_equal);
243 }
244
245 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
246 * harder than this. -DaveM
247 */
248 static struct sock *__udp4_lib_lookup(__be32 saddr, __be16 sport,
249 __be32 daddr, __be16 dport,
250 int dif, struct hlist_head udptable[])
251 {
252 struct sock *sk, *result = NULL;
253 struct hlist_node *node;
254 unsigned short hnum = ntohs(dport);
255 int badness = -1;
256
257 read_lock(&udp_hash_lock);
258 sk_for_each(sk, node, &udptable[hnum & (UDP_HTABLE_SIZE - 1)]) {
259 struct inet_sock *inet = inet_sk(sk);
260
261 if (sk->sk_hash == hnum && !ipv6_only_sock(sk)) {
262 int score = (sk->sk_family == PF_INET ? 1 : 0);
263 if (inet->rcv_saddr) {
264 if (inet->rcv_saddr != daddr)
265 continue;
266 score+=2;
267 }
268 if (inet->daddr) {
269 if (inet->daddr != saddr)
270 continue;
271 score+=2;
272 }
273 if (inet->dport) {
274 if (inet->dport != sport)
275 continue;
276 score+=2;
277 }
278 if (sk->sk_bound_dev_if) {
279 if (sk->sk_bound_dev_if != dif)
280 continue;
281 score+=2;
282 }
283 if (score == 9) {
284 result = sk;
285 break;
286 } else if (score > badness) {
287 result = sk;
288 badness = score;
289 }
290 }
291 }
292 if (result)
293 sock_hold(result);
294 read_unlock(&udp_hash_lock);
295 return result;
296 }
297
298 static inline struct sock *udp_v4_mcast_next(struct sock *sk,
299 __be16 loc_port, __be32 loc_addr,
300 __be16 rmt_port, __be32 rmt_addr,
301 int dif)
302 {
303 struct hlist_node *node;
304 struct sock *s = sk;
305 unsigned short hnum = ntohs(loc_port);
306
307 sk_for_each_from(s, node) {
308 struct inet_sock *inet = inet_sk(s);
309
310 if (s->sk_hash != hnum ||
311 (inet->daddr && inet->daddr != rmt_addr) ||
312 (inet->dport != rmt_port && inet->dport) ||
313 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) ||
314 ipv6_only_sock(s) ||
315 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
316 continue;
317 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
318 continue;
319 goto found;
320 }
321 s = NULL;
322 found:
323 return s;
324 }
325
326 /*
327 * This routine is called by the ICMP module when it gets some
328 * sort of error condition. If err < 0 then the socket should
329 * be closed and the error returned to the user. If err > 0
330 * it's just the icmp type << 8 | icmp code.
331 * Header points to the ip header of the error packet. We move
332 * on past this. Then (as it used to claim before adjustment)
333 * header points to the first 8 bytes of the udp header. We need
334 * to find the appropriate port.
335 */
336
337 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[])
338 {
339 struct inet_sock *inet;
340 struct iphdr *iph = (struct iphdr*)skb->data;
341 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2));
342 const int type = icmp_hdr(skb)->type;
343 const int code = icmp_hdr(skb)->code;
344 struct sock *sk;
345 int harderr;
346 int err;
347
348 sk = __udp4_lib_lookup(iph->daddr, uh->dest, iph->saddr, uh->source,
349 skb->dev->ifindex, udptable );
350 if (sk == NULL) {
351 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
352 return; /* No socket for error */
353 }
354
355 err = 0;
356 harderr = 0;
357 inet = inet_sk(sk);
358
359 switch (type) {
360 default:
361 case ICMP_TIME_EXCEEDED:
362 err = EHOSTUNREACH;
363 break;
364 case ICMP_SOURCE_QUENCH:
365 goto out;
366 case ICMP_PARAMETERPROB:
367 err = EPROTO;
368 harderr = 1;
369 break;
370 case ICMP_DEST_UNREACH:
371 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
372 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
373 err = EMSGSIZE;
374 harderr = 1;
375 break;
376 }
377 goto out;
378 }
379 err = EHOSTUNREACH;
380 if (code <= NR_ICMP_UNREACH) {
381 harderr = icmp_err_convert[code].fatal;
382 err = icmp_err_convert[code].errno;
383 }
384 break;
385 }
386
387 /*
388 * RFC1122: OK. Passes ICMP errors back to application, as per
389 * 4.1.3.3.
390 */
391 if (!inet->recverr) {
392 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
393 goto out;
394 } else {
395 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1));
396 }
397 sk->sk_err = err;
398 sk->sk_error_report(sk);
399 out:
400 sock_put(sk);
401 }
402
403 void udp_err(struct sk_buff *skb, u32 info)
404 {
405 return __udp4_lib_err(skb, info, udp_hash);
406 }
407
408 /*
409 * Throw away all pending data and cancel the corking. Socket is locked.
410 */
411 static void udp_flush_pending_frames(struct sock *sk)
412 {
413 struct udp_sock *up = udp_sk(sk);
414
415 if (up->pending) {
416 up->len = 0;
417 up->pending = 0;
418 ip_flush_pending_frames(sk);
419 }
420 }
421
422 /**
423 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
424 * @sk: socket we are sending on
425 * @skb: sk_buff containing the filled-in UDP header
426 * (checksum field must be zeroed out)
427 */
428 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
429 __be32 src, __be32 dst, int len )
430 {
431 unsigned int offset;
432 struct udphdr *uh = udp_hdr(skb);
433 __wsum csum = 0;
434
435 if (skb_queue_len(&sk->sk_write_queue) == 1) {
436 /*
437 * Only one fragment on the socket.
438 */
439 skb->csum_start = skb_transport_header(skb) - skb->head;
440 skb->csum_offset = offsetof(struct udphdr, check);
441 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
442 } else {
443 /*
444 * HW-checksum won't work as there are two or more
445 * fragments on the socket so that all csums of sk_buffs
446 * should be together
447 */
448 offset = skb_transport_offset(skb);
449 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
450
451 skb->ip_summed = CHECKSUM_NONE;
452
453 skb_queue_walk(&sk->sk_write_queue, skb) {
454 csum = csum_add(csum, skb->csum);
455 }
456
457 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
458 if (uh->check == 0)
459 uh->check = CSUM_MANGLED_0;
460 }
461 }
462
463 /*
464 * Push out all pending data as one UDP datagram. Socket is locked.
465 */
466 static int udp_push_pending_frames(struct sock *sk)
467 {
468 struct udp_sock *up = udp_sk(sk);
469 struct inet_sock *inet = inet_sk(sk);
470 struct flowi *fl = &inet->cork.fl;
471 struct sk_buff *skb;
472 struct udphdr *uh;
473 int err = 0;
474 int is_udplite = IS_UDPLITE(sk);
475 __wsum csum = 0;
476
477 /* Grab the skbuff where UDP header space exists. */
478 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
479 goto out;
480
481 /*
482 * Create a UDP header
483 */
484 uh = udp_hdr(skb);
485 uh->source = fl->fl_ip_sport;
486 uh->dest = fl->fl_ip_dport;
487 uh->len = htons(up->len);
488 uh->check = 0;
489
490 if (is_udplite) /* UDP-Lite */
491 csum = udplite_csum_outgoing(sk, skb);
492
493 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
494
495 skb->ip_summed = CHECKSUM_NONE;
496 goto send;
497
498 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
499
500 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len);
501 goto send;
502
503 } else /* `normal' UDP */
504 csum = udp_csum_outgoing(sk, skb);
505
506 /* add protocol-dependent pseudo-header */
507 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
508 sk->sk_protocol, csum );
509 if (uh->check == 0)
510 uh->check = CSUM_MANGLED_0;
511
512 send:
513 err = ip_push_pending_frames(sk);
514 out:
515 up->len = 0;
516 up->pending = 0;
517 if (!err)
518 UDP_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS, is_udplite);
519 return err;
520 }
521
522 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
523 size_t len)
524 {
525 struct inet_sock *inet = inet_sk(sk);
526 struct udp_sock *up = udp_sk(sk);
527 int ulen = len;
528 struct ipcm_cookie ipc;
529 struct rtable *rt = NULL;
530 int free = 0;
531 int connected = 0;
532 __be32 daddr, faddr, saddr;
533 __be16 dport;
534 u8 tos;
535 int err, is_udplite = IS_UDPLITE(sk);
536 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
537 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
538
539 if (len > 0xFFFF)
540 return -EMSGSIZE;
541
542 /*
543 * Check the flags.
544 */
545
546 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */
547 return -EOPNOTSUPP;
548
549 ipc.opt = NULL;
550
551 if (up->pending) {
552 /*
553 * There are pending frames.
554 * The socket lock must be held while it's corked.
555 */
556 lock_sock(sk);
557 if (likely(up->pending)) {
558 if (unlikely(up->pending != AF_INET)) {
559 release_sock(sk);
560 return -EINVAL;
561 }
562 goto do_append_data;
563 }
564 release_sock(sk);
565 }
566 ulen += sizeof(struct udphdr);
567
568 /*
569 * Get and verify the address.
570 */
571 if (msg->msg_name) {
572 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name;
573 if (msg->msg_namelen < sizeof(*usin))
574 return -EINVAL;
575 if (usin->sin_family != AF_INET) {
576 if (usin->sin_family != AF_UNSPEC)
577 return -EAFNOSUPPORT;
578 }
579
580 daddr = usin->sin_addr.s_addr;
581 dport = usin->sin_port;
582 if (dport == 0)
583 return -EINVAL;
584 } else {
585 if (sk->sk_state != TCP_ESTABLISHED)
586 return -EDESTADDRREQ;
587 daddr = inet->daddr;
588 dport = inet->dport;
589 /* Open fast path for connected socket.
590 Route will not be used, if at least one option is set.
591 */
592 connected = 1;
593 }
594 ipc.addr = inet->saddr;
595
596 ipc.oif = sk->sk_bound_dev_if;
597 if (msg->msg_controllen) {
598 err = ip_cmsg_send(msg, &ipc);
599 if (err)
600 return err;
601 if (ipc.opt)
602 free = 1;
603 connected = 0;
604 }
605 if (!ipc.opt)
606 ipc.opt = inet->opt;
607
608 saddr = ipc.addr;
609 ipc.addr = faddr = daddr;
610
611 if (ipc.opt && ipc.opt->srr) {
612 if (!daddr)
613 return -EINVAL;
614 faddr = ipc.opt->faddr;
615 connected = 0;
616 }
617 tos = RT_TOS(inet->tos);
618 if (sock_flag(sk, SOCK_LOCALROUTE) ||
619 (msg->msg_flags & MSG_DONTROUTE) ||
620 (ipc.opt && ipc.opt->is_strictroute)) {
621 tos |= RTO_ONLINK;
622 connected = 0;
623 }
624
625 if (MULTICAST(daddr)) {
626 if (!ipc.oif)
627 ipc.oif = inet->mc_index;
628 if (!saddr)
629 saddr = inet->mc_addr;
630 connected = 0;
631 }
632
633 if (connected)
634 rt = (struct rtable*)sk_dst_check(sk, 0);
635
636 if (rt == NULL) {
637 struct flowi fl = { .oif = ipc.oif,
638 .nl_u = { .ip4_u =
639 { .daddr = faddr,
640 .saddr = saddr,
641 .tos = tos } },
642 .proto = sk->sk_protocol,
643 .uli_u = { .ports =
644 { .sport = inet->sport,
645 .dport = dport } } };
646 security_sk_classify_flow(sk, &fl);
647 err = ip_route_output_flow(&rt, &fl, sk, 1);
648 if (err) {
649 if (err == -ENETUNREACH)
650 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
651 goto out;
652 }
653
654 err = -EACCES;
655 if ((rt->rt_flags & RTCF_BROADCAST) &&
656 !sock_flag(sk, SOCK_BROADCAST))
657 goto out;
658 if (connected)
659 sk_dst_set(sk, dst_clone(&rt->u.dst));
660 }
661
662 if (msg->msg_flags&MSG_CONFIRM)
663 goto do_confirm;
664 back_from_confirm:
665
666 saddr = rt->rt_src;
667 if (!ipc.addr)
668 daddr = ipc.addr = rt->rt_dst;
669
670 lock_sock(sk);
671 if (unlikely(up->pending)) {
672 /* The socket is already corked while preparing it. */
673 /* ... which is an evident application bug. --ANK */
674 release_sock(sk);
675
676 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
677 err = -EINVAL;
678 goto out;
679 }
680 /*
681 * Now cork the socket to pend data.
682 */
683 inet->cork.fl.fl4_dst = daddr;
684 inet->cork.fl.fl_ip_dport = dport;
685 inet->cork.fl.fl4_src = saddr;
686 inet->cork.fl.fl_ip_sport = inet->sport;
687 up->pending = AF_INET;
688
689 do_append_data:
690 up->len += ulen;
691 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
692 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
693 sizeof(struct udphdr), &ipc, rt,
694 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
695 if (err)
696 udp_flush_pending_frames(sk);
697 else if (!corkreq)
698 err = udp_push_pending_frames(sk);
699 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
700 up->pending = 0;
701 release_sock(sk);
702
703 out:
704 ip_rt_put(rt);
705 if (free)
706 kfree(ipc.opt);
707 if (!err)
708 return len;
709 /*
710 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
711 * ENOBUFS might not be good (it's not tunable per se), but otherwise
712 * we don't have a good statistic (IpOutDiscards but it can be too many
713 * things). We could add another new stat but at least for now that
714 * seems like overkill.
715 */
716 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
717 UDP_INC_STATS_USER(UDP_MIB_SNDBUFERRORS, is_udplite);
718 }
719 return err;
720
721 do_confirm:
722 dst_confirm(&rt->u.dst);
723 if (!(msg->msg_flags&MSG_PROBE) || len)
724 goto back_from_confirm;
725 err = 0;
726 goto out;
727 }
728
729 int udp_sendpage(struct sock *sk, struct page *page, int offset,
730 size_t size, int flags)
731 {
732 struct udp_sock *up = udp_sk(sk);
733 int ret;
734
735 if (!up->pending) {
736 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
737
738 /* Call udp_sendmsg to specify destination address which
739 * sendpage interface can't pass.
740 * This will succeed only when the socket is connected.
741 */
742 ret = udp_sendmsg(NULL, sk, &msg, 0);
743 if (ret < 0)
744 return ret;
745 }
746
747 lock_sock(sk);
748
749 if (unlikely(!up->pending)) {
750 release_sock(sk);
751
752 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
753 return -EINVAL;
754 }
755
756 ret = ip_append_page(sk, page, offset, size, flags);
757 if (ret == -EOPNOTSUPP) {
758 release_sock(sk);
759 return sock_no_sendpage(sk->sk_socket, page, offset,
760 size, flags);
761 }
762 if (ret < 0) {
763 udp_flush_pending_frames(sk);
764 goto out;
765 }
766
767 up->len += size;
768 if (!(up->corkflag || (flags&MSG_MORE)))
769 ret = udp_push_pending_frames(sk);
770 if (!ret)
771 ret = size;
772 out:
773 release_sock(sk);
774 return ret;
775 }
776
777 /*
778 * IOCTL requests applicable to the UDP protocol
779 */
780
781 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
782 {
783 switch (cmd) {
784 case SIOCOUTQ:
785 {
786 int amount = atomic_read(&sk->sk_wmem_alloc);
787 return put_user(amount, (int __user *)arg);
788 }
789
790 case SIOCINQ:
791 {
792 struct sk_buff *skb;
793 unsigned long amount;
794
795 amount = 0;
796 spin_lock_bh(&sk->sk_receive_queue.lock);
797 skb = skb_peek(&sk->sk_receive_queue);
798 if (skb != NULL) {
799 /*
800 * We will only return the amount
801 * of this packet since that is all
802 * that will be read.
803 */
804 amount = skb->len - sizeof(struct udphdr);
805 }
806 spin_unlock_bh(&sk->sk_receive_queue.lock);
807 return put_user(amount, (int __user *)arg);
808 }
809
810 default:
811 return -ENOIOCTLCMD;
812 }
813
814 return 0;
815 }
816
817 /*
818 * This should be easy, if there is something there we
819 * return it, otherwise we block.
820 */
821
822 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
823 size_t len, int noblock, int flags, int *addr_len)
824 {
825 struct inet_sock *inet = inet_sk(sk);
826 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
827 struct sk_buff *skb;
828 unsigned int ulen, copied;
829 int err;
830 int is_udplite = IS_UDPLITE(sk);
831
832 /*
833 * Check any passed addresses
834 */
835 if (addr_len)
836 *addr_len=sizeof(*sin);
837
838 if (flags & MSG_ERRQUEUE)
839 return ip_recv_error(sk, msg, len);
840
841 try_again:
842 skb = skb_recv_datagram(sk, flags, noblock, &err);
843 if (!skb)
844 goto out;
845
846 ulen = skb->len - sizeof(struct udphdr);
847 copied = len;
848 if (copied > ulen)
849 copied = ulen;
850 else if (copied < ulen)
851 msg->msg_flags |= MSG_TRUNC;
852
853 /*
854 * If checksum is needed at all, try to do it while copying the
855 * data. If the data is truncated, or if we only want a partial
856 * coverage checksum (UDP-Lite), do it before the copy.
857 */
858
859 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
860 if (udp_lib_checksum_complete(skb))
861 goto csum_copy_err;
862 }
863
864 if (skb_csum_unnecessary(skb))
865 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
866 msg->msg_iov, copied );
867 else {
868 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
869
870 if (err == -EINVAL)
871 goto csum_copy_err;
872 }
873
874 if (err)
875 goto out_free;
876
877 UDP_INC_STATS_USER(UDP_MIB_INDATAGRAMS, is_udplite);
878
879 sock_recv_timestamp(msg, sk, skb);
880
881 /* Copy the address. */
882 if (sin)
883 {
884 sin->sin_family = AF_INET;
885 sin->sin_port = udp_hdr(skb)->source;
886 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
887 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
888 }
889 if (inet->cmsg_flags)
890 ip_cmsg_recv(msg, skb);
891
892 err = copied;
893 if (flags & MSG_TRUNC)
894 err = ulen;
895
896 out_free:
897 skb_free_datagram(sk, skb);
898 out:
899 return err;
900
901 csum_copy_err:
902 UDP_INC_STATS_USER(UDP_MIB_INERRORS, is_udplite);
903
904 skb_kill_datagram(sk, skb, flags);
905
906 if (noblock)
907 return -EAGAIN;
908 goto try_again;
909 }
910
911
912 int udp_disconnect(struct sock *sk, int flags)
913 {
914 struct inet_sock *inet = inet_sk(sk);
915 /*
916 * 1003.1g - break association.
917 */
918
919 sk->sk_state = TCP_CLOSE;
920 inet->daddr = 0;
921 inet->dport = 0;
922 sk->sk_bound_dev_if = 0;
923 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
924 inet_reset_saddr(sk);
925
926 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
927 sk->sk_prot->unhash(sk);
928 inet->sport = 0;
929 }
930 sk_dst_reset(sk);
931 return 0;
932 }
933
934 /* returns:
935 * -1: error
936 * 0: success
937 * >0: "udp encap" protocol resubmission
938 *
939 * Note that in the success and error cases, the skb is assumed to
940 * have either been requeued or freed.
941 */
942 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
943 {
944 struct udp_sock *up = udp_sk(sk);
945 int rc;
946 int is_udplite = IS_UDPLITE(sk);
947
948 /*
949 * Charge it to the socket, dropping if the queue is full.
950 */
951 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
952 goto drop;
953 nf_reset(skb);
954
955 if (up->encap_type) {
956 /*
957 * This is an encapsulation socket so pass the skb to
958 * the socket's udp_encap_rcv() hook. Otherwise, just
959 * fall through and pass this up the UDP socket.
960 * up->encap_rcv() returns the following value:
961 * =0 if skb was successfully passed to the encap
962 * handler or was discarded by it.
963 * >0 if skb should be passed on to UDP.
964 * <0 if skb should be resubmitted as proto -N
965 */
966
967 /* if we're overly short, let UDP handle it */
968 if (skb->len > sizeof(struct udphdr) &&
969 up->encap_rcv != NULL) {
970 int ret;
971
972 ret = (*up->encap_rcv)(sk, skb);
973 if (ret <= 0)
974 return -ret;
975 }
976
977 /* FALLTHROUGH -- it's a UDP Packet */
978 }
979
980 /*
981 * UDP-Lite specific tests, ignored on UDP sockets
982 */
983 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
984
985 /*
986 * MIB statistics other than incrementing the error count are
987 * disabled for the following two types of errors: these depend
988 * on the application settings, not on the functioning of the
989 * protocol stack as such.
990 *
991 * RFC 3828 here recommends (sec 3.3): "There should also be a
992 * way ... to ... at least let the receiving application block
993 * delivery of packets with coverage values less than a value
994 * provided by the application."
995 */
996 if (up->pcrlen == 0) { /* full coverage was set */
997 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
998 "%d while full coverage %d requested\n",
999 UDP_SKB_CB(skb)->cscov, skb->len);
1000 goto drop;
1001 }
1002 /* The next case involves violating the min. coverage requested
1003 * by the receiver. This is subtle: if receiver wants x and x is
1004 * greater than the buffersize/MTU then receiver will complain
1005 * that it wants x while sender emits packets of smaller size y.
1006 * Therefore the above ...()->partial_cov statement is essential.
1007 */
1008 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1009 LIMIT_NETDEBUG(KERN_WARNING
1010 "UDPLITE: coverage %d too small, need min %d\n",
1011 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1012 goto drop;
1013 }
1014 }
1015
1016 if (sk->sk_filter) {
1017 if (udp_lib_checksum_complete(skb))
1018 goto drop;
1019 }
1020
1021 if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) {
1022 /* Note that an ENOMEM error is charged twice */
1023 if (rc == -ENOMEM)
1024 UDP_INC_STATS_BH(UDP_MIB_RCVBUFERRORS, is_udplite);
1025 goto drop;
1026 }
1027
1028 return 0;
1029
1030 drop:
1031 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite);
1032 kfree_skb(skb);
1033 return -1;
1034 }
1035
1036 /*
1037 * Multicasts and broadcasts go to each listener.
1038 *
1039 * Note: called only from the BH handler context,
1040 * so we don't need to lock the hashes.
1041 */
1042 static int __udp4_lib_mcast_deliver(struct sk_buff *skb,
1043 struct udphdr *uh,
1044 __be32 saddr, __be32 daddr,
1045 struct hlist_head udptable[])
1046 {
1047 struct sock *sk;
1048 int dif;
1049
1050 read_lock(&udp_hash_lock);
1051 sk = sk_head(&udptable[ntohs(uh->dest) & (UDP_HTABLE_SIZE - 1)]);
1052 dif = skb->dev->ifindex;
1053 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
1054 if (sk) {
1055 struct sock *sknext = NULL;
1056
1057 do {
1058 struct sk_buff *skb1 = skb;
1059
1060 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr,
1061 uh->source, saddr, dif);
1062 if (sknext)
1063 skb1 = skb_clone(skb, GFP_ATOMIC);
1064
1065 if (skb1) {
1066 int ret = udp_queue_rcv_skb(sk, skb1);
1067 if (ret > 0)
1068 /* we should probably re-process instead
1069 * of dropping packets here. */
1070 kfree_skb(skb1);
1071 }
1072 sk = sknext;
1073 } while (sknext);
1074 } else
1075 kfree_skb(skb);
1076 read_unlock(&udp_hash_lock);
1077 return 0;
1078 }
1079
1080 /* Initialize UDP checksum. If exited with zero value (success),
1081 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1082 * Otherwise, csum completion requires chacksumming packet body,
1083 * including udp header and folding it to skb->csum.
1084 */
1085 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1086 int proto)
1087 {
1088 const struct iphdr *iph;
1089 int err;
1090
1091 UDP_SKB_CB(skb)->partial_cov = 0;
1092 UDP_SKB_CB(skb)->cscov = skb->len;
1093
1094 if (proto == IPPROTO_UDPLITE) {
1095 err = udplite_checksum_init(skb, uh);
1096 if (err)
1097 return err;
1098 }
1099
1100 iph = ip_hdr(skb);
1101 if (uh->check == 0) {
1102 skb->ip_summed = CHECKSUM_UNNECESSARY;
1103 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1104 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1105 proto, skb->csum))
1106 skb->ip_summed = CHECKSUM_UNNECESSARY;
1107 }
1108 if (!skb_csum_unnecessary(skb))
1109 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1110 skb->len, proto, 0);
1111 /* Probably, we should checksum udp header (it should be in cache
1112 * in any case) and data in tiny packets (< rx copybreak).
1113 */
1114
1115 return 0;
1116 }
1117
1118 /*
1119 * All we need to do is get the socket, and then do a checksum.
1120 */
1121
1122 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
1123 int proto)
1124 {
1125 struct sock *sk;
1126 struct udphdr *uh = udp_hdr(skb);
1127 unsigned short ulen;
1128 struct rtable *rt = (struct rtable*)skb->dst;
1129 __be32 saddr = ip_hdr(skb)->saddr;
1130 __be32 daddr = ip_hdr(skb)->daddr;
1131
1132 /*
1133 * Validate the packet.
1134 */
1135 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1136 goto drop; /* No space for header. */
1137
1138 ulen = ntohs(uh->len);
1139 if (ulen > skb->len)
1140 goto short_packet;
1141
1142 if (proto == IPPROTO_UDP) {
1143 /* UDP validates ulen. */
1144 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1145 goto short_packet;
1146 uh = udp_hdr(skb);
1147 }
1148
1149 if (udp4_csum_init(skb, uh, proto))
1150 goto csum_error;
1151
1152 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1153 return __udp4_lib_mcast_deliver(skb, uh, saddr, daddr, udptable);
1154
1155 sk = __udp4_lib_lookup(saddr, uh->source, daddr, uh->dest,
1156 inet_iif(skb), udptable);
1157
1158 if (sk != NULL) {
1159 int ret = udp_queue_rcv_skb(sk, skb);
1160 sock_put(sk);
1161
1162 /* a return value > 0 means to resubmit the input, but
1163 * it wants the return to be -protocol, or 0
1164 */
1165 if (ret > 0)
1166 return -ret;
1167 return 0;
1168 }
1169
1170 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1171 goto drop;
1172 nf_reset(skb);
1173
1174 /* No socket. Drop packet silently, if checksum is wrong */
1175 if (udp_lib_checksum_complete(skb))
1176 goto csum_error;
1177
1178 UDP_INC_STATS_BH(UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1179 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1180
1181 /*
1182 * Hmm. We got an UDP packet to a port to which we
1183 * don't wanna listen. Ignore it.
1184 */
1185 kfree_skb(skb);
1186 return 0;
1187
1188 short_packet:
1189 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %u.%u.%u.%u:%u %d/%d to %u.%u.%u.%u:%u\n",
1190 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1191 NIPQUAD(saddr),
1192 ntohs(uh->source),
1193 ulen,
1194 skb->len,
1195 NIPQUAD(daddr),
1196 ntohs(uh->dest));
1197 goto drop;
1198
1199 csum_error:
1200 /*
1201 * RFC1122: OK. Discards the bad packet silently (as far as
1202 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1203 */
1204 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %d.%d.%d.%d:%d to %d.%d.%d.%d:%d ulen %d\n",
1205 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1206 NIPQUAD(saddr),
1207 ntohs(uh->source),
1208 NIPQUAD(daddr),
1209 ntohs(uh->dest),
1210 ulen);
1211 drop:
1212 UDP_INC_STATS_BH(UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1213 kfree_skb(skb);
1214 return 0;
1215 }
1216
1217 int udp_rcv(struct sk_buff *skb)
1218 {
1219 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP);
1220 }
1221
1222 int udp_destroy_sock(struct sock *sk)
1223 {
1224 lock_sock(sk);
1225 udp_flush_pending_frames(sk);
1226 release_sock(sk);
1227 return 0;
1228 }
1229
1230 /*
1231 * Socket option code for UDP
1232 */
1233 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1234 char __user *optval, int optlen,
1235 int (*push_pending_frames)(struct sock *))
1236 {
1237 struct udp_sock *up = udp_sk(sk);
1238 int val;
1239 int err = 0;
1240 int is_udplite = IS_UDPLITE(sk);
1241
1242 if (optlen<sizeof(int))
1243 return -EINVAL;
1244
1245 if (get_user(val, (int __user *)optval))
1246 return -EFAULT;
1247
1248 switch (optname) {
1249 case UDP_CORK:
1250 if (val != 0) {
1251 up->corkflag = 1;
1252 } else {
1253 up->corkflag = 0;
1254 lock_sock(sk);
1255 (*push_pending_frames)(sk);
1256 release_sock(sk);
1257 }
1258 break;
1259
1260 case UDP_ENCAP:
1261 switch (val) {
1262 case 0:
1263 case UDP_ENCAP_ESPINUDP:
1264 case UDP_ENCAP_ESPINUDP_NON_IKE:
1265 up->encap_rcv = xfrm4_udp_encap_rcv;
1266 /* FALLTHROUGH */
1267 case UDP_ENCAP_L2TPINUDP:
1268 up->encap_type = val;
1269 break;
1270 default:
1271 err = -ENOPROTOOPT;
1272 break;
1273 }
1274 break;
1275
1276 /*
1277 * UDP-Lite's partial checksum coverage (RFC 3828).
1278 */
1279 /* The sender sets actual checksum coverage length via this option.
1280 * The case coverage > packet length is handled by send module. */
1281 case UDPLITE_SEND_CSCOV:
1282 if (!is_udplite) /* Disable the option on UDP sockets */
1283 return -ENOPROTOOPT;
1284 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1285 val = 8;
1286 up->pcslen = val;
1287 up->pcflag |= UDPLITE_SEND_CC;
1288 break;
1289
1290 /* The receiver specifies a minimum checksum coverage value. To make
1291 * sense, this should be set to at least 8 (as done below). If zero is
1292 * used, this again means full checksum coverage. */
1293 case UDPLITE_RECV_CSCOV:
1294 if (!is_udplite) /* Disable the option on UDP sockets */
1295 return -ENOPROTOOPT;
1296 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1297 val = 8;
1298 up->pcrlen = val;
1299 up->pcflag |= UDPLITE_RECV_CC;
1300 break;
1301
1302 default:
1303 err = -ENOPROTOOPT;
1304 break;
1305 }
1306
1307 return err;
1308 }
1309
1310 int udp_setsockopt(struct sock *sk, int level, int optname,
1311 char __user *optval, int optlen)
1312 {
1313 if (level == SOL_UDP || level == SOL_UDPLITE)
1314 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1315 udp_push_pending_frames);
1316 return ip_setsockopt(sk, level, optname, optval, optlen);
1317 }
1318
1319 #ifdef CONFIG_COMPAT
1320 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1321 char __user *optval, int optlen)
1322 {
1323 if (level == SOL_UDP || level == SOL_UDPLITE)
1324 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1325 udp_push_pending_frames);
1326 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1327 }
1328 #endif
1329
1330 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1331 char __user *optval, int __user *optlen)
1332 {
1333 struct udp_sock *up = udp_sk(sk);
1334 int val, len;
1335
1336 if (get_user(len,optlen))
1337 return -EFAULT;
1338
1339 len = min_t(unsigned int, len, sizeof(int));
1340
1341 if (len < 0)
1342 return -EINVAL;
1343
1344 switch (optname) {
1345 case UDP_CORK:
1346 val = up->corkflag;
1347 break;
1348
1349 case UDP_ENCAP:
1350 val = up->encap_type;
1351 break;
1352
1353 /* The following two cannot be changed on UDP sockets, the return is
1354 * always 0 (which corresponds to the full checksum coverage of UDP). */
1355 case UDPLITE_SEND_CSCOV:
1356 val = up->pcslen;
1357 break;
1358
1359 case UDPLITE_RECV_CSCOV:
1360 val = up->pcrlen;
1361 break;
1362
1363 default:
1364 return -ENOPROTOOPT;
1365 }
1366
1367 if (put_user(len, optlen))
1368 return -EFAULT;
1369 if (copy_to_user(optval, &val,len))
1370 return -EFAULT;
1371 return 0;
1372 }
1373
1374 int udp_getsockopt(struct sock *sk, int level, int optname,
1375 char __user *optval, int __user *optlen)
1376 {
1377 if (level == SOL_UDP || level == SOL_UDPLITE)
1378 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1379 return ip_getsockopt(sk, level, optname, optval, optlen);
1380 }
1381
1382 #ifdef CONFIG_COMPAT
1383 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1384 char __user *optval, int __user *optlen)
1385 {
1386 if (level == SOL_UDP || level == SOL_UDPLITE)
1387 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1388 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1389 }
1390 #endif
1391 /**
1392 * udp_poll - wait for a UDP event.
1393 * @file - file struct
1394 * @sock - socket
1395 * @wait - poll table
1396 *
1397 * This is same as datagram poll, except for the special case of
1398 * blocking sockets. If application is using a blocking fd
1399 * and a packet with checksum error is in the queue;
1400 * then it could get return from select indicating data available
1401 * but then block when reading it. Add special case code
1402 * to work around these arguably broken applications.
1403 */
1404 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1405 {
1406 unsigned int mask = datagram_poll(file, sock, wait);
1407 struct sock *sk = sock->sk;
1408 int is_lite = IS_UDPLITE(sk);
1409
1410 /* Check for false positives due to checksum errors */
1411 if ( (mask & POLLRDNORM) &&
1412 !(file->f_flags & O_NONBLOCK) &&
1413 !(sk->sk_shutdown & RCV_SHUTDOWN)){
1414 struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1415 struct sk_buff *skb;
1416
1417 spin_lock_bh(&rcvq->lock);
1418 while ((skb = skb_peek(rcvq)) != NULL &&
1419 udp_lib_checksum_complete(skb)) {
1420 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_lite);
1421 __skb_unlink(skb, rcvq);
1422 kfree_skb(skb);
1423 }
1424 spin_unlock_bh(&rcvq->lock);
1425
1426 /* nothing to see, move along */
1427 if (skb == NULL)
1428 mask &= ~(POLLIN | POLLRDNORM);
1429 }
1430
1431 return mask;
1432
1433 }
1434
1435 DEFINE_PROTO_INUSE(udp)
1436
1437 struct proto udp_prot = {
1438 .name = "UDP",
1439 .owner = THIS_MODULE,
1440 .close = udp_lib_close,
1441 .connect = ip4_datagram_connect,
1442 .disconnect = udp_disconnect,
1443 .ioctl = udp_ioctl,
1444 .destroy = udp_destroy_sock,
1445 .setsockopt = udp_setsockopt,
1446 .getsockopt = udp_getsockopt,
1447 .sendmsg = udp_sendmsg,
1448 .recvmsg = udp_recvmsg,
1449 .sendpage = udp_sendpage,
1450 .backlog_rcv = udp_queue_rcv_skb,
1451 .hash = udp_lib_hash,
1452 .unhash = udp_lib_unhash,
1453 .get_port = udp_v4_get_port,
1454 .obj_size = sizeof(struct udp_sock),
1455 #ifdef CONFIG_COMPAT
1456 .compat_setsockopt = compat_udp_setsockopt,
1457 .compat_getsockopt = compat_udp_getsockopt,
1458 #endif
1459 REF_PROTO_INUSE(udp)
1460 };
1461
1462 /* ------------------------------------------------------------------------ */
1463 #ifdef CONFIG_PROC_FS
1464
1465 static struct sock *udp_get_first(struct seq_file *seq)
1466 {
1467 struct sock *sk;
1468 struct udp_iter_state *state = seq->private;
1469
1470 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) {
1471 struct hlist_node *node;
1472 sk_for_each(sk, node, state->hashtable + state->bucket) {
1473 if (sk->sk_family == state->family)
1474 goto found;
1475 }
1476 }
1477 sk = NULL;
1478 found:
1479 return sk;
1480 }
1481
1482 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1483 {
1484 struct udp_iter_state *state = seq->private;
1485
1486 do {
1487 sk = sk_next(sk);
1488 try_again:
1489 ;
1490 } while (sk && sk->sk_family != state->family);
1491
1492 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) {
1493 sk = sk_head(state->hashtable + state->bucket);
1494 goto try_again;
1495 }
1496 return sk;
1497 }
1498
1499 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1500 {
1501 struct sock *sk = udp_get_first(seq);
1502
1503 if (sk)
1504 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1505 --pos;
1506 return pos ? NULL : sk;
1507 }
1508
1509 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1510 {
1511 read_lock(&udp_hash_lock);
1512 return *pos ? udp_get_idx(seq, *pos-1) : (void *)1;
1513 }
1514
1515 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1516 {
1517 struct sock *sk;
1518
1519 if (v == (void *)1)
1520 sk = udp_get_idx(seq, 0);
1521 else
1522 sk = udp_get_next(seq, v);
1523
1524 ++*pos;
1525 return sk;
1526 }
1527
1528 static void udp_seq_stop(struct seq_file *seq, void *v)
1529 {
1530 read_unlock(&udp_hash_lock);
1531 }
1532
1533 static int udp_seq_open(struct inode *inode, struct file *file)
1534 {
1535 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1536 struct seq_file *seq;
1537 int rc = -ENOMEM;
1538 struct udp_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL);
1539
1540 if (!s)
1541 goto out;
1542 s->family = afinfo->family;
1543 s->hashtable = afinfo->hashtable;
1544 s->seq_ops.start = udp_seq_start;
1545 s->seq_ops.next = udp_seq_next;
1546 s->seq_ops.show = afinfo->seq_show;
1547 s->seq_ops.stop = udp_seq_stop;
1548
1549 rc = seq_open(file, &s->seq_ops);
1550 if (rc)
1551 goto out_kfree;
1552
1553 seq = file->private_data;
1554 seq->private = s;
1555 out:
1556 return rc;
1557 out_kfree:
1558 kfree(s);
1559 goto out;
1560 }
1561
1562 /* ------------------------------------------------------------------------ */
1563 int udp_proc_register(struct udp_seq_afinfo *afinfo)
1564 {
1565 struct proc_dir_entry *p;
1566 int rc = 0;
1567
1568 if (!afinfo)
1569 return -EINVAL;
1570 afinfo->seq_fops->owner = afinfo->owner;
1571 afinfo->seq_fops->open = udp_seq_open;
1572 afinfo->seq_fops->read = seq_read;
1573 afinfo->seq_fops->llseek = seq_lseek;
1574 afinfo->seq_fops->release = seq_release_private;
1575
1576 p = proc_net_fops_create(&init_net, afinfo->name, S_IRUGO, afinfo->seq_fops);
1577 if (p)
1578 p->data = afinfo;
1579 else
1580 rc = -ENOMEM;
1581 return rc;
1582 }
1583
1584 void udp_proc_unregister(struct udp_seq_afinfo *afinfo)
1585 {
1586 if (!afinfo)
1587 return;
1588 proc_net_remove(&init_net, afinfo->name);
1589 memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops));
1590 }
1591
1592 /* ------------------------------------------------------------------------ */
1593 static void udp4_format_sock(struct sock *sp, char *tmpbuf, int bucket)
1594 {
1595 struct inet_sock *inet = inet_sk(sp);
1596 __be32 dest = inet->daddr;
1597 __be32 src = inet->rcv_saddr;
1598 __u16 destp = ntohs(inet->dport);
1599 __u16 srcp = ntohs(inet->sport);
1600
1601 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
1602 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p",
1603 bucket, src, srcp, dest, destp, sp->sk_state,
1604 atomic_read(&sp->sk_wmem_alloc),
1605 atomic_read(&sp->sk_rmem_alloc),
1606 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1607 atomic_read(&sp->sk_refcnt), sp);
1608 }
1609
1610 int udp4_seq_show(struct seq_file *seq, void *v)
1611 {
1612 if (v == SEQ_START_TOKEN)
1613 seq_printf(seq, "%-127s\n",
1614 " sl local_address rem_address st tx_queue "
1615 "rx_queue tr tm->when retrnsmt uid timeout "
1616 "inode");
1617 else {
1618 char tmpbuf[129];
1619 struct udp_iter_state *state = seq->private;
1620
1621 udp4_format_sock(v, tmpbuf, state->bucket);
1622 seq_printf(seq, "%-127s\n", tmpbuf);
1623 }
1624 return 0;
1625 }
1626
1627 /* ------------------------------------------------------------------------ */
1628 static struct file_operations udp4_seq_fops;
1629 static struct udp_seq_afinfo udp4_seq_afinfo = {
1630 .owner = THIS_MODULE,
1631 .name = "udp",
1632 .family = AF_INET,
1633 .hashtable = udp_hash,
1634 .seq_show = udp4_seq_show,
1635 .seq_fops = &udp4_seq_fops,
1636 };
1637
1638 int __init udp4_proc_init(void)
1639 {
1640 return udp_proc_register(&udp4_seq_afinfo);
1641 }
1642
1643 void udp4_proc_exit(void)
1644 {
1645 udp_proc_unregister(&udp4_seq_afinfo);
1646 }
1647 #endif /* CONFIG_PROC_FS */
1648
1649 EXPORT_SYMBOL(udp_disconnect);
1650 EXPORT_SYMBOL(udp_hash);
1651 EXPORT_SYMBOL(udp_hash_lock);
1652 EXPORT_SYMBOL(udp_ioctl);
1653 EXPORT_SYMBOL(udp_get_port);
1654 EXPORT_SYMBOL(udp_prot);
1655 EXPORT_SYMBOL(udp_sendmsg);
1656 EXPORT_SYMBOL(udp_lib_getsockopt);
1657 EXPORT_SYMBOL(udp_lib_setsockopt);
1658 EXPORT_SYMBOL(udp_poll);
1659
1660 #ifdef CONFIG_PROC_FS
1661 EXPORT_SYMBOL(udp_proc_register);
1662 EXPORT_SYMBOL(udp_proc_unregister);
1663 #endif
Linux kernel - Deliverables
This page took 0.069239 seconds and 5 git commands to generate.