projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
[NET]: Add some acquires/releases sparse annotations.
[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/bootmem.h>
86 #include <linux/types.h>
87 #include <linux/fcntl.h>
88 #include <linux/module.h>
89 #include <linux/socket.h>
90 #include <linux/sockios.h>
91 #include <linux/igmp.h>
92 #include <linux/in.h>
93 #include <linux/errno.h>
94 #include <linux/timer.h>
95 #include <linux/mm.h>
96 #include <linux/inet.h>
97 #include <linux/netdevice.h>
98 #include <net/tcp_states.h>
99 #include <linux/skbuff.h>
100 #include <linux/proc_fs.h>
101 #include <linux/seq_file.h>
102 #include <net/net_namespace.h>
103 #include <net/icmp.h>
104 #include <net/route.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include "udp_impl.h"
108
109 /*
110 * Snmp MIB for the UDP layer
111 */
112
113 DEFINE_SNMP_STAT(struct udp_mib, udp_statistics) __read_mostly;
114 EXPORT_SYMBOL(udp_statistics);
115
116 DEFINE_SNMP_STAT(struct udp_mib, udp_stats_in6) __read_mostly;
117 EXPORT_SYMBOL(udp_stats_in6);
118
119 struct hlist_head udp_hash[UDP_HTABLE_SIZE];
120 DEFINE_RWLOCK(udp_hash_lock);
121
122 int sysctl_udp_mem[3] __read_mostly;
123 int sysctl_udp_rmem_min __read_mostly;
124 int sysctl_udp_wmem_min __read_mostly;
125
126 EXPORT_SYMBOL(sysctl_udp_mem);
127 EXPORT_SYMBOL(sysctl_udp_rmem_min);
128 EXPORT_SYMBOL(sysctl_udp_wmem_min);
129
130 atomic_t udp_memory_allocated;
131 EXPORT_SYMBOL(udp_memory_allocated);
132
133 static inline int __udp_lib_lport_inuse(__u16 num,
134 const struct hlist_head udptable[])
135 {
136 struct sock *sk;
137 struct hlist_node *node;
138
139 sk_for_each(sk, node, &udptable[num & (UDP_HTABLE_SIZE - 1)])
140 if (sk->sk_hash == num)
141 return 1;
142 return 0;
143 }
144
145 /**
146 * __udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
147 *
148 * @sk: socket struct in question
149 * @snum: port number to look up
150 * @udptable: hash list table, must be of UDP_HTABLE_SIZE
151 * @saddr_comp: AF-dependent comparison of bound local IP addresses
152 */
153 int __udp_lib_get_port(struct sock *sk, unsigned short snum,
154 struct hlist_head udptable[],
155 int (*saddr_comp)(const struct sock *sk1,
156 const struct sock *sk2 ) )
157 {
158 struct hlist_node *node;
159 struct hlist_head *head;
160 struct sock *sk2;
161 int error = 1;
162
163 write_lock_bh(&udp_hash_lock);
164
165 if (!snum) {
166 int i, low, high, remaining;
167 unsigned rover, best, best_size_so_far;
168
169 inet_get_local_port_range(&low, &high);
170 remaining = (high - low) + 1;
171
172 best_size_so_far = UINT_MAX;
173 best = rover = net_random() % remaining + low;
174
175 /* 1st pass: look for empty (or shortest) hash chain */
176 for (i = 0; i < UDP_HTABLE_SIZE; i++) {
177 int size = 0;
178
179 head = &udptable[rover & (UDP_HTABLE_SIZE - 1)];
180 if (hlist_empty(head))
181 goto gotit;
182
183 sk_for_each(sk2, node, head) {
184 if (++size >= best_size_so_far)
185 goto next;
186 }
187 best_size_so_far = size;
188 best = rover;
189 next:
190 /* fold back if end of range */
191 if (++rover > high)
192 rover = low + ((rover - low)
193 & (UDP_HTABLE_SIZE - 1));
194
195
196 }
197
198 /* 2nd pass: find hole in shortest hash chain */
199 rover = best;
200 for (i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; i++) {
201 if (! __udp_lib_lport_inuse(rover, udptable))
202 goto gotit;
203 rover += UDP_HTABLE_SIZE;
204 if (rover > high)
205 rover = low + ((rover - low)
206 & (UDP_HTABLE_SIZE - 1));
207 }
208
209
210 /* All ports in use! */
211 goto fail;
212
213 gotit:
214 snum = rover;
215 } else {
216 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
217
218 sk_for_each(sk2, node, head)
219 if (sk2->sk_hash == snum &&
220 sk2 != sk &&
221 (!sk2->sk_reuse || !sk->sk_reuse) &&
222 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if
223 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
224 (*saddr_comp)(sk, sk2) )
225 goto fail;
226 }
227
228 inet_sk(sk)->num = snum;
229 sk->sk_hash = snum;
230 if (sk_unhashed(sk)) {
231 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
232 sk_add_node(sk, head);
233 sock_prot_inc_use(sk->sk_prot);
234 }
235 error = 0;
236 fail:
237 write_unlock_bh(&udp_hash_lock);
238 return error;
239 }
240
241 int udp_get_port(struct sock *sk, unsigned short snum,
242 int (*scmp)(const struct sock *, const struct sock *))
243 {
244 return __udp_lib_get_port(sk, snum, udp_hash, scmp);
245 }
246
247 int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
248 {
249 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
250
251 return ( !ipv6_only_sock(sk2) &&
252 (!inet1->rcv_saddr || !inet2->rcv_saddr ||
253 inet1->rcv_saddr == inet2->rcv_saddr ));
254 }
255
256 static inline int udp_v4_get_port(struct sock *sk, unsigned short snum)
257 {
258 return udp_get_port(sk, snum, ipv4_rcv_saddr_equal);
259 }
260
261 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
262 * harder than this. -DaveM
263 */
264 static struct sock *__udp4_lib_lookup(__be32 saddr, __be16 sport,
265 __be32 daddr, __be16 dport,
266 int dif, struct hlist_head udptable[])
267 {
268 struct sock *sk, *result = NULL;
269 struct hlist_node *node;
270 unsigned short hnum = ntohs(dport);
271 int badness = -1;
272
273 read_lock(&udp_hash_lock);
274 sk_for_each(sk, node, &udptable[hnum & (UDP_HTABLE_SIZE - 1)]) {
275 struct inet_sock *inet = inet_sk(sk);
276
277 if (sk->sk_hash == hnum && !ipv6_only_sock(sk)) {
278 int score = (sk->sk_family == PF_INET ? 1 : 0);
279 if (inet->rcv_saddr) {
280 if (inet->rcv_saddr != daddr)
281 continue;
282 score+=2;
283 }
284 if (inet->daddr) {
285 if (inet->daddr != saddr)
286 continue;
287 score+=2;
288 }
289 if (inet->dport) {
290 if (inet->dport != sport)
291 continue;
292 score+=2;
293 }
294 if (sk->sk_bound_dev_if) {
295 if (sk->sk_bound_dev_if != dif)
296 continue;
297 score+=2;
298 }
299 if (score == 9) {
300 result = sk;
301 break;
302 } else if (score > badness) {
303 result = sk;
304 badness = score;
305 }
306 }
307 }
308 if (result)
309 sock_hold(result);
310 read_unlock(&udp_hash_lock);
311 return result;
312 }
313
314 static inline struct sock *udp_v4_mcast_next(struct sock *sk,
315 __be16 loc_port, __be32 loc_addr,
316 __be16 rmt_port, __be32 rmt_addr,
317 int dif)
318 {
319 struct hlist_node *node;
320 struct sock *s = sk;
321 unsigned short hnum = ntohs(loc_port);
322
323 sk_for_each_from(s, node) {
324 struct inet_sock *inet = inet_sk(s);
325
326 if (s->sk_hash != hnum ||
327 (inet->daddr && inet->daddr != rmt_addr) ||
328 (inet->dport != rmt_port && inet->dport) ||
329 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) ||
330 ipv6_only_sock(s) ||
331 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
332 continue;
333 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
334 continue;
335 goto found;
336 }
337 s = NULL;
338 found:
339 return s;
340 }
341
342 /*
343 * This routine is called by the ICMP module when it gets some
344 * sort of error condition. If err < 0 then the socket should
345 * be closed and the error returned to the user. If err > 0
346 * it's just the icmp type << 8 | icmp code.
347 * Header points to the ip header of the error packet. We move
348 * on past this. Then (as it used to claim before adjustment)
349 * header points to the first 8 bytes of the udp header. We need
350 * to find the appropriate port.
351 */
352
353 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[])
354 {
355 struct inet_sock *inet;
356 struct iphdr *iph = (struct iphdr*)skb->data;
357 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2));
358 const int type = icmp_hdr(skb)->type;
359 const int code = icmp_hdr(skb)->code;
360 struct sock *sk;
361 int harderr;
362 int err;
363
364 sk = __udp4_lib_lookup(iph->daddr, uh->dest, iph->saddr, uh->source,
365 skb->dev->ifindex, udptable );
366 if (sk == NULL) {
367 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
368 return; /* No socket for error */
369 }
370
371 err = 0;
372 harderr = 0;
373 inet = inet_sk(sk);
374
375 switch (type) {
376 default:
377 case ICMP_TIME_EXCEEDED:
378 err = EHOSTUNREACH;
379 break;
380 case ICMP_SOURCE_QUENCH:
381 goto out;
382 case ICMP_PARAMETERPROB:
383 err = EPROTO;
384 harderr = 1;
385 break;
386 case ICMP_DEST_UNREACH:
387 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
388 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
389 err = EMSGSIZE;
390 harderr = 1;
391 break;
392 }
393 goto out;
394 }
395 err = EHOSTUNREACH;
396 if (code <= NR_ICMP_UNREACH) {
397 harderr = icmp_err_convert[code].fatal;
398 err = icmp_err_convert[code].errno;
399 }
400 break;
401 }
402
403 /*
404 * RFC1122: OK. Passes ICMP errors back to application, as per
405 * 4.1.3.3.
406 */
407 if (!inet->recverr) {
408 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
409 goto out;
410 } else {
411 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1));
412 }
413 sk->sk_err = err;
414 sk->sk_error_report(sk);
415 out:
416 sock_put(sk);
417 }
418
419 void udp_err(struct sk_buff *skb, u32 info)
420 {
421 return __udp4_lib_err(skb, info, udp_hash);
422 }
423
424 /*
425 * Throw away all pending data and cancel the corking. Socket is locked.
426 */
427 static void udp_flush_pending_frames(struct sock *sk)
428 {
429 struct udp_sock *up = udp_sk(sk);
430
431 if (up->pending) {
432 up->len = 0;
433 up->pending = 0;
434 ip_flush_pending_frames(sk);
435 }
436 }
437
438 /**
439 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
440 * @sk: socket we are sending on
441 * @skb: sk_buff containing the filled-in UDP header
442 * (checksum field must be zeroed out)
443 */
444 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
445 __be32 src, __be32 dst, int len )
446 {
447 unsigned int offset;
448 struct udphdr *uh = udp_hdr(skb);
449 __wsum csum = 0;
450
451 if (skb_queue_len(&sk->sk_write_queue) == 1) {
452 /*
453 * Only one fragment on the socket.
454 */
455 skb->csum_start = skb_transport_header(skb) - skb->head;
456 skb->csum_offset = offsetof(struct udphdr, check);
457 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
458 } else {
459 /*
460 * HW-checksum won't work as there are two or more
461 * fragments on the socket so that all csums of sk_buffs
462 * should be together
463 */
464 offset = skb_transport_offset(skb);
465 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
466
467 skb->ip_summed = CHECKSUM_NONE;
468
469 skb_queue_walk(&sk->sk_write_queue, skb) {
470 csum = csum_add(csum, skb->csum);
471 }
472
473 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
474 if (uh->check == 0)
475 uh->check = CSUM_MANGLED_0;
476 }
477 }
478
479 /*
480 * Push out all pending data as one UDP datagram. Socket is locked.
481 */
482 static int udp_push_pending_frames(struct sock *sk)
483 {
484 struct udp_sock *up = udp_sk(sk);
485 struct inet_sock *inet = inet_sk(sk);
486 struct flowi *fl = &inet->cork.fl;
487 struct sk_buff *skb;
488 struct udphdr *uh;
489 int err = 0;
490 int is_udplite = IS_UDPLITE(sk);
491 __wsum csum = 0;
492
493 /* Grab the skbuff where UDP header space exists. */
494 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
495 goto out;
496
497 /*
498 * Create a UDP header
499 */
500 uh = udp_hdr(skb);
501 uh->source = fl->fl_ip_sport;
502 uh->dest = fl->fl_ip_dport;
503 uh->len = htons(up->len);
504 uh->check = 0;
505
506 if (is_udplite) /* UDP-Lite */
507 csum = udplite_csum_outgoing(sk, skb);
508
509 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
510
511 skb->ip_summed = CHECKSUM_NONE;
512 goto send;
513
514 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
515
516 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len);
517 goto send;
518
519 } else /* `normal' UDP */
520 csum = udp_csum_outgoing(sk, skb);
521
522 /* add protocol-dependent pseudo-header */
523 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
524 sk->sk_protocol, csum );
525 if (uh->check == 0)
526 uh->check = CSUM_MANGLED_0;
527
528 send:
529 err = ip_push_pending_frames(sk);
530 out:
531 up->len = 0;
532 up->pending = 0;
533 if (!err)
534 UDP_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS, is_udplite);
535 return err;
536 }
537
538 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
539 size_t len)
540 {
541 struct inet_sock *inet = inet_sk(sk);
542 struct udp_sock *up = udp_sk(sk);
543 int ulen = len;
544 struct ipcm_cookie ipc;
545 struct rtable *rt = NULL;
546 int free = 0;
547 int connected = 0;
548 __be32 daddr, faddr, saddr;
549 __be16 dport;
550 u8 tos;
551 int err, is_udplite = IS_UDPLITE(sk);
552 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
553 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
554
555 if (len > 0xFFFF)
556 return -EMSGSIZE;
557
558 /*
559 * Check the flags.
560 */
561
562 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */
563 return -EOPNOTSUPP;
564
565 ipc.opt = NULL;
566
567 if (up->pending) {
568 /*
569 * There are pending frames.
570 * The socket lock must be held while it's corked.
571 */
572 lock_sock(sk);
573 if (likely(up->pending)) {
574 if (unlikely(up->pending != AF_INET)) {
575 release_sock(sk);
576 return -EINVAL;
577 }
578 goto do_append_data;
579 }
580 release_sock(sk);
581 }
582 ulen += sizeof(struct udphdr);
583
584 /*
585 * Get and verify the address.
586 */
587 if (msg->msg_name) {
588 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name;
589 if (msg->msg_namelen < sizeof(*usin))
590 return -EINVAL;
591 if (usin->sin_family != AF_INET) {
592 if (usin->sin_family != AF_UNSPEC)
593 return -EAFNOSUPPORT;
594 }
595
596 daddr = usin->sin_addr.s_addr;
597 dport = usin->sin_port;
598 if (dport == 0)
599 return -EINVAL;
600 } else {
601 if (sk->sk_state != TCP_ESTABLISHED)
602 return -EDESTADDRREQ;
603 daddr = inet->daddr;
604 dport = inet->dport;
605 /* Open fast path for connected socket.
606 Route will not be used, if at least one option is set.
607 */
608 connected = 1;
609 }
610 ipc.addr = inet->saddr;
611
612 ipc.oif = sk->sk_bound_dev_if;
613 if (msg->msg_controllen) {
614 err = ip_cmsg_send(msg, &ipc);
615 if (err)
616 return err;
617 if (ipc.opt)
618 free = 1;
619 connected = 0;
620 }
621 if (!ipc.opt)
622 ipc.opt = inet->opt;
623
624 saddr = ipc.addr;
625 ipc.addr = faddr = daddr;
626
627 if (ipc.opt && ipc.opt->srr) {
628 if (!daddr)
629 return -EINVAL;
630 faddr = ipc.opt->faddr;
631 connected = 0;
632 }
633 tos = RT_TOS(inet->tos);
634 if (sock_flag(sk, SOCK_LOCALROUTE) ||
635 (msg->msg_flags & MSG_DONTROUTE) ||
636 (ipc.opt && ipc.opt->is_strictroute)) {
637 tos |= RTO_ONLINK;
638 connected = 0;
639 }
640
641 if (ipv4_is_multicast(daddr)) {
642 if (!ipc.oif)
643 ipc.oif = inet->mc_index;
644 if (!saddr)
645 saddr = inet->mc_addr;
646 connected = 0;
647 }
648
649 if (connected)
650 rt = (struct rtable*)sk_dst_check(sk, 0);
651
652 if (rt == NULL) {
653 struct flowi fl = { .oif = ipc.oif,
654 .nl_u = { .ip4_u =
655 { .daddr = faddr,
656 .saddr = saddr,
657 .tos = tos } },
658 .proto = sk->sk_protocol,
659 .uli_u = { .ports =
660 { .sport = inet->sport,
661 .dport = dport } } };
662 security_sk_classify_flow(sk, &fl);
663 err = ip_route_output_flow(&rt, &fl, sk, 1);
664 if (err) {
665 if (err == -ENETUNREACH)
666 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
667 goto out;
668 }
669
670 err = -EACCES;
671 if ((rt->rt_flags & RTCF_BROADCAST) &&
672 !sock_flag(sk, SOCK_BROADCAST))
673 goto out;
674 if (connected)
675 sk_dst_set(sk, dst_clone(&rt->u.dst));
676 }
677
678 if (msg->msg_flags&MSG_CONFIRM)
679 goto do_confirm;
680 back_from_confirm:
681
682 saddr = rt->rt_src;
683 if (!ipc.addr)
684 daddr = ipc.addr = rt->rt_dst;
685
686 lock_sock(sk);
687 if (unlikely(up->pending)) {
688 /* The socket is already corked while preparing it. */
689 /* ... which is an evident application bug. --ANK */
690 release_sock(sk);
691
692 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
693 err = -EINVAL;
694 goto out;
695 }
696 /*
697 * Now cork the socket to pend data.
698 */
699 inet->cork.fl.fl4_dst = daddr;
700 inet->cork.fl.fl_ip_dport = dport;
701 inet->cork.fl.fl4_src = saddr;
702 inet->cork.fl.fl_ip_sport = inet->sport;
703 up->pending = AF_INET;
704
705 do_append_data:
706 up->len += ulen;
707 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
708 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
709 sizeof(struct udphdr), &ipc, rt,
710 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
711 if (err)
712 udp_flush_pending_frames(sk);
713 else if (!corkreq)
714 err = udp_push_pending_frames(sk);
715 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
716 up->pending = 0;
717 release_sock(sk);
718
719 out:
720 ip_rt_put(rt);
721 if (free)
722 kfree(ipc.opt);
723 if (!err)
724 return len;
725 /*
726 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
727 * ENOBUFS might not be good (it's not tunable per se), but otherwise
728 * we don't have a good statistic (IpOutDiscards but it can be too many
729 * things). We could add another new stat but at least for now that
730 * seems like overkill.
731 */
732 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
733 UDP_INC_STATS_USER(UDP_MIB_SNDBUFERRORS, is_udplite);
734 }
735 return err;
736
737 do_confirm:
738 dst_confirm(&rt->u.dst);
739 if (!(msg->msg_flags&MSG_PROBE) || len)
740 goto back_from_confirm;
741 err = 0;
742 goto out;
743 }
744
745 int udp_sendpage(struct sock *sk, struct page *page, int offset,
746 size_t size, int flags)
747 {
748 struct udp_sock *up = udp_sk(sk);
749 int ret;
750
751 if (!up->pending) {
752 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
753
754 /* Call udp_sendmsg to specify destination address which
755 * sendpage interface can't pass.
756 * This will succeed only when the socket is connected.
757 */
758 ret = udp_sendmsg(NULL, sk, &msg, 0);
759 if (ret < 0)
760 return ret;
761 }
762
763 lock_sock(sk);
764
765 if (unlikely(!up->pending)) {
766 release_sock(sk);
767
768 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
769 return -EINVAL;
770 }
771
772 ret = ip_append_page(sk, page, offset, size, flags);
773 if (ret == -EOPNOTSUPP) {
774 release_sock(sk);
775 return sock_no_sendpage(sk->sk_socket, page, offset,
776 size, flags);
777 }
778 if (ret < 0) {
779 udp_flush_pending_frames(sk);
780 goto out;
781 }
782
783 up->len += size;
784 if (!(up->corkflag || (flags&MSG_MORE)))
785 ret = udp_push_pending_frames(sk);
786 if (!ret)
787 ret = size;
788 out:
789 release_sock(sk);
790 return ret;
791 }
792
793 /*
794 * IOCTL requests applicable to the UDP protocol
795 */
796
797 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
798 {
799 switch (cmd) {
800 case SIOCOUTQ:
801 {
802 int amount = atomic_read(&sk->sk_wmem_alloc);
803 return put_user(amount, (int __user *)arg);
804 }
805
806 case SIOCINQ:
807 {
808 struct sk_buff *skb;
809 unsigned long amount;
810
811 amount = 0;
812 spin_lock_bh(&sk->sk_receive_queue.lock);
813 skb = skb_peek(&sk->sk_receive_queue);
814 if (skb != NULL) {
815 /*
816 * We will only return the amount
817 * of this packet since that is all
818 * that will be read.
819 */
820 amount = skb->len - sizeof(struct udphdr);
821 }
822 spin_unlock_bh(&sk->sk_receive_queue.lock);
823 return put_user(amount, (int __user *)arg);
824 }
825
826 default:
827 return -ENOIOCTLCMD;
828 }
829
830 return 0;
831 }
832
833 /*
834 * This should be easy, if there is something there we
835 * return it, otherwise we block.
836 */
837
838 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
839 size_t len, int noblock, int flags, int *addr_len)
840 {
841 struct inet_sock *inet = inet_sk(sk);
842 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
843 struct sk_buff *skb;
844 unsigned int ulen, copied;
845 int peeked;
846 int err;
847 int is_udplite = IS_UDPLITE(sk);
848
849 /*
850 * Check any passed addresses
851 */
852 if (addr_len)
853 *addr_len=sizeof(*sin);
854
855 if (flags & MSG_ERRQUEUE)
856 return ip_recv_error(sk, msg, len);
857
858 try_again:
859 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
860 &peeked, &err);
861 if (!skb)
862 goto out;
863
864 ulen = skb->len - sizeof(struct udphdr);
865 copied = len;
866 if (copied > ulen)
867 copied = ulen;
868 else if (copied < ulen)
869 msg->msg_flags |= MSG_TRUNC;
870
871 /*
872 * If checksum is needed at all, try to do it while copying the
873 * data. If the data is truncated, or if we only want a partial
874 * coverage checksum (UDP-Lite), do it before the copy.
875 */
876
877 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
878 if (udp_lib_checksum_complete(skb))
879 goto csum_copy_err;
880 }
881
882 if (skb_csum_unnecessary(skb))
883 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
884 msg->msg_iov, copied );
885 else {
886 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
887
888 if (err == -EINVAL)
889 goto csum_copy_err;
890 }
891
892 if (err)
893 goto out_free;
894
895 if (!peeked)
896 UDP_INC_STATS_USER(UDP_MIB_INDATAGRAMS, is_udplite);
897
898 sock_recv_timestamp(msg, sk, skb);
899
900 /* Copy the address. */
901 if (sin)
902 {
903 sin->sin_family = AF_INET;
904 sin->sin_port = udp_hdr(skb)->source;
905 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
906 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
907 }
908 if (inet->cmsg_flags)
909 ip_cmsg_recv(msg, skb);
910
911 err = copied;
912 if (flags & MSG_TRUNC)
913 err = ulen;
914
915 out_free:
916 lock_sock(sk);
917 skb_free_datagram(sk, skb);
918 release_sock(sk);
919 out:
920 return err;
921
922 csum_copy_err:
923 lock_sock(sk);
924 if (!skb_kill_datagram(sk, skb, flags))
925 UDP_INC_STATS_USER(UDP_MIB_INERRORS, is_udplite);
926 release_sock(sk);
927
928 if (noblock)
929 return -EAGAIN;
930 goto try_again;
931 }
932
933
934 int udp_disconnect(struct sock *sk, int flags)
935 {
936 struct inet_sock *inet = inet_sk(sk);
937 /*
938 * 1003.1g - break association.
939 */
940
941 sk->sk_state = TCP_CLOSE;
942 inet->daddr = 0;
943 inet->dport = 0;
944 sk->sk_bound_dev_if = 0;
945 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
946 inet_reset_saddr(sk);
947
948 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
949 sk->sk_prot->unhash(sk);
950 inet->sport = 0;
951 }
952 sk_dst_reset(sk);
953 return 0;
954 }
955
956 /* returns:
957 * -1: error
958 * 0: success
959 * >0: "udp encap" protocol resubmission
960 *
961 * Note that in the success and error cases, the skb is assumed to
962 * have either been requeued or freed.
963 */
964 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
965 {
966 struct udp_sock *up = udp_sk(sk);
967 int rc;
968 int is_udplite = IS_UDPLITE(sk);
969
970 /*
971 * Charge it to the socket, dropping if the queue is full.
972 */
973 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
974 goto drop;
975 nf_reset(skb);
976
977 if (up->encap_type) {
978 /*
979 * This is an encapsulation socket so pass the skb to
980 * the socket's udp_encap_rcv() hook. Otherwise, just
981 * fall through and pass this up the UDP socket.
982 * up->encap_rcv() returns the following value:
983 * =0 if skb was successfully passed to the encap
984 * handler or was discarded by it.
985 * >0 if skb should be passed on to UDP.
986 * <0 if skb should be resubmitted as proto -N
987 */
988
989 /* if we're overly short, let UDP handle it */
990 if (skb->len > sizeof(struct udphdr) &&
991 up->encap_rcv != NULL) {
992 int ret;
993
994 ret = (*up->encap_rcv)(sk, skb);
995 if (ret <= 0) {
996 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS,
997 is_udplite);
998 return -ret;
999 }
1000 }
1001
1002 /* FALLTHROUGH -- it's a UDP Packet */
1003 }
1004
1005 /*
1006 * UDP-Lite specific tests, ignored on UDP sockets
1007 */
1008 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1009
1010 /*
1011 * MIB statistics other than incrementing the error count are
1012 * disabled for the following two types of errors: these depend
1013 * on the application settings, not on the functioning of the
1014 * protocol stack as such.
1015 *
1016 * RFC 3828 here recommends (sec 3.3): "There should also be a
1017 * way ... to ... at least let the receiving application block
1018 * delivery of packets with coverage values less than a value
1019 * provided by the application."
1020 */
1021 if (up->pcrlen == 0) { /* full coverage was set */
1022 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1023 "%d while full coverage %d requested\n",
1024 UDP_SKB_CB(skb)->cscov, skb->len);
1025 goto drop;
1026 }
1027 /* The next case involves violating the min. coverage requested
1028 * by the receiver. This is subtle: if receiver wants x and x is
1029 * greater than the buffersize/MTU then receiver will complain
1030 * that it wants x while sender emits packets of smaller size y.
1031 * Therefore the above ...()->partial_cov statement is essential.
1032 */
1033 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1034 LIMIT_NETDEBUG(KERN_WARNING
1035 "UDPLITE: coverage %d too small, need min %d\n",
1036 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1037 goto drop;
1038 }
1039 }
1040
1041 if (sk->sk_filter) {
1042 if (udp_lib_checksum_complete(skb))
1043 goto drop;
1044 }
1045
1046 if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) {
1047 /* Note that an ENOMEM error is charged twice */
1048 if (rc == -ENOMEM)
1049 UDP_INC_STATS_BH(UDP_MIB_RCVBUFERRORS, is_udplite);
1050 goto drop;
1051 }
1052
1053 return 0;
1054
1055 drop:
1056 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite);
1057 kfree_skb(skb);
1058 return -1;
1059 }
1060
1061 /*
1062 * Multicasts and broadcasts go to each listener.
1063 *
1064 * Note: called only from the BH handler context,
1065 * so we don't need to lock the hashes.
1066 */
1067 static int __udp4_lib_mcast_deliver(struct sk_buff *skb,
1068 struct udphdr *uh,
1069 __be32 saddr, __be32 daddr,
1070 struct hlist_head udptable[])
1071 {
1072 struct sock *sk;
1073 int dif;
1074
1075 read_lock(&udp_hash_lock);
1076 sk = sk_head(&udptable[ntohs(uh->dest) & (UDP_HTABLE_SIZE - 1)]);
1077 dif = skb->dev->ifindex;
1078 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
1079 if (sk) {
1080 struct sock *sknext = NULL;
1081
1082 do {
1083 struct sk_buff *skb1 = skb;
1084
1085 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr,
1086 uh->source, saddr, dif);
1087 if (sknext)
1088 skb1 = skb_clone(skb, GFP_ATOMIC);
1089
1090 if (skb1) {
1091 int ret = 0;
1092
1093 bh_lock_sock_nested(sk);
1094 if (!sock_owned_by_user(sk))
1095 ret = udp_queue_rcv_skb(sk, skb1);
1096 else
1097 sk_add_backlog(sk, skb1);
1098 bh_unlock_sock(sk);
1099
1100 if (ret > 0)
1101 /* we should probably re-process instead
1102 * of dropping packets here. */
1103 kfree_skb(skb1);
1104 }
1105 sk = sknext;
1106 } while (sknext);
1107 } else
1108 kfree_skb(skb);
1109 read_unlock(&udp_hash_lock);
1110 return 0;
1111 }
1112
1113 /* Initialize UDP checksum. If exited with zero value (success),
1114 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1115 * Otherwise, csum completion requires chacksumming packet body,
1116 * including udp header and folding it to skb->csum.
1117 */
1118 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1119 int proto)
1120 {
1121 const struct iphdr *iph;
1122 int err;
1123
1124 UDP_SKB_CB(skb)->partial_cov = 0;
1125 UDP_SKB_CB(skb)->cscov = skb->len;
1126
1127 if (proto == IPPROTO_UDPLITE) {
1128 err = udplite_checksum_init(skb, uh);
1129 if (err)
1130 return err;
1131 }
1132
1133 iph = ip_hdr(skb);
1134 if (uh->check == 0) {
1135 skb->ip_summed = CHECKSUM_UNNECESSARY;
1136 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1137 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1138 proto, skb->csum))
1139 skb->ip_summed = CHECKSUM_UNNECESSARY;
1140 }
1141 if (!skb_csum_unnecessary(skb))
1142 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1143 skb->len, proto, 0);
1144 /* Probably, we should checksum udp header (it should be in cache
1145 * in any case) and data in tiny packets (< rx copybreak).
1146 */
1147
1148 return 0;
1149 }
1150
1151 /*
1152 * All we need to do is get the socket, and then do a checksum.
1153 */
1154
1155 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
1156 int proto)
1157 {
1158 struct sock *sk;
1159 struct udphdr *uh = udp_hdr(skb);
1160 unsigned short ulen;
1161 struct rtable *rt = (struct rtable*)skb->dst;
1162 __be32 saddr = ip_hdr(skb)->saddr;
1163 __be32 daddr = ip_hdr(skb)->daddr;
1164
1165 /*
1166 * Validate the packet.
1167 */
1168 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1169 goto drop; /* No space for header. */
1170
1171 ulen = ntohs(uh->len);
1172 if (ulen > skb->len)
1173 goto short_packet;
1174
1175 if (proto == IPPROTO_UDP) {
1176 /* UDP validates ulen. */
1177 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1178 goto short_packet;
1179 uh = udp_hdr(skb);
1180 }
1181
1182 if (udp4_csum_init(skb, uh, proto))
1183 goto csum_error;
1184
1185 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1186 return __udp4_lib_mcast_deliver(skb, uh, saddr, daddr, udptable);
1187
1188 sk = __udp4_lib_lookup(saddr, uh->source, daddr, uh->dest,
1189 inet_iif(skb), udptable);
1190
1191 if (sk != NULL) {
1192 int ret = 0;
1193 bh_lock_sock_nested(sk);
1194 if (!sock_owned_by_user(sk))
1195 ret = udp_queue_rcv_skb(sk, skb);
1196 else
1197 sk_add_backlog(sk, skb);
1198 bh_unlock_sock(sk);
1199 sock_put(sk);
1200
1201 /* a return value > 0 means to resubmit the input, but
1202 * it wants the return to be -protocol, or 0
1203 */
1204 if (ret > 0)
1205 return -ret;
1206 return 0;
1207 }
1208
1209 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1210 goto drop;
1211 nf_reset(skb);
1212
1213 /* No socket. Drop packet silently, if checksum is wrong */
1214 if (udp_lib_checksum_complete(skb))
1215 goto csum_error;
1216
1217 UDP_INC_STATS_BH(UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1218 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1219
1220 /*
1221 * Hmm. We got an UDP packet to a port to which we
1222 * don't wanna listen. Ignore it.
1223 */
1224 kfree_skb(skb);
1225 return 0;
1226
1227 short_packet:
1228 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %u.%u.%u.%u:%u %d/%d to %u.%u.%u.%u:%u\n",
1229 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1230 NIPQUAD(saddr),
1231 ntohs(uh->source),
1232 ulen,
1233 skb->len,
1234 NIPQUAD(daddr),
1235 ntohs(uh->dest));
1236 goto drop;
1237
1238 csum_error:
1239 /*
1240 * RFC1122: OK. Discards the bad packet silently (as far as
1241 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1242 */
1243 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %d.%d.%d.%d:%d to %d.%d.%d.%d:%d ulen %d\n",
1244 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1245 NIPQUAD(saddr),
1246 ntohs(uh->source),
1247 NIPQUAD(daddr),
1248 ntohs(uh->dest),
1249 ulen);
1250 drop:
1251 UDP_INC_STATS_BH(UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1252 kfree_skb(skb);
1253 return 0;
1254 }
1255
1256 int udp_rcv(struct sk_buff *skb)
1257 {
1258 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP);
1259 }
1260
1261 int udp_destroy_sock(struct sock *sk)
1262 {
1263 lock_sock(sk);
1264 udp_flush_pending_frames(sk);
1265 release_sock(sk);
1266 return 0;
1267 }
1268
1269 /*
1270 * Socket option code for UDP
1271 */
1272 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1273 char __user *optval, int optlen,
1274 int (*push_pending_frames)(struct sock *))
1275 {
1276 struct udp_sock *up = udp_sk(sk);
1277 int val;
1278 int err = 0;
1279 int is_udplite = IS_UDPLITE(sk);
1280
1281 if (optlen<sizeof(int))
1282 return -EINVAL;
1283
1284 if (get_user(val, (int __user *)optval))
1285 return -EFAULT;
1286
1287 switch (optname) {
1288 case UDP_CORK:
1289 if (val != 0) {
1290 up->corkflag = 1;
1291 } else {
1292 up->corkflag = 0;
1293 lock_sock(sk);
1294 (*push_pending_frames)(sk);
1295 release_sock(sk);
1296 }
1297 break;
1298
1299 case UDP_ENCAP:
1300 switch (val) {
1301 case 0:
1302 case UDP_ENCAP_ESPINUDP:
1303 case UDP_ENCAP_ESPINUDP_NON_IKE:
1304 up->encap_rcv = xfrm4_udp_encap_rcv;
1305 /* FALLTHROUGH */
1306 case UDP_ENCAP_L2TPINUDP:
1307 up->encap_type = val;
1308 break;
1309 default:
1310 err = -ENOPROTOOPT;
1311 break;
1312 }
1313 break;
1314
1315 /*
1316 * UDP-Lite's partial checksum coverage (RFC 3828).
1317 */
1318 /* The sender sets actual checksum coverage length via this option.
1319 * The case coverage > packet length is handled by send module. */
1320 case UDPLITE_SEND_CSCOV:
1321 if (!is_udplite) /* Disable the option on UDP sockets */
1322 return -ENOPROTOOPT;
1323 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1324 val = 8;
1325 up->pcslen = val;
1326 up->pcflag |= UDPLITE_SEND_CC;
1327 break;
1328
1329 /* The receiver specifies a minimum checksum coverage value. To make
1330 * sense, this should be set to at least 8 (as done below). If zero is
1331 * used, this again means full checksum coverage. */
1332 case UDPLITE_RECV_CSCOV:
1333 if (!is_udplite) /* Disable the option on UDP sockets */
1334 return -ENOPROTOOPT;
1335 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1336 val = 8;
1337 up->pcrlen = val;
1338 up->pcflag |= UDPLITE_RECV_CC;
1339 break;
1340
1341 default:
1342 err = -ENOPROTOOPT;
1343 break;
1344 }
1345
1346 return err;
1347 }
1348
1349 int udp_setsockopt(struct sock *sk, int level, int optname,
1350 char __user *optval, int optlen)
1351 {
1352 if (level == SOL_UDP || level == SOL_UDPLITE)
1353 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1354 udp_push_pending_frames);
1355 return ip_setsockopt(sk, level, optname, optval, optlen);
1356 }
1357
1358 #ifdef CONFIG_COMPAT
1359 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1360 char __user *optval, int optlen)
1361 {
1362 if (level == SOL_UDP || level == SOL_UDPLITE)
1363 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1364 udp_push_pending_frames);
1365 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1366 }
1367 #endif
1368
1369 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1370 char __user *optval, int __user *optlen)
1371 {
1372 struct udp_sock *up = udp_sk(sk);
1373 int val, len;
1374
1375 if (get_user(len,optlen))
1376 return -EFAULT;
1377
1378 len = min_t(unsigned int, len, sizeof(int));
1379
1380 if (len < 0)
1381 return -EINVAL;
1382
1383 switch (optname) {
1384 case UDP_CORK:
1385 val = up->corkflag;
1386 break;
1387
1388 case UDP_ENCAP:
1389 val = up->encap_type;
1390 break;
1391
1392 /* The following two cannot be changed on UDP sockets, the return is
1393 * always 0 (which corresponds to the full checksum coverage of UDP). */
1394 case UDPLITE_SEND_CSCOV:
1395 val = up->pcslen;
1396 break;
1397
1398 case UDPLITE_RECV_CSCOV:
1399 val = up->pcrlen;
1400 break;
1401
1402 default:
1403 return -ENOPROTOOPT;
1404 }
1405
1406 if (put_user(len, optlen))
1407 return -EFAULT;
1408 if (copy_to_user(optval, &val,len))
1409 return -EFAULT;
1410 return 0;
1411 }
1412
1413 int udp_getsockopt(struct sock *sk, int level, int optname,
1414 char __user *optval, int __user *optlen)
1415 {
1416 if (level == SOL_UDP || level == SOL_UDPLITE)
1417 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1418 return ip_getsockopt(sk, level, optname, optval, optlen);
1419 }
1420
1421 #ifdef CONFIG_COMPAT
1422 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1423 char __user *optval, int __user *optlen)
1424 {
1425 if (level == SOL_UDP || level == SOL_UDPLITE)
1426 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1427 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1428 }
1429 #endif
1430 /**
1431 * udp_poll - wait for a UDP event.
1432 * @file - file struct
1433 * @sock - socket
1434 * @wait - poll table
1435 *
1436 * This is same as datagram poll, except for the special case of
1437 * blocking sockets. If application is using a blocking fd
1438 * and a packet with checksum error is in the queue;
1439 * then it could get return from select indicating data available
1440 * but then block when reading it. Add special case code
1441 * to work around these arguably broken applications.
1442 */
1443 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1444 {
1445 unsigned int mask = datagram_poll(file, sock, wait);
1446 struct sock *sk = sock->sk;
1447 int is_lite = IS_UDPLITE(sk);
1448
1449 /* Check for false positives due to checksum errors */
1450 if ( (mask & POLLRDNORM) &&
1451 !(file->f_flags & O_NONBLOCK) &&
1452 !(sk->sk_shutdown & RCV_SHUTDOWN)){
1453 struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1454 struct sk_buff *skb;
1455
1456 spin_lock_bh(&rcvq->lock);
1457 while ((skb = skb_peek(rcvq)) != NULL &&
1458 udp_lib_checksum_complete(skb)) {
1459 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_lite);
1460 __skb_unlink(skb, rcvq);
1461 kfree_skb(skb);
1462 }
1463 spin_unlock_bh(&rcvq->lock);
1464
1465 /* nothing to see, move along */
1466 if (skb == NULL)
1467 mask &= ~(POLLIN | POLLRDNORM);
1468 }
1469
1470 return mask;
1471
1472 }
1473
1474 DEFINE_PROTO_INUSE(udp)
1475
1476 struct proto udp_prot = {
1477 .name = "UDP",
1478 .owner = THIS_MODULE,
1479 .close = udp_lib_close,
1480 .connect = ip4_datagram_connect,
1481 .disconnect = udp_disconnect,
1482 .ioctl = udp_ioctl,
1483 .destroy = udp_destroy_sock,
1484 .setsockopt = udp_setsockopt,
1485 .getsockopt = udp_getsockopt,
1486 .sendmsg = udp_sendmsg,
1487 .recvmsg = udp_recvmsg,
1488 .sendpage = udp_sendpage,
1489 .backlog_rcv = udp_queue_rcv_skb,
1490 .hash = udp_lib_hash,
1491 .unhash = udp_lib_unhash,
1492 .get_port = udp_v4_get_port,
1493 .memory_allocated = &udp_memory_allocated,
1494 .sysctl_mem = sysctl_udp_mem,
1495 .sysctl_wmem = &sysctl_udp_wmem_min,
1496 .sysctl_rmem = &sysctl_udp_rmem_min,
1497 .obj_size = sizeof(struct udp_sock),
1498 #ifdef CONFIG_COMPAT
1499 .compat_setsockopt = compat_udp_setsockopt,
1500 .compat_getsockopt = compat_udp_getsockopt,
1501 #endif
1502 REF_PROTO_INUSE(udp)
1503 };
1504
1505 /* ------------------------------------------------------------------------ */
1506 #ifdef CONFIG_PROC_FS
1507
1508 static struct sock *udp_get_first(struct seq_file *seq)
1509 {
1510 struct sock *sk;
1511 struct udp_iter_state *state = seq->private;
1512
1513 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) {
1514 struct hlist_node *node;
1515 sk_for_each(sk, node, state->hashtable + state->bucket) {
1516 if (sk->sk_family == state->family)
1517 goto found;
1518 }
1519 }
1520 sk = NULL;
1521 found:
1522 return sk;
1523 }
1524
1525 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1526 {
1527 struct udp_iter_state *state = seq->private;
1528
1529 do {
1530 sk = sk_next(sk);
1531 try_again:
1532 ;
1533 } while (sk && sk->sk_family != state->family);
1534
1535 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) {
1536 sk = sk_head(state->hashtable + state->bucket);
1537 goto try_again;
1538 }
1539 return sk;
1540 }
1541
1542 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1543 {
1544 struct sock *sk = udp_get_first(seq);
1545
1546 if (sk)
1547 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1548 --pos;
1549 return pos ? NULL : sk;
1550 }
1551
1552 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1553 __acquires(udp_hash_lock)
1554 {
1555 read_lock(&udp_hash_lock);
1556 return *pos ? udp_get_idx(seq, *pos-1) : (void *)1;
1557 }
1558
1559 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1560 {
1561 struct sock *sk;
1562
1563 if (v == (void *)1)
1564 sk = udp_get_idx(seq, 0);
1565 else
1566 sk = udp_get_next(seq, v);
1567
1568 ++*pos;
1569 return sk;
1570 }
1571
1572 static void udp_seq_stop(struct seq_file *seq, void *v)
1573 __releases(udp_hash_lock)
1574 {
1575 read_unlock(&udp_hash_lock);
1576 }
1577
1578 static int udp_seq_open(struct inode *inode, struct file *file)
1579 {
1580 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1581 struct seq_file *seq;
1582 int rc = -ENOMEM;
1583 struct udp_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL);
1584
1585 if (!s)
1586 goto out;
1587 s->family = afinfo->family;
1588 s->hashtable = afinfo->hashtable;
1589 s->seq_ops.start = udp_seq_start;
1590 s->seq_ops.next = udp_seq_next;
1591 s->seq_ops.show = afinfo->seq_show;
1592 s->seq_ops.stop = udp_seq_stop;
1593
1594 rc = seq_open(file, &s->seq_ops);
1595 if (rc)
1596 goto out_kfree;
1597
1598 seq = file->private_data;
1599 seq->private = s;
1600 out:
1601 return rc;
1602 out_kfree:
1603 kfree(s);
1604 goto out;
1605 }
1606
1607 /* ------------------------------------------------------------------------ */
1608 int udp_proc_register(struct udp_seq_afinfo *afinfo)
1609 {
1610 struct proc_dir_entry *p;
1611 int rc = 0;
1612
1613 if (!afinfo)
1614 return -EINVAL;
1615 afinfo->seq_fops->owner = afinfo->owner;
1616 afinfo->seq_fops->open = udp_seq_open;
1617 afinfo->seq_fops->read = seq_read;
1618 afinfo->seq_fops->llseek = seq_lseek;
1619 afinfo->seq_fops->release = seq_release_private;
1620
1621 p = proc_net_fops_create(&init_net, afinfo->name, S_IRUGO, afinfo->seq_fops);
1622 if (p)
1623 p->data = afinfo;
1624 else
1625 rc = -ENOMEM;
1626 return rc;
1627 }
1628
1629 void udp_proc_unregister(struct udp_seq_afinfo *afinfo)
1630 {
1631 if (!afinfo)
1632 return;
1633 proc_net_remove(&init_net, afinfo->name);
1634 memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops));
1635 }
1636
1637 /* ------------------------------------------------------------------------ */
1638 static void udp4_format_sock(struct sock *sp, char *tmpbuf, int bucket)
1639 {
1640 struct inet_sock *inet = inet_sk(sp);
1641 __be32 dest = inet->daddr;
1642 __be32 src = inet->rcv_saddr;
1643 __u16 destp = ntohs(inet->dport);
1644 __u16 srcp = ntohs(inet->sport);
1645
1646 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
1647 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p",
1648 bucket, src, srcp, dest, destp, sp->sk_state,
1649 atomic_read(&sp->sk_wmem_alloc),
1650 atomic_read(&sp->sk_rmem_alloc),
1651 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1652 atomic_read(&sp->sk_refcnt), sp);
1653 }
1654
1655 int udp4_seq_show(struct seq_file *seq, void *v)
1656 {
1657 if (v == SEQ_START_TOKEN)
1658 seq_printf(seq, "%-127s\n",
1659 " sl local_address rem_address st tx_queue "
1660 "rx_queue tr tm->when retrnsmt uid timeout "
1661 "inode");
1662 else {
1663 char tmpbuf[129];
1664 struct udp_iter_state *state = seq->private;
1665
1666 udp4_format_sock(v, tmpbuf, state->bucket);
1667 seq_printf(seq, "%-127s\n", tmpbuf);
1668 }
1669 return 0;
1670 }
1671
1672 /* ------------------------------------------------------------------------ */
1673 static struct file_operations udp4_seq_fops;
1674 static struct udp_seq_afinfo udp4_seq_afinfo = {
1675 .owner = THIS_MODULE,
1676 .name = "udp",
1677 .family = AF_INET,
1678 .hashtable = udp_hash,
1679 .seq_show = udp4_seq_show,
1680 .seq_fops = &udp4_seq_fops,
1681 };
1682
1683 int __init udp4_proc_init(void)
1684 {
1685 return udp_proc_register(&udp4_seq_afinfo);
1686 }
1687
1688 void udp4_proc_exit(void)
1689 {
1690 udp_proc_unregister(&udp4_seq_afinfo);
1691 }
1692 #endif /* CONFIG_PROC_FS */
1693
1694 void __init udp_init(void)
1695 {
1696 unsigned long limit;
1697
1698 /* Set the pressure threshold up by the same strategy of TCP. It is a
1699 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
1700 * toward zero with the amount of memory, with a floor of 128 pages.
1701 */
1702 limit = min(nr_all_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
1703 limit = (limit * (nr_all_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
1704 limit = max(limit, 128UL);
1705 sysctl_udp_mem[0] = limit / 4 * 3;
1706 sysctl_udp_mem[1] = limit;
1707 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
1708
1709 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
1710 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
1711 }
1712
1713 EXPORT_SYMBOL(udp_disconnect);
1714 EXPORT_SYMBOL(udp_hash);
1715 EXPORT_SYMBOL(udp_hash_lock);
1716 EXPORT_SYMBOL(udp_ioctl);
1717 EXPORT_SYMBOL(udp_get_port);
1718 EXPORT_SYMBOL(udp_prot);
1719 EXPORT_SYMBOL(udp_sendmsg);
1720 EXPORT_SYMBOL(udp_lib_getsockopt);
1721 EXPORT_SYMBOL(udp_lib_setsockopt);
1722 EXPORT_SYMBOL(udp_poll);
1723
1724 #ifdef CONFIG_PROC_FS
1725 EXPORT_SYMBOL(udp_proc_register);
1726 EXPORT_SYMBOL(udp_proc_unregister);
1727 #endif
Linux kernel - Deliverables
This page took 0.064131 seconds and 6 git commands to generate.