[SK_BUFF]: Introduce udp_hdr(), remove skb->h.uh
[linux-2.6/mini2440.git] / net / ipv4 / udp.c
blob926404c5e58c64e60ddeea29f1e1c4b75d23b5c3
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.
6 * The User Datagram Protocol (UDP).
8 * Version: $Id: udp.c,v 1.102 2002/02/01 22:01:04 davem Exp $
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>
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
75 * This program is free software; you can redistribute it and/or
76 * modify it under the terms of the GNU General Public License
77 * as published by the Free Software Foundation; either version
78 * 2 of the License, or (at your option) any later version.
81 #include <asm/system.h>
82 #include <asm/uaccess.h>
83 #include <asm/ioctls.h>
84 #include <linux/types.h>
85 #include <linux/fcntl.h>
86 #include <linux/module.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/igmp.h>
90 #include <linux/in.h>
91 #include <linux/errno.h>
92 #include <linux/timer.h>
93 #include <linux/mm.h>
94 #include <linux/inet.h>
95 #include <linux/netdevice.h>
96 #include <net/tcp_states.h>
97 #include <linux/skbuff.h>
98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h>
100 #include <net/icmp.h>
101 #include <net/route.h>
102 #include <net/checksum.h>
103 #include <net/xfrm.h>
104 #include "udp_impl.h"
107 * Snmp MIB for the UDP layer
110 DEFINE_SNMP_STAT(struct udp_mib, udp_statistics) __read_mostly;
112 struct hlist_head udp_hash[UDP_HTABLE_SIZE];
113 DEFINE_RWLOCK(udp_hash_lock);
115 static int udp_port_rover;
117 static inline int __udp_lib_lport_inuse(__u16 num, struct hlist_head udptable[])
119 struct sock *sk;
120 struct hlist_node *node;
122 sk_for_each(sk, node, &udptable[num & (UDP_HTABLE_SIZE - 1)])
123 if (sk->sk_hash == num)
124 return 1;
125 return 0;
129 * __udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
131 * @sk: socket struct in question
132 * @snum: port number to look up
133 * @udptable: hash list table, must be of UDP_HTABLE_SIZE
134 * @port_rover: pointer to record of last unallocated port
135 * @saddr_comp: AF-dependent comparison of bound local IP addresses
137 int __udp_lib_get_port(struct sock *sk, unsigned short snum,
138 struct hlist_head udptable[], int *port_rover,
139 int (*saddr_comp)(const struct sock *sk1,
140 const struct sock *sk2 ) )
142 struct hlist_node *node;
143 struct hlist_head *head;
144 struct sock *sk2;
145 int error = 1;
147 write_lock_bh(&udp_hash_lock);
148 if (snum == 0) {
149 int best_size_so_far, best, result, i;
151 if (*port_rover > sysctl_local_port_range[1] ||
152 *port_rover < sysctl_local_port_range[0])
153 *port_rover = sysctl_local_port_range[0];
154 best_size_so_far = 32767;
155 best = result = *port_rover;
156 for (i = 0; i < UDP_HTABLE_SIZE; i++, result++) {
157 int size;
159 head = &udptable[result & (UDP_HTABLE_SIZE - 1)];
160 if (hlist_empty(head)) {
161 if (result > sysctl_local_port_range[1])
162 result = sysctl_local_port_range[0] +
163 ((result - sysctl_local_port_range[0]) &
164 (UDP_HTABLE_SIZE - 1));
165 goto gotit;
167 size = 0;
168 sk_for_each(sk2, node, head) {
169 if (++size >= best_size_so_far)
170 goto next;
172 best_size_so_far = size;
173 best = result;
174 next:
177 result = best;
178 for (i = 0; i < (1 << 16) / UDP_HTABLE_SIZE;
179 i++, result += UDP_HTABLE_SIZE) {
180 if (result > sysctl_local_port_range[1])
181 result = sysctl_local_port_range[0]
182 + ((result - sysctl_local_port_range[0]) &
183 (UDP_HTABLE_SIZE - 1));
184 if (! __udp_lib_lport_inuse(result, udptable))
185 break;
187 if (i >= (1 << 16) / UDP_HTABLE_SIZE)
188 goto fail;
189 gotit:
190 *port_rover = snum = result;
191 } else {
192 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
194 sk_for_each(sk2, node, head)
195 if (sk2->sk_hash == snum &&
196 sk2 != sk &&
197 (!sk2->sk_reuse || !sk->sk_reuse) &&
198 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if
199 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
200 (*saddr_comp)(sk, sk2) )
201 goto fail;
203 inet_sk(sk)->num = snum;
204 sk->sk_hash = snum;
205 if (sk_unhashed(sk)) {
206 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
207 sk_add_node(sk, head);
208 sock_prot_inc_use(sk->sk_prot);
210 error = 0;
211 fail:
212 write_unlock_bh(&udp_hash_lock);
213 return error;
216 int udp_get_port(struct sock *sk, unsigned short snum,
217 int (*scmp)(const struct sock *, const struct sock *))
219 return __udp_lib_get_port(sk, snum, udp_hash, &udp_port_rover, scmp);
222 int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
224 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
226 return ( !ipv6_only_sock(sk2) &&
227 (!inet1->rcv_saddr || !inet2->rcv_saddr ||
228 inet1->rcv_saddr == inet2->rcv_saddr ));
231 static inline int udp_v4_get_port(struct sock *sk, unsigned short snum)
233 return udp_get_port(sk, snum, ipv4_rcv_saddr_equal);
236 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
237 * harder than this. -DaveM
239 static struct sock *__udp4_lib_lookup(__be32 saddr, __be16 sport,
240 __be32 daddr, __be16 dport,
241 int dif, struct hlist_head udptable[])
243 struct sock *sk, *result = NULL;
244 struct hlist_node *node;
245 unsigned short hnum = ntohs(dport);
246 int badness = -1;
248 read_lock(&udp_hash_lock);
249 sk_for_each(sk, node, &udptable[hnum & (UDP_HTABLE_SIZE - 1)]) {
250 struct inet_sock *inet = inet_sk(sk);
252 if (sk->sk_hash == hnum && !ipv6_only_sock(sk)) {
253 int score = (sk->sk_family == PF_INET ? 1 : 0);
254 if (inet->rcv_saddr) {
255 if (inet->rcv_saddr != daddr)
256 continue;
257 score+=2;
259 if (inet->daddr) {
260 if (inet->daddr != saddr)
261 continue;
262 score+=2;
264 if (inet->dport) {
265 if (inet->dport != sport)
266 continue;
267 score+=2;
269 if (sk->sk_bound_dev_if) {
270 if (sk->sk_bound_dev_if != dif)
271 continue;
272 score+=2;
274 if (score == 9) {
275 result = sk;
276 break;
277 } else if (score > badness) {
278 result = sk;
279 badness = score;
283 if (result)
284 sock_hold(result);
285 read_unlock(&udp_hash_lock);
286 return result;
289 static inline struct sock *udp_v4_mcast_next(struct sock *sk,
290 __be16 loc_port, __be32 loc_addr,
291 __be16 rmt_port, __be32 rmt_addr,
292 int dif)
294 struct hlist_node *node;
295 struct sock *s = sk;
296 unsigned short hnum = ntohs(loc_port);
298 sk_for_each_from(s, node) {
299 struct inet_sock *inet = inet_sk(s);
301 if (s->sk_hash != hnum ||
302 (inet->daddr && inet->daddr != rmt_addr) ||
303 (inet->dport != rmt_port && inet->dport) ||
304 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) ||
305 ipv6_only_sock(s) ||
306 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
307 continue;
308 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
309 continue;
310 goto found;
312 s = NULL;
313 found:
314 return s;
318 * This routine is called by the ICMP module when it gets some
319 * sort of error condition. If err < 0 then the socket should
320 * be closed and the error returned to the user. If err > 0
321 * it's just the icmp type << 8 | icmp code.
322 * Header points to the ip header of the error packet. We move
323 * on past this. Then (as it used to claim before adjustment)
324 * header points to the first 8 bytes of the udp header. We need
325 * to find the appropriate port.
328 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[])
330 struct inet_sock *inet;
331 struct iphdr *iph = (struct iphdr*)skb->data;
332 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2));
333 int type = skb->h.icmph->type;
334 int code = skb->h.icmph->code;
335 struct sock *sk;
336 int harderr;
337 int err;
339 sk = __udp4_lib_lookup(iph->daddr, uh->dest, iph->saddr, uh->source,
340 skb->dev->ifindex, udptable );
341 if (sk == NULL) {
342 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
343 return; /* No socket for error */
346 err = 0;
347 harderr = 0;
348 inet = inet_sk(sk);
350 switch (type) {
351 default:
352 case ICMP_TIME_EXCEEDED:
353 err = EHOSTUNREACH;
354 break;
355 case ICMP_SOURCE_QUENCH:
356 goto out;
357 case ICMP_PARAMETERPROB:
358 err = EPROTO;
359 harderr = 1;
360 break;
361 case ICMP_DEST_UNREACH:
362 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
363 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
364 err = EMSGSIZE;
365 harderr = 1;
366 break;
368 goto out;
370 err = EHOSTUNREACH;
371 if (code <= NR_ICMP_UNREACH) {
372 harderr = icmp_err_convert[code].fatal;
373 err = icmp_err_convert[code].errno;
375 break;
379 * RFC1122: OK. Passes ICMP errors back to application, as per
380 * 4.1.3.3.
382 if (!inet->recverr) {
383 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
384 goto out;
385 } else {
386 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1));
388 sk->sk_err = err;
389 sk->sk_error_report(sk);
390 out:
391 sock_put(sk);
394 void udp_err(struct sk_buff *skb, u32 info)
396 return __udp4_lib_err(skb, info, udp_hash);
400 * Throw away all pending data and cancel the corking. Socket is locked.
402 static void udp_flush_pending_frames(struct sock *sk)
404 struct udp_sock *up = udp_sk(sk);
406 if (up->pending) {
407 up->len = 0;
408 up->pending = 0;
409 ip_flush_pending_frames(sk);
414 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
415 * @sk: socket we are sending on
416 * @skb: sk_buff containing the filled-in UDP header
417 * (checksum field must be zeroed out)
419 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
420 __be32 src, __be32 dst, int len )
422 unsigned int offset;
423 struct udphdr *uh = udp_hdr(skb);
424 __wsum csum = 0;
426 if (skb_queue_len(&sk->sk_write_queue) == 1) {
428 * Only one fragment on the socket.
430 skb->csum_offset = offsetof(struct udphdr, check);
431 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
432 } else {
434 * HW-checksum won't work as there are two or more
435 * fragments on the socket so that all csums of sk_buffs
436 * should be together
438 offset = skb_transport_offset(skb);
439 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
441 skb->ip_summed = CHECKSUM_NONE;
443 skb_queue_walk(&sk->sk_write_queue, skb) {
444 csum = csum_add(csum, skb->csum);
447 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
448 if (uh->check == 0)
449 uh->check = CSUM_MANGLED_0;
454 * Push out all pending data as one UDP datagram. Socket is locked.
456 static int udp_push_pending_frames(struct sock *sk)
458 struct udp_sock *up = udp_sk(sk);
459 struct inet_sock *inet = inet_sk(sk);
460 struct flowi *fl = &inet->cork.fl;
461 struct sk_buff *skb;
462 struct udphdr *uh;
463 int err = 0;
464 __wsum csum = 0;
466 /* Grab the skbuff where UDP header space exists. */
467 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
468 goto out;
471 * Create a UDP header
473 uh = udp_hdr(skb);
474 uh->source = fl->fl_ip_sport;
475 uh->dest = fl->fl_ip_dport;
476 uh->len = htons(up->len);
477 uh->check = 0;
479 if (up->pcflag) /* UDP-Lite */
480 csum = udplite_csum_outgoing(sk, skb);
482 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
484 skb->ip_summed = CHECKSUM_NONE;
485 goto send;
487 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
489 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len);
490 goto send;
492 } else /* `normal' UDP */
493 csum = udp_csum_outgoing(sk, skb);
495 /* add protocol-dependent pseudo-header */
496 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
497 sk->sk_protocol, csum );
498 if (uh->check == 0)
499 uh->check = CSUM_MANGLED_0;
501 send:
502 err = ip_push_pending_frames(sk);
503 out:
504 up->len = 0;
505 up->pending = 0;
506 return err;
509 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
510 size_t len)
512 struct inet_sock *inet = inet_sk(sk);
513 struct udp_sock *up = udp_sk(sk);
514 int ulen = len;
515 struct ipcm_cookie ipc;
516 struct rtable *rt = NULL;
517 int free = 0;
518 int connected = 0;
519 __be32 daddr, faddr, saddr;
520 __be16 dport;
521 u8 tos;
522 int err, is_udplite = up->pcflag;
523 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
524 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
526 if (len > 0xFFFF)
527 return -EMSGSIZE;
530 * Check the flags.
533 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */
534 return -EOPNOTSUPP;
536 ipc.opt = NULL;
538 if (up->pending) {
540 * There are pending frames.
541 * The socket lock must be held while it's corked.
543 lock_sock(sk);
544 if (likely(up->pending)) {
545 if (unlikely(up->pending != AF_INET)) {
546 release_sock(sk);
547 return -EINVAL;
549 goto do_append_data;
551 release_sock(sk);
553 ulen += sizeof(struct udphdr);
556 * Get and verify the address.
558 if (msg->msg_name) {
559 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name;
560 if (msg->msg_namelen < sizeof(*usin))
561 return -EINVAL;
562 if (usin->sin_family != AF_INET) {
563 if (usin->sin_family != AF_UNSPEC)
564 return -EAFNOSUPPORT;
567 daddr = usin->sin_addr.s_addr;
568 dport = usin->sin_port;
569 if (dport == 0)
570 return -EINVAL;
571 } else {
572 if (sk->sk_state != TCP_ESTABLISHED)
573 return -EDESTADDRREQ;
574 daddr = inet->daddr;
575 dport = inet->dport;
576 /* Open fast path for connected socket.
577 Route will not be used, if at least one option is set.
579 connected = 1;
581 ipc.addr = inet->saddr;
583 ipc.oif = sk->sk_bound_dev_if;
584 if (msg->msg_controllen) {
585 err = ip_cmsg_send(msg, &ipc);
586 if (err)
587 return err;
588 if (ipc.opt)
589 free = 1;
590 connected = 0;
592 if (!ipc.opt)
593 ipc.opt = inet->opt;
595 saddr = ipc.addr;
596 ipc.addr = faddr = daddr;
598 if (ipc.opt && ipc.opt->srr) {
599 if (!daddr)
600 return -EINVAL;
601 faddr = ipc.opt->faddr;
602 connected = 0;
604 tos = RT_TOS(inet->tos);
605 if (sock_flag(sk, SOCK_LOCALROUTE) ||
606 (msg->msg_flags & MSG_DONTROUTE) ||
607 (ipc.opt && ipc.opt->is_strictroute)) {
608 tos |= RTO_ONLINK;
609 connected = 0;
612 if (MULTICAST(daddr)) {
613 if (!ipc.oif)
614 ipc.oif = inet->mc_index;
615 if (!saddr)
616 saddr = inet->mc_addr;
617 connected = 0;
620 if (connected)
621 rt = (struct rtable*)sk_dst_check(sk, 0);
623 if (rt == NULL) {
624 struct flowi fl = { .oif = ipc.oif,
625 .nl_u = { .ip4_u =
626 { .daddr = faddr,
627 .saddr = saddr,
628 .tos = tos } },
629 .proto = sk->sk_protocol,
630 .uli_u = { .ports =
631 { .sport = inet->sport,
632 .dport = dport } } };
633 security_sk_classify_flow(sk, &fl);
634 err = ip_route_output_flow(&rt, &fl, sk, 1);
635 if (err)
636 goto out;
638 err = -EACCES;
639 if ((rt->rt_flags & RTCF_BROADCAST) &&
640 !sock_flag(sk, SOCK_BROADCAST))
641 goto out;
642 if (connected)
643 sk_dst_set(sk, dst_clone(&rt->u.dst));
646 if (msg->msg_flags&MSG_CONFIRM)
647 goto do_confirm;
648 back_from_confirm:
650 saddr = rt->rt_src;
651 if (!ipc.addr)
652 daddr = ipc.addr = rt->rt_dst;
654 lock_sock(sk);
655 if (unlikely(up->pending)) {
656 /* The socket is already corked while preparing it. */
657 /* ... which is an evident application bug. --ANK */
658 release_sock(sk);
660 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
661 err = -EINVAL;
662 goto out;
665 * Now cork the socket to pend data.
667 inet->cork.fl.fl4_dst = daddr;
668 inet->cork.fl.fl_ip_dport = dport;
669 inet->cork.fl.fl4_src = saddr;
670 inet->cork.fl.fl_ip_sport = inet->sport;
671 up->pending = AF_INET;
673 do_append_data:
674 up->len += ulen;
675 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
676 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
677 sizeof(struct udphdr), &ipc, rt,
678 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
679 if (err)
680 udp_flush_pending_frames(sk);
681 else if (!corkreq)
682 err = udp_push_pending_frames(sk);
683 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
684 up->pending = 0;
685 release_sock(sk);
687 out:
688 ip_rt_put(rt);
689 if (free)
690 kfree(ipc.opt);
691 if (!err) {
692 UDP_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS, is_udplite);
693 return len;
696 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
697 * ENOBUFS might not be good (it's not tunable per se), but otherwise
698 * we don't have a good statistic (IpOutDiscards but it can be too many
699 * things). We could add another new stat but at least for now that
700 * seems like overkill.
702 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
703 UDP_INC_STATS_USER(UDP_MIB_SNDBUFERRORS, is_udplite);
705 return err;
707 do_confirm:
708 dst_confirm(&rt->u.dst);
709 if (!(msg->msg_flags&MSG_PROBE) || len)
710 goto back_from_confirm;
711 err = 0;
712 goto out;
715 int udp_sendpage(struct sock *sk, struct page *page, int offset,
716 size_t size, int flags)
718 struct udp_sock *up = udp_sk(sk);
719 int ret;
721 if (!up->pending) {
722 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
724 /* Call udp_sendmsg to specify destination address which
725 * sendpage interface can't pass.
726 * This will succeed only when the socket is connected.
728 ret = udp_sendmsg(NULL, sk, &msg, 0);
729 if (ret < 0)
730 return ret;
733 lock_sock(sk);
735 if (unlikely(!up->pending)) {
736 release_sock(sk);
738 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
739 return -EINVAL;
742 ret = ip_append_page(sk, page, offset, size, flags);
743 if (ret == -EOPNOTSUPP) {
744 release_sock(sk);
745 return sock_no_sendpage(sk->sk_socket, page, offset,
746 size, flags);
748 if (ret < 0) {
749 udp_flush_pending_frames(sk);
750 goto out;
753 up->len += size;
754 if (!(up->corkflag || (flags&MSG_MORE)))
755 ret = udp_push_pending_frames(sk);
756 if (!ret)
757 ret = size;
758 out:
759 release_sock(sk);
760 return ret;
764 * IOCTL requests applicable to the UDP protocol
767 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
769 switch (cmd) {
770 case SIOCOUTQ:
772 int amount = atomic_read(&sk->sk_wmem_alloc);
773 return put_user(amount, (int __user *)arg);
776 case SIOCINQ:
778 struct sk_buff *skb;
779 unsigned long amount;
781 amount = 0;
782 spin_lock_bh(&sk->sk_receive_queue.lock);
783 skb = skb_peek(&sk->sk_receive_queue);
784 if (skb != NULL) {
786 * We will only return the amount
787 * of this packet since that is all
788 * that will be read.
790 amount = skb->len - sizeof(struct udphdr);
792 spin_unlock_bh(&sk->sk_receive_queue.lock);
793 return put_user(amount, (int __user *)arg);
796 default:
797 return -ENOIOCTLCMD;
800 return 0;
804 * This should be easy, if there is something there we
805 * return it, otherwise we block.
808 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
809 size_t len, int noblock, int flags, int *addr_len)
811 struct inet_sock *inet = inet_sk(sk);
812 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
813 struct sk_buff *skb;
814 unsigned int ulen, copied;
815 int err;
816 int is_udplite = IS_UDPLITE(sk);
819 * Check any passed addresses
821 if (addr_len)
822 *addr_len=sizeof(*sin);
824 if (flags & MSG_ERRQUEUE)
825 return ip_recv_error(sk, msg, len);
827 try_again:
828 skb = skb_recv_datagram(sk, flags, noblock, &err);
829 if (!skb)
830 goto out;
832 ulen = skb->len - sizeof(struct udphdr);
833 copied = len;
834 if (copied > ulen)
835 copied = ulen;
836 else if (copied < ulen)
837 msg->msg_flags |= MSG_TRUNC;
840 * If checksum is needed at all, try to do it while copying the
841 * data. If the data is truncated, or if we only want a partial
842 * coverage checksum (UDP-Lite), do it before the copy.
845 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
846 if (udp_lib_checksum_complete(skb))
847 goto csum_copy_err;
850 if (skb->ip_summed == CHECKSUM_UNNECESSARY)
851 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
852 msg->msg_iov, copied );
853 else {
854 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
856 if (err == -EINVAL)
857 goto csum_copy_err;
860 if (err)
861 goto out_free;
863 sock_recv_timestamp(msg, sk, skb);
865 /* Copy the address. */
866 if (sin)
868 sin->sin_family = AF_INET;
869 sin->sin_port = udp_hdr(skb)->source;
870 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
871 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
873 if (inet->cmsg_flags)
874 ip_cmsg_recv(msg, skb);
876 err = copied;
877 if (flags & MSG_TRUNC)
878 err = ulen;
880 out_free:
881 skb_free_datagram(sk, skb);
882 out:
883 return err;
885 csum_copy_err:
886 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite);
888 skb_kill_datagram(sk, skb, flags);
890 if (noblock)
891 return -EAGAIN;
892 goto try_again;
896 int udp_disconnect(struct sock *sk, int flags)
898 struct inet_sock *inet = inet_sk(sk);
900 * 1003.1g - break association.
903 sk->sk_state = TCP_CLOSE;
904 inet->daddr = 0;
905 inet->dport = 0;
906 sk->sk_bound_dev_if = 0;
907 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
908 inet_reset_saddr(sk);
910 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
911 sk->sk_prot->unhash(sk);
912 inet->sport = 0;
914 sk_dst_reset(sk);
915 return 0;
918 /* return:
919 * 1 if the the UDP system should process it
920 * 0 if we should drop this packet
921 * -1 if it should get processed by xfrm4_rcv_encap
923 static int udp_encap_rcv(struct sock * sk, struct sk_buff *skb)
925 #ifndef CONFIG_XFRM
926 return 1;
927 #else
928 struct udp_sock *up = udp_sk(sk);
929 struct udphdr *uh;
930 struct iphdr *iph;
931 int iphlen, len;
933 __u8 *udpdata;
934 __be32 *udpdata32;
935 __u16 encap_type = up->encap_type;
937 /* if we're overly short, let UDP handle it */
938 len = skb->len - sizeof(struct udphdr);
939 if (len <= 0)
940 return 1;
942 /* if this is not encapsulated socket, then just return now */
943 if (!encap_type)
944 return 1;
946 /* If this is a paged skb, make sure we pull up
947 * whatever data we need to look at. */
948 if (!pskb_may_pull(skb, sizeof(struct udphdr) + min(len, 8)))
949 return 1;
951 /* Now we can get the pointers */
952 uh = udp_hdr(skb);
953 udpdata = (__u8 *)uh + sizeof(struct udphdr);
954 udpdata32 = (__be32 *)udpdata;
956 switch (encap_type) {
957 default:
958 case UDP_ENCAP_ESPINUDP:
959 /* Check if this is a keepalive packet. If so, eat it. */
960 if (len == 1 && udpdata[0] == 0xff) {
961 return 0;
962 } else if (len > sizeof(struct ip_esp_hdr) && udpdata32[0] != 0) {
963 /* ESP Packet without Non-ESP header */
964 len = sizeof(struct udphdr);
965 } else
966 /* Must be an IKE packet.. pass it through */
967 return 1;
968 break;
969 case UDP_ENCAP_ESPINUDP_NON_IKE:
970 /* Check if this is a keepalive packet. If so, eat it. */
971 if (len == 1 && udpdata[0] == 0xff) {
972 return 0;
973 } else if (len > 2 * sizeof(u32) + sizeof(struct ip_esp_hdr) &&
974 udpdata32[0] == 0 && udpdata32[1] == 0) {
976 /* ESP Packet with Non-IKE marker */
977 len = sizeof(struct udphdr) + 2 * sizeof(u32);
978 } else
979 /* Must be an IKE packet.. pass it through */
980 return 1;
981 break;
984 /* At this point we are sure that this is an ESPinUDP packet,
985 * so we need to remove 'len' bytes from the packet (the UDP
986 * header and optional ESP marker bytes) and then modify the
987 * protocol to ESP, and then call into the transform receiver.
989 if (skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
990 return 0;
992 /* Now we can update and verify the packet length... */
993 iph = ip_hdr(skb);
994 iphlen = iph->ihl << 2;
995 iph->tot_len = htons(ntohs(iph->tot_len) - len);
996 if (skb->len < iphlen + len) {
997 /* packet is too small!?! */
998 return 0;
1001 /* pull the data buffer up to the ESP header and set the
1002 * transport header to point to ESP. Keep UDP on the stack
1003 * for later.
1005 __skb_pull(skb, len);
1006 skb_reset_transport_header(skb);
1008 /* modify the protocol (it's ESP!) */
1009 iph->protocol = IPPROTO_ESP;
1011 /* and let the caller know to send this into the ESP processor... */
1012 return -1;
1013 #endif
1016 /* returns:
1017 * -1: error
1018 * 0: success
1019 * >0: "udp encap" protocol resubmission
1021 * Note that in the success and error cases, the skb is assumed to
1022 * have either been requeued or freed.
1024 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
1026 struct udp_sock *up = udp_sk(sk);
1027 int rc;
1030 * Charge it to the socket, dropping if the queue is full.
1032 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1033 goto drop;
1034 nf_reset(skb);
1036 if (up->encap_type) {
1038 * This is an encapsulation socket, so let's see if this is
1039 * an encapsulated packet.
1040 * If it's a keepalive packet, then just eat it.
1041 * If it's an encapsulateed packet, then pass it to the
1042 * IPsec xfrm input and return the response
1043 * appropriately. Otherwise, just fall through and
1044 * pass this up the UDP socket.
1046 int ret;
1048 ret = udp_encap_rcv(sk, skb);
1049 if (ret == 0) {
1050 /* Eat the packet .. */
1051 kfree_skb(skb);
1052 return 0;
1054 if (ret < 0) {
1055 /* process the ESP packet */
1056 ret = xfrm4_rcv_encap(skb, up->encap_type);
1057 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS, up->pcflag);
1058 return -ret;
1060 /* FALLTHROUGH -- it's a UDP Packet */
1064 * UDP-Lite specific tests, ignored on UDP sockets
1066 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1069 * MIB statistics other than incrementing the error count are
1070 * disabled for the following two types of errors: these depend
1071 * on the application settings, not on the functioning of the
1072 * protocol stack as such.
1074 * RFC 3828 here recommends (sec 3.3): "There should also be a
1075 * way ... to ... at least let the receiving application block
1076 * delivery of packets with coverage values less than a value
1077 * provided by the application."
1079 if (up->pcrlen == 0) { /* full coverage was set */
1080 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1081 "%d while full coverage %d requested\n",
1082 UDP_SKB_CB(skb)->cscov, skb->len);
1083 goto drop;
1085 /* The next case involves violating the min. coverage requested
1086 * by the receiver. This is subtle: if receiver wants x and x is
1087 * greater than the buffersize/MTU then receiver will complain
1088 * that it wants x while sender emits packets of smaller size y.
1089 * Therefore the above ...()->partial_cov statement is essential.
1091 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1092 LIMIT_NETDEBUG(KERN_WARNING
1093 "UDPLITE: coverage %d too small, need min %d\n",
1094 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1095 goto drop;
1099 if (sk->sk_filter) {
1100 if (udp_lib_checksum_complete(skb))
1101 goto drop;
1104 if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) {
1105 /* Note that an ENOMEM error is charged twice */
1106 if (rc == -ENOMEM)
1107 UDP_INC_STATS_BH(UDP_MIB_RCVBUFERRORS, up->pcflag);
1108 goto drop;
1111 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS, up->pcflag);
1112 return 0;
1114 drop:
1115 UDP_INC_STATS_BH(UDP_MIB_INERRORS, up->pcflag);
1116 kfree_skb(skb);
1117 return -1;
1121 * Multicasts and broadcasts go to each listener.
1123 * Note: called only from the BH handler context,
1124 * so we don't need to lock the hashes.
1126 static int __udp4_lib_mcast_deliver(struct sk_buff *skb,
1127 struct udphdr *uh,
1128 __be32 saddr, __be32 daddr,
1129 struct hlist_head udptable[])
1131 struct sock *sk;
1132 int dif;
1134 read_lock(&udp_hash_lock);
1135 sk = sk_head(&udptable[ntohs(uh->dest) & (UDP_HTABLE_SIZE - 1)]);
1136 dif = skb->dev->ifindex;
1137 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
1138 if (sk) {
1139 struct sock *sknext = NULL;
1141 do {
1142 struct sk_buff *skb1 = skb;
1144 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr,
1145 uh->source, saddr, dif);
1146 if (sknext)
1147 skb1 = skb_clone(skb, GFP_ATOMIC);
1149 if (skb1) {
1150 int ret = udp_queue_rcv_skb(sk, skb1);
1151 if (ret > 0)
1152 /* we should probably re-process instead
1153 * of dropping packets here. */
1154 kfree_skb(skb1);
1156 sk = sknext;
1157 } while (sknext);
1158 } else
1159 kfree_skb(skb);
1160 read_unlock(&udp_hash_lock);
1161 return 0;
1164 /* Initialize UDP checksum. If exited with zero value (success),
1165 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1166 * Otherwise, csum completion requires chacksumming packet body,
1167 * including udp header and folding it to skb->csum.
1169 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1170 int proto)
1172 const struct iphdr *iph;
1173 int err;
1175 UDP_SKB_CB(skb)->partial_cov = 0;
1176 UDP_SKB_CB(skb)->cscov = skb->len;
1178 if (proto == IPPROTO_UDPLITE) {
1179 err = udplite_checksum_init(skb, uh);
1180 if (err)
1181 return err;
1184 iph = ip_hdr(skb);
1185 if (uh->check == 0) {
1186 skb->ip_summed = CHECKSUM_UNNECESSARY;
1187 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1188 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1189 proto, skb->csum))
1190 skb->ip_summed = CHECKSUM_UNNECESSARY;
1192 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
1193 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1194 skb->len, proto, 0);
1195 /* Probably, we should checksum udp header (it should be in cache
1196 * in any case) and data in tiny packets (< rx copybreak).
1199 return 0;
1203 * All we need to do is get the socket, and then do a checksum.
1206 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
1207 int proto)
1209 struct sock *sk;
1210 struct udphdr *uh = udp_hdr(skb);
1211 unsigned short ulen;
1212 struct rtable *rt = (struct rtable*)skb->dst;
1213 __be32 saddr = ip_hdr(skb)->saddr;
1214 __be32 daddr = ip_hdr(skb)->daddr;
1217 * Validate the packet.
1219 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1220 goto drop; /* No space for header. */
1222 ulen = ntohs(uh->len);
1223 if (ulen > skb->len)
1224 goto short_packet;
1226 if (proto == IPPROTO_UDP) {
1227 /* UDP validates ulen. */
1228 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1229 goto short_packet;
1230 uh = udp_hdr(skb);
1233 if (udp4_csum_init(skb, uh, proto))
1234 goto csum_error;
1236 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1237 return __udp4_lib_mcast_deliver(skb, uh, saddr, daddr, udptable);
1239 sk = __udp4_lib_lookup(saddr, uh->source, daddr, uh->dest,
1240 skb->dev->ifindex, udptable );
1242 if (sk != NULL) {
1243 int ret = udp_queue_rcv_skb(sk, skb);
1244 sock_put(sk);
1246 /* a return value > 0 means to resubmit the input, but
1247 * it wants the return to be -protocol, or 0
1249 if (ret > 0)
1250 return -ret;
1251 return 0;
1254 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1255 goto drop;
1256 nf_reset(skb);
1258 /* No socket. Drop packet silently, if checksum is wrong */
1259 if (udp_lib_checksum_complete(skb))
1260 goto csum_error;
1262 UDP_INC_STATS_BH(UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1263 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1266 * Hmm. We got an UDP packet to a port to which we
1267 * don't wanna listen. Ignore it.
1269 kfree_skb(skb);
1270 return 0;
1272 short_packet:
1273 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %u.%u.%u.%u:%u %d/%d to %u.%u.%u.%u:%u\n",
1274 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1275 NIPQUAD(saddr),
1276 ntohs(uh->source),
1277 ulen,
1278 skb->len,
1279 NIPQUAD(daddr),
1280 ntohs(uh->dest));
1281 goto drop;
1283 csum_error:
1285 * RFC1122: OK. Discards the bad packet silently (as far as
1286 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1288 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %d.%d.%d.%d:%d to %d.%d.%d.%d:%d ulen %d\n",
1289 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1290 NIPQUAD(saddr),
1291 ntohs(uh->source),
1292 NIPQUAD(daddr),
1293 ntohs(uh->dest),
1294 ulen);
1295 drop:
1296 UDP_INC_STATS_BH(UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1297 kfree_skb(skb);
1298 return 0;
1301 int udp_rcv(struct sk_buff *skb)
1303 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP);
1306 int udp_destroy_sock(struct sock *sk)
1308 lock_sock(sk);
1309 udp_flush_pending_frames(sk);
1310 release_sock(sk);
1311 return 0;
1315 * Socket option code for UDP
1317 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1318 char __user *optval, int optlen,
1319 int (*push_pending_frames)(struct sock *))
1321 struct udp_sock *up = udp_sk(sk);
1322 int val;
1323 int err = 0;
1325 if (optlen<sizeof(int))
1326 return -EINVAL;
1328 if (get_user(val, (int __user *)optval))
1329 return -EFAULT;
1331 switch (optname) {
1332 case UDP_CORK:
1333 if (val != 0) {
1334 up->corkflag = 1;
1335 } else {
1336 up->corkflag = 0;
1337 lock_sock(sk);
1338 (*push_pending_frames)(sk);
1339 release_sock(sk);
1341 break;
1343 case UDP_ENCAP:
1344 switch (val) {
1345 case 0:
1346 case UDP_ENCAP_ESPINUDP:
1347 case UDP_ENCAP_ESPINUDP_NON_IKE:
1348 up->encap_type = val;
1349 break;
1350 default:
1351 err = -ENOPROTOOPT;
1352 break;
1354 break;
1357 * UDP-Lite's partial checksum coverage (RFC 3828).
1359 /* The sender sets actual checksum coverage length via this option.
1360 * The case coverage > packet length is handled by send module. */
1361 case UDPLITE_SEND_CSCOV:
1362 if (!up->pcflag) /* Disable the option on UDP sockets */
1363 return -ENOPROTOOPT;
1364 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1365 val = 8;
1366 up->pcslen = val;
1367 up->pcflag |= UDPLITE_SEND_CC;
1368 break;
1370 /* The receiver specifies a minimum checksum coverage value. To make
1371 * sense, this should be set to at least 8 (as done below). If zero is
1372 * used, this again means full checksum coverage. */
1373 case UDPLITE_RECV_CSCOV:
1374 if (!up->pcflag) /* Disable the option on UDP sockets */
1375 return -ENOPROTOOPT;
1376 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1377 val = 8;
1378 up->pcrlen = val;
1379 up->pcflag |= UDPLITE_RECV_CC;
1380 break;
1382 default:
1383 err = -ENOPROTOOPT;
1384 break;
1387 return err;
1390 int udp_setsockopt(struct sock *sk, int level, int optname,
1391 char __user *optval, int optlen)
1393 if (level == SOL_UDP || level == SOL_UDPLITE)
1394 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1395 udp_push_pending_frames);
1396 return ip_setsockopt(sk, level, optname, optval, optlen);
1399 #ifdef CONFIG_COMPAT
1400 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1401 char __user *optval, int optlen)
1403 if (level == SOL_UDP || level == SOL_UDPLITE)
1404 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1405 udp_push_pending_frames);
1406 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1408 #endif
1410 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1411 char __user *optval, int __user *optlen)
1413 struct udp_sock *up = udp_sk(sk);
1414 int val, len;
1416 if (get_user(len,optlen))
1417 return -EFAULT;
1419 len = min_t(unsigned int, len, sizeof(int));
1421 if (len < 0)
1422 return -EINVAL;
1424 switch (optname) {
1425 case UDP_CORK:
1426 val = up->corkflag;
1427 break;
1429 case UDP_ENCAP:
1430 val = up->encap_type;
1431 break;
1433 /* The following two cannot be changed on UDP sockets, the return is
1434 * always 0 (which corresponds to the full checksum coverage of UDP). */
1435 case UDPLITE_SEND_CSCOV:
1436 val = up->pcslen;
1437 break;
1439 case UDPLITE_RECV_CSCOV:
1440 val = up->pcrlen;
1441 break;
1443 default:
1444 return -ENOPROTOOPT;
1447 if (put_user(len, optlen))
1448 return -EFAULT;
1449 if (copy_to_user(optval, &val,len))
1450 return -EFAULT;
1451 return 0;
1454 int udp_getsockopt(struct sock *sk, int level, int optname,
1455 char __user *optval, int __user *optlen)
1457 if (level == SOL_UDP || level == SOL_UDPLITE)
1458 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1459 return ip_getsockopt(sk, level, optname, optval, optlen);
1462 #ifdef CONFIG_COMPAT
1463 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1464 char __user *optval, int __user *optlen)
1466 if (level == SOL_UDP || level == SOL_UDPLITE)
1467 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1468 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1470 #endif
1472 * udp_poll - wait for a UDP event.
1473 * @file - file struct
1474 * @sock - socket
1475 * @wait - poll table
1477 * This is same as datagram poll, except for the special case of
1478 * blocking sockets. If application is using a blocking fd
1479 * and a packet with checksum error is in the queue;
1480 * then it could get return from select indicating data available
1481 * but then block when reading it. Add special case code
1482 * to work around these arguably broken applications.
1484 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1486 unsigned int mask = datagram_poll(file, sock, wait);
1487 struct sock *sk = sock->sk;
1488 int is_lite = IS_UDPLITE(sk);
1490 /* Check for false positives due to checksum errors */
1491 if ( (mask & POLLRDNORM) &&
1492 !(file->f_flags & O_NONBLOCK) &&
1493 !(sk->sk_shutdown & RCV_SHUTDOWN)){
1494 struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1495 struct sk_buff *skb;
1497 spin_lock_bh(&rcvq->lock);
1498 while ((skb = skb_peek(rcvq)) != NULL &&
1499 udp_lib_checksum_complete(skb)) {
1500 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_lite);
1501 __skb_unlink(skb, rcvq);
1502 kfree_skb(skb);
1504 spin_unlock_bh(&rcvq->lock);
1506 /* nothing to see, move along */
1507 if (skb == NULL)
1508 mask &= ~(POLLIN | POLLRDNORM);
1511 return mask;
1515 struct proto udp_prot = {
1516 .name = "UDP",
1517 .owner = THIS_MODULE,
1518 .close = udp_lib_close,
1519 .connect = ip4_datagram_connect,
1520 .disconnect = udp_disconnect,
1521 .ioctl = udp_ioctl,
1522 .destroy = udp_destroy_sock,
1523 .setsockopt = udp_setsockopt,
1524 .getsockopt = udp_getsockopt,
1525 .sendmsg = udp_sendmsg,
1526 .recvmsg = udp_recvmsg,
1527 .sendpage = udp_sendpage,
1528 .backlog_rcv = udp_queue_rcv_skb,
1529 .hash = udp_lib_hash,
1530 .unhash = udp_lib_unhash,
1531 .get_port = udp_v4_get_port,
1532 .obj_size = sizeof(struct udp_sock),
1533 #ifdef CONFIG_COMPAT
1534 .compat_setsockopt = compat_udp_setsockopt,
1535 .compat_getsockopt = compat_udp_getsockopt,
1536 #endif
1539 /* ------------------------------------------------------------------------ */
1540 #ifdef CONFIG_PROC_FS
1542 static struct sock *udp_get_first(struct seq_file *seq)
1544 struct sock *sk;
1545 struct udp_iter_state *state = seq->private;
1547 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) {
1548 struct hlist_node *node;
1549 sk_for_each(sk, node, state->hashtable + state->bucket) {
1550 if (sk->sk_family == state->family)
1551 goto found;
1554 sk = NULL;
1555 found:
1556 return sk;
1559 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1561 struct udp_iter_state *state = seq->private;
1563 do {
1564 sk = sk_next(sk);
1565 try_again:
1567 } while (sk && sk->sk_family != state->family);
1569 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) {
1570 sk = sk_head(state->hashtable + state->bucket);
1571 goto try_again;
1573 return sk;
1576 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1578 struct sock *sk = udp_get_first(seq);
1580 if (sk)
1581 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1582 --pos;
1583 return pos ? NULL : sk;
1586 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1588 read_lock(&udp_hash_lock);
1589 return *pos ? udp_get_idx(seq, *pos-1) : (void *)1;
1592 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1594 struct sock *sk;
1596 if (v == (void *)1)
1597 sk = udp_get_idx(seq, 0);
1598 else
1599 sk = udp_get_next(seq, v);
1601 ++*pos;
1602 return sk;
1605 static void udp_seq_stop(struct seq_file *seq, void *v)
1607 read_unlock(&udp_hash_lock);
1610 static int udp_seq_open(struct inode *inode, struct file *file)
1612 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1613 struct seq_file *seq;
1614 int rc = -ENOMEM;
1615 struct udp_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL);
1617 if (!s)
1618 goto out;
1619 s->family = afinfo->family;
1620 s->hashtable = afinfo->hashtable;
1621 s->seq_ops.start = udp_seq_start;
1622 s->seq_ops.next = udp_seq_next;
1623 s->seq_ops.show = afinfo->seq_show;
1624 s->seq_ops.stop = udp_seq_stop;
1626 rc = seq_open(file, &s->seq_ops);
1627 if (rc)
1628 goto out_kfree;
1630 seq = file->private_data;
1631 seq->private = s;
1632 out:
1633 return rc;
1634 out_kfree:
1635 kfree(s);
1636 goto out;
1639 /* ------------------------------------------------------------------------ */
1640 int udp_proc_register(struct udp_seq_afinfo *afinfo)
1642 struct proc_dir_entry *p;
1643 int rc = 0;
1645 if (!afinfo)
1646 return -EINVAL;
1647 afinfo->seq_fops->owner = afinfo->owner;
1648 afinfo->seq_fops->open = udp_seq_open;
1649 afinfo->seq_fops->read = seq_read;
1650 afinfo->seq_fops->llseek = seq_lseek;
1651 afinfo->seq_fops->release = seq_release_private;
1653 p = proc_net_fops_create(afinfo->name, S_IRUGO, afinfo->seq_fops);
1654 if (p)
1655 p->data = afinfo;
1656 else
1657 rc = -ENOMEM;
1658 return rc;
1661 void udp_proc_unregister(struct udp_seq_afinfo *afinfo)
1663 if (!afinfo)
1664 return;
1665 proc_net_remove(afinfo->name);
1666 memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops));
1669 /* ------------------------------------------------------------------------ */
1670 static void udp4_format_sock(struct sock *sp, char *tmpbuf, int bucket)
1672 struct inet_sock *inet = inet_sk(sp);
1673 __be32 dest = inet->daddr;
1674 __be32 src = inet->rcv_saddr;
1675 __u16 destp = ntohs(inet->dport);
1676 __u16 srcp = ntohs(inet->sport);
1678 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
1679 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p",
1680 bucket, src, srcp, dest, destp, sp->sk_state,
1681 atomic_read(&sp->sk_wmem_alloc),
1682 atomic_read(&sp->sk_rmem_alloc),
1683 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1684 atomic_read(&sp->sk_refcnt), sp);
1687 int udp4_seq_show(struct seq_file *seq, void *v)
1689 if (v == SEQ_START_TOKEN)
1690 seq_printf(seq, "%-127s\n",
1691 " sl local_address rem_address st tx_queue "
1692 "rx_queue tr tm->when retrnsmt uid timeout "
1693 "inode");
1694 else {
1695 char tmpbuf[129];
1696 struct udp_iter_state *state = seq->private;
1698 udp4_format_sock(v, tmpbuf, state->bucket);
1699 seq_printf(seq, "%-127s\n", tmpbuf);
1701 return 0;
1704 /* ------------------------------------------------------------------------ */
1705 static struct file_operations udp4_seq_fops;
1706 static struct udp_seq_afinfo udp4_seq_afinfo = {
1707 .owner = THIS_MODULE,
1708 .name = "udp",
1709 .family = AF_INET,
1710 .hashtable = udp_hash,
1711 .seq_show = udp4_seq_show,
1712 .seq_fops = &udp4_seq_fops,
1715 int __init udp4_proc_init(void)
1717 return udp_proc_register(&udp4_seq_afinfo);
1720 void udp4_proc_exit(void)
1722 udp_proc_unregister(&udp4_seq_afinfo);
1724 #endif /* CONFIG_PROC_FS */
1726 EXPORT_SYMBOL(udp_disconnect);
1727 EXPORT_SYMBOL(udp_hash);
1728 EXPORT_SYMBOL(udp_hash_lock);
1729 EXPORT_SYMBOL(udp_ioctl);
1730 EXPORT_SYMBOL(udp_get_port);
1731 EXPORT_SYMBOL(udp_prot);
1732 EXPORT_SYMBOL(udp_sendmsg);
1733 EXPORT_SYMBOL(udp_lib_getsockopt);
1734 EXPORT_SYMBOL(udp_lib_setsockopt);
1735 EXPORT_SYMBOL(udp_poll);
1737 #ifdef CONFIG_PROC_FS
1738 EXPORT_SYMBOL(udp_proc_register);
1739 EXPORT_SYMBOL(udp_proc_unregister);
1740 #endif