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 $
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>
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
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
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
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.
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.
82 #include <asm/system.h>
83 #include <asm/uaccess.h>
84 #include <asm/ioctls.h>
85 #include <linux/types.h>
86 #include <linux/fcntl.h>
87 #include <linux/module.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/igmp.h>
92 #include <linux/errno.h>
93 #include <linux/timer.h>
95 #include <linux/inet.h>
96 #include <linux/netdevice.h>
97 #include <net/tcp_states.h>
98 #include <linux/skbuff.h>
99 #include <linux/proc_fs.h>
100 #include <linux/seq_file.h>
101 #include <net/net_namespace.h>
102 #include <net/icmp.h>
103 #include <net/route.h>
104 #include <net/checksum.h>
105 #include <net/xfrm.h>
106 #include "udp_impl.h"
109 * Snmp MIB for the UDP layer
112 DEFINE_SNMP_STAT(struct udp_mib
, udp_statistics
) __read_mostly
;
113 EXPORT_SYMBOL(udp_statistics
);
115 struct hlist_head udp_hash
[UDP_HTABLE_SIZE
];
116 DEFINE_RWLOCK(udp_hash_lock
);
118 static inline int __udp_lib_lport_inuse(__u16 num
,
119 const struct hlist_head udptable
[])
122 struct hlist_node
*node
;
124 sk_for_each(sk
, node
, &udptable
[num
& (UDP_HTABLE_SIZE
- 1)])
125 if (sk
->sk_hash
== num
)
131 * __udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
133 * @sk: socket struct in question
134 * @snum: port number to look up
135 * @udptable: hash list table, must be of UDP_HTABLE_SIZE
136 * @saddr_comp: AF-dependent comparison of bound local IP addresses
138 int __udp_lib_get_port(struct sock
*sk
, unsigned short snum
,
139 struct hlist_head udptable
[],
140 int (*saddr_comp
)(const struct sock
*sk1
,
141 const struct sock
*sk2
) )
143 struct hlist_node
*node
;
144 struct hlist_head
*head
;
148 write_lock_bh(&udp_hash_lock
);
151 int i
, low
, high
, remaining
;
152 unsigned rover
, best
, best_size_so_far
;
154 inet_get_local_port_range(&low
, &high
);
155 remaining
= (high
- low
) + 1;
157 best_size_so_far
= UINT_MAX
;
158 best
= rover
= net_random() % remaining
+ low
;
160 /* 1st pass: look for empty (or shortest) hash chain */
161 for (i
= 0; i
< UDP_HTABLE_SIZE
; i
++) {
164 head
= &udptable
[rover
& (UDP_HTABLE_SIZE
- 1)];
165 if (hlist_empty(head
))
168 sk_for_each(sk2
, node
, head
) {
169 if (++size
>= best_size_so_far
)
172 best_size_so_far
= size
;
175 /* fold back if end of range */
177 rover
= low
+ ((rover
- low
)
178 & (UDP_HTABLE_SIZE
- 1));
183 /* 2nd pass: find hole in shortest hash chain */
185 for (i
= 0; i
< (1 << 16) / UDP_HTABLE_SIZE
; i
++) {
186 if (! __udp_lib_lport_inuse(rover
, udptable
))
188 rover
+= UDP_HTABLE_SIZE
;
190 rover
= low
+ ((rover
- low
)
191 & (UDP_HTABLE_SIZE
- 1));
195 /* All ports in use! */
201 head
= &udptable
[snum
& (UDP_HTABLE_SIZE
- 1)];
203 sk_for_each(sk2
, node
, head
)
204 if (sk2
->sk_hash
== snum
&&
206 (!sk2
->sk_reuse
|| !sk
->sk_reuse
) &&
207 (!sk2
->sk_bound_dev_if
|| !sk
->sk_bound_dev_if
208 || sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
209 (*saddr_comp
)(sk
, sk2
) )
213 inet_sk(sk
)->num
= snum
;
215 if (sk_unhashed(sk
)) {
216 head
= &udptable
[snum
& (UDP_HTABLE_SIZE
- 1)];
217 sk_add_node(sk
, head
);
218 sock_prot_inc_use(sk
->sk_prot
);
222 write_unlock_bh(&udp_hash_lock
);
226 int udp_get_port(struct sock
*sk
, unsigned short snum
,
227 int (*scmp
)(const struct sock
*, const struct sock
*))
229 return __udp_lib_get_port(sk
, snum
, udp_hash
, scmp
);
232 int ipv4_rcv_saddr_equal(const struct sock
*sk1
, const struct sock
*sk2
)
234 struct inet_sock
*inet1
= inet_sk(sk1
), *inet2
= inet_sk(sk2
);
236 return ( !ipv6_only_sock(sk2
) &&
237 (!inet1
->rcv_saddr
|| !inet2
->rcv_saddr
||
238 inet1
->rcv_saddr
== inet2
->rcv_saddr
));
241 static inline int udp_v4_get_port(struct sock
*sk
, unsigned short snum
)
243 return udp_get_port(sk
, snum
, ipv4_rcv_saddr_equal
);
246 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
247 * harder than this. -DaveM
249 static struct sock
*__udp4_lib_lookup(__be32 saddr
, __be16 sport
,
250 __be32 daddr
, __be16 dport
,
251 int dif
, struct hlist_head udptable
[])
253 struct sock
*sk
, *result
= NULL
;
254 struct hlist_node
*node
;
255 unsigned short hnum
= ntohs(dport
);
258 read_lock(&udp_hash_lock
);
259 sk_for_each(sk
, node
, &udptable
[hnum
& (UDP_HTABLE_SIZE
- 1)]) {
260 struct inet_sock
*inet
= inet_sk(sk
);
262 if (sk
->sk_hash
== hnum
&& !ipv6_only_sock(sk
)) {
263 int score
= (sk
->sk_family
== PF_INET
? 1 : 0);
264 if (inet
->rcv_saddr
) {
265 if (inet
->rcv_saddr
!= daddr
)
270 if (inet
->daddr
!= saddr
)
275 if (inet
->dport
!= sport
)
279 if (sk
->sk_bound_dev_if
) {
280 if (sk
->sk_bound_dev_if
!= dif
)
287 } else if (score
> badness
) {
295 read_unlock(&udp_hash_lock
);
299 static inline struct sock
*udp_v4_mcast_next(struct sock
*sk
,
300 __be16 loc_port
, __be32 loc_addr
,
301 __be16 rmt_port
, __be32 rmt_addr
,
304 struct hlist_node
*node
;
306 unsigned short hnum
= ntohs(loc_port
);
308 sk_for_each_from(s
, node
) {
309 struct inet_sock
*inet
= inet_sk(s
);
311 if (s
->sk_hash
!= hnum
||
312 (inet
->daddr
&& inet
->daddr
!= rmt_addr
) ||
313 (inet
->dport
!= rmt_port
&& inet
->dport
) ||
314 (inet
->rcv_saddr
&& inet
->rcv_saddr
!= loc_addr
) ||
316 (s
->sk_bound_dev_if
&& s
->sk_bound_dev_if
!= dif
))
318 if (!ip_mc_sf_allow(s
, loc_addr
, rmt_addr
, dif
))
328 * This routine is called by the ICMP module when it gets some
329 * sort of error condition. If err < 0 then the socket should
330 * be closed and the error returned to the user. If err > 0
331 * it's just the icmp type << 8 | icmp code.
332 * Header points to the ip header of the error packet. We move
333 * on past this. Then (as it used to claim before adjustment)
334 * header points to the first 8 bytes of the udp header. We need
335 * to find the appropriate port.
338 void __udp4_lib_err(struct sk_buff
*skb
, u32 info
, struct hlist_head udptable
[])
340 struct inet_sock
*inet
;
341 struct iphdr
*iph
= (struct iphdr
*)skb
->data
;
342 struct udphdr
*uh
= (struct udphdr
*)(skb
->data
+(iph
->ihl
<<2));
343 const int type
= icmp_hdr(skb
)->type
;
344 const int code
= icmp_hdr(skb
)->code
;
349 sk
= __udp4_lib_lookup(iph
->daddr
, uh
->dest
, iph
->saddr
, uh
->source
,
350 skb
->dev
->ifindex
, udptable
);
352 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS
);
353 return; /* No socket for error */
362 case ICMP_TIME_EXCEEDED
:
365 case ICMP_SOURCE_QUENCH
:
367 case ICMP_PARAMETERPROB
:
371 case ICMP_DEST_UNREACH
:
372 if (code
== ICMP_FRAG_NEEDED
) { /* Path MTU discovery */
373 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
) {
381 if (code
<= NR_ICMP_UNREACH
) {
382 harderr
= icmp_err_convert
[code
].fatal
;
383 err
= icmp_err_convert
[code
].errno
;
389 * RFC1122: OK. Passes ICMP errors back to application, as per
392 if (!inet
->recverr
) {
393 if (!harderr
|| sk
->sk_state
!= TCP_ESTABLISHED
)
396 ip_icmp_error(sk
, skb
, err
, uh
->dest
, info
, (u8
*)(uh
+1));
399 sk
->sk_error_report(sk
);
404 void udp_err(struct sk_buff
*skb
, u32 info
)
406 return __udp4_lib_err(skb
, info
, udp_hash
);
410 * Throw away all pending data and cancel the corking. Socket is locked.
412 static void udp_flush_pending_frames(struct sock
*sk
)
414 struct udp_sock
*up
= udp_sk(sk
);
419 ip_flush_pending_frames(sk
);
424 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
425 * @sk: socket we are sending on
426 * @skb: sk_buff containing the filled-in UDP header
427 * (checksum field must be zeroed out)
429 static void udp4_hwcsum_outgoing(struct sock
*sk
, struct sk_buff
*skb
,
430 __be32 src
, __be32 dst
, int len
)
433 struct udphdr
*uh
= udp_hdr(skb
);
436 if (skb_queue_len(&sk
->sk_write_queue
) == 1) {
438 * Only one fragment on the socket.
440 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
441 skb
->csum_offset
= offsetof(struct udphdr
, check
);
442 uh
->check
= ~csum_tcpudp_magic(src
, dst
, len
, IPPROTO_UDP
, 0);
445 * HW-checksum won't work as there are two or more
446 * fragments on the socket so that all csums of sk_buffs
449 offset
= skb_transport_offset(skb
);
450 skb
->csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
452 skb
->ip_summed
= CHECKSUM_NONE
;
454 skb_queue_walk(&sk
->sk_write_queue
, skb
) {
455 csum
= csum_add(csum
, skb
->csum
);
458 uh
->check
= csum_tcpudp_magic(src
, dst
, len
, IPPROTO_UDP
, csum
);
460 uh
->check
= CSUM_MANGLED_0
;
465 * Push out all pending data as one UDP datagram. Socket is locked.
467 static int udp_push_pending_frames(struct sock
*sk
)
469 struct udp_sock
*up
= udp_sk(sk
);
470 struct inet_sock
*inet
= inet_sk(sk
);
471 struct flowi
*fl
= &inet
->cork
.fl
;
475 int is_udplite
= IS_UDPLITE(sk
);
478 /* Grab the skbuff where UDP header space exists. */
479 if ((skb
= skb_peek(&sk
->sk_write_queue
)) == NULL
)
483 * Create a UDP header
486 uh
->source
= fl
->fl_ip_sport
;
487 uh
->dest
= fl
->fl_ip_dport
;
488 uh
->len
= htons(up
->len
);
491 if (is_udplite
) /* UDP-Lite */
492 csum
= udplite_csum_outgoing(sk
, skb
);
494 else if (sk
->sk_no_check
== UDP_CSUM_NOXMIT
) { /* UDP csum disabled */
496 skb
->ip_summed
= CHECKSUM_NONE
;
499 } else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) { /* UDP hardware csum */
501 udp4_hwcsum_outgoing(sk
, skb
, fl
->fl4_src
,fl
->fl4_dst
, up
->len
);
504 } else /* `normal' UDP */
505 csum
= udp_csum_outgoing(sk
, skb
);
507 /* add protocol-dependent pseudo-header */
508 uh
->check
= csum_tcpudp_magic(fl
->fl4_src
, fl
->fl4_dst
, up
->len
,
509 sk
->sk_protocol
, csum
);
511 uh
->check
= CSUM_MANGLED_0
;
514 err
= ip_push_pending_frames(sk
);
519 UDP_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS
, is_udplite
);
523 int udp_sendmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
526 struct inet_sock
*inet
= inet_sk(sk
);
527 struct udp_sock
*up
= udp_sk(sk
);
529 struct ipcm_cookie ipc
;
530 struct rtable
*rt
= NULL
;
533 __be32 daddr
, faddr
, saddr
;
536 int err
, is_udplite
= IS_UDPLITE(sk
);
537 int corkreq
= up
->corkflag
|| msg
->msg_flags
&MSG_MORE
;
538 int (*getfrag
)(void *, char *, int, int, int, struct sk_buff
*);
547 if (msg
->msg_flags
&MSG_OOB
) /* Mirror BSD error message compatibility */
554 * There are pending frames.
555 * The socket lock must be held while it's corked.
558 if (likely(up
->pending
)) {
559 if (unlikely(up
->pending
!= AF_INET
)) {
567 ulen
+= sizeof(struct udphdr
);
570 * Get and verify the address.
573 struct sockaddr_in
* usin
= (struct sockaddr_in
*)msg
->msg_name
;
574 if (msg
->msg_namelen
< sizeof(*usin
))
576 if (usin
->sin_family
!= AF_INET
) {
577 if (usin
->sin_family
!= AF_UNSPEC
)
578 return -EAFNOSUPPORT
;
581 daddr
= usin
->sin_addr
.s_addr
;
582 dport
= usin
->sin_port
;
586 if (sk
->sk_state
!= TCP_ESTABLISHED
)
587 return -EDESTADDRREQ
;
590 /* Open fast path for connected socket.
591 Route will not be used, if at least one option is set.
595 ipc
.addr
= inet
->saddr
;
597 ipc
.oif
= sk
->sk_bound_dev_if
;
598 if (msg
->msg_controllen
) {
599 err
= ip_cmsg_send(msg
, &ipc
);
610 ipc
.addr
= faddr
= daddr
;
612 if (ipc
.opt
&& ipc
.opt
->srr
) {
615 faddr
= ipc
.opt
->faddr
;
618 tos
= RT_TOS(inet
->tos
);
619 if (sock_flag(sk
, SOCK_LOCALROUTE
) ||
620 (msg
->msg_flags
& MSG_DONTROUTE
) ||
621 (ipc
.opt
&& ipc
.opt
->is_strictroute
)) {
626 if (MULTICAST(daddr
)) {
628 ipc
.oif
= inet
->mc_index
;
630 saddr
= inet
->mc_addr
;
635 rt
= (struct rtable
*)sk_dst_check(sk
, 0);
638 struct flowi fl
= { .oif
= ipc
.oif
,
643 .proto
= sk
->sk_protocol
,
645 { .sport
= inet
->sport
,
646 .dport
= dport
} } };
647 security_sk_classify_flow(sk
, &fl
);
648 err
= ip_route_output_flow(&rt
, &fl
, sk
, 1);
650 if (err
== -ENETUNREACH
)
651 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES
);
656 if ((rt
->rt_flags
& RTCF_BROADCAST
) &&
657 !sock_flag(sk
, SOCK_BROADCAST
))
660 sk_dst_set(sk
, dst_clone(&rt
->u
.dst
));
663 if (msg
->msg_flags
&MSG_CONFIRM
)
669 daddr
= ipc
.addr
= rt
->rt_dst
;
672 if (unlikely(up
->pending
)) {
673 /* The socket is already corked while preparing it. */
674 /* ... which is an evident application bug. --ANK */
677 LIMIT_NETDEBUG(KERN_DEBUG
"udp cork app bug 2\n");
682 * Now cork the socket to pend data.
684 inet
->cork
.fl
.fl4_dst
= daddr
;
685 inet
->cork
.fl
.fl_ip_dport
= dport
;
686 inet
->cork
.fl
.fl4_src
= saddr
;
687 inet
->cork
.fl
.fl_ip_sport
= inet
->sport
;
688 up
->pending
= AF_INET
;
692 getfrag
= is_udplite
? udplite_getfrag
: ip_generic_getfrag
;
693 err
= ip_append_data(sk
, getfrag
, msg
->msg_iov
, ulen
,
694 sizeof(struct udphdr
), &ipc
, rt
,
695 corkreq
? msg
->msg_flags
|MSG_MORE
: msg
->msg_flags
);
697 udp_flush_pending_frames(sk
);
699 err
= udp_push_pending_frames(sk
);
700 else if (unlikely(skb_queue_empty(&sk
->sk_write_queue
)))
711 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
712 * ENOBUFS might not be good (it's not tunable per se), but otherwise
713 * we don't have a good statistic (IpOutDiscards but it can be too many
714 * things). We could add another new stat but at least for now that
715 * seems like overkill.
717 if (err
== -ENOBUFS
|| test_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
)) {
718 UDP_INC_STATS_USER(UDP_MIB_SNDBUFERRORS
, is_udplite
);
723 dst_confirm(&rt
->u
.dst
);
724 if (!(msg
->msg_flags
&MSG_PROBE
) || len
)
725 goto back_from_confirm
;
730 int udp_sendpage(struct sock
*sk
, struct page
*page
, int offset
,
731 size_t size
, int flags
)
733 struct udp_sock
*up
= udp_sk(sk
);
737 struct msghdr msg
= { .msg_flags
= flags
|MSG_MORE
};
739 /* Call udp_sendmsg to specify destination address which
740 * sendpage interface can't pass.
741 * This will succeed only when the socket is connected.
743 ret
= udp_sendmsg(NULL
, sk
, &msg
, 0);
750 if (unlikely(!up
->pending
)) {
753 LIMIT_NETDEBUG(KERN_DEBUG
"udp cork app bug 3\n");
757 ret
= ip_append_page(sk
, page
, offset
, size
, flags
);
758 if (ret
== -EOPNOTSUPP
) {
760 return sock_no_sendpage(sk
->sk_socket
, page
, offset
,
764 udp_flush_pending_frames(sk
);
769 if (!(up
->corkflag
|| (flags
&MSG_MORE
)))
770 ret
= udp_push_pending_frames(sk
);
779 * IOCTL requests applicable to the UDP protocol
782 int udp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
)
787 int amount
= atomic_read(&sk
->sk_wmem_alloc
);
788 return put_user(amount
, (int __user
*)arg
);
794 unsigned long amount
;
797 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
798 skb
= skb_peek(&sk
->sk_receive_queue
);
801 * We will only return the amount
802 * of this packet since that is all
805 amount
= skb
->len
- sizeof(struct udphdr
);
807 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
808 return put_user(amount
, (int __user
*)arg
);
819 * This should be easy, if there is something there we
820 * return it, otherwise we block.
823 int udp_recvmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
824 size_t len
, int noblock
, int flags
, int *addr_len
)
826 struct inet_sock
*inet
= inet_sk(sk
);
827 struct sockaddr_in
*sin
= (struct sockaddr_in
*)msg
->msg_name
;
829 unsigned int ulen
, copied
;
831 int is_udplite
= IS_UDPLITE(sk
);
834 * Check any passed addresses
837 *addr_len
=sizeof(*sin
);
839 if (flags
& MSG_ERRQUEUE
)
840 return ip_recv_error(sk
, msg
, len
);
843 skb
= skb_recv_datagram(sk
, flags
, noblock
, &err
);
847 ulen
= skb
->len
- sizeof(struct udphdr
);
851 else if (copied
< ulen
)
852 msg
->msg_flags
|= MSG_TRUNC
;
855 * If checksum is needed at all, try to do it while copying the
856 * data. If the data is truncated, or if we only want a partial
857 * coverage checksum (UDP-Lite), do it before the copy.
860 if (copied
< ulen
|| UDP_SKB_CB(skb
)->partial_cov
) {
861 if (udp_lib_checksum_complete(skb
))
865 if (skb_csum_unnecessary(skb
))
866 err
= skb_copy_datagram_iovec(skb
, sizeof(struct udphdr
),
867 msg
->msg_iov
, copied
);
869 err
= skb_copy_and_csum_datagram_iovec(skb
, sizeof(struct udphdr
), msg
->msg_iov
);
878 UDP_INC_STATS_USER(UDP_MIB_INDATAGRAMS
, is_udplite
);
880 sock_recv_timestamp(msg
, sk
, skb
);
882 /* Copy the address. */
885 sin
->sin_family
= AF_INET
;
886 sin
->sin_port
= udp_hdr(skb
)->source
;
887 sin
->sin_addr
.s_addr
= ip_hdr(skb
)->saddr
;
888 memset(sin
->sin_zero
, 0, sizeof(sin
->sin_zero
));
890 if (inet
->cmsg_flags
)
891 ip_cmsg_recv(msg
, skb
);
894 if (flags
& MSG_TRUNC
)
898 skb_free_datagram(sk
, skb
);
903 if (!skb_kill_datagram(sk
, skb
, flags
))
904 UDP_INC_STATS_USER(UDP_MIB_INERRORS
, is_udplite
);
912 int udp_disconnect(struct sock
*sk
, int flags
)
914 struct inet_sock
*inet
= inet_sk(sk
);
916 * 1003.1g - break association.
919 sk
->sk_state
= TCP_CLOSE
;
922 sk
->sk_bound_dev_if
= 0;
923 if (!(sk
->sk_userlocks
& SOCK_BINDADDR_LOCK
))
924 inet_reset_saddr(sk
);
926 if (!(sk
->sk_userlocks
& SOCK_BINDPORT_LOCK
)) {
927 sk
->sk_prot
->unhash(sk
);
937 * >0: "udp encap" protocol resubmission
939 * Note that in the success and error cases, the skb is assumed to
940 * have either been requeued or freed.
942 int udp_queue_rcv_skb(struct sock
* sk
, struct sk_buff
*skb
)
944 struct udp_sock
*up
= udp_sk(sk
);
946 int is_udplite
= IS_UDPLITE(sk
);
949 * Charge it to the socket, dropping if the queue is full.
951 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
955 if (up
->encap_type
) {
957 * This is an encapsulation socket so pass the skb to
958 * the socket's udp_encap_rcv() hook. Otherwise, just
959 * fall through and pass this up the UDP socket.
960 * up->encap_rcv() returns the following value:
961 * =0 if skb was successfully passed to the encap
962 * handler or was discarded by it.
963 * >0 if skb should be passed on to UDP.
964 * <0 if skb should be resubmitted as proto -N
967 /* if we're overly short, let UDP handle it */
968 if (skb
->len
> sizeof(struct udphdr
) &&
969 up
->encap_rcv
!= NULL
) {
972 ret
= (*up
->encap_rcv
)(sk
, skb
);
974 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS
,
980 /* FALLTHROUGH -- it's a UDP Packet */
984 * UDP-Lite specific tests, ignored on UDP sockets
986 if ((is_udplite
& UDPLITE_RECV_CC
) && UDP_SKB_CB(skb
)->partial_cov
) {
989 * MIB statistics other than incrementing the error count are
990 * disabled for the following two types of errors: these depend
991 * on the application settings, not on the functioning of the
992 * protocol stack as such.
994 * RFC 3828 here recommends (sec 3.3): "There should also be a
995 * way ... to ... at least let the receiving application block
996 * delivery of packets with coverage values less than a value
997 * provided by the application."
999 if (up
->pcrlen
== 0) { /* full coverage was set */
1000 LIMIT_NETDEBUG(KERN_WARNING
"UDPLITE: partial coverage "
1001 "%d while full coverage %d requested\n",
1002 UDP_SKB_CB(skb
)->cscov
, skb
->len
);
1005 /* The next case involves violating the min. coverage requested
1006 * by the receiver. This is subtle: if receiver wants x and x is
1007 * greater than the buffersize/MTU then receiver will complain
1008 * that it wants x while sender emits packets of smaller size y.
1009 * Therefore the above ...()->partial_cov statement is essential.
1011 if (UDP_SKB_CB(skb
)->cscov
< up
->pcrlen
) {
1012 LIMIT_NETDEBUG(KERN_WARNING
1013 "UDPLITE: coverage %d too small, need min %d\n",
1014 UDP_SKB_CB(skb
)->cscov
, up
->pcrlen
);
1019 if (sk
->sk_filter
) {
1020 if (udp_lib_checksum_complete(skb
))
1024 if ((rc
= sock_queue_rcv_skb(sk
,skb
)) < 0) {
1025 /* Note that an ENOMEM error is charged twice */
1027 UDP_INC_STATS_BH(UDP_MIB_RCVBUFERRORS
, is_udplite
);
1034 UDP_INC_STATS_BH(UDP_MIB_INERRORS
, is_udplite
);
1040 * Multicasts and broadcasts go to each listener.
1042 * Note: called only from the BH handler context,
1043 * so we don't need to lock the hashes.
1045 static int __udp4_lib_mcast_deliver(struct sk_buff
*skb
,
1047 __be32 saddr
, __be32 daddr
,
1048 struct hlist_head udptable
[])
1053 read_lock(&udp_hash_lock
);
1054 sk
= sk_head(&udptable
[ntohs(uh
->dest
) & (UDP_HTABLE_SIZE
- 1)]);
1055 dif
= skb
->dev
->ifindex
;
1056 sk
= udp_v4_mcast_next(sk
, uh
->dest
, daddr
, uh
->source
, saddr
, dif
);
1058 struct sock
*sknext
= NULL
;
1061 struct sk_buff
*skb1
= skb
;
1063 sknext
= udp_v4_mcast_next(sk_next(sk
), uh
->dest
, daddr
,
1064 uh
->source
, saddr
, dif
);
1066 skb1
= skb_clone(skb
, GFP_ATOMIC
);
1069 int ret
= udp_queue_rcv_skb(sk
, skb1
);
1071 /* we should probably re-process instead
1072 * of dropping packets here. */
1079 read_unlock(&udp_hash_lock
);
1083 /* Initialize UDP checksum. If exited with zero value (success),
1084 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1085 * Otherwise, csum completion requires chacksumming packet body,
1086 * including udp header and folding it to skb->csum.
1088 static inline int udp4_csum_init(struct sk_buff
*skb
, struct udphdr
*uh
,
1091 const struct iphdr
*iph
;
1094 UDP_SKB_CB(skb
)->partial_cov
= 0;
1095 UDP_SKB_CB(skb
)->cscov
= skb
->len
;
1097 if (proto
== IPPROTO_UDPLITE
) {
1098 err
= udplite_checksum_init(skb
, uh
);
1104 if (uh
->check
== 0) {
1105 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1106 } else if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
1107 if (!csum_tcpudp_magic(iph
->saddr
, iph
->daddr
, skb
->len
,
1109 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1111 if (!skb_csum_unnecessary(skb
))
1112 skb
->csum
= csum_tcpudp_nofold(iph
->saddr
, iph
->daddr
,
1113 skb
->len
, proto
, 0);
1114 /* Probably, we should checksum udp header (it should be in cache
1115 * in any case) and data in tiny packets (< rx copybreak).
1122 * All we need to do is get the socket, and then do a checksum.
1125 int __udp4_lib_rcv(struct sk_buff
*skb
, struct hlist_head udptable
[],
1129 struct udphdr
*uh
= udp_hdr(skb
);
1130 unsigned short ulen
;
1131 struct rtable
*rt
= (struct rtable
*)skb
->dst
;
1132 __be32 saddr
= ip_hdr(skb
)->saddr
;
1133 __be32 daddr
= ip_hdr(skb
)->daddr
;
1136 * Validate the packet.
1138 if (!pskb_may_pull(skb
, sizeof(struct udphdr
)))
1139 goto drop
; /* No space for header. */
1141 ulen
= ntohs(uh
->len
);
1142 if (ulen
> skb
->len
)
1145 if (proto
== IPPROTO_UDP
) {
1146 /* UDP validates ulen. */
1147 if (ulen
< sizeof(*uh
) || pskb_trim_rcsum(skb
, ulen
))
1152 if (udp4_csum_init(skb
, uh
, proto
))
1155 if (rt
->rt_flags
& (RTCF_BROADCAST
|RTCF_MULTICAST
))
1156 return __udp4_lib_mcast_deliver(skb
, uh
, saddr
, daddr
, udptable
);
1158 sk
= __udp4_lib_lookup(saddr
, uh
->source
, daddr
, uh
->dest
,
1159 inet_iif(skb
), udptable
);
1162 int ret
= udp_queue_rcv_skb(sk
, skb
);
1165 /* a return value > 0 means to resubmit the input, but
1166 * it wants the return to be -protocol, or 0
1173 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
1177 /* No socket. Drop packet silently, if checksum is wrong */
1178 if (udp_lib_checksum_complete(skb
))
1181 UDP_INC_STATS_BH(UDP_MIB_NOPORTS
, proto
== IPPROTO_UDPLITE
);
1182 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_PORT_UNREACH
, 0);
1185 * Hmm. We got an UDP packet to a port to which we
1186 * don't wanna listen. Ignore it.
1192 LIMIT_NETDEBUG(KERN_DEBUG
"UDP%s: short packet: From %u.%u.%u.%u:%u %d/%d to %u.%u.%u.%u:%u\n",
1193 proto
== IPPROTO_UDPLITE
? "-Lite" : "",
1204 * RFC1122: OK. Discards the bad packet silently (as far as
1205 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1207 LIMIT_NETDEBUG(KERN_DEBUG
"UDP%s: bad checksum. From %d.%d.%d.%d:%d to %d.%d.%d.%d:%d ulen %d\n",
1208 proto
== IPPROTO_UDPLITE
? "-Lite" : "",
1215 UDP_INC_STATS_BH(UDP_MIB_INERRORS
, proto
== IPPROTO_UDPLITE
);
1220 int udp_rcv(struct sk_buff
*skb
)
1222 return __udp4_lib_rcv(skb
, udp_hash
, IPPROTO_UDP
);
1225 int udp_destroy_sock(struct sock
*sk
)
1228 udp_flush_pending_frames(sk
);
1234 * Socket option code for UDP
1236 int udp_lib_setsockopt(struct sock
*sk
, int level
, int optname
,
1237 char __user
*optval
, int optlen
,
1238 int (*push_pending_frames
)(struct sock
*))
1240 struct udp_sock
*up
= udp_sk(sk
);
1243 int is_udplite
= IS_UDPLITE(sk
);
1245 if (optlen
<sizeof(int))
1248 if (get_user(val
, (int __user
*)optval
))
1258 (*push_pending_frames
)(sk
);
1266 case UDP_ENCAP_ESPINUDP
:
1267 case UDP_ENCAP_ESPINUDP_NON_IKE
:
1268 up
->encap_rcv
= xfrm4_udp_encap_rcv
;
1270 case UDP_ENCAP_L2TPINUDP
:
1271 up
->encap_type
= val
;
1280 * UDP-Lite's partial checksum coverage (RFC 3828).
1282 /* The sender sets actual checksum coverage length via this option.
1283 * The case coverage > packet length is handled by send module. */
1284 case UDPLITE_SEND_CSCOV
:
1285 if (!is_udplite
) /* Disable the option on UDP sockets */
1286 return -ENOPROTOOPT
;
1287 if (val
!= 0 && val
< 8) /* Illegal coverage: use default (8) */
1290 up
->pcflag
|= UDPLITE_SEND_CC
;
1293 /* The receiver specifies a minimum checksum coverage value. To make
1294 * sense, this should be set to at least 8 (as done below). If zero is
1295 * used, this again means full checksum coverage. */
1296 case UDPLITE_RECV_CSCOV
:
1297 if (!is_udplite
) /* Disable the option on UDP sockets */
1298 return -ENOPROTOOPT
;
1299 if (val
!= 0 && val
< 8) /* Avoid silly minimal values. */
1302 up
->pcflag
|= UDPLITE_RECV_CC
;
1313 int udp_setsockopt(struct sock
*sk
, int level
, int optname
,
1314 char __user
*optval
, int optlen
)
1316 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1317 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1318 udp_push_pending_frames
);
1319 return ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
1322 #ifdef CONFIG_COMPAT
1323 int compat_udp_setsockopt(struct sock
*sk
, int level
, int optname
,
1324 char __user
*optval
, int optlen
)
1326 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1327 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1328 udp_push_pending_frames
);
1329 return compat_ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
1333 int udp_lib_getsockopt(struct sock
*sk
, int level
, int optname
,
1334 char __user
*optval
, int __user
*optlen
)
1336 struct udp_sock
*up
= udp_sk(sk
);
1339 if (get_user(len
,optlen
))
1342 len
= min_t(unsigned int, len
, sizeof(int));
1353 val
= up
->encap_type
;
1356 /* The following two cannot be changed on UDP sockets, the return is
1357 * always 0 (which corresponds to the full checksum coverage of UDP). */
1358 case UDPLITE_SEND_CSCOV
:
1362 case UDPLITE_RECV_CSCOV
:
1367 return -ENOPROTOOPT
;
1370 if (put_user(len
, optlen
))
1372 if (copy_to_user(optval
, &val
,len
))
1377 int udp_getsockopt(struct sock
*sk
, int level
, int optname
,
1378 char __user
*optval
, int __user
*optlen
)
1380 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1381 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1382 return ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
1385 #ifdef CONFIG_COMPAT
1386 int compat_udp_getsockopt(struct sock
*sk
, int level
, int optname
,
1387 char __user
*optval
, int __user
*optlen
)
1389 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1390 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1391 return compat_ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
1395 * udp_poll - wait for a UDP event.
1396 * @file - file struct
1398 * @wait - poll table
1400 * This is same as datagram poll, except for the special case of
1401 * blocking sockets. If application is using a blocking fd
1402 * and a packet with checksum error is in the queue;
1403 * then it could get return from select indicating data available
1404 * but then block when reading it. Add special case code
1405 * to work around these arguably broken applications.
1407 unsigned int udp_poll(struct file
*file
, struct socket
*sock
, poll_table
*wait
)
1409 unsigned int mask
= datagram_poll(file
, sock
, wait
);
1410 struct sock
*sk
= sock
->sk
;
1411 int is_lite
= IS_UDPLITE(sk
);
1413 /* Check for false positives due to checksum errors */
1414 if ( (mask
& POLLRDNORM
) &&
1415 !(file
->f_flags
& O_NONBLOCK
) &&
1416 !(sk
->sk_shutdown
& RCV_SHUTDOWN
)){
1417 struct sk_buff_head
*rcvq
= &sk
->sk_receive_queue
;
1418 struct sk_buff
*skb
;
1420 spin_lock_bh(&rcvq
->lock
);
1421 while ((skb
= skb_peek(rcvq
)) != NULL
&&
1422 udp_lib_checksum_complete(skb
)) {
1423 UDP_INC_STATS_BH(UDP_MIB_INERRORS
, is_lite
);
1424 __skb_unlink(skb
, rcvq
);
1427 spin_unlock_bh(&rcvq
->lock
);
1429 /* nothing to see, move along */
1431 mask
&= ~(POLLIN
| POLLRDNORM
);
1438 DEFINE_PROTO_INUSE(udp
)
1440 struct proto udp_prot
= {
1442 .owner
= THIS_MODULE
,
1443 .close
= udp_lib_close
,
1444 .connect
= ip4_datagram_connect
,
1445 .disconnect
= udp_disconnect
,
1447 .destroy
= udp_destroy_sock
,
1448 .setsockopt
= udp_setsockopt
,
1449 .getsockopt
= udp_getsockopt
,
1450 .sendmsg
= udp_sendmsg
,
1451 .recvmsg
= udp_recvmsg
,
1452 .sendpage
= udp_sendpage
,
1453 .backlog_rcv
= udp_queue_rcv_skb
,
1454 .hash
= udp_lib_hash
,
1455 .unhash
= udp_lib_unhash
,
1456 .get_port
= udp_v4_get_port
,
1457 .obj_size
= sizeof(struct udp_sock
),
1458 #ifdef CONFIG_COMPAT
1459 .compat_setsockopt
= compat_udp_setsockopt
,
1460 .compat_getsockopt
= compat_udp_getsockopt
,
1462 REF_PROTO_INUSE(udp
)
1465 /* ------------------------------------------------------------------------ */
1466 #ifdef CONFIG_PROC_FS
1468 static struct sock
*udp_get_first(struct seq_file
*seq
)
1471 struct udp_iter_state
*state
= seq
->private;
1473 for (state
->bucket
= 0; state
->bucket
< UDP_HTABLE_SIZE
; ++state
->bucket
) {
1474 struct hlist_node
*node
;
1475 sk_for_each(sk
, node
, state
->hashtable
+ state
->bucket
) {
1476 if (sk
->sk_family
== state
->family
)
1485 static struct sock
*udp_get_next(struct seq_file
*seq
, struct sock
*sk
)
1487 struct udp_iter_state
*state
= seq
->private;
1493 } while (sk
&& sk
->sk_family
!= state
->family
);
1495 if (!sk
&& ++state
->bucket
< UDP_HTABLE_SIZE
) {
1496 sk
= sk_head(state
->hashtable
+ state
->bucket
);
1502 static struct sock
*udp_get_idx(struct seq_file
*seq
, loff_t pos
)
1504 struct sock
*sk
= udp_get_first(seq
);
1507 while (pos
&& (sk
= udp_get_next(seq
, sk
)) != NULL
)
1509 return pos
? NULL
: sk
;
1512 static void *udp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1514 read_lock(&udp_hash_lock
);
1515 return *pos
? udp_get_idx(seq
, *pos
-1) : (void *)1;
1518 static void *udp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1523 sk
= udp_get_idx(seq
, 0);
1525 sk
= udp_get_next(seq
, v
);
1531 static void udp_seq_stop(struct seq_file
*seq
, void *v
)
1533 read_unlock(&udp_hash_lock
);
1536 static int udp_seq_open(struct inode
*inode
, struct file
*file
)
1538 struct udp_seq_afinfo
*afinfo
= PDE(inode
)->data
;
1539 struct seq_file
*seq
;
1541 struct udp_iter_state
*s
= kzalloc(sizeof(*s
), GFP_KERNEL
);
1545 s
->family
= afinfo
->family
;
1546 s
->hashtable
= afinfo
->hashtable
;
1547 s
->seq_ops
.start
= udp_seq_start
;
1548 s
->seq_ops
.next
= udp_seq_next
;
1549 s
->seq_ops
.show
= afinfo
->seq_show
;
1550 s
->seq_ops
.stop
= udp_seq_stop
;
1552 rc
= seq_open(file
, &s
->seq_ops
);
1556 seq
= file
->private_data
;
1565 /* ------------------------------------------------------------------------ */
1566 int udp_proc_register(struct udp_seq_afinfo
*afinfo
)
1568 struct proc_dir_entry
*p
;
1573 afinfo
->seq_fops
->owner
= afinfo
->owner
;
1574 afinfo
->seq_fops
->open
= udp_seq_open
;
1575 afinfo
->seq_fops
->read
= seq_read
;
1576 afinfo
->seq_fops
->llseek
= seq_lseek
;
1577 afinfo
->seq_fops
->release
= seq_release_private
;
1579 p
= proc_net_fops_create(&init_net
, afinfo
->name
, S_IRUGO
, afinfo
->seq_fops
);
1587 void udp_proc_unregister(struct udp_seq_afinfo
*afinfo
)
1591 proc_net_remove(&init_net
, afinfo
->name
);
1592 memset(afinfo
->seq_fops
, 0, sizeof(*afinfo
->seq_fops
));
1595 /* ------------------------------------------------------------------------ */
1596 static void udp4_format_sock(struct sock
*sp
, char *tmpbuf
, int bucket
)
1598 struct inet_sock
*inet
= inet_sk(sp
);
1599 __be32 dest
= inet
->daddr
;
1600 __be32 src
= inet
->rcv_saddr
;
1601 __u16 destp
= ntohs(inet
->dport
);
1602 __u16 srcp
= ntohs(inet
->sport
);
1604 sprintf(tmpbuf
, "%4d: %08X:%04X %08X:%04X"
1605 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p",
1606 bucket
, src
, srcp
, dest
, destp
, sp
->sk_state
,
1607 atomic_read(&sp
->sk_wmem_alloc
),
1608 atomic_read(&sp
->sk_rmem_alloc
),
1609 0, 0L, 0, sock_i_uid(sp
), 0, sock_i_ino(sp
),
1610 atomic_read(&sp
->sk_refcnt
), sp
);
1613 int udp4_seq_show(struct seq_file
*seq
, void *v
)
1615 if (v
== SEQ_START_TOKEN
)
1616 seq_printf(seq
, "%-127s\n",
1617 " sl local_address rem_address st tx_queue "
1618 "rx_queue tr tm->when retrnsmt uid timeout "
1622 struct udp_iter_state
*state
= seq
->private;
1624 udp4_format_sock(v
, tmpbuf
, state
->bucket
);
1625 seq_printf(seq
, "%-127s\n", tmpbuf
);
1630 /* ------------------------------------------------------------------------ */
1631 static struct file_operations udp4_seq_fops
;
1632 static struct udp_seq_afinfo udp4_seq_afinfo
= {
1633 .owner
= THIS_MODULE
,
1636 .hashtable
= udp_hash
,
1637 .seq_show
= udp4_seq_show
,
1638 .seq_fops
= &udp4_seq_fops
,
1641 int __init
udp4_proc_init(void)
1643 return udp_proc_register(&udp4_seq_afinfo
);
1646 void udp4_proc_exit(void)
1648 udp_proc_unregister(&udp4_seq_afinfo
);
1650 #endif /* CONFIG_PROC_FS */
1652 EXPORT_SYMBOL(udp_disconnect
);
1653 EXPORT_SYMBOL(udp_hash
);
1654 EXPORT_SYMBOL(udp_hash_lock
);
1655 EXPORT_SYMBOL(udp_ioctl
);
1656 EXPORT_SYMBOL(udp_get_port
);
1657 EXPORT_SYMBOL(udp_prot
);
1658 EXPORT_SYMBOL(udp_sendmsg
);
1659 EXPORT_SYMBOL(udp_lib_getsockopt
);
1660 EXPORT_SYMBOL(udp_lib_setsockopt
);
1661 EXPORT_SYMBOL(udp_poll
);
1663 #ifdef CONFIG_PROC_FS
1664 EXPORT_SYMBOL(udp_proc_register
);
1665 EXPORT_SYMBOL(udp_proc_unregister
);