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/icmp.h>
102 #include <net/route.h>
103 #include <net/checksum.h>
104 #include <net/xfrm.h>
105 #include "udp_impl.h"
108 * Snmp MIB for the UDP layer
111 DEFINE_SNMP_STAT(struct udp_mib
, udp_statistics
) __read_mostly
;
113 struct hlist_head udp_hash
[UDP_HTABLE_SIZE
];
114 DEFINE_RWLOCK(udp_hash_lock
);
116 static int udp_port_rover
;
118 static inline int __udp_lib_lport_inuse(__u16 num
, struct hlist_head udptable
[])
121 struct hlist_node
*node
;
123 sk_for_each(sk
, node
, &udptable
[num
& (UDP_HTABLE_SIZE
- 1)])
124 if (sk
->sk_hash
== num
)
130 * __udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
132 * @sk: socket struct in question
133 * @snum: port number to look up
134 * @udptable: hash list table, must be of UDP_HTABLE_SIZE
135 * @port_rover: pointer to record of last unallocated port
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
[], int *port_rover
,
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
);
150 int best_size_so_far
, best
, result
, i
;
152 if (*port_rover
> sysctl_local_port_range
[1] ||
153 *port_rover
< sysctl_local_port_range
[0])
154 *port_rover
= sysctl_local_port_range
[0];
155 best_size_so_far
= 32767;
156 best
= result
= *port_rover
;
157 for (i
= 0; i
< UDP_HTABLE_SIZE
; i
++, result
++) {
160 head
= &udptable
[result
& (UDP_HTABLE_SIZE
- 1)];
161 if (hlist_empty(head
)) {
162 if (result
> sysctl_local_port_range
[1])
163 result
= sysctl_local_port_range
[0] +
164 ((result
- sysctl_local_port_range
[0]) &
165 (UDP_HTABLE_SIZE
- 1));
169 sk_for_each(sk2
, node
, head
) {
170 if (++size
>= best_size_so_far
)
173 best_size_so_far
= size
;
179 for (i
= 0; i
< (1 << 16) / UDP_HTABLE_SIZE
;
180 i
++, result
+= UDP_HTABLE_SIZE
) {
181 if (result
> sysctl_local_port_range
[1])
182 result
= sysctl_local_port_range
[0]
183 + ((result
- sysctl_local_port_range
[0]) &
184 (UDP_HTABLE_SIZE
- 1));
185 if (! __udp_lib_lport_inuse(result
, udptable
))
188 if (i
>= (1 << 16) / UDP_HTABLE_SIZE
)
191 *port_rover
= snum
= result
;
193 head
= &udptable
[snum
& (UDP_HTABLE_SIZE
- 1)];
195 sk_for_each(sk2
, node
, head
)
196 if (sk2
->sk_hash
== snum
&&
198 (!sk2
->sk_reuse
|| !sk
->sk_reuse
) &&
199 (!sk2
->sk_bound_dev_if
|| !sk
->sk_bound_dev_if
200 || sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
201 (*saddr_comp
)(sk
, sk2
) )
204 inet_sk(sk
)->num
= snum
;
206 if (sk_unhashed(sk
)) {
207 head
= &udptable
[snum
& (UDP_HTABLE_SIZE
- 1)];
208 sk_add_node(sk
, head
);
209 sock_prot_inc_use(sk
->sk_prot
);
213 write_unlock_bh(&udp_hash_lock
);
217 int udp_get_port(struct sock
*sk
, unsigned short snum
,
218 int (*scmp
)(const struct sock
*, const struct sock
*))
220 return __udp_lib_get_port(sk
, snum
, udp_hash
, &udp_port_rover
, scmp
);
223 int ipv4_rcv_saddr_equal(const struct sock
*sk1
, const struct sock
*sk2
)
225 struct inet_sock
*inet1
= inet_sk(sk1
), *inet2
= inet_sk(sk2
);
227 return ( !ipv6_only_sock(sk2
) &&
228 (!inet1
->rcv_saddr
|| !inet2
->rcv_saddr
||
229 inet1
->rcv_saddr
== inet2
->rcv_saddr
));
232 static inline int udp_v4_get_port(struct sock
*sk
, unsigned short snum
)
234 return udp_get_port(sk
, snum
, ipv4_rcv_saddr_equal
);
237 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
238 * harder than this. -DaveM
240 static struct sock
*__udp4_lib_lookup(__be32 saddr
, __be16 sport
,
241 __be32 daddr
, __be16 dport
,
242 int dif
, struct hlist_head udptable
[])
244 struct sock
*sk
, *result
= NULL
;
245 struct hlist_node
*node
;
246 unsigned short hnum
= ntohs(dport
);
249 read_lock(&udp_hash_lock
);
250 sk_for_each(sk
, node
, &udptable
[hnum
& (UDP_HTABLE_SIZE
- 1)]) {
251 struct inet_sock
*inet
= inet_sk(sk
);
253 if (sk
->sk_hash
== hnum
&& !ipv6_only_sock(sk
)) {
254 int score
= (sk
->sk_family
== PF_INET
? 1 : 0);
255 if (inet
->rcv_saddr
) {
256 if (inet
->rcv_saddr
!= daddr
)
261 if (inet
->daddr
!= saddr
)
266 if (inet
->dport
!= sport
)
270 if (sk
->sk_bound_dev_if
) {
271 if (sk
->sk_bound_dev_if
!= dif
)
278 } else if (score
> badness
) {
286 read_unlock(&udp_hash_lock
);
290 static inline struct sock
*udp_v4_mcast_next(struct sock
*sk
,
291 __be16 loc_port
, __be32 loc_addr
,
292 __be16 rmt_port
, __be32 rmt_addr
,
295 struct hlist_node
*node
;
297 unsigned short hnum
= ntohs(loc_port
);
299 sk_for_each_from(s
, node
) {
300 struct inet_sock
*inet
= inet_sk(s
);
302 if (s
->sk_hash
!= hnum
||
303 (inet
->daddr
&& inet
->daddr
!= rmt_addr
) ||
304 (inet
->dport
!= rmt_port
&& inet
->dport
) ||
305 (inet
->rcv_saddr
&& inet
->rcv_saddr
!= loc_addr
) ||
307 (s
->sk_bound_dev_if
&& s
->sk_bound_dev_if
!= dif
))
309 if (!ip_mc_sf_allow(s
, loc_addr
, rmt_addr
, dif
))
319 * This routine is called by the ICMP module when it gets some
320 * sort of error condition. If err < 0 then the socket should
321 * be closed and the error returned to the user. If err > 0
322 * it's just the icmp type << 8 | icmp code.
323 * Header points to the ip header of the error packet. We move
324 * on past this. Then (as it used to claim before adjustment)
325 * header points to the first 8 bytes of the udp header. We need
326 * to find the appropriate port.
329 void __udp4_lib_err(struct sk_buff
*skb
, u32 info
, struct hlist_head udptable
[])
331 struct inet_sock
*inet
;
332 struct iphdr
*iph
= (struct iphdr
*)skb
->data
;
333 struct udphdr
*uh
= (struct udphdr
*)(skb
->data
+(iph
->ihl
<<2));
334 const int type
= icmp_hdr(skb
)->type
;
335 const int code
= icmp_hdr(skb
)->code
;
340 sk
= __udp4_lib_lookup(iph
->daddr
, uh
->dest
, iph
->saddr
, uh
->source
,
341 skb
->dev
->ifindex
, udptable
);
343 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS
);
344 return; /* No socket for error */
353 case ICMP_TIME_EXCEEDED
:
356 case ICMP_SOURCE_QUENCH
:
358 case ICMP_PARAMETERPROB
:
362 case ICMP_DEST_UNREACH
:
363 if (code
== ICMP_FRAG_NEEDED
) { /* Path MTU discovery */
364 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
) {
372 if (code
<= NR_ICMP_UNREACH
) {
373 harderr
= icmp_err_convert
[code
].fatal
;
374 err
= icmp_err_convert
[code
].errno
;
380 * RFC1122: OK. Passes ICMP errors back to application, as per
383 if (!inet
->recverr
) {
384 if (!harderr
|| sk
->sk_state
!= TCP_ESTABLISHED
)
387 ip_icmp_error(sk
, skb
, err
, uh
->dest
, info
, (u8
*)(uh
+1));
390 sk
->sk_error_report(sk
);
395 void udp_err(struct sk_buff
*skb
, u32 info
)
397 return __udp4_lib_err(skb
, info
, udp_hash
);
401 * Throw away all pending data and cancel the corking. Socket is locked.
403 static void udp_flush_pending_frames(struct sock
*sk
)
405 struct udp_sock
*up
= udp_sk(sk
);
410 ip_flush_pending_frames(sk
);
415 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
416 * @sk: socket we are sending on
417 * @skb: sk_buff containing the filled-in UDP header
418 * (checksum field must be zeroed out)
420 static void udp4_hwcsum_outgoing(struct sock
*sk
, struct sk_buff
*skb
,
421 __be32 src
, __be32 dst
, int len
)
424 struct udphdr
*uh
= udp_hdr(skb
);
427 if (skb_queue_len(&sk
->sk_write_queue
) == 1) {
429 * Only one fragment on the socket.
431 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
432 skb
->csum_offset
= offsetof(struct udphdr
, check
);
433 uh
->check
= ~csum_tcpudp_magic(src
, dst
, len
, IPPROTO_UDP
, 0);
436 * HW-checksum won't work as there are two or more
437 * fragments on the socket so that all csums of sk_buffs
440 offset
= skb_transport_offset(skb
);
441 skb
->csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
443 skb
->ip_summed
= CHECKSUM_NONE
;
445 skb_queue_walk(&sk
->sk_write_queue
, skb
) {
446 csum
= csum_add(csum
, skb
->csum
);
449 uh
->check
= csum_tcpudp_magic(src
, dst
, len
, IPPROTO_UDP
, csum
);
451 uh
->check
= CSUM_MANGLED_0
;
456 * Push out all pending data as one UDP datagram. Socket is locked.
458 static int udp_push_pending_frames(struct sock
*sk
)
460 struct udp_sock
*up
= udp_sk(sk
);
461 struct inet_sock
*inet
= inet_sk(sk
);
462 struct flowi
*fl
= &inet
->cork
.fl
;
468 /* Grab the skbuff where UDP header space exists. */
469 if ((skb
= skb_peek(&sk
->sk_write_queue
)) == NULL
)
473 * Create a UDP header
476 uh
->source
= fl
->fl_ip_sport
;
477 uh
->dest
= fl
->fl_ip_dport
;
478 uh
->len
= htons(up
->len
);
481 if (up
->pcflag
) /* UDP-Lite */
482 csum
= udplite_csum_outgoing(sk
, skb
);
484 else if (sk
->sk_no_check
== UDP_CSUM_NOXMIT
) { /* UDP csum disabled */
486 skb
->ip_summed
= CHECKSUM_NONE
;
489 } else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) { /* UDP hardware csum */
491 udp4_hwcsum_outgoing(sk
, skb
, fl
->fl4_src
,fl
->fl4_dst
, up
->len
);
494 } else /* `normal' UDP */
495 csum
= udp_csum_outgoing(sk
, skb
);
497 /* add protocol-dependent pseudo-header */
498 uh
->check
= csum_tcpudp_magic(fl
->fl4_src
, fl
->fl4_dst
, up
->len
,
499 sk
->sk_protocol
, csum
);
501 uh
->check
= CSUM_MANGLED_0
;
504 err
= ip_push_pending_frames(sk
);
511 int udp_sendmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
514 struct inet_sock
*inet
= inet_sk(sk
);
515 struct udp_sock
*up
= udp_sk(sk
);
517 struct ipcm_cookie ipc
;
518 struct rtable
*rt
= NULL
;
521 __be32 daddr
, faddr
, saddr
;
524 int err
, is_udplite
= up
->pcflag
;
525 int corkreq
= up
->corkflag
|| msg
->msg_flags
&MSG_MORE
;
526 int (*getfrag
)(void *, char *, int, int, int, struct sk_buff
*);
535 if (msg
->msg_flags
&MSG_OOB
) /* Mirror BSD error message compatibility */
542 * There are pending frames.
543 * The socket lock must be held while it's corked.
546 if (likely(up
->pending
)) {
547 if (unlikely(up
->pending
!= AF_INET
)) {
555 ulen
+= sizeof(struct udphdr
);
558 * Get and verify the address.
561 struct sockaddr_in
* usin
= (struct sockaddr_in
*)msg
->msg_name
;
562 if (msg
->msg_namelen
< sizeof(*usin
))
564 if (usin
->sin_family
!= AF_INET
) {
565 if (usin
->sin_family
!= AF_UNSPEC
)
566 return -EAFNOSUPPORT
;
569 daddr
= usin
->sin_addr
.s_addr
;
570 dport
= usin
->sin_port
;
574 if (sk
->sk_state
!= TCP_ESTABLISHED
)
575 return -EDESTADDRREQ
;
578 /* Open fast path for connected socket.
579 Route will not be used, if at least one option is set.
583 ipc
.addr
= inet
->saddr
;
585 ipc
.oif
= sk
->sk_bound_dev_if
;
586 if (msg
->msg_controllen
) {
587 err
= ip_cmsg_send(msg
, &ipc
);
598 ipc
.addr
= faddr
= daddr
;
600 if (ipc
.opt
&& ipc
.opt
->srr
) {
603 faddr
= ipc
.opt
->faddr
;
606 tos
= RT_TOS(inet
->tos
);
607 if (sock_flag(sk
, SOCK_LOCALROUTE
) ||
608 (msg
->msg_flags
& MSG_DONTROUTE
) ||
609 (ipc
.opt
&& ipc
.opt
->is_strictroute
)) {
614 if (MULTICAST(daddr
)) {
616 ipc
.oif
= inet
->mc_index
;
618 saddr
= inet
->mc_addr
;
623 rt
= (struct rtable
*)sk_dst_check(sk
, 0);
626 struct flowi fl
= { .oif
= ipc
.oif
,
631 .proto
= sk
->sk_protocol
,
633 { .sport
= inet
->sport
,
634 .dport
= dport
} } };
635 security_sk_classify_flow(sk
, &fl
);
636 err
= ip_route_output_flow(&rt
, &fl
, sk
, 1);
638 if (err
== -ENETUNREACH
)
639 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES
);
644 if ((rt
->rt_flags
& RTCF_BROADCAST
) &&
645 !sock_flag(sk
, SOCK_BROADCAST
))
648 sk_dst_set(sk
, dst_clone(&rt
->u
.dst
));
651 if (msg
->msg_flags
&MSG_CONFIRM
)
657 daddr
= ipc
.addr
= rt
->rt_dst
;
660 if (unlikely(up
->pending
)) {
661 /* The socket is already corked while preparing it. */
662 /* ... which is an evident application bug. --ANK */
665 LIMIT_NETDEBUG(KERN_DEBUG
"udp cork app bug 2\n");
670 * Now cork the socket to pend data.
672 inet
->cork
.fl
.fl4_dst
= daddr
;
673 inet
->cork
.fl
.fl_ip_dport
= dport
;
674 inet
->cork
.fl
.fl4_src
= saddr
;
675 inet
->cork
.fl
.fl_ip_sport
= inet
->sport
;
676 up
->pending
= AF_INET
;
680 getfrag
= is_udplite
? udplite_getfrag
: ip_generic_getfrag
;
681 err
= ip_append_data(sk
, getfrag
, msg
->msg_iov
, ulen
,
682 sizeof(struct udphdr
), &ipc
, rt
,
683 corkreq
? msg
->msg_flags
|MSG_MORE
: msg
->msg_flags
);
685 udp_flush_pending_frames(sk
);
687 err
= udp_push_pending_frames(sk
);
688 else if (unlikely(skb_queue_empty(&sk
->sk_write_queue
)))
697 UDP_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS
, is_udplite
);
701 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
702 * ENOBUFS might not be good (it's not tunable per se), but otherwise
703 * we don't have a good statistic (IpOutDiscards but it can be too many
704 * things). We could add another new stat but at least for now that
705 * seems like overkill.
707 if (err
== -ENOBUFS
|| test_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
)) {
708 UDP_INC_STATS_USER(UDP_MIB_SNDBUFERRORS
, is_udplite
);
713 dst_confirm(&rt
->u
.dst
);
714 if (!(msg
->msg_flags
&MSG_PROBE
) || len
)
715 goto back_from_confirm
;
720 int udp_sendpage(struct sock
*sk
, struct page
*page
, int offset
,
721 size_t size
, int flags
)
723 struct udp_sock
*up
= udp_sk(sk
);
727 struct msghdr msg
= { .msg_flags
= flags
|MSG_MORE
};
729 /* Call udp_sendmsg to specify destination address which
730 * sendpage interface can't pass.
731 * This will succeed only when the socket is connected.
733 ret
= udp_sendmsg(NULL
, sk
, &msg
, 0);
740 if (unlikely(!up
->pending
)) {
743 LIMIT_NETDEBUG(KERN_DEBUG
"udp cork app bug 3\n");
747 ret
= ip_append_page(sk
, page
, offset
, size
, flags
);
748 if (ret
== -EOPNOTSUPP
) {
750 return sock_no_sendpage(sk
->sk_socket
, page
, offset
,
754 udp_flush_pending_frames(sk
);
759 if (!(up
->corkflag
|| (flags
&MSG_MORE
)))
760 ret
= udp_push_pending_frames(sk
);
769 * IOCTL requests applicable to the UDP protocol
772 int udp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
)
777 int amount
= atomic_read(&sk
->sk_wmem_alloc
);
778 return put_user(amount
, (int __user
*)arg
);
784 unsigned long amount
;
787 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
788 skb
= skb_peek(&sk
->sk_receive_queue
);
791 * We will only return the amount
792 * of this packet since that is all
795 amount
= skb
->len
- sizeof(struct udphdr
);
797 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
798 return put_user(amount
, (int __user
*)arg
);
809 * This should be easy, if there is something there we
810 * return it, otherwise we block.
813 int udp_recvmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
814 size_t len
, int noblock
, int flags
, int *addr_len
)
816 struct inet_sock
*inet
= inet_sk(sk
);
817 struct sockaddr_in
*sin
= (struct sockaddr_in
*)msg
->msg_name
;
819 unsigned int ulen
, copied
;
821 int is_udplite
= IS_UDPLITE(sk
);
824 * Check any passed addresses
827 *addr_len
=sizeof(*sin
);
829 if (flags
& MSG_ERRQUEUE
)
830 return ip_recv_error(sk
, msg
, len
);
833 skb
= skb_recv_datagram(sk
, flags
, noblock
, &err
);
837 ulen
= skb
->len
- sizeof(struct udphdr
);
841 else if (copied
< ulen
)
842 msg
->msg_flags
|= MSG_TRUNC
;
845 * If checksum is needed at all, try to do it while copying the
846 * data. If the data is truncated, or if we only want a partial
847 * coverage checksum (UDP-Lite), do it before the copy.
850 if (copied
< ulen
|| UDP_SKB_CB(skb
)->partial_cov
) {
851 if (udp_lib_checksum_complete(skb
))
855 if (skb_csum_unnecessary(skb
))
856 err
= skb_copy_datagram_iovec(skb
, sizeof(struct udphdr
),
857 msg
->msg_iov
, copied
);
859 err
= skb_copy_and_csum_datagram_iovec(skb
, sizeof(struct udphdr
), msg
->msg_iov
);
868 sock_recv_timestamp(msg
, sk
, skb
);
870 /* Copy the address. */
873 sin
->sin_family
= AF_INET
;
874 sin
->sin_port
= udp_hdr(skb
)->source
;
875 sin
->sin_addr
.s_addr
= ip_hdr(skb
)->saddr
;
876 memset(sin
->sin_zero
, 0, sizeof(sin
->sin_zero
));
878 if (inet
->cmsg_flags
)
879 ip_cmsg_recv(msg
, skb
);
882 if (flags
& MSG_TRUNC
)
886 skb_free_datagram(sk
, skb
);
891 UDP_INC_STATS_BH(UDP_MIB_INERRORS
, is_udplite
);
893 skb_kill_datagram(sk
, skb
, flags
);
901 int udp_disconnect(struct sock
*sk
, int flags
)
903 struct inet_sock
*inet
= inet_sk(sk
);
905 * 1003.1g - break association.
908 sk
->sk_state
= TCP_CLOSE
;
911 sk
->sk_bound_dev_if
= 0;
912 if (!(sk
->sk_userlocks
& SOCK_BINDADDR_LOCK
))
913 inet_reset_saddr(sk
);
915 if (!(sk
->sk_userlocks
& SOCK_BINDPORT_LOCK
)) {
916 sk
->sk_prot
->unhash(sk
);
926 * >0: "udp encap" protocol resubmission
928 * Note that in the success and error cases, the skb is assumed to
929 * have either been requeued or freed.
931 int udp_queue_rcv_skb(struct sock
* sk
, struct sk_buff
*skb
)
933 struct udp_sock
*up
= udp_sk(sk
);
937 * Charge it to the socket, dropping if the queue is full.
939 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
943 if (up
->encap_type
) {
945 * This is an encapsulation socket so pass the skb to
946 * the socket's udp_encap_rcv() hook. Otherwise, just
947 * fall through and pass this up the UDP socket.
948 * up->encap_rcv() returns the following value:
949 * =0 if skb was successfully passed to the encap
950 * handler or was discarded by it.
951 * >0 if skb should be passed on to UDP.
952 * <0 if skb should be resubmitted as proto -N
955 /* if we're overly short, let UDP handle it */
956 if (skb
->len
> sizeof(struct udphdr
) &&
957 up
->encap_rcv
!= NULL
) {
960 ret
= (*up
->encap_rcv
)(sk
, skb
);
962 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS
, up
->pcflag
);
967 /* FALLTHROUGH -- it's a UDP Packet */
971 * UDP-Lite specific tests, ignored on UDP sockets
973 if ((up
->pcflag
& UDPLITE_RECV_CC
) && UDP_SKB_CB(skb
)->partial_cov
) {
976 * MIB statistics other than incrementing the error count are
977 * disabled for the following two types of errors: these depend
978 * on the application settings, not on the functioning of the
979 * protocol stack as such.
981 * RFC 3828 here recommends (sec 3.3): "There should also be a
982 * way ... to ... at least let the receiving application block
983 * delivery of packets with coverage values less than a value
984 * provided by the application."
986 if (up
->pcrlen
== 0) { /* full coverage was set */
987 LIMIT_NETDEBUG(KERN_WARNING
"UDPLITE: partial coverage "
988 "%d while full coverage %d requested\n",
989 UDP_SKB_CB(skb
)->cscov
, skb
->len
);
992 /* The next case involves violating the min. coverage requested
993 * by the receiver. This is subtle: if receiver wants x and x is
994 * greater than the buffersize/MTU then receiver will complain
995 * that it wants x while sender emits packets of smaller size y.
996 * Therefore the above ...()->partial_cov statement is essential.
998 if (UDP_SKB_CB(skb
)->cscov
< up
->pcrlen
) {
999 LIMIT_NETDEBUG(KERN_WARNING
1000 "UDPLITE: coverage %d too small, need min %d\n",
1001 UDP_SKB_CB(skb
)->cscov
, up
->pcrlen
);
1006 if (sk
->sk_filter
) {
1007 if (udp_lib_checksum_complete(skb
))
1011 if ((rc
= sock_queue_rcv_skb(sk
,skb
)) < 0) {
1012 /* Note that an ENOMEM error is charged twice */
1014 UDP_INC_STATS_BH(UDP_MIB_RCVBUFERRORS
, up
->pcflag
);
1018 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS
, up
->pcflag
);
1022 UDP_INC_STATS_BH(UDP_MIB_INERRORS
, up
->pcflag
);
1028 * Multicasts and broadcasts go to each listener.
1030 * Note: called only from the BH handler context,
1031 * so we don't need to lock the hashes.
1033 static int __udp4_lib_mcast_deliver(struct sk_buff
*skb
,
1035 __be32 saddr
, __be32 daddr
,
1036 struct hlist_head udptable
[])
1041 read_lock(&udp_hash_lock
);
1042 sk
= sk_head(&udptable
[ntohs(uh
->dest
) & (UDP_HTABLE_SIZE
- 1)]);
1043 dif
= skb
->dev
->ifindex
;
1044 sk
= udp_v4_mcast_next(sk
, uh
->dest
, daddr
, uh
->source
, saddr
, dif
);
1046 struct sock
*sknext
= NULL
;
1049 struct sk_buff
*skb1
= skb
;
1051 sknext
= udp_v4_mcast_next(sk_next(sk
), uh
->dest
, daddr
,
1052 uh
->source
, saddr
, dif
);
1054 skb1
= skb_clone(skb
, GFP_ATOMIC
);
1057 int ret
= udp_queue_rcv_skb(sk
, skb1
);
1059 /* we should probably re-process instead
1060 * of dropping packets here. */
1067 read_unlock(&udp_hash_lock
);
1071 /* Initialize UDP checksum. If exited with zero value (success),
1072 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1073 * Otherwise, csum completion requires chacksumming packet body,
1074 * including udp header and folding it to skb->csum.
1076 static inline int udp4_csum_init(struct sk_buff
*skb
, struct udphdr
*uh
,
1079 const struct iphdr
*iph
;
1082 UDP_SKB_CB(skb
)->partial_cov
= 0;
1083 UDP_SKB_CB(skb
)->cscov
= skb
->len
;
1085 if (proto
== IPPROTO_UDPLITE
) {
1086 err
= udplite_checksum_init(skb
, uh
);
1092 if (uh
->check
== 0) {
1093 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1094 } else if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
1095 if (!csum_tcpudp_magic(iph
->saddr
, iph
->daddr
, skb
->len
,
1097 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1099 if (!skb_csum_unnecessary(skb
))
1100 skb
->csum
= csum_tcpudp_nofold(iph
->saddr
, iph
->daddr
,
1101 skb
->len
, proto
, 0);
1102 /* Probably, we should checksum udp header (it should be in cache
1103 * in any case) and data in tiny packets (< rx copybreak).
1110 * All we need to do is get the socket, and then do a checksum.
1113 int __udp4_lib_rcv(struct sk_buff
*skb
, struct hlist_head udptable
[],
1117 struct udphdr
*uh
= udp_hdr(skb
);
1118 unsigned short ulen
;
1119 struct rtable
*rt
= (struct rtable
*)skb
->dst
;
1120 __be32 saddr
= ip_hdr(skb
)->saddr
;
1121 __be32 daddr
= ip_hdr(skb
)->daddr
;
1124 * Validate the packet.
1126 if (!pskb_may_pull(skb
, sizeof(struct udphdr
)))
1127 goto drop
; /* No space for header. */
1129 ulen
= ntohs(uh
->len
);
1130 if (ulen
> skb
->len
)
1133 if (proto
== IPPROTO_UDP
) {
1134 /* UDP validates ulen. */
1135 if (ulen
< sizeof(*uh
) || pskb_trim_rcsum(skb
, ulen
))
1140 if (udp4_csum_init(skb
, uh
, proto
))
1143 if (rt
->rt_flags
& (RTCF_BROADCAST
|RTCF_MULTICAST
))
1144 return __udp4_lib_mcast_deliver(skb
, uh
, saddr
, daddr
, udptable
);
1146 sk
= __udp4_lib_lookup(saddr
, uh
->source
, daddr
, uh
->dest
,
1147 skb
->dev
->ifindex
, udptable
);
1150 int ret
= udp_queue_rcv_skb(sk
, skb
);
1153 /* a return value > 0 means to resubmit the input, but
1154 * it wants the return to be -protocol, or 0
1161 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
1165 /* No socket. Drop packet silently, if checksum is wrong */
1166 if (udp_lib_checksum_complete(skb
))
1169 UDP_INC_STATS_BH(UDP_MIB_NOPORTS
, proto
== IPPROTO_UDPLITE
);
1170 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_PORT_UNREACH
, 0);
1173 * Hmm. We got an UDP packet to a port to which we
1174 * don't wanna listen. Ignore it.
1180 LIMIT_NETDEBUG(KERN_DEBUG
"UDP%s: short packet: From %u.%u.%u.%u:%u %d/%d to %u.%u.%u.%u:%u\n",
1181 proto
== IPPROTO_UDPLITE
? "-Lite" : "",
1192 * RFC1122: OK. Discards the bad packet silently (as far as
1193 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1195 LIMIT_NETDEBUG(KERN_DEBUG
"UDP%s: bad checksum. From %d.%d.%d.%d:%d to %d.%d.%d.%d:%d ulen %d\n",
1196 proto
== IPPROTO_UDPLITE
? "-Lite" : "",
1203 UDP_INC_STATS_BH(UDP_MIB_INERRORS
, proto
== IPPROTO_UDPLITE
);
1208 int udp_rcv(struct sk_buff
*skb
)
1210 return __udp4_lib_rcv(skb
, udp_hash
, IPPROTO_UDP
);
1213 int udp_destroy_sock(struct sock
*sk
)
1216 udp_flush_pending_frames(sk
);
1222 * Socket option code for UDP
1224 int udp_lib_setsockopt(struct sock
*sk
, int level
, int optname
,
1225 char __user
*optval
, int optlen
,
1226 int (*push_pending_frames
)(struct sock
*))
1228 struct udp_sock
*up
= udp_sk(sk
);
1232 if (optlen
<sizeof(int))
1235 if (get_user(val
, (int __user
*)optval
))
1245 (*push_pending_frames
)(sk
);
1253 case UDP_ENCAP_ESPINUDP
:
1254 case UDP_ENCAP_ESPINUDP_NON_IKE
:
1255 up
->encap_rcv
= xfrm4_udp_encap_rcv
;
1257 case UDP_ENCAP_L2TPINUDP
:
1258 up
->encap_type
= val
;
1267 * UDP-Lite's partial checksum coverage (RFC 3828).
1269 /* The sender sets actual checksum coverage length via this option.
1270 * The case coverage > packet length is handled by send module. */
1271 case UDPLITE_SEND_CSCOV
:
1272 if (!up
->pcflag
) /* Disable the option on UDP sockets */
1273 return -ENOPROTOOPT
;
1274 if (val
!= 0 && val
< 8) /* Illegal coverage: use default (8) */
1277 up
->pcflag
|= UDPLITE_SEND_CC
;
1280 /* The receiver specifies a minimum checksum coverage value. To make
1281 * sense, this should be set to at least 8 (as done below). If zero is
1282 * used, this again means full checksum coverage. */
1283 case UDPLITE_RECV_CSCOV
:
1284 if (!up
->pcflag
) /* Disable the option on UDP sockets */
1285 return -ENOPROTOOPT
;
1286 if (val
!= 0 && val
< 8) /* Avoid silly minimal values. */
1289 up
->pcflag
|= UDPLITE_RECV_CC
;
1300 int udp_setsockopt(struct sock
*sk
, int level
, int optname
,
1301 char __user
*optval
, int optlen
)
1303 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1304 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1305 udp_push_pending_frames
);
1306 return ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
1309 #ifdef CONFIG_COMPAT
1310 int compat_udp_setsockopt(struct sock
*sk
, int level
, int optname
,
1311 char __user
*optval
, int optlen
)
1313 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1314 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1315 udp_push_pending_frames
);
1316 return compat_ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
1320 int udp_lib_getsockopt(struct sock
*sk
, int level
, int optname
,
1321 char __user
*optval
, int __user
*optlen
)
1323 struct udp_sock
*up
= udp_sk(sk
);
1326 if (get_user(len
,optlen
))
1329 len
= min_t(unsigned int, len
, sizeof(int));
1340 val
= up
->encap_type
;
1343 /* The following two cannot be changed on UDP sockets, the return is
1344 * always 0 (which corresponds to the full checksum coverage of UDP). */
1345 case UDPLITE_SEND_CSCOV
:
1349 case UDPLITE_RECV_CSCOV
:
1354 return -ENOPROTOOPT
;
1357 if (put_user(len
, optlen
))
1359 if (copy_to_user(optval
, &val
,len
))
1364 int udp_getsockopt(struct sock
*sk
, int level
, int optname
,
1365 char __user
*optval
, int __user
*optlen
)
1367 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1368 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1369 return ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
1372 #ifdef CONFIG_COMPAT
1373 int compat_udp_getsockopt(struct sock
*sk
, int level
, int optname
,
1374 char __user
*optval
, int __user
*optlen
)
1376 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1377 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1378 return compat_ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
1382 * udp_poll - wait for a UDP event.
1383 * @file - file struct
1385 * @wait - poll table
1387 * This is same as datagram poll, except for the special case of
1388 * blocking sockets. If application is using a blocking fd
1389 * and a packet with checksum error is in the queue;
1390 * then it could get return from select indicating data available
1391 * but then block when reading it. Add special case code
1392 * to work around these arguably broken applications.
1394 unsigned int udp_poll(struct file
*file
, struct socket
*sock
, poll_table
*wait
)
1396 unsigned int mask
= datagram_poll(file
, sock
, wait
);
1397 struct sock
*sk
= sock
->sk
;
1398 int is_lite
= IS_UDPLITE(sk
);
1400 /* Check for false positives due to checksum errors */
1401 if ( (mask
& POLLRDNORM
) &&
1402 !(file
->f_flags
& O_NONBLOCK
) &&
1403 !(sk
->sk_shutdown
& RCV_SHUTDOWN
)){
1404 struct sk_buff_head
*rcvq
= &sk
->sk_receive_queue
;
1405 struct sk_buff
*skb
;
1407 spin_lock_bh(&rcvq
->lock
);
1408 while ((skb
= skb_peek(rcvq
)) != NULL
&&
1409 udp_lib_checksum_complete(skb
)) {
1410 UDP_INC_STATS_BH(UDP_MIB_INERRORS
, is_lite
);
1411 __skb_unlink(skb
, rcvq
);
1414 spin_unlock_bh(&rcvq
->lock
);
1416 /* nothing to see, move along */
1418 mask
&= ~(POLLIN
| POLLRDNORM
);
1425 struct proto udp_prot
= {
1427 .owner
= THIS_MODULE
,
1428 .close
= udp_lib_close
,
1429 .connect
= ip4_datagram_connect
,
1430 .disconnect
= udp_disconnect
,
1432 .destroy
= udp_destroy_sock
,
1433 .setsockopt
= udp_setsockopt
,
1434 .getsockopt
= udp_getsockopt
,
1435 .sendmsg
= udp_sendmsg
,
1436 .recvmsg
= udp_recvmsg
,
1437 .sendpage
= udp_sendpage
,
1438 .backlog_rcv
= udp_queue_rcv_skb
,
1439 .hash
= udp_lib_hash
,
1440 .unhash
= udp_lib_unhash
,
1441 .get_port
= udp_v4_get_port
,
1442 .obj_size
= sizeof(struct udp_sock
),
1443 #ifdef CONFIG_COMPAT
1444 .compat_setsockopt
= compat_udp_setsockopt
,
1445 .compat_getsockopt
= compat_udp_getsockopt
,
1449 /* ------------------------------------------------------------------------ */
1450 #ifdef CONFIG_PROC_FS
1452 static struct sock
*udp_get_first(struct seq_file
*seq
)
1455 struct udp_iter_state
*state
= seq
->private;
1457 for (state
->bucket
= 0; state
->bucket
< UDP_HTABLE_SIZE
; ++state
->bucket
) {
1458 struct hlist_node
*node
;
1459 sk_for_each(sk
, node
, state
->hashtable
+ state
->bucket
) {
1460 if (sk
->sk_family
== state
->family
)
1469 static struct sock
*udp_get_next(struct seq_file
*seq
, struct sock
*sk
)
1471 struct udp_iter_state
*state
= seq
->private;
1477 } while (sk
&& sk
->sk_family
!= state
->family
);
1479 if (!sk
&& ++state
->bucket
< UDP_HTABLE_SIZE
) {
1480 sk
= sk_head(state
->hashtable
+ state
->bucket
);
1486 static struct sock
*udp_get_idx(struct seq_file
*seq
, loff_t pos
)
1488 struct sock
*sk
= udp_get_first(seq
);
1491 while (pos
&& (sk
= udp_get_next(seq
, sk
)) != NULL
)
1493 return pos
? NULL
: sk
;
1496 static void *udp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1498 read_lock(&udp_hash_lock
);
1499 return *pos
? udp_get_idx(seq
, *pos
-1) : (void *)1;
1502 static void *udp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1507 sk
= udp_get_idx(seq
, 0);
1509 sk
= udp_get_next(seq
, v
);
1515 static void udp_seq_stop(struct seq_file
*seq
, void *v
)
1517 read_unlock(&udp_hash_lock
);
1520 static int udp_seq_open(struct inode
*inode
, struct file
*file
)
1522 struct udp_seq_afinfo
*afinfo
= PDE(inode
)->data
;
1523 struct seq_file
*seq
;
1525 struct udp_iter_state
*s
= kzalloc(sizeof(*s
), GFP_KERNEL
);
1529 s
->family
= afinfo
->family
;
1530 s
->hashtable
= afinfo
->hashtable
;
1531 s
->seq_ops
.start
= udp_seq_start
;
1532 s
->seq_ops
.next
= udp_seq_next
;
1533 s
->seq_ops
.show
= afinfo
->seq_show
;
1534 s
->seq_ops
.stop
= udp_seq_stop
;
1536 rc
= seq_open(file
, &s
->seq_ops
);
1540 seq
= file
->private_data
;
1549 /* ------------------------------------------------------------------------ */
1550 int udp_proc_register(struct udp_seq_afinfo
*afinfo
)
1552 struct proc_dir_entry
*p
;
1557 afinfo
->seq_fops
->owner
= afinfo
->owner
;
1558 afinfo
->seq_fops
->open
= udp_seq_open
;
1559 afinfo
->seq_fops
->read
= seq_read
;
1560 afinfo
->seq_fops
->llseek
= seq_lseek
;
1561 afinfo
->seq_fops
->release
= seq_release_private
;
1563 p
= proc_net_fops_create(afinfo
->name
, S_IRUGO
, afinfo
->seq_fops
);
1571 void udp_proc_unregister(struct udp_seq_afinfo
*afinfo
)
1575 proc_net_remove(afinfo
->name
);
1576 memset(afinfo
->seq_fops
, 0, sizeof(*afinfo
->seq_fops
));
1579 /* ------------------------------------------------------------------------ */
1580 static void udp4_format_sock(struct sock
*sp
, char *tmpbuf
, int bucket
)
1582 struct inet_sock
*inet
= inet_sk(sp
);
1583 __be32 dest
= inet
->daddr
;
1584 __be32 src
= inet
->rcv_saddr
;
1585 __u16 destp
= ntohs(inet
->dport
);
1586 __u16 srcp
= ntohs(inet
->sport
);
1588 sprintf(tmpbuf
, "%4d: %08X:%04X %08X:%04X"
1589 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p",
1590 bucket
, src
, srcp
, dest
, destp
, sp
->sk_state
,
1591 atomic_read(&sp
->sk_wmem_alloc
),
1592 atomic_read(&sp
->sk_rmem_alloc
),
1593 0, 0L, 0, sock_i_uid(sp
), 0, sock_i_ino(sp
),
1594 atomic_read(&sp
->sk_refcnt
), sp
);
1597 int udp4_seq_show(struct seq_file
*seq
, void *v
)
1599 if (v
== SEQ_START_TOKEN
)
1600 seq_printf(seq
, "%-127s\n",
1601 " sl local_address rem_address st tx_queue "
1602 "rx_queue tr tm->when retrnsmt uid timeout "
1606 struct udp_iter_state
*state
= seq
->private;
1608 udp4_format_sock(v
, tmpbuf
, state
->bucket
);
1609 seq_printf(seq
, "%-127s\n", tmpbuf
);
1614 /* ------------------------------------------------------------------------ */
1615 static struct file_operations udp4_seq_fops
;
1616 static struct udp_seq_afinfo udp4_seq_afinfo
= {
1617 .owner
= THIS_MODULE
,
1620 .hashtable
= udp_hash
,
1621 .seq_show
= udp4_seq_show
,
1622 .seq_fops
= &udp4_seq_fops
,
1625 int __init
udp4_proc_init(void)
1627 return udp_proc_register(&udp4_seq_afinfo
);
1630 void udp4_proc_exit(void)
1632 udp_proc_unregister(&udp4_seq_afinfo
);
1634 #endif /* CONFIG_PROC_FS */
1636 EXPORT_SYMBOL(udp_disconnect
);
1637 EXPORT_SYMBOL(udp_hash
);
1638 EXPORT_SYMBOL(udp_hash_lock
);
1639 EXPORT_SYMBOL(udp_ioctl
);
1640 EXPORT_SYMBOL(udp_get_port
);
1641 EXPORT_SYMBOL(udp_prot
);
1642 EXPORT_SYMBOL(udp_sendmsg
);
1643 EXPORT_SYMBOL(udp_lib_getsockopt
);
1644 EXPORT_SYMBOL(udp_lib_setsockopt
);
1645 EXPORT_SYMBOL(udp_poll
);
1647 #ifdef CONFIG_PROC_FS
1648 EXPORT_SYMBOL(udp_proc_register
);
1649 EXPORT_SYMBOL(udp_proc_unregister
);