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).
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
11 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
12 * Hirokazu Takahashi, <taka@valinux.co.jp>
15 * Alan Cox : verify_area() calls
16 * Alan Cox : stopped close while in use off icmp
17 * messages. Not a fix but a botch that
18 * for udp at least is 'valid'.
19 * Alan Cox : Fixed icmp handling properly
20 * Alan Cox : Correct error for oversized datagrams
21 * Alan Cox : Tidied select() semantics.
22 * Alan Cox : udp_err() fixed properly, also now
23 * select and read wake correctly on errors
24 * Alan Cox : udp_send verify_area moved to avoid mem leak
25 * Alan Cox : UDP can count its memory
26 * Alan Cox : send to an unknown connection causes
27 * an ECONNREFUSED off the icmp, but
29 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
30 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
31 * bug no longer crashes it.
32 * Fred Van Kempen : Net2e support for sk->broadcast.
33 * Alan Cox : Uses skb_free_datagram
34 * Alan Cox : Added get/set sockopt support.
35 * Alan Cox : Broadcasting without option set returns EACCES.
36 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
37 * Alan Cox : Use ip_tos and ip_ttl
38 * Alan Cox : SNMP Mibs
39 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
40 * Matt Dillon : UDP length checks.
41 * Alan Cox : Smarter af_inet used properly.
42 * Alan Cox : Use new kernel side addressing.
43 * Alan Cox : Incorrect return on truncated datagram receive.
44 * Arnt Gulbrandsen : New udp_send and stuff
45 * Alan Cox : Cache last socket
46 * Alan Cox : Route cache
47 * Jon Peatfield : Minor efficiency fix to sendto().
48 * Mike Shaver : RFC1122 checks.
49 * Alan Cox : Nonblocking error fix.
50 * Willy Konynenberg : Transparent proxying support.
51 * Mike McLagan : Routing by source
52 * David S. Miller : New socket lookup architecture.
53 * Last socket cache retained as it
54 * does have a high hit rate.
55 * Olaf Kirch : Don't linearise iovec on sendmsg.
56 * Andi Kleen : Some cleanups, cache destination entry
58 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
59 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
60 * return ENOTCONN for unconnected sockets (POSIX)
61 * Janos Farkas : don't deliver multi/broadcasts to a different
62 * bound-to-device socket
63 * Hirokazu Takahashi : HW checksumming for outgoing UDP
65 * Hirokazu Takahashi : sendfile() on UDP works now.
66 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
67 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
68 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
69 * a single port at the same time.
70 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
71 * James Chapman : Add L2TP encapsulation type.
74 * This program is free software; you can redistribute it and/or
75 * modify it under the terms of the GNU General Public License
76 * as published by the Free Software Foundation; either version
77 * 2 of the License, or (at your option) any later version.
80 #include <asm/system.h>
81 #include <asm/uaccess.h>
82 #include <asm/ioctls.h>
83 #include <linux/bootmem.h>
84 #include <linux/highmem.h>
85 #include <linux/swap.h>
86 #include <linux/types.h>
87 #include <linux/fcntl.h>
88 #include <linux/module.h>
89 #include <linux/socket.h>
90 #include <linux/sockios.h>
91 #include <linux/igmp.h>
93 #include <linux/errno.h>
94 #include <linux/timer.h>
96 #include <linux/inet.h>
97 #include <linux/netdevice.h>
98 #include <net/tcp_states.h>
99 #include <linux/skbuff.h>
100 #include <linux/proc_fs.h>
101 #include <linux/seq_file.h>
102 #include <net/net_namespace.h>
103 #include <net/icmp.h>
104 #include <net/route.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include "udp_impl.h"
109 struct udp_table udp_table __read_mostly
;
110 EXPORT_SYMBOL(udp_table
);
112 int sysctl_udp_mem
[3] __read_mostly
;
113 EXPORT_SYMBOL(sysctl_udp_mem
);
115 int sysctl_udp_rmem_min __read_mostly
;
116 EXPORT_SYMBOL(sysctl_udp_rmem_min
);
118 int sysctl_udp_wmem_min __read_mostly
;
119 EXPORT_SYMBOL(sysctl_udp_wmem_min
);
121 atomic_t udp_memory_allocated
;
122 EXPORT_SYMBOL(udp_memory_allocated
);
124 #define MAX_UDP_PORTS 65536
125 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
127 static int udp_lib_lport_inuse(struct net
*net
, __u16 num
,
128 const struct udp_hslot
*hslot
,
129 unsigned long *bitmap
,
131 int (*saddr_comp
)(const struct sock
*sk1
,
132 const struct sock
*sk2
),
136 struct hlist_nulls_node
*node
;
138 sk_nulls_for_each(sk2
, node
, &hslot
->head
)
139 if (net_eq(sock_net(sk2
), net
) &&
141 (bitmap
|| sk2
->sk_hash
== num
) &&
142 (!sk2
->sk_reuse
|| !sk
->sk_reuse
) &&
143 (!sk2
->sk_bound_dev_if
|| !sk
->sk_bound_dev_if
144 || sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
145 (*saddr_comp
)(sk
, sk2
)) {
147 __set_bit(sk2
->sk_hash
>> log
, bitmap
);
155 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
157 * @sk: socket struct in question
158 * @snum: port number to look up
159 * @saddr_comp: AF-dependent comparison of bound local IP addresses
161 int udp_lib_get_port(struct sock
*sk
, unsigned short snum
,
162 int (*saddr_comp
)(const struct sock
*sk1
,
163 const struct sock
*sk2
))
165 struct udp_hslot
*hslot
;
166 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
168 struct net
*net
= sock_net(sk
);
171 int low
, high
, remaining
;
173 unsigned short first
, last
;
174 DECLARE_BITMAP(bitmap
, PORTS_PER_CHAIN
);
176 inet_get_local_port_range(&low
, &high
);
177 remaining
= (high
- low
) + 1;
180 first
= (((u64
)rand
* remaining
) >> 32) + low
;
182 * force rand to be an odd multiple of UDP_HTABLE_SIZE
184 rand
= (rand
| 1) * (udptable
->mask
+ 1);
185 for (last
= first
+ udptable
->mask
+ 1;
188 hslot
= udp_hashslot(udptable
, net
, first
);
189 bitmap_zero(bitmap
, PORTS_PER_CHAIN
);
190 spin_lock_bh(&hslot
->lock
);
191 udp_lib_lport_inuse(net
, snum
, hslot
, bitmap
, sk
,
192 saddr_comp
, udptable
->log
);
196 * Iterate on all possible values of snum for this hash.
197 * Using steps of an odd multiple of UDP_HTABLE_SIZE
198 * give us randomization and full range coverage.
201 if (low
<= snum
&& snum
<= high
&&
202 !test_bit(snum
>> udptable
->log
, bitmap
))
205 } while (snum
!= first
);
206 spin_unlock_bh(&hslot
->lock
);
210 hslot
= udp_hashslot(udptable
, net
, snum
);
211 spin_lock_bh(&hslot
->lock
);
212 if (udp_lib_lport_inuse(net
, snum
, hslot
, NULL
, sk
,
217 inet_sk(sk
)->inet_num
= snum
;
219 if (sk_unhashed(sk
)) {
220 sk_nulls_add_node_rcu(sk
, &hslot
->head
);
221 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, 1);
225 spin_unlock_bh(&hslot
->lock
);
229 EXPORT_SYMBOL(udp_lib_get_port
);
231 static int ipv4_rcv_saddr_equal(const struct sock
*sk1
, const struct sock
*sk2
)
233 struct inet_sock
*inet1
= inet_sk(sk1
), *inet2
= inet_sk(sk2
);
235 return (!ipv6_only_sock(sk2
) &&
236 (!inet1
->inet_rcv_saddr
|| !inet2
->inet_rcv_saddr
||
237 inet1
->inet_rcv_saddr
== inet2
->inet_rcv_saddr
));
240 int udp_v4_get_port(struct sock
*sk
, unsigned short snum
)
242 return udp_lib_get_port(sk
, snum
, ipv4_rcv_saddr_equal
);
245 static inline int compute_score(struct sock
*sk
, struct net
*net
, __be32 saddr
,
247 __be16 sport
, __be32 daddr
, __be16 dport
, int dif
)
251 if (net_eq(sock_net(sk
), net
) && sk
->sk_hash
== hnum
&&
252 !ipv6_only_sock(sk
)) {
253 struct inet_sock
*inet
= inet_sk(sk
);
255 score
= (sk
->sk_family
== PF_INET
? 1 : 0);
256 if (inet
->inet_rcv_saddr
) {
257 if (inet
->inet_rcv_saddr
!= daddr
)
261 if (inet
->inet_daddr
) {
262 if (inet
->inet_daddr
!= saddr
)
266 if (inet
->inet_dport
) {
267 if (inet
->inet_dport
!= sport
)
271 if (sk
->sk_bound_dev_if
) {
272 if (sk
->sk_bound_dev_if
!= dif
)
280 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
281 * harder than this. -DaveM
283 static struct sock
*__udp4_lib_lookup(struct net
*net
, __be32 saddr
,
284 __be16 sport
, __be32 daddr
, __be16 dport
,
285 int dif
, struct udp_table
*udptable
)
287 struct sock
*sk
, *result
;
288 struct hlist_nulls_node
*node
;
289 unsigned short hnum
= ntohs(dport
);
290 unsigned int hash
= udp_hashfn(net
, hnum
, udptable
->mask
);
291 struct udp_hslot
*hslot
= &udptable
->hash
[hash
];
298 sk_nulls_for_each_rcu(sk
, node
, &hslot
->head
) {
299 score
= compute_score(sk
, net
, saddr
, hnum
, sport
,
301 if (score
> badness
) {
307 * if the nulls value we got at the end of this lookup is
308 * not the expected one, we must restart lookup.
309 * We probably met an item that was moved to another chain.
311 if (get_nulls_value(node
) != hash
)
315 if (unlikely(!atomic_inc_not_zero(&result
->sk_refcnt
)))
317 else if (unlikely(compute_score(result
, net
, saddr
, hnum
, sport
,
318 daddr
, dport
, dif
) < badness
)) {
327 static inline struct sock
*__udp4_lib_lookup_skb(struct sk_buff
*skb
,
328 __be16 sport
, __be16 dport
,
329 struct udp_table
*udptable
)
332 const struct iphdr
*iph
= ip_hdr(skb
);
334 if (unlikely(sk
= skb_steal_sock(skb
)))
337 return __udp4_lib_lookup(dev_net(skb_dst(skb
)->dev
), iph
->saddr
, sport
,
338 iph
->daddr
, dport
, inet_iif(skb
),
342 struct sock
*udp4_lib_lookup(struct net
*net
, __be32 saddr
, __be16 sport
,
343 __be32 daddr
, __be16 dport
, int dif
)
345 return __udp4_lib_lookup(net
, saddr
, sport
, daddr
, dport
, dif
, &udp_table
);
347 EXPORT_SYMBOL_GPL(udp4_lib_lookup
);
349 static inline struct sock
*udp_v4_mcast_next(struct net
*net
, struct sock
*sk
,
350 __be16 loc_port
, __be32 loc_addr
,
351 __be16 rmt_port
, __be32 rmt_addr
,
354 struct hlist_nulls_node
*node
;
356 unsigned short hnum
= ntohs(loc_port
);
358 sk_nulls_for_each_from(s
, node
) {
359 struct inet_sock
*inet
= inet_sk(s
);
361 if (!net_eq(sock_net(s
), net
) ||
362 s
->sk_hash
!= hnum
||
363 (inet
->inet_daddr
&& inet
->inet_daddr
!= rmt_addr
) ||
364 (inet
->inet_dport
!= rmt_port
&& inet
->inet_dport
) ||
365 (inet
->inet_rcv_saddr
&&
366 inet
->inet_rcv_saddr
!= loc_addr
) ||
368 (s
->sk_bound_dev_if
&& s
->sk_bound_dev_if
!= dif
))
370 if (!ip_mc_sf_allow(s
, loc_addr
, rmt_addr
, dif
))
380 * This routine is called by the ICMP module when it gets some
381 * sort of error condition. If err < 0 then the socket should
382 * be closed and the error returned to the user. If err > 0
383 * it's just the icmp type << 8 | icmp code.
384 * Header points to the ip header of the error packet. We move
385 * on past this. Then (as it used to claim before adjustment)
386 * header points to the first 8 bytes of the udp header. We need
387 * to find the appropriate port.
390 void __udp4_lib_err(struct sk_buff
*skb
, u32 info
, struct udp_table
*udptable
)
392 struct inet_sock
*inet
;
393 struct iphdr
*iph
= (struct iphdr
*)skb
->data
;
394 struct udphdr
*uh
= (struct udphdr
*)(skb
->data
+(iph
->ihl
<<2));
395 const int type
= icmp_hdr(skb
)->type
;
396 const int code
= icmp_hdr(skb
)->code
;
400 struct net
*net
= dev_net(skb
->dev
);
402 sk
= __udp4_lib_lookup(net
, iph
->daddr
, uh
->dest
,
403 iph
->saddr
, uh
->source
, skb
->dev
->ifindex
, udptable
);
405 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
406 return; /* No socket for error */
415 case ICMP_TIME_EXCEEDED
:
418 case ICMP_SOURCE_QUENCH
:
420 case ICMP_PARAMETERPROB
:
424 case ICMP_DEST_UNREACH
:
425 if (code
== ICMP_FRAG_NEEDED
) { /* Path MTU discovery */
426 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
) {
434 if (code
<= NR_ICMP_UNREACH
) {
435 harderr
= icmp_err_convert
[code
].fatal
;
436 err
= icmp_err_convert
[code
].errno
;
442 * RFC1122: OK. Passes ICMP errors back to application, as per
445 if (!inet
->recverr
) {
446 if (!harderr
|| sk
->sk_state
!= TCP_ESTABLISHED
)
449 ip_icmp_error(sk
, skb
, err
, uh
->dest
, info
, (u8
*)(uh
+1));
452 sk
->sk_error_report(sk
);
457 void udp_err(struct sk_buff
*skb
, u32 info
)
459 __udp4_lib_err(skb
, info
, &udp_table
);
463 * Throw away all pending data and cancel the corking. Socket is locked.
465 void udp_flush_pending_frames(struct sock
*sk
)
467 struct udp_sock
*up
= udp_sk(sk
);
472 ip_flush_pending_frames(sk
);
475 EXPORT_SYMBOL(udp_flush_pending_frames
);
478 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
479 * @sk: socket we are sending on
480 * @skb: sk_buff containing the filled-in UDP header
481 * (checksum field must be zeroed out)
483 static void udp4_hwcsum_outgoing(struct sock
*sk
, struct sk_buff
*skb
,
484 __be32 src
, __be32 dst
, int len
)
487 struct udphdr
*uh
= udp_hdr(skb
);
490 if (skb_queue_len(&sk
->sk_write_queue
) == 1) {
492 * Only one fragment on the socket.
494 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
495 skb
->csum_offset
= offsetof(struct udphdr
, check
);
496 uh
->check
= ~csum_tcpudp_magic(src
, dst
, len
, IPPROTO_UDP
, 0);
499 * HW-checksum won't work as there are two or more
500 * fragments on the socket so that all csums of sk_buffs
503 offset
= skb_transport_offset(skb
);
504 skb
->csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
506 skb
->ip_summed
= CHECKSUM_NONE
;
508 skb_queue_walk(&sk
->sk_write_queue
, skb
) {
509 csum
= csum_add(csum
, skb
->csum
);
512 uh
->check
= csum_tcpudp_magic(src
, dst
, len
, IPPROTO_UDP
, csum
);
514 uh
->check
= CSUM_MANGLED_0
;
519 * Push out all pending data as one UDP datagram. Socket is locked.
521 static int udp_push_pending_frames(struct sock
*sk
)
523 struct udp_sock
*up
= udp_sk(sk
);
524 struct inet_sock
*inet
= inet_sk(sk
);
525 struct flowi
*fl
= &inet
->cork
.fl
;
529 int is_udplite
= IS_UDPLITE(sk
);
532 /* Grab the skbuff where UDP header space exists. */
533 if ((skb
= skb_peek(&sk
->sk_write_queue
)) == NULL
)
537 * Create a UDP header
540 uh
->source
= fl
->fl_ip_sport
;
541 uh
->dest
= fl
->fl_ip_dport
;
542 uh
->len
= htons(up
->len
);
545 if (is_udplite
) /* UDP-Lite */
546 csum
= udplite_csum_outgoing(sk
, skb
);
548 else if (sk
->sk_no_check
== UDP_CSUM_NOXMIT
) { /* UDP csum disabled */
550 skb
->ip_summed
= CHECKSUM_NONE
;
553 } else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) { /* UDP hardware csum */
555 udp4_hwcsum_outgoing(sk
, skb
, fl
->fl4_src
, fl
->fl4_dst
, up
->len
);
558 } else /* `normal' UDP */
559 csum
= udp_csum_outgoing(sk
, skb
);
561 /* add protocol-dependent pseudo-header */
562 uh
->check
= csum_tcpudp_magic(fl
->fl4_src
, fl
->fl4_dst
, up
->len
,
563 sk
->sk_protocol
, csum
);
565 uh
->check
= CSUM_MANGLED_0
;
568 err
= ip_push_pending_frames(sk
);
570 if (err
== -ENOBUFS
&& !inet
->recverr
) {
571 UDP_INC_STATS_USER(sock_net(sk
),
572 UDP_MIB_SNDBUFERRORS
, is_udplite
);
576 UDP_INC_STATS_USER(sock_net(sk
),
577 UDP_MIB_OUTDATAGRAMS
, is_udplite
);
584 int udp_sendmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
587 struct inet_sock
*inet
= inet_sk(sk
);
588 struct udp_sock
*up
= udp_sk(sk
);
590 struct ipcm_cookie ipc
;
591 struct rtable
*rt
= NULL
;
594 __be32 daddr
, faddr
, saddr
;
597 int err
, is_udplite
= IS_UDPLITE(sk
);
598 int corkreq
= up
->corkflag
|| msg
->msg_flags
&MSG_MORE
;
599 int (*getfrag
)(void *, char *, int, int, int, struct sk_buff
*);
608 if (msg
->msg_flags
& MSG_OOB
) /* Mirror BSD error message compatibility */
616 * There are pending frames.
617 * The socket lock must be held while it's corked.
620 if (likely(up
->pending
)) {
621 if (unlikely(up
->pending
!= AF_INET
)) {
629 ulen
+= sizeof(struct udphdr
);
632 * Get and verify the address.
635 struct sockaddr_in
* usin
= (struct sockaddr_in
*)msg
->msg_name
;
636 if (msg
->msg_namelen
< sizeof(*usin
))
638 if (usin
->sin_family
!= AF_INET
) {
639 if (usin
->sin_family
!= AF_UNSPEC
)
640 return -EAFNOSUPPORT
;
643 daddr
= usin
->sin_addr
.s_addr
;
644 dport
= usin
->sin_port
;
648 if (sk
->sk_state
!= TCP_ESTABLISHED
)
649 return -EDESTADDRREQ
;
650 daddr
= inet
->inet_daddr
;
651 dport
= inet
->inet_dport
;
652 /* Open fast path for connected socket.
653 Route will not be used, if at least one option is set.
657 ipc
.addr
= inet
->inet_saddr
;
659 ipc
.oif
= sk
->sk_bound_dev_if
;
660 err
= sock_tx_timestamp(msg
, sk
, &ipc
.shtx
);
663 if (msg
->msg_controllen
) {
664 err
= ip_cmsg_send(sock_net(sk
), msg
, &ipc
);
675 ipc
.addr
= faddr
= daddr
;
677 if (ipc
.opt
&& ipc
.opt
->srr
) {
680 faddr
= ipc
.opt
->faddr
;
683 tos
= RT_TOS(inet
->tos
);
684 if (sock_flag(sk
, SOCK_LOCALROUTE
) ||
685 (msg
->msg_flags
& MSG_DONTROUTE
) ||
686 (ipc
.opt
&& ipc
.opt
->is_strictroute
)) {
691 if (ipv4_is_multicast(daddr
)) {
693 ipc
.oif
= inet
->mc_index
;
695 saddr
= inet
->mc_addr
;
700 rt
= (struct rtable
*)sk_dst_check(sk
, 0);
703 struct flowi fl
= { .oif
= ipc
.oif
,
709 .proto
= sk
->sk_protocol
,
710 .flags
= inet_sk_flowi_flags(sk
),
712 { .sport
= inet
->inet_sport
,
713 .dport
= dport
} } };
714 struct net
*net
= sock_net(sk
);
716 security_sk_classify_flow(sk
, &fl
);
717 err
= ip_route_output_flow(net
, &rt
, &fl
, sk
, 1);
719 if (err
== -ENETUNREACH
)
720 IP_INC_STATS_BH(net
, IPSTATS_MIB_OUTNOROUTES
);
725 if ((rt
->rt_flags
& RTCF_BROADCAST
) &&
726 !sock_flag(sk
, SOCK_BROADCAST
))
729 sk_dst_set(sk
, dst_clone(&rt
->u
.dst
));
732 if (msg
->msg_flags
&MSG_CONFIRM
)
738 daddr
= ipc
.addr
= rt
->rt_dst
;
741 if (unlikely(up
->pending
)) {
742 /* The socket is already corked while preparing it. */
743 /* ... which is an evident application bug. --ANK */
746 LIMIT_NETDEBUG(KERN_DEBUG
"udp cork app bug 2\n");
751 * Now cork the socket to pend data.
753 inet
->cork
.fl
.fl4_dst
= daddr
;
754 inet
->cork
.fl
.fl_ip_dport
= dport
;
755 inet
->cork
.fl
.fl4_src
= saddr
;
756 inet
->cork
.fl
.fl_ip_sport
= inet
->inet_sport
;
757 up
->pending
= AF_INET
;
761 getfrag
= is_udplite
? udplite_getfrag
: ip_generic_getfrag
;
762 err
= ip_append_data(sk
, getfrag
, msg
->msg_iov
, ulen
,
763 sizeof(struct udphdr
), &ipc
, &rt
,
764 corkreq
? msg
->msg_flags
|MSG_MORE
: msg
->msg_flags
);
766 udp_flush_pending_frames(sk
);
768 err
= udp_push_pending_frames(sk
);
769 else if (unlikely(skb_queue_empty(&sk
->sk_write_queue
)))
780 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
781 * ENOBUFS might not be good (it's not tunable per se), but otherwise
782 * we don't have a good statistic (IpOutDiscards but it can be too many
783 * things). We could add another new stat but at least for now that
784 * seems like overkill.
786 if (err
== -ENOBUFS
|| test_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
)) {
787 UDP_INC_STATS_USER(sock_net(sk
),
788 UDP_MIB_SNDBUFERRORS
, is_udplite
);
793 dst_confirm(&rt
->u
.dst
);
794 if (!(msg
->msg_flags
&MSG_PROBE
) || len
)
795 goto back_from_confirm
;
799 EXPORT_SYMBOL(udp_sendmsg
);
801 int udp_sendpage(struct sock
*sk
, struct page
*page
, int offset
,
802 size_t size
, int flags
)
804 struct udp_sock
*up
= udp_sk(sk
);
808 struct msghdr msg
= { .msg_flags
= flags
|MSG_MORE
};
810 /* Call udp_sendmsg to specify destination address which
811 * sendpage interface can't pass.
812 * This will succeed only when the socket is connected.
814 ret
= udp_sendmsg(NULL
, sk
, &msg
, 0);
821 if (unlikely(!up
->pending
)) {
824 LIMIT_NETDEBUG(KERN_DEBUG
"udp cork app bug 3\n");
828 ret
= ip_append_page(sk
, page
, offset
, size
, flags
);
829 if (ret
== -EOPNOTSUPP
) {
831 return sock_no_sendpage(sk
->sk_socket
, page
, offset
,
835 udp_flush_pending_frames(sk
);
840 if (!(up
->corkflag
|| (flags
&MSG_MORE
)))
841 ret
= udp_push_pending_frames(sk
);
851 * first_packet_length - return length of first packet in receive queue
854 * Drops all bad checksum frames, until a valid one is found.
855 * Returns the length of found skb, or 0 if none is found.
857 static unsigned int first_packet_length(struct sock
*sk
)
859 struct sk_buff_head list_kill
, *rcvq
= &sk
->sk_receive_queue
;
863 __skb_queue_head_init(&list_kill
);
865 spin_lock_bh(&rcvq
->lock
);
866 while ((skb
= skb_peek(rcvq
)) != NULL
&&
867 udp_lib_checksum_complete(skb
)) {
868 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
,
870 atomic_inc(&sk
->sk_drops
);
871 __skb_unlink(skb
, rcvq
);
872 __skb_queue_tail(&list_kill
, skb
);
874 res
= skb
? skb
->len
: 0;
875 spin_unlock_bh(&rcvq
->lock
);
877 if (!skb_queue_empty(&list_kill
)) {
879 __skb_queue_purge(&list_kill
);
880 sk_mem_reclaim_partial(sk
);
887 * IOCTL requests applicable to the UDP protocol
890 int udp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
)
895 int amount
= sk_wmem_alloc_get(sk
);
897 return put_user(amount
, (int __user
*)arg
);
902 unsigned int amount
= first_packet_length(sk
);
906 * We will only return the amount
907 * of this packet since that is all
910 amount
-= sizeof(struct udphdr
);
912 return put_user(amount
, (int __user
*)arg
);
921 EXPORT_SYMBOL(udp_ioctl
);
924 * This should be easy, if there is something there we
925 * return it, otherwise we block.
928 int udp_recvmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
929 size_t len
, int noblock
, int flags
, int *addr_len
)
931 struct inet_sock
*inet
= inet_sk(sk
);
932 struct sockaddr_in
*sin
= (struct sockaddr_in
*)msg
->msg_name
;
934 unsigned int ulen
, copied
;
937 int is_udplite
= IS_UDPLITE(sk
);
940 * Check any passed addresses
943 *addr_len
= sizeof(*sin
);
945 if (flags
& MSG_ERRQUEUE
)
946 return ip_recv_error(sk
, msg
, len
);
949 skb
= __skb_recv_datagram(sk
, flags
| (noblock
? MSG_DONTWAIT
: 0),
954 ulen
= skb
->len
- sizeof(struct udphdr
);
958 else if (copied
< ulen
)
959 msg
->msg_flags
|= MSG_TRUNC
;
962 * If checksum is needed at all, try to do it while copying the
963 * data. If the data is truncated, or if we only want a partial
964 * coverage checksum (UDP-Lite), do it before the copy.
967 if (copied
< ulen
|| UDP_SKB_CB(skb
)->partial_cov
) {
968 if (udp_lib_checksum_complete(skb
))
972 if (skb_csum_unnecessary(skb
))
973 err
= skb_copy_datagram_iovec(skb
, sizeof(struct udphdr
),
974 msg
->msg_iov
, copied
);
976 err
= skb_copy_and_csum_datagram_iovec(skb
,
977 sizeof(struct udphdr
),
988 UDP_INC_STATS_USER(sock_net(sk
),
989 UDP_MIB_INDATAGRAMS
, is_udplite
);
991 sock_recv_ts_and_drops(msg
, sk
, skb
);
993 /* Copy the address. */
995 sin
->sin_family
= AF_INET
;
996 sin
->sin_port
= udp_hdr(skb
)->source
;
997 sin
->sin_addr
.s_addr
= ip_hdr(skb
)->saddr
;
998 memset(sin
->sin_zero
, 0, sizeof(sin
->sin_zero
));
1000 if (inet
->cmsg_flags
)
1001 ip_cmsg_recv(msg
, skb
);
1004 if (flags
& MSG_TRUNC
)
1008 skb_free_datagram_locked(sk
, skb
);
1014 if (!skb_kill_datagram(sk
, skb
, flags
))
1015 UDP_INC_STATS_USER(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1024 int udp_disconnect(struct sock
*sk
, int flags
)
1026 struct inet_sock
*inet
= inet_sk(sk
);
1028 * 1003.1g - break association.
1031 sk
->sk_state
= TCP_CLOSE
;
1032 inet
->inet_daddr
= 0;
1033 inet
->inet_dport
= 0;
1034 sk
->sk_bound_dev_if
= 0;
1035 if (!(sk
->sk_userlocks
& SOCK_BINDADDR_LOCK
))
1036 inet_reset_saddr(sk
);
1038 if (!(sk
->sk_userlocks
& SOCK_BINDPORT_LOCK
)) {
1039 sk
->sk_prot
->unhash(sk
);
1040 inet
->inet_sport
= 0;
1045 EXPORT_SYMBOL(udp_disconnect
);
1047 void udp_lib_unhash(struct sock
*sk
)
1049 if (sk_hashed(sk
)) {
1050 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
1051 struct udp_hslot
*hslot
= udp_hashslot(udptable
, sock_net(sk
),
1054 spin_lock_bh(&hslot
->lock
);
1055 if (sk_nulls_del_node_init_rcu(sk
)) {
1056 inet_sk(sk
)->inet_num
= 0;
1057 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, -1);
1059 spin_unlock_bh(&hslot
->lock
);
1062 EXPORT_SYMBOL(udp_lib_unhash
);
1064 static int __udp_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
1066 int rc
= sock_queue_rcv_skb(sk
, skb
);
1069 int is_udplite
= IS_UDPLITE(sk
);
1071 /* Note that an ENOMEM error is charged twice */
1073 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_RCVBUFERRORS
,
1075 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1087 * >0: "udp encap" protocol resubmission
1089 * Note that in the success and error cases, the skb is assumed to
1090 * have either been requeued or freed.
1092 int udp_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
1094 struct udp_sock
*up
= udp_sk(sk
);
1096 int is_udplite
= IS_UDPLITE(sk
);
1099 * Charge it to the socket, dropping if the queue is full.
1101 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1105 if (up
->encap_type
) {
1107 * This is an encapsulation socket so pass the skb to
1108 * the socket's udp_encap_rcv() hook. Otherwise, just
1109 * fall through and pass this up the UDP socket.
1110 * up->encap_rcv() returns the following value:
1111 * =0 if skb was successfully passed to the encap
1112 * handler or was discarded by it.
1113 * >0 if skb should be passed on to UDP.
1114 * <0 if skb should be resubmitted as proto -N
1117 /* if we're overly short, let UDP handle it */
1118 if (skb
->len
> sizeof(struct udphdr
) &&
1119 up
->encap_rcv
!= NULL
) {
1122 ret
= (*up
->encap_rcv
)(sk
, skb
);
1124 UDP_INC_STATS_BH(sock_net(sk
),
1125 UDP_MIB_INDATAGRAMS
,
1131 /* FALLTHROUGH -- it's a UDP Packet */
1135 * UDP-Lite specific tests, ignored on UDP sockets
1137 if ((is_udplite
& UDPLITE_RECV_CC
) && UDP_SKB_CB(skb
)->partial_cov
) {
1140 * MIB statistics other than incrementing the error count are
1141 * disabled for the following two types of errors: these depend
1142 * on the application settings, not on the functioning of the
1143 * protocol stack as such.
1145 * RFC 3828 here recommends (sec 3.3): "There should also be a
1146 * way ... to ... at least let the receiving application block
1147 * delivery of packets with coverage values less than a value
1148 * provided by the application."
1150 if (up
->pcrlen
== 0) { /* full coverage was set */
1151 LIMIT_NETDEBUG(KERN_WARNING
"UDPLITE: partial coverage "
1152 "%d while full coverage %d requested\n",
1153 UDP_SKB_CB(skb
)->cscov
, skb
->len
);
1156 /* The next case involves violating the min. coverage requested
1157 * by the receiver. This is subtle: if receiver wants x and x is
1158 * greater than the buffersize/MTU then receiver will complain
1159 * that it wants x while sender emits packets of smaller size y.
1160 * Therefore the above ...()->partial_cov statement is essential.
1162 if (UDP_SKB_CB(skb
)->cscov
< up
->pcrlen
) {
1163 LIMIT_NETDEBUG(KERN_WARNING
1164 "UDPLITE: coverage %d too small, need min %d\n",
1165 UDP_SKB_CB(skb
)->cscov
, up
->pcrlen
);
1170 if (sk
->sk_filter
) {
1171 if (udp_lib_checksum_complete(skb
))
1178 if (!sock_owned_by_user(sk
))
1179 rc
= __udp_queue_rcv_skb(sk
, skb
);
1181 sk_add_backlog(sk
, skb
);
1187 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1188 atomic_inc(&sk
->sk_drops
);
1194 * Multicasts and broadcasts go to each listener.
1196 * Note: called only from the BH handler context,
1197 * so we don't need to lock the hashes.
1199 static int __udp4_lib_mcast_deliver(struct net
*net
, struct sk_buff
*skb
,
1201 __be32 saddr
, __be32 daddr
,
1202 struct udp_table
*udptable
)
1205 struct udp_hslot
*hslot
= udp_hashslot(udptable
, net
, ntohs(uh
->dest
));
1208 spin_lock(&hslot
->lock
);
1209 sk
= sk_nulls_head(&hslot
->head
);
1210 dif
= skb
->dev
->ifindex
;
1211 sk
= udp_v4_mcast_next(net
, sk
, uh
->dest
, daddr
, uh
->source
, saddr
, dif
);
1213 struct sock
*sknext
= NULL
;
1216 struct sk_buff
*skb1
= skb
;
1218 sknext
= udp_v4_mcast_next(net
, sk_nulls_next(sk
), uh
->dest
,
1219 daddr
, uh
->source
, saddr
,
1222 skb1
= skb_clone(skb
, GFP_ATOMIC
);
1225 int ret
= udp_queue_rcv_skb(sk
, skb1
);
1227 /* we should probably re-process instead
1228 * of dropping packets here. */
1235 spin_unlock(&hslot
->lock
);
1239 /* Initialize UDP checksum. If exited with zero value (success),
1240 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1241 * Otherwise, csum completion requires chacksumming packet body,
1242 * including udp header and folding it to skb->csum.
1244 static inline int udp4_csum_init(struct sk_buff
*skb
, struct udphdr
*uh
,
1247 const struct iphdr
*iph
;
1250 UDP_SKB_CB(skb
)->partial_cov
= 0;
1251 UDP_SKB_CB(skb
)->cscov
= skb
->len
;
1253 if (proto
== IPPROTO_UDPLITE
) {
1254 err
= udplite_checksum_init(skb
, uh
);
1260 if (uh
->check
== 0) {
1261 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1262 } else if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
1263 if (!csum_tcpudp_magic(iph
->saddr
, iph
->daddr
, skb
->len
,
1265 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1267 if (!skb_csum_unnecessary(skb
))
1268 skb
->csum
= csum_tcpudp_nofold(iph
->saddr
, iph
->daddr
,
1269 skb
->len
, proto
, 0);
1270 /* Probably, we should checksum udp header (it should be in cache
1271 * in any case) and data in tiny packets (< rx copybreak).
1278 * All we need to do is get the socket, and then do a checksum.
1281 int __udp4_lib_rcv(struct sk_buff
*skb
, struct udp_table
*udptable
,
1286 unsigned short ulen
;
1287 struct rtable
*rt
= skb_rtable(skb
);
1288 __be32 saddr
, daddr
;
1289 struct net
*net
= dev_net(skb
->dev
);
1292 * Validate the packet.
1294 if (!pskb_may_pull(skb
, sizeof(struct udphdr
)))
1295 goto drop
; /* No space for header. */
1298 ulen
= ntohs(uh
->len
);
1299 if (ulen
> skb
->len
)
1302 if (proto
== IPPROTO_UDP
) {
1303 /* UDP validates ulen. */
1304 if (ulen
< sizeof(*uh
) || pskb_trim_rcsum(skb
, ulen
))
1309 if (udp4_csum_init(skb
, uh
, proto
))
1312 saddr
= ip_hdr(skb
)->saddr
;
1313 daddr
= ip_hdr(skb
)->daddr
;
1315 if (rt
->rt_flags
& (RTCF_BROADCAST
|RTCF_MULTICAST
))
1316 return __udp4_lib_mcast_deliver(net
, skb
, uh
,
1317 saddr
, daddr
, udptable
);
1319 sk
= __udp4_lib_lookup_skb(skb
, uh
->source
, uh
->dest
, udptable
);
1322 int ret
= udp_queue_rcv_skb(sk
, skb
);
1325 /* a return value > 0 means to resubmit the input, but
1326 * it wants the return to be -protocol, or 0
1333 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
1337 /* No socket. Drop packet silently, if checksum is wrong */
1338 if (udp_lib_checksum_complete(skb
))
1341 UDP_INC_STATS_BH(net
, UDP_MIB_NOPORTS
, proto
== IPPROTO_UDPLITE
);
1342 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_PORT_UNREACH
, 0);
1345 * Hmm. We got an UDP packet to a port to which we
1346 * don't wanna listen. Ignore it.
1352 LIMIT_NETDEBUG(KERN_DEBUG
"UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1353 proto
== IPPROTO_UDPLITE
? "-Lite" : "",
1364 * RFC1122: OK. Discards the bad packet silently (as far as
1365 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1367 LIMIT_NETDEBUG(KERN_DEBUG
"UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1368 proto
== IPPROTO_UDPLITE
? "-Lite" : "",
1375 UDP_INC_STATS_BH(net
, UDP_MIB_INERRORS
, proto
== IPPROTO_UDPLITE
);
1380 int udp_rcv(struct sk_buff
*skb
)
1382 return __udp4_lib_rcv(skb
, &udp_table
, IPPROTO_UDP
);
1385 void udp_destroy_sock(struct sock
*sk
)
1388 udp_flush_pending_frames(sk
);
1393 * Socket option code for UDP
1395 int udp_lib_setsockopt(struct sock
*sk
, int level
, int optname
,
1396 char __user
*optval
, unsigned int optlen
,
1397 int (*push_pending_frames
)(struct sock
*))
1399 struct udp_sock
*up
= udp_sk(sk
);
1402 int is_udplite
= IS_UDPLITE(sk
);
1404 if (optlen
< sizeof(int))
1407 if (get_user(val
, (int __user
*)optval
))
1417 (*push_pending_frames
)(sk
);
1425 case UDP_ENCAP_ESPINUDP
:
1426 case UDP_ENCAP_ESPINUDP_NON_IKE
:
1427 up
->encap_rcv
= xfrm4_udp_encap_rcv
;
1429 case UDP_ENCAP_L2TPINUDP
:
1430 up
->encap_type
= val
;
1439 * UDP-Lite's partial checksum coverage (RFC 3828).
1441 /* The sender sets actual checksum coverage length via this option.
1442 * The case coverage > packet length is handled by send module. */
1443 case UDPLITE_SEND_CSCOV
:
1444 if (!is_udplite
) /* Disable the option on UDP sockets */
1445 return -ENOPROTOOPT
;
1446 if (val
!= 0 && val
< 8) /* Illegal coverage: use default (8) */
1448 else if (val
> USHORT_MAX
)
1451 up
->pcflag
|= UDPLITE_SEND_CC
;
1454 /* The receiver specifies a minimum checksum coverage value. To make
1455 * sense, this should be set to at least 8 (as done below). If zero is
1456 * used, this again means full checksum coverage. */
1457 case UDPLITE_RECV_CSCOV
:
1458 if (!is_udplite
) /* Disable the option on UDP sockets */
1459 return -ENOPROTOOPT
;
1460 if (val
!= 0 && val
< 8) /* Avoid silly minimal values. */
1462 else if (val
> USHORT_MAX
)
1465 up
->pcflag
|= UDPLITE_RECV_CC
;
1475 EXPORT_SYMBOL(udp_lib_setsockopt
);
1477 int udp_setsockopt(struct sock
*sk
, int level
, int optname
,
1478 char __user
*optval
, unsigned int optlen
)
1480 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1481 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1482 udp_push_pending_frames
);
1483 return ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
1486 #ifdef CONFIG_COMPAT
1487 int compat_udp_setsockopt(struct sock
*sk
, int level
, int optname
,
1488 char __user
*optval
, unsigned int optlen
)
1490 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1491 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1492 udp_push_pending_frames
);
1493 return compat_ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
1497 int udp_lib_getsockopt(struct sock
*sk
, int level
, int optname
,
1498 char __user
*optval
, int __user
*optlen
)
1500 struct udp_sock
*up
= udp_sk(sk
);
1503 if (get_user(len
, optlen
))
1506 len
= min_t(unsigned int, len
, sizeof(int));
1517 val
= up
->encap_type
;
1520 /* The following two cannot be changed on UDP sockets, the return is
1521 * always 0 (which corresponds to the full checksum coverage of UDP). */
1522 case UDPLITE_SEND_CSCOV
:
1526 case UDPLITE_RECV_CSCOV
:
1531 return -ENOPROTOOPT
;
1534 if (put_user(len
, optlen
))
1536 if (copy_to_user(optval
, &val
, len
))
1540 EXPORT_SYMBOL(udp_lib_getsockopt
);
1542 int udp_getsockopt(struct sock
*sk
, int level
, int optname
,
1543 char __user
*optval
, int __user
*optlen
)
1545 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1546 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1547 return ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
1550 #ifdef CONFIG_COMPAT
1551 int compat_udp_getsockopt(struct sock
*sk
, int level
, int optname
,
1552 char __user
*optval
, int __user
*optlen
)
1554 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1555 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1556 return compat_ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
1560 * udp_poll - wait for a UDP event.
1561 * @file - file struct
1563 * @wait - poll table
1565 * This is same as datagram poll, except for the special case of
1566 * blocking sockets. If application is using a blocking fd
1567 * and a packet with checksum error is in the queue;
1568 * then it could get return from select indicating data available
1569 * but then block when reading it. Add special case code
1570 * to work around these arguably broken applications.
1572 unsigned int udp_poll(struct file
*file
, struct socket
*sock
, poll_table
*wait
)
1574 unsigned int mask
= datagram_poll(file
, sock
, wait
);
1575 struct sock
*sk
= sock
->sk
;
1577 /* Check for false positives due to checksum errors */
1578 if ((mask
& POLLRDNORM
) && !(file
->f_flags
& O_NONBLOCK
) &&
1579 !(sk
->sk_shutdown
& RCV_SHUTDOWN
) && !first_packet_length(sk
))
1580 mask
&= ~(POLLIN
| POLLRDNORM
);
1585 EXPORT_SYMBOL(udp_poll
);
1587 struct proto udp_prot
= {
1589 .owner
= THIS_MODULE
,
1590 .close
= udp_lib_close
,
1591 .connect
= ip4_datagram_connect
,
1592 .disconnect
= udp_disconnect
,
1594 .destroy
= udp_destroy_sock
,
1595 .setsockopt
= udp_setsockopt
,
1596 .getsockopt
= udp_getsockopt
,
1597 .sendmsg
= udp_sendmsg
,
1598 .recvmsg
= udp_recvmsg
,
1599 .sendpage
= udp_sendpage
,
1600 .backlog_rcv
= __udp_queue_rcv_skb
,
1601 .hash
= udp_lib_hash
,
1602 .unhash
= udp_lib_unhash
,
1603 .get_port
= udp_v4_get_port
,
1604 .memory_allocated
= &udp_memory_allocated
,
1605 .sysctl_mem
= sysctl_udp_mem
,
1606 .sysctl_wmem
= &sysctl_udp_wmem_min
,
1607 .sysctl_rmem
= &sysctl_udp_rmem_min
,
1608 .obj_size
= sizeof(struct udp_sock
),
1609 .slab_flags
= SLAB_DESTROY_BY_RCU
,
1610 .h
.udp_table
= &udp_table
,
1611 #ifdef CONFIG_COMPAT
1612 .compat_setsockopt
= compat_udp_setsockopt
,
1613 .compat_getsockopt
= compat_udp_getsockopt
,
1616 EXPORT_SYMBOL(udp_prot
);
1618 /* ------------------------------------------------------------------------ */
1619 #ifdef CONFIG_PROC_FS
1621 static struct sock
*udp_get_first(struct seq_file
*seq
, int start
)
1624 struct udp_iter_state
*state
= seq
->private;
1625 struct net
*net
= seq_file_net(seq
);
1627 for (state
->bucket
= start
; state
->bucket
<= state
->udp_table
->mask
;
1629 struct hlist_nulls_node
*node
;
1630 struct udp_hslot
*hslot
= &state
->udp_table
->hash
[state
->bucket
];
1632 if (hlist_nulls_empty(&hslot
->head
))
1635 spin_lock_bh(&hslot
->lock
);
1636 sk_nulls_for_each(sk
, node
, &hslot
->head
) {
1637 if (!net_eq(sock_net(sk
), net
))
1639 if (sk
->sk_family
== state
->family
)
1642 spin_unlock_bh(&hslot
->lock
);
1649 static struct sock
*udp_get_next(struct seq_file
*seq
, struct sock
*sk
)
1651 struct udp_iter_state
*state
= seq
->private;
1652 struct net
*net
= seq_file_net(seq
);
1655 sk
= sk_nulls_next(sk
);
1656 } while (sk
&& (!net_eq(sock_net(sk
), net
) || sk
->sk_family
!= state
->family
));
1659 if (state
->bucket
<= state
->udp_table
->mask
)
1660 spin_unlock_bh(&state
->udp_table
->hash
[state
->bucket
].lock
);
1661 return udp_get_first(seq
, state
->bucket
+ 1);
1666 static struct sock
*udp_get_idx(struct seq_file
*seq
, loff_t pos
)
1668 struct sock
*sk
= udp_get_first(seq
, 0);
1671 while (pos
&& (sk
= udp_get_next(seq
, sk
)) != NULL
)
1673 return pos
? NULL
: sk
;
1676 static void *udp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1678 struct udp_iter_state
*state
= seq
->private;
1679 state
->bucket
= MAX_UDP_PORTS
;
1681 return *pos
? udp_get_idx(seq
, *pos
-1) : SEQ_START_TOKEN
;
1684 static void *udp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1688 if (v
== SEQ_START_TOKEN
)
1689 sk
= udp_get_idx(seq
, 0);
1691 sk
= udp_get_next(seq
, v
);
1697 static void udp_seq_stop(struct seq_file
*seq
, void *v
)
1699 struct udp_iter_state
*state
= seq
->private;
1701 if (state
->bucket
<= state
->udp_table
->mask
)
1702 spin_unlock_bh(&state
->udp_table
->hash
[state
->bucket
].lock
);
1705 static int udp_seq_open(struct inode
*inode
, struct file
*file
)
1707 struct udp_seq_afinfo
*afinfo
= PDE(inode
)->data
;
1708 struct udp_iter_state
*s
;
1711 err
= seq_open_net(inode
, file
, &afinfo
->seq_ops
,
1712 sizeof(struct udp_iter_state
));
1716 s
= ((struct seq_file
*)file
->private_data
)->private;
1717 s
->family
= afinfo
->family
;
1718 s
->udp_table
= afinfo
->udp_table
;
1722 /* ------------------------------------------------------------------------ */
1723 int udp_proc_register(struct net
*net
, struct udp_seq_afinfo
*afinfo
)
1725 struct proc_dir_entry
*p
;
1728 afinfo
->seq_fops
.open
= udp_seq_open
;
1729 afinfo
->seq_fops
.read
= seq_read
;
1730 afinfo
->seq_fops
.llseek
= seq_lseek
;
1731 afinfo
->seq_fops
.release
= seq_release_net
;
1733 afinfo
->seq_ops
.start
= udp_seq_start
;
1734 afinfo
->seq_ops
.next
= udp_seq_next
;
1735 afinfo
->seq_ops
.stop
= udp_seq_stop
;
1737 p
= proc_create_data(afinfo
->name
, S_IRUGO
, net
->proc_net
,
1738 &afinfo
->seq_fops
, afinfo
);
1743 EXPORT_SYMBOL(udp_proc_register
);
1745 void udp_proc_unregister(struct net
*net
, struct udp_seq_afinfo
*afinfo
)
1747 proc_net_remove(net
, afinfo
->name
);
1749 EXPORT_SYMBOL(udp_proc_unregister
);
1751 /* ------------------------------------------------------------------------ */
1752 static void udp4_format_sock(struct sock
*sp
, struct seq_file
*f
,
1753 int bucket
, int *len
)
1755 struct inet_sock
*inet
= inet_sk(sp
);
1756 __be32 dest
= inet
->inet_daddr
;
1757 __be32 src
= inet
->inet_rcv_saddr
;
1758 __u16 destp
= ntohs(inet
->inet_dport
);
1759 __u16 srcp
= ntohs(inet
->inet_sport
);
1761 seq_printf(f
, "%5d: %08X:%04X %08X:%04X"
1762 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n",
1763 bucket
, src
, srcp
, dest
, destp
, sp
->sk_state
,
1764 sk_wmem_alloc_get(sp
),
1765 sk_rmem_alloc_get(sp
),
1766 0, 0L, 0, sock_i_uid(sp
), 0, sock_i_ino(sp
),
1767 atomic_read(&sp
->sk_refcnt
), sp
,
1768 atomic_read(&sp
->sk_drops
), len
);
1771 int udp4_seq_show(struct seq_file
*seq
, void *v
)
1773 if (v
== SEQ_START_TOKEN
)
1774 seq_printf(seq
, "%-127s\n",
1775 " sl local_address rem_address st tx_queue "
1776 "rx_queue tr tm->when retrnsmt uid timeout "
1777 "inode ref pointer drops");
1779 struct udp_iter_state
*state
= seq
->private;
1782 udp4_format_sock(v
, seq
, state
->bucket
, &len
);
1783 seq_printf(seq
, "%*s\n", 127 - len
, "");
1788 /* ------------------------------------------------------------------------ */
1789 static struct udp_seq_afinfo udp4_seq_afinfo
= {
1792 .udp_table
= &udp_table
,
1794 .owner
= THIS_MODULE
,
1797 .show
= udp4_seq_show
,
1801 static int udp4_proc_init_net(struct net
*net
)
1803 return udp_proc_register(net
, &udp4_seq_afinfo
);
1806 static void udp4_proc_exit_net(struct net
*net
)
1808 udp_proc_unregister(net
, &udp4_seq_afinfo
);
1811 static struct pernet_operations udp4_net_ops
= {
1812 .init
= udp4_proc_init_net
,
1813 .exit
= udp4_proc_exit_net
,
1816 int __init
udp4_proc_init(void)
1818 return register_pernet_subsys(&udp4_net_ops
);
1821 void udp4_proc_exit(void)
1823 unregister_pernet_subsys(&udp4_net_ops
);
1825 #endif /* CONFIG_PROC_FS */
1827 static __initdata
unsigned long uhash_entries
;
1828 static int __init
set_uhash_entries(char *str
)
1832 uhash_entries
= simple_strtoul(str
, &str
, 0);
1833 if (uhash_entries
&& uhash_entries
< UDP_HTABLE_SIZE_MIN
)
1834 uhash_entries
= UDP_HTABLE_SIZE_MIN
;
1837 __setup("uhash_entries=", set_uhash_entries
);
1839 void __init
udp_table_init(struct udp_table
*table
, const char *name
)
1843 if (!CONFIG_BASE_SMALL
)
1844 table
->hash
= alloc_large_system_hash(name
,
1845 sizeof(struct udp_hslot
),
1847 21, /* one slot per 2 MB */
1853 * Make sure hash table has the minimum size
1855 if (CONFIG_BASE_SMALL
|| table
->mask
< UDP_HTABLE_SIZE_MIN
- 1) {
1856 table
->hash
= kmalloc(UDP_HTABLE_SIZE_MIN
*
1857 sizeof(struct udp_hslot
), GFP_KERNEL
);
1860 table
->log
= ilog2(UDP_HTABLE_SIZE_MIN
);
1861 table
->mask
= UDP_HTABLE_SIZE_MIN
- 1;
1863 for (i
= 0; i
<= table
->mask
; i
++) {
1864 INIT_HLIST_NULLS_HEAD(&table
->hash
[i
].head
, i
);
1865 spin_lock_init(&table
->hash
[i
].lock
);
1869 void __init
udp_init(void)
1871 unsigned long nr_pages
, limit
;
1873 udp_table_init(&udp_table
, "UDP");
1874 /* Set the pressure threshold up by the same strategy of TCP. It is a
1875 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
1876 * toward zero with the amount of memory, with a floor of 128 pages.
1878 nr_pages
= totalram_pages
- totalhigh_pages
;
1879 limit
= min(nr_pages
, 1UL<<(28-PAGE_SHIFT
)) >> (20-PAGE_SHIFT
);
1880 limit
= (limit
* (nr_pages
>> (20-PAGE_SHIFT
))) >> (PAGE_SHIFT
-11);
1881 limit
= max(limit
, 128UL);
1882 sysctl_udp_mem
[0] = limit
/ 4 * 3;
1883 sysctl_udp_mem
[1] = limit
;
1884 sysctl_udp_mem
[2] = sysctl_udp_mem
[0] * 2;
1886 sysctl_udp_rmem_min
= SK_MEM_QUANTUM
;
1887 sysctl_udp_wmem_min
= SK_MEM_QUANTUM
;
1890 int udp4_ufo_send_check(struct sk_buff
*skb
)
1892 const struct iphdr
*iph
;
1895 if (!pskb_may_pull(skb
, sizeof(*uh
)))
1901 uh
->check
= ~csum_tcpudp_magic(iph
->saddr
, iph
->daddr
, skb
->len
,
1903 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
1904 skb
->csum_offset
= offsetof(struct udphdr
, check
);
1905 skb
->ip_summed
= CHECKSUM_PARTIAL
;
1909 struct sk_buff
*udp4_ufo_fragment(struct sk_buff
*skb
, int features
)
1911 struct sk_buff
*segs
= ERR_PTR(-EINVAL
);
1916 mss
= skb_shinfo(skb
)->gso_size
;
1917 if (unlikely(skb
->len
<= mss
))
1920 if (skb_gso_ok(skb
, features
| NETIF_F_GSO_ROBUST
)) {
1921 /* Packet is from an untrusted source, reset gso_segs. */
1922 int type
= skb_shinfo(skb
)->gso_type
;
1924 if (unlikely(type
& ~(SKB_GSO_UDP
| SKB_GSO_DODGY
) ||
1925 !(type
& (SKB_GSO_UDP
))))
1928 skb_shinfo(skb
)->gso_segs
= DIV_ROUND_UP(skb
->len
, mss
);
1934 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
1935 * do checksum of UDP packets sent as multiple IP fragments.
1937 offset
= skb
->csum_start
- skb_headroom(skb
);
1938 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1939 offset
+= skb
->csum_offset
;
1940 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1941 skb
->ip_summed
= CHECKSUM_NONE
;
1943 /* Fragment the skb. IP headers of the fragments are updated in
1944 * inet_gso_segment()
1946 segs
= skb_segment(skb
, features
);