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
|| udp_sk(sk2
)->udp_port_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(udp_sk(sk2
)->udp_port_hash
>> log
,
156 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
158 * @sk: socket struct in question
159 * @snum: port number to look up
160 * @saddr_comp: AF-dependent comparison of bound local IP addresses
162 int udp_lib_get_port(struct sock
*sk
, unsigned short snum
,
163 int (*saddr_comp
)(const struct sock
*sk1
,
164 const struct sock
*sk2
))
166 struct udp_hslot
*hslot
, *hslot2
;
167 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
169 struct net
*net
= sock_net(sk
);
172 int low
, high
, remaining
;
174 unsigned short first
, last
;
175 DECLARE_BITMAP(bitmap
, PORTS_PER_CHAIN
);
177 inet_get_local_port_range(&low
, &high
);
178 remaining
= (high
- low
) + 1;
181 first
= (((u64
)rand
* remaining
) >> 32) + low
;
183 * force rand to be an odd multiple of UDP_HTABLE_SIZE
185 rand
= (rand
| 1) * (udptable
->mask
+ 1);
186 for (last
= first
+ udptable
->mask
+ 1;
189 hslot
= udp_hashslot(udptable
, net
, first
);
190 bitmap_zero(bitmap
, PORTS_PER_CHAIN
);
191 spin_lock_bh(&hslot
->lock
);
192 udp_lib_lport_inuse(net
, snum
, hslot
, bitmap
, sk
,
193 saddr_comp
, udptable
->log
);
197 * Iterate on all possible values of snum for this hash.
198 * Using steps of an odd multiple of UDP_HTABLE_SIZE
199 * give us randomization and full range coverage.
202 if (low
<= snum
&& snum
<= high
&&
203 !test_bit(snum
>> udptable
->log
, bitmap
))
206 } while (snum
!= first
);
207 spin_unlock_bh(&hslot
->lock
);
211 hslot
= udp_hashslot(udptable
, net
, snum
);
212 spin_lock_bh(&hslot
->lock
);
213 if (udp_lib_lport_inuse(net
, snum
, hslot
, NULL
, sk
,
218 inet_sk(sk
)->inet_num
= snum
;
219 udp_sk(sk
)->udp_port_hash
= snum
;
220 udp_sk(sk
)->udp_portaddr_hash
^= snum
;
221 if (sk_unhashed(sk
)) {
222 sk_nulls_add_node_rcu(sk
, &hslot
->head
);
224 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, 1);
226 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
227 spin_lock(&hslot2
->lock
);
228 hlist_nulls_add_head_rcu(&udp_sk(sk
)->udp_portaddr_node
,
231 spin_unlock(&hslot2
->lock
);
235 spin_unlock_bh(&hslot
->lock
);
239 EXPORT_SYMBOL(udp_lib_get_port
);
241 static int ipv4_rcv_saddr_equal(const struct sock
*sk1
, const struct sock
*sk2
)
243 struct inet_sock
*inet1
= inet_sk(sk1
), *inet2
= inet_sk(sk2
);
245 return (!ipv6_only_sock(sk2
) &&
246 (!inet1
->inet_rcv_saddr
|| !inet2
->inet_rcv_saddr
||
247 inet1
->inet_rcv_saddr
== inet2
->inet_rcv_saddr
));
250 static unsigned int udp4_portaddr_hash(struct net
*net
, __be32 saddr
,
253 return jhash_1word(saddr
, net_hash_mix(net
)) ^ port
;
256 int udp_v4_get_port(struct sock
*sk
, unsigned short snum
)
258 /* precompute partial secondary hash */
259 udp_sk(sk
)->udp_portaddr_hash
=
260 udp4_portaddr_hash(sock_net(sk
),
261 inet_sk(sk
)->inet_rcv_saddr
,
263 return udp_lib_get_port(sk
, snum
, ipv4_rcv_saddr_equal
);
266 static inline int compute_score(struct sock
*sk
, struct net
*net
, __be32 saddr
,
268 __be16 sport
, __be32 daddr
, __be16 dport
, int dif
)
272 if (net_eq(sock_net(sk
), net
) && udp_sk(sk
)->udp_port_hash
== hnum
&&
273 !ipv6_only_sock(sk
)) {
274 struct inet_sock
*inet
= inet_sk(sk
);
276 score
= (sk
->sk_family
== PF_INET
? 1 : 0);
277 if (inet
->inet_rcv_saddr
) {
278 if (inet
->inet_rcv_saddr
!= daddr
)
282 if (inet
->inet_daddr
) {
283 if (inet
->inet_daddr
!= saddr
)
287 if (inet
->inet_dport
) {
288 if (inet
->inet_dport
!= sport
)
292 if (sk
->sk_bound_dev_if
) {
293 if (sk
->sk_bound_dev_if
!= dif
)
301 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
302 * harder than this. -DaveM
304 static struct sock
*__udp4_lib_lookup(struct net
*net
, __be32 saddr
,
305 __be16 sport
, __be32 daddr
, __be16 dport
,
306 int dif
, struct udp_table
*udptable
)
308 struct sock
*sk
, *result
;
309 struct hlist_nulls_node
*node
;
310 unsigned short hnum
= ntohs(dport
);
311 unsigned int hash
= udp_hashfn(net
, hnum
, udptable
->mask
);
312 struct udp_hslot
*hslot
= &udptable
->hash
[hash
];
319 sk_nulls_for_each_rcu(sk
, node
, &hslot
->head
) {
320 score
= compute_score(sk
, net
, saddr
, hnum
, sport
,
322 if (score
> badness
) {
328 * if the nulls value we got at the end of this lookup is
329 * not the expected one, we must restart lookup.
330 * We probably met an item that was moved to another chain.
332 if (get_nulls_value(node
) != hash
)
336 if (unlikely(!atomic_inc_not_zero(&result
->sk_refcnt
)))
338 else if (unlikely(compute_score(result
, net
, saddr
, hnum
, sport
,
339 daddr
, dport
, dif
) < badness
)) {
348 static inline struct sock
*__udp4_lib_lookup_skb(struct sk_buff
*skb
,
349 __be16 sport
, __be16 dport
,
350 struct udp_table
*udptable
)
353 const struct iphdr
*iph
= ip_hdr(skb
);
355 if (unlikely(sk
= skb_steal_sock(skb
)))
358 return __udp4_lib_lookup(dev_net(skb_dst(skb
)->dev
), iph
->saddr
, sport
,
359 iph
->daddr
, dport
, inet_iif(skb
),
363 struct sock
*udp4_lib_lookup(struct net
*net
, __be32 saddr
, __be16 sport
,
364 __be32 daddr
, __be16 dport
, int dif
)
366 return __udp4_lib_lookup(net
, saddr
, sport
, daddr
, dport
, dif
, &udp_table
);
368 EXPORT_SYMBOL_GPL(udp4_lib_lookup
);
370 static inline struct sock
*udp_v4_mcast_next(struct net
*net
, struct sock
*sk
,
371 __be16 loc_port
, __be32 loc_addr
,
372 __be16 rmt_port
, __be32 rmt_addr
,
375 struct hlist_nulls_node
*node
;
377 unsigned short hnum
= ntohs(loc_port
);
379 sk_nulls_for_each_from(s
, node
) {
380 struct inet_sock
*inet
= inet_sk(s
);
382 if (!net_eq(sock_net(s
), net
) ||
383 udp_sk(s
)->udp_port_hash
!= hnum
||
384 (inet
->inet_daddr
&& inet
->inet_daddr
!= rmt_addr
) ||
385 (inet
->inet_dport
!= rmt_port
&& inet
->inet_dport
) ||
386 (inet
->inet_rcv_saddr
&&
387 inet
->inet_rcv_saddr
!= loc_addr
) ||
389 (s
->sk_bound_dev_if
&& s
->sk_bound_dev_if
!= dif
))
391 if (!ip_mc_sf_allow(s
, loc_addr
, rmt_addr
, dif
))
401 * This routine is called by the ICMP module when it gets some
402 * sort of error condition. If err < 0 then the socket should
403 * be closed and the error returned to the user. If err > 0
404 * it's just the icmp type << 8 | icmp code.
405 * Header points to the ip header of the error packet. We move
406 * on past this. Then (as it used to claim before adjustment)
407 * header points to the first 8 bytes of the udp header. We need
408 * to find the appropriate port.
411 void __udp4_lib_err(struct sk_buff
*skb
, u32 info
, struct udp_table
*udptable
)
413 struct inet_sock
*inet
;
414 struct iphdr
*iph
= (struct iphdr
*)skb
->data
;
415 struct udphdr
*uh
= (struct udphdr
*)(skb
->data
+(iph
->ihl
<<2));
416 const int type
= icmp_hdr(skb
)->type
;
417 const int code
= icmp_hdr(skb
)->code
;
421 struct net
*net
= dev_net(skb
->dev
);
423 sk
= __udp4_lib_lookup(net
, iph
->daddr
, uh
->dest
,
424 iph
->saddr
, uh
->source
, skb
->dev
->ifindex
, udptable
);
426 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
427 return; /* No socket for error */
436 case ICMP_TIME_EXCEEDED
:
439 case ICMP_SOURCE_QUENCH
:
441 case ICMP_PARAMETERPROB
:
445 case ICMP_DEST_UNREACH
:
446 if (code
== ICMP_FRAG_NEEDED
) { /* Path MTU discovery */
447 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
) {
455 if (code
<= NR_ICMP_UNREACH
) {
456 harderr
= icmp_err_convert
[code
].fatal
;
457 err
= icmp_err_convert
[code
].errno
;
463 * RFC1122: OK. Passes ICMP errors back to application, as per
466 if (!inet
->recverr
) {
467 if (!harderr
|| sk
->sk_state
!= TCP_ESTABLISHED
)
470 ip_icmp_error(sk
, skb
, err
, uh
->dest
, info
, (u8
*)(uh
+1));
473 sk
->sk_error_report(sk
);
478 void udp_err(struct sk_buff
*skb
, u32 info
)
480 __udp4_lib_err(skb
, info
, &udp_table
);
484 * Throw away all pending data and cancel the corking. Socket is locked.
486 void udp_flush_pending_frames(struct sock
*sk
)
488 struct udp_sock
*up
= udp_sk(sk
);
493 ip_flush_pending_frames(sk
);
496 EXPORT_SYMBOL(udp_flush_pending_frames
);
499 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
500 * @sk: socket we are sending on
501 * @skb: sk_buff containing the filled-in UDP header
502 * (checksum field must be zeroed out)
504 static void udp4_hwcsum_outgoing(struct sock
*sk
, struct sk_buff
*skb
,
505 __be32 src
, __be32 dst
, int len
)
508 struct udphdr
*uh
= udp_hdr(skb
);
511 if (skb_queue_len(&sk
->sk_write_queue
) == 1) {
513 * Only one fragment on the socket.
515 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
516 skb
->csum_offset
= offsetof(struct udphdr
, check
);
517 uh
->check
= ~csum_tcpudp_magic(src
, dst
, len
, IPPROTO_UDP
, 0);
520 * HW-checksum won't work as there are two or more
521 * fragments on the socket so that all csums of sk_buffs
524 offset
= skb_transport_offset(skb
);
525 skb
->csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
527 skb
->ip_summed
= CHECKSUM_NONE
;
529 skb_queue_walk(&sk
->sk_write_queue
, skb
) {
530 csum
= csum_add(csum
, skb
->csum
);
533 uh
->check
= csum_tcpudp_magic(src
, dst
, len
, IPPROTO_UDP
, csum
);
535 uh
->check
= CSUM_MANGLED_0
;
540 * Push out all pending data as one UDP datagram. Socket is locked.
542 static int udp_push_pending_frames(struct sock
*sk
)
544 struct udp_sock
*up
= udp_sk(sk
);
545 struct inet_sock
*inet
= inet_sk(sk
);
546 struct flowi
*fl
= &inet
->cork
.fl
;
550 int is_udplite
= IS_UDPLITE(sk
);
553 /* Grab the skbuff where UDP header space exists. */
554 if ((skb
= skb_peek(&sk
->sk_write_queue
)) == NULL
)
558 * Create a UDP header
561 uh
->source
= fl
->fl_ip_sport
;
562 uh
->dest
= fl
->fl_ip_dport
;
563 uh
->len
= htons(up
->len
);
566 if (is_udplite
) /* UDP-Lite */
567 csum
= udplite_csum_outgoing(sk
, skb
);
569 else if (sk
->sk_no_check
== UDP_CSUM_NOXMIT
) { /* UDP csum disabled */
571 skb
->ip_summed
= CHECKSUM_NONE
;
574 } else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) { /* UDP hardware csum */
576 udp4_hwcsum_outgoing(sk
, skb
, fl
->fl4_src
, fl
->fl4_dst
, up
->len
);
579 } else /* `normal' UDP */
580 csum
= udp_csum_outgoing(sk
, skb
);
582 /* add protocol-dependent pseudo-header */
583 uh
->check
= csum_tcpudp_magic(fl
->fl4_src
, fl
->fl4_dst
, up
->len
,
584 sk
->sk_protocol
, csum
);
586 uh
->check
= CSUM_MANGLED_0
;
589 err
= ip_push_pending_frames(sk
);
591 if (err
== -ENOBUFS
&& !inet
->recverr
) {
592 UDP_INC_STATS_USER(sock_net(sk
),
593 UDP_MIB_SNDBUFERRORS
, is_udplite
);
597 UDP_INC_STATS_USER(sock_net(sk
),
598 UDP_MIB_OUTDATAGRAMS
, is_udplite
);
605 int udp_sendmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
608 struct inet_sock
*inet
= inet_sk(sk
);
609 struct udp_sock
*up
= udp_sk(sk
);
611 struct ipcm_cookie ipc
;
612 struct rtable
*rt
= NULL
;
615 __be32 daddr
, faddr
, saddr
;
618 int err
, is_udplite
= IS_UDPLITE(sk
);
619 int corkreq
= up
->corkflag
|| msg
->msg_flags
&MSG_MORE
;
620 int (*getfrag
)(void *, char *, int, int, int, struct sk_buff
*);
629 if (msg
->msg_flags
& MSG_OOB
) /* Mirror BSD error message compatibility */
637 * There are pending frames.
638 * The socket lock must be held while it's corked.
641 if (likely(up
->pending
)) {
642 if (unlikely(up
->pending
!= AF_INET
)) {
650 ulen
+= sizeof(struct udphdr
);
653 * Get and verify the address.
656 struct sockaddr_in
* usin
= (struct sockaddr_in
*)msg
->msg_name
;
657 if (msg
->msg_namelen
< sizeof(*usin
))
659 if (usin
->sin_family
!= AF_INET
) {
660 if (usin
->sin_family
!= AF_UNSPEC
)
661 return -EAFNOSUPPORT
;
664 daddr
= usin
->sin_addr
.s_addr
;
665 dport
= usin
->sin_port
;
669 if (sk
->sk_state
!= TCP_ESTABLISHED
)
670 return -EDESTADDRREQ
;
671 daddr
= inet
->inet_daddr
;
672 dport
= inet
->inet_dport
;
673 /* Open fast path for connected socket.
674 Route will not be used, if at least one option is set.
678 ipc
.addr
= inet
->inet_saddr
;
680 ipc
.oif
= sk
->sk_bound_dev_if
;
681 err
= sock_tx_timestamp(msg
, sk
, &ipc
.shtx
);
684 if (msg
->msg_controllen
) {
685 err
= ip_cmsg_send(sock_net(sk
), msg
, &ipc
);
696 ipc
.addr
= faddr
= daddr
;
698 if (ipc
.opt
&& ipc
.opt
->srr
) {
701 faddr
= ipc
.opt
->faddr
;
704 tos
= RT_TOS(inet
->tos
);
705 if (sock_flag(sk
, SOCK_LOCALROUTE
) ||
706 (msg
->msg_flags
& MSG_DONTROUTE
) ||
707 (ipc
.opt
&& ipc
.opt
->is_strictroute
)) {
712 if (ipv4_is_multicast(daddr
)) {
714 ipc
.oif
= inet
->mc_index
;
716 saddr
= inet
->mc_addr
;
721 rt
= (struct rtable
*)sk_dst_check(sk
, 0);
724 struct flowi fl
= { .oif
= ipc
.oif
,
730 .proto
= sk
->sk_protocol
,
731 .flags
= inet_sk_flowi_flags(sk
),
733 { .sport
= inet
->inet_sport
,
734 .dport
= dport
} } };
735 struct net
*net
= sock_net(sk
);
737 security_sk_classify_flow(sk
, &fl
);
738 err
= ip_route_output_flow(net
, &rt
, &fl
, sk
, 1);
740 if (err
== -ENETUNREACH
)
741 IP_INC_STATS_BH(net
, IPSTATS_MIB_OUTNOROUTES
);
746 if ((rt
->rt_flags
& RTCF_BROADCAST
) &&
747 !sock_flag(sk
, SOCK_BROADCAST
))
750 sk_dst_set(sk
, dst_clone(&rt
->u
.dst
));
753 if (msg
->msg_flags
&MSG_CONFIRM
)
759 daddr
= ipc
.addr
= rt
->rt_dst
;
762 if (unlikely(up
->pending
)) {
763 /* The socket is already corked while preparing it. */
764 /* ... which is an evident application bug. --ANK */
767 LIMIT_NETDEBUG(KERN_DEBUG
"udp cork app bug 2\n");
772 * Now cork the socket to pend data.
774 inet
->cork
.fl
.fl4_dst
= daddr
;
775 inet
->cork
.fl
.fl_ip_dport
= dport
;
776 inet
->cork
.fl
.fl4_src
= saddr
;
777 inet
->cork
.fl
.fl_ip_sport
= inet
->inet_sport
;
778 up
->pending
= AF_INET
;
782 getfrag
= is_udplite
? udplite_getfrag
: ip_generic_getfrag
;
783 err
= ip_append_data(sk
, getfrag
, msg
->msg_iov
, ulen
,
784 sizeof(struct udphdr
), &ipc
, &rt
,
785 corkreq
? msg
->msg_flags
|MSG_MORE
: msg
->msg_flags
);
787 udp_flush_pending_frames(sk
);
789 err
= udp_push_pending_frames(sk
);
790 else if (unlikely(skb_queue_empty(&sk
->sk_write_queue
)))
801 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
802 * ENOBUFS might not be good (it's not tunable per se), but otherwise
803 * we don't have a good statistic (IpOutDiscards but it can be too many
804 * things). We could add another new stat but at least for now that
805 * seems like overkill.
807 if (err
== -ENOBUFS
|| test_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
)) {
808 UDP_INC_STATS_USER(sock_net(sk
),
809 UDP_MIB_SNDBUFERRORS
, is_udplite
);
814 dst_confirm(&rt
->u
.dst
);
815 if (!(msg
->msg_flags
&MSG_PROBE
) || len
)
816 goto back_from_confirm
;
820 EXPORT_SYMBOL(udp_sendmsg
);
822 int udp_sendpage(struct sock
*sk
, struct page
*page
, int offset
,
823 size_t size
, int flags
)
825 struct udp_sock
*up
= udp_sk(sk
);
829 struct msghdr msg
= { .msg_flags
= flags
|MSG_MORE
};
831 /* Call udp_sendmsg to specify destination address which
832 * sendpage interface can't pass.
833 * This will succeed only when the socket is connected.
835 ret
= udp_sendmsg(NULL
, sk
, &msg
, 0);
842 if (unlikely(!up
->pending
)) {
845 LIMIT_NETDEBUG(KERN_DEBUG
"udp cork app bug 3\n");
849 ret
= ip_append_page(sk
, page
, offset
, size
, flags
);
850 if (ret
== -EOPNOTSUPP
) {
852 return sock_no_sendpage(sk
->sk_socket
, page
, offset
,
856 udp_flush_pending_frames(sk
);
861 if (!(up
->corkflag
|| (flags
&MSG_MORE
)))
862 ret
= udp_push_pending_frames(sk
);
872 * first_packet_length - return length of first packet in receive queue
875 * Drops all bad checksum frames, until a valid one is found.
876 * Returns the length of found skb, or 0 if none is found.
878 static unsigned int first_packet_length(struct sock
*sk
)
880 struct sk_buff_head list_kill
, *rcvq
= &sk
->sk_receive_queue
;
884 __skb_queue_head_init(&list_kill
);
886 spin_lock_bh(&rcvq
->lock
);
887 while ((skb
= skb_peek(rcvq
)) != NULL
&&
888 udp_lib_checksum_complete(skb
)) {
889 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
,
891 atomic_inc(&sk
->sk_drops
);
892 __skb_unlink(skb
, rcvq
);
893 __skb_queue_tail(&list_kill
, skb
);
895 res
= skb
? skb
->len
: 0;
896 spin_unlock_bh(&rcvq
->lock
);
898 if (!skb_queue_empty(&list_kill
)) {
900 __skb_queue_purge(&list_kill
);
901 sk_mem_reclaim_partial(sk
);
908 * IOCTL requests applicable to the UDP protocol
911 int udp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
)
916 int amount
= sk_wmem_alloc_get(sk
);
918 return put_user(amount
, (int __user
*)arg
);
923 unsigned int amount
= first_packet_length(sk
);
927 * We will only return the amount
928 * of this packet since that is all
931 amount
-= sizeof(struct udphdr
);
933 return put_user(amount
, (int __user
*)arg
);
942 EXPORT_SYMBOL(udp_ioctl
);
945 * This should be easy, if there is something there we
946 * return it, otherwise we block.
949 int udp_recvmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
950 size_t len
, int noblock
, int flags
, int *addr_len
)
952 struct inet_sock
*inet
= inet_sk(sk
);
953 struct sockaddr_in
*sin
= (struct sockaddr_in
*)msg
->msg_name
;
955 unsigned int ulen
, copied
;
958 int is_udplite
= IS_UDPLITE(sk
);
961 * Check any passed addresses
964 *addr_len
= sizeof(*sin
);
966 if (flags
& MSG_ERRQUEUE
)
967 return ip_recv_error(sk
, msg
, len
);
970 skb
= __skb_recv_datagram(sk
, flags
| (noblock
? MSG_DONTWAIT
: 0),
975 ulen
= skb
->len
- sizeof(struct udphdr
);
979 else if (copied
< ulen
)
980 msg
->msg_flags
|= MSG_TRUNC
;
983 * If checksum is needed at all, try to do it while copying the
984 * data. If the data is truncated, or if we only want a partial
985 * coverage checksum (UDP-Lite), do it before the copy.
988 if (copied
< ulen
|| UDP_SKB_CB(skb
)->partial_cov
) {
989 if (udp_lib_checksum_complete(skb
))
993 if (skb_csum_unnecessary(skb
))
994 err
= skb_copy_datagram_iovec(skb
, sizeof(struct udphdr
),
995 msg
->msg_iov
, copied
);
997 err
= skb_copy_and_csum_datagram_iovec(skb
,
998 sizeof(struct udphdr
),
1009 UDP_INC_STATS_USER(sock_net(sk
),
1010 UDP_MIB_INDATAGRAMS
, is_udplite
);
1012 sock_recv_ts_and_drops(msg
, sk
, skb
);
1014 /* Copy the address. */
1016 sin
->sin_family
= AF_INET
;
1017 sin
->sin_port
= udp_hdr(skb
)->source
;
1018 sin
->sin_addr
.s_addr
= ip_hdr(skb
)->saddr
;
1019 memset(sin
->sin_zero
, 0, sizeof(sin
->sin_zero
));
1021 if (inet
->cmsg_flags
)
1022 ip_cmsg_recv(msg
, skb
);
1025 if (flags
& MSG_TRUNC
)
1029 skb_free_datagram_locked(sk
, skb
);
1035 if (!skb_kill_datagram(sk
, skb
, flags
))
1036 UDP_INC_STATS_USER(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1045 int udp_disconnect(struct sock
*sk
, int flags
)
1047 struct inet_sock
*inet
= inet_sk(sk
);
1049 * 1003.1g - break association.
1052 sk
->sk_state
= TCP_CLOSE
;
1053 inet
->inet_daddr
= 0;
1054 inet
->inet_dport
= 0;
1055 sk
->sk_bound_dev_if
= 0;
1056 if (!(sk
->sk_userlocks
& SOCK_BINDADDR_LOCK
))
1057 inet_reset_saddr(sk
);
1059 if (!(sk
->sk_userlocks
& SOCK_BINDPORT_LOCK
)) {
1060 sk
->sk_prot
->unhash(sk
);
1061 inet
->inet_sport
= 0;
1066 EXPORT_SYMBOL(udp_disconnect
);
1068 void udp_lib_unhash(struct sock
*sk
)
1070 if (sk_hashed(sk
)) {
1071 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
1072 struct udp_hslot
*hslot
, *hslot2
;
1074 hslot
= udp_hashslot(udptable
, sock_net(sk
),
1075 udp_sk(sk
)->udp_port_hash
);
1076 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
1078 spin_lock_bh(&hslot
->lock
);
1079 if (sk_nulls_del_node_init_rcu(sk
)) {
1081 inet_sk(sk
)->inet_num
= 0;
1082 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, -1);
1084 spin_lock(&hslot2
->lock
);
1085 hlist_nulls_del_init_rcu(&udp_sk(sk
)->udp_portaddr_node
);
1087 spin_unlock(&hslot2
->lock
);
1089 spin_unlock_bh(&hslot
->lock
);
1092 EXPORT_SYMBOL(udp_lib_unhash
);
1094 static int __udp_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
1096 int rc
= sock_queue_rcv_skb(sk
, skb
);
1099 int is_udplite
= IS_UDPLITE(sk
);
1101 /* Note that an ENOMEM error is charged twice */
1103 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_RCVBUFERRORS
,
1105 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1117 * >0: "udp encap" protocol resubmission
1119 * Note that in the success and error cases, the skb is assumed to
1120 * have either been requeued or freed.
1122 int udp_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
1124 struct udp_sock
*up
= udp_sk(sk
);
1126 int is_udplite
= IS_UDPLITE(sk
);
1129 * Charge it to the socket, dropping if the queue is full.
1131 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1135 if (up
->encap_type
) {
1137 * This is an encapsulation socket so pass the skb to
1138 * the socket's udp_encap_rcv() hook. Otherwise, just
1139 * fall through and pass this up the UDP socket.
1140 * up->encap_rcv() returns the following value:
1141 * =0 if skb was successfully passed to the encap
1142 * handler or was discarded by it.
1143 * >0 if skb should be passed on to UDP.
1144 * <0 if skb should be resubmitted as proto -N
1147 /* if we're overly short, let UDP handle it */
1148 if (skb
->len
> sizeof(struct udphdr
) &&
1149 up
->encap_rcv
!= NULL
) {
1152 ret
= (*up
->encap_rcv
)(sk
, skb
);
1154 UDP_INC_STATS_BH(sock_net(sk
),
1155 UDP_MIB_INDATAGRAMS
,
1161 /* FALLTHROUGH -- it's a UDP Packet */
1165 * UDP-Lite specific tests, ignored on UDP sockets
1167 if ((is_udplite
& UDPLITE_RECV_CC
) && UDP_SKB_CB(skb
)->partial_cov
) {
1170 * MIB statistics other than incrementing the error count are
1171 * disabled for the following two types of errors: these depend
1172 * on the application settings, not on the functioning of the
1173 * protocol stack as such.
1175 * RFC 3828 here recommends (sec 3.3): "There should also be a
1176 * way ... to ... at least let the receiving application block
1177 * delivery of packets with coverage values less than a value
1178 * provided by the application."
1180 if (up
->pcrlen
== 0) { /* full coverage was set */
1181 LIMIT_NETDEBUG(KERN_WARNING
"UDPLITE: partial coverage "
1182 "%d while full coverage %d requested\n",
1183 UDP_SKB_CB(skb
)->cscov
, skb
->len
);
1186 /* The next case involves violating the min. coverage requested
1187 * by the receiver. This is subtle: if receiver wants x and x is
1188 * greater than the buffersize/MTU then receiver will complain
1189 * that it wants x while sender emits packets of smaller size y.
1190 * Therefore the above ...()->partial_cov statement is essential.
1192 if (UDP_SKB_CB(skb
)->cscov
< up
->pcrlen
) {
1193 LIMIT_NETDEBUG(KERN_WARNING
1194 "UDPLITE: coverage %d too small, need min %d\n",
1195 UDP_SKB_CB(skb
)->cscov
, up
->pcrlen
);
1200 if (sk
->sk_filter
) {
1201 if (udp_lib_checksum_complete(skb
))
1208 if (!sock_owned_by_user(sk
))
1209 rc
= __udp_queue_rcv_skb(sk
, skb
);
1211 sk_add_backlog(sk
, skb
);
1217 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1218 atomic_inc(&sk
->sk_drops
);
1224 * Multicasts and broadcasts go to each listener.
1226 * Note: called only from the BH handler context,
1227 * so we don't need to lock the hashes.
1229 static int __udp4_lib_mcast_deliver(struct net
*net
, struct sk_buff
*skb
,
1231 __be32 saddr
, __be32 daddr
,
1232 struct udp_table
*udptable
)
1235 struct udp_hslot
*hslot
= udp_hashslot(udptable
, net
, ntohs(uh
->dest
));
1238 spin_lock(&hslot
->lock
);
1239 sk
= sk_nulls_head(&hslot
->head
);
1240 dif
= skb
->dev
->ifindex
;
1241 sk
= udp_v4_mcast_next(net
, sk
, uh
->dest
, daddr
, uh
->source
, saddr
, dif
);
1243 struct sock
*sknext
= NULL
;
1246 struct sk_buff
*skb1
= skb
;
1248 sknext
= udp_v4_mcast_next(net
, sk_nulls_next(sk
), uh
->dest
,
1249 daddr
, uh
->source
, saddr
,
1252 skb1
= skb_clone(skb
, GFP_ATOMIC
);
1255 int ret
= udp_queue_rcv_skb(sk
, skb1
);
1257 /* we should probably re-process instead
1258 * of dropping packets here. */
1265 spin_unlock(&hslot
->lock
);
1269 /* Initialize UDP checksum. If exited with zero value (success),
1270 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1271 * Otherwise, csum completion requires chacksumming packet body,
1272 * including udp header and folding it to skb->csum.
1274 static inline int udp4_csum_init(struct sk_buff
*skb
, struct udphdr
*uh
,
1277 const struct iphdr
*iph
;
1280 UDP_SKB_CB(skb
)->partial_cov
= 0;
1281 UDP_SKB_CB(skb
)->cscov
= skb
->len
;
1283 if (proto
== IPPROTO_UDPLITE
) {
1284 err
= udplite_checksum_init(skb
, uh
);
1290 if (uh
->check
== 0) {
1291 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1292 } else if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
1293 if (!csum_tcpudp_magic(iph
->saddr
, iph
->daddr
, skb
->len
,
1295 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1297 if (!skb_csum_unnecessary(skb
))
1298 skb
->csum
= csum_tcpudp_nofold(iph
->saddr
, iph
->daddr
,
1299 skb
->len
, proto
, 0);
1300 /* Probably, we should checksum udp header (it should be in cache
1301 * in any case) and data in tiny packets (< rx copybreak).
1308 * All we need to do is get the socket, and then do a checksum.
1311 int __udp4_lib_rcv(struct sk_buff
*skb
, struct udp_table
*udptable
,
1316 unsigned short ulen
;
1317 struct rtable
*rt
= skb_rtable(skb
);
1318 __be32 saddr
, daddr
;
1319 struct net
*net
= dev_net(skb
->dev
);
1322 * Validate the packet.
1324 if (!pskb_may_pull(skb
, sizeof(struct udphdr
)))
1325 goto drop
; /* No space for header. */
1328 ulen
= ntohs(uh
->len
);
1329 if (ulen
> skb
->len
)
1332 if (proto
== IPPROTO_UDP
) {
1333 /* UDP validates ulen. */
1334 if (ulen
< sizeof(*uh
) || pskb_trim_rcsum(skb
, ulen
))
1339 if (udp4_csum_init(skb
, uh
, proto
))
1342 saddr
= ip_hdr(skb
)->saddr
;
1343 daddr
= ip_hdr(skb
)->daddr
;
1345 if (rt
->rt_flags
& (RTCF_BROADCAST
|RTCF_MULTICAST
))
1346 return __udp4_lib_mcast_deliver(net
, skb
, uh
,
1347 saddr
, daddr
, udptable
);
1349 sk
= __udp4_lib_lookup_skb(skb
, uh
->source
, uh
->dest
, udptable
);
1352 int ret
= udp_queue_rcv_skb(sk
, skb
);
1355 /* a return value > 0 means to resubmit the input, but
1356 * it wants the return to be -protocol, or 0
1363 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
1367 /* No socket. Drop packet silently, if checksum is wrong */
1368 if (udp_lib_checksum_complete(skb
))
1371 UDP_INC_STATS_BH(net
, UDP_MIB_NOPORTS
, proto
== IPPROTO_UDPLITE
);
1372 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_PORT_UNREACH
, 0);
1375 * Hmm. We got an UDP packet to a port to which we
1376 * don't wanna listen. Ignore it.
1382 LIMIT_NETDEBUG(KERN_DEBUG
"UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1383 proto
== IPPROTO_UDPLITE
? "-Lite" : "",
1394 * RFC1122: OK. Discards the bad packet silently (as far as
1395 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1397 LIMIT_NETDEBUG(KERN_DEBUG
"UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1398 proto
== IPPROTO_UDPLITE
? "-Lite" : "",
1405 UDP_INC_STATS_BH(net
, UDP_MIB_INERRORS
, proto
== IPPROTO_UDPLITE
);
1410 int udp_rcv(struct sk_buff
*skb
)
1412 return __udp4_lib_rcv(skb
, &udp_table
, IPPROTO_UDP
);
1415 void udp_destroy_sock(struct sock
*sk
)
1418 udp_flush_pending_frames(sk
);
1423 * Socket option code for UDP
1425 int udp_lib_setsockopt(struct sock
*sk
, int level
, int optname
,
1426 char __user
*optval
, unsigned int optlen
,
1427 int (*push_pending_frames
)(struct sock
*))
1429 struct udp_sock
*up
= udp_sk(sk
);
1432 int is_udplite
= IS_UDPLITE(sk
);
1434 if (optlen
< sizeof(int))
1437 if (get_user(val
, (int __user
*)optval
))
1447 (*push_pending_frames
)(sk
);
1455 case UDP_ENCAP_ESPINUDP
:
1456 case UDP_ENCAP_ESPINUDP_NON_IKE
:
1457 up
->encap_rcv
= xfrm4_udp_encap_rcv
;
1459 case UDP_ENCAP_L2TPINUDP
:
1460 up
->encap_type
= val
;
1469 * UDP-Lite's partial checksum coverage (RFC 3828).
1471 /* The sender sets actual checksum coverage length via this option.
1472 * The case coverage > packet length is handled by send module. */
1473 case UDPLITE_SEND_CSCOV
:
1474 if (!is_udplite
) /* Disable the option on UDP sockets */
1475 return -ENOPROTOOPT
;
1476 if (val
!= 0 && val
< 8) /* Illegal coverage: use default (8) */
1478 else if (val
> USHORT_MAX
)
1481 up
->pcflag
|= UDPLITE_SEND_CC
;
1484 /* The receiver specifies a minimum checksum coverage value. To make
1485 * sense, this should be set to at least 8 (as done below). If zero is
1486 * used, this again means full checksum coverage. */
1487 case UDPLITE_RECV_CSCOV
:
1488 if (!is_udplite
) /* Disable the option on UDP sockets */
1489 return -ENOPROTOOPT
;
1490 if (val
!= 0 && val
< 8) /* Avoid silly minimal values. */
1492 else if (val
> USHORT_MAX
)
1495 up
->pcflag
|= UDPLITE_RECV_CC
;
1505 EXPORT_SYMBOL(udp_lib_setsockopt
);
1507 int udp_setsockopt(struct sock
*sk
, int level
, int optname
,
1508 char __user
*optval
, unsigned int optlen
)
1510 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1511 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1512 udp_push_pending_frames
);
1513 return ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
1516 #ifdef CONFIG_COMPAT
1517 int compat_udp_setsockopt(struct sock
*sk
, int level
, int optname
,
1518 char __user
*optval
, unsigned int optlen
)
1520 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1521 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1522 udp_push_pending_frames
);
1523 return compat_ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
1527 int udp_lib_getsockopt(struct sock
*sk
, int level
, int optname
,
1528 char __user
*optval
, int __user
*optlen
)
1530 struct udp_sock
*up
= udp_sk(sk
);
1533 if (get_user(len
, optlen
))
1536 len
= min_t(unsigned int, len
, sizeof(int));
1547 val
= up
->encap_type
;
1550 /* The following two cannot be changed on UDP sockets, the return is
1551 * always 0 (which corresponds to the full checksum coverage of UDP). */
1552 case UDPLITE_SEND_CSCOV
:
1556 case UDPLITE_RECV_CSCOV
:
1561 return -ENOPROTOOPT
;
1564 if (put_user(len
, optlen
))
1566 if (copy_to_user(optval
, &val
, len
))
1570 EXPORT_SYMBOL(udp_lib_getsockopt
);
1572 int udp_getsockopt(struct sock
*sk
, int level
, int optname
,
1573 char __user
*optval
, int __user
*optlen
)
1575 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1576 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1577 return ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
1580 #ifdef CONFIG_COMPAT
1581 int compat_udp_getsockopt(struct sock
*sk
, int level
, int optname
,
1582 char __user
*optval
, int __user
*optlen
)
1584 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1585 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1586 return compat_ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
1590 * udp_poll - wait for a UDP event.
1591 * @file - file struct
1593 * @wait - poll table
1595 * This is same as datagram poll, except for the special case of
1596 * blocking sockets. If application is using a blocking fd
1597 * and a packet with checksum error is in the queue;
1598 * then it could get return from select indicating data available
1599 * but then block when reading it. Add special case code
1600 * to work around these arguably broken applications.
1602 unsigned int udp_poll(struct file
*file
, struct socket
*sock
, poll_table
*wait
)
1604 unsigned int mask
= datagram_poll(file
, sock
, wait
);
1605 struct sock
*sk
= sock
->sk
;
1607 /* Check for false positives due to checksum errors */
1608 if ((mask
& POLLRDNORM
) && !(file
->f_flags
& O_NONBLOCK
) &&
1609 !(sk
->sk_shutdown
& RCV_SHUTDOWN
) && !first_packet_length(sk
))
1610 mask
&= ~(POLLIN
| POLLRDNORM
);
1615 EXPORT_SYMBOL(udp_poll
);
1617 struct proto udp_prot
= {
1619 .owner
= THIS_MODULE
,
1620 .close
= udp_lib_close
,
1621 .connect
= ip4_datagram_connect
,
1622 .disconnect
= udp_disconnect
,
1624 .destroy
= udp_destroy_sock
,
1625 .setsockopt
= udp_setsockopt
,
1626 .getsockopt
= udp_getsockopt
,
1627 .sendmsg
= udp_sendmsg
,
1628 .recvmsg
= udp_recvmsg
,
1629 .sendpage
= udp_sendpage
,
1630 .backlog_rcv
= __udp_queue_rcv_skb
,
1631 .hash
= udp_lib_hash
,
1632 .unhash
= udp_lib_unhash
,
1633 .get_port
= udp_v4_get_port
,
1634 .memory_allocated
= &udp_memory_allocated
,
1635 .sysctl_mem
= sysctl_udp_mem
,
1636 .sysctl_wmem
= &sysctl_udp_wmem_min
,
1637 .sysctl_rmem
= &sysctl_udp_rmem_min
,
1638 .obj_size
= sizeof(struct udp_sock
),
1639 .slab_flags
= SLAB_DESTROY_BY_RCU
,
1640 .h
.udp_table
= &udp_table
,
1641 #ifdef CONFIG_COMPAT
1642 .compat_setsockopt
= compat_udp_setsockopt
,
1643 .compat_getsockopt
= compat_udp_getsockopt
,
1646 EXPORT_SYMBOL(udp_prot
);
1648 /* ------------------------------------------------------------------------ */
1649 #ifdef CONFIG_PROC_FS
1651 static struct sock
*udp_get_first(struct seq_file
*seq
, int start
)
1654 struct udp_iter_state
*state
= seq
->private;
1655 struct net
*net
= seq_file_net(seq
);
1657 for (state
->bucket
= start
; state
->bucket
<= state
->udp_table
->mask
;
1659 struct hlist_nulls_node
*node
;
1660 struct udp_hslot
*hslot
= &state
->udp_table
->hash
[state
->bucket
];
1662 if (hlist_nulls_empty(&hslot
->head
))
1665 spin_lock_bh(&hslot
->lock
);
1666 sk_nulls_for_each(sk
, node
, &hslot
->head
) {
1667 if (!net_eq(sock_net(sk
), net
))
1669 if (sk
->sk_family
== state
->family
)
1672 spin_unlock_bh(&hslot
->lock
);
1679 static struct sock
*udp_get_next(struct seq_file
*seq
, struct sock
*sk
)
1681 struct udp_iter_state
*state
= seq
->private;
1682 struct net
*net
= seq_file_net(seq
);
1685 sk
= sk_nulls_next(sk
);
1686 } while (sk
&& (!net_eq(sock_net(sk
), net
) || sk
->sk_family
!= state
->family
));
1689 if (state
->bucket
<= state
->udp_table
->mask
)
1690 spin_unlock_bh(&state
->udp_table
->hash
[state
->bucket
].lock
);
1691 return udp_get_first(seq
, state
->bucket
+ 1);
1696 static struct sock
*udp_get_idx(struct seq_file
*seq
, loff_t pos
)
1698 struct sock
*sk
= udp_get_first(seq
, 0);
1701 while (pos
&& (sk
= udp_get_next(seq
, sk
)) != NULL
)
1703 return pos
? NULL
: sk
;
1706 static void *udp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1708 struct udp_iter_state
*state
= seq
->private;
1709 state
->bucket
= MAX_UDP_PORTS
;
1711 return *pos
? udp_get_idx(seq
, *pos
-1) : SEQ_START_TOKEN
;
1714 static void *udp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1718 if (v
== SEQ_START_TOKEN
)
1719 sk
= udp_get_idx(seq
, 0);
1721 sk
= udp_get_next(seq
, v
);
1727 static void udp_seq_stop(struct seq_file
*seq
, void *v
)
1729 struct udp_iter_state
*state
= seq
->private;
1731 if (state
->bucket
<= state
->udp_table
->mask
)
1732 spin_unlock_bh(&state
->udp_table
->hash
[state
->bucket
].lock
);
1735 static int udp_seq_open(struct inode
*inode
, struct file
*file
)
1737 struct udp_seq_afinfo
*afinfo
= PDE(inode
)->data
;
1738 struct udp_iter_state
*s
;
1741 err
= seq_open_net(inode
, file
, &afinfo
->seq_ops
,
1742 sizeof(struct udp_iter_state
));
1746 s
= ((struct seq_file
*)file
->private_data
)->private;
1747 s
->family
= afinfo
->family
;
1748 s
->udp_table
= afinfo
->udp_table
;
1752 /* ------------------------------------------------------------------------ */
1753 int udp_proc_register(struct net
*net
, struct udp_seq_afinfo
*afinfo
)
1755 struct proc_dir_entry
*p
;
1758 afinfo
->seq_fops
.open
= udp_seq_open
;
1759 afinfo
->seq_fops
.read
= seq_read
;
1760 afinfo
->seq_fops
.llseek
= seq_lseek
;
1761 afinfo
->seq_fops
.release
= seq_release_net
;
1763 afinfo
->seq_ops
.start
= udp_seq_start
;
1764 afinfo
->seq_ops
.next
= udp_seq_next
;
1765 afinfo
->seq_ops
.stop
= udp_seq_stop
;
1767 p
= proc_create_data(afinfo
->name
, S_IRUGO
, net
->proc_net
,
1768 &afinfo
->seq_fops
, afinfo
);
1773 EXPORT_SYMBOL(udp_proc_register
);
1775 void udp_proc_unregister(struct net
*net
, struct udp_seq_afinfo
*afinfo
)
1777 proc_net_remove(net
, afinfo
->name
);
1779 EXPORT_SYMBOL(udp_proc_unregister
);
1781 /* ------------------------------------------------------------------------ */
1782 static void udp4_format_sock(struct sock
*sp
, struct seq_file
*f
,
1783 int bucket
, int *len
)
1785 struct inet_sock
*inet
= inet_sk(sp
);
1786 __be32 dest
= inet
->inet_daddr
;
1787 __be32 src
= inet
->inet_rcv_saddr
;
1788 __u16 destp
= ntohs(inet
->inet_dport
);
1789 __u16 srcp
= ntohs(inet
->inet_sport
);
1791 seq_printf(f
, "%5d: %08X:%04X %08X:%04X"
1792 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n",
1793 bucket
, src
, srcp
, dest
, destp
, sp
->sk_state
,
1794 sk_wmem_alloc_get(sp
),
1795 sk_rmem_alloc_get(sp
),
1796 0, 0L, 0, sock_i_uid(sp
), 0, sock_i_ino(sp
),
1797 atomic_read(&sp
->sk_refcnt
), sp
,
1798 atomic_read(&sp
->sk_drops
), len
);
1801 int udp4_seq_show(struct seq_file
*seq
, void *v
)
1803 if (v
== SEQ_START_TOKEN
)
1804 seq_printf(seq
, "%-127s\n",
1805 " sl local_address rem_address st tx_queue "
1806 "rx_queue tr tm->when retrnsmt uid timeout "
1807 "inode ref pointer drops");
1809 struct udp_iter_state
*state
= seq
->private;
1812 udp4_format_sock(v
, seq
, state
->bucket
, &len
);
1813 seq_printf(seq
, "%*s\n", 127 - len
, "");
1818 /* ------------------------------------------------------------------------ */
1819 static struct udp_seq_afinfo udp4_seq_afinfo
= {
1822 .udp_table
= &udp_table
,
1824 .owner
= THIS_MODULE
,
1827 .show
= udp4_seq_show
,
1831 static int udp4_proc_init_net(struct net
*net
)
1833 return udp_proc_register(net
, &udp4_seq_afinfo
);
1836 static void udp4_proc_exit_net(struct net
*net
)
1838 udp_proc_unregister(net
, &udp4_seq_afinfo
);
1841 static struct pernet_operations udp4_net_ops
= {
1842 .init
= udp4_proc_init_net
,
1843 .exit
= udp4_proc_exit_net
,
1846 int __init
udp4_proc_init(void)
1848 return register_pernet_subsys(&udp4_net_ops
);
1851 void udp4_proc_exit(void)
1853 unregister_pernet_subsys(&udp4_net_ops
);
1855 #endif /* CONFIG_PROC_FS */
1857 static __initdata
unsigned long uhash_entries
;
1858 static int __init
set_uhash_entries(char *str
)
1862 uhash_entries
= simple_strtoul(str
, &str
, 0);
1863 if (uhash_entries
&& uhash_entries
< UDP_HTABLE_SIZE_MIN
)
1864 uhash_entries
= UDP_HTABLE_SIZE_MIN
;
1867 __setup("uhash_entries=", set_uhash_entries
);
1869 void __init
udp_table_init(struct udp_table
*table
, const char *name
)
1873 if (!CONFIG_BASE_SMALL
)
1874 table
->hash
= alloc_large_system_hash(name
,
1875 2 * sizeof(struct udp_hslot
),
1877 21, /* one slot per 2 MB */
1883 * Make sure hash table has the minimum size
1885 if (CONFIG_BASE_SMALL
|| table
->mask
< UDP_HTABLE_SIZE_MIN
- 1) {
1886 table
->hash
= kmalloc(UDP_HTABLE_SIZE_MIN
*
1887 2 * sizeof(struct udp_hslot
), GFP_KERNEL
);
1890 table
->log
= ilog2(UDP_HTABLE_SIZE_MIN
);
1891 table
->mask
= UDP_HTABLE_SIZE_MIN
- 1;
1893 table
->hash2
= table
->hash
+ (table
->mask
+ 1);
1894 for (i
= 0; i
<= table
->mask
; i
++) {
1895 INIT_HLIST_NULLS_HEAD(&table
->hash
[i
].head
, i
);
1896 table
->hash
[i
].count
= 0;
1897 spin_lock_init(&table
->hash
[i
].lock
);
1899 for (i
= 0; i
<= table
->mask
; i
++) {
1900 INIT_HLIST_NULLS_HEAD(&table
->hash2
[i
].head
, i
);
1901 table
->hash2
[i
].count
= 0;
1902 spin_lock_init(&table
->hash2
[i
].lock
);
1906 void __init
udp_init(void)
1908 unsigned long nr_pages
, limit
;
1910 udp_table_init(&udp_table
, "UDP");
1911 /* Set the pressure threshold up by the same strategy of TCP. It is a
1912 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
1913 * toward zero with the amount of memory, with a floor of 128 pages.
1915 nr_pages
= totalram_pages
- totalhigh_pages
;
1916 limit
= min(nr_pages
, 1UL<<(28-PAGE_SHIFT
)) >> (20-PAGE_SHIFT
);
1917 limit
= (limit
* (nr_pages
>> (20-PAGE_SHIFT
))) >> (PAGE_SHIFT
-11);
1918 limit
= max(limit
, 128UL);
1919 sysctl_udp_mem
[0] = limit
/ 4 * 3;
1920 sysctl_udp_mem
[1] = limit
;
1921 sysctl_udp_mem
[2] = sysctl_udp_mem
[0] * 2;
1923 sysctl_udp_rmem_min
= SK_MEM_QUANTUM
;
1924 sysctl_udp_wmem_min
= SK_MEM_QUANTUM
;
1927 int udp4_ufo_send_check(struct sk_buff
*skb
)
1929 const struct iphdr
*iph
;
1932 if (!pskb_may_pull(skb
, sizeof(*uh
)))
1938 uh
->check
= ~csum_tcpudp_magic(iph
->saddr
, iph
->daddr
, skb
->len
,
1940 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
1941 skb
->csum_offset
= offsetof(struct udphdr
, check
);
1942 skb
->ip_summed
= CHECKSUM_PARTIAL
;
1946 struct sk_buff
*udp4_ufo_fragment(struct sk_buff
*skb
, int features
)
1948 struct sk_buff
*segs
= ERR_PTR(-EINVAL
);
1953 mss
= skb_shinfo(skb
)->gso_size
;
1954 if (unlikely(skb
->len
<= mss
))
1957 if (skb_gso_ok(skb
, features
| NETIF_F_GSO_ROBUST
)) {
1958 /* Packet is from an untrusted source, reset gso_segs. */
1959 int type
= skb_shinfo(skb
)->gso_type
;
1961 if (unlikely(type
& ~(SKB_GSO_UDP
| SKB_GSO_DODGY
) ||
1962 !(type
& (SKB_GSO_UDP
))))
1965 skb_shinfo(skb
)->gso_segs
= DIV_ROUND_UP(skb
->len
, mss
);
1971 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
1972 * do checksum of UDP packets sent as multiple IP fragments.
1974 offset
= skb
->csum_start
- skb_headroom(skb
);
1975 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1976 offset
+= skb
->csum_offset
;
1977 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1978 skb
->ip_summed
= CHECKSUM_NONE
;
1980 /* Fragment the skb. IP headers of the fragments are updated in
1981 * inet_gso_segment()
1983 segs
= skb_segment(skb
, features
);