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 <linux/slab.h>
99 #include <net/tcp_states.h>
100 #include <linux/skbuff.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <net/net_namespace.h>
104 #include <net/icmp.h>
105 #include <net/route.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include "udp_impl.h"
110 struct udp_table udp_table __read_mostly
;
111 EXPORT_SYMBOL(udp_table
);
113 long sysctl_udp_mem
[3] __read_mostly
;
114 EXPORT_SYMBOL(sysctl_udp_mem
);
116 int sysctl_udp_rmem_min __read_mostly
;
117 EXPORT_SYMBOL(sysctl_udp_rmem_min
);
119 int sysctl_udp_wmem_min __read_mostly
;
120 EXPORT_SYMBOL(sysctl_udp_wmem_min
);
122 atomic_long_t udp_memory_allocated
;
123 EXPORT_SYMBOL(udp_memory_allocated
);
125 #define MAX_UDP_PORTS 65536
126 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
128 static int udp_lib_lport_inuse(struct net
*net
, __u16 num
,
129 const struct udp_hslot
*hslot
,
130 unsigned long *bitmap
,
132 int (*saddr_comp
)(const struct sock
*sk1
,
133 const struct sock
*sk2
),
137 struct hlist_nulls_node
*node
;
139 sk_nulls_for_each(sk2
, node
, &hslot
->head
)
140 if (net_eq(sock_net(sk2
), net
) &&
142 (bitmap
|| udp_sk(sk2
)->udp_port_hash
== num
) &&
143 (!sk2
->sk_reuse
|| !sk
->sk_reuse
) &&
144 (!sk2
->sk_bound_dev_if
|| !sk
->sk_bound_dev_if
||
145 sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
146 (*saddr_comp
)(sk
, sk2
)) {
148 __set_bit(udp_sk(sk2
)->udp_port_hash
>> log
,
157 * Note: we still hold spinlock of primary hash chain, so no other writer
158 * can insert/delete a socket with local_port == num
160 static int udp_lib_lport_inuse2(struct net
*net
, __u16 num
,
161 struct udp_hslot
*hslot2
,
163 int (*saddr_comp
)(const struct sock
*sk1
,
164 const struct sock
*sk2
))
167 struct hlist_nulls_node
*node
;
170 spin_lock(&hslot2
->lock
);
171 udp_portaddr_for_each_entry(sk2
, node
, &hslot2
->head
)
172 if (net_eq(sock_net(sk2
), net
) &&
174 (udp_sk(sk2
)->udp_port_hash
== num
) &&
175 (!sk2
->sk_reuse
|| !sk
->sk_reuse
) &&
176 (!sk2
->sk_bound_dev_if
|| !sk
->sk_bound_dev_if
||
177 sk2
->sk_bound_dev_if
== sk
->sk_bound_dev_if
) &&
178 (*saddr_comp
)(sk
, sk2
)) {
182 spin_unlock(&hslot2
->lock
);
187 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
189 * @sk: socket struct in question
190 * @snum: port number to look up
191 * @saddr_comp: AF-dependent comparison of bound local IP addresses
192 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
195 int udp_lib_get_port(struct sock
*sk
, unsigned short snum
,
196 int (*saddr_comp
)(const struct sock
*sk1
,
197 const struct sock
*sk2
),
198 unsigned int hash2_nulladdr
)
200 struct udp_hslot
*hslot
, *hslot2
;
201 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
203 struct net
*net
= sock_net(sk
);
206 int low
, high
, remaining
;
208 unsigned short first
, last
;
209 DECLARE_BITMAP(bitmap
, PORTS_PER_CHAIN
);
211 inet_get_local_port_range(&low
, &high
);
212 remaining
= (high
- low
) + 1;
215 first
= (((u64
)rand
* remaining
) >> 32) + low
;
217 * force rand to be an odd multiple of UDP_HTABLE_SIZE
219 rand
= (rand
| 1) * (udptable
->mask
+ 1);
220 last
= first
+ udptable
->mask
+ 1;
222 hslot
= udp_hashslot(udptable
, net
, first
);
223 bitmap_zero(bitmap
, PORTS_PER_CHAIN
);
224 spin_lock_bh(&hslot
->lock
);
225 udp_lib_lport_inuse(net
, snum
, hslot
, bitmap
, sk
,
226 saddr_comp
, udptable
->log
);
230 * Iterate on all possible values of snum for this hash.
231 * Using steps of an odd multiple of UDP_HTABLE_SIZE
232 * give us randomization and full range coverage.
235 if (low
<= snum
&& snum
<= high
&&
236 !test_bit(snum
>> udptable
->log
, bitmap
) &&
237 !inet_is_reserved_local_port(snum
))
240 } while (snum
!= first
);
241 spin_unlock_bh(&hslot
->lock
);
242 } while (++first
!= last
);
245 hslot
= udp_hashslot(udptable
, net
, snum
);
246 spin_lock_bh(&hslot
->lock
);
247 if (hslot
->count
> 10) {
249 unsigned int slot2
= udp_sk(sk
)->udp_portaddr_hash
^ snum
;
251 slot2
&= udptable
->mask
;
252 hash2_nulladdr
&= udptable
->mask
;
254 hslot2
= udp_hashslot2(udptable
, slot2
);
255 if (hslot
->count
< hslot2
->count
)
256 goto scan_primary_hash
;
258 exist
= udp_lib_lport_inuse2(net
, snum
, hslot2
,
260 if (!exist
&& (hash2_nulladdr
!= slot2
)) {
261 hslot2
= udp_hashslot2(udptable
, hash2_nulladdr
);
262 exist
= udp_lib_lport_inuse2(net
, snum
, hslot2
,
271 if (udp_lib_lport_inuse(net
, snum
, hslot
, NULL
, sk
,
276 inet_sk(sk
)->inet_num
= snum
;
277 udp_sk(sk
)->udp_port_hash
= snum
;
278 udp_sk(sk
)->udp_portaddr_hash
^= snum
;
279 if (sk_unhashed(sk
)) {
280 sk_nulls_add_node_rcu(sk
, &hslot
->head
);
282 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, 1);
284 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
285 spin_lock(&hslot2
->lock
);
286 hlist_nulls_add_head_rcu(&udp_sk(sk
)->udp_portaddr_node
,
289 spin_unlock(&hslot2
->lock
);
293 spin_unlock_bh(&hslot
->lock
);
297 EXPORT_SYMBOL(udp_lib_get_port
);
299 static int ipv4_rcv_saddr_equal(const struct sock
*sk1
, const struct sock
*sk2
)
301 struct inet_sock
*inet1
= inet_sk(sk1
), *inet2
= inet_sk(sk2
);
303 return (!ipv6_only_sock(sk2
) &&
304 (!inet1
->inet_rcv_saddr
|| !inet2
->inet_rcv_saddr
||
305 inet1
->inet_rcv_saddr
== inet2
->inet_rcv_saddr
));
308 static unsigned int udp4_portaddr_hash(struct net
*net
, __be32 saddr
,
311 return jhash_1word((__force u32
)saddr
, net_hash_mix(net
)) ^ port
;
314 int udp_v4_get_port(struct sock
*sk
, unsigned short snum
)
316 unsigned int hash2_nulladdr
=
317 udp4_portaddr_hash(sock_net(sk
), htonl(INADDR_ANY
), snum
);
318 unsigned int hash2_partial
=
319 udp4_portaddr_hash(sock_net(sk
), inet_sk(sk
)->inet_rcv_saddr
, 0);
321 /* precompute partial secondary hash */
322 udp_sk(sk
)->udp_portaddr_hash
= hash2_partial
;
323 return udp_lib_get_port(sk
, snum
, ipv4_rcv_saddr_equal
, hash2_nulladdr
);
326 static inline int compute_score(struct sock
*sk
, struct net
*net
, __be32 saddr
,
328 __be16 sport
, __be32 daddr
, __be16 dport
, int dif
)
332 if (net_eq(sock_net(sk
), net
) && udp_sk(sk
)->udp_port_hash
== hnum
&&
333 !ipv6_only_sock(sk
)) {
334 struct inet_sock
*inet
= inet_sk(sk
);
336 score
= (sk
->sk_family
== PF_INET
? 1 : 0);
337 if (inet
->inet_rcv_saddr
) {
338 if (inet
->inet_rcv_saddr
!= daddr
)
342 if (inet
->inet_daddr
) {
343 if (inet
->inet_daddr
!= saddr
)
347 if (inet
->inet_dport
) {
348 if (inet
->inet_dport
!= sport
)
352 if (sk
->sk_bound_dev_if
) {
353 if (sk
->sk_bound_dev_if
!= dif
)
362 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
364 #define SCORE2_MAX (1 + 2 + 2 + 2)
365 static inline int compute_score2(struct sock
*sk
, struct net
*net
,
366 __be32 saddr
, __be16 sport
,
367 __be32 daddr
, unsigned int hnum
, int dif
)
371 if (net_eq(sock_net(sk
), net
) && !ipv6_only_sock(sk
)) {
372 struct inet_sock
*inet
= inet_sk(sk
);
374 if (inet
->inet_rcv_saddr
!= daddr
)
376 if (inet
->inet_num
!= hnum
)
379 score
= (sk
->sk_family
== PF_INET
? 1 : 0);
380 if (inet
->inet_daddr
) {
381 if (inet
->inet_daddr
!= saddr
)
385 if (inet
->inet_dport
) {
386 if (inet
->inet_dport
!= sport
)
390 if (sk
->sk_bound_dev_if
) {
391 if (sk
->sk_bound_dev_if
!= dif
)
400 /* called with read_rcu_lock() */
401 static struct sock
*udp4_lib_lookup2(struct net
*net
,
402 __be32 saddr
, __be16 sport
,
403 __be32 daddr
, unsigned int hnum
, int dif
,
404 struct udp_hslot
*hslot2
, unsigned int slot2
)
406 struct sock
*sk
, *result
;
407 struct hlist_nulls_node
*node
;
413 udp_portaddr_for_each_entry_rcu(sk
, node
, &hslot2
->head
) {
414 score
= compute_score2(sk
, net
, saddr
, sport
,
416 if (score
> badness
) {
419 if (score
== SCORE2_MAX
)
424 * if the nulls value we got at the end of this lookup is
425 * not the expected one, we must restart lookup.
426 * We probably met an item that was moved to another chain.
428 if (get_nulls_value(node
) != slot2
)
433 if (unlikely(!atomic_inc_not_zero_hint(&result
->sk_refcnt
, 2)))
435 else if (unlikely(compute_score2(result
, net
, saddr
, sport
,
436 daddr
, hnum
, dif
) < badness
)) {
444 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
445 * harder than this. -DaveM
447 static struct sock
*__udp4_lib_lookup(struct net
*net
, __be32 saddr
,
448 __be16 sport
, __be32 daddr
, __be16 dport
,
449 int dif
, struct udp_table
*udptable
)
451 struct sock
*sk
, *result
;
452 struct hlist_nulls_node
*node
;
453 unsigned short hnum
= ntohs(dport
);
454 unsigned int hash2
, slot2
, slot
= udp_hashfn(net
, hnum
, udptable
->mask
);
455 struct udp_hslot
*hslot2
, *hslot
= &udptable
->hash
[slot
];
459 if (hslot
->count
> 10) {
460 hash2
= udp4_portaddr_hash(net
, daddr
, hnum
);
461 slot2
= hash2
& udptable
->mask
;
462 hslot2
= &udptable
->hash2
[slot2
];
463 if (hslot
->count
< hslot2
->count
)
466 result
= udp4_lib_lookup2(net
, saddr
, sport
,
470 hash2
= udp4_portaddr_hash(net
, htonl(INADDR_ANY
), hnum
);
471 slot2
= hash2
& udptable
->mask
;
472 hslot2
= &udptable
->hash2
[slot2
];
473 if (hslot
->count
< hslot2
->count
)
476 result
= udp4_lib_lookup2(net
, saddr
, sport
,
477 htonl(INADDR_ANY
), hnum
, dif
,
486 sk_nulls_for_each_rcu(sk
, node
, &hslot
->head
) {
487 score
= compute_score(sk
, net
, saddr
, hnum
, sport
,
489 if (score
> badness
) {
495 * if the nulls value we got at the end of this lookup is
496 * not the expected one, we must restart lookup.
497 * We probably met an item that was moved to another chain.
499 if (get_nulls_value(node
) != slot
)
503 if (unlikely(!atomic_inc_not_zero_hint(&result
->sk_refcnt
, 2)))
505 else if (unlikely(compute_score(result
, net
, saddr
, hnum
, sport
,
506 daddr
, dport
, dif
) < badness
)) {
515 static inline struct sock
*__udp4_lib_lookup_skb(struct sk_buff
*skb
,
516 __be16 sport
, __be16 dport
,
517 struct udp_table
*udptable
)
520 const struct iphdr
*iph
= ip_hdr(skb
);
522 if (unlikely(sk
= skb_steal_sock(skb
)))
525 return __udp4_lib_lookup(dev_net(skb_dst(skb
)->dev
), iph
->saddr
, sport
,
526 iph
->daddr
, dport
, inet_iif(skb
),
530 struct sock
*udp4_lib_lookup(struct net
*net
, __be32 saddr
, __be16 sport
,
531 __be32 daddr
, __be16 dport
, int dif
)
533 return __udp4_lib_lookup(net
, saddr
, sport
, daddr
, dport
, dif
, &udp_table
);
535 EXPORT_SYMBOL_GPL(udp4_lib_lookup
);
537 static inline struct sock
*udp_v4_mcast_next(struct net
*net
, struct sock
*sk
,
538 __be16 loc_port
, __be32 loc_addr
,
539 __be16 rmt_port
, __be32 rmt_addr
,
542 struct hlist_nulls_node
*node
;
544 unsigned short hnum
= ntohs(loc_port
);
546 sk_nulls_for_each_from(s
, node
) {
547 struct inet_sock
*inet
= inet_sk(s
);
549 if (!net_eq(sock_net(s
), net
) ||
550 udp_sk(s
)->udp_port_hash
!= hnum
||
551 (inet
->inet_daddr
&& inet
->inet_daddr
!= rmt_addr
) ||
552 (inet
->inet_dport
!= rmt_port
&& inet
->inet_dport
) ||
553 (inet
->inet_rcv_saddr
&&
554 inet
->inet_rcv_saddr
!= loc_addr
) ||
556 (s
->sk_bound_dev_if
&& s
->sk_bound_dev_if
!= dif
))
558 if (!ip_mc_sf_allow(s
, loc_addr
, rmt_addr
, dif
))
568 * This routine is called by the ICMP module when it gets some
569 * sort of error condition. If err < 0 then the socket should
570 * be closed and the error returned to the user. If err > 0
571 * it's just the icmp type << 8 | icmp code.
572 * Header points to the ip header of the error packet. We move
573 * on past this. Then (as it used to claim before adjustment)
574 * header points to the first 8 bytes of the udp header. We need
575 * to find the appropriate port.
578 void __udp4_lib_err(struct sk_buff
*skb
, u32 info
, struct udp_table
*udptable
)
580 struct inet_sock
*inet
;
581 const struct iphdr
*iph
= (const struct iphdr
*)skb
->data
;
582 struct udphdr
*uh
= (struct udphdr
*)(skb
->data
+(iph
->ihl
<<2));
583 const int type
= icmp_hdr(skb
)->type
;
584 const int code
= icmp_hdr(skb
)->code
;
588 struct net
*net
= dev_net(skb
->dev
);
590 sk
= __udp4_lib_lookup(net
, iph
->daddr
, uh
->dest
,
591 iph
->saddr
, uh
->source
, skb
->dev
->ifindex
, udptable
);
593 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
594 return; /* No socket for error */
603 case ICMP_TIME_EXCEEDED
:
606 case ICMP_SOURCE_QUENCH
:
608 case ICMP_PARAMETERPROB
:
612 case ICMP_DEST_UNREACH
:
613 if (code
== ICMP_FRAG_NEEDED
) { /* Path MTU discovery */
614 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
) {
622 if (code
<= NR_ICMP_UNREACH
) {
623 harderr
= icmp_err_convert
[code
].fatal
;
624 err
= icmp_err_convert
[code
].errno
;
630 * RFC1122: OK. Passes ICMP errors back to application, as per
633 if (!inet
->recverr
) {
634 if (!harderr
|| sk
->sk_state
!= TCP_ESTABLISHED
)
637 ip_icmp_error(sk
, skb
, err
, uh
->dest
, info
, (u8
*)(uh
+1));
640 sk
->sk_error_report(sk
);
645 void udp_err(struct sk_buff
*skb
, u32 info
)
647 __udp4_lib_err(skb
, info
, &udp_table
);
651 * Throw away all pending data and cancel the corking. Socket is locked.
653 void udp_flush_pending_frames(struct sock
*sk
)
655 struct udp_sock
*up
= udp_sk(sk
);
660 ip_flush_pending_frames(sk
);
663 EXPORT_SYMBOL(udp_flush_pending_frames
);
666 * udp4_hwcsum - handle outgoing HW checksumming
667 * @skb: sk_buff containing the filled-in UDP header
668 * (checksum field must be zeroed out)
669 * @src: source IP address
670 * @dst: destination IP address
672 static void udp4_hwcsum(struct sk_buff
*skb
, __be32 src
, __be32 dst
)
674 struct udphdr
*uh
= udp_hdr(skb
);
675 struct sk_buff
*frags
= skb_shinfo(skb
)->frag_list
;
676 int offset
= skb_transport_offset(skb
);
677 int len
= skb
->len
- offset
;
683 * Only one fragment on the socket.
685 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
686 skb
->csum_offset
= offsetof(struct udphdr
, check
);
687 uh
->check
= ~csum_tcpudp_magic(src
, dst
, len
,
691 * HW-checksum won't work as there are two or more
692 * fragments on the socket so that all csums of sk_buffs
696 csum
= csum_add(csum
, frags
->csum
);
698 } while ((frags
= frags
->next
));
700 csum
= skb_checksum(skb
, offset
, hlen
, csum
);
701 skb
->ip_summed
= CHECKSUM_NONE
;
703 uh
->check
= csum_tcpudp_magic(src
, dst
, len
, IPPROTO_UDP
, csum
);
705 uh
->check
= CSUM_MANGLED_0
;
709 static int udp_send_skb(struct sk_buff
*skb
, __be32 daddr
, __be32 dport
)
711 struct sock
*sk
= skb
->sk
;
712 struct inet_sock
*inet
= inet_sk(sk
);
714 struct rtable
*rt
= (struct rtable
*)skb_dst(skb
);
716 int is_udplite
= IS_UDPLITE(sk
);
717 int offset
= skb_transport_offset(skb
);
718 int len
= skb
->len
- offset
;
722 * Create a UDP header
725 uh
->source
= inet
->inet_sport
;
727 uh
->len
= htons(len
);
730 if (is_udplite
) /* UDP-Lite */
731 csum
= udplite_csum(skb
);
733 else if (sk
->sk_no_check
== UDP_CSUM_NOXMIT
) { /* UDP csum disabled */
735 skb
->ip_summed
= CHECKSUM_NONE
;
738 } else if (skb
->ip_summed
== CHECKSUM_PARTIAL
) { /* UDP hardware csum */
740 udp4_hwcsum(skb
, rt
->rt_src
, daddr
);
744 csum
= udp_csum(skb
);
746 /* add protocol-dependent pseudo-header */
747 uh
->check
= csum_tcpudp_magic(rt
->rt_src
, daddr
, len
,
748 sk
->sk_protocol
, csum
);
750 uh
->check
= CSUM_MANGLED_0
;
753 err
= ip_send_skb(skb
);
755 if (err
== -ENOBUFS
&& !inet
->recverr
) {
756 UDP_INC_STATS_USER(sock_net(sk
),
757 UDP_MIB_SNDBUFERRORS
, is_udplite
);
761 UDP_INC_STATS_USER(sock_net(sk
),
762 UDP_MIB_OUTDATAGRAMS
, is_udplite
);
767 * Push out all pending data as one UDP datagram. Socket is locked.
769 static int udp_push_pending_frames(struct sock
*sk
)
771 struct udp_sock
*up
= udp_sk(sk
);
772 struct inet_sock
*inet
= inet_sk(sk
);
773 struct flowi4
*fl4
= &inet
->cork
.fl
.u
.ip4
;
777 skb
= ip_finish_skb(sk
, fl4
);
781 err
= udp_send_skb(skb
, fl4
->daddr
, fl4
->fl4_dport
);
789 int udp_sendmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
792 struct inet_sock
*inet
= inet_sk(sk
);
793 struct udp_sock
*up
= udp_sk(sk
);
794 struct flowi4 fl4_stack
;
797 struct ipcm_cookie ipc
;
798 struct rtable
*rt
= NULL
;
801 __be32 daddr
, faddr
, saddr
;
804 int err
, is_udplite
= IS_UDPLITE(sk
);
805 int corkreq
= up
->corkflag
|| msg
->msg_flags
&MSG_MORE
;
806 int (*getfrag
)(void *, char *, int, int, int, struct sk_buff
*);
808 struct ip_options_data opt_copy
;
817 if (msg
->msg_flags
& MSG_OOB
) /* Mirror BSD error message compatibility */
823 getfrag
= is_udplite
? udplite_getfrag
: ip_generic_getfrag
;
827 * There are pending frames.
828 * The socket lock must be held while it's corked.
831 if (likely(up
->pending
)) {
832 if (unlikely(up
->pending
!= AF_INET
)) {
840 ulen
+= sizeof(struct udphdr
);
843 * Get and verify the address.
846 struct sockaddr_in
* usin
= (struct sockaddr_in
*)msg
->msg_name
;
847 if (msg
->msg_namelen
< sizeof(*usin
))
849 if (usin
->sin_family
!= AF_INET
) {
850 if (usin
->sin_family
!= AF_UNSPEC
)
851 return -EAFNOSUPPORT
;
854 daddr
= usin
->sin_addr
.s_addr
;
855 dport
= usin
->sin_port
;
859 if (sk
->sk_state
!= TCP_ESTABLISHED
)
860 return -EDESTADDRREQ
;
861 daddr
= inet
->inet_daddr
;
862 dport
= inet
->inet_dport
;
863 /* Open fast path for connected socket.
864 Route will not be used, if at least one option is set.
868 ipc
.addr
= inet
->inet_saddr
;
870 ipc
.oif
= sk
->sk_bound_dev_if
;
871 err
= sock_tx_timestamp(sk
, &ipc
.tx_flags
);
874 if (msg
->msg_controllen
) {
875 err
= ip_cmsg_send(sock_net(sk
), msg
, &ipc
);
883 struct ip_options_rcu
*inet_opt
;
886 inet_opt
= rcu_dereference(inet
->inet_opt
);
888 memcpy(&opt_copy
, inet_opt
,
889 sizeof(*inet_opt
) + inet_opt
->opt
.optlen
);
890 ipc
.opt
= &opt_copy
.opt
;
896 ipc
.addr
= faddr
= daddr
;
898 if (ipc
.opt
&& ipc
.opt
->opt
.srr
) {
901 faddr
= ipc
.opt
->opt
.faddr
;
904 tos
= RT_TOS(inet
->tos
);
905 if (sock_flag(sk
, SOCK_LOCALROUTE
) ||
906 (msg
->msg_flags
& MSG_DONTROUTE
) ||
907 (ipc
.opt
&& ipc
.opt
->opt
.is_strictroute
)) {
912 if (ipv4_is_multicast(daddr
)) {
914 ipc
.oif
= inet
->mc_index
;
916 saddr
= inet
->mc_addr
;
921 rt
= (struct rtable
*)sk_dst_check(sk
, 0);
923 fl4
= &inet
->cork
.fl
.u
.ip4
;
925 struct net
*net
= sock_net(sk
);
928 flowi4_init_output(fl4
, ipc
.oif
, sk
->sk_mark
, tos
,
929 RT_SCOPE_UNIVERSE
, sk
->sk_protocol
,
930 inet_sk_flowi_flags(sk
)|FLOWI_FLAG_CAN_SLEEP
,
931 faddr
, saddr
, dport
, inet
->inet_sport
);
933 security_sk_classify_flow(sk
, flowi4_to_flowi(fl4
));
934 rt
= ip_route_output_flow(net
, fl4
, sk
);
938 if (err
== -ENETUNREACH
)
939 IP_INC_STATS_BH(net
, IPSTATS_MIB_OUTNOROUTES
);
944 if ((rt
->rt_flags
& RTCF_BROADCAST
) &&
945 !sock_flag(sk
, SOCK_BROADCAST
))
948 sk_dst_set(sk
, dst_clone(&rt
->dst
));
951 if (msg
->msg_flags
&MSG_CONFIRM
)
957 daddr
= ipc
.addr
= fl4
->daddr
;
959 /* Lockless fast path for the non-corking case. */
961 skb
= ip_make_skb(sk
, fl4
, getfrag
, msg
->msg_iov
, ulen
,
962 sizeof(struct udphdr
), &ipc
, &rt
,
965 if (skb
&& !IS_ERR(skb
))
966 err
= udp_send_skb(skb
, daddr
, dport
);
971 if (unlikely(up
->pending
)) {
972 /* The socket is already corked while preparing it. */
973 /* ... which is an evident application bug. --ANK */
976 LIMIT_NETDEBUG(KERN_DEBUG
"udp cork app bug 2\n");
981 * Now cork the socket to pend data.
983 fl4
= &inet
->cork
.fl
.u
.ip4
;
986 fl4
->fl4_dport
= dport
;
987 fl4
->fl4_sport
= inet
->inet_sport
;
988 up
->pending
= AF_INET
;
992 err
= ip_append_data(sk
, getfrag
, msg
->msg_iov
, ulen
,
993 sizeof(struct udphdr
), &ipc
, &rt
,
994 corkreq
? msg
->msg_flags
|MSG_MORE
: msg
->msg_flags
);
996 udp_flush_pending_frames(sk
);
998 err
= udp_push_pending_frames(sk
);
999 else if (unlikely(skb_queue_empty(&sk
->sk_write_queue
)))
1010 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1011 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1012 * we don't have a good statistic (IpOutDiscards but it can be too many
1013 * things). We could add another new stat but at least for now that
1014 * seems like overkill.
1016 if (err
== -ENOBUFS
|| test_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
)) {
1017 UDP_INC_STATS_USER(sock_net(sk
),
1018 UDP_MIB_SNDBUFERRORS
, is_udplite
);
1023 dst_confirm(&rt
->dst
);
1024 if (!(msg
->msg_flags
&MSG_PROBE
) || len
)
1025 goto back_from_confirm
;
1029 EXPORT_SYMBOL(udp_sendmsg
);
1031 int udp_sendpage(struct sock
*sk
, struct page
*page
, int offset
,
1032 size_t size
, int flags
)
1034 struct udp_sock
*up
= udp_sk(sk
);
1038 struct msghdr msg
= { .msg_flags
= flags
|MSG_MORE
};
1040 /* Call udp_sendmsg to specify destination address which
1041 * sendpage interface can't pass.
1042 * This will succeed only when the socket is connected.
1044 ret
= udp_sendmsg(NULL
, sk
, &msg
, 0);
1051 if (unlikely(!up
->pending
)) {
1054 LIMIT_NETDEBUG(KERN_DEBUG
"udp cork app bug 3\n");
1058 ret
= ip_append_page(sk
, page
, offset
, size
, flags
);
1059 if (ret
== -EOPNOTSUPP
) {
1061 return sock_no_sendpage(sk
->sk_socket
, page
, offset
,
1065 udp_flush_pending_frames(sk
);
1070 if (!(up
->corkflag
|| (flags
&MSG_MORE
)))
1071 ret
= udp_push_pending_frames(sk
);
1081 * first_packet_length - return length of first packet in receive queue
1084 * Drops all bad checksum frames, until a valid one is found.
1085 * Returns the length of found skb, or 0 if none is found.
1087 static unsigned int first_packet_length(struct sock
*sk
)
1089 struct sk_buff_head list_kill
, *rcvq
= &sk
->sk_receive_queue
;
1090 struct sk_buff
*skb
;
1093 __skb_queue_head_init(&list_kill
);
1095 spin_lock_bh(&rcvq
->lock
);
1096 while ((skb
= skb_peek(rcvq
)) != NULL
&&
1097 udp_lib_checksum_complete(skb
)) {
1098 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
,
1100 atomic_inc(&sk
->sk_drops
);
1101 __skb_unlink(skb
, rcvq
);
1102 __skb_queue_tail(&list_kill
, skb
);
1104 res
= skb
? skb
->len
: 0;
1105 spin_unlock_bh(&rcvq
->lock
);
1107 if (!skb_queue_empty(&list_kill
)) {
1108 bool slow
= lock_sock_fast(sk
);
1110 __skb_queue_purge(&list_kill
);
1111 sk_mem_reclaim_partial(sk
);
1112 unlock_sock_fast(sk
, slow
);
1118 * IOCTL requests applicable to the UDP protocol
1121 int udp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
)
1126 int amount
= sk_wmem_alloc_get(sk
);
1128 return put_user(amount
, (int __user
*)arg
);
1133 unsigned int amount
= first_packet_length(sk
);
1137 * We will only return the amount
1138 * of this packet since that is all
1139 * that will be read.
1141 amount
-= sizeof(struct udphdr
);
1143 return put_user(amount
, (int __user
*)arg
);
1147 return -ENOIOCTLCMD
;
1152 EXPORT_SYMBOL(udp_ioctl
);
1155 * This should be easy, if there is something there we
1156 * return it, otherwise we block.
1159 int udp_recvmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
1160 size_t len
, int noblock
, int flags
, int *addr_len
)
1162 struct inet_sock
*inet
= inet_sk(sk
);
1163 struct sockaddr_in
*sin
= (struct sockaddr_in
*)msg
->msg_name
;
1164 struct sk_buff
*skb
;
1168 int is_udplite
= IS_UDPLITE(sk
);
1172 * Check any passed addresses
1175 *addr_len
= sizeof(*sin
);
1177 if (flags
& MSG_ERRQUEUE
)
1178 return ip_recv_error(sk
, msg
, len
);
1181 skb
= __skb_recv_datagram(sk
, flags
| (noblock
? MSG_DONTWAIT
: 0),
1186 ulen
= skb
->len
- sizeof(struct udphdr
);
1189 else if (len
< ulen
)
1190 msg
->msg_flags
|= MSG_TRUNC
;
1193 * If checksum is needed at all, try to do it while copying the
1194 * data. If the data is truncated, or if we only want a partial
1195 * coverage checksum (UDP-Lite), do it before the copy.
1198 if (len
< ulen
|| UDP_SKB_CB(skb
)->partial_cov
) {
1199 if (udp_lib_checksum_complete(skb
))
1203 if (skb_csum_unnecessary(skb
))
1204 err
= skb_copy_datagram_iovec(skb
, sizeof(struct udphdr
),
1207 err
= skb_copy_and_csum_datagram_iovec(skb
,
1208 sizeof(struct udphdr
),
1219 UDP_INC_STATS_USER(sock_net(sk
),
1220 UDP_MIB_INDATAGRAMS
, is_udplite
);
1222 sock_recv_ts_and_drops(msg
, sk
, skb
);
1224 /* Copy the address. */
1226 sin
->sin_family
= AF_INET
;
1227 sin
->sin_port
= udp_hdr(skb
)->source
;
1228 sin
->sin_addr
.s_addr
= ip_hdr(skb
)->saddr
;
1229 memset(sin
->sin_zero
, 0, sizeof(sin
->sin_zero
));
1231 if (inet
->cmsg_flags
)
1232 ip_cmsg_recv(msg
, skb
);
1235 if (flags
& MSG_TRUNC
)
1239 skb_free_datagram_locked(sk
, skb
);
1244 slow
= lock_sock_fast(sk
);
1245 if (!skb_kill_datagram(sk
, skb
, flags
))
1246 UDP_INC_STATS_USER(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1247 unlock_sock_fast(sk
, slow
);
1255 int udp_disconnect(struct sock
*sk
, int flags
)
1257 struct inet_sock
*inet
= inet_sk(sk
);
1259 * 1003.1g - break association.
1262 sk
->sk_state
= TCP_CLOSE
;
1263 inet
->inet_daddr
= 0;
1264 inet
->inet_dport
= 0;
1265 sock_rps_save_rxhash(sk
, 0);
1266 sk
->sk_bound_dev_if
= 0;
1267 if (!(sk
->sk_userlocks
& SOCK_BINDADDR_LOCK
))
1268 inet_reset_saddr(sk
);
1270 if (!(sk
->sk_userlocks
& SOCK_BINDPORT_LOCK
)) {
1271 sk
->sk_prot
->unhash(sk
);
1272 inet
->inet_sport
= 0;
1277 EXPORT_SYMBOL(udp_disconnect
);
1279 void udp_lib_unhash(struct sock
*sk
)
1281 if (sk_hashed(sk
)) {
1282 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
1283 struct udp_hslot
*hslot
, *hslot2
;
1285 hslot
= udp_hashslot(udptable
, sock_net(sk
),
1286 udp_sk(sk
)->udp_port_hash
);
1287 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
1289 spin_lock_bh(&hslot
->lock
);
1290 if (sk_nulls_del_node_init_rcu(sk
)) {
1292 inet_sk(sk
)->inet_num
= 0;
1293 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, -1);
1295 spin_lock(&hslot2
->lock
);
1296 hlist_nulls_del_init_rcu(&udp_sk(sk
)->udp_portaddr_node
);
1298 spin_unlock(&hslot2
->lock
);
1300 spin_unlock_bh(&hslot
->lock
);
1303 EXPORT_SYMBOL(udp_lib_unhash
);
1306 * inet_rcv_saddr was changed, we must rehash secondary hash
1308 void udp_lib_rehash(struct sock
*sk
, u16 newhash
)
1310 if (sk_hashed(sk
)) {
1311 struct udp_table
*udptable
= sk
->sk_prot
->h
.udp_table
;
1312 struct udp_hslot
*hslot
, *hslot2
, *nhslot2
;
1314 hslot2
= udp_hashslot2(udptable
, udp_sk(sk
)->udp_portaddr_hash
);
1315 nhslot2
= udp_hashslot2(udptable
, newhash
);
1316 udp_sk(sk
)->udp_portaddr_hash
= newhash
;
1317 if (hslot2
!= nhslot2
) {
1318 hslot
= udp_hashslot(udptable
, sock_net(sk
),
1319 udp_sk(sk
)->udp_port_hash
);
1320 /* we must lock primary chain too */
1321 spin_lock_bh(&hslot
->lock
);
1323 spin_lock(&hslot2
->lock
);
1324 hlist_nulls_del_init_rcu(&udp_sk(sk
)->udp_portaddr_node
);
1326 spin_unlock(&hslot2
->lock
);
1328 spin_lock(&nhslot2
->lock
);
1329 hlist_nulls_add_head_rcu(&udp_sk(sk
)->udp_portaddr_node
,
1332 spin_unlock(&nhslot2
->lock
);
1334 spin_unlock_bh(&hslot
->lock
);
1338 EXPORT_SYMBOL(udp_lib_rehash
);
1340 static void udp_v4_rehash(struct sock
*sk
)
1342 u16 new_hash
= udp4_portaddr_hash(sock_net(sk
),
1343 inet_sk(sk
)->inet_rcv_saddr
,
1344 inet_sk(sk
)->inet_num
);
1345 udp_lib_rehash(sk
, new_hash
);
1348 static int __udp_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
1352 if (inet_sk(sk
)->inet_daddr
)
1353 sock_rps_save_rxhash(sk
, skb
->rxhash
);
1355 rc
= ip_queue_rcv_skb(sk
, skb
);
1357 int is_udplite
= IS_UDPLITE(sk
);
1359 /* Note that an ENOMEM error is charged twice */
1361 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_RCVBUFERRORS
,
1363 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1375 * >0: "udp encap" protocol resubmission
1377 * Note that in the success and error cases, the skb is assumed to
1378 * have either been requeued or freed.
1380 int udp_queue_rcv_skb(struct sock
*sk
, struct sk_buff
*skb
)
1382 struct udp_sock
*up
= udp_sk(sk
);
1384 int is_udplite
= IS_UDPLITE(sk
);
1387 * Charge it to the socket, dropping if the queue is full.
1389 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1393 if (up
->encap_type
) {
1395 * This is an encapsulation socket so pass the skb to
1396 * the socket's udp_encap_rcv() hook. Otherwise, just
1397 * fall through and pass this up the UDP socket.
1398 * up->encap_rcv() returns the following value:
1399 * =0 if skb was successfully passed to the encap
1400 * handler or was discarded by it.
1401 * >0 if skb should be passed on to UDP.
1402 * <0 if skb should be resubmitted as proto -N
1405 /* if we're overly short, let UDP handle it */
1406 if (skb
->len
> sizeof(struct udphdr
) &&
1407 up
->encap_rcv
!= NULL
) {
1410 ret
= (*up
->encap_rcv
)(sk
, skb
);
1412 UDP_INC_STATS_BH(sock_net(sk
),
1413 UDP_MIB_INDATAGRAMS
,
1419 /* FALLTHROUGH -- it's a UDP Packet */
1423 * UDP-Lite specific tests, ignored on UDP sockets
1425 if ((is_udplite
& UDPLITE_RECV_CC
) && UDP_SKB_CB(skb
)->partial_cov
) {
1428 * MIB statistics other than incrementing the error count are
1429 * disabled for the following two types of errors: these depend
1430 * on the application settings, not on the functioning of the
1431 * protocol stack as such.
1433 * RFC 3828 here recommends (sec 3.3): "There should also be a
1434 * way ... to ... at least let the receiving application block
1435 * delivery of packets with coverage values less than a value
1436 * provided by the application."
1438 if (up
->pcrlen
== 0) { /* full coverage was set */
1439 LIMIT_NETDEBUG(KERN_WARNING
"UDPLITE: partial coverage "
1440 "%d while full coverage %d requested\n",
1441 UDP_SKB_CB(skb
)->cscov
, skb
->len
);
1444 /* The next case involves violating the min. coverage requested
1445 * by the receiver. This is subtle: if receiver wants x and x is
1446 * greater than the buffersize/MTU then receiver will complain
1447 * that it wants x while sender emits packets of smaller size y.
1448 * Therefore the above ...()->partial_cov statement is essential.
1450 if (UDP_SKB_CB(skb
)->cscov
< up
->pcrlen
) {
1451 LIMIT_NETDEBUG(KERN_WARNING
1452 "UDPLITE: coverage %d too small, need min %d\n",
1453 UDP_SKB_CB(skb
)->cscov
, up
->pcrlen
);
1458 if (rcu_dereference_raw(sk
->sk_filter
)) {
1459 if (udp_lib_checksum_complete(skb
))
1464 if (sk_rcvqueues_full(sk
, skb
))
1470 if (!sock_owned_by_user(sk
))
1471 rc
= __udp_queue_rcv_skb(sk
, skb
);
1472 else if (sk_add_backlog(sk
, skb
)) {
1481 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
, is_udplite
);
1482 atomic_inc(&sk
->sk_drops
);
1488 static void flush_stack(struct sock
**stack
, unsigned int count
,
1489 struct sk_buff
*skb
, unsigned int final
)
1492 struct sk_buff
*skb1
= NULL
;
1495 for (i
= 0; i
< count
; i
++) {
1497 if (likely(skb1
== NULL
))
1498 skb1
= (i
== final
) ? skb
: skb_clone(skb
, GFP_ATOMIC
);
1501 atomic_inc(&sk
->sk_drops
);
1502 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_RCVBUFERRORS
,
1504 UDP_INC_STATS_BH(sock_net(sk
), UDP_MIB_INERRORS
,
1508 if (skb1
&& udp_queue_rcv_skb(sk
, skb1
) <= 0)
1516 * Multicasts and broadcasts go to each listener.
1518 * Note: called only from the BH handler context.
1520 static int __udp4_lib_mcast_deliver(struct net
*net
, struct sk_buff
*skb
,
1522 __be32 saddr
, __be32 daddr
,
1523 struct udp_table
*udptable
)
1525 struct sock
*sk
, *stack
[256 / sizeof(struct sock
*)];
1526 struct udp_hslot
*hslot
= udp_hashslot(udptable
, net
, ntohs(uh
->dest
));
1528 unsigned int i
, count
= 0;
1530 spin_lock(&hslot
->lock
);
1531 sk
= sk_nulls_head(&hslot
->head
);
1532 dif
= skb
->dev
->ifindex
;
1533 sk
= udp_v4_mcast_next(net
, sk
, uh
->dest
, daddr
, uh
->source
, saddr
, dif
);
1535 stack
[count
++] = sk
;
1536 sk
= udp_v4_mcast_next(net
, sk_nulls_next(sk
), uh
->dest
,
1537 daddr
, uh
->source
, saddr
, dif
);
1538 if (unlikely(count
== ARRAY_SIZE(stack
))) {
1541 flush_stack(stack
, count
, skb
, ~0);
1546 * before releasing chain lock, we must take a reference on sockets
1548 for (i
= 0; i
< count
; i
++)
1549 sock_hold(stack
[i
]);
1551 spin_unlock(&hslot
->lock
);
1554 * do the slow work with no lock held
1557 flush_stack(stack
, count
, skb
, count
- 1);
1559 for (i
= 0; i
< count
; i
++)
1567 /* Initialize UDP checksum. If exited with zero value (success),
1568 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1569 * Otherwise, csum completion requires chacksumming packet body,
1570 * including udp header and folding it to skb->csum.
1572 static inline int udp4_csum_init(struct sk_buff
*skb
, struct udphdr
*uh
,
1575 const struct iphdr
*iph
;
1578 UDP_SKB_CB(skb
)->partial_cov
= 0;
1579 UDP_SKB_CB(skb
)->cscov
= skb
->len
;
1581 if (proto
== IPPROTO_UDPLITE
) {
1582 err
= udplite_checksum_init(skb
, uh
);
1588 if (uh
->check
== 0) {
1589 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1590 } else if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
1591 if (!csum_tcpudp_magic(iph
->saddr
, iph
->daddr
, skb
->len
,
1593 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1595 if (!skb_csum_unnecessary(skb
))
1596 skb
->csum
= csum_tcpudp_nofold(iph
->saddr
, iph
->daddr
,
1597 skb
->len
, proto
, 0);
1598 /* Probably, we should checksum udp header (it should be in cache
1599 * in any case) and data in tiny packets (< rx copybreak).
1606 * All we need to do is get the socket, and then do a checksum.
1609 int __udp4_lib_rcv(struct sk_buff
*skb
, struct udp_table
*udptable
,
1614 unsigned short ulen
;
1615 struct rtable
*rt
= skb_rtable(skb
);
1616 __be32 saddr
, daddr
;
1617 struct net
*net
= dev_net(skb
->dev
);
1620 * Validate the packet.
1622 if (!pskb_may_pull(skb
, sizeof(struct udphdr
)))
1623 goto drop
; /* No space for header. */
1626 ulen
= ntohs(uh
->len
);
1627 saddr
= ip_hdr(skb
)->saddr
;
1628 daddr
= ip_hdr(skb
)->daddr
;
1630 if (ulen
> skb
->len
)
1633 if (proto
== IPPROTO_UDP
) {
1634 /* UDP validates ulen. */
1635 if (ulen
< sizeof(*uh
) || pskb_trim_rcsum(skb
, ulen
))
1640 if (udp4_csum_init(skb
, uh
, proto
))
1643 if (rt
->rt_flags
& (RTCF_BROADCAST
|RTCF_MULTICAST
))
1644 return __udp4_lib_mcast_deliver(net
, skb
, uh
,
1645 saddr
, daddr
, udptable
);
1647 sk
= __udp4_lib_lookup_skb(skb
, uh
->source
, uh
->dest
, udptable
);
1650 int ret
= udp_queue_rcv_skb(sk
, skb
);
1653 /* a return value > 0 means to resubmit the input, but
1654 * it wants the return to be -protocol, or 0
1661 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
1665 /* No socket. Drop packet silently, if checksum is wrong */
1666 if (udp_lib_checksum_complete(skb
))
1669 UDP_INC_STATS_BH(net
, UDP_MIB_NOPORTS
, proto
== IPPROTO_UDPLITE
);
1670 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_PORT_UNREACH
, 0);
1673 * Hmm. We got an UDP packet to a port to which we
1674 * don't wanna listen. Ignore it.
1680 LIMIT_NETDEBUG(KERN_DEBUG
"UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1681 proto
== IPPROTO_UDPLITE
? "-Lite" : "",
1692 * RFC1122: OK. Discards the bad packet silently (as far as
1693 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1695 LIMIT_NETDEBUG(KERN_DEBUG
"UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1696 proto
== IPPROTO_UDPLITE
? "-Lite" : "",
1703 UDP_INC_STATS_BH(net
, UDP_MIB_INERRORS
, proto
== IPPROTO_UDPLITE
);
1708 int udp_rcv(struct sk_buff
*skb
)
1710 return __udp4_lib_rcv(skb
, &udp_table
, IPPROTO_UDP
);
1713 void udp_destroy_sock(struct sock
*sk
)
1715 bool slow
= lock_sock_fast(sk
);
1716 udp_flush_pending_frames(sk
);
1717 unlock_sock_fast(sk
, slow
);
1721 * Socket option code for UDP
1723 int udp_lib_setsockopt(struct sock
*sk
, int level
, int optname
,
1724 char __user
*optval
, unsigned int optlen
,
1725 int (*push_pending_frames
)(struct sock
*))
1727 struct udp_sock
*up
= udp_sk(sk
);
1730 int is_udplite
= IS_UDPLITE(sk
);
1732 if (optlen
< sizeof(int))
1735 if (get_user(val
, (int __user
*)optval
))
1745 (*push_pending_frames
)(sk
);
1753 case UDP_ENCAP_ESPINUDP
:
1754 case UDP_ENCAP_ESPINUDP_NON_IKE
:
1755 up
->encap_rcv
= xfrm4_udp_encap_rcv
;
1757 case UDP_ENCAP_L2TPINUDP
:
1758 up
->encap_type
= val
;
1767 * UDP-Lite's partial checksum coverage (RFC 3828).
1769 /* The sender sets actual checksum coverage length via this option.
1770 * The case coverage > packet length is handled by send module. */
1771 case UDPLITE_SEND_CSCOV
:
1772 if (!is_udplite
) /* Disable the option on UDP sockets */
1773 return -ENOPROTOOPT
;
1774 if (val
!= 0 && val
< 8) /* Illegal coverage: use default (8) */
1776 else if (val
> USHRT_MAX
)
1779 up
->pcflag
|= UDPLITE_SEND_CC
;
1782 /* The receiver specifies a minimum checksum coverage value. To make
1783 * sense, this should be set to at least 8 (as done below). If zero is
1784 * used, this again means full checksum coverage. */
1785 case UDPLITE_RECV_CSCOV
:
1786 if (!is_udplite
) /* Disable the option on UDP sockets */
1787 return -ENOPROTOOPT
;
1788 if (val
!= 0 && val
< 8) /* Avoid silly minimal values. */
1790 else if (val
> USHRT_MAX
)
1793 up
->pcflag
|= UDPLITE_RECV_CC
;
1803 EXPORT_SYMBOL(udp_lib_setsockopt
);
1805 int udp_setsockopt(struct sock
*sk
, int level
, int optname
,
1806 char __user
*optval
, unsigned int optlen
)
1808 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1809 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1810 udp_push_pending_frames
);
1811 return ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
1814 #ifdef CONFIG_COMPAT
1815 int compat_udp_setsockopt(struct sock
*sk
, int level
, int optname
,
1816 char __user
*optval
, unsigned int optlen
)
1818 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1819 return udp_lib_setsockopt(sk
, level
, optname
, optval
, optlen
,
1820 udp_push_pending_frames
);
1821 return compat_ip_setsockopt(sk
, level
, optname
, optval
, optlen
);
1825 int udp_lib_getsockopt(struct sock
*sk
, int level
, int optname
,
1826 char __user
*optval
, int __user
*optlen
)
1828 struct udp_sock
*up
= udp_sk(sk
);
1831 if (get_user(len
, optlen
))
1834 len
= min_t(unsigned int, len
, sizeof(int));
1845 val
= up
->encap_type
;
1848 /* The following two cannot be changed on UDP sockets, the return is
1849 * always 0 (which corresponds to the full checksum coverage of UDP). */
1850 case UDPLITE_SEND_CSCOV
:
1854 case UDPLITE_RECV_CSCOV
:
1859 return -ENOPROTOOPT
;
1862 if (put_user(len
, optlen
))
1864 if (copy_to_user(optval
, &val
, len
))
1868 EXPORT_SYMBOL(udp_lib_getsockopt
);
1870 int udp_getsockopt(struct sock
*sk
, int level
, int optname
,
1871 char __user
*optval
, int __user
*optlen
)
1873 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1874 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1875 return ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
1878 #ifdef CONFIG_COMPAT
1879 int compat_udp_getsockopt(struct sock
*sk
, int level
, int optname
,
1880 char __user
*optval
, int __user
*optlen
)
1882 if (level
== SOL_UDP
|| level
== SOL_UDPLITE
)
1883 return udp_lib_getsockopt(sk
, level
, optname
, optval
, optlen
);
1884 return compat_ip_getsockopt(sk
, level
, optname
, optval
, optlen
);
1888 * udp_poll - wait for a UDP event.
1889 * @file - file struct
1891 * @wait - poll table
1893 * This is same as datagram poll, except for the special case of
1894 * blocking sockets. If application is using a blocking fd
1895 * and a packet with checksum error is in the queue;
1896 * then it could get return from select indicating data available
1897 * but then block when reading it. Add special case code
1898 * to work around these arguably broken applications.
1900 unsigned int udp_poll(struct file
*file
, struct socket
*sock
, poll_table
*wait
)
1902 unsigned int mask
= datagram_poll(file
, sock
, wait
);
1903 struct sock
*sk
= sock
->sk
;
1905 /* Check for false positives due to checksum errors */
1906 if ((mask
& POLLRDNORM
) && !(file
->f_flags
& O_NONBLOCK
) &&
1907 !(sk
->sk_shutdown
& RCV_SHUTDOWN
) && !first_packet_length(sk
))
1908 mask
&= ~(POLLIN
| POLLRDNORM
);
1913 EXPORT_SYMBOL(udp_poll
);
1915 struct proto udp_prot
= {
1917 .owner
= THIS_MODULE
,
1918 .close
= udp_lib_close
,
1919 .connect
= ip4_datagram_connect
,
1920 .disconnect
= udp_disconnect
,
1922 .destroy
= udp_destroy_sock
,
1923 .setsockopt
= udp_setsockopt
,
1924 .getsockopt
= udp_getsockopt
,
1925 .sendmsg
= udp_sendmsg
,
1926 .recvmsg
= udp_recvmsg
,
1927 .sendpage
= udp_sendpage
,
1928 .backlog_rcv
= __udp_queue_rcv_skb
,
1929 .hash
= udp_lib_hash
,
1930 .unhash
= udp_lib_unhash
,
1931 .rehash
= udp_v4_rehash
,
1932 .get_port
= udp_v4_get_port
,
1933 .memory_allocated
= &udp_memory_allocated
,
1934 .sysctl_mem
= sysctl_udp_mem
,
1935 .sysctl_wmem
= &sysctl_udp_wmem_min
,
1936 .sysctl_rmem
= &sysctl_udp_rmem_min
,
1937 .obj_size
= sizeof(struct udp_sock
),
1938 .slab_flags
= SLAB_DESTROY_BY_RCU
,
1939 .h
.udp_table
= &udp_table
,
1940 #ifdef CONFIG_COMPAT
1941 .compat_setsockopt
= compat_udp_setsockopt
,
1942 .compat_getsockopt
= compat_udp_getsockopt
,
1944 .clear_sk
= sk_prot_clear_portaddr_nulls
,
1946 EXPORT_SYMBOL(udp_prot
);
1948 /* ------------------------------------------------------------------------ */
1949 #ifdef CONFIG_PROC_FS
1951 static struct sock
*udp_get_first(struct seq_file
*seq
, int start
)
1954 struct udp_iter_state
*state
= seq
->private;
1955 struct net
*net
= seq_file_net(seq
);
1957 for (state
->bucket
= start
; state
->bucket
<= state
->udp_table
->mask
;
1959 struct hlist_nulls_node
*node
;
1960 struct udp_hslot
*hslot
= &state
->udp_table
->hash
[state
->bucket
];
1962 if (hlist_nulls_empty(&hslot
->head
))
1965 spin_lock_bh(&hslot
->lock
);
1966 sk_nulls_for_each(sk
, node
, &hslot
->head
) {
1967 if (!net_eq(sock_net(sk
), net
))
1969 if (sk
->sk_family
== state
->family
)
1972 spin_unlock_bh(&hslot
->lock
);
1979 static struct sock
*udp_get_next(struct seq_file
*seq
, struct sock
*sk
)
1981 struct udp_iter_state
*state
= seq
->private;
1982 struct net
*net
= seq_file_net(seq
);
1985 sk
= sk_nulls_next(sk
);
1986 } while (sk
&& (!net_eq(sock_net(sk
), net
) || sk
->sk_family
!= state
->family
));
1989 if (state
->bucket
<= state
->udp_table
->mask
)
1990 spin_unlock_bh(&state
->udp_table
->hash
[state
->bucket
].lock
);
1991 return udp_get_first(seq
, state
->bucket
+ 1);
1996 static struct sock
*udp_get_idx(struct seq_file
*seq
, loff_t pos
)
1998 struct sock
*sk
= udp_get_first(seq
, 0);
2001 while (pos
&& (sk
= udp_get_next(seq
, sk
)) != NULL
)
2003 return pos
? NULL
: sk
;
2006 static void *udp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2008 struct udp_iter_state
*state
= seq
->private;
2009 state
->bucket
= MAX_UDP_PORTS
;
2011 return *pos
? udp_get_idx(seq
, *pos
-1) : SEQ_START_TOKEN
;
2014 static void *udp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2018 if (v
== SEQ_START_TOKEN
)
2019 sk
= udp_get_idx(seq
, 0);
2021 sk
= udp_get_next(seq
, v
);
2027 static void udp_seq_stop(struct seq_file
*seq
, void *v
)
2029 struct udp_iter_state
*state
= seq
->private;
2031 if (state
->bucket
<= state
->udp_table
->mask
)
2032 spin_unlock_bh(&state
->udp_table
->hash
[state
->bucket
].lock
);
2035 static int udp_seq_open(struct inode
*inode
, struct file
*file
)
2037 struct udp_seq_afinfo
*afinfo
= PDE(inode
)->data
;
2038 struct udp_iter_state
*s
;
2041 err
= seq_open_net(inode
, file
, &afinfo
->seq_ops
,
2042 sizeof(struct udp_iter_state
));
2046 s
= ((struct seq_file
*)file
->private_data
)->private;
2047 s
->family
= afinfo
->family
;
2048 s
->udp_table
= afinfo
->udp_table
;
2052 /* ------------------------------------------------------------------------ */
2053 int udp_proc_register(struct net
*net
, struct udp_seq_afinfo
*afinfo
)
2055 struct proc_dir_entry
*p
;
2058 afinfo
->seq_fops
.open
= udp_seq_open
;
2059 afinfo
->seq_fops
.read
= seq_read
;
2060 afinfo
->seq_fops
.llseek
= seq_lseek
;
2061 afinfo
->seq_fops
.release
= seq_release_net
;
2063 afinfo
->seq_ops
.start
= udp_seq_start
;
2064 afinfo
->seq_ops
.next
= udp_seq_next
;
2065 afinfo
->seq_ops
.stop
= udp_seq_stop
;
2067 p
= proc_create_data(afinfo
->name
, S_IRUGO
, net
->proc_net
,
2068 &afinfo
->seq_fops
, afinfo
);
2073 EXPORT_SYMBOL(udp_proc_register
);
2075 void udp_proc_unregister(struct net
*net
, struct udp_seq_afinfo
*afinfo
)
2077 proc_net_remove(net
, afinfo
->name
);
2079 EXPORT_SYMBOL(udp_proc_unregister
);
2081 /* ------------------------------------------------------------------------ */
2082 static void udp4_format_sock(struct sock
*sp
, struct seq_file
*f
,
2083 int bucket
, int *len
)
2085 struct inet_sock
*inet
= inet_sk(sp
);
2086 __be32 dest
= inet
->inet_daddr
;
2087 __be32 src
= inet
->inet_rcv_saddr
;
2088 __u16 destp
= ntohs(inet
->inet_dport
);
2089 __u16 srcp
= ntohs(inet
->inet_sport
);
2091 seq_printf(f
, "%5d: %08X:%04X %08X:%04X"
2092 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n",
2093 bucket
, src
, srcp
, dest
, destp
, sp
->sk_state
,
2094 sk_wmem_alloc_get(sp
),
2095 sk_rmem_alloc_get(sp
),
2096 0, 0L, 0, sock_i_uid(sp
), 0, sock_i_ino(sp
),
2097 atomic_read(&sp
->sk_refcnt
), sp
,
2098 atomic_read(&sp
->sk_drops
), len
);
2101 int udp4_seq_show(struct seq_file
*seq
, void *v
)
2103 if (v
== SEQ_START_TOKEN
)
2104 seq_printf(seq
, "%-127s\n",
2105 " sl local_address rem_address st tx_queue "
2106 "rx_queue tr tm->when retrnsmt uid timeout "
2107 "inode ref pointer drops");
2109 struct udp_iter_state
*state
= seq
->private;
2112 udp4_format_sock(v
, seq
, state
->bucket
, &len
);
2113 seq_printf(seq
, "%*s\n", 127 - len
, "");
2118 /* ------------------------------------------------------------------------ */
2119 static struct udp_seq_afinfo udp4_seq_afinfo
= {
2122 .udp_table
= &udp_table
,
2124 .owner
= THIS_MODULE
,
2127 .show
= udp4_seq_show
,
2131 static int __net_init
udp4_proc_init_net(struct net
*net
)
2133 return udp_proc_register(net
, &udp4_seq_afinfo
);
2136 static void __net_exit
udp4_proc_exit_net(struct net
*net
)
2138 udp_proc_unregister(net
, &udp4_seq_afinfo
);
2141 static struct pernet_operations udp4_net_ops
= {
2142 .init
= udp4_proc_init_net
,
2143 .exit
= udp4_proc_exit_net
,
2146 int __init
udp4_proc_init(void)
2148 return register_pernet_subsys(&udp4_net_ops
);
2151 void udp4_proc_exit(void)
2153 unregister_pernet_subsys(&udp4_net_ops
);
2155 #endif /* CONFIG_PROC_FS */
2157 static __initdata
unsigned long uhash_entries
;
2158 static int __init
set_uhash_entries(char *str
)
2162 uhash_entries
= simple_strtoul(str
, &str
, 0);
2163 if (uhash_entries
&& uhash_entries
< UDP_HTABLE_SIZE_MIN
)
2164 uhash_entries
= UDP_HTABLE_SIZE_MIN
;
2167 __setup("uhash_entries=", set_uhash_entries
);
2169 void __init
udp_table_init(struct udp_table
*table
, const char *name
)
2173 if (!CONFIG_BASE_SMALL
)
2174 table
->hash
= alloc_large_system_hash(name
,
2175 2 * sizeof(struct udp_hslot
),
2177 21, /* one slot per 2 MB */
2183 * Make sure hash table has the minimum size
2185 if (CONFIG_BASE_SMALL
|| table
->mask
< UDP_HTABLE_SIZE_MIN
- 1) {
2186 table
->hash
= kmalloc(UDP_HTABLE_SIZE_MIN
*
2187 2 * sizeof(struct udp_hslot
), GFP_KERNEL
);
2190 table
->log
= ilog2(UDP_HTABLE_SIZE_MIN
);
2191 table
->mask
= UDP_HTABLE_SIZE_MIN
- 1;
2193 table
->hash2
= table
->hash
+ (table
->mask
+ 1);
2194 for (i
= 0; i
<= table
->mask
; i
++) {
2195 INIT_HLIST_NULLS_HEAD(&table
->hash
[i
].head
, i
);
2196 table
->hash
[i
].count
= 0;
2197 spin_lock_init(&table
->hash
[i
].lock
);
2199 for (i
= 0; i
<= table
->mask
; i
++) {
2200 INIT_HLIST_NULLS_HEAD(&table
->hash2
[i
].head
, i
);
2201 table
->hash2
[i
].count
= 0;
2202 spin_lock_init(&table
->hash2
[i
].lock
);
2206 void __init
udp_init(void)
2208 unsigned long nr_pages
, limit
;
2210 udp_table_init(&udp_table
, "UDP");
2211 /* Set the pressure threshold up by the same strategy of TCP. It is a
2212 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
2213 * toward zero with the amount of memory, with a floor of 128 pages.
2215 nr_pages
= totalram_pages
- totalhigh_pages
;
2216 limit
= min(nr_pages
, 1UL<<(28-PAGE_SHIFT
)) >> (20-PAGE_SHIFT
);
2217 limit
= (limit
* (nr_pages
>> (20-PAGE_SHIFT
))) >> (PAGE_SHIFT
-11);
2218 limit
= max(limit
, 128UL);
2219 sysctl_udp_mem
[0] = limit
/ 4 * 3;
2220 sysctl_udp_mem
[1] = limit
;
2221 sysctl_udp_mem
[2] = sysctl_udp_mem
[0] * 2;
2223 sysctl_udp_rmem_min
= SK_MEM_QUANTUM
;
2224 sysctl_udp_wmem_min
= SK_MEM_QUANTUM
;
2227 int udp4_ufo_send_check(struct sk_buff
*skb
)
2229 const struct iphdr
*iph
;
2232 if (!pskb_may_pull(skb
, sizeof(*uh
)))
2238 uh
->check
= ~csum_tcpudp_magic(iph
->saddr
, iph
->daddr
, skb
->len
,
2240 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
2241 skb
->csum_offset
= offsetof(struct udphdr
, check
);
2242 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2246 struct sk_buff
*udp4_ufo_fragment(struct sk_buff
*skb
, u32 features
)
2248 struct sk_buff
*segs
= ERR_PTR(-EINVAL
);
2253 mss
= skb_shinfo(skb
)->gso_size
;
2254 if (unlikely(skb
->len
<= mss
))
2257 if (skb_gso_ok(skb
, features
| NETIF_F_GSO_ROBUST
)) {
2258 /* Packet is from an untrusted source, reset gso_segs. */
2259 int type
= skb_shinfo(skb
)->gso_type
;
2261 if (unlikely(type
& ~(SKB_GSO_UDP
| SKB_GSO_DODGY
) ||
2262 !(type
& (SKB_GSO_UDP
))))
2265 skb_shinfo(skb
)->gso_segs
= DIV_ROUND_UP(skb
->len
, mss
);
2271 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
2272 * do checksum of UDP packets sent as multiple IP fragments.
2274 offset
= skb_checksum_start_offset(skb
);
2275 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
2276 offset
+= skb
->csum_offset
;
2277 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
2278 skb
->ip_summed
= CHECKSUM_NONE
;
2280 /* Fragment the skb. IP headers of the fragments are updated in
2281 * inet_gso_segment()
2283 segs
= skb_segment(skb
, features
);