blob: c36522a0f1130005f96bd11fc385c42d949c6e80 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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.
5 *
6 * The User Datagram Protocol (UDP).
7 *
Jesper Juhl02c30a82005-05-05 16:16:16 -07008 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
Alan Cox113aa832008-10-13 19:01:08 -070011 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012 * Hirokazu Takahashi, <taka@valinux.co.jp>
13 *
14 * Fixes:
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
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +090021 * Alan Cox : Tidied select() semantics.
22 * Alan Cox : udp_err() fixed properly, also now
Linus Torvalds1da177e2005-04-16 15:20:36 -070023 * 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
28 * does NOT close.
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
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +090057 * for connect.
Linus Torvalds1da177e2005-04-16 15:20:36 -070058 * 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
64 * datagrams.
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
James Chapman342f0232007-06-27 15:37:46 -070071 * James Chapman : Add L2TP encapsulation type.
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 *
73 *
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.
78 */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +090079
Linus Torvalds1da177e2005-04-16 15:20:36 -070080#include <asm/system.h>
81#include <asm/uaccess.h>
82#include <asm/ioctls.h>
Hideo Aoki95766ff2007-12-31 00:29:24 -080083#include <linux/bootmem.h>
Eric Dumazet8203efb2008-10-29 02:32:32 -070084#include <linux/highmem.h>
85#include <linux/swap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070086#include <linux/types.h>
87#include <linux/fcntl.h>
88#include <linux/module.h>
89#include <linux/socket.h>
90#include <linux/sockios.h>
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -020091#include <linux/igmp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070092#include <linux/in.h>
93#include <linux/errno.h>
94#include <linux/timer.h>
95#include <linux/mm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070096#include <linux/inet.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070097#include <linux/netdevice.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090098#include <linux/slab.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070099#include <net/tcp_states.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100#include <linux/skbuff.h>
101#include <linux/proc_fs.h>
102#include <linux/seq_file.h>
Eric W. Biederman457c4cb2007-09-12 12:01:34 +0200103#include <net/net_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104#include <net/icmp.h>
105#include <net/route.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106#include <net/checksum.h>
107#include <net/xfrm.h>
Gerrit Renkerba4e58e2006-11-27 11:10:57 -0800108#include "udp_impl.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109
Eric Dumazetf86dcc52009-10-07 00:37:59 +0000110struct udp_table udp_table __read_mostly;
Eric Dumazet645ca702008-10-29 01:41:45 -0700111EXPORT_SYMBOL(udp_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112
Hideo Aoki95766ff2007-12-31 00:29:24 -0800113int sysctl_udp_mem[3] __read_mostly;
Hideo Aoki95766ff2007-12-31 00:29:24 -0800114EXPORT_SYMBOL(sysctl_udp_mem);
Eric Dumazetc482c562009-07-17 00:26:32 +0000115
116int sysctl_udp_rmem_min __read_mostly;
Hideo Aoki95766ff2007-12-31 00:29:24 -0800117EXPORT_SYMBOL(sysctl_udp_rmem_min);
Eric Dumazetc482c562009-07-17 00:26:32 +0000118
119int sysctl_udp_wmem_min __read_mostly;
Hideo Aoki95766ff2007-12-31 00:29:24 -0800120EXPORT_SYMBOL(sysctl_udp_wmem_min);
121
122atomic_t udp_memory_allocated;
123EXPORT_SYMBOL(udp_memory_allocated);
124
Eric Dumazetf86dcc52009-10-07 00:37:59 +0000125#define MAX_UDP_PORTS 65536
126#define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
Eric Dumazet98322f22009-01-26 21:35:35 -0800127
Eric Dumazetf24d43c2008-10-09 14:51:27 -0700128static int udp_lib_lport_inuse(struct net *net, __u16 num,
Eric Dumazet645ca702008-10-29 01:41:45 -0700129 const struct udp_hslot *hslot,
Eric Dumazet98322f22009-01-26 21:35:35 -0800130 unsigned long *bitmap,
Eric Dumazetf24d43c2008-10-09 14:51:27 -0700131 struct sock *sk,
132 int (*saddr_comp)(const struct sock *sk1,
Eric Dumazetf86dcc52009-10-07 00:37:59 +0000133 const struct sock *sk2),
134 unsigned int log)
Gerrit Renker25030a72006-08-26 20:06:05 -0700135{
Eric Dumazetf24d43c2008-10-09 14:51:27 -0700136 struct sock *sk2;
Eric Dumazet88ab1932008-11-16 19:39:21 -0800137 struct hlist_nulls_node *node;
Gerrit Renker25030a72006-08-26 20:06:05 -0700138
Eric Dumazet88ab1932008-11-16 19:39:21 -0800139 sk_nulls_for_each(sk2, node, &hslot->head)
Joe Perches9d4fb272009-11-23 10:41:23 -0800140 if (net_eq(sock_net(sk2), net) &&
141 sk2 != sk &&
Eric Dumazetd4cada42009-11-08 10:17:30 +0000142 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
Joe Perches9d4fb272009-11-23 10:41:23 -0800143 (!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) &&
Eric Dumazet98322f22009-01-26 21:35:35 -0800146 (*saddr_comp)(sk, sk2)) {
147 if (bitmap)
Eric Dumazetd4cada42009-11-08 10:17:30 +0000148 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
149 bitmap);
Eric Dumazet98322f22009-01-26 21:35:35 -0800150 else
151 return 1;
152 }
Gerrit Renker25030a72006-08-26 20:06:05 -0700153 return 0;
154}
155
Eric Dumazet30fff922009-11-09 05:26:33 +0000156/*
157 * Note: we still hold spinlock of primary hash chain, so no other writer
158 * can insert/delete a socket with local_port == num
159 */
160static int udp_lib_lport_inuse2(struct net *net, __u16 num,
161 struct udp_hslot *hslot2,
162 struct sock *sk,
163 int (*saddr_comp)(const struct sock *sk1,
164 const struct sock *sk2))
165{
166 struct sock *sk2;
167 struct hlist_nulls_node *node;
168 int res = 0;
169
170 spin_lock(&hslot2->lock);
171 udp_portaddr_for_each_entry(sk2, node, &hslot2->head)
Joe Perches9d4fb272009-11-23 10:41:23 -0800172 if (net_eq(sock_net(sk2), net) &&
173 sk2 != sk &&
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) &&
Eric Dumazet30fff922009-11-09 05:26:33 +0000178 (*saddr_comp)(sk, sk2)) {
179 res = 1;
180 break;
181 }
182 spin_unlock(&hslot2->lock);
183 return res;
184}
185
Gerrit Renker25030a72006-08-26 20:06:05 -0700186/**
Pavel Emelyanov6ba5a3c2008-03-22 16:51:21 -0700187 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
Gerrit Renker25030a72006-08-26 20:06:05 -0700188 *
189 * @sk: socket struct in question
190 * @snum: port number to look up
David S. Millerdf2bc452007-06-05 15:18:43 -0700191 * @saddr_comp: AF-dependent comparison of bound local IP addresses
Eric Dumazet30fff922009-11-09 05:26:33 +0000192 * @hash2_nulladdr: AF-dependant hash value in secondary hash chains,
193 * with NULL address
Gerrit Renker25030a72006-08-26 20:06:05 -0700194 */
Pavel Emelyanov6ba5a3c2008-03-22 16:51:21 -0700195int udp_lib_get_port(struct sock *sk, unsigned short snum,
David S. Millerdf2bc452007-06-05 15:18:43 -0700196 int (*saddr_comp)(const struct sock *sk1,
Eric Dumazet30fff922009-11-09 05:26:33 +0000197 const struct sock *sk2),
198 unsigned int hash2_nulladdr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199{
Eric Dumazet512615b2009-11-08 10:17:58 +0000200 struct udp_hslot *hslot, *hslot2;
Eric Dumazet645ca702008-10-29 01:41:45 -0700201 struct udp_table *udptable = sk->sk_prot->h.udp_table;
Gerrit Renker25030a72006-08-26 20:06:05 -0700202 int error = 1;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900203 struct net *net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
Stephen Hemminger32c1da72007-08-24 23:09:41 -0700205 if (!snum) {
Eric Dumazet9088c562008-10-08 11:44:17 -0700206 int low, high, remaining;
207 unsigned rand;
Eric Dumazet98322f22009-01-26 21:35:35 -0800208 unsigned short first, last;
209 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
Stephen Hemminger227b60f2007-10-10 17:30:46 -0700211 inet_get_local_port_range(&low, &high);
Anton Arapova25de532007-10-18 22:00:17 -0700212 remaining = (high - low) + 1;
Stephen Hemminger227b60f2007-10-10 17:30:46 -0700213
Eric Dumazet9088c562008-10-08 11:44:17 -0700214 rand = net_random();
Eric Dumazet98322f22009-01-26 21:35:35 -0800215 first = (((u64)rand * remaining) >> 32) + low;
216 /*
217 * force rand to be an odd multiple of UDP_HTABLE_SIZE
218 */
Eric Dumazetf86dcc52009-10-07 00:37:59 +0000219 rand = (rand | 1) * (udptable->mask + 1);
Eric Dumazet5781b232009-12-13 19:32:39 -0800220 last = first + udptable->mask + 1;
221 do {
Eric Dumazetf86dcc52009-10-07 00:37:59 +0000222 hslot = udp_hashslot(udptable, net, first);
Eric Dumazet98322f22009-01-26 21:35:35 -0800223 bitmap_zero(bitmap, PORTS_PER_CHAIN);
Eric Dumazet645ca702008-10-29 01:41:45 -0700224 spin_lock_bh(&hslot->lock);
Eric Dumazet98322f22009-01-26 21:35:35 -0800225 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
Eric Dumazetf86dcc52009-10-07 00:37:59 +0000226 saddr_comp, udptable->log);
Eric Dumazet98322f22009-01-26 21:35:35 -0800227
228 snum = first;
229 /*
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.
233 */
Eric Dumazet9088c562008-10-08 11:44:17 -0700234 do {
Eric Dumazet98322f22009-01-26 21:35:35 -0800235 if (low <= snum && snum <= high &&
Eric Dumazetf86dcc52009-10-07 00:37:59 +0000236 !test_bit(snum >> udptable->log, bitmap))
Eric Dumazet98322f22009-01-26 21:35:35 -0800237 goto found;
238 snum += rand;
239 } while (snum != first);
240 spin_unlock_bh(&hslot->lock);
Eric Dumazet5781b232009-12-13 19:32:39 -0800241 } while (++first != last);
Eric Dumazet98322f22009-01-26 21:35:35 -0800242 goto fail;
Eric Dumazet645ca702008-10-29 01:41:45 -0700243 } else {
Eric Dumazetf86dcc52009-10-07 00:37:59 +0000244 hslot = udp_hashslot(udptable, net, snum);
Eric Dumazet645ca702008-10-29 01:41:45 -0700245 spin_lock_bh(&hslot->lock);
Eric Dumazet30fff922009-11-09 05:26:33 +0000246 if (hslot->count > 10) {
247 int exist;
248 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
249
250 slot2 &= udptable->mask;
251 hash2_nulladdr &= udptable->mask;
252
253 hslot2 = udp_hashslot2(udptable, slot2);
254 if (hslot->count < hslot2->count)
255 goto scan_primary_hash;
256
257 exist = udp_lib_lport_inuse2(net, snum, hslot2,
258 sk, saddr_comp);
259 if (!exist && (hash2_nulladdr != slot2)) {
260 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
261 exist = udp_lib_lport_inuse2(net, snum, hslot2,
262 sk, saddr_comp);
263 }
264 if (exist)
265 goto fail_unlock;
266 else
267 goto found;
268 }
269scan_primary_hash:
Eric Dumazetf86dcc52009-10-07 00:37:59 +0000270 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
271 saddr_comp, 0))
Eric Dumazet645ca702008-10-29 01:41:45 -0700272 goto fail_unlock;
273 }
Eric Dumazet98322f22009-01-26 21:35:35 -0800274found:
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000275 inet_sk(sk)->inet_num = snum;
Eric Dumazetd4cada42009-11-08 10:17:30 +0000276 udp_sk(sk)->udp_port_hash = snum;
277 udp_sk(sk)->udp_portaddr_hash ^= snum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 if (sk_unhashed(sk)) {
Eric Dumazet88ab1932008-11-16 19:39:21 -0800279 sk_nulls_add_node_rcu(sk, &hslot->head);
Eric Dumazetfdcc8aa92009-11-08 10:17:05 +0000280 hslot->count++;
Pavel Emelyanovc29a0bc2008-03-31 19:41:46 -0700281 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
Eric Dumazet512615b2009-11-08 10:17:58 +0000282
283 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
284 spin_lock(&hslot2->lock);
285 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
286 &hslot2->head);
287 hslot2->count++;
288 spin_unlock(&hslot2->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 }
Gerrit Renker25030a72006-08-26 20:06:05 -0700290 error = 0;
Eric Dumazet645ca702008-10-29 01:41:45 -0700291fail_unlock:
292 spin_unlock_bh(&hslot->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293fail:
Gerrit Renker25030a72006-08-26 20:06:05 -0700294 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295}
Eric Dumazetc482c562009-07-17 00:26:32 +0000296EXPORT_SYMBOL(udp_lib_get_port);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297
Vlad Yasevich499923c2009-04-09 17:37:33 +0000298static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
David S. Millerdb8dac22008-03-06 16:22:02 -0800299{
300 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
301
Eric Dumazetc482c562009-07-17 00:26:32 +0000302 return (!ipv6_only_sock(sk2) &&
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000303 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
304 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
David S. Millerdb8dac22008-03-06 16:22:02 -0800305}
306
Eric Dumazetd4cada42009-11-08 10:17:30 +0000307static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr,
308 unsigned int port)
309{
310 return jhash_1word(saddr, net_hash_mix(net)) ^ port;
311}
312
Pavel Emelyanov6ba5a3c2008-03-22 16:51:21 -0700313int udp_v4_get_port(struct sock *sk, unsigned short snum)
David S. Millerdb8dac22008-03-06 16:22:02 -0800314{
Eric Dumazet30fff922009-11-09 05:26:33 +0000315 unsigned int hash2_nulladdr =
316 udp4_portaddr_hash(sock_net(sk), INADDR_ANY, snum);
317 unsigned int hash2_partial =
318 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
319
Eric Dumazetd4cada42009-11-08 10:17:30 +0000320 /* precompute partial secondary hash */
Eric Dumazet30fff922009-11-09 05:26:33 +0000321 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
322 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
David S. Millerdb8dac22008-03-06 16:22:02 -0800323}
324
Eric Dumazet645ca702008-10-29 01:41:45 -0700325static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
326 unsigned short hnum,
327 __be16 sport, __be32 daddr, __be16 dport, int dif)
328{
329 int score = -1;
330
Eric Dumazetd4cada42009-11-08 10:17:30 +0000331 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
Eric Dumazet645ca702008-10-29 01:41:45 -0700332 !ipv6_only_sock(sk)) {
333 struct inet_sock *inet = inet_sk(sk);
334
335 score = (sk->sk_family == PF_INET ? 1 : 0);
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000336 if (inet->inet_rcv_saddr) {
337 if (inet->inet_rcv_saddr != daddr)
Eric Dumazet645ca702008-10-29 01:41:45 -0700338 return -1;
339 score += 2;
340 }
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000341 if (inet->inet_daddr) {
342 if (inet->inet_daddr != saddr)
Eric Dumazet645ca702008-10-29 01:41:45 -0700343 return -1;
344 score += 2;
345 }
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000346 if (inet->inet_dport) {
347 if (inet->inet_dport != sport)
Eric Dumazet645ca702008-10-29 01:41:45 -0700348 return -1;
349 score += 2;
350 }
351 if (sk->sk_bound_dev_if) {
352 if (sk->sk_bound_dev_if != dif)
353 return -1;
354 score += 2;
355 }
356 }
357 return score;
358}
359
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000360/*
361 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
362 */
363#define SCORE2_MAX (1 + 2 + 2 + 2)
364static inline int compute_score2(struct sock *sk, struct net *net,
365 __be32 saddr, __be16 sport,
366 __be32 daddr, unsigned int hnum, int dif)
367{
368 int score = -1;
369
370 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) {
371 struct inet_sock *inet = inet_sk(sk);
372
373 if (inet->inet_rcv_saddr != daddr)
374 return -1;
375 if (inet->inet_num != hnum)
376 return -1;
377
378 score = (sk->sk_family == PF_INET ? 1 : 0);
379 if (inet->inet_daddr) {
380 if (inet->inet_daddr != saddr)
381 return -1;
382 score += 2;
383 }
384 if (inet->inet_dport) {
385 if (inet->inet_dport != sport)
386 return -1;
387 score += 2;
388 }
389 if (sk->sk_bound_dev_if) {
390 if (sk->sk_bound_dev_if != dif)
391 return -1;
392 score += 2;
393 }
394 }
395 return score;
396}
397
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000398
399/* called with read_rcu_lock() */
400static struct sock *udp4_lib_lookup2(struct net *net,
401 __be32 saddr, __be16 sport,
402 __be32 daddr, unsigned int hnum, int dif,
403 struct udp_hslot *hslot2, unsigned int slot2)
404{
405 struct sock *sk, *result;
406 struct hlist_nulls_node *node;
407 int score, badness;
408
409begin:
410 result = NULL;
411 badness = -1;
412 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
413 score = compute_score2(sk, net, saddr, sport,
414 daddr, hnum, dif);
415 if (score > badness) {
416 result = sk;
417 badness = score;
418 if (score == SCORE2_MAX)
419 goto exact_match;
420 }
421 }
422 /*
423 * if the nulls value we got at the end of this lookup is
424 * not the expected one, we must restart lookup.
425 * We probably met an item that was moved to another chain.
426 */
427 if (get_nulls_value(node) != slot2)
428 goto begin;
429
430 if (result) {
431exact_match:
432 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
433 result = NULL;
434 else if (unlikely(compute_score2(result, net, saddr, sport,
435 daddr, hnum, dif) < badness)) {
436 sock_put(result);
437 goto begin;
438 }
439 }
440 return result;
441}
442
David S. Millerdb8dac22008-03-06 16:22:02 -0800443/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
444 * harder than this. -DaveM
445 */
446static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
447 __be16 sport, __be32 daddr, __be16 dport,
Eric Dumazet645ca702008-10-29 01:41:45 -0700448 int dif, struct udp_table *udptable)
David S. Millerdb8dac22008-03-06 16:22:02 -0800449{
Eric Dumazet271b72c2008-10-29 02:11:14 -0700450 struct sock *sk, *result;
Eric Dumazet88ab1932008-11-16 19:39:21 -0800451 struct hlist_nulls_node *node;
David S. Millerdb8dac22008-03-06 16:22:02 -0800452 unsigned short hnum = ntohs(dport);
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000453 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
454 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
Eric Dumazet271b72c2008-10-29 02:11:14 -0700455 int score, badness;
David S. Millerdb8dac22008-03-06 16:22:02 -0800456
Eric Dumazet271b72c2008-10-29 02:11:14 -0700457 rcu_read_lock();
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000458 if (hslot->count > 10) {
459 hash2 = udp4_portaddr_hash(net, daddr, hnum);
460 slot2 = hash2 & udptable->mask;
461 hslot2 = &udptable->hash2[slot2];
462 if (hslot->count < hslot2->count)
463 goto begin;
464
465 result = udp4_lib_lookup2(net, saddr, sport,
466 daddr, hnum, dif,
467 hslot2, slot2);
468 if (!result) {
469 hash2 = udp4_portaddr_hash(net, INADDR_ANY, hnum);
470 slot2 = hash2 & udptable->mask;
471 hslot2 = &udptable->hash2[slot2];
472 if (hslot->count < hslot2->count)
473 goto begin;
474
Jorge Boncompte [DTI2]1223c672010-04-08 04:56:48 +0000475 result = udp4_lib_lookup2(net, saddr, sport,
476 INADDR_ANY, hnum, dif,
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000477 hslot2, slot2);
478 }
479 rcu_read_unlock();
480 return result;
481 }
Eric Dumazet271b72c2008-10-29 02:11:14 -0700482begin:
483 result = NULL;
484 badness = -1;
Eric Dumazet88ab1932008-11-16 19:39:21 -0800485 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
Eric Dumazet645ca702008-10-29 01:41:45 -0700486 score = compute_score(sk, net, saddr, hnum, sport,
487 daddr, dport, dif);
488 if (score > badness) {
489 result = sk;
490 badness = score;
David S. Millerdb8dac22008-03-06 16:22:02 -0800491 }
492 }
Eric Dumazet88ab1932008-11-16 19:39:21 -0800493 /*
494 * if the nulls value we got at the end of this lookup is
495 * not the expected one, we must restart lookup.
496 * We probably met an item that was moved to another chain.
497 */
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000498 if (get_nulls_value(node) != slot)
Eric Dumazet88ab1932008-11-16 19:39:21 -0800499 goto begin;
500
Eric Dumazet271b72c2008-10-29 02:11:14 -0700501 if (result) {
502 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
503 result = NULL;
504 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
505 daddr, dport, dif) < badness)) {
506 sock_put(result);
507 goto begin;
508 }
509 }
510 rcu_read_unlock();
David S. Millerdb8dac22008-03-06 16:22:02 -0800511 return result;
512}
513
KOVACS Krisztian607c4aa2008-10-07 12:38:32 -0700514static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
515 __be16 sport, __be16 dport,
Eric Dumazet645ca702008-10-29 01:41:45 -0700516 struct udp_table *udptable)
KOVACS Krisztian607c4aa2008-10-07 12:38:32 -0700517{
KOVACS Krisztian23542612008-10-07 12:41:01 -0700518 struct sock *sk;
KOVACS Krisztian607c4aa2008-10-07 12:38:32 -0700519 const struct iphdr *iph = ip_hdr(skb);
520
KOVACS Krisztian23542612008-10-07 12:41:01 -0700521 if (unlikely(sk = skb_steal_sock(skb)))
522 return sk;
523 else
Eric Dumazetadf30902009-06-02 05:19:30 +0000524 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
KOVACS Krisztian23542612008-10-07 12:41:01 -0700525 iph->daddr, dport, inet_iif(skb),
526 udptable);
KOVACS Krisztian607c4aa2008-10-07 12:38:32 -0700527}
528
KOVACS Krisztianbcd41302008-10-01 07:48:10 -0700529struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
530 __be32 daddr, __be16 dport, int dif)
531{
Eric Dumazet645ca702008-10-29 01:41:45 -0700532 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
KOVACS Krisztianbcd41302008-10-01 07:48:10 -0700533}
534EXPORT_SYMBOL_GPL(udp4_lib_lookup);
535
Eric Dumazet920a4612008-11-01 21:22:23 -0700536static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
David S. Millerdb8dac22008-03-06 16:22:02 -0800537 __be16 loc_port, __be32 loc_addr,
538 __be16 rmt_port, __be32 rmt_addr,
539 int dif)
540{
Eric Dumazet88ab1932008-11-16 19:39:21 -0800541 struct hlist_nulls_node *node;
David S. Millerdb8dac22008-03-06 16:22:02 -0800542 struct sock *s = sk;
543 unsigned short hnum = ntohs(loc_port);
544
Eric Dumazet88ab1932008-11-16 19:39:21 -0800545 sk_nulls_for_each_from(s, node) {
David S. Millerdb8dac22008-03-06 16:22:02 -0800546 struct inet_sock *inet = inet_sk(s);
547
Joe Perches9d4fb272009-11-23 10:41:23 -0800548 if (!net_eq(sock_net(s), net) ||
549 udp_sk(s)->udp_port_hash != hnum ||
550 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
551 (inet->inet_dport != rmt_port && inet->inet_dport) ||
552 (inet->inet_rcv_saddr &&
553 inet->inet_rcv_saddr != loc_addr) ||
554 ipv6_only_sock(s) ||
David S. Millerdb8dac22008-03-06 16:22:02 -0800555 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
556 continue;
557 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
558 continue;
559 goto found;
560 }
561 s = NULL;
562found:
563 return s;
564}
565
566/*
567 * This routine is called by the ICMP module when it gets some
568 * sort of error condition. If err < 0 then the socket should
569 * be closed and the error returned to the user. If err > 0
570 * it's just the icmp type << 8 | icmp code.
571 * Header points to the ip header of the error packet. We move
572 * on past this. Then (as it used to claim before adjustment)
573 * header points to the first 8 bytes of the udp header. We need
574 * to find the appropriate port.
575 */
576
Eric Dumazet645ca702008-10-29 01:41:45 -0700577void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
David S. Millerdb8dac22008-03-06 16:22:02 -0800578{
579 struct inet_sock *inet;
Eric Dumazetc482c562009-07-17 00:26:32 +0000580 struct iphdr *iph = (struct iphdr *)skb->data;
581 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
David S. Millerdb8dac22008-03-06 16:22:02 -0800582 const int type = icmp_hdr(skb)->type;
583 const int code = icmp_hdr(skb)->code;
584 struct sock *sk;
585 int harderr;
586 int err;
Pavel Emelyanovfd54d712008-07-14 23:01:40 -0700587 struct net *net = dev_net(skb->dev);
David S. Millerdb8dac22008-03-06 16:22:02 -0800588
Pavel Emelyanovfd54d712008-07-14 23:01:40 -0700589 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
David S. Millerdb8dac22008-03-06 16:22:02 -0800590 iph->saddr, uh->source, skb->dev->ifindex, udptable);
591 if (sk == NULL) {
Pavel Emelyanovdcfc23c2008-07-14 23:03:00 -0700592 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
David S. Millerdb8dac22008-03-06 16:22:02 -0800593 return; /* No socket for error */
594 }
595
596 err = 0;
597 harderr = 0;
598 inet = inet_sk(sk);
599
600 switch (type) {
601 default:
602 case ICMP_TIME_EXCEEDED:
603 err = EHOSTUNREACH;
604 break;
605 case ICMP_SOURCE_QUENCH:
606 goto out;
607 case ICMP_PARAMETERPROB:
608 err = EPROTO;
609 harderr = 1;
610 break;
611 case ICMP_DEST_UNREACH:
612 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
613 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
614 err = EMSGSIZE;
615 harderr = 1;
616 break;
617 }
618 goto out;
619 }
620 err = EHOSTUNREACH;
621 if (code <= NR_ICMP_UNREACH) {
622 harderr = icmp_err_convert[code].fatal;
623 err = icmp_err_convert[code].errno;
624 }
625 break;
626 }
627
628 /*
629 * RFC1122: OK. Passes ICMP errors back to application, as per
630 * 4.1.3.3.
631 */
632 if (!inet->recverr) {
633 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
634 goto out;
635 } else {
Eric Dumazetc482c562009-07-17 00:26:32 +0000636 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
David S. Millerdb8dac22008-03-06 16:22:02 -0800637 }
638 sk->sk_err = err;
639 sk->sk_error_report(sk);
640out:
641 sock_put(sk);
642}
643
644void udp_err(struct sk_buff *skb, u32 info)
645{
Eric Dumazet645ca702008-10-29 01:41:45 -0700646 __udp4_lib_err(skb, info, &udp_table);
David S. Millerdb8dac22008-03-06 16:22:02 -0800647}
648
649/*
650 * Throw away all pending data and cancel the corking. Socket is locked.
651 */
Denis V. Lunev36d926b2008-06-04 15:49:07 +0400652void udp_flush_pending_frames(struct sock *sk)
David S. Millerdb8dac22008-03-06 16:22:02 -0800653{
654 struct udp_sock *up = udp_sk(sk);
655
656 if (up->pending) {
657 up->len = 0;
658 up->pending = 0;
659 ip_flush_pending_frames(sk);
660 }
661}
Denis V. Lunev36d926b2008-06-04 15:49:07 +0400662EXPORT_SYMBOL(udp_flush_pending_frames);
David S. Millerdb8dac22008-03-06 16:22:02 -0800663
664/**
665 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
666 * @sk: socket we are sending on
667 * @skb: sk_buff containing the filled-in UDP header
668 * (checksum field must be zeroed out)
669 */
670static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
Eric Dumazetc482c562009-07-17 00:26:32 +0000671 __be32 src, __be32 dst, int len)
David S. Millerdb8dac22008-03-06 16:22:02 -0800672{
673 unsigned int offset;
674 struct udphdr *uh = udp_hdr(skb);
675 __wsum csum = 0;
676
677 if (skb_queue_len(&sk->sk_write_queue) == 1) {
678 /*
679 * Only one fragment on the socket.
680 */
681 skb->csum_start = skb_transport_header(skb) - skb->head;
682 skb->csum_offset = offsetof(struct udphdr, check);
683 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
684 } else {
685 /*
686 * HW-checksum won't work as there are two or more
687 * fragments on the socket so that all csums of sk_buffs
688 * should be together
689 */
690 offset = skb_transport_offset(skb);
691 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
692
693 skb->ip_summed = CHECKSUM_NONE;
694
695 skb_queue_walk(&sk->sk_write_queue, skb) {
696 csum = csum_add(csum, skb->csum);
697 }
698
699 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
700 if (uh->check == 0)
701 uh->check = CSUM_MANGLED_0;
702 }
703}
704
705/*
706 * Push out all pending data as one UDP datagram. Socket is locked.
707 */
708static int udp_push_pending_frames(struct sock *sk)
709{
710 struct udp_sock *up = udp_sk(sk);
711 struct inet_sock *inet = inet_sk(sk);
712 struct flowi *fl = &inet->cork.fl;
713 struct sk_buff *skb;
714 struct udphdr *uh;
715 int err = 0;
716 int is_udplite = IS_UDPLITE(sk);
717 __wsum csum = 0;
718
719 /* Grab the skbuff where UDP header space exists. */
720 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
721 goto out;
722
723 /*
724 * Create a UDP header
725 */
726 uh = udp_hdr(skb);
727 uh->source = fl->fl_ip_sport;
728 uh->dest = fl->fl_ip_dport;
729 uh->len = htons(up->len);
730 uh->check = 0;
731
732 if (is_udplite) /* UDP-Lite */
733 csum = udplite_csum_outgoing(sk, skb);
734
735 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
736
737 skb->ip_summed = CHECKSUM_NONE;
738 goto send;
739
740 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
741
Eric Dumazetc482c562009-07-17 00:26:32 +0000742 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src, fl->fl4_dst, up->len);
David S. Millerdb8dac22008-03-06 16:22:02 -0800743 goto send;
744
745 } else /* `normal' UDP */
746 csum = udp_csum_outgoing(sk, skb);
747
748 /* add protocol-dependent pseudo-header */
749 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
Eric Dumazetc482c562009-07-17 00:26:32 +0000750 sk->sk_protocol, csum);
David S. Millerdb8dac22008-03-06 16:22:02 -0800751 if (uh->check == 0)
752 uh->check = CSUM_MANGLED_0;
753
754send:
755 err = ip_push_pending_frames(sk);
Eric Dumazet6ce9e7b2009-09-02 18:05:33 -0700756 if (err) {
757 if (err == -ENOBUFS && !inet->recverr) {
758 UDP_INC_STATS_USER(sock_net(sk),
759 UDP_MIB_SNDBUFERRORS, is_udplite);
760 err = 0;
761 }
762 } else
763 UDP_INC_STATS_USER(sock_net(sk),
764 UDP_MIB_OUTDATAGRAMS, is_udplite);
David S. Millerdb8dac22008-03-06 16:22:02 -0800765out:
766 up->len = 0;
767 up->pending = 0;
David S. Millerdb8dac22008-03-06 16:22:02 -0800768 return err;
769}
770
771int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
772 size_t len)
773{
774 struct inet_sock *inet = inet_sk(sk);
775 struct udp_sock *up = udp_sk(sk);
776 int ulen = len;
777 struct ipcm_cookie ipc;
778 struct rtable *rt = NULL;
779 int free = 0;
780 int connected = 0;
781 __be32 daddr, faddr, saddr;
782 __be16 dport;
783 u8 tos;
784 int err, is_udplite = IS_UDPLITE(sk);
785 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
786 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
787
788 if (len > 0xFFFF)
789 return -EMSGSIZE;
790
791 /*
792 * Check the flags.
793 */
794
Eric Dumazetc482c562009-07-17 00:26:32 +0000795 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
David S. Millerdb8dac22008-03-06 16:22:02 -0800796 return -EOPNOTSUPP;
797
798 ipc.opt = NULL;
Patrick Ohly51f31ca2009-02-12 05:03:39 +0000799 ipc.shtx.flags = 0;
David S. Millerdb8dac22008-03-06 16:22:02 -0800800
801 if (up->pending) {
802 /*
803 * There are pending frames.
804 * The socket lock must be held while it's corked.
805 */
806 lock_sock(sk);
807 if (likely(up->pending)) {
808 if (unlikely(up->pending != AF_INET)) {
809 release_sock(sk);
810 return -EINVAL;
811 }
812 goto do_append_data;
813 }
814 release_sock(sk);
815 }
816 ulen += sizeof(struct udphdr);
817
818 /*
819 * Get and verify the address.
820 */
821 if (msg->msg_name) {
Eric Dumazetc482c562009-07-17 00:26:32 +0000822 struct sockaddr_in * usin = (struct sockaddr_in *)msg->msg_name;
David S. Millerdb8dac22008-03-06 16:22:02 -0800823 if (msg->msg_namelen < sizeof(*usin))
824 return -EINVAL;
825 if (usin->sin_family != AF_INET) {
826 if (usin->sin_family != AF_UNSPEC)
827 return -EAFNOSUPPORT;
828 }
829
830 daddr = usin->sin_addr.s_addr;
831 dport = usin->sin_port;
832 if (dport == 0)
833 return -EINVAL;
834 } else {
835 if (sk->sk_state != TCP_ESTABLISHED)
836 return -EDESTADDRREQ;
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000837 daddr = inet->inet_daddr;
838 dport = inet->inet_dport;
David S. Millerdb8dac22008-03-06 16:22:02 -0800839 /* Open fast path for connected socket.
840 Route will not be used, if at least one option is set.
841 */
842 connected = 1;
843 }
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000844 ipc.addr = inet->inet_saddr;
David S. Millerdb8dac22008-03-06 16:22:02 -0800845
846 ipc.oif = sk->sk_bound_dev_if;
Patrick Ohly51f31ca2009-02-12 05:03:39 +0000847 err = sock_tx_timestamp(msg, sk, &ipc.shtx);
848 if (err)
849 return err;
David S. Millerdb8dac22008-03-06 16:22:02 -0800850 if (msg->msg_controllen) {
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900851 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
David S. Millerdb8dac22008-03-06 16:22:02 -0800852 if (err)
853 return err;
854 if (ipc.opt)
855 free = 1;
856 connected = 0;
857 }
858 if (!ipc.opt)
859 ipc.opt = inet->opt;
860
861 saddr = ipc.addr;
862 ipc.addr = faddr = daddr;
863
864 if (ipc.opt && ipc.opt->srr) {
865 if (!daddr)
866 return -EINVAL;
867 faddr = ipc.opt->faddr;
868 connected = 0;
869 }
870 tos = RT_TOS(inet->tos);
871 if (sock_flag(sk, SOCK_LOCALROUTE) ||
872 (msg->msg_flags & MSG_DONTROUTE) ||
873 (ipc.opt && ipc.opt->is_strictroute)) {
874 tos |= RTO_ONLINK;
875 connected = 0;
876 }
877
878 if (ipv4_is_multicast(daddr)) {
879 if (!ipc.oif)
880 ipc.oif = inet->mc_index;
881 if (!saddr)
882 saddr = inet->mc_addr;
883 connected = 0;
884 }
885
886 if (connected)
Eric Dumazetc482c562009-07-17 00:26:32 +0000887 rt = (struct rtable *)sk_dst_check(sk, 0);
David S. Millerdb8dac22008-03-06 16:22:02 -0800888
889 if (rt == NULL) {
890 struct flowi fl = { .oif = ipc.oif,
Atis Elsts914a9ab2009-10-01 15:16:49 -0700891 .mark = sk->sk_mark,
David S. Millerdb8dac22008-03-06 16:22:02 -0800892 .nl_u = { .ip4_u =
893 { .daddr = faddr,
894 .saddr = saddr,
895 .tos = tos } },
896 .proto = sk->sk_protocol,
Balazs Scheidlera134f852008-11-20 01:07:24 -0800897 .flags = inet_sk_flowi_flags(sk),
David S. Millerdb8dac22008-03-06 16:22:02 -0800898 .uli_u = { .ports =
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000899 { .sport = inet->inet_sport,
David S. Millerdb8dac22008-03-06 16:22:02 -0800900 .dport = dport } } };
Pavel Emelyanov84a3aa02008-07-16 20:19:08 -0700901 struct net *net = sock_net(sk);
902
David S. Millerdb8dac22008-03-06 16:22:02 -0800903 security_sk_classify_flow(sk, &fl);
Pavel Emelyanov84a3aa02008-07-16 20:19:08 -0700904 err = ip_route_output_flow(net, &rt, &fl, sk, 1);
David S. Millerdb8dac22008-03-06 16:22:02 -0800905 if (err) {
906 if (err == -ENETUNREACH)
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700907 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
David S. Millerdb8dac22008-03-06 16:22:02 -0800908 goto out;
909 }
910
911 err = -EACCES;
912 if ((rt->rt_flags & RTCF_BROADCAST) &&
913 !sock_flag(sk, SOCK_BROADCAST))
914 goto out;
915 if (connected)
916 sk_dst_set(sk, dst_clone(&rt->u.dst));
917 }
918
919 if (msg->msg_flags&MSG_CONFIRM)
920 goto do_confirm;
921back_from_confirm:
922
923 saddr = rt->rt_src;
924 if (!ipc.addr)
925 daddr = ipc.addr = rt->rt_dst;
926
927 lock_sock(sk);
928 if (unlikely(up->pending)) {
929 /* The socket is already corked while preparing it. */
930 /* ... which is an evident application bug. --ANK */
931 release_sock(sk);
932
933 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
934 err = -EINVAL;
935 goto out;
936 }
937 /*
938 * Now cork the socket to pend data.
939 */
940 inet->cork.fl.fl4_dst = daddr;
941 inet->cork.fl.fl_ip_dport = dport;
942 inet->cork.fl.fl4_src = saddr;
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000943 inet->cork.fl.fl_ip_sport = inet->inet_sport;
David S. Millerdb8dac22008-03-06 16:22:02 -0800944 up->pending = AF_INET;
945
946do_append_data:
947 up->len += ulen;
948 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
949 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
Eric Dumazet2e77d892008-11-24 15:52:46 -0800950 sizeof(struct udphdr), &ipc, &rt,
David S. Millerdb8dac22008-03-06 16:22:02 -0800951 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
952 if (err)
953 udp_flush_pending_frames(sk);
954 else if (!corkreq)
955 err = udp_push_pending_frames(sk);
956 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
957 up->pending = 0;
958 release_sock(sk);
959
960out:
961 ip_rt_put(rt);
962 if (free)
963 kfree(ipc.opt);
964 if (!err)
965 return len;
966 /*
967 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
968 * ENOBUFS might not be good (it's not tunable per se), but otherwise
969 * we don't have a good statistic (IpOutDiscards but it can be too many
970 * things). We could add another new stat but at least for now that
971 * seems like overkill.
972 */
973 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
Pavel Emelyanov629ca232008-07-05 21:18:07 -0700974 UDP_INC_STATS_USER(sock_net(sk),
975 UDP_MIB_SNDBUFERRORS, is_udplite);
David S. Millerdb8dac22008-03-06 16:22:02 -0800976 }
977 return err;
978
979do_confirm:
980 dst_confirm(&rt->u.dst);
981 if (!(msg->msg_flags&MSG_PROBE) || len)
982 goto back_from_confirm;
983 err = 0;
984 goto out;
985}
Eric Dumazetc482c562009-07-17 00:26:32 +0000986EXPORT_SYMBOL(udp_sendmsg);
David S. Millerdb8dac22008-03-06 16:22:02 -0800987
988int udp_sendpage(struct sock *sk, struct page *page, int offset,
989 size_t size, int flags)
990{
991 struct udp_sock *up = udp_sk(sk);
992 int ret;
993
994 if (!up->pending) {
995 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
996
997 /* Call udp_sendmsg to specify destination address which
998 * sendpage interface can't pass.
999 * This will succeed only when the socket is connected.
1000 */
1001 ret = udp_sendmsg(NULL, sk, &msg, 0);
1002 if (ret < 0)
1003 return ret;
1004 }
1005
1006 lock_sock(sk);
1007
1008 if (unlikely(!up->pending)) {
1009 release_sock(sk);
1010
1011 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
1012 return -EINVAL;
1013 }
1014
1015 ret = ip_append_page(sk, page, offset, size, flags);
1016 if (ret == -EOPNOTSUPP) {
1017 release_sock(sk);
1018 return sock_no_sendpage(sk->sk_socket, page, offset,
1019 size, flags);
1020 }
1021 if (ret < 0) {
1022 udp_flush_pending_frames(sk);
1023 goto out;
1024 }
1025
1026 up->len += size;
1027 if (!(up->corkflag || (flags&MSG_MORE)))
1028 ret = udp_push_pending_frames(sk);
1029 if (!ret)
1030 ret = size;
1031out:
1032 release_sock(sk);
1033 return ret;
1034}
1035
Eric Dumazet85584672009-10-09 04:43:40 +00001036
1037/**
1038 * first_packet_length - return length of first packet in receive queue
1039 * @sk: socket
1040 *
1041 * Drops all bad checksum frames, until a valid one is found.
1042 * Returns the length of found skb, or 0 if none is found.
1043 */
1044static unsigned int first_packet_length(struct sock *sk)
1045{
1046 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1047 struct sk_buff *skb;
1048 unsigned int res;
1049
1050 __skb_queue_head_init(&list_kill);
1051
1052 spin_lock_bh(&rcvq->lock);
1053 while ((skb = skb_peek(rcvq)) != NULL &&
1054 udp_lib_checksum_complete(skb)) {
1055 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1056 IS_UDPLITE(sk));
Eric Dumazet8edf19c2009-10-15 00:12:40 +00001057 atomic_inc(&sk->sk_drops);
Eric Dumazet85584672009-10-09 04:43:40 +00001058 __skb_unlink(skb, rcvq);
1059 __skb_queue_tail(&list_kill, skb);
1060 }
1061 res = skb ? skb->len : 0;
1062 spin_unlock_bh(&rcvq->lock);
1063
1064 if (!skb_queue_empty(&list_kill)) {
1065 lock_sock(sk);
1066 __skb_queue_purge(&list_kill);
1067 sk_mem_reclaim_partial(sk);
1068 release_sock(sk);
1069 }
1070 return res;
1071}
1072
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073/*
1074 * IOCTL requests applicable to the UDP protocol
1075 */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001076
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1078{
Stephen Hemminger6516c652007-03-08 20:41:55 -08001079 switch (cmd) {
1080 case SIOCOUTQ:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 {
Eric Dumazet31e6d362009-06-17 19:05:41 -07001082 int amount = sk_wmem_alloc_get(sk);
1083
Stephen Hemminger6516c652007-03-08 20:41:55 -08001084 return put_user(amount, (int __user *)arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 }
Stephen Hemminger6516c652007-03-08 20:41:55 -08001086
1087 case SIOCINQ:
1088 {
Eric Dumazet85584672009-10-09 04:43:40 +00001089 unsigned int amount = first_packet_length(sk);
Stephen Hemminger6516c652007-03-08 20:41:55 -08001090
Eric Dumazet85584672009-10-09 04:43:40 +00001091 if (amount)
Stephen Hemminger6516c652007-03-08 20:41:55 -08001092 /*
1093 * We will only return the amount
1094 * of this packet since that is all
1095 * that will be read.
1096 */
Eric Dumazet85584672009-10-09 04:43:40 +00001097 amount -= sizeof(struct udphdr);
1098
Stephen Hemminger6516c652007-03-08 20:41:55 -08001099 return put_user(amount, (int __user *)arg);
1100 }
1101
1102 default:
1103 return -ENOIOCTLCMD;
1104 }
1105
1106 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107}
Eric Dumazetc482c562009-07-17 00:26:32 +00001108EXPORT_SYMBOL(udp_ioctl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109
David S. Millerdb8dac22008-03-06 16:22:02 -08001110/*
1111 * This should be easy, if there is something there we
1112 * return it, otherwise we block.
1113 */
1114
1115int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1116 size_t len, int noblock, int flags, int *addr_len)
1117{
1118 struct inet_sock *inet = inet_sk(sk);
1119 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
1120 struct sk_buff *skb;
Gerrit Renker81d54ec2010-02-10 20:26:19 +00001121 unsigned int ulen;
David S. Millerdb8dac22008-03-06 16:22:02 -08001122 int peeked;
1123 int err;
1124 int is_udplite = IS_UDPLITE(sk);
1125
1126 /*
1127 * Check any passed addresses
1128 */
1129 if (addr_len)
Eric Dumazetc482c562009-07-17 00:26:32 +00001130 *addr_len = sizeof(*sin);
David S. Millerdb8dac22008-03-06 16:22:02 -08001131
1132 if (flags & MSG_ERRQUEUE)
1133 return ip_recv_error(sk, msg, len);
1134
1135try_again:
1136 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1137 &peeked, &err);
1138 if (!skb)
1139 goto out;
1140
1141 ulen = skb->len - sizeof(struct udphdr);
Gerrit Renker81d54ec2010-02-10 20:26:19 +00001142 if (len > ulen)
1143 len = ulen;
1144 else if (len < ulen)
David S. Millerdb8dac22008-03-06 16:22:02 -08001145 msg->msg_flags |= MSG_TRUNC;
1146
1147 /*
1148 * If checksum is needed at all, try to do it while copying the
1149 * data. If the data is truncated, or if we only want a partial
1150 * coverage checksum (UDP-Lite), do it before the copy.
1151 */
1152
Gerrit Renker81d54ec2010-02-10 20:26:19 +00001153 if (len < ulen || UDP_SKB_CB(skb)->partial_cov) {
David S. Millerdb8dac22008-03-06 16:22:02 -08001154 if (udp_lib_checksum_complete(skb))
1155 goto csum_copy_err;
1156 }
1157
1158 if (skb_csum_unnecessary(skb))
1159 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
Gerrit Renker81d54ec2010-02-10 20:26:19 +00001160 msg->msg_iov, len);
David S. Millerdb8dac22008-03-06 16:22:02 -08001161 else {
Eric Dumazetc482c562009-07-17 00:26:32 +00001162 err = skb_copy_and_csum_datagram_iovec(skb,
1163 sizeof(struct udphdr),
1164 msg->msg_iov);
David S. Millerdb8dac22008-03-06 16:22:02 -08001165
1166 if (err == -EINVAL)
1167 goto csum_copy_err;
1168 }
1169
1170 if (err)
1171 goto out_free;
1172
1173 if (!peeked)
Pavel Emelyanov629ca232008-07-05 21:18:07 -07001174 UDP_INC_STATS_USER(sock_net(sk),
1175 UDP_MIB_INDATAGRAMS, is_udplite);
David S. Millerdb8dac22008-03-06 16:22:02 -08001176
Neil Horman3b885782009-10-12 13:26:31 -07001177 sock_recv_ts_and_drops(msg, sk, skb);
David S. Millerdb8dac22008-03-06 16:22:02 -08001178
1179 /* Copy the address. */
Eric Dumazetc482c562009-07-17 00:26:32 +00001180 if (sin) {
David S. Millerdb8dac22008-03-06 16:22:02 -08001181 sin->sin_family = AF_INET;
1182 sin->sin_port = udp_hdr(skb)->source;
1183 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1184 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1185 }
1186 if (inet->cmsg_flags)
1187 ip_cmsg_recv(msg, skb);
1188
Gerrit Renker81d54ec2010-02-10 20:26:19 +00001189 err = len;
David S. Millerdb8dac22008-03-06 16:22:02 -08001190 if (flags & MSG_TRUNC)
1191 err = ulen;
1192
1193out_free:
Eric Dumazet9d410c72009-10-30 05:03:53 +00001194 skb_free_datagram_locked(sk, skb);
David S. Millerdb8dac22008-03-06 16:22:02 -08001195out:
1196 return err;
1197
1198csum_copy_err:
1199 lock_sock(sk);
1200 if (!skb_kill_datagram(sk, skb, flags))
Pavel Emelyanov629ca232008-07-05 21:18:07 -07001201 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
David S. Millerdb8dac22008-03-06 16:22:02 -08001202 release_sock(sk);
1203
1204 if (noblock)
1205 return -EAGAIN;
1206 goto try_again;
1207}
1208
1209
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210int udp_disconnect(struct sock *sk, int flags)
1211{
1212 struct inet_sock *inet = inet_sk(sk);
1213 /*
1214 * 1003.1g - break association.
1215 */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001216
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 sk->sk_state = TCP_CLOSE;
Eric Dumazetc720c7e2009-10-15 06:30:45 +00001218 inet->inet_daddr = 0;
1219 inet->inet_dport = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 sk->sk_bound_dev_if = 0;
1221 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1222 inet_reset_saddr(sk);
1223
1224 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1225 sk->sk_prot->unhash(sk);
Eric Dumazetc720c7e2009-10-15 06:30:45 +00001226 inet->inet_sport = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 }
1228 sk_dst_reset(sk);
1229 return 0;
1230}
Eric Dumazetc482c562009-07-17 00:26:32 +00001231EXPORT_SYMBOL(udp_disconnect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232
Eric Dumazet645ca702008-10-29 01:41:45 -07001233void udp_lib_unhash(struct sock *sk)
1234{
Eric Dumazet723b4612008-11-25 13:55:15 -08001235 if (sk_hashed(sk)) {
1236 struct udp_table *udptable = sk->sk_prot->h.udp_table;
Eric Dumazet512615b2009-11-08 10:17:58 +00001237 struct udp_hslot *hslot, *hslot2;
1238
1239 hslot = udp_hashslot(udptable, sock_net(sk),
1240 udp_sk(sk)->udp_port_hash);
1241 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
Eric Dumazet645ca702008-10-29 01:41:45 -07001242
Eric Dumazet723b4612008-11-25 13:55:15 -08001243 spin_lock_bh(&hslot->lock);
1244 if (sk_nulls_del_node_init_rcu(sk)) {
Eric Dumazetfdcc8aa92009-11-08 10:17:05 +00001245 hslot->count--;
Eric Dumazetc720c7e2009-10-15 06:30:45 +00001246 inet_sk(sk)->inet_num = 0;
Eric Dumazet723b4612008-11-25 13:55:15 -08001247 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
Eric Dumazet512615b2009-11-08 10:17:58 +00001248
1249 spin_lock(&hslot2->lock);
1250 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1251 hslot2->count--;
1252 spin_unlock(&hslot2->lock);
Eric Dumazet723b4612008-11-25 13:55:15 -08001253 }
1254 spin_unlock_bh(&hslot->lock);
Eric Dumazet645ca702008-10-29 01:41:45 -07001255 }
Eric Dumazet645ca702008-10-29 01:41:45 -07001256}
1257EXPORT_SYMBOL(udp_lib_unhash);
1258
Herbert Xu93821772008-09-15 11:48:46 -07001259static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1260{
Eric Dumazet766e90372009-10-14 20:40:11 -07001261 int rc = sock_queue_rcv_skb(sk, skb);
Herbert Xu93821772008-09-15 11:48:46 -07001262
Eric Dumazet766e90372009-10-14 20:40:11 -07001263 if (rc < 0) {
1264 int is_udplite = IS_UDPLITE(sk);
1265
Herbert Xu93821772008-09-15 11:48:46 -07001266 /* Note that an ENOMEM error is charged twice */
Eric Dumazet766e90372009-10-14 20:40:11 -07001267 if (rc == -ENOMEM)
Herbert Xu93821772008-09-15 11:48:46 -07001268 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1269 is_udplite);
Eric Dumazet766e90372009-10-14 20:40:11 -07001270 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1271 kfree_skb(skb);
1272 return -1;
Herbert Xu93821772008-09-15 11:48:46 -07001273 }
1274
1275 return 0;
1276
Herbert Xu93821772008-09-15 11:48:46 -07001277}
1278
David S. Millerdb8dac22008-03-06 16:22:02 -08001279/* returns:
1280 * -1: error
1281 * 0: success
1282 * >0: "udp encap" protocol resubmission
1283 *
1284 * Note that in the success and error cases, the skb is assumed to
1285 * have either been requeued or freed.
1286 */
Eric Dumazetc482c562009-07-17 00:26:32 +00001287int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
David S. Millerdb8dac22008-03-06 16:22:02 -08001288{
1289 struct udp_sock *up = udp_sk(sk);
1290 int rc;
1291 int is_udplite = IS_UDPLITE(sk);
1292
1293 /*
1294 * Charge it to the socket, dropping if the queue is full.
1295 */
1296 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1297 goto drop;
1298 nf_reset(skb);
1299
1300 if (up->encap_type) {
1301 /*
1302 * This is an encapsulation socket so pass the skb to
1303 * the socket's udp_encap_rcv() hook. Otherwise, just
1304 * fall through and pass this up the UDP socket.
1305 * up->encap_rcv() returns the following value:
1306 * =0 if skb was successfully passed to the encap
1307 * handler or was discarded by it.
1308 * >0 if skb should be passed on to UDP.
1309 * <0 if skb should be resubmitted as proto -N
1310 */
1311
1312 /* if we're overly short, let UDP handle it */
1313 if (skb->len > sizeof(struct udphdr) &&
1314 up->encap_rcv != NULL) {
1315 int ret;
1316
1317 ret = (*up->encap_rcv)(sk, skb);
1318 if (ret <= 0) {
Pavel Emelyanov02833282008-07-05 21:18:48 -07001319 UDP_INC_STATS_BH(sock_net(sk),
1320 UDP_MIB_INDATAGRAMS,
David S. Millerdb8dac22008-03-06 16:22:02 -08001321 is_udplite);
1322 return -ret;
1323 }
1324 }
1325
1326 /* FALLTHROUGH -- it's a UDP Packet */
1327 }
1328
1329 /*
1330 * UDP-Lite specific tests, ignored on UDP sockets
1331 */
1332 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1333
1334 /*
1335 * MIB statistics other than incrementing the error count are
1336 * disabled for the following two types of errors: these depend
1337 * on the application settings, not on the functioning of the
1338 * protocol stack as such.
1339 *
1340 * RFC 3828 here recommends (sec 3.3): "There should also be a
1341 * way ... to ... at least let the receiving application block
1342 * delivery of packets with coverage values less than a value
1343 * provided by the application."
1344 */
1345 if (up->pcrlen == 0) { /* full coverage was set */
1346 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1347 "%d while full coverage %d requested\n",
1348 UDP_SKB_CB(skb)->cscov, skb->len);
1349 goto drop;
1350 }
1351 /* The next case involves violating the min. coverage requested
1352 * by the receiver. This is subtle: if receiver wants x and x is
1353 * greater than the buffersize/MTU then receiver will complain
1354 * that it wants x while sender emits packets of smaller size y.
1355 * Therefore the above ...()->partial_cov statement is essential.
1356 */
1357 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1358 LIMIT_NETDEBUG(KERN_WARNING
1359 "UDPLITE: coverage %d too small, need min %d\n",
1360 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1361 goto drop;
1362 }
1363 }
1364
1365 if (sk->sk_filter) {
1366 if (udp_lib_checksum_complete(skb))
1367 goto drop;
1368 }
1369
Herbert Xu93821772008-09-15 11:48:46 -07001370 rc = 0;
David S. Millerdb8dac22008-03-06 16:22:02 -08001371
Herbert Xu93821772008-09-15 11:48:46 -07001372 bh_lock_sock(sk);
1373 if (!sock_owned_by_user(sk))
1374 rc = __udp_queue_rcv_skb(sk, skb);
Zhu Yia3a858f2010-03-04 18:01:47 +00001375 else if (sk_add_backlog(sk, skb)) {
Zhu Yi55349792010-03-04 18:01:42 +00001376 bh_unlock_sock(sk);
1377 goto drop;
1378 }
Herbert Xu93821772008-09-15 11:48:46 -07001379 bh_unlock_sock(sk);
1380
1381 return rc;
David S. Millerdb8dac22008-03-06 16:22:02 -08001382
1383drop:
Pavel Emelyanov02833282008-07-05 21:18:48 -07001384 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
Eric Dumazet8edf19c2009-10-15 00:12:40 +00001385 atomic_inc(&sk->sk_drops);
David S. Millerdb8dac22008-03-06 16:22:02 -08001386 kfree_skb(skb);
1387 return -1;
1388}
1389
Eric Dumazet1240d132009-11-08 10:18:44 +00001390
1391static void flush_stack(struct sock **stack, unsigned int count,
1392 struct sk_buff *skb, unsigned int final)
1393{
1394 unsigned int i;
1395 struct sk_buff *skb1 = NULL;
Eric Dumazetf6b8f322009-11-08 10:20:19 +00001396 struct sock *sk;
Eric Dumazet1240d132009-11-08 10:18:44 +00001397
1398 for (i = 0; i < count; i++) {
Eric Dumazetf6b8f322009-11-08 10:20:19 +00001399 sk = stack[i];
Eric Dumazet1240d132009-11-08 10:18:44 +00001400 if (likely(skb1 == NULL))
1401 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1402
Eric Dumazetf6b8f322009-11-08 10:20:19 +00001403 if (!skb1) {
1404 atomic_inc(&sk->sk_drops);
1405 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1406 IS_UDPLITE(sk));
1407 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1408 IS_UDPLITE(sk));
1409 }
1410
1411 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
Eric Dumazet1240d132009-11-08 10:18:44 +00001412 skb1 = NULL;
1413 }
1414 if (unlikely(skb1))
1415 kfree_skb(skb1);
1416}
1417
David S. Millerdb8dac22008-03-06 16:22:02 -08001418/*
1419 * Multicasts and broadcasts go to each listener.
1420 *
Eric Dumazet1240d132009-11-08 10:18:44 +00001421 * Note: called only from the BH handler context.
David S. Millerdb8dac22008-03-06 16:22:02 -08001422 */
Pavel Emelyanove3163492008-06-16 17:12:11 -07001423static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
David S. Millerdb8dac22008-03-06 16:22:02 -08001424 struct udphdr *uh,
1425 __be32 saddr, __be32 daddr,
Eric Dumazet645ca702008-10-29 01:41:45 -07001426 struct udp_table *udptable)
David S. Millerdb8dac22008-03-06 16:22:02 -08001427{
Eric Dumazet1240d132009-11-08 10:18:44 +00001428 struct sock *sk, *stack[256 / sizeof(struct sock *)];
Eric Dumazetf86dcc52009-10-07 00:37:59 +00001429 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
David S. Millerdb8dac22008-03-06 16:22:02 -08001430 int dif;
Eric Dumazet1240d132009-11-08 10:18:44 +00001431 unsigned int i, count = 0;
David S. Millerdb8dac22008-03-06 16:22:02 -08001432
Eric Dumazet645ca702008-10-29 01:41:45 -07001433 spin_lock(&hslot->lock);
Eric Dumazet88ab1932008-11-16 19:39:21 -08001434 sk = sk_nulls_head(&hslot->head);
David S. Millerdb8dac22008-03-06 16:22:02 -08001435 dif = skb->dev->ifindex;
Eric Dumazet920a4612008-11-01 21:22:23 -07001436 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
Eric Dumazet1240d132009-11-08 10:18:44 +00001437 while (sk) {
1438 stack[count++] = sk;
1439 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1440 daddr, uh->source, saddr, dif);
1441 if (unlikely(count == ARRAY_SIZE(stack))) {
1442 if (!sk)
1443 break;
1444 flush_stack(stack, count, skb, ~0);
1445 count = 0;
1446 }
1447 }
1448 /*
1449 * before releasing chain lock, we must take a reference on sockets
1450 */
1451 for (i = 0; i < count; i++)
1452 sock_hold(stack[i]);
David S. Millerdb8dac22008-03-06 16:22:02 -08001453
Eric Dumazet645ca702008-10-29 01:41:45 -07001454 spin_unlock(&hslot->lock);
Eric Dumazet1240d132009-11-08 10:18:44 +00001455
1456 /*
1457 * do the slow work with no lock held
1458 */
1459 if (count) {
1460 flush_stack(stack, count, skb, count - 1);
1461
1462 for (i = 0; i < count; i++)
1463 sock_put(stack[i]);
1464 } else {
1465 kfree_skb(skb);
1466 }
David S. Millerdb8dac22008-03-06 16:22:02 -08001467 return 0;
1468}
1469
1470/* Initialize UDP checksum. If exited with zero value (success),
1471 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1472 * Otherwise, csum completion requires chacksumming packet body,
1473 * including udp header and folding it to skb->csum.
1474 */
1475static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1476 int proto)
1477{
1478 const struct iphdr *iph;
1479 int err;
1480
1481 UDP_SKB_CB(skb)->partial_cov = 0;
1482 UDP_SKB_CB(skb)->cscov = skb->len;
1483
1484 if (proto == IPPROTO_UDPLITE) {
1485 err = udplite_checksum_init(skb, uh);
1486 if (err)
1487 return err;
1488 }
1489
1490 iph = ip_hdr(skb);
1491 if (uh->check == 0) {
1492 skb->ip_summed = CHECKSUM_UNNECESSARY;
1493 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
Eric Dumazetc482c562009-07-17 00:26:32 +00001494 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
David S. Millerdb8dac22008-03-06 16:22:02 -08001495 proto, skb->csum))
1496 skb->ip_summed = CHECKSUM_UNNECESSARY;
1497 }
1498 if (!skb_csum_unnecessary(skb))
1499 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1500 skb->len, proto, 0);
1501 /* Probably, we should checksum udp header (it should be in cache
1502 * in any case) and data in tiny packets (< rx copybreak).
1503 */
1504
1505 return 0;
1506}
1507
1508/*
1509 * All we need to do is get the socket, and then do a checksum.
1510 */
1511
Eric Dumazet645ca702008-10-29 01:41:45 -07001512int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
David S. Millerdb8dac22008-03-06 16:22:02 -08001513 int proto)
1514{
1515 struct sock *sk;
Jesper Dangaard Brouer7b5e56f2009-02-05 15:05:45 -08001516 struct udphdr *uh;
David S. Millerdb8dac22008-03-06 16:22:02 -08001517 unsigned short ulen;
Eric Dumazetadf30902009-06-02 05:19:30 +00001518 struct rtable *rt = skb_rtable(skb);
Jesper Dangaard Brouer2783ef22009-02-06 01:59:12 -08001519 __be32 saddr, daddr;
Pavel Emelyanov02833282008-07-05 21:18:48 -07001520 struct net *net = dev_net(skb->dev);
David S. Millerdb8dac22008-03-06 16:22:02 -08001521
1522 /*
1523 * Validate the packet.
1524 */
1525 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1526 goto drop; /* No space for header. */
1527
Jesper Dangaard Brouer7b5e56f2009-02-05 15:05:45 -08001528 uh = udp_hdr(skb);
David S. Millerdb8dac22008-03-06 16:22:02 -08001529 ulen = ntohs(uh->len);
Bjørn Morkccc2d972010-05-06 03:44:34 +00001530 saddr = ip_hdr(skb)->saddr;
1531 daddr = ip_hdr(skb)->daddr;
1532
David S. Millerdb8dac22008-03-06 16:22:02 -08001533 if (ulen > skb->len)
1534 goto short_packet;
1535
1536 if (proto == IPPROTO_UDP) {
1537 /* UDP validates ulen. */
1538 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1539 goto short_packet;
1540 uh = udp_hdr(skb);
1541 }
1542
1543 if (udp4_csum_init(skb, uh, proto))
1544 goto csum_error;
1545
1546 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
Pavel Emelyanove3163492008-06-16 17:12:11 -07001547 return __udp4_lib_mcast_deliver(net, skb, uh,
1548 saddr, daddr, udptable);
David S. Millerdb8dac22008-03-06 16:22:02 -08001549
KOVACS Krisztian607c4aa2008-10-07 12:38:32 -07001550 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
David S. Millerdb8dac22008-03-06 16:22:02 -08001551
1552 if (sk != NULL) {
Herbert Xu93821772008-09-15 11:48:46 -07001553 int ret = udp_queue_rcv_skb(sk, skb);
David S. Millerdb8dac22008-03-06 16:22:02 -08001554 sock_put(sk);
1555
1556 /* a return value > 0 means to resubmit the input, but
1557 * it wants the return to be -protocol, or 0
1558 */
1559 if (ret > 0)
1560 return -ret;
1561 return 0;
1562 }
1563
1564 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1565 goto drop;
1566 nf_reset(skb);
1567
1568 /* No socket. Drop packet silently, if checksum is wrong */
1569 if (udp_lib_checksum_complete(skb))
1570 goto csum_error;
1571
Pavel Emelyanov02833282008-07-05 21:18:48 -07001572 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
David S. Millerdb8dac22008-03-06 16:22:02 -08001573 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1574
1575 /*
1576 * Hmm. We got an UDP packet to a port to which we
1577 * don't wanna listen. Ignore it.
1578 */
1579 kfree_skb(skb);
1580 return 0;
1581
1582short_packet:
Harvey Harrison673d57e2008-10-31 00:53:57 -07001583 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
David S. Millerdb8dac22008-03-06 16:22:02 -08001584 proto == IPPROTO_UDPLITE ? "-Lite" : "",
Harvey Harrison673d57e2008-10-31 00:53:57 -07001585 &saddr,
David S. Millerdb8dac22008-03-06 16:22:02 -08001586 ntohs(uh->source),
1587 ulen,
1588 skb->len,
Harvey Harrison673d57e2008-10-31 00:53:57 -07001589 &daddr,
David S. Millerdb8dac22008-03-06 16:22:02 -08001590 ntohs(uh->dest));
1591 goto drop;
1592
1593csum_error:
1594 /*
1595 * RFC1122: OK. Discards the bad packet silently (as far as
1596 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1597 */
Harvey Harrison673d57e2008-10-31 00:53:57 -07001598 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
David S. Millerdb8dac22008-03-06 16:22:02 -08001599 proto == IPPROTO_UDPLITE ? "-Lite" : "",
Harvey Harrison673d57e2008-10-31 00:53:57 -07001600 &saddr,
David S. Millerdb8dac22008-03-06 16:22:02 -08001601 ntohs(uh->source),
Harvey Harrison673d57e2008-10-31 00:53:57 -07001602 &daddr,
David S. Millerdb8dac22008-03-06 16:22:02 -08001603 ntohs(uh->dest),
1604 ulen);
1605drop:
Pavel Emelyanov02833282008-07-05 21:18:48 -07001606 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
David S. Millerdb8dac22008-03-06 16:22:02 -08001607 kfree_skb(skb);
1608 return 0;
1609}
1610
1611int udp_rcv(struct sk_buff *skb)
1612{
Eric Dumazet645ca702008-10-29 01:41:45 -07001613 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
David S. Millerdb8dac22008-03-06 16:22:02 -08001614}
1615
Brian Haley7d06b2e2008-06-14 17:04:49 -07001616void udp_destroy_sock(struct sock *sk)
David S. Millerdb8dac22008-03-06 16:22:02 -08001617{
1618 lock_sock(sk);
1619 udp_flush_pending_frames(sk);
1620 release_sock(sk);
David S. Millerdb8dac22008-03-06 16:22:02 -08001621}
1622
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623/*
1624 * Socket option code for UDP
1625 */
Gerrit Renker4c0a6cb2006-11-27 09:29:59 -08001626int udp_lib_setsockopt(struct sock *sk, int level, int optname,
David S. Millerb7058842009-09-30 16:12:20 -07001627 char __user *optval, unsigned int optlen,
Gerrit Renker4c0a6cb2006-11-27 09:29:59 -08001628 int (*push_pending_frames)(struct sock *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629{
1630 struct udp_sock *up = udp_sk(sk);
1631 int val;
1632 int err = 0;
Wang Chenb2bf1e22007-12-03 22:34:16 +11001633 int is_udplite = IS_UDPLITE(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634
Eric Dumazetc482c562009-07-17 00:26:32 +00001635 if (optlen < sizeof(int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 return -EINVAL;
1637
1638 if (get_user(val, (int __user *)optval))
1639 return -EFAULT;
1640
Stephen Hemminger6516c652007-03-08 20:41:55 -08001641 switch (optname) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 case UDP_CORK:
1643 if (val != 0) {
1644 up->corkflag = 1;
1645 } else {
1646 up->corkflag = 0;
1647 lock_sock(sk);
Gerrit Renker4c0a6cb2006-11-27 09:29:59 -08001648 (*push_pending_frames)(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 release_sock(sk);
1650 }
1651 break;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001652
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 case UDP_ENCAP:
1654 switch (val) {
1655 case 0:
1656 case UDP_ENCAP_ESPINUDP:
1657 case UDP_ENCAP_ESPINUDP_NON_IKE:
James Chapman067b2072007-07-05 17:08:05 -07001658 up->encap_rcv = xfrm4_udp_encap_rcv;
1659 /* FALLTHROUGH */
James Chapman342f0232007-06-27 15:37:46 -07001660 case UDP_ENCAP_L2TPINUDP:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 up->encap_type = val;
1662 break;
1663 default:
1664 err = -ENOPROTOOPT;
1665 break;
1666 }
1667 break;
1668
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08001669 /*
1670 * UDP-Lite's partial checksum coverage (RFC 3828).
1671 */
1672 /* The sender sets actual checksum coverage length via this option.
1673 * The case coverage > packet length is handled by send module. */
1674 case UDPLITE_SEND_CSCOV:
Wang Chenb2bf1e22007-12-03 22:34:16 +11001675 if (!is_udplite) /* Disable the option on UDP sockets */
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08001676 return -ENOPROTOOPT;
1677 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1678 val = 8;
Gerrit Renker47112e22008-07-21 13:35:08 -07001679 else if (val > USHORT_MAX)
1680 val = USHORT_MAX;
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08001681 up->pcslen = val;
1682 up->pcflag |= UDPLITE_SEND_CC;
1683 break;
1684
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001685 /* The receiver specifies a minimum checksum coverage value. To make
1686 * sense, this should be set to at least 8 (as done below). If zero is
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08001687 * used, this again means full checksum coverage. */
1688 case UDPLITE_RECV_CSCOV:
Wang Chenb2bf1e22007-12-03 22:34:16 +11001689 if (!is_udplite) /* Disable the option on UDP sockets */
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08001690 return -ENOPROTOOPT;
1691 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1692 val = 8;
Gerrit Renker47112e22008-07-21 13:35:08 -07001693 else if (val > USHORT_MAX)
1694 val = USHORT_MAX;
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08001695 up->pcrlen = val;
1696 up->pcflag |= UDPLITE_RECV_CC;
1697 break;
1698
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 default:
1700 err = -ENOPROTOOPT;
1701 break;
Stephen Hemminger6516c652007-03-08 20:41:55 -08001702 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703
1704 return err;
1705}
Eric Dumazetc482c562009-07-17 00:26:32 +00001706EXPORT_SYMBOL(udp_lib_setsockopt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707
David S. Millerdb8dac22008-03-06 16:22:02 -08001708int udp_setsockopt(struct sock *sk, int level, int optname,
David S. Millerb7058842009-09-30 16:12:20 -07001709 char __user *optval, unsigned int optlen)
David S. Millerdb8dac22008-03-06 16:22:02 -08001710{
1711 if (level == SOL_UDP || level == SOL_UDPLITE)
1712 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1713 udp_push_pending_frames);
1714 return ip_setsockopt(sk, level, optname, optval, optlen);
1715}
1716
1717#ifdef CONFIG_COMPAT
1718int compat_udp_setsockopt(struct sock *sk, int level, int optname,
David S. Millerb7058842009-09-30 16:12:20 -07001719 char __user *optval, unsigned int optlen)
David S. Millerdb8dac22008-03-06 16:22:02 -08001720{
1721 if (level == SOL_UDP || level == SOL_UDPLITE)
1722 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1723 udp_push_pending_frames);
1724 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1725}
1726#endif
1727
Gerrit Renker4c0a6cb2006-11-27 09:29:59 -08001728int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1729 char __user *optval, int __user *optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730{
1731 struct udp_sock *up = udp_sk(sk);
1732 int val, len;
1733
Eric Dumazetc482c562009-07-17 00:26:32 +00001734 if (get_user(len, optlen))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 return -EFAULT;
1736
1737 len = min_t(unsigned int, len, sizeof(int));
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001738
Stephen Hemminger6516c652007-03-08 20:41:55 -08001739 if (len < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 return -EINVAL;
1741
Stephen Hemminger6516c652007-03-08 20:41:55 -08001742 switch (optname) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 case UDP_CORK:
1744 val = up->corkflag;
1745 break;
1746
1747 case UDP_ENCAP:
1748 val = up->encap_type;
1749 break;
1750
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08001751 /* The following two cannot be changed on UDP sockets, the return is
1752 * always 0 (which corresponds to the full checksum coverage of UDP). */
1753 case UDPLITE_SEND_CSCOV:
1754 val = up->pcslen;
1755 break;
1756
1757 case UDPLITE_RECV_CSCOV:
1758 val = up->pcrlen;
1759 break;
1760
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 default:
1762 return -ENOPROTOOPT;
Stephen Hemminger6516c652007-03-08 20:41:55 -08001763 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764
Stephen Hemminger6516c652007-03-08 20:41:55 -08001765 if (put_user(len, optlen))
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001766 return -EFAULT;
Eric Dumazetc482c562009-07-17 00:26:32 +00001767 if (copy_to_user(optval, &val, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 return -EFAULT;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001769 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770}
Eric Dumazetc482c562009-07-17 00:26:32 +00001771EXPORT_SYMBOL(udp_lib_getsockopt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772
David S. Millerdb8dac22008-03-06 16:22:02 -08001773int udp_getsockopt(struct sock *sk, int level, int optname,
1774 char __user *optval, int __user *optlen)
1775{
1776 if (level == SOL_UDP || level == SOL_UDPLITE)
1777 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1778 return ip_getsockopt(sk, level, optname, optval, optlen);
1779}
1780
1781#ifdef CONFIG_COMPAT
1782int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1783 char __user *optval, int __user *optlen)
1784{
1785 if (level == SOL_UDP || level == SOL_UDPLITE)
1786 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1787 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1788}
1789#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790/**
1791 * udp_poll - wait for a UDP event.
1792 * @file - file struct
1793 * @sock - socket
1794 * @wait - poll table
1795 *
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001796 * This is same as datagram poll, except for the special case of
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 * blocking sockets. If application is using a blocking fd
1798 * and a packet with checksum error is in the queue;
1799 * then it could get return from select indicating data available
1800 * but then block when reading it. Add special case code
1801 * to work around these arguably broken applications.
1802 */
1803unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1804{
1805 unsigned int mask = datagram_poll(file, sock, wait);
1806 struct sock *sk = sock->sk;
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08001807
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 /* Check for false positives due to checksum errors */
Eric Dumazet85584672009-10-09 04:43:40 +00001809 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
1810 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
1811 mask &= ~(POLLIN | POLLRDNORM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812
1813 return mask;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001814
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815}
Eric Dumazetc482c562009-07-17 00:26:32 +00001816EXPORT_SYMBOL(udp_poll);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817
David S. Millerdb8dac22008-03-06 16:22:02 -08001818struct proto udp_prot = {
1819 .name = "UDP",
1820 .owner = THIS_MODULE,
1821 .close = udp_lib_close,
1822 .connect = ip4_datagram_connect,
1823 .disconnect = udp_disconnect,
1824 .ioctl = udp_ioctl,
1825 .destroy = udp_destroy_sock,
1826 .setsockopt = udp_setsockopt,
1827 .getsockopt = udp_getsockopt,
1828 .sendmsg = udp_sendmsg,
1829 .recvmsg = udp_recvmsg,
1830 .sendpage = udp_sendpage,
Herbert Xu93821772008-09-15 11:48:46 -07001831 .backlog_rcv = __udp_queue_rcv_skb,
David S. Millerdb8dac22008-03-06 16:22:02 -08001832 .hash = udp_lib_hash,
1833 .unhash = udp_lib_unhash,
1834 .get_port = udp_v4_get_port,
1835 .memory_allocated = &udp_memory_allocated,
1836 .sysctl_mem = sysctl_udp_mem,
1837 .sysctl_wmem = &sysctl_udp_wmem_min,
1838 .sysctl_rmem = &sysctl_udp_rmem_min,
1839 .obj_size = sizeof(struct udp_sock),
Eric Dumazet271b72c2008-10-29 02:11:14 -07001840 .slab_flags = SLAB_DESTROY_BY_RCU,
Eric Dumazet645ca702008-10-29 01:41:45 -07001841 .h.udp_table = &udp_table,
David S. Millerdb8dac22008-03-06 16:22:02 -08001842#ifdef CONFIG_COMPAT
1843 .compat_setsockopt = compat_udp_setsockopt,
1844 .compat_getsockopt = compat_udp_getsockopt,
1845#endif
David S. Millerdb8dac22008-03-06 16:22:02 -08001846};
Eric Dumazetc482c562009-07-17 00:26:32 +00001847EXPORT_SYMBOL(udp_prot);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848
1849/* ------------------------------------------------------------------------ */
1850#ifdef CONFIG_PROC_FS
1851
Eric Dumazet645ca702008-10-29 01:41:45 -07001852static struct sock *udp_get_first(struct seq_file *seq, int start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853{
1854 struct sock *sk;
1855 struct udp_iter_state *state = seq->private;
Denis V. Lunev6f191ef2008-03-28 18:23:33 -07001856 struct net *net = seq_file_net(seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857
Eric Dumazetf86dcc52009-10-07 00:37:59 +00001858 for (state->bucket = start; state->bucket <= state->udp_table->mask;
1859 ++state->bucket) {
Eric Dumazet88ab1932008-11-16 19:39:21 -08001860 struct hlist_nulls_node *node;
Eric Dumazet645ca702008-10-29 01:41:45 -07001861 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
Eric Dumazetf86dcc52009-10-07 00:37:59 +00001862
1863 if (hlist_nulls_empty(&hslot->head))
1864 continue;
1865
Eric Dumazet645ca702008-10-29 01:41:45 -07001866 spin_lock_bh(&hslot->lock);
Eric Dumazet88ab1932008-11-16 19:39:21 -08001867 sk_nulls_for_each(sk, node, &hslot->head) {
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +09001868 if (!net_eq(sock_net(sk), net))
Daniel Lezcanoa91275e2008-03-21 04:11:58 -07001869 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 if (sk->sk_family == state->family)
1871 goto found;
1872 }
Eric Dumazet645ca702008-10-29 01:41:45 -07001873 spin_unlock_bh(&hslot->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 }
1875 sk = NULL;
1876found:
1877 return sk;
1878}
1879
1880static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1881{
1882 struct udp_iter_state *state = seq->private;
Denis V. Lunev6f191ef2008-03-28 18:23:33 -07001883 struct net *net = seq_file_net(seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884
1885 do {
Eric Dumazet88ab1932008-11-16 19:39:21 -08001886 sk = sk_nulls_next(sk);
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +09001887 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888
Eric Dumazet645ca702008-10-29 01:41:45 -07001889 if (!sk) {
Eric Dumazetf86dcc52009-10-07 00:37:59 +00001890 if (state->bucket <= state->udp_table->mask)
Vitaly Mayatskikh30842f22009-03-23 15:22:33 -07001891 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
Eric Dumazet645ca702008-10-29 01:41:45 -07001892 return udp_get_first(seq, state->bucket + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 }
1894 return sk;
1895}
1896
1897static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1898{
Eric Dumazet645ca702008-10-29 01:41:45 -07001899 struct sock *sk = udp_get_first(seq, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900
1901 if (sk)
Stephen Hemminger6516c652007-03-08 20:41:55 -08001902 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 --pos;
1904 return pos ? NULL : sk;
1905}
1906
1907static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1908{
Vitaly Mayatskikh30842f22009-03-23 15:22:33 -07001909 struct udp_iter_state *state = seq->private;
Eric Dumazetf86dcc52009-10-07 00:37:59 +00001910 state->bucket = MAX_UDP_PORTS;
Vitaly Mayatskikh30842f22009-03-23 15:22:33 -07001911
YOSHIFUJI Hideakib50660f2008-03-31 19:38:15 -07001912 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913}
1914
1915static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1916{
1917 struct sock *sk;
1918
YOSHIFUJI Hideakib50660f2008-03-31 19:38:15 -07001919 if (v == SEQ_START_TOKEN)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 sk = udp_get_idx(seq, 0);
1921 else
1922 sk = udp_get_next(seq, v);
1923
1924 ++*pos;
1925 return sk;
1926}
1927
1928static void udp_seq_stop(struct seq_file *seq, void *v)
1929{
Eric Dumazet645ca702008-10-29 01:41:45 -07001930 struct udp_iter_state *state = seq->private;
1931
Eric Dumazetf86dcc52009-10-07 00:37:59 +00001932 if (state->bucket <= state->udp_table->mask)
Eric Dumazet645ca702008-10-29 01:41:45 -07001933 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934}
1935
1936static int udp_seq_open(struct inode *inode, struct file *file)
1937{
1938 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
Denis V. Luneva2be75c2008-03-28 18:25:06 -07001939 struct udp_iter_state *s;
1940 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941
Denis V. Luneva2be75c2008-03-28 18:25:06 -07001942 err = seq_open_net(inode, file, &afinfo->seq_ops,
1943 sizeof(struct udp_iter_state));
1944 if (err < 0)
1945 return err;
Daniel Lezcanoa91275e2008-03-21 04:11:58 -07001946
Denis V. Luneva2be75c2008-03-28 18:25:06 -07001947 s = ((struct seq_file *)file->private_data)->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 s->family = afinfo->family;
Eric Dumazet645ca702008-10-29 01:41:45 -07001949 s->udp_table = afinfo->udp_table;
Denis V. Luneva2be75c2008-03-28 18:25:06 -07001950 return err;
Daniel Lezcanoa91275e2008-03-21 04:11:58 -07001951}
1952
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953/* ------------------------------------------------------------------------ */
Daniel Lezcano0c96d8c2008-03-21 04:14:17 -07001954int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955{
1956 struct proc_dir_entry *p;
1957 int rc = 0;
1958
Denis V. Lunev3ba94412008-03-28 18:25:32 -07001959 afinfo->seq_fops.open = udp_seq_open;
1960 afinfo->seq_fops.read = seq_read;
1961 afinfo->seq_fops.llseek = seq_lseek;
1962 afinfo->seq_fops.release = seq_release_net;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963
Denis V. Lunevdda61922008-03-28 18:24:26 -07001964 afinfo->seq_ops.start = udp_seq_start;
1965 afinfo->seq_ops.next = udp_seq_next;
1966 afinfo->seq_ops.stop = udp_seq_stop;
1967
Denis V. Lunev84841c32008-05-02 04:10:08 -07001968 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
1969 &afinfo->seq_fops, afinfo);
1970 if (!p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 rc = -ENOMEM;
1972 return rc;
1973}
Eric Dumazetc482c562009-07-17 00:26:32 +00001974EXPORT_SYMBOL(udp_proc_register);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975
Daniel Lezcano0c96d8c2008-03-21 04:14:17 -07001976void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977{
Daniel Lezcano0c96d8c2008-03-21 04:14:17 -07001978 proc_net_remove(net, afinfo->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979}
Eric Dumazetc482c562009-07-17 00:26:32 +00001980EXPORT_SYMBOL(udp_proc_unregister);
David S. Millerdb8dac22008-03-06 16:22:02 -08001981
1982/* ------------------------------------------------------------------------ */
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07001983static void udp4_format_sock(struct sock *sp, struct seq_file *f,
1984 int bucket, int *len)
David S. Millerdb8dac22008-03-06 16:22:02 -08001985{
1986 struct inet_sock *inet = inet_sk(sp);
Eric Dumazetc720c7e2009-10-15 06:30:45 +00001987 __be32 dest = inet->inet_daddr;
1988 __be32 src = inet->inet_rcv_saddr;
1989 __u16 destp = ntohs(inet->inet_dport);
1990 __u16 srcp = ntohs(inet->inet_sport);
David S. Millerdb8dac22008-03-06 16:22:02 -08001991
Eric Dumazetf86dcc52009-10-07 00:37:59 +00001992 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
Eric Dumazetcb61cb92008-06-17 21:04:56 -07001993 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n",
David S. Millerdb8dac22008-03-06 16:22:02 -08001994 bucket, src, srcp, dest, destp, sp->sk_state,
Eric Dumazet31e6d362009-06-17 19:05:41 -07001995 sk_wmem_alloc_get(sp),
1996 sk_rmem_alloc_get(sp),
David S. Millerdb8dac22008-03-06 16:22:02 -08001997 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
Eric Dumazetcb61cb92008-06-17 21:04:56 -07001998 atomic_read(&sp->sk_refcnt), sp,
1999 atomic_read(&sp->sk_drops), len);
David S. Millerdb8dac22008-03-06 16:22:02 -08002000}
2001
2002int udp4_seq_show(struct seq_file *seq, void *v)
2003{
2004 if (v == SEQ_START_TOKEN)
2005 seq_printf(seq, "%-127s\n",
2006 " sl local_address rem_address st tx_queue "
2007 "rx_queue tr tm->when retrnsmt uid timeout "
Eric Dumazetcb61cb92008-06-17 21:04:56 -07002008 "inode ref pointer drops");
David S. Millerdb8dac22008-03-06 16:22:02 -08002009 else {
David S. Millerdb8dac22008-03-06 16:22:02 -08002010 struct udp_iter_state *state = seq->private;
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002011 int len;
David S. Millerdb8dac22008-03-06 16:22:02 -08002012
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002013 udp4_format_sock(v, seq, state->bucket, &len);
Eric Dumazetc482c562009-07-17 00:26:32 +00002014 seq_printf(seq, "%*s\n", 127 - len, "");
David S. Millerdb8dac22008-03-06 16:22:02 -08002015 }
2016 return 0;
2017}
2018
2019/* ------------------------------------------------------------------------ */
David S. Millerdb8dac22008-03-06 16:22:02 -08002020static struct udp_seq_afinfo udp4_seq_afinfo = {
David S. Millerdb8dac22008-03-06 16:22:02 -08002021 .name = "udp",
2022 .family = AF_INET,
Eric Dumazet645ca702008-10-29 01:41:45 -07002023 .udp_table = &udp_table,
Denis V. Lunev4ad96d32008-03-28 18:25:53 -07002024 .seq_fops = {
2025 .owner = THIS_MODULE,
2026 },
Denis V. Lunevdda61922008-03-28 18:24:26 -07002027 .seq_ops = {
2028 .show = udp4_seq_show,
2029 },
David S. Millerdb8dac22008-03-06 16:22:02 -08002030};
2031
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00002032static int __net_init udp4_proc_init_net(struct net *net)
Pavel Emelyanov15439fe2008-03-24 14:53:49 -07002033{
2034 return udp_proc_register(net, &udp4_seq_afinfo);
2035}
2036
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00002037static void __net_exit udp4_proc_exit_net(struct net *net)
Pavel Emelyanov15439fe2008-03-24 14:53:49 -07002038{
2039 udp_proc_unregister(net, &udp4_seq_afinfo);
2040}
2041
2042static struct pernet_operations udp4_net_ops = {
2043 .init = udp4_proc_init_net,
2044 .exit = udp4_proc_exit_net,
2045};
2046
David S. Millerdb8dac22008-03-06 16:22:02 -08002047int __init udp4_proc_init(void)
2048{
Pavel Emelyanov15439fe2008-03-24 14:53:49 -07002049 return register_pernet_subsys(&udp4_net_ops);
David S. Millerdb8dac22008-03-06 16:22:02 -08002050}
2051
2052void udp4_proc_exit(void)
2053{
Pavel Emelyanov15439fe2008-03-24 14:53:49 -07002054 unregister_pernet_subsys(&udp4_net_ops);
David S. Millerdb8dac22008-03-06 16:22:02 -08002055}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056#endif /* CONFIG_PROC_FS */
2057
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002058static __initdata unsigned long uhash_entries;
2059static int __init set_uhash_entries(char *str)
Eric Dumazet645ca702008-10-29 01:41:45 -07002060{
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002061 if (!str)
2062 return 0;
2063 uhash_entries = simple_strtoul(str, &str, 0);
2064 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2065 uhash_entries = UDP_HTABLE_SIZE_MIN;
2066 return 1;
2067}
2068__setup("uhash_entries=", set_uhash_entries);
Eric Dumazet645ca702008-10-29 01:41:45 -07002069
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002070void __init udp_table_init(struct udp_table *table, const char *name)
2071{
2072 unsigned int i;
2073
2074 if (!CONFIG_BASE_SMALL)
2075 table->hash = alloc_large_system_hash(name,
Eric Dumazet512615b2009-11-08 10:17:58 +00002076 2 * sizeof(struct udp_hslot),
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002077 uhash_entries,
2078 21, /* one slot per 2 MB */
2079 0,
2080 &table->log,
2081 &table->mask,
2082 64 * 1024);
2083 /*
2084 * Make sure hash table has the minimum size
2085 */
2086 if (CONFIG_BASE_SMALL || table->mask < UDP_HTABLE_SIZE_MIN - 1) {
2087 table->hash = kmalloc(UDP_HTABLE_SIZE_MIN *
Eric Dumazet512615b2009-11-08 10:17:58 +00002088 2 * sizeof(struct udp_hslot), GFP_KERNEL);
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002089 if (!table->hash)
2090 panic(name);
2091 table->log = ilog2(UDP_HTABLE_SIZE_MIN);
2092 table->mask = UDP_HTABLE_SIZE_MIN - 1;
2093 }
Eric Dumazet512615b2009-11-08 10:17:58 +00002094 table->hash2 = table->hash + (table->mask + 1);
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002095 for (i = 0; i <= table->mask; i++) {
Eric Dumazet88ab1932008-11-16 19:39:21 -08002096 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
Eric Dumazetfdcc8aa92009-11-08 10:17:05 +00002097 table->hash[i].count = 0;
Eric Dumazet645ca702008-10-29 01:41:45 -07002098 spin_lock_init(&table->hash[i].lock);
2099 }
Eric Dumazet512615b2009-11-08 10:17:58 +00002100 for (i = 0; i <= table->mask; i++) {
2101 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2102 table->hash2[i].count = 0;
2103 spin_lock_init(&table->hash2[i].lock);
2104 }
Eric Dumazet645ca702008-10-29 01:41:45 -07002105}
2106
Hideo Aoki95766ff2007-12-31 00:29:24 -08002107void __init udp_init(void)
2108{
Eric Dumazet8203efb2008-10-29 02:32:32 -07002109 unsigned long nr_pages, limit;
Hideo Aoki95766ff2007-12-31 00:29:24 -08002110
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002111 udp_table_init(&udp_table, "UDP");
Hideo Aoki95766ff2007-12-31 00:29:24 -08002112 /* Set the pressure threshold up by the same strategy of TCP. It is a
2113 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
2114 * toward zero with the amount of memory, with a floor of 128 pages.
2115 */
Eric Dumazet8203efb2008-10-29 02:32:32 -07002116 nr_pages = totalram_pages - totalhigh_pages;
2117 limit = min(nr_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
2118 limit = (limit * (nr_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
Hideo Aoki95766ff2007-12-31 00:29:24 -08002119 limit = max(limit, 128UL);
2120 sysctl_udp_mem[0] = limit / 4 * 3;
2121 sysctl_udp_mem[1] = limit;
2122 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2123
2124 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2125 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2126}
2127
Sridhar Samudralad7ca4cc2009-07-09 08:09:47 +00002128int udp4_ufo_send_check(struct sk_buff *skb)
2129{
2130 const struct iphdr *iph;
2131 struct udphdr *uh;
2132
2133 if (!pskb_may_pull(skb, sizeof(*uh)))
2134 return -EINVAL;
2135
2136 iph = ip_hdr(skb);
2137 uh = udp_hdr(skb);
2138
2139 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
2140 IPPROTO_UDP, 0);
2141 skb->csum_start = skb_transport_header(skb) - skb->head;
2142 skb->csum_offset = offsetof(struct udphdr, check);
2143 skb->ip_summed = CHECKSUM_PARTIAL;
2144 return 0;
2145}
2146
2147struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb, int features)
2148{
2149 struct sk_buff *segs = ERR_PTR(-EINVAL);
2150 unsigned int mss;
2151 int offset;
2152 __wsum csum;
2153
2154 mss = skb_shinfo(skb)->gso_size;
2155 if (unlikely(skb->len <= mss))
2156 goto out;
2157
2158 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
2159 /* Packet is from an untrusted source, reset gso_segs. */
2160 int type = skb_shinfo(skb)->gso_type;
2161
2162 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
2163 !(type & (SKB_GSO_UDP))))
2164 goto out;
2165
2166 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
2167
2168 segs = NULL;
2169 goto out;
2170 }
2171
2172 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
2173 * do checksum of UDP packets sent as multiple IP fragments.
2174 */
2175 offset = skb->csum_start - skb_headroom(skb);
Eric Dumazetc482c562009-07-17 00:26:32 +00002176 csum = skb_checksum(skb, offset, skb->len - offset, 0);
Sridhar Samudralad7ca4cc2009-07-09 08:09:47 +00002177 offset += skb->csum_offset;
2178 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2179 skb->ip_summed = CHECKSUM_NONE;
2180
2181 /* Fragment the skb. IP headers of the fragments are updated in
2182 * inet_gso_segment()
2183 */
2184 segs = skb_segment(skb, features);
2185out:
2186 return segs;
2187}
2188