blob: 32e0bef60d0afdbedd3d4cbdd3c57a9f9beafcdc [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 &&
Amerigo Wange3826f12010-05-05 00:27:06 +0000236 !test_bit(snum >> udptable->log, bitmap) &&
237 !inet_is_reserved_local_port(snum))
Eric Dumazet98322f22009-01-26 21:35:35 -0800238 goto found;
239 snum += rand;
240 } while (snum != first);
241 spin_unlock_bh(&hslot->lock);
Eric Dumazet5781b232009-12-13 19:32:39 -0800242 } while (++first != last);
Eric Dumazet98322f22009-01-26 21:35:35 -0800243 goto fail;
Eric Dumazet645ca702008-10-29 01:41:45 -0700244 } else {
Eric Dumazetf86dcc52009-10-07 00:37:59 +0000245 hslot = udp_hashslot(udptable, net, snum);
Eric Dumazet645ca702008-10-29 01:41:45 -0700246 spin_lock_bh(&hslot->lock);
Eric Dumazet30fff922009-11-09 05:26:33 +0000247 if (hslot->count > 10) {
248 int exist;
249 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
250
251 slot2 &= udptable->mask;
252 hash2_nulladdr &= udptable->mask;
253
254 hslot2 = udp_hashslot2(udptable, slot2);
255 if (hslot->count < hslot2->count)
256 goto scan_primary_hash;
257
258 exist = udp_lib_lport_inuse2(net, snum, hslot2,
259 sk, saddr_comp);
260 if (!exist && (hash2_nulladdr != slot2)) {
261 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
262 exist = udp_lib_lport_inuse2(net, snum, hslot2,
263 sk, saddr_comp);
264 }
265 if (exist)
266 goto fail_unlock;
267 else
268 goto found;
269 }
270scan_primary_hash:
Eric Dumazetf86dcc52009-10-07 00:37:59 +0000271 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
272 saddr_comp, 0))
Eric Dumazet645ca702008-10-29 01:41:45 -0700273 goto fail_unlock;
274 }
Eric Dumazet98322f22009-01-26 21:35:35 -0800275found:
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000276 inet_sk(sk)->inet_num = snum;
Eric Dumazetd4cada42009-11-08 10:17:30 +0000277 udp_sk(sk)->udp_port_hash = snum;
278 udp_sk(sk)->udp_portaddr_hash ^= snum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 if (sk_unhashed(sk)) {
Eric Dumazet88ab1932008-11-16 19:39:21 -0800280 sk_nulls_add_node_rcu(sk, &hslot->head);
Eric Dumazetfdcc8aa92009-11-08 10:17:05 +0000281 hslot->count++;
Pavel Emelyanovc29a0bc2008-03-31 19:41:46 -0700282 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
Eric Dumazet512615b2009-11-08 10:17:58 +0000283
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,
287 &hslot2->head);
288 hslot2->count++;
289 spin_unlock(&hslot2->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 }
Gerrit Renker25030a72006-08-26 20:06:05 -0700291 error = 0;
Eric Dumazet645ca702008-10-29 01:41:45 -0700292fail_unlock:
293 spin_unlock_bh(&hslot->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294fail:
Gerrit Renker25030a72006-08-26 20:06:05 -0700295 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296}
Eric Dumazetc482c562009-07-17 00:26:32 +0000297EXPORT_SYMBOL(udp_lib_get_port);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298
Vlad Yasevich499923c2009-04-09 17:37:33 +0000299static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
David S. Millerdb8dac22008-03-06 16:22:02 -0800300{
301 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
302
Eric Dumazetc482c562009-07-17 00:26:32 +0000303 return (!ipv6_only_sock(sk2) &&
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000304 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
305 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
David S. Millerdb8dac22008-03-06 16:22:02 -0800306}
307
Eric Dumazetd4cada42009-11-08 10:17:30 +0000308static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr,
309 unsigned int port)
310{
Eric Dumazet0eae88f2010-04-20 19:06:52 -0700311 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
Eric Dumazetd4cada42009-11-08 10:17:30 +0000312}
313
Pavel Emelyanov6ba5a3c2008-03-22 16:51:21 -0700314int udp_v4_get_port(struct sock *sk, unsigned short snum)
David S. Millerdb8dac22008-03-06 16:22:02 -0800315{
Eric Dumazet30fff922009-11-09 05:26:33 +0000316 unsigned int hash2_nulladdr =
Eric Dumazet0eae88f2010-04-20 19:06:52 -0700317 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
Eric Dumazet30fff922009-11-09 05:26:33 +0000318 unsigned int hash2_partial =
319 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
320
Eric Dumazetd4cada42009-11-08 10:17:30 +0000321 /* precompute partial secondary hash */
Eric Dumazet30fff922009-11-09 05:26:33 +0000322 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
323 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
David S. Millerdb8dac22008-03-06 16:22:02 -0800324}
325
Eric Dumazet645ca702008-10-29 01:41:45 -0700326static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
327 unsigned short hnum,
328 __be16 sport, __be32 daddr, __be16 dport, int dif)
329{
330 int score = -1;
331
Eric Dumazetd4cada42009-11-08 10:17:30 +0000332 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
Eric Dumazet645ca702008-10-29 01:41:45 -0700333 !ipv6_only_sock(sk)) {
334 struct inet_sock *inet = inet_sk(sk);
335
336 score = (sk->sk_family == PF_INET ? 1 : 0);
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000337 if (inet->inet_rcv_saddr) {
338 if (inet->inet_rcv_saddr != daddr)
Eric Dumazet645ca702008-10-29 01:41:45 -0700339 return -1;
340 score += 2;
341 }
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000342 if (inet->inet_daddr) {
343 if (inet->inet_daddr != saddr)
Eric Dumazet645ca702008-10-29 01:41:45 -0700344 return -1;
345 score += 2;
346 }
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000347 if (inet->inet_dport) {
348 if (inet->inet_dport != sport)
Eric Dumazet645ca702008-10-29 01:41:45 -0700349 return -1;
350 score += 2;
351 }
352 if (sk->sk_bound_dev_if) {
353 if (sk->sk_bound_dev_if != dif)
354 return -1;
355 score += 2;
356 }
357 }
358 return score;
359}
360
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000361/*
362 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
363 */
364#define SCORE2_MAX (1 + 2 + 2 + 2)
365static inline int compute_score2(struct sock *sk, struct net *net,
366 __be32 saddr, __be16 sport,
367 __be32 daddr, unsigned int hnum, int dif)
368{
369 int score = -1;
370
371 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) {
372 struct inet_sock *inet = inet_sk(sk);
373
374 if (inet->inet_rcv_saddr != daddr)
375 return -1;
376 if (inet->inet_num != hnum)
377 return -1;
378
379 score = (sk->sk_family == PF_INET ? 1 : 0);
380 if (inet->inet_daddr) {
381 if (inet->inet_daddr != saddr)
382 return -1;
383 score += 2;
384 }
385 if (inet->inet_dport) {
386 if (inet->inet_dport != sport)
387 return -1;
388 score += 2;
389 }
390 if (sk->sk_bound_dev_if) {
391 if (sk->sk_bound_dev_if != dif)
392 return -1;
393 score += 2;
394 }
395 }
396 return score;
397}
398
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000399
400/* called with read_rcu_lock() */
401static 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)
405{
406 struct sock *sk, *result;
407 struct hlist_nulls_node *node;
408 int score, badness;
409
410begin:
411 result = NULL;
412 badness = -1;
413 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
414 score = compute_score2(sk, net, saddr, sport,
415 daddr, hnum, dif);
416 if (score > badness) {
417 result = sk;
418 badness = score;
419 if (score == SCORE2_MAX)
420 goto exact_match;
421 }
422 }
423 /*
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.
427 */
428 if (get_nulls_value(node) != slot2)
429 goto begin;
430
431 if (result) {
432exact_match:
433 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
434 result = NULL;
435 else if (unlikely(compute_score2(result, net, saddr, sport,
436 daddr, hnum, dif) < badness)) {
437 sock_put(result);
438 goto begin;
439 }
440 }
441 return result;
442}
443
David S. Millerdb8dac22008-03-06 16:22:02 -0800444/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
445 * harder than this. -DaveM
446 */
447static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
448 __be16 sport, __be32 daddr, __be16 dport,
Eric Dumazet645ca702008-10-29 01:41:45 -0700449 int dif, struct udp_table *udptable)
David S. Millerdb8dac22008-03-06 16:22:02 -0800450{
Eric Dumazet271b72c2008-10-29 02:11:14 -0700451 struct sock *sk, *result;
Eric Dumazet88ab1932008-11-16 19:39:21 -0800452 struct hlist_nulls_node *node;
David S. Millerdb8dac22008-03-06 16:22:02 -0800453 unsigned short hnum = ntohs(dport);
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000454 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
455 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
Eric Dumazet271b72c2008-10-29 02:11:14 -0700456 int score, badness;
David S. Millerdb8dac22008-03-06 16:22:02 -0800457
Eric Dumazet271b72c2008-10-29 02:11:14 -0700458 rcu_read_lock();
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000459 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)
464 goto begin;
465
466 result = udp4_lib_lookup2(net, saddr, sport,
467 daddr, hnum, dif,
468 hslot2, slot2);
469 if (!result) {
Eric Dumazet0eae88f2010-04-20 19:06:52 -0700470 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000471 slot2 = hash2 & udptable->mask;
472 hslot2 = &udptable->hash2[slot2];
473 if (hslot->count < hslot2->count)
474 goto begin;
475
Jorge Boncompte [DTI2]1223c672010-04-08 04:56:48 +0000476 result = udp4_lib_lookup2(net, saddr, sport,
Eric Dumazet0eae88f2010-04-20 19:06:52 -0700477 htonl(INADDR_ANY), hnum, dif,
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000478 hslot2, slot2);
479 }
480 rcu_read_unlock();
481 return result;
482 }
Eric Dumazet271b72c2008-10-29 02:11:14 -0700483begin:
484 result = NULL;
485 badness = -1;
Eric Dumazet88ab1932008-11-16 19:39:21 -0800486 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
Eric Dumazet645ca702008-10-29 01:41:45 -0700487 score = compute_score(sk, net, saddr, hnum, sport,
488 daddr, dport, dif);
489 if (score > badness) {
490 result = sk;
491 badness = score;
David S. Millerdb8dac22008-03-06 16:22:02 -0800492 }
493 }
Eric Dumazet88ab1932008-11-16 19:39:21 -0800494 /*
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.
498 */
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000499 if (get_nulls_value(node) != slot)
Eric Dumazet88ab1932008-11-16 19:39:21 -0800500 goto begin;
501
Eric Dumazet271b72c2008-10-29 02:11:14 -0700502 if (result) {
503 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
504 result = NULL;
505 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
506 daddr, dport, dif) < badness)) {
507 sock_put(result);
508 goto begin;
509 }
510 }
511 rcu_read_unlock();
David S. Millerdb8dac22008-03-06 16:22:02 -0800512 return result;
513}
514
KOVACS Krisztian607c4aa2008-10-07 12:38:32 -0700515static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
516 __be16 sport, __be16 dport,
Eric Dumazet645ca702008-10-29 01:41:45 -0700517 struct udp_table *udptable)
KOVACS Krisztian607c4aa2008-10-07 12:38:32 -0700518{
KOVACS Krisztian23542612008-10-07 12:41:01 -0700519 struct sock *sk;
KOVACS Krisztian607c4aa2008-10-07 12:38:32 -0700520 const struct iphdr *iph = ip_hdr(skb);
521
KOVACS Krisztian23542612008-10-07 12:41:01 -0700522 if (unlikely(sk = skb_steal_sock(skb)))
523 return sk;
524 else
Eric Dumazetadf30902009-06-02 05:19:30 +0000525 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
KOVACS Krisztian23542612008-10-07 12:41:01 -0700526 iph->daddr, dport, inet_iif(skb),
527 udptable);
KOVACS Krisztian607c4aa2008-10-07 12:38:32 -0700528}
529
KOVACS Krisztianbcd41302008-10-01 07:48:10 -0700530struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
531 __be32 daddr, __be16 dport, int dif)
532{
Eric Dumazet645ca702008-10-29 01:41:45 -0700533 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
KOVACS Krisztianbcd41302008-10-01 07:48:10 -0700534}
535EXPORT_SYMBOL_GPL(udp4_lib_lookup);
536
Eric Dumazet920a4612008-11-01 21:22:23 -0700537static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
David S. Millerdb8dac22008-03-06 16:22:02 -0800538 __be16 loc_port, __be32 loc_addr,
539 __be16 rmt_port, __be32 rmt_addr,
540 int dif)
541{
Eric Dumazet88ab1932008-11-16 19:39:21 -0800542 struct hlist_nulls_node *node;
David S. Millerdb8dac22008-03-06 16:22:02 -0800543 struct sock *s = sk;
544 unsigned short hnum = ntohs(loc_port);
545
Eric Dumazet88ab1932008-11-16 19:39:21 -0800546 sk_nulls_for_each_from(s, node) {
David S. Millerdb8dac22008-03-06 16:22:02 -0800547 struct inet_sock *inet = inet_sk(s);
548
Joe Perches9d4fb272009-11-23 10:41:23 -0800549 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) ||
555 ipv6_only_sock(s) ||
David S. Millerdb8dac22008-03-06 16:22:02 -0800556 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
557 continue;
558 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
559 continue;
560 goto found;
561 }
562 s = NULL;
563found:
564 return s;
565}
566
567/*
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.
576 */
577
Eric Dumazet645ca702008-10-29 01:41:45 -0700578void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
David S. Millerdb8dac22008-03-06 16:22:02 -0800579{
580 struct inet_sock *inet;
Eric Dumazetc482c562009-07-17 00:26:32 +0000581 struct iphdr *iph = (struct iphdr *)skb->data;
582 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
David S. Millerdb8dac22008-03-06 16:22:02 -0800583 const int type = icmp_hdr(skb)->type;
584 const int code = icmp_hdr(skb)->code;
585 struct sock *sk;
586 int harderr;
587 int err;
Pavel Emelyanovfd54d712008-07-14 23:01:40 -0700588 struct net *net = dev_net(skb->dev);
David S. Millerdb8dac22008-03-06 16:22:02 -0800589
Pavel Emelyanovfd54d712008-07-14 23:01:40 -0700590 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
David S. Millerdb8dac22008-03-06 16:22:02 -0800591 iph->saddr, uh->source, skb->dev->ifindex, udptable);
592 if (sk == NULL) {
Pavel Emelyanovdcfc23c2008-07-14 23:03:00 -0700593 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
David S. Millerdb8dac22008-03-06 16:22:02 -0800594 return; /* No socket for error */
595 }
596
597 err = 0;
598 harderr = 0;
599 inet = inet_sk(sk);
600
601 switch (type) {
602 default:
603 case ICMP_TIME_EXCEEDED:
604 err = EHOSTUNREACH;
605 break;
606 case ICMP_SOURCE_QUENCH:
607 goto out;
608 case ICMP_PARAMETERPROB:
609 err = EPROTO;
610 harderr = 1;
611 break;
612 case ICMP_DEST_UNREACH:
613 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
614 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
615 err = EMSGSIZE;
616 harderr = 1;
617 break;
618 }
619 goto out;
620 }
621 err = EHOSTUNREACH;
622 if (code <= NR_ICMP_UNREACH) {
623 harderr = icmp_err_convert[code].fatal;
624 err = icmp_err_convert[code].errno;
625 }
626 break;
627 }
628
629 /*
630 * RFC1122: OK. Passes ICMP errors back to application, as per
631 * 4.1.3.3.
632 */
633 if (!inet->recverr) {
634 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
635 goto out;
Eric Dumazetb1faf562010-05-31 23:44:05 -0700636 } else
Eric Dumazetc482c562009-07-17 00:26:32 +0000637 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
Eric Dumazetb1faf562010-05-31 23:44:05 -0700638
David S. Millerdb8dac22008-03-06 16:22:02 -0800639 sk->sk_err = err;
640 sk->sk_error_report(sk);
641out:
642 sock_put(sk);
643}
644
645void udp_err(struct sk_buff *skb, u32 info)
646{
Eric Dumazet645ca702008-10-29 01:41:45 -0700647 __udp4_lib_err(skb, info, &udp_table);
David S. Millerdb8dac22008-03-06 16:22:02 -0800648}
649
650/*
651 * Throw away all pending data and cancel the corking. Socket is locked.
652 */
Denis V. Lunev36d926b2008-06-04 15:49:07 +0400653void udp_flush_pending_frames(struct sock *sk)
David S. Millerdb8dac22008-03-06 16:22:02 -0800654{
655 struct udp_sock *up = udp_sk(sk);
656
657 if (up->pending) {
658 up->len = 0;
659 up->pending = 0;
660 ip_flush_pending_frames(sk);
661 }
662}
Denis V. Lunev36d926b2008-06-04 15:49:07 +0400663EXPORT_SYMBOL(udp_flush_pending_frames);
David S. Millerdb8dac22008-03-06 16:22:02 -0800664
665/**
666 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
667 * @sk: socket we are sending on
668 * @skb: sk_buff containing the filled-in UDP header
669 * (checksum field must be zeroed out)
670 */
671static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
Eric Dumazetc482c562009-07-17 00:26:32 +0000672 __be32 src, __be32 dst, int len)
David S. Millerdb8dac22008-03-06 16:22:02 -0800673{
674 unsigned int offset;
675 struct udphdr *uh = udp_hdr(skb);
676 __wsum csum = 0;
677
678 if (skb_queue_len(&sk->sk_write_queue) == 1) {
679 /*
680 * Only one fragment on the socket.
681 */
682 skb->csum_start = skb_transport_header(skb) - skb->head;
683 skb->csum_offset = offsetof(struct udphdr, check);
684 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
685 } else {
686 /*
687 * HW-checksum won't work as there are two or more
688 * fragments on the socket so that all csums of sk_buffs
689 * should be together
690 */
691 offset = skb_transport_offset(skb);
692 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
693
694 skb->ip_summed = CHECKSUM_NONE;
695
696 skb_queue_walk(&sk->sk_write_queue, skb) {
697 csum = csum_add(csum, skb->csum);
698 }
699
700 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
701 if (uh->check == 0)
702 uh->check = CSUM_MANGLED_0;
703 }
704}
705
706/*
707 * Push out all pending data as one UDP datagram. Socket is locked.
708 */
709static int udp_push_pending_frames(struct sock *sk)
710{
711 struct udp_sock *up = udp_sk(sk);
712 struct inet_sock *inet = inet_sk(sk);
713 struct flowi *fl = &inet->cork.fl;
714 struct sk_buff *skb;
715 struct udphdr *uh;
716 int err = 0;
717 int is_udplite = IS_UDPLITE(sk);
718 __wsum csum = 0;
719
720 /* Grab the skbuff where UDP header space exists. */
721 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
722 goto out;
723
724 /*
725 * Create a UDP header
726 */
727 uh = udp_hdr(skb);
728 uh->source = fl->fl_ip_sport;
729 uh->dest = fl->fl_ip_dport;
730 uh->len = htons(up->len);
731 uh->check = 0;
732
733 if (is_udplite) /* UDP-Lite */
734 csum = udplite_csum_outgoing(sk, skb);
735
736 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
737
738 skb->ip_summed = CHECKSUM_NONE;
739 goto send;
740
741 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
742
Eric Dumazetc482c562009-07-17 00:26:32 +0000743 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src, fl->fl4_dst, up->len);
David S. Millerdb8dac22008-03-06 16:22:02 -0800744 goto send;
745
746 } else /* `normal' UDP */
747 csum = udp_csum_outgoing(sk, skb);
748
749 /* add protocol-dependent pseudo-header */
750 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
Eric Dumazetc482c562009-07-17 00:26:32 +0000751 sk->sk_protocol, csum);
David S. Millerdb8dac22008-03-06 16:22:02 -0800752 if (uh->check == 0)
753 uh->check = CSUM_MANGLED_0;
754
755send:
756 err = ip_push_pending_frames(sk);
Eric Dumazet6ce9e7b2009-09-02 18:05:33 -0700757 if (err) {
758 if (err == -ENOBUFS && !inet->recverr) {
759 UDP_INC_STATS_USER(sock_net(sk),
760 UDP_MIB_SNDBUFERRORS, is_udplite);
761 err = 0;
762 }
763 } else
764 UDP_INC_STATS_USER(sock_net(sk),
765 UDP_MIB_OUTDATAGRAMS, is_udplite);
David S. Millerdb8dac22008-03-06 16:22:02 -0800766out:
767 up->len = 0;
768 up->pending = 0;
David S. Millerdb8dac22008-03-06 16:22:02 -0800769 return err;
770}
771
772int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
773 size_t len)
774{
775 struct inet_sock *inet = inet_sk(sk);
776 struct udp_sock *up = udp_sk(sk);
777 int ulen = len;
778 struct ipcm_cookie ipc;
779 struct rtable *rt = NULL;
780 int free = 0;
781 int connected = 0;
782 __be32 daddr, faddr, saddr;
783 __be16 dport;
784 u8 tos;
785 int err, is_udplite = IS_UDPLITE(sk);
786 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
787 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
788
789 if (len > 0xFFFF)
790 return -EMSGSIZE;
791
792 /*
793 * Check the flags.
794 */
795
Eric Dumazetc482c562009-07-17 00:26:32 +0000796 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
David S. Millerdb8dac22008-03-06 16:22:02 -0800797 return -EOPNOTSUPP;
798
799 ipc.opt = NULL;
Patrick Ohly51f31ca2009-02-12 05:03:39 +0000800 ipc.shtx.flags = 0;
David S. Millerdb8dac22008-03-06 16:22:02 -0800801
802 if (up->pending) {
803 /*
804 * There are pending frames.
805 * The socket lock must be held while it's corked.
806 */
807 lock_sock(sk);
808 if (likely(up->pending)) {
809 if (unlikely(up->pending != AF_INET)) {
810 release_sock(sk);
811 return -EINVAL;
812 }
813 goto do_append_data;
814 }
815 release_sock(sk);
816 }
817 ulen += sizeof(struct udphdr);
818
819 /*
820 * Get and verify the address.
821 */
822 if (msg->msg_name) {
Eric Dumazetc482c562009-07-17 00:26:32 +0000823 struct sockaddr_in * usin = (struct sockaddr_in *)msg->msg_name;
David S. Millerdb8dac22008-03-06 16:22:02 -0800824 if (msg->msg_namelen < sizeof(*usin))
825 return -EINVAL;
826 if (usin->sin_family != AF_INET) {
827 if (usin->sin_family != AF_UNSPEC)
828 return -EAFNOSUPPORT;
829 }
830
831 daddr = usin->sin_addr.s_addr;
832 dport = usin->sin_port;
833 if (dport == 0)
834 return -EINVAL;
835 } else {
836 if (sk->sk_state != TCP_ESTABLISHED)
837 return -EDESTADDRREQ;
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000838 daddr = inet->inet_daddr;
839 dport = inet->inet_dport;
David S. Millerdb8dac22008-03-06 16:22:02 -0800840 /* Open fast path for connected socket.
841 Route will not be used, if at least one option is set.
842 */
843 connected = 1;
844 }
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000845 ipc.addr = inet->inet_saddr;
David S. Millerdb8dac22008-03-06 16:22:02 -0800846
847 ipc.oif = sk->sk_bound_dev_if;
Patrick Ohly51f31ca2009-02-12 05:03:39 +0000848 err = sock_tx_timestamp(msg, sk, &ipc.shtx);
849 if (err)
850 return err;
David S. Millerdb8dac22008-03-06 16:22:02 -0800851 if (msg->msg_controllen) {
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900852 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
David S. Millerdb8dac22008-03-06 16:22:02 -0800853 if (err)
854 return err;
855 if (ipc.opt)
856 free = 1;
857 connected = 0;
858 }
859 if (!ipc.opt)
860 ipc.opt = inet->opt;
861
862 saddr = ipc.addr;
863 ipc.addr = faddr = daddr;
864
865 if (ipc.opt && ipc.opt->srr) {
866 if (!daddr)
867 return -EINVAL;
868 faddr = ipc.opt->faddr;
869 connected = 0;
870 }
871 tos = RT_TOS(inet->tos);
872 if (sock_flag(sk, SOCK_LOCALROUTE) ||
873 (msg->msg_flags & MSG_DONTROUTE) ||
874 (ipc.opt && ipc.opt->is_strictroute)) {
875 tos |= RTO_ONLINK;
876 connected = 0;
877 }
878
879 if (ipv4_is_multicast(daddr)) {
880 if (!ipc.oif)
881 ipc.oif = inet->mc_index;
882 if (!saddr)
883 saddr = inet->mc_addr;
884 connected = 0;
885 }
886
887 if (connected)
Eric Dumazetc482c562009-07-17 00:26:32 +0000888 rt = (struct rtable *)sk_dst_check(sk, 0);
David S. Millerdb8dac22008-03-06 16:22:02 -0800889
890 if (rt == NULL) {
891 struct flowi fl = { .oif = ipc.oif,
Atis Elsts914a9ab2009-10-01 15:16:49 -0700892 .mark = sk->sk_mark,
David S. Millerdb8dac22008-03-06 16:22:02 -0800893 .nl_u = { .ip4_u =
894 { .daddr = faddr,
895 .saddr = saddr,
896 .tos = tos } },
897 .proto = sk->sk_protocol,
Balazs Scheidlera134f852008-11-20 01:07:24 -0800898 .flags = inet_sk_flowi_flags(sk),
David S. Millerdb8dac22008-03-06 16:22:02 -0800899 .uli_u = { .ports =
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000900 { .sport = inet->inet_sport,
David S. Millerdb8dac22008-03-06 16:22:02 -0800901 .dport = dport } } };
Pavel Emelyanov84a3aa02008-07-16 20:19:08 -0700902 struct net *net = sock_net(sk);
903
David S. Millerdb8dac22008-03-06 16:22:02 -0800904 security_sk_classify_flow(sk, &fl);
Pavel Emelyanov84a3aa02008-07-16 20:19:08 -0700905 err = ip_route_output_flow(net, &rt, &fl, sk, 1);
David S. Millerdb8dac22008-03-06 16:22:02 -0800906 if (err) {
907 if (err == -ENETUNREACH)
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700908 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
David S. Millerdb8dac22008-03-06 16:22:02 -0800909 goto out;
910 }
911
912 err = -EACCES;
913 if ((rt->rt_flags & RTCF_BROADCAST) &&
914 !sock_flag(sk, SOCK_BROADCAST))
915 goto out;
916 if (connected)
Changli Gaod8d1f302010-06-10 23:31:35 -0700917 sk_dst_set(sk, dst_clone(&rt->dst));
David S. Millerdb8dac22008-03-06 16:22:02 -0800918 }
919
920 if (msg->msg_flags&MSG_CONFIRM)
921 goto do_confirm;
922back_from_confirm:
923
924 saddr = rt->rt_src;
925 if (!ipc.addr)
926 daddr = ipc.addr = rt->rt_dst;
927
928 lock_sock(sk);
929 if (unlikely(up->pending)) {
930 /* The socket is already corked while preparing it. */
931 /* ... which is an evident application bug. --ANK */
932 release_sock(sk);
933
934 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
935 err = -EINVAL;
936 goto out;
937 }
938 /*
939 * Now cork the socket to pend data.
940 */
941 inet->cork.fl.fl4_dst = daddr;
942 inet->cork.fl.fl_ip_dport = dport;
943 inet->cork.fl.fl4_src = saddr;
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000944 inet->cork.fl.fl_ip_sport = inet->inet_sport;
David S. Millerdb8dac22008-03-06 16:22:02 -0800945 up->pending = AF_INET;
946
947do_append_data:
948 up->len += ulen;
949 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
950 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
Eric Dumazet2e77d892008-11-24 15:52:46 -0800951 sizeof(struct udphdr), &ipc, &rt,
David S. Millerdb8dac22008-03-06 16:22:02 -0800952 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
953 if (err)
954 udp_flush_pending_frames(sk);
955 else if (!corkreq)
956 err = udp_push_pending_frames(sk);
957 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
958 up->pending = 0;
959 release_sock(sk);
960
961out:
962 ip_rt_put(rt);
963 if (free)
964 kfree(ipc.opt);
965 if (!err)
966 return len;
967 /*
968 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
969 * ENOBUFS might not be good (it's not tunable per se), but otherwise
970 * we don't have a good statistic (IpOutDiscards but it can be too many
971 * things). We could add another new stat but at least for now that
972 * seems like overkill.
973 */
974 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
Pavel Emelyanov629ca232008-07-05 21:18:07 -0700975 UDP_INC_STATS_USER(sock_net(sk),
976 UDP_MIB_SNDBUFERRORS, is_udplite);
David S. Millerdb8dac22008-03-06 16:22:02 -0800977 }
978 return err;
979
980do_confirm:
Changli Gaod8d1f302010-06-10 23:31:35 -0700981 dst_confirm(&rt->dst);
David S. Millerdb8dac22008-03-06 16:22:02 -0800982 if (!(msg->msg_flags&MSG_PROBE) || len)
983 goto back_from_confirm;
984 err = 0;
985 goto out;
986}
Eric Dumazetc482c562009-07-17 00:26:32 +0000987EXPORT_SYMBOL(udp_sendmsg);
David S. Millerdb8dac22008-03-06 16:22:02 -0800988
989int udp_sendpage(struct sock *sk, struct page *page, int offset,
990 size_t size, int flags)
991{
992 struct udp_sock *up = udp_sk(sk);
993 int ret;
994
995 if (!up->pending) {
996 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
997
998 /* Call udp_sendmsg to specify destination address which
999 * sendpage interface can't pass.
1000 * This will succeed only when the socket is connected.
1001 */
1002 ret = udp_sendmsg(NULL, sk, &msg, 0);
1003 if (ret < 0)
1004 return ret;
1005 }
1006
1007 lock_sock(sk);
1008
1009 if (unlikely(!up->pending)) {
1010 release_sock(sk);
1011
1012 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
1013 return -EINVAL;
1014 }
1015
1016 ret = ip_append_page(sk, page, offset, size, flags);
1017 if (ret == -EOPNOTSUPP) {
1018 release_sock(sk);
1019 return sock_no_sendpage(sk->sk_socket, page, offset,
1020 size, flags);
1021 }
1022 if (ret < 0) {
1023 udp_flush_pending_frames(sk);
1024 goto out;
1025 }
1026
1027 up->len += size;
1028 if (!(up->corkflag || (flags&MSG_MORE)))
1029 ret = udp_push_pending_frames(sk);
1030 if (!ret)
1031 ret = size;
1032out:
1033 release_sock(sk);
1034 return ret;
1035}
1036
Eric Dumazet85584672009-10-09 04:43:40 +00001037
1038/**
1039 * first_packet_length - return length of first packet in receive queue
1040 * @sk: socket
1041 *
1042 * Drops all bad checksum frames, until a valid one is found.
1043 * Returns the length of found skb, or 0 if none is found.
1044 */
1045static unsigned int first_packet_length(struct sock *sk)
1046{
1047 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1048 struct sk_buff *skb;
1049 unsigned int res;
1050
1051 __skb_queue_head_init(&list_kill);
1052
1053 spin_lock_bh(&rcvq->lock);
1054 while ((skb = skb_peek(rcvq)) != NULL &&
1055 udp_lib_checksum_complete(skb)) {
1056 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1057 IS_UDPLITE(sk));
Eric Dumazet8edf19c2009-10-15 00:12:40 +00001058 atomic_inc(&sk->sk_drops);
Eric Dumazet85584672009-10-09 04:43:40 +00001059 __skb_unlink(skb, rcvq);
1060 __skb_queue_tail(&list_kill, skb);
1061 }
1062 res = skb ? skb->len : 0;
1063 spin_unlock_bh(&rcvq->lock);
1064
1065 if (!skb_queue_empty(&list_kill)) {
Eric Dumazet8a74ad62010-05-26 19:20:18 +00001066 bool slow = lock_sock_fast(sk);
1067
Eric Dumazet85584672009-10-09 04:43:40 +00001068 __skb_queue_purge(&list_kill);
1069 sk_mem_reclaim_partial(sk);
Eric Dumazet8a74ad62010-05-26 19:20:18 +00001070 unlock_sock_fast(sk, slow);
Eric Dumazet85584672009-10-09 04:43:40 +00001071 }
1072 return res;
1073}
1074
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075/*
1076 * IOCTL requests applicable to the UDP protocol
1077 */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001078
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1080{
Stephen Hemminger6516c652007-03-08 20:41:55 -08001081 switch (cmd) {
1082 case SIOCOUTQ:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 {
Eric Dumazet31e6d362009-06-17 19:05:41 -07001084 int amount = sk_wmem_alloc_get(sk);
1085
Stephen Hemminger6516c652007-03-08 20:41:55 -08001086 return put_user(amount, (int __user *)arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 }
Stephen Hemminger6516c652007-03-08 20:41:55 -08001088
1089 case SIOCINQ:
1090 {
Eric Dumazet85584672009-10-09 04:43:40 +00001091 unsigned int amount = first_packet_length(sk);
Stephen Hemminger6516c652007-03-08 20:41:55 -08001092
Eric Dumazet85584672009-10-09 04:43:40 +00001093 if (amount)
Stephen Hemminger6516c652007-03-08 20:41:55 -08001094 /*
1095 * We will only return the amount
1096 * of this packet since that is all
1097 * that will be read.
1098 */
Eric Dumazet85584672009-10-09 04:43:40 +00001099 amount -= sizeof(struct udphdr);
1100
Stephen Hemminger6516c652007-03-08 20:41:55 -08001101 return put_user(amount, (int __user *)arg);
1102 }
1103
1104 default:
1105 return -ENOIOCTLCMD;
1106 }
1107
1108 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109}
Eric Dumazetc482c562009-07-17 00:26:32 +00001110EXPORT_SYMBOL(udp_ioctl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111
David S. Millerdb8dac22008-03-06 16:22:02 -08001112/*
1113 * This should be easy, if there is something there we
1114 * return it, otherwise we block.
1115 */
1116
1117int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1118 size_t len, int noblock, int flags, int *addr_len)
1119{
1120 struct inet_sock *inet = inet_sk(sk);
1121 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
1122 struct sk_buff *skb;
Gerrit Renker81d54ec2010-02-10 20:26:19 +00001123 unsigned int ulen;
David S. Millerdb8dac22008-03-06 16:22:02 -08001124 int peeked;
1125 int err;
1126 int is_udplite = IS_UDPLITE(sk);
Eric Dumazet8a74ad62010-05-26 19:20:18 +00001127 bool slow;
David S. Millerdb8dac22008-03-06 16:22:02 -08001128
1129 /*
1130 * Check any passed addresses
1131 */
1132 if (addr_len)
Eric Dumazetc482c562009-07-17 00:26:32 +00001133 *addr_len = sizeof(*sin);
David S. Millerdb8dac22008-03-06 16:22:02 -08001134
1135 if (flags & MSG_ERRQUEUE)
1136 return ip_recv_error(sk, msg, len);
1137
1138try_again:
1139 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1140 &peeked, &err);
1141 if (!skb)
1142 goto out;
1143
1144 ulen = skb->len - sizeof(struct udphdr);
Gerrit Renker81d54ec2010-02-10 20:26:19 +00001145 if (len > ulen)
1146 len = ulen;
1147 else if (len < ulen)
David S. Millerdb8dac22008-03-06 16:22:02 -08001148 msg->msg_flags |= MSG_TRUNC;
1149
1150 /*
1151 * If checksum is needed at all, try to do it while copying the
1152 * data. If the data is truncated, or if we only want a partial
1153 * coverage checksum (UDP-Lite), do it before the copy.
1154 */
1155
Gerrit Renker81d54ec2010-02-10 20:26:19 +00001156 if (len < ulen || UDP_SKB_CB(skb)->partial_cov) {
David S. Millerdb8dac22008-03-06 16:22:02 -08001157 if (udp_lib_checksum_complete(skb))
1158 goto csum_copy_err;
1159 }
1160
1161 if (skb_csum_unnecessary(skb))
1162 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
Gerrit Renker81d54ec2010-02-10 20:26:19 +00001163 msg->msg_iov, len);
David S. Millerdb8dac22008-03-06 16:22:02 -08001164 else {
Eric Dumazetc482c562009-07-17 00:26:32 +00001165 err = skb_copy_and_csum_datagram_iovec(skb,
1166 sizeof(struct udphdr),
1167 msg->msg_iov);
David S. Millerdb8dac22008-03-06 16:22:02 -08001168
1169 if (err == -EINVAL)
1170 goto csum_copy_err;
1171 }
1172
1173 if (err)
1174 goto out_free;
1175
1176 if (!peeked)
Pavel Emelyanov629ca232008-07-05 21:18:07 -07001177 UDP_INC_STATS_USER(sock_net(sk),
1178 UDP_MIB_INDATAGRAMS, is_udplite);
David S. Millerdb8dac22008-03-06 16:22:02 -08001179
Neil Horman3b885782009-10-12 13:26:31 -07001180 sock_recv_ts_and_drops(msg, sk, skb);
David S. Millerdb8dac22008-03-06 16:22:02 -08001181
1182 /* Copy the address. */
Eric Dumazetc482c562009-07-17 00:26:32 +00001183 if (sin) {
David S. Millerdb8dac22008-03-06 16:22:02 -08001184 sin->sin_family = AF_INET;
1185 sin->sin_port = udp_hdr(skb)->source;
1186 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1187 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1188 }
1189 if (inet->cmsg_flags)
1190 ip_cmsg_recv(msg, skb);
1191
Gerrit Renker81d54ec2010-02-10 20:26:19 +00001192 err = len;
David S. Millerdb8dac22008-03-06 16:22:02 -08001193 if (flags & MSG_TRUNC)
1194 err = ulen;
1195
1196out_free:
Eric Dumazet9d410c72009-10-30 05:03:53 +00001197 skb_free_datagram_locked(sk, skb);
David S. Millerdb8dac22008-03-06 16:22:02 -08001198out:
1199 return err;
1200
1201csum_copy_err:
Eric Dumazet8a74ad62010-05-26 19:20:18 +00001202 slow = lock_sock_fast(sk);
David S. Millerdb8dac22008-03-06 16:22:02 -08001203 if (!skb_kill_datagram(sk, skb, flags))
Pavel Emelyanov629ca232008-07-05 21:18:07 -07001204 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
Eric Dumazet8a74ad62010-05-26 19:20:18 +00001205 unlock_sock_fast(sk, slow);
David S. Millerdb8dac22008-03-06 16:22:02 -08001206
1207 if (noblock)
1208 return -EAGAIN;
1209 goto try_again;
1210}
1211
1212
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213int udp_disconnect(struct sock *sk, int flags)
1214{
1215 struct inet_sock *inet = inet_sk(sk);
1216 /*
1217 * 1003.1g - break association.
1218 */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001219
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 sk->sk_state = TCP_CLOSE;
Eric Dumazetc720c7e2009-10-15 06:30:45 +00001221 inet->inet_daddr = 0;
1222 inet->inet_dport = 0;
David S. Millerc58dc012010-04-27 15:05:31 -07001223 sock_rps_save_rxhash(sk, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 sk->sk_bound_dev_if = 0;
1225 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1226 inet_reset_saddr(sk);
1227
1228 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1229 sk->sk_prot->unhash(sk);
Eric Dumazetc720c7e2009-10-15 06:30:45 +00001230 inet->inet_sport = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 }
1232 sk_dst_reset(sk);
1233 return 0;
1234}
Eric Dumazetc482c562009-07-17 00:26:32 +00001235EXPORT_SYMBOL(udp_disconnect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236
Eric Dumazet645ca702008-10-29 01:41:45 -07001237void udp_lib_unhash(struct sock *sk)
1238{
Eric Dumazet723b4612008-11-25 13:55:15 -08001239 if (sk_hashed(sk)) {
1240 struct udp_table *udptable = sk->sk_prot->h.udp_table;
Eric Dumazet512615b2009-11-08 10:17:58 +00001241 struct udp_hslot *hslot, *hslot2;
1242
1243 hslot = udp_hashslot(udptable, sock_net(sk),
1244 udp_sk(sk)->udp_port_hash);
1245 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
Eric Dumazet645ca702008-10-29 01:41:45 -07001246
Eric Dumazet723b4612008-11-25 13:55:15 -08001247 spin_lock_bh(&hslot->lock);
1248 if (sk_nulls_del_node_init_rcu(sk)) {
Eric Dumazetfdcc8aa92009-11-08 10:17:05 +00001249 hslot->count--;
Eric Dumazetc720c7e2009-10-15 06:30:45 +00001250 inet_sk(sk)->inet_num = 0;
Eric Dumazet723b4612008-11-25 13:55:15 -08001251 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
Eric Dumazet512615b2009-11-08 10:17:58 +00001252
1253 spin_lock(&hslot2->lock);
1254 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1255 hslot2->count--;
1256 spin_unlock(&hslot2->lock);
Eric Dumazet723b4612008-11-25 13:55:15 -08001257 }
1258 spin_unlock_bh(&hslot->lock);
Eric Dumazet645ca702008-10-29 01:41:45 -07001259 }
Eric Dumazet645ca702008-10-29 01:41:45 -07001260}
1261EXPORT_SYMBOL(udp_lib_unhash);
1262
Herbert Xu93821772008-09-15 11:48:46 -07001263static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1264{
Tom Herbertfec5e652010-04-16 16:01:27 -07001265 int rc;
Herbert Xu93821772008-09-15 11:48:46 -07001266
Tom Herbertfec5e652010-04-16 16:01:27 -07001267 if (inet_sk(sk)->inet_daddr)
David S. Millerc58dc012010-04-27 15:05:31 -07001268 sock_rps_save_rxhash(sk, skb->rxhash);
Tom Herbertfec5e652010-04-16 16:01:27 -07001269
Eric Dumazetf84af322010-04-28 15:31:51 -07001270 rc = ip_queue_rcv_skb(sk, skb);
Eric Dumazet766e90372009-10-14 20:40:11 -07001271 if (rc < 0) {
1272 int is_udplite = IS_UDPLITE(sk);
1273
Herbert Xu93821772008-09-15 11:48:46 -07001274 /* Note that an ENOMEM error is charged twice */
Eric Dumazet766e90372009-10-14 20:40:11 -07001275 if (rc == -ENOMEM)
Herbert Xu93821772008-09-15 11:48:46 -07001276 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1277 is_udplite);
Eric Dumazet766e90372009-10-14 20:40:11 -07001278 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1279 kfree_skb(skb);
1280 return -1;
Herbert Xu93821772008-09-15 11:48:46 -07001281 }
1282
1283 return 0;
1284
Herbert Xu93821772008-09-15 11:48:46 -07001285}
1286
David S. Millerdb8dac22008-03-06 16:22:02 -08001287/* returns:
1288 * -1: error
1289 * 0: success
1290 * >0: "udp encap" protocol resubmission
1291 *
1292 * Note that in the success and error cases, the skb is assumed to
1293 * have either been requeued or freed.
1294 */
Eric Dumazetc482c562009-07-17 00:26:32 +00001295int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
David S. Millerdb8dac22008-03-06 16:22:02 -08001296{
1297 struct udp_sock *up = udp_sk(sk);
1298 int rc;
1299 int is_udplite = IS_UDPLITE(sk);
1300
1301 /*
1302 * Charge it to the socket, dropping if the queue is full.
1303 */
1304 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1305 goto drop;
1306 nf_reset(skb);
1307
1308 if (up->encap_type) {
1309 /*
1310 * This is an encapsulation socket so pass the skb to
1311 * the socket's udp_encap_rcv() hook. Otherwise, just
1312 * fall through and pass this up the UDP socket.
1313 * up->encap_rcv() returns the following value:
1314 * =0 if skb was successfully passed to the encap
1315 * handler or was discarded by it.
1316 * >0 if skb should be passed on to UDP.
1317 * <0 if skb should be resubmitted as proto -N
1318 */
1319
1320 /* if we're overly short, let UDP handle it */
1321 if (skb->len > sizeof(struct udphdr) &&
1322 up->encap_rcv != NULL) {
1323 int ret;
1324
1325 ret = (*up->encap_rcv)(sk, skb);
1326 if (ret <= 0) {
Pavel Emelyanov02833282008-07-05 21:18:48 -07001327 UDP_INC_STATS_BH(sock_net(sk),
1328 UDP_MIB_INDATAGRAMS,
David S. Millerdb8dac22008-03-06 16:22:02 -08001329 is_udplite);
1330 return -ret;
1331 }
1332 }
1333
1334 /* FALLTHROUGH -- it's a UDP Packet */
1335 }
1336
1337 /*
1338 * UDP-Lite specific tests, ignored on UDP sockets
1339 */
1340 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1341
1342 /*
1343 * MIB statistics other than incrementing the error count are
1344 * disabled for the following two types of errors: these depend
1345 * on the application settings, not on the functioning of the
1346 * protocol stack as such.
1347 *
1348 * RFC 3828 here recommends (sec 3.3): "There should also be a
1349 * way ... to ... at least let the receiving application block
1350 * delivery of packets with coverage values less than a value
1351 * provided by the application."
1352 */
1353 if (up->pcrlen == 0) { /* full coverage was set */
1354 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1355 "%d while full coverage %d requested\n",
1356 UDP_SKB_CB(skb)->cscov, skb->len);
1357 goto drop;
1358 }
1359 /* The next case involves violating the min. coverage requested
1360 * by the receiver. This is subtle: if receiver wants x and x is
1361 * greater than the buffersize/MTU then receiver will complain
1362 * that it wants x while sender emits packets of smaller size y.
1363 * Therefore the above ...()->partial_cov statement is essential.
1364 */
1365 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1366 LIMIT_NETDEBUG(KERN_WARNING
1367 "UDPLITE: coverage %d too small, need min %d\n",
1368 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1369 goto drop;
1370 }
1371 }
1372
1373 if (sk->sk_filter) {
1374 if (udp_lib_checksum_complete(skb))
1375 goto drop;
1376 }
1377
Eric Dumazetc3774112010-04-27 15:13:20 -07001378
1379 if (sk_rcvqueues_full(sk, skb))
1380 goto drop;
1381
Herbert Xu93821772008-09-15 11:48:46 -07001382 rc = 0;
David S. Millerdb8dac22008-03-06 16:22:02 -08001383
Herbert Xu93821772008-09-15 11:48:46 -07001384 bh_lock_sock(sk);
1385 if (!sock_owned_by_user(sk))
1386 rc = __udp_queue_rcv_skb(sk, skb);
Zhu Yia3a858f2010-03-04 18:01:47 +00001387 else if (sk_add_backlog(sk, skb)) {
Zhu Yi55349792010-03-04 18:01:42 +00001388 bh_unlock_sock(sk);
1389 goto drop;
1390 }
Herbert Xu93821772008-09-15 11:48:46 -07001391 bh_unlock_sock(sk);
1392
1393 return rc;
David S. Millerdb8dac22008-03-06 16:22:02 -08001394
1395drop:
Pavel Emelyanov02833282008-07-05 21:18:48 -07001396 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
Eric Dumazet8edf19c2009-10-15 00:12:40 +00001397 atomic_inc(&sk->sk_drops);
David S. Millerdb8dac22008-03-06 16:22:02 -08001398 kfree_skb(skb);
1399 return -1;
1400}
1401
Eric Dumazet1240d132009-11-08 10:18:44 +00001402
1403static void flush_stack(struct sock **stack, unsigned int count,
1404 struct sk_buff *skb, unsigned int final)
1405{
1406 unsigned int i;
1407 struct sk_buff *skb1 = NULL;
Eric Dumazetf6b8f322009-11-08 10:20:19 +00001408 struct sock *sk;
Eric Dumazet1240d132009-11-08 10:18:44 +00001409
1410 for (i = 0; i < count; i++) {
Eric Dumazetf6b8f322009-11-08 10:20:19 +00001411 sk = stack[i];
Eric Dumazet1240d132009-11-08 10:18:44 +00001412 if (likely(skb1 == NULL))
1413 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1414
Eric Dumazetf6b8f322009-11-08 10:20:19 +00001415 if (!skb1) {
1416 atomic_inc(&sk->sk_drops);
1417 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1418 IS_UDPLITE(sk));
1419 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1420 IS_UDPLITE(sk));
1421 }
1422
1423 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
Eric Dumazet1240d132009-11-08 10:18:44 +00001424 skb1 = NULL;
1425 }
1426 if (unlikely(skb1))
1427 kfree_skb(skb1);
1428}
1429
David S. Millerdb8dac22008-03-06 16:22:02 -08001430/*
1431 * Multicasts and broadcasts go to each listener.
1432 *
Eric Dumazet1240d132009-11-08 10:18:44 +00001433 * Note: called only from the BH handler context.
David S. Millerdb8dac22008-03-06 16:22:02 -08001434 */
Pavel Emelyanove3163492008-06-16 17:12:11 -07001435static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
David S. Millerdb8dac22008-03-06 16:22:02 -08001436 struct udphdr *uh,
1437 __be32 saddr, __be32 daddr,
Eric Dumazet645ca702008-10-29 01:41:45 -07001438 struct udp_table *udptable)
David S. Millerdb8dac22008-03-06 16:22:02 -08001439{
Eric Dumazet1240d132009-11-08 10:18:44 +00001440 struct sock *sk, *stack[256 / sizeof(struct sock *)];
Eric Dumazetf86dcc52009-10-07 00:37:59 +00001441 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
David S. Millerdb8dac22008-03-06 16:22:02 -08001442 int dif;
Eric Dumazet1240d132009-11-08 10:18:44 +00001443 unsigned int i, count = 0;
David S. Millerdb8dac22008-03-06 16:22:02 -08001444
Eric Dumazet645ca702008-10-29 01:41:45 -07001445 spin_lock(&hslot->lock);
Eric Dumazet88ab1932008-11-16 19:39:21 -08001446 sk = sk_nulls_head(&hslot->head);
David S. Millerdb8dac22008-03-06 16:22:02 -08001447 dif = skb->dev->ifindex;
Eric Dumazet920a4612008-11-01 21:22:23 -07001448 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
Eric Dumazet1240d132009-11-08 10:18:44 +00001449 while (sk) {
1450 stack[count++] = sk;
1451 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1452 daddr, uh->source, saddr, dif);
1453 if (unlikely(count == ARRAY_SIZE(stack))) {
1454 if (!sk)
1455 break;
1456 flush_stack(stack, count, skb, ~0);
1457 count = 0;
1458 }
1459 }
1460 /*
1461 * before releasing chain lock, we must take a reference on sockets
1462 */
1463 for (i = 0; i < count; i++)
1464 sock_hold(stack[i]);
David S. Millerdb8dac22008-03-06 16:22:02 -08001465
Eric Dumazet645ca702008-10-29 01:41:45 -07001466 spin_unlock(&hslot->lock);
Eric Dumazet1240d132009-11-08 10:18:44 +00001467
1468 /*
1469 * do the slow work with no lock held
1470 */
1471 if (count) {
1472 flush_stack(stack, count, skb, count - 1);
1473
1474 for (i = 0; i < count; i++)
1475 sock_put(stack[i]);
1476 } else {
1477 kfree_skb(skb);
1478 }
David S. Millerdb8dac22008-03-06 16:22:02 -08001479 return 0;
1480}
1481
1482/* Initialize UDP checksum. If exited with zero value (success),
1483 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1484 * Otherwise, csum completion requires chacksumming packet body,
1485 * including udp header and folding it to skb->csum.
1486 */
1487static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1488 int proto)
1489{
1490 const struct iphdr *iph;
1491 int err;
1492
1493 UDP_SKB_CB(skb)->partial_cov = 0;
1494 UDP_SKB_CB(skb)->cscov = skb->len;
1495
1496 if (proto == IPPROTO_UDPLITE) {
1497 err = udplite_checksum_init(skb, uh);
1498 if (err)
1499 return err;
1500 }
1501
1502 iph = ip_hdr(skb);
1503 if (uh->check == 0) {
1504 skb->ip_summed = CHECKSUM_UNNECESSARY;
1505 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
Eric Dumazetc482c562009-07-17 00:26:32 +00001506 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
David S. Millerdb8dac22008-03-06 16:22:02 -08001507 proto, skb->csum))
1508 skb->ip_summed = CHECKSUM_UNNECESSARY;
1509 }
1510 if (!skb_csum_unnecessary(skb))
1511 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1512 skb->len, proto, 0);
1513 /* Probably, we should checksum udp header (it should be in cache
1514 * in any case) and data in tiny packets (< rx copybreak).
1515 */
1516
1517 return 0;
1518}
1519
1520/*
1521 * All we need to do is get the socket, and then do a checksum.
1522 */
1523
Eric Dumazet645ca702008-10-29 01:41:45 -07001524int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
David S. Millerdb8dac22008-03-06 16:22:02 -08001525 int proto)
1526{
1527 struct sock *sk;
Jesper Dangaard Brouer7b5e56f2009-02-05 15:05:45 -08001528 struct udphdr *uh;
David S. Millerdb8dac22008-03-06 16:22:02 -08001529 unsigned short ulen;
Eric Dumazetadf30902009-06-02 05:19:30 +00001530 struct rtable *rt = skb_rtable(skb);
Jesper Dangaard Brouer2783ef22009-02-06 01:59:12 -08001531 __be32 saddr, daddr;
Pavel Emelyanov02833282008-07-05 21:18:48 -07001532 struct net *net = dev_net(skb->dev);
David S. Millerdb8dac22008-03-06 16:22:02 -08001533
1534 /*
1535 * Validate the packet.
1536 */
1537 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1538 goto drop; /* No space for header. */
1539
Jesper Dangaard Brouer7b5e56f2009-02-05 15:05:45 -08001540 uh = udp_hdr(skb);
David S. Millerdb8dac22008-03-06 16:22:02 -08001541 ulen = ntohs(uh->len);
Bjørn Morkccc2d972010-05-06 03:44:34 +00001542 saddr = ip_hdr(skb)->saddr;
1543 daddr = ip_hdr(skb)->daddr;
1544
David S. Millerdb8dac22008-03-06 16:22:02 -08001545 if (ulen > skb->len)
1546 goto short_packet;
1547
1548 if (proto == IPPROTO_UDP) {
1549 /* UDP validates ulen. */
1550 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1551 goto short_packet;
1552 uh = udp_hdr(skb);
1553 }
1554
1555 if (udp4_csum_init(skb, uh, proto))
1556 goto csum_error;
1557
1558 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
Pavel Emelyanove3163492008-06-16 17:12:11 -07001559 return __udp4_lib_mcast_deliver(net, skb, uh,
1560 saddr, daddr, udptable);
David S. Millerdb8dac22008-03-06 16:22:02 -08001561
KOVACS Krisztian607c4aa2008-10-07 12:38:32 -07001562 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
David S. Millerdb8dac22008-03-06 16:22:02 -08001563
1564 if (sk != NULL) {
Herbert Xu93821772008-09-15 11:48:46 -07001565 int ret = udp_queue_rcv_skb(sk, skb);
David S. Millerdb8dac22008-03-06 16:22:02 -08001566 sock_put(sk);
1567
1568 /* a return value > 0 means to resubmit the input, but
1569 * it wants the return to be -protocol, or 0
1570 */
1571 if (ret > 0)
1572 return -ret;
1573 return 0;
1574 }
1575
1576 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1577 goto drop;
1578 nf_reset(skb);
1579
1580 /* No socket. Drop packet silently, if checksum is wrong */
1581 if (udp_lib_checksum_complete(skb))
1582 goto csum_error;
1583
Pavel Emelyanov02833282008-07-05 21:18:48 -07001584 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
David S. Millerdb8dac22008-03-06 16:22:02 -08001585 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1586
1587 /*
1588 * Hmm. We got an UDP packet to a port to which we
1589 * don't wanna listen. Ignore it.
1590 */
1591 kfree_skb(skb);
1592 return 0;
1593
1594short_packet:
Harvey Harrison673d57e2008-10-31 00:53:57 -07001595 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 -08001596 proto == IPPROTO_UDPLITE ? "-Lite" : "",
Harvey Harrison673d57e2008-10-31 00:53:57 -07001597 &saddr,
David S. Millerdb8dac22008-03-06 16:22:02 -08001598 ntohs(uh->source),
1599 ulen,
1600 skb->len,
Harvey Harrison673d57e2008-10-31 00:53:57 -07001601 &daddr,
David S. Millerdb8dac22008-03-06 16:22:02 -08001602 ntohs(uh->dest));
1603 goto drop;
1604
1605csum_error:
1606 /*
1607 * RFC1122: OK. Discards the bad packet silently (as far as
1608 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1609 */
Harvey Harrison673d57e2008-10-31 00:53:57 -07001610 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 -08001611 proto == IPPROTO_UDPLITE ? "-Lite" : "",
Harvey Harrison673d57e2008-10-31 00:53:57 -07001612 &saddr,
David S. Millerdb8dac22008-03-06 16:22:02 -08001613 ntohs(uh->source),
Harvey Harrison673d57e2008-10-31 00:53:57 -07001614 &daddr,
David S. Millerdb8dac22008-03-06 16:22:02 -08001615 ntohs(uh->dest),
1616 ulen);
1617drop:
Pavel Emelyanov02833282008-07-05 21:18:48 -07001618 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
David S. Millerdb8dac22008-03-06 16:22:02 -08001619 kfree_skb(skb);
1620 return 0;
1621}
1622
1623int udp_rcv(struct sk_buff *skb)
1624{
Eric Dumazet645ca702008-10-29 01:41:45 -07001625 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
David S. Millerdb8dac22008-03-06 16:22:02 -08001626}
1627
Brian Haley7d06b2e2008-06-14 17:04:49 -07001628void udp_destroy_sock(struct sock *sk)
David S. Millerdb8dac22008-03-06 16:22:02 -08001629{
Eric Dumazet8a74ad62010-05-26 19:20:18 +00001630 bool slow = lock_sock_fast(sk);
David S. Millerdb8dac22008-03-06 16:22:02 -08001631 udp_flush_pending_frames(sk);
Eric Dumazet8a74ad62010-05-26 19:20:18 +00001632 unlock_sock_fast(sk, slow);
David S. Millerdb8dac22008-03-06 16:22:02 -08001633}
1634
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635/*
1636 * Socket option code for UDP
1637 */
Gerrit Renker4c0a6cb2006-11-27 09:29:59 -08001638int udp_lib_setsockopt(struct sock *sk, int level, int optname,
David S. Millerb7058842009-09-30 16:12:20 -07001639 char __user *optval, unsigned int optlen,
Gerrit Renker4c0a6cb2006-11-27 09:29:59 -08001640 int (*push_pending_frames)(struct sock *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641{
1642 struct udp_sock *up = udp_sk(sk);
1643 int val;
1644 int err = 0;
Wang Chenb2bf1e22007-12-03 22:34:16 +11001645 int is_udplite = IS_UDPLITE(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646
Eric Dumazetc482c562009-07-17 00:26:32 +00001647 if (optlen < sizeof(int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 return -EINVAL;
1649
1650 if (get_user(val, (int __user *)optval))
1651 return -EFAULT;
1652
Stephen Hemminger6516c652007-03-08 20:41:55 -08001653 switch (optname) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 case UDP_CORK:
1655 if (val != 0) {
1656 up->corkflag = 1;
1657 } else {
1658 up->corkflag = 0;
1659 lock_sock(sk);
Gerrit Renker4c0a6cb2006-11-27 09:29:59 -08001660 (*push_pending_frames)(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 release_sock(sk);
1662 }
1663 break;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001664
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 case UDP_ENCAP:
1666 switch (val) {
1667 case 0:
1668 case UDP_ENCAP_ESPINUDP:
1669 case UDP_ENCAP_ESPINUDP_NON_IKE:
James Chapman067b2072007-07-05 17:08:05 -07001670 up->encap_rcv = xfrm4_udp_encap_rcv;
1671 /* FALLTHROUGH */
James Chapman342f0232007-06-27 15:37:46 -07001672 case UDP_ENCAP_L2TPINUDP:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 up->encap_type = val;
1674 break;
1675 default:
1676 err = -ENOPROTOOPT;
1677 break;
1678 }
1679 break;
1680
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08001681 /*
1682 * UDP-Lite's partial checksum coverage (RFC 3828).
1683 */
1684 /* The sender sets actual checksum coverage length via this option.
1685 * The case coverage > packet length is handled by send module. */
1686 case UDPLITE_SEND_CSCOV:
Wang Chenb2bf1e22007-12-03 22:34:16 +11001687 if (!is_udplite) /* Disable the option on UDP sockets */
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08001688 return -ENOPROTOOPT;
1689 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1690 val = 8;
Alexey Dobriyan4be929b2010-05-24 14:33:03 -07001691 else if (val > USHRT_MAX)
1692 val = USHRT_MAX;
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08001693 up->pcslen = val;
1694 up->pcflag |= UDPLITE_SEND_CC;
1695 break;
1696
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001697 /* The receiver specifies a minimum checksum coverage value. To make
1698 * sense, this should be set to at least 8 (as done below). If zero is
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08001699 * used, this again means full checksum coverage. */
1700 case UDPLITE_RECV_CSCOV:
Wang Chenb2bf1e22007-12-03 22:34:16 +11001701 if (!is_udplite) /* Disable the option on UDP sockets */
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08001702 return -ENOPROTOOPT;
1703 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1704 val = 8;
Alexey Dobriyan4be929b2010-05-24 14:33:03 -07001705 else if (val > USHRT_MAX)
1706 val = USHRT_MAX;
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08001707 up->pcrlen = val;
1708 up->pcflag |= UDPLITE_RECV_CC;
1709 break;
1710
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 default:
1712 err = -ENOPROTOOPT;
1713 break;
Stephen Hemminger6516c652007-03-08 20:41:55 -08001714 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715
1716 return err;
1717}
Eric Dumazetc482c562009-07-17 00:26:32 +00001718EXPORT_SYMBOL(udp_lib_setsockopt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719
David S. Millerdb8dac22008-03-06 16:22:02 -08001720int udp_setsockopt(struct sock *sk, int level, int optname,
David S. Millerb7058842009-09-30 16:12:20 -07001721 char __user *optval, unsigned int optlen)
David S. Millerdb8dac22008-03-06 16:22:02 -08001722{
1723 if (level == SOL_UDP || level == SOL_UDPLITE)
1724 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1725 udp_push_pending_frames);
1726 return ip_setsockopt(sk, level, optname, optval, optlen);
1727}
1728
1729#ifdef CONFIG_COMPAT
1730int compat_udp_setsockopt(struct sock *sk, int level, int optname,
David S. Millerb7058842009-09-30 16:12:20 -07001731 char __user *optval, unsigned int optlen)
David S. Millerdb8dac22008-03-06 16:22:02 -08001732{
1733 if (level == SOL_UDP || level == SOL_UDPLITE)
1734 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1735 udp_push_pending_frames);
1736 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1737}
1738#endif
1739
Gerrit Renker4c0a6cb2006-11-27 09:29:59 -08001740int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1741 char __user *optval, int __user *optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742{
1743 struct udp_sock *up = udp_sk(sk);
1744 int val, len;
1745
Eric Dumazetc482c562009-07-17 00:26:32 +00001746 if (get_user(len, optlen))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 return -EFAULT;
1748
1749 len = min_t(unsigned int, len, sizeof(int));
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001750
Stephen Hemminger6516c652007-03-08 20:41:55 -08001751 if (len < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 return -EINVAL;
1753
Stephen Hemminger6516c652007-03-08 20:41:55 -08001754 switch (optname) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 case UDP_CORK:
1756 val = up->corkflag;
1757 break;
1758
1759 case UDP_ENCAP:
1760 val = up->encap_type;
1761 break;
1762
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08001763 /* The following two cannot be changed on UDP sockets, the return is
1764 * always 0 (which corresponds to the full checksum coverage of UDP). */
1765 case UDPLITE_SEND_CSCOV:
1766 val = up->pcslen;
1767 break;
1768
1769 case UDPLITE_RECV_CSCOV:
1770 val = up->pcrlen;
1771 break;
1772
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 default:
1774 return -ENOPROTOOPT;
Stephen Hemminger6516c652007-03-08 20:41:55 -08001775 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776
Stephen Hemminger6516c652007-03-08 20:41:55 -08001777 if (put_user(len, optlen))
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001778 return -EFAULT;
Eric Dumazetc482c562009-07-17 00:26:32 +00001779 if (copy_to_user(optval, &val, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 return -EFAULT;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001781 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782}
Eric Dumazetc482c562009-07-17 00:26:32 +00001783EXPORT_SYMBOL(udp_lib_getsockopt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784
David S. Millerdb8dac22008-03-06 16:22:02 -08001785int udp_getsockopt(struct sock *sk, int level, int optname,
1786 char __user *optval, int __user *optlen)
1787{
1788 if (level == SOL_UDP || level == SOL_UDPLITE)
1789 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1790 return ip_getsockopt(sk, level, optname, optval, optlen);
1791}
1792
1793#ifdef CONFIG_COMPAT
1794int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1795 char __user *optval, int __user *optlen)
1796{
1797 if (level == SOL_UDP || level == SOL_UDPLITE)
1798 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1799 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1800}
1801#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802/**
1803 * udp_poll - wait for a UDP event.
1804 * @file - file struct
1805 * @sock - socket
1806 * @wait - poll table
1807 *
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001808 * This is same as datagram poll, except for the special case of
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 * blocking sockets. If application is using a blocking fd
1810 * and a packet with checksum error is in the queue;
1811 * then it could get return from select indicating data available
1812 * but then block when reading it. Add special case code
1813 * to work around these arguably broken applications.
1814 */
1815unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1816{
1817 unsigned int mask = datagram_poll(file, sock, wait);
1818 struct sock *sk = sock->sk;
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08001819
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 /* Check for false positives due to checksum errors */
Eric Dumazet85584672009-10-09 04:43:40 +00001821 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
1822 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
1823 mask &= ~(POLLIN | POLLRDNORM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824
1825 return mask;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001826
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827}
Eric Dumazetc482c562009-07-17 00:26:32 +00001828EXPORT_SYMBOL(udp_poll);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829
David S. Millerdb8dac22008-03-06 16:22:02 -08001830struct proto udp_prot = {
1831 .name = "UDP",
1832 .owner = THIS_MODULE,
1833 .close = udp_lib_close,
1834 .connect = ip4_datagram_connect,
1835 .disconnect = udp_disconnect,
1836 .ioctl = udp_ioctl,
1837 .destroy = udp_destroy_sock,
1838 .setsockopt = udp_setsockopt,
1839 .getsockopt = udp_getsockopt,
1840 .sendmsg = udp_sendmsg,
1841 .recvmsg = udp_recvmsg,
1842 .sendpage = udp_sendpage,
Herbert Xu93821772008-09-15 11:48:46 -07001843 .backlog_rcv = __udp_queue_rcv_skb,
David S. Millerdb8dac22008-03-06 16:22:02 -08001844 .hash = udp_lib_hash,
1845 .unhash = udp_lib_unhash,
1846 .get_port = udp_v4_get_port,
1847 .memory_allocated = &udp_memory_allocated,
1848 .sysctl_mem = sysctl_udp_mem,
1849 .sysctl_wmem = &sysctl_udp_wmem_min,
1850 .sysctl_rmem = &sysctl_udp_rmem_min,
1851 .obj_size = sizeof(struct udp_sock),
Eric Dumazet271b72c2008-10-29 02:11:14 -07001852 .slab_flags = SLAB_DESTROY_BY_RCU,
Eric Dumazet645ca702008-10-29 01:41:45 -07001853 .h.udp_table = &udp_table,
David S. Millerdb8dac22008-03-06 16:22:02 -08001854#ifdef CONFIG_COMPAT
1855 .compat_setsockopt = compat_udp_setsockopt,
1856 .compat_getsockopt = compat_udp_getsockopt,
1857#endif
David S. Millerdb8dac22008-03-06 16:22:02 -08001858};
Eric Dumazetc482c562009-07-17 00:26:32 +00001859EXPORT_SYMBOL(udp_prot);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860
1861/* ------------------------------------------------------------------------ */
1862#ifdef CONFIG_PROC_FS
1863
Eric Dumazet645ca702008-10-29 01:41:45 -07001864static struct sock *udp_get_first(struct seq_file *seq, int start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865{
1866 struct sock *sk;
1867 struct udp_iter_state *state = seq->private;
Denis V. Lunev6f191ef2008-03-28 18:23:33 -07001868 struct net *net = seq_file_net(seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869
Eric Dumazetf86dcc52009-10-07 00:37:59 +00001870 for (state->bucket = start; state->bucket <= state->udp_table->mask;
1871 ++state->bucket) {
Eric Dumazet88ab1932008-11-16 19:39:21 -08001872 struct hlist_nulls_node *node;
Eric Dumazet645ca702008-10-29 01:41:45 -07001873 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
Eric Dumazetf86dcc52009-10-07 00:37:59 +00001874
1875 if (hlist_nulls_empty(&hslot->head))
1876 continue;
1877
Eric Dumazet645ca702008-10-29 01:41:45 -07001878 spin_lock_bh(&hslot->lock);
Eric Dumazet88ab1932008-11-16 19:39:21 -08001879 sk_nulls_for_each(sk, node, &hslot->head) {
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +09001880 if (!net_eq(sock_net(sk), net))
Daniel Lezcanoa91275e2008-03-21 04:11:58 -07001881 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 if (sk->sk_family == state->family)
1883 goto found;
1884 }
Eric Dumazet645ca702008-10-29 01:41:45 -07001885 spin_unlock_bh(&hslot->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 }
1887 sk = NULL;
1888found:
1889 return sk;
1890}
1891
1892static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1893{
1894 struct udp_iter_state *state = seq->private;
Denis V. Lunev6f191ef2008-03-28 18:23:33 -07001895 struct net *net = seq_file_net(seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896
1897 do {
Eric Dumazet88ab1932008-11-16 19:39:21 -08001898 sk = sk_nulls_next(sk);
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +09001899 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900
Eric Dumazet645ca702008-10-29 01:41:45 -07001901 if (!sk) {
Eric Dumazetf86dcc52009-10-07 00:37:59 +00001902 if (state->bucket <= state->udp_table->mask)
Vitaly Mayatskikh30842f22009-03-23 15:22:33 -07001903 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
Eric Dumazet645ca702008-10-29 01:41:45 -07001904 return udp_get_first(seq, state->bucket + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 }
1906 return sk;
1907}
1908
1909static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1910{
Eric Dumazet645ca702008-10-29 01:41:45 -07001911 struct sock *sk = udp_get_first(seq, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912
1913 if (sk)
Stephen Hemminger6516c652007-03-08 20:41:55 -08001914 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 --pos;
1916 return pos ? NULL : sk;
1917}
1918
1919static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1920{
Vitaly Mayatskikh30842f22009-03-23 15:22:33 -07001921 struct udp_iter_state *state = seq->private;
Eric Dumazetf86dcc52009-10-07 00:37:59 +00001922 state->bucket = MAX_UDP_PORTS;
Vitaly Mayatskikh30842f22009-03-23 15:22:33 -07001923
YOSHIFUJI Hideakib50660f2008-03-31 19:38:15 -07001924 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925}
1926
1927static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1928{
1929 struct sock *sk;
1930
YOSHIFUJI Hideakib50660f2008-03-31 19:38:15 -07001931 if (v == SEQ_START_TOKEN)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 sk = udp_get_idx(seq, 0);
1933 else
1934 sk = udp_get_next(seq, v);
1935
1936 ++*pos;
1937 return sk;
1938}
1939
1940static void udp_seq_stop(struct seq_file *seq, void *v)
1941{
Eric Dumazet645ca702008-10-29 01:41:45 -07001942 struct udp_iter_state *state = seq->private;
1943
Eric Dumazetf86dcc52009-10-07 00:37:59 +00001944 if (state->bucket <= state->udp_table->mask)
Eric Dumazet645ca702008-10-29 01:41:45 -07001945 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946}
1947
1948static int udp_seq_open(struct inode *inode, struct file *file)
1949{
1950 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
Denis V. Luneva2be75c2008-03-28 18:25:06 -07001951 struct udp_iter_state *s;
1952 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953
Denis V. Luneva2be75c2008-03-28 18:25:06 -07001954 err = seq_open_net(inode, file, &afinfo->seq_ops,
1955 sizeof(struct udp_iter_state));
1956 if (err < 0)
1957 return err;
Daniel Lezcanoa91275e2008-03-21 04:11:58 -07001958
Denis V. Luneva2be75c2008-03-28 18:25:06 -07001959 s = ((struct seq_file *)file->private_data)->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 s->family = afinfo->family;
Eric Dumazet645ca702008-10-29 01:41:45 -07001961 s->udp_table = afinfo->udp_table;
Denis V. Luneva2be75c2008-03-28 18:25:06 -07001962 return err;
Daniel Lezcanoa91275e2008-03-21 04:11:58 -07001963}
1964
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965/* ------------------------------------------------------------------------ */
Daniel Lezcano0c96d8c2008-03-21 04:14:17 -07001966int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967{
1968 struct proc_dir_entry *p;
1969 int rc = 0;
1970
Denis V. Lunev3ba94412008-03-28 18:25:32 -07001971 afinfo->seq_fops.open = udp_seq_open;
1972 afinfo->seq_fops.read = seq_read;
1973 afinfo->seq_fops.llseek = seq_lseek;
1974 afinfo->seq_fops.release = seq_release_net;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975
Denis V. Lunevdda61922008-03-28 18:24:26 -07001976 afinfo->seq_ops.start = udp_seq_start;
1977 afinfo->seq_ops.next = udp_seq_next;
1978 afinfo->seq_ops.stop = udp_seq_stop;
1979
Denis V. Lunev84841c32008-05-02 04:10:08 -07001980 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
1981 &afinfo->seq_fops, afinfo);
1982 if (!p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 rc = -ENOMEM;
1984 return rc;
1985}
Eric Dumazetc482c562009-07-17 00:26:32 +00001986EXPORT_SYMBOL(udp_proc_register);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987
Daniel Lezcano0c96d8c2008-03-21 04:14:17 -07001988void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989{
Daniel Lezcano0c96d8c2008-03-21 04:14:17 -07001990 proc_net_remove(net, afinfo->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991}
Eric Dumazetc482c562009-07-17 00:26:32 +00001992EXPORT_SYMBOL(udp_proc_unregister);
David S. Millerdb8dac22008-03-06 16:22:02 -08001993
1994/* ------------------------------------------------------------------------ */
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07001995static void udp4_format_sock(struct sock *sp, struct seq_file *f,
1996 int bucket, int *len)
David S. Millerdb8dac22008-03-06 16:22:02 -08001997{
1998 struct inet_sock *inet = inet_sk(sp);
Eric Dumazetc720c7e2009-10-15 06:30:45 +00001999 __be32 dest = inet->inet_daddr;
2000 __be32 src = inet->inet_rcv_saddr;
2001 __u16 destp = ntohs(inet->inet_dport);
2002 __u16 srcp = ntohs(inet->inet_sport);
David S. Millerdb8dac22008-03-06 16:22:02 -08002003
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002004 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
Eric Dumazetcb61cb92008-06-17 21:04:56 -07002005 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n",
David S. Millerdb8dac22008-03-06 16:22:02 -08002006 bucket, src, srcp, dest, destp, sp->sk_state,
Eric Dumazet31e6d362009-06-17 19:05:41 -07002007 sk_wmem_alloc_get(sp),
2008 sk_rmem_alloc_get(sp),
David S. Millerdb8dac22008-03-06 16:22:02 -08002009 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
Eric Dumazetcb61cb92008-06-17 21:04:56 -07002010 atomic_read(&sp->sk_refcnt), sp,
2011 atomic_read(&sp->sk_drops), len);
David S. Millerdb8dac22008-03-06 16:22:02 -08002012}
2013
2014int udp4_seq_show(struct seq_file *seq, void *v)
2015{
2016 if (v == SEQ_START_TOKEN)
2017 seq_printf(seq, "%-127s\n",
2018 " sl local_address rem_address st tx_queue "
2019 "rx_queue tr tm->when retrnsmt uid timeout "
Eric Dumazetcb61cb92008-06-17 21:04:56 -07002020 "inode ref pointer drops");
David S. Millerdb8dac22008-03-06 16:22:02 -08002021 else {
David S. Millerdb8dac22008-03-06 16:22:02 -08002022 struct udp_iter_state *state = seq->private;
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002023 int len;
David S. Millerdb8dac22008-03-06 16:22:02 -08002024
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002025 udp4_format_sock(v, seq, state->bucket, &len);
Eric Dumazetc482c562009-07-17 00:26:32 +00002026 seq_printf(seq, "%*s\n", 127 - len, "");
David S. Millerdb8dac22008-03-06 16:22:02 -08002027 }
2028 return 0;
2029}
2030
2031/* ------------------------------------------------------------------------ */
David S. Millerdb8dac22008-03-06 16:22:02 -08002032static struct udp_seq_afinfo udp4_seq_afinfo = {
David S. Millerdb8dac22008-03-06 16:22:02 -08002033 .name = "udp",
2034 .family = AF_INET,
Eric Dumazet645ca702008-10-29 01:41:45 -07002035 .udp_table = &udp_table,
Denis V. Lunev4ad96d32008-03-28 18:25:53 -07002036 .seq_fops = {
2037 .owner = THIS_MODULE,
2038 },
Denis V. Lunevdda61922008-03-28 18:24:26 -07002039 .seq_ops = {
2040 .show = udp4_seq_show,
2041 },
David S. Millerdb8dac22008-03-06 16:22:02 -08002042};
2043
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00002044static int __net_init udp4_proc_init_net(struct net *net)
Pavel Emelyanov15439fe2008-03-24 14:53:49 -07002045{
2046 return udp_proc_register(net, &udp4_seq_afinfo);
2047}
2048
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00002049static void __net_exit udp4_proc_exit_net(struct net *net)
Pavel Emelyanov15439fe2008-03-24 14:53:49 -07002050{
2051 udp_proc_unregister(net, &udp4_seq_afinfo);
2052}
2053
2054static struct pernet_operations udp4_net_ops = {
2055 .init = udp4_proc_init_net,
2056 .exit = udp4_proc_exit_net,
2057};
2058
David S. Millerdb8dac22008-03-06 16:22:02 -08002059int __init udp4_proc_init(void)
2060{
Pavel Emelyanov15439fe2008-03-24 14:53:49 -07002061 return register_pernet_subsys(&udp4_net_ops);
David S. Millerdb8dac22008-03-06 16:22:02 -08002062}
2063
2064void udp4_proc_exit(void)
2065{
Pavel Emelyanov15439fe2008-03-24 14:53:49 -07002066 unregister_pernet_subsys(&udp4_net_ops);
David S. Millerdb8dac22008-03-06 16:22:02 -08002067}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068#endif /* CONFIG_PROC_FS */
2069
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002070static __initdata unsigned long uhash_entries;
2071static int __init set_uhash_entries(char *str)
Eric Dumazet645ca702008-10-29 01:41:45 -07002072{
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002073 if (!str)
2074 return 0;
2075 uhash_entries = simple_strtoul(str, &str, 0);
2076 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2077 uhash_entries = UDP_HTABLE_SIZE_MIN;
2078 return 1;
2079}
2080__setup("uhash_entries=", set_uhash_entries);
Eric Dumazet645ca702008-10-29 01:41:45 -07002081
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002082void __init udp_table_init(struct udp_table *table, const char *name)
2083{
2084 unsigned int i;
2085
2086 if (!CONFIG_BASE_SMALL)
2087 table->hash = alloc_large_system_hash(name,
Eric Dumazet512615b2009-11-08 10:17:58 +00002088 2 * sizeof(struct udp_hslot),
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002089 uhash_entries,
2090 21, /* one slot per 2 MB */
2091 0,
2092 &table->log,
2093 &table->mask,
2094 64 * 1024);
2095 /*
2096 * Make sure hash table has the minimum size
2097 */
2098 if (CONFIG_BASE_SMALL || table->mask < UDP_HTABLE_SIZE_MIN - 1) {
2099 table->hash = kmalloc(UDP_HTABLE_SIZE_MIN *
Eric Dumazet512615b2009-11-08 10:17:58 +00002100 2 * sizeof(struct udp_hslot), GFP_KERNEL);
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002101 if (!table->hash)
2102 panic(name);
2103 table->log = ilog2(UDP_HTABLE_SIZE_MIN);
2104 table->mask = UDP_HTABLE_SIZE_MIN - 1;
2105 }
Eric Dumazet512615b2009-11-08 10:17:58 +00002106 table->hash2 = table->hash + (table->mask + 1);
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002107 for (i = 0; i <= table->mask; i++) {
Eric Dumazet88ab1932008-11-16 19:39:21 -08002108 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
Eric Dumazetfdcc8aa92009-11-08 10:17:05 +00002109 table->hash[i].count = 0;
Eric Dumazet645ca702008-10-29 01:41:45 -07002110 spin_lock_init(&table->hash[i].lock);
2111 }
Eric Dumazet512615b2009-11-08 10:17:58 +00002112 for (i = 0; i <= table->mask; i++) {
2113 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2114 table->hash2[i].count = 0;
2115 spin_lock_init(&table->hash2[i].lock);
2116 }
Eric Dumazet645ca702008-10-29 01:41:45 -07002117}
2118
Hideo Aoki95766ff2007-12-31 00:29:24 -08002119void __init udp_init(void)
2120{
Eric Dumazet8203efb2008-10-29 02:32:32 -07002121 unsigned long nr_pages, limit;
Hideo Aoki95766ff2007-12-31 00:29:24 -08002122
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002123 udp_table_init(&udp_table, "UDP");
Hideo Aoki95766ff2007-12-31 00:29:24 -08002124 /* Set the pressure threshold up by the same strategy of TCP. It is a
2125 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
2126 * toward zero with the amount of memory, with a floor of 128 pages.
2127 */
Eric Dumazet8203efb2008-10-29 02:32:32 -07002128 nr_pages = totalram_pages - totalhigh_pages;
2129 limit = min(nr_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
2130 limit = (limit * (nr_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
Hideo Aoki95766ff2007-12-31 00:29:24 -08002131 limit = max(limit, 128UL);
2132 sysctl_udp_mem[0] = limit / 4 * 3;
2133 sysctl_udp_mem[1] = limit;
2134 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2135
2136 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2137 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2138}
2139
Sridhar Samudralad7ca4cc2009-07-09 08:09:47 +00002140int udp4_ufo_send_check(struct sk_buff *skb)
2141{
2142 const struct iphdr *iph;
2143 struct udphdr *uh;
2144
2145 if (!pskb_may_pull(skb, sizeof(*uh)))
2146 return -EINVAL;
2147
2148 iph = ip_hdr(skb);
2149 uh = udp_hdr(skb);
2150
2151 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
2152 IPPROTO_UDP, 0);
2153 skb->csum_start = skb_transport_header(skb) - skb->head;
2154 skb->csum_offset = offsetof(struct udphdr, check);
2155 skb->ip_summed = CHECKSUM_PARTIAL;
2156 return 0;
2157}
2158
2159struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb, int features)
2160{
2161 struct sk_buff *segs = ERR_PTR(-EINVAL);
2162 unsigned int mss;
2163 int offset;
2164 __wsum csum;
2165
2166 mss = skb_shinfo(skb)->gso_size;
2167 if (unlikely(skb->len <= mss))
2168 goto out;
2169
2170 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
2171 /* Packet is from an untrusted source, reset gso_segs. */
2172 int type = skb_shinfo(skb)->gso_type;
2173
2174 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
2175 !(type & (SKB_GSO_UDP))))
2176 goto out;
2177
2178 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
2179
2180 segs = NULL;
2181 goto out;
2182 }
2183
2184 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
2185 * do checksum of UDP packets sent as multiple IP fragments.
2186 */
2187 offset = skb->csum_start - skb_headroom(skb);
Eric Dumazetc482c562009-07-17 00:26:32 +00002188 csum = skb_checksum(skb, offset, skb->len - offset, 0);
Sridhar Samudralad7ca4cc2009-07-09 08:09:47 +00002189 offset += skb->csum_offset;
2190 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2191 skb->ip_summed = CHECKSUM_NONE;
2192
2193 /* Fragment the skb. IP headers of the fragments are updated in
2194 * inet_gso_segment()
2195 */
2196 segs = skb_segment(skb, features);
2197out:
2198 return segs;
2199}
2200