blob: bf2b85b25491287f1f7f6517b8d2897dd1454ec1 [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
Joe Perchesafd465032012-03-12 07:03:32 +000080#define pr_fmt(fmt) "UDP: " fmt
81
Linus Torvalds1da177e2005-04-16 15:20:36 -070082#include <asm/uaccess.h>
83#include <asm/ioctls.h>
Hideo Aoki95766ff2007-12-31 00:29:24 -080084#include <linux/bootmem.h>
Eric Dumazet8203efb2008-10-29 02:32:32 -070085#include <linux/highmem.h>
86#include <linux/swap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070087#include <linux/types.h>
88#include <linux/fcntl.h>
89#include <linux/module.h>
90#include <linux/socket.h>
91#include <linux/sockios.h>
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -020092#include <linux/igmp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070093#include <linux/in.h>
94#include <linux/errno.h>
95#include <linux/timer.h>
96#include <linux/mm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070097#include <linux/inet.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070098#include <linux/netdevice.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090099#include <linux/slab.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -0700100#include <net/tcp_states.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101#include <linux/skbuff.h>
102#include <linux/proc_fs.h>
103#include <linux/seq_file.h>
Eric W. Biederman457c4cb2007-09-12 12:01:34 +0200104#include <net/net_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105#include <net/icmp.h>
Shawn Bohrer421b3882013-10-07 11:01:39 -0500106#include <net/inet_hashtables.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107#include <net/route.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108#include <net/checksum.h>
109#include <net/xfrm.h>
Satoru Moriya296f7ea2011-06-17 11:58:39 +0000110#include <trace/events/udp.h>
Eric Dumazet447167b2012-04-11 23:05:28 +0000111#include <linux/static_key.h>
Eric Dumazet22911fc2012-06-27 00:23:44 +0000112#include <trace/events/skb.h>
Eliezer Tamir076bb0c2013-07-10 17:13:17 +0300113#include <net/busy_poll.h>
Gerrit Renkerba4e58e2006-11-27 11:10:57 -0800114#include "udp_impl.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
Eric Dumazetf86dcc52009-10-07 00:37:59 +0000116struct udp_table udp_table __read_mostly;
Eric Dumazet645ca702008-10-29 01:41:45 -0700117EXPORT_SYMBOL(udp_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118
Eric Dumazet8d987e52010-11-09 23:24:26 +0000119long sysctl_udp_mem[3] __read_mostly;
Hideo Aoki95766ff2007-12-31 00:29:24 -0800120EXPORT_SYMBOL(sysctl_udp_mem);
Eric Dumazetc482c562009-07-17 00:26:32 +0000121
122int sysctl_udp_rmem_min __read_mostly;
Hideo Aoki95766ff2007-12-31 00:29:24 -0800123EXPORT_SYMBOL(sysctl_udp_rmem_min);
Eric Dumazetc482c562009-07-17 00:26:32 +0000124
125int sysctl_udp_wmem_min __read_mostly;
Hideo Aoki95766ff2007-12-31 00:29:24 -0800126EXPORT_SYMBOL(sysctl_udp_wmem_min);
127
Eric Dumazet8d987e52010-11-09 23:24:26 +0000128atomic_long_t udp_memory_allocated;
Hideo Aoki95766ff2007-12-31 00:29:24 -0800129EXPORT_SYMBOL(udp_memory_allocated);
130
Eric Dumazetf86dcc52009-10-07 00:37:59 +0000131#define MAX_UDP_PORTS 65536
132#define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
Eric Dumazet98322f22009-01-26 21:35:35 -0800133
Eric Dumazetf24d43c2008-10-09 14:51:27 -0700134static int udp_lib_lport_inuse(struct net *net, __u16 num,
Eric Dumazet645ca702008-10-29 01:41:45 -0700135 const struct udp_hslot *hslot,
Eric Dumazet98322f22009-01-26 21:35:35 -0800136 unsigned long *bitmap,
Eric Dumazetf24d43c2008-10-09 14:51:27 -0700137 struct sock *sk,
138 int (*saddr_comp)(const struct sock *sk1,
Eric Dumazetf86dcc52009-10-07 00:37:59 +0000139 const struct sock *sk2),
140 unsigned int log)
Gerrit Renker25030a72006-08-26 20:06:05 -0700141{
Eric Dumazetf24d43c2008-10-09 14:51:27 -0700142 struct sock *sk2;
Eric Dumazet88ab1932008-11-16 19:39:21 -0800143 struct hlist_nulls_node *node;
Tom Herbertba418fa2013-01-22 09:50:32 +0000144 kuid_t uid = sock_i_uid(sk);
Gerrit Renker25030a72006-08-26 20:06:05 -0700145
Eric Dumazet88ab1932008-11-16 19:39:21 -0800146 sk_nulls_for_each(sk2, node, &hslot->head)
Joe Perches9d4fb272009-11-23 10:41:23 -0800147 if (net_eq(sock_net(sk2), net) &&
148 sk2 != sk &&
Eric Dumazetd4cada42009-11-08 10:17:30 +0000149 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
Joe Perches9d4fb272009-11-23 10:41:23 -0800150 (!sk2->sk_reuse || !sk->sk_reuse) &&
151 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
152 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
Tom Herbertba418fa2013-01-22 09:50:32 +0000153 (!sk2->sk_reuseport || !sk->sk_reuseport ||
154 !uid_eq(uid, sock_i_uid(sk2))) &&
Eric Dumazet98322f22009-01-26 21:35:35 -0800155 (*saddr_comp)(sk, sk2)) {
156 if (bitmap)
Eric Dumazetd4cada42009-11-08 10:17:30 +0000157 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
158 bitmap);
Eric Dumazet98322f22009-01-26 21:35:35 -0800159 else
160 return 1;
161 }
Gerrit Renker25030a72006-08-26 20:06:05 -0700162 return 0;
163}
164
Eric Dumazet30fff922009-11-09 05:26:33 +0000165/*
166 * Note: we still hold spinlock of primary hash chain, so no other writer
167 * can insert/delete a socket with local_port == num
168 */
169static int udp_lib_lport_inuse2(struct net *net, __u16 num,
170 struct udp_hslot *hslot2,
171 struct sock *sk,
172 int (*saddr_comp)(const struct sock *sk1,
173 const struct sock *sk2))
174{
175 struct sock *sk2;
176 struct hlist_nulls_node *node;
Tom Herbertba418fa2013-01-22 09:50:32 +0000177 kuid_t uid = sock_i_uid(sk);
Eric Dumazet30fff922009-11-09 05:26:33 +0000178 int res = 0;
179
180 spin_lock(&hslot2->lock);
181 udp_portaddr_for_each_entry(sk2, node, &hslot2->head)
Joe Perches9d4fb272009-11-23 10:41:23 -0800182 if (net_eq(sock_net(sk2), net) &&
183 sk2 != sk &&
184 (udp_sk(sk2)->udp_port_hash == num) &&
185 (!sk2->sk_reuse || !sk->sk_reuse) &&
186 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
187 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
Tom Herbertba418fa2013-01-22 09:50:32 +0000188 (!sk2->sk_reuseport || !sk->sk_reuseport ||
189 !uid_eq(uid, sock_i_uid(sk2))) &&
Eric Dumazet30fff922009-11-09 05:26:33 +0000190 (*saddr_comp)(sk, sk2)) {
191 res = 1;
192 break;
193 }
194 spin_unlock(&hslot2->lock);
195 return res;
196}
197
Gerrit Renker25030a72006-08-26 20:06:05 -0700198/**
Pavel Emelyanov6ba5a3c2008-03-22 16:51:21 -0700199 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
Gerrit Renker25030a72006-08-26 20:06:05 -0700200 *
201 * @sk: socket struct in question
202 * @snum: port number to look up
David S. Millerdf2bc452007-06-05 15:18:43 -0700203 * @saddr_comp: AF-dependent comparison of bound local IP addresses
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300204 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
Eric Dumazet30fff922009-11-09 05:26:33 +0000205 * with NULL address
Gerrit Renker25030a72006-08-26 20:06:05 -0700206 */
Pavel Emelyanov6ba5a3c2008-03-22 16:51:21 -0700207int udp_lib_get_port(struct sock *sk, unsigned short snum,
David S. Millerdf2bc452007-06-05 15:18:43 -0700208 int (*saddr_comp)(const struct sock *sk1,
Eric Dumazet30fff922009-11-09 05:26:33 +0000209 const struct sock *sk2),
210 unsigned int hash2_nulladdr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211{
Eric Dumazet512615b2009-11-08 10:17:58 +0000212 struct udp_hslot *hslot, *hslot2;
Eric Dumazet645ca702008-10-29 01:41:45 -0700213 struct udp_table *udptable = sk->sk_prot->h.udp_table;
Gerrit Renker25030a72006-08-26 20:06:05 -0700214 int error = 1;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900215 struct net *net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216
Stephen Hemminger32c1da72007-08-24 23:09:41 -0700217 if (!snum) {
Eric Dumazet9088c562008-10-08 11:44:17 -0700218 int low, high, remaining;
Eric Dumazet95c96172012-04-15 05:58:06 +0000219 unsigned int rand;
Eric Dumazet98322f22009-01-26 21:35:35 -0800220 unsigned short first, last;
221 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222
Eric W. Biederman0bbf87d2013-09-28 14:10:59 -0700223 inet_get_local_port_range(net, &low, &high);
Anton Arapova25de532007-10-18 22:00:17 -0700224 remaining = (high - low) + 1;
Stephen Hemminger227b60f2007-10-10 17:30:46 -0700225
Eric Dumazet9088c562008-10-08 11:44:17 -0700226 rand = net_random();
Eric Dumazet98322f22009-01-26 21:35:35 -0800227 first = (((u64)rand * remaining) >> 32) + low;
228 /*
229 * force rand to be an odd multiple of UDP_HTABLE_SIZE
230 */
Eric Dumazetf86dcc52009-10-07 00:37:59 +0000231 rand = (rand | 1) * (udptable->mask + 1);
Eric Dumazet5781b232009-12-13 19:32:39 -0800232 last = first + udptable->mask + 1;
233 do {
Eric Dumazetf86dcc52009-10-07 00:37:59 +0000234 hslot = udp_hashslot(udptable, net, first);
Eric Dumazet98322f22009-01-26 21:35:35 -0800235 bitmap_zero(bitmap, PORTS_PER_CHAIN);
Eric Dumazet645ca702008-10-29 01:41:45 -0700236 spin_lock_bh(&hslot->lock);
Eric Dumazet98322f22009-01-26 21:35:35 -0800237 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
Eric Dumazetf86dcc52009-10-07 00:37:59 +0000238 saddr_comp, udptable->log);
Eric Dumazet98322f22009-01-26 21:35:35 -0800239
240 snum = first;
241 /*
242 * Iterate on all possible values of snum for this hash.
243 * Using steps of an odd multiple of UDP_HTABLE_SIZE
244 * give us randomization and full range coverage.
245 */
Eric Dumazet9088c562008-10-08 11:44:17 -0700246 do {
Eric Dumazet98322f22009-01-26 21:35:35 -0800247 if (low <= snum && snum <= high &&
Amerigo Wange3826f12010-05-05 00:27:06 +0000248 !test_bit(snum >> udptable->log, bitmap) &&
249 !inet_is_reserved_local_port(snum))
Eric Dumazet98322f22009-01-26 21:35:35 -0800250 goto found;
251 snum += rand;
252 } while (snum != first);
253 spin_unlock_bh(&hslot->lock);
Eric Dumazet5781b232009-12-13 19:32:39 -0800254 } while (++first != last);
Eric Dumazet98322f22009-01-26 21:35:35 -0800255 goto fail;
Eric Dumazet645ca702008-10-29 01:41:45 -0700256 } else {
Eric Dumazetf86dcc52009-10-07 00:37:59 +0000257 hslot = udp_hashslot(udptable, net, snum);
Eric Dumazet645ca702008-10-29 01:41:45 -0700258 spin_lock_bh(&hslot->lock);
Eric Dumazet30fff922009-11-09 05:26:33 +0000259 if (hslot->count > 10) {
260 int exist;
261 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
262
263 slot2 &= udptable->mask;
264 hash2_nulladdr &= udptable->mask;
265
266 hslot2 = udp_hashslot2(udptable, slot2);
267 if (hslot->count < hslot2->count)
268 goto scan_primary_hash;
269
270 exist = udp_lib_lport_inuse2(net, snum, hslot2,
271 sk, saddr_comp);
272 if (!exist && (hash2_nulladdr != slot2)) {
273 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
274 exist = udp_lib_lport_inuse2(net, snum, hslot2,
275 sk, saddr_comp);
276 }
277 if (exist)
278 goto fail_unlock;
279 else
280 goto found;
281 }
282scan_primary_hash:
Eric Dumazetf86dcc52009-10-07 00:37:59 +0000283 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
284 saddr_comp, 0))
Eric Dumazet645ca702008-10-29 01:41:45 -0700285 goto fail_unlock;
286 }
Eric Dumazet98322f22009-01-26 21:35:35 -0800287found:
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000288 inet_sk(sk)->inet_num = snum;
Eric Dumazetd4cada42009-11-08 10:17:30 +0000289 udp_sk(sk)->udp_port_hash = snum;
290 udp_sk(sk)->udp_portaddr_hash ^= snum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 if (sk_unhashed(sk)) {
Eric Dumazet88ab1932008-11-16 19:39:21 -0800292 sk_nulls_add_node_rcu(sk, &hslot->head);
Eric Dumazetfdcc8aa92009-11-08 10:17:05 +0000293 hslot->count++;
Pavel Emelyanovc29a0bc2008-03-31 19:41:46 -0700294 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
Eric Dumazet512615b2009-11-08 10:17:58 +0000295
296 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
297 spin_lock(&hslot2->lock);
298 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
299 &hslot2->head);
300 hslot2->count++;
301 spin_unlock(&hslot2->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 }
Gerrit Renker25030a72006-08-26 20:06:05 -0700303 error = 0;
Eric Dumazet645ca702008-10-29 01:41:45 -0700304fail_unlock:
305 spin_unlock_bh(&hslot->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306fail:
Gerrit Renker25030a72006-08-26 20:06:05 -0700307 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308}
Eric Dumazetc482c562009-07-17 00:26:32 +0000309EXPORT_SYMBOL(udp_lib_get_port);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310
Vlad Yasevich499923c2009-04-09 17:37:33 +0000311static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
David S. Millerdb8dac22008-03-06 16:22:02 -0800312{
313 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
314
Eric Dumazetc482c562009-07-17 00:26:32 +0000315 return (!ipv6_only_sock(sk2) &&
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000316 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
317 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
David S. Millerdb8dac22008-03-06 16:22:02 -0800318}
319
Eric Dumazetd4cada42009-11-08 10:17:30 +0000320static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr,
321 unsigned int port)
322{
Eric Dumazet0eae88f2010-04-20 19:06:52 -0700323 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
Eric Dumazetd4cada42009-11-08 10:17:30 +0000324}
325
Pavel Emelyanov6ba5a3c2008-03-22 16:51:21 -0700326int udp_v4_get_port(struct sock *sk, unsigned short snum)
David S. Millerdb8dac22008-03-06 16:22:02 -0800327{
Eric Dumazet30fff922009-11-09 05:26:33 +0000328 unsigned int hash2_nulladdr =
Eric Dumazet0eae88f2010-04-20 19:06:52 -0700329 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
Eric Dumazet30fff922009-11-09 05:26:33 +0000330 unsigned int hash2_partial =
331 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
332
Eric Dumazetd4cada42009-11-08 10:17:30 +0000333 /* precompute partial secondary hash */
Eric Dumazet30fff922009-11-09 05:26:33 +0000334 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
335 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
David S. Millerdb8dac22008-03-06 16:22:02 -0800336}
337
Eric Dumazet645ca702008-10-29 01:41:45 -0700338static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
339 unsigned short hnum,
340 __be16 sport, __be32 daddr, __be16 dport, int dif)
341{
342 int score = -1;
343
Eric Dumazetd4cada42009-11-08 10:17:30 +0000344 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
Eric Dumazet645ca702008-10-29 01:41:45 -0700345 !ipv6_only_sock(sk)) {
346 struct inet_sock *inet = inet_sk(sk);
347
Tom Herbertba418fa2013-01-22 09:50:32 +0000348 score = (sk->sk_family == PF_INET ? 2 : 1);
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000349 if (inet->inet_rcv_saddr) {
350 if (inet->inet_rcv_saddr != daddr)
Eric Dumazet645ca702008-10-29 01:41:45 -0700351 return -1;
Tom Herbertba418fa2013-01-22 09:50:32 +0000352 score += 4;
Eric Dumazet645ca702008-10-29 01:41:45 -0700353 }
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000354 if (inet->inet_daddr) {
355 if (inet->inet_daddr != saddr)
Eric Dumazet645ca702008-10-29 01:41:45 -0700356 return -1;
Tom Herbertba418fa2013-01-22 09:50:32 +0000357 score += 4;
Eric Dumazet645ca702008-10-29 01:41:45 -0700358 }
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000359 if (inet->inet_dport) {
360 if (inet->inet_dport != sport)
Eric Dumazet645ca702008-10-29 01:41:45 -0700361 return -1;
Tom Herbertba418fa2013-01-22 09:50:32 +0000362 score += 4;
Eric Dumazet645ca702008-10-29 01:41:45 -0700363 }
364 if (sk->sk_bound_dev_if) {
365 if (sk->sk_bound_dev_if != dif)
366 return -1;
Tom Herbertba418fa2013-01-22 09:50:32 +0000367 score += 4;
Eric Dumazet645ca702008-10-29 01:41:45 -0700368 }
369 }
370 return score;
371}
372
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000373/*
374 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
375 */
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000376static inline int compute_score2(struct sock *sk, struct net *net,
377 __be32 saddr, __be16 sport,
378 __be32 daddr, unsigned int hnum, int dif)
379{
380 int score = -1;
381
382 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) {
383 struct inet_sock *inet = inet_sk(sk);
384
385 if (inet->inet_rcv_saddr != daddr)
386 return -1;
387 if (inet->inet_num != hnum)
388 return -1;
389
Tom Herbertba418fa2013-01-22 09:50:32 +0000390 score = (sk->sk_family == PF_INET ? 2 : 1);
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000391 if (inet->inet_daddr) {
392 if (inet->inet_daddr != saddr)
393 return -1;
Tom Herbertba418fa2013-01-22 09:50:32 +0000394 score += 4;
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000395 }
396 if (inet->inet_dport) {
397 if (inet->inet_dport != sport)
398 return -1;
Tom Herbertba418fa2013-01-22 09:50:32 +0000399 score += 4;
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000400 }
401 if (sk->sk_bound_dev_if) {
402 if (sk->sk_bound_dev_if != dif)
403 return -1;
Tom Herbertba418fa2013-01-22 09:50:32 +0000404 score += 4;
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000405 }
406 }
407 return score;
408}
409
Hannes Frederic Sowa65cd8032013-10-19 21:48:51 +0200410static unsigned int udp_ehashfn(struct net *net, const __be32 laddr,
411 const __u16 lport, const __be32 faddr,
412 const __be16 fport)
413{
Hannes Frederic Sowa1bbdcee2013-10-19 21:48:57 +0200414 static u32 udp_ehash_secret __read_mostly;
415
416 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
417
Hannes Frederic Sowa65cd8032013-10-19 21:48:51 +0200418 return __inet_ehashfn(laddr, lport, faddr, fport,
Hannes Frederic Sowa1bbdcee2013-10-19 21:48:57 +0200419 udp_ehash_secret + net_hash_mix(net));
Hannes Frederic Sowa65cd8032013-10-19 21:48:51 +0200420}
421
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000422
423/* called with read_rcu_lock() */
424static struct sock *udp4_lib_lookup2(struct net *net,
425 __be32 saddr, __be16 sport,
426 __be32 daddr, unsigned int hnum, int dif,
427 struct udp_hslot *hslot2, unsigned int slot2)
428{
429 struct sock *sk, *result;
430 struct hlist_nulls_node *node;
Tom Herbertba418fa2013-01-22 09:50:32 +0000431 int score, badness, matches = 0, reuseport = 0;
432 u32 hash = 0;
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000433
434begin:
435 result = NULL;
Tom Herbertba418fa2013-01-22 09:50:32 +0000436 badness = 0;
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000437 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
438 score = compute_score2(sk, net, saddr, sport,
439 daddr, hnum, dif);
440 if (score > badness) {
441 result = sk;
442 badness = score;
Tom Herbertba418fa2013-01-22 09:50:32 +0000443 reuseport = sk->sk_reuseport;
444 if (reuseport) {
Hannes Frederic Sowa65cd8032013-10-19 21:48:51 +0200445 hash = udp_ehashfn(net, daddr, hnum,
446 saddr, sport);
Tom Herbertba418fa2013-01-22 09:50:32 +0000447 matches = 1;
448 }
449 } else if (score == badness && reuseport) {
450 matches++;
451 if (((u64)hash * matches) >> 32 == 0)
452 result = sk;
453 hash = next_pseudo_random32(hash);
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000454 }
455 }
456 /*
457 * if the nulls value we got at the end of this lookup is
458 * not the expected one, we must restart lookup.
459 * We probably met an item that was moved to another chain.
460 */
461 if (get_nulls_value(node) != slot2)
462 goto begin;
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000463 if (result) {
Eric Dumazetc31504d2010-11-15 19:58:26 +0000464 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000465 result = NULL;
466 else if (unlikely(compute_score2(result, net, saddr, sport,
467 daddr, hnum, dif) < badness)) {
468 sock_put(result);
469 goto begin;
470 }
471 }
472 return result;
473}
474
David S. Millerdb8dac22008-03-06 16:22:02 -0800475/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
476 * harder than this. -DaveM
477 */
Pavel Emelyanovfce82332011-12-09 06:23:34 +0000478struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
David S. Millerdb8dac22008-03-06 16:22:02 -0800479 __be16 sport, __be32 daddr, __be16 dport,
Eric Dumazet645ca702008-10-29 01:41:45 -0700480 int dif, struct udp_table *udptable)
David S. Millerdb8dac22008-03-06 16:22:02 -0800481{
Eric Dumazet271b72c2008-10-29 02:11:14 -0700482 struct sock *sk, *result;
Eric Dumazet88ab1932008-11-16 19:39:21 -0800483 struct hlist_nulls_node *node;
David S. Millerdb8dac22008-03-06 16:22:02 -0800484 unsigned short hnum = ntohs(dport);
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000485 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
486 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
Tom Herbertba418fa2013-01-22 09:50:32 +0000487 int score, badness, matches = 0, reuseport = 0;
488 u32 hash = 0;
David S. Millerdb8dac22008-03-06 16:22:02 -0800489
Eric Dumazet271b72c2008-10-29 02:11:14 -0700490 rcu_read_lock();
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000491 if (hslot->count > 10) {
492 hash2 = udp4_portaddr_hash(net, daddr, hnum);
493 slot2 = hash2 & udptable->mask;
494 hslot2 = &udptable->hash2[slot2];
495 if (hslot->count < hslot2->count)
496 goto begin;
497
498 result = udp4_lib_lookup2(net, saddr, sport,
499 daddr, hnum, dif,
500 hslot2, slot2);
501 if (!result) {
Eric Dumazet0eae88f2010-04-20 19:06:52 -0700502 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000503 slot2 = hash2 & udptable->mask;
504 hslot2 = &udptable->hash2[slot2];
505 if (hslot->count < hslot2->count)
506 goto begin;
507
Jorge Boncompte [DTI2]1223c672010-04-08 04:56:48 +0000508 result = udp4_lib_lookup2(net, saddr, sport,
Eric Dumazet0eae88f2010-04-20 19:06:52 -0700509 htonl(INADDR_ANY), hnum, dif,
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000510 hslot2, slot2);
511 }
512 rcu_read_unlock();
513 return result;
514 }
Eric Dumazet271b72c2008-10-29 02:11:14 -0700515begin:
516 result = NULL;
Tom Herbertba418fa2013-01-22 09:50:32 +0000517 badness = 0;
Eric Dumazet88ab1932008-11-16 19:39:21 -0800518 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
Eric Dumazet645ca702008-10-29 01:41:45 -0700519 score = compute_score(sk, net, saddr, hnum, sport,
520 daddr, dport, dif);
521 if (score > badness) {
522 result = sk;
523 badness = score;
Tom Herbertba418fa2013-01-22 09:50:32 +0000524 reuseport = sk->sk_reuseport;
525 if (reuseport) {
Hannes Frederic Sowa65cd8032013-10-19 21:48:51 +0200526 hash = udp_ehashfn(net, daddr, hnum,
527 saddr, sport);
Tom Herbertba418fa2013-01-22 09:50:32 +0000528 matches = 1;
529 }
530 } else if (score == badness && reuseport) {
531 matches++;
532 if (((u64)hash * matches) >> 32 == 0)
533 result = sk;
534 hash = next_pseudo_random32(hash);
David S. Millerdb8dac22008-03-06 16:22:02 -0800535 }
536 }
Eric Dumazet88ab1932008-11-16 19:39:21 -0800537 /*
538 * if the nulls value we got at the end of this lookup is
539 * not the expected one, we must restart lookup.
540 * We probably met an item that was moved to another chain.
541 */
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000542 if (get_nulls_value(node) != slot)
Eric Dumazet88ab1932008-11-16 19:39:21 -0800543 goto begin;
544
Eric Dumazet271b72c2008-10-29 02:11:14 -0700545 if (result) {
Eric Dumazetc31504d2010-11-15 19:58:26 +0000546 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
Eric Dumazet271b72c2008-10-29 02:11:14 -0700547 result = NULL;
548 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
549 daddr, dport, dif) < badness)) {
550 sock_put(result);
551 goto begin;
552 }
553 }
554 rcu_read_unlock();
David S. Millerdb8dac22008-03-06 16:22:02 -0800555 return result;
556}
Pavel Emelyanovfce82332011-12-09 06:23:34 +0000557EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
David S. Millerdb8dac22008-03-06 16:22:02 -0800558
KOVACS Krisztian607c4aa2008-10-07 12:38:32 -0700559static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
560 __be16 sport, __be16 dport,
Eric Dumazet645ca702008-10-29 01:41:45 -0700561 struct udp_table *udptable)
KOVACS Krisztian607c4aa2008-10-07 12:38:32 -0700562{
KOVACS Krisztian23542612008-10-07 12:41:01 -0700563 struct sock *sk;
KOVACS Krisztian607c4aa2008-10-07 12:38:32 -0700564 const struct iphdr *iph = ip_hdr(skb);
565
KOVACS Krisztian23542612008-10-07 12:41:01 -0700566 if (unlikely(sk = skb_steal_sock(skb)))
567 return sk;
568 else
Eric Dumazetadf30902009-06-02 05:19:30 +0000569 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
KOVACS Krisztian23542612008-10-07 12:41:01 -0700570 iph->daddr, dport, inet_iif(skb),
571 udptable);
KOVACS Krisztian607c4aa2008-10-07 12:38:32 -0700572}
573
KOVACS Krisztianbcd41302008-10-01 07:48:10 -0700574struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
575 __be32 daddr, __be16 dport, int dif)
576{
Eric Dumazet645ca702008-10-29 01:41:45 -0700577 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
KOVACS Krisztianbcd41302008-10-01 07:48:10 -0700578}
579EXPORT_SYMBOL_GPL(udp4_lib_lookup);
580
Shawn Bohrer421b3882013-10-07 11:01:39 -0500581static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
582 __be16 loc_port, __be32 loc_addr,
583 __be16 rmt_port, __be32 rmt_addr,
584 int dif, unsigned short hnum)
585{
586 struct inet_sock *inet = inet_sk(sk);
587
588 if (!net_eq(sock_net(sk), net) ||
589 udp_sk(sk)->udp_port_hash != hnum ||
590 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
591 (inet->inet_dport != rmt_port && inet->inet_dport) ||
592 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
593 ipv6_only_sock(sk) ||
594 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
595 return false;
596 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif))
597 return false;
598 return true;
599}
600
Eric Dumazet920a4612008-11-01 21:22:23 -0700601static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
David S. Millerdb8dac22008-03-06 16:22:02 -0800602 __be16 loc_port, __be32 loc_addr,
603 __be16 rmt_port, __be32 rmt_addr,
604 int dif)
605{
Eric Dumazet88ab1932008-11-16 19:39:21 -0800606 struct hlist_nulls_node *node;
David S. Millerdb8dac22008-03-06 16:22:02 -0800607 struct sock *s = sk;
608 unsigned short hnum = ntohs(loc_port);
609
Eric Dumazet88ab1932008-11-16 19:39:21 -0800610 sk_nulls_for_each_from(s, node) {
Shawn Bohrer421b3882013-10-07 11:01:39 -0500611 if (__udp_is_mcast_sock(net, s,
612 loc_port, loc_addr,
613 rmt_port, rmt_addr,
614 dif, hnum))
615 goto found;
David S. Millerdb8dac22008-03-06 16:22:02 -0800616 }
617 s = NULL;
618found:
619 return s;
620}
621
622/*
623 * This routine is called by the ICMP module when it gets some
624 * sort of error condition. If err < 0 then the socket should
625 * be closed and the error returned to the user. If err > 0
626 * it's just the icmp type << 8 | icmp code.
627 * Header points to the ip header of the error packet. We move
628 * on past this. Then (as it used to claim before adjustment)
629 * header points to the first 8 bytes of the udp header. We need
630 * to find the appropriate port.
631 */
632
Eric Dumazet645ca702008-10-29 01:41:45 -0700633void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
David S. Millerdb8dac22008-03-06 16:22:02 -0800634{
635 struct inet_sock *inet;
Eric Dumazetb71d1d42011-04-22 04:53:02 +0000636 const struct iphdr *iph = (const struct iphdr *)skb->data;
Eric Dumazetc482c562009-07-17 00:26:32 +0000637 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
David S. Millerdb8dac22008-03-06 16:22:02 -0800638 const int type = icmp_hdr(skb)->type;
639 const int code = icmp_hdr(skb)->code;
640 struct sock *sk;
641 int harderr;
642 int err;
Pavel Emelyanovfd54d712008-07-14 23:01:40 -0700643 struct net *net = dev_net(skb->dev);
David S. Millerdb8dac22008-03-06 16:22:02 -0800644
Pavel Emelyanovfd54d712008-07-14 23:01:40 -0700645 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
David S. Millerdb8dac22008-03-06 16:22:02 -0800646 iph->saddr, uh->source, skb->dev->ifindex, udptable);
647 if (sk == NULL) {
Pavel Emelyanovdcfc23c2008-07-14 23:03:00 -0700648 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
David S. Millerdb8dac22008-03-06 16:22:02 -0800649 return; /* No socket for error */
650 }
651
652 err = 0;
653 harderr = 0;
654 inet = inet_sk(sk);
655
656 switch (type) {
657 default:
658 case ICMP_TIME_EXCEEDED:
659 err = EHOSTUNREACH;
660 break;
661 case ICMP_SOURCE_QUENCH:
662 goto out;
663 case ICMP_PARAMETERPROB:
664 err = EPROTO;
665 harderr = 1;
666 break;
667 case ICMP_DEST_UNREACH:
668 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
David S. Miller36393392012-06-14 22:21:46 -0700669 ipv4_sk_update_pmtu(skb, sk, info);
David S. Millerdb8dac22008-03-06 16:22:02 -0800670 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
671 err = EMSGSIZE;
672 harderr = 1;
673 break;
674 }
675 goto out;
676 }
677 err = EHOSTUNREACH;
678 if (code <= NR_ICMP_UNREACH) {
679 harderr = icmp_err_convert[code].fatal;
680 err = icmp_err_convert[code].errno;
681 }
682 break;
David S. Miller55be7a92012-07-11 21:27:49 -0700683 case ICMP_REDIRECT:
684 ipv4_sk_redirect(skb, sk);
Duan Jiong1a462d12013-09-20 18:20:28 +0800685 goto out;
David S. Millerdb8dac22008-03-06 16:22:02 -0800686 }
687
688 /*
689 * RFC1122: OK. Passes ICMP errors back to application, as per
690 * 4.1.3.3.
691 */
692 if (!inet->recverr) {
693 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
694 goto out;
Eric Dumazetb1faf562010-05-31 23:44:05 -0700695 } else
Eric Dumazetc482c562009-07-17 00:26:32 +0000696 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
Eric Dumazetb1faf562010-05-31 23:44:05 -0700697
David S. Millerdb8dac22008-03-06 16:22:02 -0800698 sk->sk_err = err;
699 sk->sk_error_report(sk);
700out:
701 sock_put(sk);
702}
703
704void udp_err(struct sk_buff *skb, u32 info)
705{
Eric Dumazet645ca702008-10-29 01:41:45 -0700706 __udp4_lib_err(skb, info, &udp_table);
David S. Millerdb8dac22008-03-06 16:22:02 -0800707}
708
709/*
710 * Throw away all pending data and cancel the corking. Socket is locked.
711 */
Denis V. Lunev36d926b2008-06-04 15:49:07 +0400712void udp_flush_pending_frames(struct sock *sk)
David S. Millerdb8dac22008-03-06 16:22:02 -0800713{
714 struct udp_sock *up = udp_sk(sk);
715
716 if (up->pending) {
717 up->len = 0;
718 up->pending = 0;
719 ip_flush_pending_frames(sk);
720 }
721}
Denis V. Lunev36d926b2008-06-04 15:49:07 +0400722EXPORT_SYMBOL(udp_flush_pending_frames);
David S. Millerdb8dac22008-03-06 16:22:02 -0800723
724/**
Herbert Xuf6b96642011-03-01 02:36:48 +0000725 * udp4_hwcsum - handle outgoing HW checksumming
David S. Millerdb8dac22008-03-06 16:22:02 -0800726 * @skb: sk_buff containing the filled-in UDP header
727 * (checksum field must be zeroed out)
Herbert Xuf6b96642011-03-01 02:36:48 +0000728 * @src: source IP address
729 * @dst: destination IP address
David S. Millerdb8dac22008-03-06 16:22:02 -0800730 */
Thomas Grafc26bf4a2013-07-25 18:12:18 +0200731void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
David S. Millerdb8dac22008-03-06 16:22:02 -0800732{
David S. Millerdb8dac22008-03-06 16:22:02 -0800733 struct udphdr *uh = udp_hdr(skb);
Herbert Xuf6b96642011-03-01 02:36:48 +0000734 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
735 int offset = skb_transport_offset(skb);
736 int len = skb->len - offset;
737 int hlen = len;
David S. Millerdb8dac22008-03-06 16:22:02 -0800738 __wsum csum = 0;
739
Herbert Xuf6b96642011-03-01 02:36:48 +0000740 if (!frags) {
David S. Millerdb8dac22008-03-06 16:22:02 -0800741 /*
742 * Only one fragment on the socket.
743 */
744 skb->csum_start = skb_transport_header(skb) - skb->head;
745 skb->csum_offset = offsetof(struct udphdr, check);
Herbert Xuf6b96642011-03-01 02:36:48 +0000746 uh->check = ~csum_tcpudp_magic(src, dst, len,
747 IPPROTO_UDP, 0);
David S. Millerdb8dac22008-03-06 16:22:02 -0800748 } else {
749 /*
750 * HW-checksum won't work as there are two or more
751 * fragments on the socket so that all csums of sk_buffs
752 * should be together
753 */
Herbert Xuf6b96642011-03-01 02:36:48 +0000754 do {
755 csum = csum_add(csum, frags->csum);
756 hlen -= frags->len;
757 } while ((frags = frags->next));
David S. Millerdb8dac22008-03-06 16:22:02 -0800758
Herbert Xuf6b96642011-03-01 02:36:48 +0000759 csum = skb_checksum(skb, offset, hlen, csum);
David S. Millerdb8dac22008-03-06 16:22:02 -0800760 skb->ip_summed = CHECKSUM_NONE;
761
David S. Millerdb8dac22008-03-06 16:22:02 -0800762 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
763 if (uh->check == 0)
764 uh->check = CSUM_MANGLED_0;
765 }
766}
Thomas Grafc26bf4a2013-07-25 18:12:18 +0200767EXPORT_SYMBOL_GPL(udp4_hwcsum);
David S. Millerdb8dac22008-03-06 16:22:02 -0800768
David S. Miller79ab0532011-05-09 13:31:04 -0700769static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
Herbert Xuf6b96642011-03-01 02:36:48 +0000770{
771 struct sock *sk = skb->sk;
772 struct inet_sock *inet = inet_sk(sk);
773 struct udphdr *uh;
Herbert Xuf6b96642011-03-01 02:36:48 +0000774 int err = 0;
775 int is_udplite = IS_UDPLITE(sk);
776 int offset = skb_transport_offset(skb);
777 int len = skb->len - offset;
778 __wsum csum = 0;
779
780 /*
781 * Create a UDP header
782 */
783 uh = udp_hdr(skb);
784 uh->source = inet->inet_sport;
David S. Miller79ab0532011-05-09 13:31:04 -0700785 uh->dest = fl4->fl4_dport;
Herbert Xuf6b96642011-03-01 02:36:48 +0000786 uh->len = htons(len);
787 uh->check = 0;
788
789 if (is_udplite) /* UDP-Lite */
790 csum = udplite_csum(skb);
791
792 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
793
794 skb->ip_summed = CHECKSUM_NONE;
795 goto send;
796
797 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
798
David S. Miller79ab0532011-05-09 13:31:04 -0700799 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
Herbert Xuf6b96642011-03-01 02:36:48 +0000800 goto send;
801
802 } else
803 csum = udp_csum(skb);
804
805 /* add protocol-dependent pseudo-header */
David S. Miller79ab0532011-05-09 13:31:04 -0700806 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
Herbert Xuf6b96642011-03-01 02:36:48 +0000807 sk->sk_protocol, csum);
808 if (uh->check == 0)
809 uh->check = CSUM_MANGLED_0;
810
811send:
Eric Dumazetb5ec8ee2012-08-10 02:22:47 +0000812 err = ip_send_skb(sock_net(sk), skb);
Herbert Xuf6b96642011-03-01 02:36:48 +0000813 if (err) {
814 if (err == -ENOBUFS && !inet->recverr) {
815 UDP_INC_STATS_USER(sock_net(sk),
816 UDP_MIB_SNDBUFERRORS, is_udplite);
817 err = 0;
818 }
819 } else
820 UDP_INC_STATS_USER(sock_net(sk),
821 UDP_MIB_OUTDATAGRAMS, is_udplite);
822 return err;
823}
824
David S. Millerdb8dac22008-03-06 16:22:02 -0800825/*
826 * Push out all pending data as one UDP datagram. Socket is locked.
827 */
Hannes Frederic Sowa8822b642013-07-01 20:21:30 +0200828int udp_push_pending_frames(struct sock *sk)
David S. Millerdb8dac22008-03-06 16:22:02 -0800829{
830 struct udp_sock *up = udp_sk(sk);
831 struct inet_sock *inet = inet_sk(sk);
David S. Millerb6f21b22011-03-12 02:09:18 -0500832 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
David S. Millerdb8dac22008-03-06 16:22:02 -0800833 struct sk_buff *skb;
David S. Millerdb8dac22008-03-06 16:22:02 -0800834 int err = 0;
David S. Millerdb8dac22008-03-06 16:22:02 -0800835
David S. Miller77968b72011-05-08 17:12:19 -0700836 skb = ip_finish_skb(sk, fl4);
Herbert Xuf6b96642011-03-01 02:36:48 +0000837 if (!skb)
David S. Millerdb8dac22008-03-06 16:22:02 -0800838 goto out;
839
David S. Miller79ab0532011-05-09 13:31:04 -0700840 err = udp_send_skb(skb, fl4);
David S. Millerdb8dac22008-03-06 16:22:02 -0800841
David S. Millerdb8dac22008-03-06 16:22:02 -0800842out:
843 up->len = 0;
844 up->pending = 0;
David S. Millerdb8dac22008-03-06 16:22:02 -0800845 return err;
846}
Hannes Frederic Sowa8822b642013-07-01 20:21:30 +0200847EXPORT_SYMBOL(udp_push_pending_frames);
David S. Millerdb8dac22008-03-06 16:22:02 -0800848
849int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
850 size_t len)
851{
852 struct inet_sock *inet = inet_sk(sk);
853 struct udp_sock *up = udp_sk(sk);
David S. Millere4749952011-05-08 16:38:45 -0700854 struct flowi4 fl4_stack;
David S. Millerb6f21b22011-03-12 02:09:18 -0500855 struct flowi4 *fl4;
David S. Millerdb8dac22008-03-06 16:22:02 -0800856 int ulen = len;
857 struct ipcm_cookie ipc;
858 struct rtable *rt = NULL;
859 int free = 0;
860 int connected = 0;
861 __be32 daddr, faddr, saddr;
862 __be16 dport;
863 u8 tos;
864 int err, is_udplite = IS_UDPLITE(sk);
865 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
866 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
Herbert Xu903ab862011-03-01 02:36:48 +0000867 struct sk_buff *skb;
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000868 struct ip_options_data opt_copy;
David S. Millerdb8dac22008-03-06 16:22:02 -0800869
870 if (len > 0xFFFF)
871 return -EMSGSIZE;
872
873 /*
874 * Check the flags.
875 */
876
Eric Dumazetc482c562009-07-17 00:26:32 +0000877 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
David S. Millerdb8dac22008-03-06 16:22:02 -0800878 return -EOPNOTSUPP;
879
880 ipc.opt = NULL;
Oliver Hartkopp2244d072010-08-17 08:59:14 +0000881 ipc.tx_flags = 0;
Francesco Fuscoaa661582013-09-24 15:43:09 +0200882 ipc.ttl = 0;
883 ipc.tos = -1;
David S. Millerdb8dac22008-03-06 16:22:02 -0800884
Herbert Xu903ab862011-03-01 02:36:48 +0000885 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
886
David S. Millerf5fca602011-05-08 17:24:10 -0700887 fl4 = &inet->cork.fl.u.ip4;
David S. Millerdb8dac22008-03-06 16:22:02 -0800888 if (up->pending) {
889 /*
890 * There are pending frames.
891 * The socket lock must be held while it's corked.
892 */
893 lock_sock(sk);
894 if (likely(up->pending)) {
895 if (unlikely(up->pending != AF_INET)) {
896 release_sock(sk);
897 return -EINVAL;
898 }
899 goto do_append_data;
900 }
901 release_sock(sk);
902 }
903 ulen += sizeof(struct udphdr);
904
905 /*
906 * Get and verify the address.
907 */
908 if (msg->msg_name) {
Daniel Baluta5e73ea12012-04-15 01:34:41 +0000909 struct sockaddr_in *usin = (struct sockaddr_in *)msg->msg_name;
David S. Millerdb8dac22008-03-06 16:22:02 -0800910 if (msg->msg_namelen < sizeof(*usin))
911 return -EINVAL;
912 if (usin->sin_family != AF_INET) {
913 if (usin->sin_family != AF_UNSPEC)
914 return -EAFNOSUPPORT;
915 }
916
917 daddr = usin->sin_addr.s_addr;
918 dport = usin->sin_port;
919 if (dport == 0)
920 return -EINVAL;
921 } else {
922 if (sk->sk_state != TCP_ESTABLISHED)
923 return -EDESTADDRREQ;
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000924 daddr = inet->inet_daddr;
925 dport = inet->inet_dport;
David S. Millerdb8dac22008-03-06 16:22:02 -0800926 /* Open fast path for connected socket.
927 Route will not be used, if at least one option is set.
928 */
929 connected = 1;
930 }
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000931 ipc.addr = inet->inet_saddr;
David S. Millerdb8dac22008-03-06 16:22:02 -0800932
933 ipc.oif = sk->sk_bound_dev_if;
Daniel Borkmannbf84a012013-04-14 08:08:13 +0000934
935 sock_tx_timestamp(sk, &ipc.tx_flags);
936
David S. Millerdb8dac22008-03-06 16:22:02 -0800937 if (msg->msg_controllen) {
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900938 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
David S. Millerdb8dac22008-03-06 16:22:02 -0800939 if (err)
940 return err;
941 if (ipc.opt)
942 free = 1;
943 connected = 0;
944 }
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000945 if (!ipc.opt) {
946 struct ip_options_rcu *inet_opt;
947
948 rcu_read_lock();
949 inet_opt = rcu_dereference(inet->inet_opt);
950 if (inet_opt) {
951 memcpy(&opt_copy, inet_opt,
952 sizeof(*inet_opt) + inet_opt->opt.optlen);
953 ipc.opt = &opt_copy.opt;
954 }
955 rcu_read_unlock();
956 }
David S. Millerdb8dac22008-03-06 16:22:02 -0800957
958 saddr = ipc.addr;
959 ipc.addr = faddr = daddr;
960
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000961 if (ipc.opt && ipc.opt->opt.srr) {
David S. Millerdb8dac22008-03-06 16:22:02 -0800962 if (!daddr)
963 return -EINVAL;
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000964 faddr = ipc.opt->opt.faddr;
David S. Millerdb8dac22008-03-06 16:22:02 -0800965 connected = 0;
966 }
Francesco Fuscoaa661582013-09-24 15:43:09 +0200967 tos = get_rttos(&ipc, inet);
David S. Millerdb8dac22008-03-06 16:22:02 -0800968 if (sock_flag(sk, SOCK_LOCALROUTE) ||
969 (msg->msg_flags & MSG_DONTROUTE) ||
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000970 (ipc.opt && ipc.opt->opt.is_strictroute)) {
David S. Millerdb8dac22008-03-06 16:22:02 -0800971 tos |= RTO_ONLINK;
972 connected = 0;
973 }
974
975 if (ipv4_is_multicast(daddr)) {
976 if (!ipc.oif)
977 ipc.oif = inet->mc_index;
978 if (!saddr)
979 saddr = inet->mc_addr;
980 connected = 0;
Erich E. Hoover76e21052012-02-08 09:11:07 +0000981 } else if (!ipc.oif)
982 ipc.oif = inet->uc_index;
David S. Millerdb8dac22008-03-06 16:22:02 -0800983
984 if (connected)
Eric Dumazetc482c562009-07-17 00:26:32 +0000985 rt = (struct rtable *)sk_dst_check(sk, 0);
David S. Millerdb8dac22008-03-06 16:22:02 -0800986
987 if (rt == NULL) {
Pavel Emelyanov84a3aa02008-07-16 20:19:08 -0700988 struct net *net = sock_net(sk);
989
David S. Millere4749952011-05-08 16:38:45 -0700990 fl4 = &fl4_stack;
991 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
David S. Millerc0951cb2011-03-31 04:54:27 -0700992 RT_SCOPE_UNIVERSE, sk->sk_protocol,
993 inet_sk_flowi_flags(sk)|FLOWI_FLAG_CAN_SLEEP,
994 faddr, saddr, dport, inet->inet_sport);
995
David S. Millere4749952011-05-08 16:38:45 -0700996 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
997 rt = ip_route_output_flow(net, fl4, sk);
David S. Millerb23dd4f2011-03-02 14:31:35 -0800998 if (IS_ERR(rt)) {
999 err = PTR_ERR(rt);
David S. Miller06dc94b2011-03-03 10:38:01 -08001000 rt = NULL;
David S. Millerdb8dac22008-03-06 16:22:02 -08001001 if (err == -ENETUNREACH)
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -07001002 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
David S. Millerdb8dac22008-03-06 16:22:02 -08001003 goto out;
1004 }
1005
1006 err = -EACCES;
1007 if ((rt->rt_flags & RTCF_BROADCAST) &&
1008 !sock_flag(sk, SOCK_BROADCAST))
1009 goto out;
1010 if (connected)
Changli Gaod8d1f302010-06-10 23:31:35 -07001011 sk_dst_set(sk, dst_clone(&rt->dst));
David S. Millerdb8dac22008-03-06 16:22:02 -08001012 }
1013
1014 if (msg->msg_flags&MSG_CONFIRM)
1015 goto do_confirm;
1016back_from_confirm:
1017
David S. Millere4749952011-05-08 16:38:45 -07001018 saddr = fl4->saddr;
David S. Millerdb8dac22008-03-06 16:22:02 -08001019 if (!ipc.addr)
David S. Millere4749952011-05-08 16:38:45 -07001020 daddr = ipc.addr = fl4->daddr;
David S. Millerdb8dac22008-03-06 16:22:02 -08001021
Herbert Xu903ab862011-03-01 02:36:48 +00001022 /* Lockless fast path for the non-corking case. */
1023 if (!corkreq) {
David S. Miller77968b72011-05-08 17:12:19 -07001024 skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen,
Herbert Xu903ab862011-03-01 02:36:48 +00001025 sizeof(struct udphdr), &ipc, &rt,
1026 msg->msg_flags);
1027 err = PTR_ERR(skb);
YOSHIFUJI Hideaki / 吉藤英明50c3a482013-01-22 06:32:49 +00001028 if (!IS_ERR_OR_NULL(skb))
David S. Miller79ab0532011-05-09 13:31:04 -07001029 err = udp_send_skb(skb, fl4);
Herbert Xu903ab862011-03-01 02:36:48 +00001030 goto out;
1031 }
1032
David S. Millerdb8dac22008-03-06 16:22:02 -08001033 lock_sock(sk);
1034 if (unlikely(up->pending)) {
1035 /* The socket is already corked while preparing it. */
1036 /* ... which is an evident application bug. --ANK */
1037 release_sock(sk);
1038
Joe Perchesafd465032012-03-12 07:03:32 +00001039 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("cork app bug 2\n"));
David S. Millerdb8dac22008-03-06 16:22:02 -08001040 err = -EINVAL;
1041 goto out;
1042 }
1043 /*
1044 * Now cork the socket to pend data.
1045 */
David S. Millerb6f21b22011-03-12 02:09:18 -05001046 fl4 = &inet->cork.fl.u.ip4;
1047 fl4->daddr = daddr;
1048 fl4->saddr = saddr;
David S. Miller9cce96d2011-03-12 03:00:33 -05001049 fl4->fl4_dport = dport;
1050 fl4->fl4_sport = inet->inet_sport;
David S. Millerdb8dac22008-03-06 16:22:02 -08001051 up->pending = AF_INET;
1052
1053do_append_data:
1054 up->len += ulen;
David S. Millerf5fca602011-05-08 17:24:10 -07001055 err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen,
1056 sizeof(struct udphdr), &ipc, &rt,
1057 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
David S. Millerdb8dac22008-03-06 16:22:02 -08001058 if (err)
1059 udp_flush_pending_frames(sk);
1060 else if (!corkreq)
1061 err = udp_push_pending_frames(sk);
1062 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1063 up->pending = 0;
1064 release_sock(sk);
1065
1066out:
1067 ip_rt_put(rt);
1068 if (free)
1069 kfree(ipc.opt);
1070 if (!err)
1071 return len;
1072 /*
1073 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1074 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1075 * we don't have a good statistic (IpOutDiscards but it can be too many
1076 * things). We could add another new stat but at least for now that
1077 * seems like overkill.
1078 */
1079 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
Pavel Emelyanov629ca232008-07-05 21:18:07 -07001080 UDP_INC_STATS_USER(sock_net(sk),
1081 UDP_MIB_SNDBUFERRORS, is_udplite);
David S. Millerdb8dac22008-03-06 16:22:02 -08001082 }
1083 return err;
1084
1085do_confirm:
Changli Gaod8d1f302010-06-10 23:31:35 -07001086 dst_confirm(&rt->dst);
David S. Millerdb8dac22008-03-06 16:22:02 -08001087 if (!(msg->msg_flags&MSG_PROBE) || len)
1088 goto back_from_confirm;
1089 err = 0;
1090 goto out;
1091}
Eric Dumazetc482c562009-07-17 00:26:32 +00001092EXPORT_SYMBOL(udp_sendmsg);
David S. Millerdb8dac22008-03-06 16:22:02 -08001093
1094int udp_sendpage(struct sock *sk, struct page *page, int offset,
1095 size_t size, int flags)
1096{
David S. Millerf5fca602011-05-08 17:24:10 -07001097 struct inet_sock *inet = inet_sk(sk);
David S. Millerdb8dac22008-03-06 16:22:02 -08001098 struct udp_sock *up = udp_sk(sk);
1099 int ret;
1100
Shawn Landdend3f7d562013-11-24 22:36:28 -08001101 if (flags & MSG_SENDPAGE_NOTLAST)
1102 flags |= MSG_MORE;
1103
David S. Millerdb8dac22008-03-06 16:22:02 -08001104 if (!up->pending) {
1105 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1106
1107 /* Call udp_sendmsg to specify destination address which
1108 * sendpage interface can't pass.
1109 * This will succeed only when the socket is connected.
1110 */
1111 ret = udp_sendmsg(NULL, sk, &msg, 0);
1112 if (ret < 0)
1113 return ret;
1114 }
1115
1116 lock_sock(sk);
1117
1118 if (unlikely(!up->pending)) {
1119 release_sock(sk);
1120
Joe Perchesafd465032012-03-12 07:03:32 +00001121 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("udp cork app bug 3\n"));
David S. Millerdb8dac22008-03-06 16:22:02 -08001122 return -EINVAL;
1123 }
1124
David S. Millerf5fca602011-05-08 17:24:10 -07001125 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1126 page, offset, size, flags);
David S. Millerdb8dac22008-03-06 16:22:02 -08001127 if (ret == -EOPNOTSUPP) {
1128 release_sock(sk);
1129 return sock_no_sendpage(sk->sk_socket, page, offset,
1130 size, flags);
1131 }
1132 if (ret < 0) {
1133 udp_flush_pending_frames(sk);
1134 goto out;
1135 }
1136
1137 up->len += size;
1138 if (!(up->corkflag || (flags&MSG_MORE)))
1139 ret = udp_push_pending_frames(sk);
1140 if (!ret)
1141 ret = size;
1142out:
1143 release_sock(sk);
1144 return ret;
1145}
1146
Eric Dumazet85584672009-10-09 04:43:40 +00001147
1148/**
1149 * first_packet_length - return length of first packet in receive queue
1150 * @sk: socket
1151 *
1152 * Drops all bad checksum frames, until a valid one is found.
1153 * Returns the length of found skb, or 0 if none is found.
1154 */
1155static unsigned int first_packet_length(struct sock *sk)
1156{
1157 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1158 struct sk_buff *skb;
1159 unsigned int res;
1160
1161 __skb_queue_head_init(&list_kill);
1162
1163 spin_lock_bh(&rcvq->lock);
1164 while ((skb = skb_peek(rcvq)) != NULL &&
1165 udp_lib_checksum_complete(skb)) {
Eric Dumazet6a5dc9e2013-04-29 08:39:56 +00001166 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS,
1167 IS_UDPLITE(sk));
Eric Dumazet85584672009-10-09 04:43:40 +00001168 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1169 IS_UDPLITE(sk));
Eric Dumazet8edf19c2009-10-15 00:12:40 +00001170 atomic_inc(&sk->sk_drops);
Eric Dumazet85584672009-10-09 04:43:40 +00001171 __skb_unlink(skb, rcvq);
1172 __skb_queue_tail(&list_kill, skb);
1173 }
1174 res = skb ? skb->len : 0;
1175 spin_unlock_bh(&rcvq->lock);
1176
1177 if (!skb_queue_empty(&list_kill)) {
Eric Dumazet8a74ad62010-05-26 19:20:18 +00001178 bool slow = lock_sock_fast(sk);
1179
Eric Dumazet85584672009-10-09 04:43:40 +00001180 __skb_queue_purge(&list_kill);
1181 sk_mem_reclaim_partial(sk);
Eric Dumazet8a74ad62010-05-26 19:20:18 +00001182 unlock_sock_fast(sk, slow);
Eric Dumazet85584672009-10-09 04:43:40 +00001183 }
1184 return res;
1185}
1186
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187/*
1188 * IOCTL requests applicable to the UDP protocol
1189 */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001190
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1192{
Stephen Hemminger6516c652007-03-08 20:41:55 -08001193 switch (cmd) {
1194 case SIOCOUTQ:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 {
Eric Dumazet31e6d362009-06-17 19:05:41 -07001196 int amount = sk_wmem_alloc_get(sk);
1197
Stephen Hemminger6516c652007-03-08 20:41:55 -08001198 return put_user(amount, (int __user *)arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 }
Stephen Hemminger6516c652007-03-08 20:41:55 -08001200
1201 case SIOCINQ:
1202 {
Eric Dumazet85584672009-10-09 04:43:40 +00001203 unsigned int amount = first_packet_length(sk);
Stephen Hemminger6516c652007-03-08 20:41:55 -08001204
Eric Dumazet85584672009-10-09 04:43:40 +00001205 if (amount)
Stephen Hemminger6516c652007-03-08 20:41:55 -08001206 /*
1207 * We will only return the amount
1208 * of this packet since that is all
1209 * that will be read.
1210 */
Eric Dumazet85584672009-10-09 04:43:40 +00001211 amount -= sizeof(struct udphdr);
1212
Stephen Hemminger6516c652007-03-08 20:41:55 -08001213 return put_user(amount, (int __user *)arg);
1214 }
1215
1216 default:
1217 return -ENOIOCTLCMD;
1218 }
1219
1220 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221}
Eric Dumazetc482c562009-07-17 00:26:32 +00001222EXPORT_SYMBOL(udp_ioctl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223
David S. Millerdb8dac22008-03-06 16:22:02 -08001224/*
1225 * This should be easy, if there is something there we
1226 * return it, otherwise we block.
1227 */
1228
1229int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1230 size_t len, int noblock, int flags, int *addr_len)
1231{
1232 struct inet_sock *inet = inet_sk(sk);
1233 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
1234 struct sk_buff *skb;
David S. Miller59c2cda2011-12-01 14:12:55 -05001235 unsigned int ulen, copied;
Pavel Emelyanov3f518bf2012-02-21 07:30:58 +00001236 int peeked, off = 0;
David S. Millerdb8dac22008-03-06 16:22:02 -08001237 int err;
1238 int is_udplite = IS_UDPLITE(sk);
Eric Dumazet8a74ad62010-05-26 19:20:18 +00001239 bool slow;
David S. Millerdb8dac22008-03-06 16:22:02 -08001240
David S. Millerdb8dac22008-03-06 16:22:02 -08001241 if (flags & MSG_ERRQUEUE)
Hannes Frederic Sowa85fbaa72013-11-23 00:46:12 +01001242 return ip_recv_error(sk, msg, len, addr_len);
David S. Millerdb8dac22008-03-06 16:22:02 -08001243
1244try_again:
1245 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
Pavel Emelyanov3f518bf2012-02-21 07:30:58 +00001246 &peeked, &off, &err);
David S. Millerdb8dac22008-03-06 16:22:02 -08001247 if (!skb)
1248 goto out;
1249
1250 ulen = skb->len - sizeof(struct udphdr);
David S. Miller59c2cda2011-12-01 14:12:55 -05001251 copied = len;
1252 if (copied > ulen)
1253 copied = ulen;
1254 else if (copied < ulen)
David S. Millerdb8dac22008-03-06 16:22:02 -08001255 msg->msg_flags |= MSG_TRUNC;
1256
1257 /*
1258 * If checksum is needed at all, try to do it while copying the
1259 * data. If the data is truncated, or if we only want a partial
1260 * coverage checksum (UDP-Lite), do it before the copy.
1261 */
1262
David S. Miller59c2cda2011-12-01 14:12:55 -05001263 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
David S. Millerdb8dac22008-03-06 16:22:02 -08001264 if (udp_lib_checksum_complete(skb))
1265 goto csum_copy_err;
1266 }
1267
1268 if (skb_csum_unnecessary(skb))
1269 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
David S. Miller59c2cda2011-12-01 14:12:55 -05001270 msg->msg_iov, copied);
David S. Millerdb8dac22008-03-06 16:22:02 -08001271 else {
Eric Dumazetc482c562009-07-17 00:26:32 +00001272 err = skb_copy_and_csum_datagram_iovec(skb,
1273 sizeof(struct udphdr),
1274 msg->msg_iov);
David S. Millerdb8dac22008-03-06 16:22:02 -08001275
1276 if (err == -EINVAL)
1277 goto csum_copy_err;
1278 }
1279
Eric Dumazet22911fc2012-06-27 00:23:44 +00001280 if (unlikely(err)) {
1281 trace_kfree_skb(skb, udp_recvmsg);
Eric Dumazet979402b2012-09-05 23:34:44 +00001282 if (!peeked) {
1283 atomic_inc(&sk->sk_drops);
1284 UDP_INC_STATS_USER(sock_net(sk),
1285 UDP_MIB_INERRORS, is_udplite);
1286 }
David S. Millerdb8dac22008-03-06 16:22:02 -08001287 goto out_free;
Eric Dumazet22911fc2012-06-27 00:23:44 +00001288 }
David S. Millerdb8dac22008-03-06 16:22:02 -08001289
1290 if (!peeked)
Pavel Emelyanov629ca232008-07-05 21:18:07 -07001291 UDP_INC_STATS_USER(sock_net(sk),
1292 UDP_MIB_INDATAGRAMS, is_udplite);
David S. Millerdb8dac22008-03-06 16:22:02 -08001293
Neil Horman3b885782009-10-12 13:26:31 -07001294 sock_recv_ts_and_drops(msg, sk, skb);
David S. Millerdb8dac22008-03-06 16:22:02 -08001295
1296 /* Copy the address. */
Eric Dumazetc482c562009-07-17 00:26:32 +00001297 if (sin) {
David S. Millerdb8dac22008-03-06 16:22:02 -08001298 sin->sin_family = AF_INET;
1299 sin->sin_port = udp_hdr(skb)->source;
1300 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1301 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
Hannes Frederic Sowabceaa902013-11-18 04:20:45 +01001302 *addr_len = sizeof(*sin);
David S. Millerdb8dac22008-03-06 16:22:02 -08001303 }
1304 if (inet->cmsg_flags)
1305 ip_cmsg_recv(msg, skb);
1306
David S. Miller59c2cda2011-12-01 14:12:55 -05001307 err = copied;
David S. Millerdb8dac22008-03-06 16:22:02 -08001308 if (flags & MSG_TRUNC)
1309 err = ulen;
1310
1311out_free:
Eric Dumazet9d410c72009-10-30 05:03:53 +00001312 skb_free_datagram_locked(sk, skb);
David S. Millerdb8dac22008-03-06 16:22:02 -08001313out:
1314 return err;
1315
1316csum_copy_err:
Eric Dumazet8a74ad62010-05-26 19:20:18 +00001317 slow = lock_sock_fast(sk);
Eric Dumazet6a5dc9e2013-04-29 08:39:56 +00001318 if (!skb_kill_datagram(sk, skb, flags)) {
1319 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
Pavel Emelyanov629ca232008-07-05 21:18:07 -07001320 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
Eric Dumazet6a5dc9e2013-04-29 08:39:56 +00001321 }
Eric Dumazet8a74ad62010-05-26 19:20:18 +00001322 unlock_sock_fast(sk, slow);
David S. Millerdb8dac22008-03-06 16:22:02 -08001323
1324 if (noblock)
1325 return -EAGAIN;
Xufeng Zhang9cfaa8d2011-06-21 10:43:40 +00001326
1327 /* starting over for a new packet */
1328 msg->msg_flags &= ~MSG_TRUNC;
David S. Millerdb8dac22008-03-06 16:22:02 -08001329 goto try_again;
1330}
1331
1332
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333int udp_disconnect(struct sock *sk, int flags)
1334{
1335 struct inet_sock *inet = inet_sk(sk);
1336 /*
1337 * 1003.1g - break association.
1338 */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001339
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 sk->sk_state = TCP_CLOSE;
Eric Dumazetc720c7e2009-10-15 06:30:45 +00001341 inet->inet_daddr = 0;
1342 inet->inet_dport = 0;
Tom Herbertbdeab992011-08-14 19:45:55 +00001343 sock_rps_reset_rxhash(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 sk->sk_bound_dev_if = 0;
1345 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1346 inet_reset_saddr(sk);
1347
1348 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1349 sk->sk_prot->unhash(sk);
Eric Dumazetc720c7e2009-10-15 06:30:45 +00001350 inet->inet_sport = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 }
1352 sk_dst_reset(sk);
1353 return 0;
1354}
Eric Dumazetc482c562009-07-17 00:26:32 +00001355EXPORT_SYMBOL(udp_disconnect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356
Eric Dumazet645ca702008-10-29 01:41:45 -07001357void udp_lib_unhash(struct sock *sk)
1358{
Eric Dumazet723b4612008-11-25 13:55:15 -08001359 if (sk_hashed(sk)) {
1360 struct udp_table *udptable = sk->sk_prot->h.udp_table;
Eric Dumazet512615b2009-11-08 10:17:58 +00001361 struct udp_hslot *hslot, *hslot2;
1362
1363 hslot = udp_hashslot(udptable, sock_net(sk),
1364 udp_sk(sk)->udp_port_hash);
1365 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
Eric Dumazet645ca702008-10-29 01:41:45 -07001366
Eric Dumazet723b4612008-11-25 13:55:15 -08001367 spin_lock_bh(&hslot->lock);
1368 if (sk_nulls_del_node_init_rcu(sk)) {
Eric Dumazetfdcc8aa92009-11-08 10:17:05 +00001369 hslot->count--;
Eric Dumazetc720c7e2009-10-15 06:30:45 +00001370 inet_sk(sk)->inet_num = 0;
Eric Dumazet723b4612008-11-25 13:55:15 -08001371 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
Eric Dumazet512615b2009-11-08 10:17:58 +00001372
1373 spin_lock(&hslot2->lock);
1374 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1375 hslot2->count--;
1376 spin_unlock(&hslot2->lock);
Eric Dumazet723b4612008-11-25 13:55:15 -08001377 }
1378 spin_unlock_bh(&hslot->lock);
Eric Dumazet645ca702008-10-29 01:41:45 -07001379 }
Eric Dumazet645ca702008-10-29 01:41:45 -07001380}
1381EXPORT_SYMBOL(udp_lib_unhash);
1382
Eric Dumazet719f8352010-09-08 05:08:44 +00001383/*
1384 * inet_rcv_saddr was changed, we must rehash secondary hash
1385 */
1386void udp_lib_rehash(struct sock *sk, u16 newhash)
1387{
1388 if (sk_hashed(sk)) {
1389 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1390 struct udp_hslot *hslot, *hslot2, *nhslot2;
1391
1392 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1393 nhslot2 = udp_hashslot2(udptable, newhash);
1394 udp_sk(sk)->udp_portaddr_hash = newhash;
1395 if (hslot2 != nhslot2) {
1396 hslot = udp_hashslot(udptable, sock_net(sk),
1397 udp_sk(sk)->udp_port_hash);
1398 /* we must lock primary chain too */
1399 spin_lock_bh(&hslot->lock);
1400
1401 spin_lock(&hslot2->lock);
1402 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1403 hslot2->count--;
1404 spin_unlock(&hslot2->lock);
1405
1406 spin_lock(&nhslot2->lock);
1407 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1408 &nhslot2->head);
1409 nhslot2->count++;
1410 spin_unlock(&nhslot2->lock);
1411
1412 spin_unlock_bh(&hslot->lock);
1413 }
1414 }
1415}
1416EXPORT_SYMBOL(udp_lib_rehash);
1417
1418static void udp_v4_rehash(struct sock *sk)
1419{
1420 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1421 inet_sk(sk)->inet_rcv_saddr,
1422 inet_sk(sk)->inet_num);
1423 udp_lib_rehash(sk, new_hash);
1424}
1425
Herbert Xu93821772008-09-15 11:48:46 -07001426static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1427{
Tom Herbertfec5e652010-04-16 16:01:27 -07001428 int rc;
Herbert Xu93821772008-09-15 11:48:46 -07001429
Shawn Bohrer005ec972013-10-07 11:01:38 -05001430 if (inet_sk(sk)->inet_daddr) {
Tom Herbertbdeab992011-08-14 19:45:55 +00001431 sock_rps_save_rxhash(sk, skb);
Shawn Bohrer005ec972013-10-07 11:01:38 -05001432 sk_mark_napi_id(sk, skb);
1433 }
Tom Herbertfec5e652010-04-16 16:01:27 -07001434
Eric Dumazetd826eb12011-11-09 07:24:35 +00001435 rc = sock_queue_rcv_skb(sk, skb);
Eric Dumazet766e90372009-10-14 20:40:11 -07001436 if (rc < 0) {
1437 int is_udplite = IS_UDPLITE(sk);
1438
Herbert Xu93821772008-09-15 11:48:46 -07001439 /* Note that an ENOMEM error is charged twice */
Eric Dumazet766e90372009-10-14 20:40:11 -07001440 if (rc == -ENOMEM)
Herbert Xu93821772008-09-15 11:48:46 -07001441 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1442 is_udplite);
Eric Dumazet766e90372009-10-14 20:40:11 -07001443 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1444 kfree_skb(skb);
Satoru Moriya296f7ea2011-06-17 11:58:39 +00001445 trace_udp_fail_queue_rcv_skb(rc, sk);
Eric Dumazet766e90372009-10-14 20:40:11 -07001446 return -1;
Herbert Xu93821772008-09-15 11:48:46 -07001447 }
1448
1449 return 0;
1450
Herbert Xu93821772008-09-15 11:48:46 -07001451}
1452
Eric Dumazet447167b2012-04-11 23:05:28 +00001453static struct static_key udp_encap_needed __read_mostly;
1454void udp_encap_enable(void)
1455{
1456 if (!static_key_enabled(&udp_encap_needed))
1457 static_key_slow_inc(&udp_encap_needed);
1458}
1459EXPORT_SYMBOL(udp_encap_enable);
1460
David S. Millerdb8dac22008-03-06 16:22:02 -08001461/* returns:
1462 * -1: error
1463 * 0: success
1464 * >0: "udp encap" protocol resubmission
1465 *
1466 * Note that in the success and error cases, the skb is assumed to
1467 * have either been requeued or freed.
1468 */
Eric Dumazetc482c562009-07-17 00:26:32 +00001469int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
David S. Millerdb8dac22008-03-06 16:22:02 -08001470{
1471 struct udp_sock *up = udp_sk(sk);
1472 int rc;
1473 int is_udplite = IS_UDPLITE(sk);
1474
1475 /*
1476 * Charge it to the socket, dropping if the queue is full.
1477 */
1478 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1479 goto drop;
1480 nf_reset(skb);
1481
Eric Dumazet447167b2012-04-11 23:05:28 +00001482 if (static_key_false(&udp_encap_needed) && up->encap_type) {
Eric Dumazet0ad92ad2011-11-01 12:56:59 +00001483 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1484
David S. Millerdb8dac22008-03-06 16:22:02 -08001485 /*
1486 * This is an encapsulation socket so pass the skb to
1487 * the socket's udp_encap_rcv() hook. Otherwise, just
1488 * fall through and pass this up the UDP socket.
1489 * up->encap_rcv() returns the following value:
1490 * =0 if skb was successfully passed to the encap
1491 * handler or was discarded by it.
1492 * >0 if skb should be passed on to UDP.
1493 * <0 if skb should be resubmitted as proto -N
1494 */
1495
1496 /* if we're overly short, let UDP handle it */
Eric Dumazet0ad92ad2011-11-01 12:56:59 +00001497 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1498 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
David S. Millerdb8dac22008-03-06 16:22:02 -08001499 int ret;
1500
Eric Dumazet0ad92ad2011-11-01 12:56:59 +00001501 ret = encap_rcv(sk, skb);
David S. Millerdb8dac22008-03-06 16:22:02 -08001502 if (ret <= 0) {
Pavel Emelyanov02833282008-07-05 21:18:48 -07001503 UDP_INC_STATS_BH(sock_net(sk),
1504 UDP_MIB_INDATAGRAMS,
David S. Millerdb8dac22008-03-06 16:22:02 -08001505 is_udplite);
1506 return -ret;
1507 }
1508 }
1509
1510 /* FALLTHROUGH -- it's a UDP Packet */
1511 }
1512
1513 /*
1514 * UDP-Lite specific tests, ignored on UDP sockets
1515 */
1516 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1517
1518 /*
1519 * MIB statistics other than incrementing the error count are
1520 * disabled for the following two types of errors: these depend
1521 * on the application settings, not on the functioning of the
1522 * protocol stack as such.
1523 *
1524 * RFC 3828 here recommends (sec 3.3): "There should also be a
1525 * way ... to ... at least let the receiving application block
1526 * delivery of packets with coverage values less than a value
1527 * provided by the application."
1528 */
1529 if (up->pcrlen == 0) { /* full coverage was set */
Joe Perchesafd465032012-03-12 07:03:32 +00001530 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: partial coverage %d while full coverage %d requested\n",
1531 UDP_SKB_CB(skb)->cscov, skb->len);
David S. Millerdb8dac22008-03-06 16:22:02 -08001532 goto drop;
1533 }
1534 /* The next case involves violating the min. coverage requested
1535 * by the receiver. This is subtle: if receiver wants x and x is
1536 * greater than the buffersize/MTU then receiver will complain
1537 * that it wants x while sender emits packets of smaller size y.
1538 * Therefore the above ...()->partial_cov statement is essential.
1539 */
1540 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
Joe Perchesafd465032012-03-12 07:03:32 +00001541 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: coverage %d too small, need min %d\n",
1542 UDP_SKB_CB(skb)->cscov, up->pcrlen);
David S. Millerdb8dac22008-03-06 16:22:02 -08001543 goto drop;
1544 }
1545 }
1546
Eric Dumazet33d480c2011-08-11 19:30:52 +00001547 if (rcu_access_pointer(sk->sk_filter) &&
1548 udp_lib_checksum_complete(skb))
Eric Dumazet6a5dc9e2013-04-29 08:39:56 +00001549 goto csum_error;
David S. Millerdb8dac22008-03-06 16:22:02 -08001550
Eric Dumazetc3774112010-04-27 15:13:20 -07001551
Eric Dumazetf545a382012-04-22 23:34:26 +00001552 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf))
Eric Dumazetc3774112010-04-27 15:13:20 -07001553 goto drop;
1554
Herbert Xu93821772008-09-15 11:48:46 -07001555 rc = 0;
David S. Millerdb8dac22008-03-06 16:22:02 -08001556
Shawn Bohrerfbf88662013-10-07 11:01:40 -05001557 ipv4_pktinfo_prepare(sk, skb);
Herbert Xu93821772008-09-15 11:48:46 -07001558 bh_lock_sock(sk);
1559 if (!sock_owned_by_user(sk))
1560 rc = __udp_queue_rcv_skb(sk, skb);
Eric Dumazetf545a382012-04-22 23:34:26 +00001561 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
Zhu Yi55349792010-03-04 18:01:42 +00001562 bh_unlock_sock(sk);
1563 goto drop;
1564 }
Herbert Xu93821772008-09-15 11:48:46 -07001565 bh_unlock_sock(sk);
1566
1567 return rc;
David S. Millerdb8dac22008-03-06 16:22:02 -08001568
Eric Dumazet6a5dc9e2013-04-29 08:39:56 +00001569csum_error:
1570 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
David S. Millerdb8dac22008-03-06 16:22:02 -08001571drop:
Pavel Emelyanov02833282008-07-05 21:18:48 -07001572 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
Eric Dumazet8edf19c2009-10-15 00:12:40 +00001573 atomic_inc(&sk->sk_drops);
David S. Millerdb8dac22008-03-06 16:22:02 -08001574 kfree_skb(skb);
1575 return -1;
1576}
1577
Eric Dumazet1240d132009-11-08 10:18:44 +00001578
1579static void flush_stack(struct sock **stack, unsigned int count,
1580 struct sk_buff *skb, unsigned int final)
1581{
1582 unsigned int i;
1583 struct sk_buff *skb1 = NULL;
Eric Dumazetf6b8f322009-11-08 10:20:19 +00001584 struct sock *sk;
Eric Dumazet1240d132009-11-08 10:18:44 +00001585
1586 for (i = 0; i < count; i++) {
Eric Dumazetf6b8f322009-11-08 10:20:19 +00001587 sk = stack[i];
Eric Dumazet1240d132009-11-08 10:18:44 +00001588 if (likely(skb1 == NULL))
1589 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1590
Eric Dumazetf6b8f322009-11-08 10:20:19 +00001591 if (!skb1) {
1592 atomic_inc(&sk->sk_drops);
1593 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1594 IS_UDPLITE(sk));
1595 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1596 IS_UDPLITE(sk));
1597 }
1598
1599 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
Eric Dumazet1240d132009-11-08 10:18:44 +00001600 skb1 = NULL;
1601 }
1602 if (unlikely(skb1))
1603 kfree_skb(skb1);
1604}
1605
Shawn Bohrer421b3882013-10-07 11:01:39 -05001606static void udp_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
1607{
1608 struct dst_entry *dst = skb_dst(skb);
1609
1610 dst_hold(dst);
1611 sk->sk_rx_dst = dst;
1612}
1613
David S. Millerdb8dac22008-03-06 16:22:02 -08001614/*
1615 * Multicasts and broadcasts go to each listener.
1616 *
Eric Dumazet1240d132009-11-08 10:18:44 +00001617 * Note: called only from the BH handler context.
David S. Millerdb8dac22008-03-06 16:22:02 -08001618 */
Pavel Emelyanove3163492008-06-16 17:12:11 -07001619static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
David S. Millerdb8dac22008-03-06 16:22:02 -08001620 struct udphdr *uh,
1621 __be32 saddr, __be32 daddr,
Eric Dumazet645ca702008-10-29 01:41:45 -07001622 struct udp_table *udptable)
David S. Millerdb8dac22008-03-06 16:22:02 -08001623{
Eric Dumazet1240d132009-11-08 10:18:44 +00001624 struct sock *sk, *stack[256 / sizeof(struct sock *)];
Eric Dumazetf86dcc52009-10-07 00:37:59 +00001625 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
David S. Millerdb8dac22008-03-06 16:22:02 -08001626 int dif;
Eric Dumazet1240d132009-11-08 10:18:44 +00001627 unsigned int i, count = 0;
David S. Millerdb8dac22008-03-06 16:22:02 -08001628
Eric Dumazet645ca702008-10-29 01:41:45 -07001629 spin_lock(&hslot->lock);
Eric Dumazet88ab1932008-11-16 19:39:21 -08001630 sk = sk_nulls_head(&hslot->head);
David S. Millerdb8dac22008-03-06 16:22:02 -08001631 dif = skb->dev->ifindex;
Eric Dumazet920a4612008-11-01 21:22:23 -07001632 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
Eric Dumazet1240d132009-11-08 10:18:44 +00001633 while (sk) {
1634 stack[count++] = sk;
1635 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1636 daddr, uh->source, saddr, dif);
1637 if (unlikely(count == ARRAY_SIZE(stack))) {
1638 if (!sk)
1639 break;
1640 flush_stack(stack, count, skb, ~0);
1641 count = 0;
1642 }
1643 }
1644 /*
1645 * before releasing chain lock, we must take a reference on sockets
1646 */
1647 for (i = 0; i < count; i++)
1648 sock_hold(stack[i]);
David S. Millerdb8dac22008-03-06 16:22:02 -08001649
Eric Dumazet645ca702008-10-29 01:41:45 -07001650 spin_unlock(&hslot->lock);
Eric Dumazet1240d132009-11-08 10:18:44 +00001651
1652 /*
1653 * do the slow work with no lock held
1654 */
1655 if (count) {
1656 flush_stack(stack, count, skb, count - 1);
1657
1658 for (i = 0; i < count; i++)
1659 sock_put(stack[i]);
1660 } else {
1661 kfree_skb(skb);
1662 }
David S. Millerdb8dac22008-03-06 16:22:02 -08001663 return 0;
1664}
1665
1666/* Initialize UDP checksum. If exited with zero value (success),
1667 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1668 * Otherwise, csum completion requires chacksumming packet body,
1669 * including udp header and folding it to skb->csum.
1670 */
1671static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1672 int proto)
1673{
1674 const struct iphdr *iph;
1675 int err;
1676
1677 UDP_SKB_CB(skb)->partial_cov = 0;
1678 UDP_SKB_CB(skb)->cscov = skb->len;
1679
1680 if (proto == IPPROTO_UDPLITE) {
1681 err = udplite_checksum_init(skb, uh);
1682 if (err)
1683 return err;
1684 }
1685
1686 iph = ip_hdr(skb);
1687 if (uh->check == 0) {
1688 skb->ip_summed = CHECKSUM_UNNECESSARY;
1689 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
Eric Dumazetc482c562009-07-17 00:26:32 +00001690 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
David S. Millerdb8dac22008-03-06 16:22:02 -08001691 proto, skb->csum))
1692 skb->ip_summed = CHECKSUM_UNNECESSARY;
1693 }
1694 if (!skb_csum_unnecessary(skb))
1695 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1696 skb->len, proto, 0);
1697 /* Probably, we should checksum udp header (it should be in cache
1698 * in any case) and data in tiny packets (< rx copybreak).
1699 */
1700
1701 return 0;
1702}
1703
1704/*
1705 * All we need to do is get the socket, and then do a checksum.
1706 */
1707
Eric Dumazet645ca702008-10-29 01:41:45 -07001708int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
David S. Millerdb8dac22008-03-06 16:22:02 -08001709 int proto)
1710{
1711 struct sock *sk;
Jesper Dangaard Brouer7b5e56f2009-02-05 15:05:45 -08001712 struct udphdr *uh;
David S. Millerdb8dac22008-03-06 16:22:02 -08001713 unsigned short ulen;
Eric Dumazetadf30902009-06-02 05:19:30 +00001714 struct rtable *rt = skb_rtable(skb);
Jesper Dangaard Brouer2783ef22009-02-06 01:59:12 -08001715 __be32 saddr, daddr;
Pavel Emelyanov02833282008-07-05 21:18:48 -07001716 struct net *net = dev_net(skb->dev);
David S. Millerdb8dac22008-03-06 16:22:02 -08001717
1718 /*
1719 * Validate the packet.
1720 */
1721 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1722 goto drop; /* No space for header. */
1723
Jesper Dangaard Brouer7b5e56f2009-02-05 15:05:45 -08001724 uh = udp_hdr(skb);
David S. Millerdb8dac22008-03-06 16:22:02 -08001725 ulen = ntohs(uh->len);
Bjørn Morkccc2d972010-05-06 03:44:34 +00001726 saddr = ip_hdr(skb)->saddr;
1727 daddr = ip_hdr(skb)->daddr;
1728
David S. Millerdb8dac22008-03-06 16:22:02 -08001729 if (ulen > skb->len)
1730 goto short_packet;
1731
1732 if (proto == IPPROTO_UDP) {
1733 /* UDP validates ulen. */
1734 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1735 goto short_packet;
1736 uh = udp_hdr(skb);
1737 }
1738
1739 if (udp4_csum_init(skb, uh, proto))
1740 goto csum_error;
1741
Shawn Bohrer421b3882013-10-07 11:01:39 -05001742 if (skb->sk) {
1743 int ret;
1744 sk = skb->sk;
David S. Millerdb8dac22008-03-06 16:22:02 -08001745
Shawn Bohrer421b3882013-10-07 11:01:39 -05001746 if (unlikely(sk->sk_rx_dst == NULL))
1747 udp_sk_rx_dst_set(sk, skb);
1748
1749 ret = udp_queue_rcv_skb(sk, skb);
1750
1751 /* a return value > 0 means to resubmit the input, but
1752 * it wants the return to be -protocol, or 0
1753 */
1754 if (ret > 0)
1755 return -ret;
1756 return 0;
1757 } else {
1758 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1759 return __udp4_lib_mcast_deliver(net, skb, uh,
1760 saddr, daddr, udptable);
1761
1762 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1763 }
David S. Millerdb8dac22008-03-06 16:22:02 -08001764
1765 if (sk != NULL) {
Eliezer Tamira5b50472013-06-10 11:40:00 +03001766 int ret;
1767
Eliezer Tamira5b50472013-06-10 11:40:00 +03001768 ret = udp_queue_rcv_skb(sk, skb);
David S. Millerdb8dac22008-03-06 16:22:02 -08001769 sock_put(sk);
1770
1771 /* a return value > 0 means to resubmit the input, but
1772 * it wants the return to be -protocol, or 0
1773 */
1774 if (ret > 0)
1775 return -ret;
1776 return 0;
1777 }
1778
1779 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1780 goto drop;
1781 nf_reset(skb);
1782
1783 /* No socket. Drop packet silently, if checksum is wrong */
1784 if (udp_lib_checksum_complete(skb))
1785 goto csum_error;
1786
Pavel Emelyanov02833282008-07-05 21:18:48 -07001787 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
David S. Millerdb8dac22008-03-06 16:22:02 -08001788 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1789
1790 /*
1791 * Hmm. We got an UDP packet to a port to which we
1792 * don't wanna listen. Ignore it.
1793 */
1794 kfree_skb(skb);
1795 return 0;
1796
1797short_packet:
Harvey Harrison673d57e2008-10-31 00:53:57 -07001798 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
Joe Perchesafd465032012-03-12 07:03:32 +00001799 proto == IPPROTO_UDPLITE ? "Lite" : "",
1800 &saddr, ntohs(uh->source),
1801 ulen, skb->len,
1802 &daddr, ntohs(uh->dest));
David S. Millerdb8dac22008-03-06 16:22:02 -08001803 goto drop;
1804
1805csum_error:
1806 /*
1807 * RFC1122: OK. Discards the bad packet silently (as far as
1808 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1809 */
Harvey Harrison673d57e2008-10-31 00:53:57 -07001810 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
Joe Perchesafd465032012-03-12 07:03:32 +00001811 proto == IPPROTO_UDPLITE ? "Lite" : "",
1812 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
David S. Millerdb8dac22008-03-06 16:22:02 -08001813 ulen);
Eric Dumazet6a5dc9e2013-04-29 08:39:56 +00001814 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
David S. Millerdb8dac22008-03-06 16:22:02 -08001815drop:
Pavel Emelyanov02833282008-07-05 21:18:48 -07001816 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
David S. Millerdb8dac22008-03-06 16:22:02 -08001817 kfree_skb(skb);
1818 return 0;
1819}
1820
Shawn Bohrer421b3882013-10-07 11:01:39 -05001821/* We can only early demux multicast if there is a single matching socket.
1822 * If more than one socket found returns NULL
1823 */
1824static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
1825 __be16 loc_port, __be32 loc_addr,
1826 __be16 rmt_port, __be32 rmt_addr,
1827 int dif)
1828{
1829 struct sock *sk, *result;
1830 struct hlist_nulls_node *node;
1831 unsigned short hnum = ntohs(loc_port);
1832 unsigned int count, slot = udp_hashfn(net, hnum, udp_table.mask);
1833 struct udp_hslot *hslot = &udp_table.hash[slot];
1834
1835 rcu_read_lock();
1836begin:
1837 count = 0;
1838 result = NULL;
1839 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
1840 if (__udp_is_mcast_sock(net, sk,
1841 loc_port, loc_addr,
1842 rmt_port, rmt_addr,
1843 dif, hnum)) {
1844 result = sk;
1845 ++count;
1846 }
1847 }
1848 /*
1849 * if the nulls value we got at the end of this lookup is
1850 * not the expected one, we must restart lookup.
1851 * We probably met an item that was moved to another chain.
1852 */
1853 if (get_nulls_value(node) != slot)
1854 goto begin;
1855
1856 if (result) {
1857 if (count != 1 ||
1858 unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
1859 result = NULL;
Eric Dumazetf69b9232013-10-08 21:47:29 -07001860 else if (unlikely(!__udp_is_mcast_sock(net, result,
Shawn Bohrer421b3882013-10-07 11:01:39 -05001861 loc_port, loc_addr,
1862 rmt_port, rmt_addr,
1863 dif, hnum))) {
1864 sock_put(result);
1865 result = NULL;
1866 }
1867 }
1868 rcu_read_unlock();
1869 return result;
1870}
1871
1872/* For unicast we should only early demux connected sockets or we can
1873 * break forwarding setups. The chains here can be long so only check
1874 * if the first socket is an exact match and if not move on.
1875 */
1876static struct sock *__udp4_lib_demux_lookup(struct net *net,
1877 __be16 loc_port, __be32 loc_addr,
1878 __be16 rmt_port, __be32 rmt_addr,
1879 int dif)
1880{
1881 struct sock *sk, *result;
1882 struct hlist_nulls_node *node;
1883 unsigned short hnum = ntohs(loc_port);
1884 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
1885 unsigned int slot2 = hash2 & udp_table.mask;
1886 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
1887 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr)
1888 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
1889
1890 rcu_read_lock();
1891 result = NULL;
1892 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
1893 if (INET_MATCH(sk, net, acookie,
1894 rmt_addr, loc_addr, ports, dif))
1895 result = sk;
1896 /* Only check first socket in chain */
1897 break;
1898 }
1899
1900 if (result) {
1901 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
1902 result = NULL;
1903 else if (unlikely(!INET_MATCH(sk, net, acookie,
1904 rmt_addr, loc_addr,
1905 ports, dif))) {
1906 sock_put(result);
1907 result = NULL;
1908 }
1909 }
1910 rcu_read_unlock();
1911 return result;
1912}
1913
1914void udp_v4_early_demux(struct sk_buff *skb)
1915{
1916 const struct iphdr *iph = ip_hdr(skb);
1917 const struct udphdr *uh = udp_hdr(skb);
1918 struct sock *sk;
1919 struct dst_entry *dst;
1920 struct net *net = dev_net(skb->dev);
1921 int dif = skb->dev->ifindex;
1922
1923 /* validate the packet */
1924 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
1925 return;
1926
1927 if (skb->pkt_type == PACKET_BROADCAST ||
1928 skb->pkt_type == PACKET_MULTICAST)
1929 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
1930 uh->source, iph->saddr, dif);
1931 else if (skb->pkt_type == PACKET_HOST)
1932 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
1933 uh->source, iph->saddr, dif);
1934 else
1935 return;
1936
1937 if (!sk)
1938 return;
1939
1940 skb->sk = sk;
1941 skb->destructor = sock_edemux;
1942 dst = sk->sk_rx_dst;
1943
1944 if (dst)
1945 dst = dst_check(dst, 0);
1946 if (dst)
1947 skb_dst_set_noref(skb, dst);
1948}
1949
David S. Millerdb8dac22008-03-06 16:22:02 -08001950int udp_rcv(struct sk_buff *skb)
1951{
Eric Dumazet645ca702008-10-29 01:41:45 -07001952 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
David S. Millerdb8dac22008-03-06 16:22:02 -08001953}
1954
Brian Haley7d06b2e2008-06-14 17:04:49 -07001955void udp_destroy_sock(struct sock *sk)
David S. Millerdb8dac22008-03-06 16:22:02 -08001956{
Tom Parkin44046a52013-03-19 06:11:12 +00001957 struct udp_sock *up = udp_sk(sk);
Eric Dumazet8a74ad62010-05-26 19:20:18 +00001958 bool slow = lock_sock_fast(sk);
David S. Millerdb8dac22008-03-06 16:22:02 -08001959 udp_flush_pending_frames(sk);
Eric Dumazet8a74ad62010-05-26 19:20:18 +00001960 unlock_sock_fast(sk, slow);
Tom Parkin44046a52013-03-19 06:11:12 +00001961 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1962 void (*encap_destroy)(struct sock *sk);
1963 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1964 if (encap_destroy)
1965 encap_destroy(sk);
1966 }
David S. Millerdb8dac22008-03-06 16:22:02 -08001967}
1968
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969/*
1970 * Socket option code for UDP
1971 */
Gerrit Renker4c0a6cb2006-11-27 09:29:59 -08001972int udp_lib_setsockopt(struct sock *sk, int level, int optname,
David S. Millerb7058842009-09-30 16:12:20 -07001973 char __user *optval, unsigned int optlen,
Gerrit Renker4c0a6cb2006-11-27 09:29:59 -08001974 int (*push_pending_frames)(struct sock *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975{
1976 struct udp_sock *up = udp_sk(sk);
1977 int val;
1978 int err = 0;
Wang Chenb2bf1e22007-12-03 22:34:16 +11001979 int is_udplite = IS_UDPLITE(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980
Eric Dumazetc482c562009-07-17 00:26:32 +00001981 if (optlen < sizeof(int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 return -EINVAL;
1983
1984 if (get_user(val, (int __user *)optval))
1985 return -EFAULT;
1986
Stephen Hemminger6516c652007-03-08 20:41:55 -08001987 switch (optname) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 case UDP_CORK:
1989 if (val != 0) {
1990 up->corkflag = 1;
1991 } else {
1992 up->corkflag = 0;
1993 lock_sock(sk);
Gerrit Renker4c0a6cb2006-11-27 09:29:59 -08001994 (*push_pending_frames)(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 release_sock(sk);
1996 }
1997 break;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001998
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 case UDP_ENCAP:
2000 switch (val) {
2001 case 0:
2002 case UDP_ENCAP_ESPINUDP:
2003 case UDP_ENCAP_ESPINUDP_NON_IKE:
James Chapman067b2072007-07-05 17:08:05 -07002004 up->encap_rcv = xfrm4_udp_encap_rcv;
2005 /* FALLTHROUGH */
James Chapman342f0232007-06-27 15:37:46 -07002006 case UDP_ENCAP_L2TPINUDP:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 up->encap_type = val;
Eric Dumazet447167b2012-04-11 23:05:28 +00002008 udp_encap_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 break;
2010 default:
2011 err = -ENOPROTOOPT;
2012 break;
2013 }
2014 break;
2015
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08002016 /*
2017 * UDP-Lite's partial checksum coverage (RFC 3828).
2018 */
2019 /* The sender sets actual checksum coverage length via this option.
2020 * The case coverage > packet length is handled by send module. */
2021 case UDPLITE_SEND_CSCOV:
Wang Chenb2bf1e22007-12-03 22:34:16 +11002022 if (!is_udplite) /* Disable the option on UDP sockets */
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08002023 return -ENOPROTOOPT;
2024 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2025 val = 8;
Alexey Dobriyan4be929b2010-05-24 14:33:03 -07002026 else if (val > USHRT_MAX)
2027 val = USHRT_MAX;
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08002028 up->pcslen = val;
2029 up->pcflag |= UDPLITE_SEND_CC;
2030 break;
2031
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09002032 /* The receiver specifies a minimum checksum coverage value. To make
2033 * sense, this should be set to at least 8 (as done below). If zero is
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08002034 * used, this again means full checksum coverage. */
2035 case UDPLITE_RECV_CSCOV:
Wang Chenb2bf1e22007-12-03 22:34:16 +11002036 if (!is_udplite) /* Disable the option on UDP sockets */
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08002037 return -ENOPROTOOPT;
2038 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2039 val = 8;
Alexey Dobriyan4be929b2010-05-24 14:33:03 -07002040 else if (val > USHRT_MAX)
2041 val = USHRT_MAX;
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08002042 up->pcrlen = val;
2043 up->pcflag |= UDPLITE_RECV_CC;
2044 break;
2045
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 default:
2047 err = -ENOPROTOOPT;
2048 break;
Stephen Hemminger6516c652007-03-08 20:41:55 -08002049 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050
2051 return err;
2052}
Eric Dumazetc482c562009-07-17 00:26:32 +00002053EXPORT_SYMBOL(udp_lib_setsockopt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054
David S. Millerdb8dac22008-03-06 16:22:02 -08002055int udp_setsockopt(struct sock *sk, int level, int optname,
David S. Millerb7058842009-09-30 16:12:20 -07002056 char __user *optval, unsigned int optlen)
David S. Millerdb8dac22008-03-06 16:22:02 -08002057{
2058 if (level == SOL_UDP || level == SOL_UDPLITE)
2059 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2060 udp_push_pending_frames);
2061 return ip_setsockopt(sk, level, optname, optval, optlen);
2062}
2063
2064#ifdef CONFIG_COMPAT
2065int compat_udp_setsockopt(struct sock *sk, int level, int optname,
David S. Millerb7058842009-09-30 16:12:20 -07002066 char __user *optval, unsigned int optlen)
David S. Millerdb8dac22008-03-06 16:22:02 -08002067{
2068 if (level == SOL_UDP || level == SOL_UDPLITE)
2069 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2070 udp_push_pending_frames);
2071 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2072}
2073#endif
2074
Gerrit Renker4c0a6cb2006-11-27 09:29:59 -08002075int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2076 char __user *optval, int __user *optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077{
2078 struct udp_sock *up = udp_sk(sk);
2079 int val, len;
2080
Eric Dumazetc482c562009-07-17 00:26:32 +00002081 if (get_user(len, optlen))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 return -EFAULT;
2083
2084 len = min_t(unsigned int, len, sizeof(int));
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09002085
Stephen Hemminger6516c652007-03-08 20:41:55 -08002086 if (len < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 return -EINVAL;
2088
Stephen Hemminger6516c652007-03-08 20:41:55 -08002089 switch (optname) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 case UDP_CORK:
2091 val = up->corkflag;
2092 break;
2093
2094 case UDP_ENCAP:
2095 val = up->encap_type;
2096 break;
2097
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08002098 /* The following two cannot be changed on UDP sockets, the return is
2099 * always 0 (which corresponds to the full checksum coverage of UDP). */
2100 case UDPLITE_SEND_CSCOV:
2101 val = up->pcslen;
2102 break;
2103
2104 case UDPLITE_RECV_CSCOV:
2105 val = up->pcrlen;
2106 break;
2107
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 default:
2109 return -ENOPROTOOPT;
Stephen Hemminger6516c652007-03-08 20:41:55 -08002110 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111
Stephen Hemminger6516c652007-03-08 20:41:55 -08002112 if (put_user(len, optlen))
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09002113 return -EFAULT;
Eric Dumazetc482c562009-07-17 00:26:32 +00002114 if (copy_to_user(optval, &val, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 return -EFAULT;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09002116 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117}
Eric Dumazetc482c562009-07-17 00:26:32 +00002118EXPORT_SYMBOL(udp_lib_getsockopt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119
David S. Millerdb8dac22008-03-06 16:22:02 -08002120int udp_getsockopt(struct sock *sk, int level, int optname,
2121 char __user *optval, int __user *optlen)
2122{
2123 if (level == SOL_UDP || level == SOL_UDPLITE)
2124 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2125 return ip_getsockopt(sk, level, optname, optval, optlen);
2126}
2127
2128#ifdef CONFIG_COMPAT
2129int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2130 char __user *optval, int __user *optlen)
2131{
2132 if (level == SOL_UDP || level == SOL_UDPLITE)
2133 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2134 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2135}
2136#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137/**
2138 * udp_poll - wait for a UDP event.
2139 * @file - file struct
2140 * @sock - socket
2141 * @wait - poll table
2142 *
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09002143 * This is same as datagram poll, except for the special case of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 * blocking sockets. If application is using a blocking fd
2145 * and a packet with checksum error is in the queue;
2146 * then it could get return from select indicating data available
2147 * but then block when reading it. Add special case code
2148 * to work around these arguably broken applications.
2149 */
2150unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2151{
2152 unsigned int mask = datagram_poll(file, sock, wait);
2153 struct sock *sk = sock->sk;
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08002154
David Majnemerc3f1dba2013-05-31 13:15:38 +00002155 sock_rps_record_flow(sk);
2156
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 /* Check for false positives due to checksum errors */
Eric Dumazet85584672009-10-09 04:43:40 +00002158 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2159 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
2160 mask &= ~(POLLIN | POLLRDNORM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161
2162 return mask;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09002163
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164}
Eric Dumazetc482c562009-07-17 00:26:32 +00002165EXPORT_SYMBOL(udp_poll);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166
David S. Millerdb8dac22008-03-06 16:22:02 -08002167struct proto udp_prot = {
2168 .name = "UDP",
2169 .owner = THIS_MODULE,
2170 .close = udp_lib_close,
2171 .connect = ip4_datagram_connect,
2172 .disconnect = udp_disconnect,
2173 .ioctl = udp_ioctl,
2174 .destroy = udp_destroy_sock,
2175 .setsockopt = udp_setsockopt,
2176 .getsockopt = udp_getsockopt,
2177 .sendmsg = udp_sendmsg,
2178 .recvmsg = udp_recvmsg,
2179 .sendpage = udp_sendpage,
Herbert Xu93821772008-09-15 11:48:46 -07002180 .backlog_rcv = __udp_queue_rcv_skb,
Steffen Klassert8141ed92013-01-21 02:00:03 +00002181 .release_cb = ip4_datagram_release_cb,
David S. Millerdb8dac22008-03-06 16:22:02 -08002182 .hash = udp_lib_hash,
2183 .unhash = udp_lib_unhash,
Eric Dumazet719f8352010-09-08 05:08:44 +00002184 .rehash = udp_v4_rehash,
David S. Millerdb8dac22008-03-06 16:22:02 -08002185 .get_port = udp_v4_get_port,
2186 .memory_allocated = &udp_memory_allocated,
2187 .sysctl_mem = sysctl_udp_mem,
2188 .sysctl_wmem = &sysctl_udp_wmem_min,
2189 .sysctl_rmem = &sysctl_udp_rmem_min,
2190 .obj_size = sizeof(struct udp_sock),
Eric Dumazet271b72c2008-10-29 02:11:14 -07002191 .slab_flags = SLAB_DESTROY_BY_RCU,
Eric Dumazet645ca702008-10-29 01:41:45 -07002192 .h.udp_table = &udp_table,
David S. Millerdb8dac22008-03-06 16:22:02 -08002193#ifdef CONFIG_COMPAT
2194 .compat_setsockopt = compat_udp_setsockopt,
2195 .compat_getsockopt = compat_udp_getsockopt,
2196#endif
Octavian Purdilafcbdf092010-12-16 14:26:56 -08002197 .clear_sk = sk_prot_clear_portaddr_nulls,
David S. Millerdb8dac22008-03-06 16:22:02 -08002198};
Eric Dumazetc482c562009-07-17 00:26:32 +00002199EXPORT_SYMBOL(udp_prot);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200
2201/* ------------------------------------------------------------------------ */
2202#ifdef CONFIG_PROC_FS
2203
Eric Dumazet645ca702008-10-29 01:41:45 -07002204static struct sock *udp_get_first(struct seq_file *seq, int start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205{
2206 struct sock *sk;
2207 struct udp_iter_state *state = seq->private;
Denis V. Lunev6f191ef2008-03-28 18:23:33 -07002208 struct net *net = seq_file_net(seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002210 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2211 ++state->bucket) {
Eric Dumazet88ab1932008-11-16 19:39:21 -08002212 struct hlist_nulls_node *node;
Eric Dumazet645ca702008-10-29 01:41:45 -07002213 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002214
2215 if (hlist_nulls_empty(&hslot->head))
2216 continue;
2217
Eric Dumazet645ca702008-10-29 01:41:45 -07002218 spin_lock_bh(&hslot->lock);
Eric Dumazet88ab1932008-11-16 19:39:21 -08002219 sk_nulls_for_each(sk, node, &hslot->head) {
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +09002220 if (!net_eq(sock_net(sk), net))
Daniel Lezcanoa91275e2008-03-21 04:11:58 -07002221 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 if (sk->sk_family == state->family)
2223 goto found;
2224 }
Eric Dumazet645ca702008-10-29 01:41:45 -07002225 spin_unlock_bh(&hslot->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 }
2227 sk = NULL;
2228found:
2229 return sk;
2230}
2231
2232static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2233{
2234 struct udp_iter_state *state = seq->private;
Denis V. Lunev6f191ef2008-03-28 18:23:33 -07002235 struct net *net = seq_file_net(seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236
2237 do {
Eric Dumazet88ab1932008-11-16 19:39:21 -08002238 sk = sk_nulls_next(sk);
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +09002239 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240
Eric Dumazet645ca702008-10-29 01:41:45 -07002241 if (!sk) {
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002242 if (state->bucket <= state->udp_table->mask)
Vitaly Mayatskikh30842f22009-03-23 15:22:33 -07002243 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
Eric Dumazet645ca702008-10-29 01:41:45 -07002244 return udp_get_first(seq, state->bucket + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 }
2246 return sk;
2247}
2248
2249static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2250{
Eric Dumazet645ca702008-10-29 01:41:45 -07002251 struct sock *sk = udp_get_first(seq, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252
2253 if (sk)
Stephen Hemminger6516c652007-03-08 20:41:55 -08002254 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 --pos;
2256 return pos ? NULL : sk;
2257}
2258
2259static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2260{
Vitaly Mayatskikh30842f22009-03-23 15:22:33 -07002261 struct udp_iter_state *state = seq->private;
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002262 state->bucket = MAX_UDP_PORTS;
Vitaly Mayatskikh30842f22009-03-23 15:22:33 -07002263
YOSHIFUJI Hideakib50660f2008-03-31 19:38:15 -07002264 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265}
2266
2267static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2268{
2269 struct sock *sk;
2270
YOSHIFUJI Hideakib50660f2008-03-31 19:38:15 -07002271 if (v == SEQ_START_TOKEN)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 sk = udp_get_idx(seq, 0);
2273 else
2274 sk = udp_get_next(seq, v);
2275
2276 ++*pos;
2277 return sk;
2278}
2279
2280static void udp_seq_stop(struct seq_file *seq, void *v)
2281{
Eric Dumazet645ca702008-10-29 01:41:45 -07002282 struct udp_iter_state *state = seq->private;
2283
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002284 if (state->bucket <= state->udp_table->mask)
Eric Dumazet645ca702008-10-29 01:41:45 -07002285 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286}
2287
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00002288int udp_seq_open(struct inode *inode, struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289{
Al Virod9dda782013-03-31 18:16:14 -04002290 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
Denis V. Luneva2be75c2008-03-28 18:25:06 -07002291 struct udp_iter_state *s;
2292 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293
Denis V. Luneva2be75c2008-03-28 18:25:06 -07002294 err = seq_open_net(inode, file, &afinfo->seq_ops,
2295 sizeof(struct udp_iter_state));
2296 if (err < 0)
2297 return err;
Daniel Lezcanoa91275e2008-03-21 04:11:58 -07002298
Denis V. Luneva2be75c2008-03-28 18:25:06 -07002299 s = ((struct seq_file *)file->private_data)->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 s->family = afinfo->family;
Eric Dumazet645ca702008-10-29 01:41:45 -07002301 s->udp_table = afinfo->udp_table;
Denis V. Luneva2be75c2008-03-28 18:25:06 -07002302 return err;
Daniel Lezcanoa91275e2008-03-21 04:11:58 -07002303}
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00002304EXPORT_SYMBOL(udp_seq_open);
Daniel Lezcanoa91275e2008-03-21 04:11:58 -07002305
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306/* ------------------------------------------------------------------------ */
Daniel Lezcano0c96d8c2008-03-21 04:14:17 -07002307int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308{
2309 struct proc_dir_entry *p;
2310 int rc = 0;
2311
Denis V. Lunevdda61922008-03-28 18:24:26 -07002312 afinfo->seq_ops.start = udp_seq_start;
2313 afinfo->seq_ops.next = udp_seq_next;
2314 afinfo->seq_ops.stop = udp_seq_stop;
2315
Denis V. Lunev84841c32008-05-02 04:10:08 -07002316 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00002317 afinfo->seq_fops, afinfo);
Denis V. Lunev84841c32008-05-02 04:10:08 -07002318 if (!p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 rc = -ENOMEM;
2320 return rc;
2321}
Eric Dumazetc482c562009-07-17 00:26:32 +00002322EXPORT_SYMBOL(udp_proc_register);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323
Daniel Lezcano0c96d8c2008-03-21 04:14:17 -07002324void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325{
Gao fengece31ff2013-02-18 01:34:56 +00002326 remove_proc_entry(afinfo->name, net->proc_net);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327}
Eric Dumazetc482c562009-07-17 00:26:32 +00002328EXPORT_SYMBOL(udp_proc_unregister);
David S. Millerdb8dac22008-03-06 16:22:02 -08002329
2330/* ------------------------------------------------------------------------ */
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002331static void udp4_format_sock(struct sock *sp, struct seq_file *f,
Tetsuo Handa652586d2013-11-14 14:31:57 -08002332 int bucket)
David S. Millerdb8dac22008-03-06 16:22:02 -08002333{
2334 struct inet_sock *inet = inet_sk(sp);
Eric Dumazetc720c7e2009-10-15 06:30:45 +00002335 __be32 dest = inet->inet_daddr;
2336 __be32 src = inet->inet_rcv_saddr;
2337 __u16 destp = ntohs(inet->inet_dport);
2338 __u16 srcp = ntohs(inet->inet_sport);
David S. Millerdb8dac22008-03-06 16:22:02 -08002339
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002340 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
Tetsuo Handa652586d2013-11-14 14:31:57 -08002341 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
David S. Millerdb8dac22008-03-06 16:22:02 -08002342 bucket, src, srcp, dest, destp, sp->sk_state,
Eric Dumazet31e6d362009-06-17 19:05:41 -07002343 sk_wmem_alloc_get(sp),
2344 sk_rmem_alloc_get(sp),
Eric W. Biedermana7cb5a42012-05-24 01:10:10 -06002345 0, 0L, 0,
2346 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2347 0, sock_i_ino(sp),
Eric Dumazetcb61cb92008-06-17 21:04:56 -07002348 atomic_read(&sp->sk_refcnt), sp,
Tetsuo Handa652586d2013-11-14 14:31:57 -08002349 atomic_read(&sp->sk_drops));
David S. Millerdb8dac22008-03-06 16:22:02 -08002350}
2351
2352int udp4_seq_show(struct seq_file *seq, void *v)
2353{
Tetsuo Handa652586d2013-11-14 14:31:57 -08002354 seq_setwidth(seq, 127);
David S. Millerdb8dac22008-03-06 16:22:02 -08002355 if (v == SEQ_START_TOKEN)
Tetsuo Handa652586d2013-11-14 14:31:57 -08002356 seq_puts(seq, " sl local_address rem_address st tx_queue "
David S. Millerdb8dac22008-03-06 16:22:02 -08002357 "rx_queue tr tm->when retrnsmt uid timeout "
Eric Dumazetcb61cb92008-06-17 21:04:56 -07002358 "inode ref pointer drops");
David S. Millerdb8dac22008-03-06 16:22:02 -08002359 else {
David S. Millerdb8dac22008-03-06 16:22:02 -08002360 struct udp_iter_state *state = seq->private;
2361
Tetsuo Handa652586d2013-11-14 14:31:57 -08002362 udp4_format_sock(v, seq, state->bucket);
David S. Millerdb8dac22008-03-06 16:22:02 -08002363 }
Tetsuo Handa652586d2013-11-14 14:31:57 -08002364 seq_pad(seq, '\n');
David S. Millerdb8dac22008-03-06 16:22:02 -08002365 return 0;
2366}
2367
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00002368static const struct file_operations udp_afinfo_seq_fops = {
2369 .owner = THIS_MODULE,
2370 .open = udp_seq_open,
2371 .read = seq_read,
2372 .llseek = seq_lseek,
2373 .release = seq_release_net
2374};
2375
David S. Millerdb8dac22008-03-06 16:22:02 -08002376/* ------------------------------------------------------------------------ */
David S. Millerdb8dac22008-03-06 16:22:02 -08002377static struct udp_seq_afinfo udp4_seq_afinfo = {
David S. Millerdb8dac22008-03-06 16:22:02 -08002378 .name = "udp",
2379 .family = AF_INET,
Eric Dumazet645ca702008-10-29 01:41:45 -07002380 .udp_table = &udp_table,
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00002381 .seq_fops = &udp_afinfo_seq_fops,
Denis V. Lunevdda61922008-03-28 18:24:26 -07002382 .seq_ops = {
2383 .show = udp4_seq_show,
2384 },
David S. Millerdb8dac22008-03-06 16:22:02 -08002385};
2386
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00002387static int __net_init udp4_proc_init_net(struct net *net)
Pavel Emelyanov15439fe2008-03-24 14:53:49 -07002388{
2389 return udp_proc_register(net, &udp4_seq_afinfo);
2390}
2391
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00002392static void __net_exit udp4_proc_exit_net(struct net *net)
Pavel Emelyanov15439fe2008-03-24 14:53:49 -07002393{
2394 udp_proc_unregister(net, &udp4_seq_afinfo);
2395}
2396
2397static struct pernet_operations udp4_net_ops = {
2398 .init = udp4_proc_init_net,
2399 .exit = udp4_proc_exit_net,
2400};
2401
David S. Millerdb8dac22008-03-06 16:22:02 -08002402int __init udp4_proc_init(void)
2403{
Pavel Emelyanov15439fe2008-03-24 14:53:49 -07002404 return register_pernet_subsys(&udp4_net_ops);
David S. Millerdb8dac22008-03-06 16:22:02 -08002405}
2406
2407void udp4_proc_exit(void)
2408{
Pavel Emelyanov15439fe2008-03-24 14:53:49 -07002409 unregister_pernet_subsys(&udp4_net_ops);
David S. Millerdb8dac22008-03-06 16:22:02 -08002410}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411#endif /* CONFIG_PROC_FS */
2412
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002413static __initdata unsigned long uhash_entries;
2414static int __init set_uhash_entries(char *str)
Eric Dumazet645ca702008-10-29 01:41:45 -07002415{
Eldad Zack413c27d2012-05-19 14:13:18 +00002416 ssize_t ret;
2417
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002418 if (!str)
2419 return 0;
Eldad Zack413c27d2012-05-19 14:13:18 +00002420
2421 ret = kstrtoul(str, 0, &uhash_entries);
2422 if (ret)
2423 return 0;
2424
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002425 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2426 uhash_entries = UDP_HTABLE_SIZE_MIN;
2427 return 1;
2428}
2429__setup("uhash_entries=", set_uhash_entries);
Eric Dumazet645ca702008-10-29 01:41:45 -07002430
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002431void __init udp_table_init(struct udp_table *table, const char *name)
2432{
2433 unsigned int i;
2434
Tim Bird31fe62b2012-05-23 13:33:35 +00002435 table->hash = alloc_large_system_hash(name,
2436 2 * sizeof(struct udp_hslot),
2437 uhash_entries,
2438 21, /* one slot per 2 MB */
2439 0,
2440 &table->log,
2441 &table->mask,
2442 UDP_HTABLE_SIZE_MIN,
2443 64 * 1024);
2444
Eric Dumazet512615b2009-11-08 10:17:58 +00002445 table->hash2 = table->hash + (table->mask + 1);
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002446 for (i = 0; i <= table->mask; i++) {
Eric Dumazet88ab1932008-11-16 19:39:21 -08002447 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
Eric Dumazetfdcc8aa92009-11-08 10:17:05 +00002448 table->hash[i].count = 0;
Eric Dumazet645ca702008-10-29 01:41:45 -07002449 spin_lock_init(&table->hash[i].lock);
2450 }
Eric Dumazet512615b2009-11-08 10:17:58 +00002451 for (i = 0; i <= table->mask; i++) {
2452 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2453 table->hash2[i].count = 0;
2454 spin_lock_init(&table->hash2[i].lock);
2455 }
Eric Dumazet645ca702008-10-29 01:41:45 -07002456}
2457
Hideo Aoki95766ff2007-12-31 00:29:24 -08002458void __init udp_init(void)
2459{
Eric Dumazetf03d78d2011-07-07 00:27:05 -07002460 unsigned long limit;
Hideo Aoki95766ff2007-12-31 00:29:24 -08002461
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002462 udp_table_init(&udp_table, "UDP");
Eric Dumazetf03d78d2011-07-07 00:27:05 -07002463 limit = nr_free_buffer_pages() / 8;
Hideo Aoki95766ff2007-12-31 00:29:24 -08002464 limit = max(limit, 128UL);
2465 sysctl_udp_mem[0] = limit / 4 * 3;
2466 sysctl_udp_mem[1] = limit;
2467 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2468
2469 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2470 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2471}
2472
Daniel Borkmannda5bab02013-06-08 12:56:03 +02002473struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
2474 netdev_features_t features)
Pravin B Shelar73136262013-03-07 13:21:51 +00002475{
2476 struct sk_buff *segs = ERR_PTR(-EINVAL);
2477 int mac_len = skb->mac_len;
2478 int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb);
Pravin B Shelar0d055352013-05-02 16:17:02 +00002479 __be16 protocol = skb->protocol;
Pravin B Shelar73136262013-03-07 13:21:51 +00002480 netdev_features_t enc_features;
Pravin B Shelar0d055352013-05-02 16:17:02 +00002481 int outer_hlen;
Pravin B Shelar73136262013-03-07 13:21:51 +00002482
2483 if (unlikely(!pskb_may_pull(skb, tnl_hlen)))
2484 goto out;
2485
2486 skb->encapsulation = 0;
2487 __skb_pull(skb, tnl_hlen);
2488 skb_reset_mac_header(skb);
2489 skb_set_network_header(skb, skb_inner_network_offset(skb));
2490 skb->mac_len = skb_inner_network_offset(skb);
Pravin B Shelar19acc322013-05-07 20:41:07 +00002491 skb->protocol = htons(ETH_P_TEB);
Pravin B Shelar73136262013-03-07 13:21:51 +00002492
2493 /* segment inner packet. */
2494 enc_features = skb->dev->hw_enc_features & netif_skb_features(skb);
2495 segs = skb_mac_gso_segment(skb, enc_features);
2496 if (!segs || IS_ERR(segs))
2497 goto out;
2498
2499 outer_hlen = skb_tnl_header_len(skb);
2500 skb = segs;
2501 do {
2502 struct udphdr *uh;
2503 int udp_offset = outer_hlen - tnl_hlen;
2504
Alexander Duyckcdbaa0b2013-07-10 17:05:06 -07002505 skb_reset_inner_headers(skb);
2506 skb->encapsulation = 1;
2507
Pravin B Shelar73136262013-03-07 13:21:51 +00002508 skb->mac_len = mac_len;
2509
2510 skb_push(skb, outer_hlen);
2511 skb_reset_mac_header(skb);
2512 skb_set_network_header(skb, mac_len);
2513 skb_set_transport_header(skb, udp_offset);
2514 uh = udp_hdr(skb);
2515 uh->len = htons(skb->len - udp_offset);
2516
2517 /* csum segment if tunnel sets skb with csum. */
Cong Wangeb3c0d82013-08-31 13:44:38 +08002518 if (protocol == htons(ETH_P_IP) && unlikely(uh->check)) {
Pravin B Shelar73136262013-03-07 13:21:51 +00002519 struct iphdr *iph = ip_hdr(skb);
2520
2521 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
2522 skb->len - udp_offset,
2523 IPPROTO_UDP, 0);
2524 uh->check = csum_fold(skb_checksum(skb, udp_offset,
2525 skb->len - udp_offset, 0));
2526 if (uh->check == 0)
2527 uh->check = CSUM_MANGLED_0;
2528
Cong Wangeb3c0d82013-08-31 13:44:38 +08002529 } else if (protocol == htons(ETH_P_IPV6)) {
2530 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
2531 u32 len = skb->len - udp_offset;
2532
2533 uh->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
2534 len, IPPROTO_UDP, 0);
2535 uh->check = csum_fold(skb_checksum(skb, udp_offset, len, 0));
2536 if (uh->check == 0)
2537 uh->check = CSUM_MANGLED_0;
2538 skb->ip_summed = CHECKSUM_NONE;
Pravin B Shelar73136262013-03-07 13:21:51 +00002539 }
Cong Wangeb3c0d82013-08-31 13:44:38 +08002540
Pravin B Shelar0d055352013-05-02 16:17:02 +00002541 skb->protocol = protocol;
Pravin B Shelar73136262013-03-07 13:21:51 +00002542 } while ((skb = skb->next));
2543out:
2544 return segs;
2545}