blob: 47c7aa0501af70e53332cc9e4ae79151f548d6f9 [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 Torvalds7c0f6ba2016-12-24 11:46:01 -080082#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070083#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>
Shawn Bohrer6e540302015-06-03 16:27:38 -050093#include <linux/inetdevice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070094#include <linux/in.h>
95#include <linux/errno.h>
96#include <linux/timer.h>
97#include <linux/mm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070098#include <linux/inet.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070099#include <linux/netdevice.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +0900100#include <linux/slab.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -0700101#include <net/tcp_states.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102#include <linux/skbuff.h>
103#include <linux/proc_fs.h>
104#include <linux/seq_file.h>
Eric W. Biederman457c4cb2007-09-12 12:01:34 +0200105#include <net/net_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106#include <net/icmp.h>
Shawn Bohrer421b3882013-10-07 11:01:39 -0500107#include <net/inet_hashtables.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108#include <net/route.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109#include <net/checksum.h>
110#include <net/xfrm.h>
Satoru Moriya296f7ea2011-06-17 11:58:39 +0000111#include <trace/events/udp.h>
Eric Dumazet447167b2012-04-11 23:05:28 +0000112#include <linux/static_key.h>
Eric Dumazet22911fc2012-06-27 00:23:44 +0000113#include <trace/events/skb.h>
Eliezer Tamir076bb0c2013-07-10 17:13:17 +0300114#include <net/busy_poll.h>
Gerrit Renkerba4e58e2006-11-27 11:10:57 -0800115#include "udp_impl.h"
Craig Galleke32ea7e2016-01-04 17:41:46 -0500116#include <net/sock_reuseport.h>
Eric Dumazet217375a2016-08-18 09:59:12 -0700117#include <net/addrconf.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118
Eric Dumazetf86dcc52009-10-07 00:37:59 +0000119struct udp_table udp_table __read_mostly;
Eric Dumazet645ca702008-10-29 01:41:45 -0700120EXPORT_SYMBOL(udp_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121
Eric Dumazet8d987e52010-11-09 23:24:26 +0000122long sysctl_udp_mem[3] __read_mostly;
Hideo Aoki95766ff2007-12-31 00:29:24 -0800123EXPORT_SYMBOL(sysctl_udp_mem);
Eric Dumazetc482c562009-07-17 00:26:32 +0000124
125int sysctl_udp_rmem_min __read_mostly;
Hideo Aoki95766ff2007-12-31 00:29:24 -0800126EXPORT_SYMBOL(sysctl_udp_rmem_min);
Eric Dumazetc482c562009-07-17 00:26:32 +0000127
128int sysctl_udp_wmem_min __read_mostly;
Hideo Aoki95766ff2007-12-31 00:29:24 -0800129EXPORT_SYMBOL(sysctl_udp_wmem_min);
130
Eric Dumazet8d987e52010-11-09 23:24:26 +0000131atomic_long_t udp_memory_allocated;
Hideo Aoki95766ff2007-12-31 00:29:24 -0800132EXPORT_SYMBOL(udp_memory_allocated);
133
Eric Dumazetf86dcc52009-10-07 00:37:59 +0000134#define MAX_UDP_PORTS 65536
135#define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
Eric Dumazet98322f22009-01-26 21:35:35 -0800136
Robert Shearman63a6fff2017-01-26 18:02:24 +0000137/* IPCB reference means this can not be used from early demux */
138static bool udp_lib_exact_dif_match(struct net *net, struct sk_buff *skb)
139{
140#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
141 if (!net->ipv4.sysctl_udp_l3mdev_accept &&
142 skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
143 return true;
144#endif
145 return false;
146}
147
Eric Dumazetf24d43c2008-10-09 14:51:27 -0700148static int udp_lib_lport_inuse(struct net *net, __u16 num,
Eric Dumazet645ca702008-10-29 01:41:45 -0700149 const struct udp_hslot *hslot,
Eric Dumazet98322f22009-01-26 21:35:35 -0800150 unsigned long *bitmap,
Josef Bacikfe38d2a2017-01-17 07:51:01 -0800151 struct sock *sk, unsigned int log)
Gerrit Renker25030a72006-08-26 20:06:05 -0700152{
Eric Dumazetf24d43c2008-10-09 14:51:27 -0700153 struct sock *sk2;
Tom Herbertba418fa2013-01-22 09:50:32 +0000154 kuid_t uid = sock_i_uid(sk);
Gerrit Renker25030a72006-08-26 20:06:05 -0700155
Eric Dumazetca065d02016-04-01 08:52:13 -0700156 sk_for_each(sk2, &hslot->head) {
Joe Perches9d4fb272009-11-23 10:41:23 -0800157 if (net_eq(sock_net(sk2), net) &&
158 sk2 != sk &&
Eric Dumazetd4cada42009-11-08 10:17:30 +0000159 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
Joe Perches9d4fb272009-11-23 10:41:23 -0800160 (!sk2->sk_reuse || !sk->sk_reuse) &&
161 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
162 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
Josef Bacikfe38d2a2017-01-17 07:51:01 -0800163 inet_rcv_saddr_equal(sk, sk2, true)) {
Eric Garverdf560052017-01-05 20:22:36 -0500164 if (sk2->sk_reuseport && sk->sk_reuseport &&
165 !rcu_access_pointer(sk->sk_reuseport_cb) &&
166 uid_eq(uid, sock_i_uid(sk2))) {
167 if (!bitmap)
168 return 0;
169 } else {
170 if (!bitmap)
171 return 1;
172 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
173 bitmap);
174 }
Eric Dumazet98322f22009-01-26 21:35:35 -0800175 }
Joe Perches4243cdc2014-11-11 21:59:20 -0800176 }
Gerrit Renker25030a72006-08-26 20:06:05 -0700177 return 0;
178}
179
Eric Dumazet30fff922009-11-09 05:26:33 +0000180/*
181 * Note: we still hold spinlock of primary hash chain, so no other writer
182 * can insert/delete a socket with local_port == num
183 */
184static int udp_lib_lport_inuse2(struct net *net, __u16 num,
Joe Perches4243cdc2014-11-11 21:59:20 -0800185 struct udp_hslot *hslot2,
Josef Bacikfe38d2a2017-01-17 07:51:01 -0800186 struct sock *sk)
Eric Dumazet30fff922009-11-09 05:26:33 +0000187{
188 struct sock *sk2;
Tom Herbertba418fa2013-01-22 09:50:32 +0000189 kuid_t uid = sock_i_uid(sk);
Eric Dumazet30fff922009-11-09 05:26:33 +0000190 int res = 0;
191
192 spin_lock(&hslot2->lock);
Eric Dumazetca065d02016-04-01 08:52:13 -0700193 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
Joe Perches9d4fb272009-11-23 10:41:23 -0800194 if (net_eq(sock_net(sk2), net) &&
195 sk2 != sk &&
196 (udp_sk(sk2)->udp_port_hash == num) &&
197 (!sk2->sk_reuse || !sk->sk_reuse) &&
198 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
199 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
Josef Bacikfe38d2a2017-01-17 07:51:01 -0800200 inet_rcv_saddr_equal(sk, sk2, true)) {
Eric Garverdf560052017-01-05 20:22:36 -0500201 if (sk2->sk_reuseport && sk->sk_reuseport &&
202 !rcu_access_pointer(sk->sk_reuseport_cb) &&
203 uid_eq(uid, sock_i_uid(sk2))) {
204 res = 0;
205 } else {
206 res = 1;
207 }
Eric Dumazet30fff922009-11-09 05:26:33 +0000208 break;
209 }
Joe Perches4243cdc2014-11-11 21:59:20 -0800210 }
Eric Dumazet30fff922009-11-09 05:26:33 +0000211 spin_unlock(&hslot2->lock);
212 return res;
213}
214
Josef Bacikfe38d2a2017-01-17 07:51:01 -0800215static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
Craig Galleke32ea7e2016-01-04 17:41:46 -0500216{
217 struct net *net = sock_net(sk);
Craig Galleke32ea7e2016-01-04 17:41:46 -0500218 kuid_t uid = sock_i_uid(sk);
219 struct sock *sk2;
220
Eric Dumazetca065d02016-04-01 08:52:13 -0700221 sk_for_each(sk2, &hslot->head) {
Craig Galleke32ea7e2016-01-04 17:41:46 -0500222 if (net_eq(sock_net(sk2), net) &&
223 sk2 != sk &&
224 sk2->sk_family == sk->sk_family &&
225 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
226 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
227 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
228 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
Josef Bacikfe38d2a2017-01-17 07:51:01 -0800229 inet_rcv_saddr_equal(sk, sk2, false)) {
Craig Galleke32ea7e2016-01-04 17:41:46 -0500230 return reuseport_add_sock(sk, sk2);
231 }
232 }
233
234 /* Initial allocation may have already happened via setsockopt */
235 if (!rcu_access_pointer(sk->sk_reuseport_cb))
236 return reuseport_alloc(sk);
237 return 0;
238}
239
Gerrit Renker25030a72006-08-26 20:06:05 -0700240/**
Pavel Emelyanov6ba5a3c2008-03-22 16:51:21 -0700241 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
Gerrit Renker25030a72006-08-26 20:06:05 -0700242 *
243 * @sk: socket struct in question
244 * @snum: port number to look up
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300245 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
Eric Dumazet30fff922009-11-09 05:26:33 +0000246 * with NULL address
Gerrit Renker25030a72006-08-26 20:06:05 -0700247 */
Pavel Emelyanov6ba5a3c2008-03-22 16:51:21 -0700248int udp_lib_get_port(struct sock *sk, unsigned short snum,
Eric Dumazet30fff922009-11-09 05:26:33 +0000249 unsigned int hash2_nulladdr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250{
Eric Dumazet512615b2009-11-08 10:17:58 +0000251 struct udp_hslot *hslot, *hslot2;
Eric Dumazet645ca702008-10-29 01:41:45 -0700252 struct udp_table *udptable = sk->sk_prot->h.udp_table;
Gerrit Renker25030a72006-08-26 20:06:05 -0700253 int error = 1;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900254 struct net *net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255
Stephen Hemminger32c1da72007-08-24 23:09:41 -0700256 if (!snum) {
Eric Dumazet9088c562008-10-08 11:44:17 -0700257 int low, high, remaining;
Eric Dumazet95c96172012-04-15 05:58:06 +0000258 unsigned int rand;
Eric Dumazet98322f22009-01-26 21:35:35 -0800259 unsigned short first, last;
260 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261
Eric W. Biederman0bbf87d2013-09-28 14:10:59 -0700262 inet_get_local_port_range(net, &low, &high);
Anton Arapova25de532007-10-18 22:00:17 -0700263 remaining = (high - low) + 1;
Stephen Hemminger227b60f2007-10-10 17:30:46 -0700264
Aruna-Hewapathirane63862b52014-01-11 07:15:59 -0500265 rand = prandom_u32();
Daniel Borkmann8fc54f62014-08-23 20:58:54 +0200266 first = reciprocal_scale(rand, remaining) + low;
Eric Dumazet98322f22009-01-26 21:35:35 -0800267 /*
268 * force rand to be an odd multiple of UDP_HTABLE_SIZE
269 */
Eric Dumazetf86dcc52009-10-07 00:37:59 +0000270 rand = (rand | 1) * (udptable->mask + 1);
Eric Dumazet5781b232009-12-13 19:32:39 -0800271 last = first + udptable->mask + 1;
272 do {
Eric Dumazetf86dcc52009-10-07 00:37:59 +0000273 hslot = udp_hashslot(udptable, net, first);
Eric Dumazet98322f22009-01-26 21:35:35 -0800274 bitmap_zero(bitmap, PORTS_PER_CHAIN);
Eric Dumazet645ca702008-10-29 01:41:45 -0700275 spin_lock_bh(&hslot->lock);
Eric Dumazet98322f22009-01-26 21:35:35 -0800276 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
Josef Bacikfe38d2a2017-01-17 07:51:01 -0800277 udptable->log);
Eric Dumazet98322f22009-01-26 21:35:35 -0800278
279 snum = first;
280 /*
281 * Iterate on all possible values of snum for this hash.
282 * Using steps of an odd multiple of UDP_HTABLE_SIZE
283 * give us randomization and full range coverage.
284 */
Eric Dumazet9088c562008-10-08 11:44:17 -0700285 do {
Eric Dumazet98322f22009-01-26 21:35:35 -0800286 if (low <= snum && snum <= high &&
Amerigo Wange3826f12010-05-05 00:27:06 +0000287 !test_bit(snum >> udptable->log, bitmap) &&
WANG Cong122ff242014-05-12 16:04:53 -0700288 !inet_is_local_reserved_port(net, snum))
Eric Dumazet98322f22009-01-26 21:35:35 -0800289 goto found;
290 snum += rand;
291 } while (snum != first);
292 spin_unlock_bh(&hslot->lock);
Eric Garverdf560052017-01-05 20:22:36 -0500293 cond_resched();
Eric Dumazet5781b232009-12-13 19:32:39 -0800294 } while (++first != last);
Eric Dumazet98322f22009-01-26 21:35:35 -0800295 goto fail;
Eric Dumazet645ca702008-10-29 01:41:45 -0700296 } else {
Eric Dumazetf86dcc52009-10-07 00:37:59 +0000297 hslot = udp_hashslot(udptable, net, snum);
Eric Dumazet645ca702008-10-29 01:41:45 -0700298 spin_lock_bh(&hslot->lock);
Eric Dumazet30fff922009-11-09 05:26:33 +0000299 if (hslot->count > 10) {
300 int exist;
301 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
302
303 slot2 &= udptable->mask;
304 hash2_nulladdr &= udptable->mask;
305
306 hslot2 = udp_hashslot2(udptable, slot2);
307 if (hslot->count < hslot2->count)
308 goto scan_primary_hash;
309
Josef Bacikfe38d2a2017-01-17 07:51:01 -0800310 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
Eric Dumazet30fff922009-11-09 05:26:33 +0000311 if (!exist && (hash2_nulladdr != slot2)) {
312 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
313 exist = udp_lib_lport_inuse2(net, snum, hslot2,
Josef Bacikfe38d2a2017-01-17 07:51:01 -0800314 sk);
Eric Dumazet30fff922009-11-09 05:26:33 +0000315 }
316 if (exist)
317 goto fail_unlock;
318 else
319 goto found;
320 }
321scan_primary_hash:
Josef Bacikfe38d2a2017-01-17 07:51:01 -0800322 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
Eric Dumazet645ca702008-10-29 01:41:45 -0700323 goto fail_unlock;
324 }
Eric Dumazet98322f22009-01-26 21:35:35 -0800325found:
Eric Dumazetc720c7e82009-10-15 06:30:45 +0000326 inet_sk(sk)->inet_num = snum;
Eric Dumazetd4cada42009-11-08 10:17:30 +0000327 udp_sk(sk)->udp_port_hash = snum;
328 udp_sk(sk)->udp_portaddr_hash ^= snum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 if (sk_unhashed(sk)) {
Craig Galleke32ea7e2016-01-04 17:41:46 -0500330 if (sk->sk_reuseport &&
Josef Bacikfe38d2a2017-01-17 07:51:01 -0800331 udp_reuseport_add_sock(sk, hslot)) {
Craig Galleke32ea7e2016-01-04 17:41:46 -0500332 inet_sk(sk)->inet_num = 0;
333 udp_sk(sk)->udp_port_hash = 0;
334 udp_sk(sk)->udp_portaddr_hash ^= snum;
335 goto fail_unlock;
336 }
337
Eric Dumazetca065d02016-04-01 08:52:13 -0700338 sk_add_node_rcu(sk, &hslot->head);
Eric Dumazetfdcc8aa92009-11-08 10:17:05 +0000339 hslot->count++;
Pavel Emelyanovc29a0bc2008-03-31 19:41:46 -0700340 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
Eric Dumazet512615b2009-11-08 10:17:58 +0000341
342 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
343 spin_lock(&hslot2->lock);
Craig Gallekd894ba12016-04-12 13:11:25 -0400344 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
David S. Miller1602f492016-04-23 18:26:24 -0400345 sk->sk_family == AF_INET6)
346 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
347 &hslot2->head);
Craig Gallekd894ba12016-04-12 13:11:25 -0400348 else
David S. Miller1602f492016-04-23 18:26:24 -0400349 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
350 &hslot2->head);
Eric Dumazet512615b2009-11-08 10:17:58 +0000351 hslot2->count++;
352 spin_unlock(&hslot2->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 }
Eric Dumazetca065d02016-04-01 08:52:13 -0700354 sock_set_flag(sk, SOCK_RCU_FREE);
Gerrit Renker25030a72006-08-26 20:06:05 -0700355 error = 0;
Eric Dumazet645ca702008-10-29 01:41:45 -0700356fail_unlock:
357 spin_unlock_bh(&hslot->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358fail:
Gerrit Renker25030a72006-08-26 20:06:05 -0700359 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360}
Eric Dumazetc482c562009-07-17 00:26:32 +0000361EXPORT_SYMBOL(udp_lib_get_port);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362
Eric Dumazet6eada012015-03-18 14:05:33 -0700363static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr,
364 unsigned int port)
Eric Dumazetd4cada42009-11-08 10:17:30 +0000365{
Eric Dumazet0eae88f2010-04-20 19:06:52 -0700366 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
Eric Dumazetd4cada42009-11-08 10:17:30 +0000367}
368
Pavel Emelyanov6ba5a3c2008-03-22 16:51:21 -0700369int udp_v4_get_port(struct sock *sk, unsigned short snum)
David S. Millerdb8dac22008-03-06 16:22:02 -0800370{
Eric Dumazet30fff922009-11-09 05:26:33 +0000371 unsigned int hash2_nulladdr =
Eric Dumazet0eae88f2010-04-20 19:06:52 -0700372 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
Eric Dumazet30fff922009-11-09 05:26:33 +0000373 unsigned int hash2_partial =
374 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
375
Eric Dumazetd4cada42009-11-08 10:17:30 +0000376 /* precompute partial secondary hash */
Eric Dumazet30fff922009-11-09 05:26:33 +0000377 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
Josef Bacikfe38d2a2017-01-17 07:51:01 -0800378 return udp_lib_get_port(sk, snum, hash2_nulladdr);
David S. Millerdb8dac22008-03-06 16:22:02 -0800379}
380
Su, Xuemind1e37282016-06-13 11:02:50 +0800381static int compute_score(struct sock *sk, struct net *net,
382 __be32 saddr, __be16 sport,
Robert Shearman63a6fff2017-01-26 18:02:24 +0000383 __be32 daddr, unsigned short hnum, int dif,
384 bool exact_dif)
Eric Dumazet645ca702008-10-29 01:41:45 -0700385{
Joe Perches60c04ae2014-12-01 20:29:06 -0800386 int score;
387 struct inet_sock *inet;
Eric Dumazet645ca702008-10-29 01:41:45 -0700388
Joe Perches60c04ae2014-12-01 20:29:06 -0800389 if (!net_eq(sock_net(sk), net) ||
390 udp_sk(sk)->udp_port_hash != hnum ||
391 ipv6_only_sock(sk))
392 return -1;
Eric Dumazet645ca702008-10-29 01:41:45 -0700393
Joe Perches60c04ae2014-12-01 20:29:06 -0800394 score = (sk->sk_family == PF_INET) ? 2 : 1;
395 inet = inet_sk(sk);
396
397 if (inet->inet_rcv_saddr) {
398 if (inet->inet_rcv_saddr != daddr)
399 return -1;
400 score += 4;
Eric Dumazet645ca702008-10-29 01:41:45 -0700401 }
Joe Perches60c04ae2014-12-01 20:29:06 -0800402
403 if (inet->inet_daddr) {
404 if (inet->inet_daddr != saddr)
405 return -1;
406 score += 4;
407 }
408
409 if (inet->inet_dport) {
410 if (inet->inet_dport != sport)
411 return -1;
412 score += 4;
413 }
414
Robert Shearman63a6fff2017-01-26 18:02:24 +0000415 if (sk->sk_bound_dev_if || exact_dif) {
Joe Perches60c04ae2014-12-01 20:29:06 -0800416 if (sk->sk_bound_dev_if != dif)
417 return -1;
418 score += 4;
419 }
Eric Dumazet70da2682015-10-08 19:33:21 -0700420 if (sk->sk_incoming_cpu == raw_smp_processor_id())
421 score++;
Eric Dumazet645ca702008-10-29 01:41:45 -0700422 return score;
423}
424
Eric Dumazet6eada012015-03-18 14:05:33 -0700425static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
426 const __u16 lport, const __be32 faddr,
427 const __be16 fport)
Hannes Frederic Sowa65cd8032013-10-19 21:48:51 +0200428{
Hannes Frederic Sowa1bbdcee2013-10-19 21:48:57 +0200429 static u32 udp_ehash_secret __read_mostly;
430
431 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
432
Hannes Frederic Sowa65cd8032013-10-19 21:48:51 +0200433 return __inet_ehashfn(laddr, lport, faddr, fport,
Hannes Frederic Sowa1bbdcee2013-10-19 21:48:57 +0200434 udp_ehash_secret + net_hash_mix(net));
Hannes Frederic Sowa65cd8032013-10-19 21:48:51 +0200435}
436
Su, Xuemind1e37282016-06-13 11:02:50 +0800437/* called with rcu_read_lock() */
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000438static struct sock *udp4_lib_lookup2(struct net *net,
439 __be32 saddr, __be16 sport,
Robert Shearman63a6fff2017-01-26 18:02:24 +0000440 __be32 daddr, unsigned int hnum, int dif, bool exact_dif,
Su, Xuemind1e37282016-06-13 11:02:50 +0800441 struct udp_hslot *hslot2,
Craig Gallek11341582016-01-05 15:08:07 -0500442 struct sk_buff *skb)
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000443{
444 struct sock *sk, *result;
Tom Herbertba418fa2013-01-22 09:50:32 +0000445 int score, badness, matches = 0, reuseport = 0;
446 u32 hash = 0;
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000447
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000448 result = NULL;
Tom Herbertba418fa2013-01-22 09:50:32 +0000449 badness = 0;
Eric Dumazetca065d02016-04-01 08:52:13 -0700450 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
Su, Xuemind1e37282016-06-13 11:02:50 +0800451 score = compute_score(sk, net, saddr, sport,
Robert Shearman63a6fff2017-01-26 18:02:24 +0000452 daddr, hnum, dif, exact_dif);
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000453 if (score > badness) {
Tom Herbertba418fa2013-01-22 09:50:32 +0000454 reuseport = sk->sk_reuseport;
455 if (reuseport) {
Hannes Frederic Sowa65cd8032013-10-19 21:48:51 +0200456 hash = udp_ehashfn(net, daddr, hnum,
457 saddr, sport);
Eric Dumazetca065d02016-04-01 08:52:13 -0700458 result = reuseport_select_sock(sk, hash, skb,
Eric Dumazeted0dfff2016-01-19 08:36:43 -0800459 sizeof(struct udphdr));
Eric Dumazetca065d02016-04-01 08:52:13 -0700460 if (result)
461 return result;
Tom Herbertba418fa2013-01-22 09:50:32 +0000462 matches = 1;
463 }
Eric Dumazetca065d02016-04-01 08:52:13 -0700464 badness = score;
465 result = sk;
Tom Herbertba418fa2013-01-22 09:50:32 +0000466 } else if (score == badness && reuseport) {
467 matches++;
Daniel Borkmann8fc54f62014-08-23 20:58:54 +0200468 if (reciprocal_scale(hash, matches) == 0)
Tom Herbertba418fa2013-01-22 09:50:32 +0000469 result = sk;
470 hash = next_pseudo_random32(hash);
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000471 }
472 }
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000473 return result;
474}
475
David S. Millerdb8dac22008-03-06 16:22:02 -0800476/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
477 * harder than this. -DaveM
478 */
Pavel Emelyanovfce82332011-12-09 06:23:34 +0000479struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
David S. Millerdb8dac22008-03-06 16:22:02 -0800480 __be16 sport, __be32 daddr, __be16 dport,
Craig Gallek538950a2016-01-04 17:41:47 -0500481 int dif, struct udp_table *udptable, struct sk_buff *skb)
David S. Millerdb8dac22008-03-06 16:22:02 -0800482{
Eric Dumazet271b72c2008-10-29 02:11:14 -0700483 struct sock *sk, *result;
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];
Robert Shearman63a6fff2017-01-26 18:02:24 +0000487 bool exact_dif = udp_lib_exact_dif_match(net, skb);
Tom Herbertba418fa2013-01-22 09:50:32 +0000488 int score, badness, matches = 0, reuseport = 0;
489 u32 hash = 0;
David S. Millerdb8dac22008-03-06 16:22:02 -0800490
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,
Robert Shearman63a6fff2017-01-26 18:02:24 +0000500 exact_dif, hslot2, skb);
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000501 if (!result) {
Su, Xuemind1e37282016-06-13 11:02:50 +0800502 unsigned int old_slot2 = slot2;
Eric Dumazet0eae88f2010-04-20 19:06:52 -0700503 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000504 slot2 = hash2 & udptable->mask;
Su, Xuemind1e37282016-06-13 11:02:50 +0800505 /* avoid searching the same slot again. */
506 if (unlikely(slot2 == old_slot2))
507 return result;
508
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000509 hslot2 = &udptable->hash2[slot2];
510 if (hslot->count < hslot2->count)
511 goto begin;
512
Jorge Boncompte [DTI2]1223c672010-04-08 04:56:48 +0000513 result = udp4_lib_lookup2(net, saddr, sport,
Su, Xuemind1e37282016-06-13 11:02:50 +0800514 daddr, hnum, dif,
Robert Shearman63a6fff2017-01-26 18:02:24 +0000515 exact_dif, hslot2, skb);
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000516 }
Eric Dumazet5051ebd2009-11-08 10:18:11 +0000517 return result;
518 }
Eric Dumazet271b72c2008-10-29 02:11:14 -0700519begin:
520 result = NULL;
Tom Herbertba418fa2013-01-22 09:50:32 +0000521 badness = 0;
Eric Dumazetca065d02016-04-01 08:52:13 -0700522 sk_for_each_rcu(sk, &hslot->head) {
Su, Xuemind1e37282016-06-13 11:02:50 +0800523 score = compute_score(sk, net, saddr, sport,
Robert Shearman63a6fff2017-01-26 18:02:24 +0000524 daddr, hnum, dif, exact_dif);
Eric Dumazet645ca702008-10-29 01:41:45 -0700525 if (score > badness) {
Tom Herbertba418fa2013-01-22 09:50:32 +0000526 reuseport = sk->sk_reuseport;
527 if (reuseport) {
Hannes Frederic Sowa65cd8032013-10-19 21:48:51 +0200528 hash = udp_ehashfn(net, daddr, hnum,
529 saddr, sport);
Eric Dumazetca065d02016-04-01 08:52:13 -0700530 result = reuseport_select_sock(sk, hash, skb,
Craig Gallek538950a2016-01-04 17:41:47 -0500531 sizeof(struct udphdr));
Eric Dumazetca065d02016-04-01 08:52:13 -0700532 if (result)
533 return result;
Tom Herbertba418fa2013-01-22 09:50:32 +0000534 matches = 1;
535 }
Eric Dumazetca065d02016-04-01 08:52:13 -0700536 result = sk;
537 badness = score;
Tom Herbertba418fa2013-01-22 09:50:32 +0000538 } else if (score == badness && reuseport) {
539 matches++;
Daniel Borkmann8fc54f62014-08-23 20:58:54 +0200540 if (reciprocal_scale(hash, matches) == 0)
Tom Herbertba418fa2013-01-22 09:50:32 +0000541 result = sk;
542 hash = next_pseudo_random32(hash);
David S. Millerdb8dac22008-03-06 16:22:02 -0800543 }
544 }
David S. Millerdb8dac22008-03-06 16:22:02 -0800545 return result;
546}
Pavel Emelyanovfce82332011-12-09 06:23:34 +0000547EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
David S. Millerdb8dac22008-03-06 16:22:02 -0800548
KOVACS Krisztian607c4aa2008-10-07 12:38:32 -0700549static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
550 __be16 sport, __be16 dport,
Eric Dumazet645ca702008-10-29 01:41:45 -0700551 struct udp_table *udptable)
KOVACS Krisztian607c4aa2008-10-07 12:38:32 -0700552{
553 const struct iphdr *iph = ip_hdr(skb);
554
Alexander Duycked7cbbc2016-05-12 16:23:44 -0700555 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
Eric Dumazet8afdd992013-12-10 18:07:23 -0800556 iph->daddr, dport, inet_iif(skb),
Craig Gallek538950a2016-01-04 17:41:47 -0500557 udptable, skb);
KOVACS Krisztian607c4aa2008-10-07 12:38:32 -0700558}
559
Tom Herbert63058302016-04-05 08:22:50 -0700560struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
561 __be16 sport, __be16 dport)
562{
Alexander Duycked7cbbc2016-05-12 16:23:44 -0700563 return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table);
Tom Herbert63058302016-04-05 08:22:50 -0700564}
565EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
566
Eric Dumazetca065d02016-04-01 08:52:13 -0700567/* Must be called under rcu_read_lock().
568 * Does increment socket refcount.
569 */
570#if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
Arnd Bergmann30f58152016-11-08 14:28:18 +0100571 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \
572 IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
KOVACS Krisztianbcd41302008-10-01 07:48:10 -0700573struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
574 __be32 daddr, __be16 dport, int dif)
575{
Eric Dumazetca065d02016-04-01 08:52:13 -0700576 struct sock *sk;
577
578 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
579 dif, &udp_table, NULL);
580 if (sk && !atomic_inc_not_zero(&sk->sk_refcnt))
581 sk = NULL;
582 return sk;
KOVACS Krisztianbcd41302008-10-01 07:48:10 -0700583}
584EXPORT_SYMBOL_GPL(udp4_lib_lookup);
Eric Dumazetca065d02016-04-01 08:52:13 -0700585#endif
KOVACS Krisztianbcd41302008-10-01 07:48:10 -0700586
Shawn Bohrer421b3882013-10-07 11:01:39 -0500587static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
588 __be16 loc_port, __be32 loc_addr,
589 __be16 rmt_port, __be32 rmt_addr,
590 int dif, unsigned short hnum)
591{
592 struct inet_sock *inet = inet_sk(sk);
593
594 if (!net_eq(sock_net(sk), net) ||
595 udp_sk(sk)->udp_port_hash != hnum ||
596 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
597 (inet->inet_dport != rmt_port && inet->inet_dport) ||
598 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
599 ipv6_only_sock(sk) ||
600 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
601 return false;
602 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif))
603 return false;
604 return true;
605}
606
David S. Millerdb8dac22008-03-06 16:22:02 -0800607/*
608 * This routine is called by the ICMP module when it gets some
609 * sort of error condition. If err < 0 then the socket should
610 * be closed and the error returned to the user. If err > 0
611 * it's just the icmp type << 8 | icmp code.
612 * Header points to the ip header of the error packet. We move
613 * on past this. Then (as it used to claim before adjustment)
614 * header points to the first 8 bytes of the udp header. We need
615 * to find the appropriate port.
616 */
617
Eric Dumazet645ca702008-10-29 01:41:45 -0700618void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
David S. Millerdb8dac22008-03-06 16:22:02 -0800619{
620 struct inet_sock *inet;
Eric Dumazetb71d1d42011-04-22 04:53:02 +0000621 const struct iphdr *iph = (const struct iphdr *)skb->data;
Eric Dumazetc482c562009-07-17 00:26:32 +0000622 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
David S. Millerdb8dac22008-03-06 16:22:02 -0800623 const int type = icmp_hdr(skb)->type;
624 const int code = icmp_hdr(skb)->code;
625 struct sock *sk;
626 int harderr;
627 int err;
Pavel Emelyanovfd54d712008-07-14 23:01:40 -0700628 struct net *net = dev_net(skb->dev);
David S. Millerdb8dac22008-03-06 16:22:02 -0800629
Pavel Emelyanovfd54d712008-07-14 23:01:40 -0700630 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
Craig Gallek538950a2016-01-04 17:41:47 -0500631 iph->saddr, uh->source, skb->dev->ifindex, udptable,
632 NULL);
Ian Morris51456b22015-04-03 09:17:26 +0100633 if (!sk) {
Eric Dumazet5d3848b2016-04-27 16:44:29 -0700634 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
David S. Millerdb8dac22008-03-06 16:22:02 -0800635 return; /* No socket for error */
636 }
637
638 err = 0;
639 harderr = 0;
640 inet = inet_sk(sk);
641
642 switch (type) {
643 default:
644 case ICMP_TIME_EXCEEDED:
645 err = EHOSTUNREACH;
646 break;
647 case ICMP_SOURCE_QUENCH:
648 goto out;
649 case ICMP_PARAMETERPROB:
650 err = EPROTO;
651 harderr = 1;
652 break;
653 case ICMP_DEST_UNREACH:
654 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
David S. Miller36393392012-06-14 22:21:46 -0700655 ipv4_sk_update_pmtu(skb, sk, info);
David S. Millerdb8dac22008-03-06 16:22:02 -0800656 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
657 err = EMSGSIZE;
658 harderr = 1;
659 break;
660 }
661 goto out;
662 }
663 err = EHOSTUNREACH;
664 if (code <= NR_ICMP_UNREACH) {
665 harderr = icmp_err_convert[code].fatal;
666 err = icmp_err_convert[code].errno;
667 }
668 break;
David S. Miller55be7a92012-07-11 21:27:49 -0700669 case ICMP_REDIRECT:
670 ipv4_sk_redirect(skb, sk);
Duan Jiong1a462d12013-09-20 18:20:28 +0800671 goto out;
David S. Millerdb8dac22008-03-06 16:22:02 -0800672 }
673
674 /*
675 * RFC1122: OK. Passes ICMP errors back to application, as per
676 * 4.1.3.3.
677 */
678 if (!inet->recverr) {
679 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
680 goto out;
Eric Dumazetb1faf562010-05-31 23:44:05 -0700681 } else
Eric Dumazetc482c562009-07-17 00:26:32 +0000682 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
Eric Dumazetb1faf562010-05-31 23:44:05 -0700683
David S. Millerdb8dac22008-03-06 16:22:02 -0800684 sk->sk_err = err;
685 sk->sk_error_report(sk);
686out:
Eric Dumazetca065d02016-04-01 08:52:13 -0700687 return;
David S. Millerdb8dac22008-03-06 16:22:02 -0800688}
689
690void udp_err(struct sk_buff *skb, u32 info)
691{
Eric Dumazet645ca702008-10-29 01:41:45 -0700692 __udp4_lib_err(skb, info, &udp_table);
David S. Millerdb8dac22008-03-06 16:22:02 -0800693}
694
695/*
696 * Throw away all pending data and cancel the corking. Socket is locked.
697 */
Denis V. Lunev36d926b2008-06-04 15:49:07 +0400698void udp_flush_pending_frames(struct sock *sk)
David S. Millerdb8dac22008-03-06 16:22:02 -0800699{
700 struct udp_sock *up = udp_sk(sk);
701
702 if (up->pending) {
703 up->len = 0;
704 up->pending = 0;
705 ip_flush_pending_frames(sk);
706 }
707}
Denis V. Lunev36d926b2008-06-04 15:49:07 +0400708EXPORT_SYMBOL(udp_flush_pending_frames);
David S. Millerdb8dac22008-03-06 16:22:02 -0800709
710/**
Herbert Xuf6b9664f2011-03-01 02:36:48 +0000711 * udp4_hwcsum - handle outgoing HW checksumming
David S. Millerdb8dac22008-03-06 16:22:02 -0800712 * @skb: sk_buff containing the filled-in UDP header
713 * (checksum field must be zeroed out)
Herbert Xuf6b9664f2011-03-01 02:36:48 +0000714 * @src: source IP address
715 * @dst: destination IP address
David S. Millerdb8dac22008-03-06 16:22:02 -0800716 */
Thomas Grafc26bf4a2013-07-25 18:12:18 +0200717void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
David S. Millerdb8dac22008-03-06 16:22:02 -0800718{
David S. Millerdb8dac22008-03-06 16:22:02 -0800719 struct udphdr *uh = udp_hdr(skb);
Herbert Xuf6b9664f2011-03-01 02:36:48 +0000720 int offset = skb_transport_offset(skb);
721 int len = skb->len - offset;
722 int hlen = len;
David S. Millerdb8dac22008-03-06 16:22:02 -0800723 __wsum csum = 0;
724
WANG Congebbe4952014-06-02 16:12:02 -0700725 if (!skb_has_frag_list(skb)) {
David S. Millerdb8dac22008-03-06 16:22:02 -0800726 /*
727 * Only one fragment on the socket.
728 */
729 skb->csum_start = skb_transport_header(skb) - skb->head;
730 skb->csum_offset = offsetof(struct udphdr, check);
Herbert Xuf6b9664f2011-03-01 02:36:48 +0000731 uh->check = ~csum_tcpudp_magic(src, dst, len,
732 IPPROTO_UDP, 0);
David S. Millerdb8dac22008-03-06 16:22:02 -0800733 } else {
WANG Congebbe4952014-06-02 16:12:02 -0700734 struct sk_buff *frags;
735
David S. Millerdb8dac22008-03-06 16:22:02 -0800736 /*
737 * HW-checksum won't work as there are two or more
738 * fragments on the socket so that all csums of sk_buffs
739 * should be together
740 */
WANG Congebbe4952014-06-02 16:12:02 -0700741 skb_walk_frags(skb, frags) {
Herbert Xuf6b9664f2011-03-01 02:36:48 +0000742 csum = csum_add(csum, frags->csum);
743 hlen -= frags->len;
WANG Congebbe4952014-06-02 16:12:02 -0700744 }
David S. Millerdb8dac22008-03-06 16:22:02 -0800745
Herbert Xuf6b9664f2011-03-01 02:36:48 +0000746 csum = skb_checksum(skb, offset, hlen, csum);
David S. Millerdb8dac22008-03-06 16:22:02 -0800747 skb->ip_summed = CHECKSUM_NONE;
748
David S. Millerdb8dac22008-03-06 16:22:02 -0800749 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
750 if (uh->check == 0)
751 uh->check = CSUM_MANGLED_0;
752 }
753}
Thomas Grafc26bf4a2013-07-25 18:12:18 +0200754EXPORT_SYMBOL_GPL(udp4_hwcsum);
David S. Millerdb8dac22008-03-06 16:22:02 -0800755
Tom Herbertaf5fcba2014-06-04 17:19:48 -0700756/* Function to set UDP checksum for an IPv4 UDP packet. This is intended
757 * for the simple case like when setting the checksum for a UDP tunnel.
758 */
759void udp_set_csum(bool nocheck, struct sk_buff *skb,
760 __be32 saddr, __be32 daddr, int len)
761{
762 struct udphdr *uh = udp_hdr(skb);
763
Edward Cree179bc672016-02-11 20:48:04 +0000764 if (nocheck) {
Tom Herbertaf5fcba2014-06-04 17:19:48 -0700765 uh->check = 0;
Edward Cree179bc672016-02-11 20:48:04 +0000766 } else if (skb_is_gso(skb)) {
Tom Herbertaf5fcba2014-06-04 17:19:48 -0700767 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
Edward Cree179bc672016-02-11 20:48:04 +0000768 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
769 uh->check = 0;
770 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
771 if (uh->check == 0)
772 uh->check = CSUM_MANGLED_0;
Edward Creed75f1302016-02-11 20:49:40 +0000773 } else {
Tom Herbertaf5fcba2014-06-04 17:19:48 -0700774 skb->ip_summed = CHECKSUM_PARTIAL;
775 skb->csum_start = skb_transport_header(skb) - skb->head;
776 skb->csum_offset = offsetof(struct udphdr, check);
777 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
Tom Herbertaf5fcba2014-06-04 17:19:48 -0700778 }
779}
780EXPORT_SYMBOL(udp_set_csum);
781
David S. Miller79ab0532011-05-09 13:31:04 -0700782static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
Herbert Xuf6b9664f2011-03-01 02:36:48 +0000783{
784 struct sock *sk = skb->sk;
785 struct inet_sock *inet = inet_sk(sk);
786 struct udphdr *uh;
Herbert Xuf6b9664f2011-03-01 02:36:48 +0000787 int err = 0;
788 int is_udplite = IS_UDPLITE(sk);
789 int offset = skb_transport_offset(skb);
790 int len = skb->len - offset;
791 __wsum csum = 0;
792
793 /*
794 * Create a UDP header
795 */
796 uh = udp_hdr(skb);
797 uh->source = inet->inet_sport;
David S. Miller79ab0532011-05-09 13:31:04 -0700798 uh->dest = fl4->fl4_dport;
Herbert Xuf6b9664f2011-03-01 02:36:48 +0000799 uh->len = htons(len);
800 uh->check = 0;
801
802 if (is_udplite) /* UDP-Lite */
803 csum = udplite_csum(skb);
804
Tom Herbert28448b82014-05-23 08:47:19 -0700805 else if (sk->sk_no_check_tx) { /* UDP csum disabled */
Herbert Xuf6b9664f2011-03-01 02:36:48 +0000806
807 skb->ip_summed = CHECKSUM_NONE;
808 goto send;
809
810 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
811
David S. Miller79ab0532011-05-09 13:31:04 -0700812 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
Herbert Xuf6b9664f2011-03-01 02:36:48 +0000813 goto send;
814
815 } else
816 csum = udp_csum(skb);
817
818 /* add protocol-dependent pseudo-header */
David S. Miller79ab0532011-05-09 13:31:04 -0700819 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
Herbert Xuf6b9664f2011-03-01 02:36:48 +0000820 sk->sk_protocol, csum);
821 if (uh->check == 0)
822 uh->check = CSUM_MANGLED_0;
823
824send:
Eric Dumazetb5ec8ee2012-08-10 02:22:47 +0000825 err = ip_send_skb(sock_net(sk), skb);
Herbert Xuf6b9664f2011-03-01 02:36:48 +0000826 if (err) {
827 if (err == -ENOBUFS && !inet->recverr) {
Eric Dumazet6aef70a2016-04-27 16:44:27 -0700828 UDP_INC_STATS(sock_net(sk),
829 UDP_MIB_SNDBUFERRORS, is_udplite);
Herbert Xuf6b9664f2011-03-01 02:36:48 +0000830 err = 0;
831 }
832 } else
Eric Dumazet6aef70a2016-04-27 16:44:27 -0700833 UDP_INC_STATS(sock_net(sk),
834 UDP_MIB_OUTDATAGRAMS, is_udplite);
Herbert Xuf6b9664f2011-03-01 02:36:48 +0000835 return err;
836}
837
David S. Millerdb8dac22008-03-06 16:22:02 -0800838/*
839 * Push out all pending data as one UDP datagram. Socket is locked.
840 */
Hannes Frederic Sowa8822b642013-07-01 20:21:30 +0200841int udp_push_pending_frames(struct sock *sk)
David S. Millerdb8dac22008-03-06 16:22:02 -0800842{
843 struct udp_sock *up = udp_sk(sk);
844 struct inet_sock *inet = inet_sk(sk);
David S. Millerb6f21b22011-03-12 02:09:18 -0500845 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
David S. Millerdb8dac22008-03-06 16:22:02 -0800846 struct sk_buff *skb;
David S. Millerdb8dac22008-03-06 16:22:02 -0800847 int err = 0;
David S. Millerdb8dac22008-03-06 16:22:02 -0800848
David S. Miller77968b72011-05-08 17:12:19 -0700849 skb = ip_finish_skb(sk, fl4);
Herbert Xuf6b9664f2011-03-01 02:36:48 +0000850 if (!skb)
David S. Millerdb8dac22008-03-06 16:22:02 -0800851 goto out;
852
David S. Miller79ab0532011-05-09 13:31:04 -0700853 err = udp_send_skb(skb, fl4);
David S. Millerdb8dac22008-03-06 16:22:02 -0800854
David S. Millerdb8dac22008-03-06 16:22:02 -0800855out:
856 up->len = 0;
857 up->pending = 0;
David S. Millerdb8dac22008-03-06 16:22:02 -0800858 return err;
859}
Hannes Frederic Sowa8822b642013-07-01 20:21:30 +0200860EXPORT_SYMBOL(udp_push_pending_frames);
David S. Millerdb8dac22008-03-06 16:22:02 -0800861
Ying Xue1b784142015-03-02 15:37:48 +0800862int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
David S. Millerdb8dac22008-03-06 16:22:02 -0800863{
864 struct inet_sock *inet = inet_sk(sk);
865 struct udp_sock *up = udp_sk(sk);
David S. Millere4749952011-05-08 16:38:45 -0700866 struct flowi4 fl4_stack;
David S. Millerb6f21b22011-03-12 02:09:18 -0500867 struct flowi4 *fl4;
David S. Millerdb8dac22008-03-06 16:22:02 -0800868 int ulen = len;
869 struct ipcm_cookie ipc;
870 struct rtable *rt = NULL;
871 int free = 0;
872 int connected = 0;
873 __be32 daddr, faddr, saddr;
874 __be16 dport;
875 u8 tos;
876 int err, is_udplite = IS_UDPLITE(sk);
877 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
878 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
Herbert Xu903ab862011-03-01 02:36:48 +0000879 struct sk_buff *skb;
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000880 struct ip_options_data opt_copy;
David S. Millerdb8dac22008-03-06 16:22:02 -0800881
882 if (len > 0xFFFF)
883 return -EMSGSIZE;
884
885 /*
886 * Check the flags.
887 */
888
Eric Dumazetc482c562009-07-17 00:26:32 +0000889 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
David S. Millerdb8dac22008-03-06 16:22:02 -0800890 return -EOPNOTSUPP;
891
892 ipc.opt = NULL;
Oliver Hartkopp2244d072010-08-17 08:59:14 +0000893 ipc.tx_flags = 0;
Francesco Fuscoaa661582013-09-24 15:43:09 +0200894 ipc.ttl = 0;
895 ipc.tos = -1;
David S. Millerdb8dac22008-03-06 16:22:02 -0800896
Herbert Xu903ab862011-03-01 02:36:48 +0000897 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
898
David S. Millerf5fca602011-05-08 17:24:10 -0700899 fl4 = &inet->cork.fl.u.ip4;
David S. Millerdb8dac22008-03-06 16:22:02 -0800900 if (up->pending) {
901 /*
902 * There are pending frames.
903 * The socket lock must be held while it's corked.
904 */
905 lock_sock(sk);
906 if (likely(up->pending)) {
907 if (unlikely(up->pending != AF_INET)) {
908 release_sock(sk);
909 return -EINVAL;
910 }
911 goto do_append_data;
912 }
913 release_sock(sk);
914 }
915 ulen += sizeof(struct udphdr);
916
917 /*
918 * Get and verify the address.
919 */
920 if (msg->msg_name) {
Steffen Hurrle342dfc32014-01-17 22:53:15 +0100921 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
David S. Millerdb8dac22008-03-06 16:22:02 -0800922 if (msg->msg_namelen < sizeof(*usin))
923 return -EINVAL;
924 if (usin->sin_family != AF_INET) {
925 if (usin->sin_family != AF_UNSPEC)
926 return -EAFNOSUPPORT;
927 }
928
929 daddr = usin->sin_addr.s_addr;
930 dport = usin->sin_port;
931 if (dport == 0)
932 return -EINVAL;
933 } else {
934 if (sk->sk_state != TCP_ESTABLISHED)
935 return -EDESTADDRREQ;
Eric Dumazetc720c7e82009-10-15 06:30:45 +0000936 daddr = inet->inet_daddr;
937 dport = inet->inet_dport;
David S. Millerdb8dac22008-03-06 16:22:02 -0800938 /* Open fast path for connected socket.
939 Route will not be used, if at least one option is set.
940 */
941 connected = 1;
942 }
Soheil Hassas Yeganehc14ac942016-04-02 23:08:12 -0400943
944 ipc.sockc.tsflags = sk->sk_tsflags;
Eric Dumazetc720c7e82009-10-15 06:30:45 +0000945 ipc.addr = inet->inet_saddr;
David S. Millerdb8dac22008-03-06 16:22:02 -0800946 ipc.oif = sk->sk_bound_dev_if;
Daniel Borkmannbf84a012013-04-14 08:08:13 +0000947
David S. Millerdb8dac22008-03-06 16:22:02 -0800948 if (msg->msg_controllen) {
Soheil Hassas Yeganeh24025c42016-04-02 23:08:10 -0400949 err = ip_cmsg_send(sk, msg, &ipc, sk->sk_family == AF_INET6);
Eric Dumazet91948302016-02-04 06:23:28 -0800950 if (unlikely(err)) {
951 kfree(ipc.opt);
David S. Millerdb8dac22008-03-06 16:22:02 -0800952 return err;
Eric Dumazet91948302016-02-04 06:23:28 -0800953 }
David S. Millerdb8dac22008-03-06 16:22:02 -0800954 if (ipc.opt)
955 free = 1;
956 connected = 0;
957 }
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000958 if (!ipc.opt) {
959 struct ip_options_rcu *inet_opt;
960
961 rcu_read_lock();
962 inet_opt = rcu_dereference(inet->inet_opt);
963 if (inet_opt) {
964 memcpy(&opt_copy, inet_opt,
965 sizeof(*inet_opt) + inet_opt->opt.optlen);
966 ipc.opt = &opt_copy.opt;
967 }
968 rcu_read_unlock();
969 }
David S. Millerdb8dac22008-03-06 16:22:02 -0800970
971 saddr = ipc.addr;
972 ipc.addr = faddr = daddr;
973
Soheil Hassas Yeganehc14ac942016-04-02 23:08:12 -0400974 sock_tx_timestamp(sk, ipc.sockc.tsflags, &ipc.tx_flags);
975
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000976 if (ipc.opt && ipc.opt->opt.srr) {
David S. Millerdb8dac22008-03-06 16:22:02 -0800977 if (!daddr)
978 return -EINVAL;
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000979 faddr = ipc.opt->opt.faddr;
David S. Millerdb8dac22008-03-06 16:22:02 -0800980 connected = 0;
981 }
Francesco Fuscoaa661582013-09-24 15:43:09 +0200982 tos = get_rttos(&ipc, inet);
David S. Millerdb8dac22008-03-06 16:22:02 -0800983 if (sock_flag(sk, SOCK_LOCALROUTE) ||
984 (msg->msg_flags & MSG_DONTROUTE) ||
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000985 (ipc.opt && ipc.opt->opt.is_strictroute)) {
David S. Millerdb8dac22008-03-06 16:22:02 -0800986 tos |= RTO_ONLINK;
987 connected = 0;
988 }
989
990 if (ipv4_is_multicast(daddr)) {
991 if (!ipc.oif)
992 ipc.oif = inet->mc_index;
993 if (!saddr)
994 saddr = inet->mc_addr;
995 connected = 0;
Erich E. Hoover76e21052012-02-08 09:11:07 +0000996 } else if (!ipc.oif)
997 ipc.oif = inet->uc_index;
David S. Millerdb8dac22008-03-06 16:22:02 -0800998
999 if (connected)
Eric Dumazetc482c562009-07-17 00:26:32 +00001000 rt = (struct rtable *)sk_dst_check(sk, 0);
David S. Millerdb8dac22008-03-06 16:22:02 -08001001
Ian Morris51456b22015-04-03 09:17:26 +01001002 if (!rt) {
Pavel Emelyanov84a3aa02008-07-16 20:19:08 -07001003 struct net *net = sock_net(sk);
David Ahern9a24abf2015-08-13 14:59:03 -06001004 __u8 flow_flags = inet_sk_flowi_flags(sk);
Pavel Emelyanov84a3aa02008-07-16 20:19:08 -07001005
David S. Millere4749952011-05-08 16:38:45 -07001006 fl4 = &fl4_stack;
David Ahern9a24abf2015-08-13 14:59:03 -06001007
David S. Millere4749952011-05-08 16:38:45 -07001008 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
David S. Millerc0951cb2011-03-31 04:54:27 -07001009 RT_SCOPE_UNIVERSE, sk->sk_protocol,
David Ahern9a24abf2015-08-13 14:59:03 -06001010 flow_flags,
Lorenzo Colittie2d118a2016-11-04 02:23:43 +09001011 faddr, saddr, dport, inet->inet_sport,
1012 sk->sk_uid);
David S. Millerc0951cb2011-03-31 04:54:27 -07001013
David S. Millere4749952011-05-08 16:38:45 -07001014 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1015 rt = ip_route_output_flow(net, fl4, sk);
David S. Millerb23dd4f2011-03-02 14:31:35 -08001016 if (IS_ERR(rt)) {
1017 err = PTR_ERR(rt);
David S. Miller06dc94b2011-03-03 10:38:01 -08001018 rt = NULL;
David S. Millerdb8dac22008-03-06 16:22:02 -08001019 if (err == -ENETUNREACH)
Eric Dumazetf1d8cba2013-11-28 09:51:22 -08001020 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
David S. Millerdb8dac22008-03-06 16:22:02 -08001021 goto out;
1022 }
1023
1024 err = -EACCES;
1025 if ((rt->rt_flags & RTCF_BROADCAST) &&
1026 !sock_flag(sk, SOCK_BROADCAST))
1027 goto out;
1028 if (connected)
Changli Gaod8d1f302010-06-10 23:31:35 -07001029 sk_dst_set(sk, dst_clone(&rt->dst));
David S. Millerdb8dac22008-03-06 16:22:02 -08001030 }
1031
1032 if (msg->msg_flags&MSG_CONFIRM)
1033 goto do_confirm;
1034back_from_confirm:
1035
David S. Millere4749952011-05-08 16:38:45 -07001036 saddr = fl4->saddr;
David S. Millerdb8dac22008-03-06 16:22:02 -08001037 if (!ipc.addr)
David S. Millere4749952011-05-08 16:38:45 -07001038 daddr = ipc.addr = fl4->daddr;
David S. Millerdb8dac22008-03-06 16:22:02 -08001039
Herbert Xu903ab862011-03-01 02:36:48 +00001040 /* Lockless fast path for the non-corking case. */
1041 if (!corkreq) {
Al Virof69e6d12014-11-24 13:23:40 -05001042 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
Herbert Xu903ab862011-03-01 02:36:48 +00001043 sizeof(struct udphdr), &ipc, &rt,
1044 msg->msg_flags);
1045 err = PTR_ERR(skb);
YOSHIFUJI Hideaki / 吉藤英明50c3a482013-01-22 06:32:49 +00001046 if (!IS_ERR_OR_NULL(skb))
David S. Miller79ab0532011-05-09 13:31:04 -07001047 err = udp_send_skb(skb, fl4);
Herbert Xu903ab862011-03-01 02:36:48 +00001048 goto out;
1049 }
1050
David S. Millerdb8dac22008-03-06 16:22:02 -08001051 lock_sock(sk);
1052 if (unlikely(up->pending)) {
1053 /* The socket is already corked while preparing it. */
1054 /* ... which is an evident application bug. --ANK */
1055 release_sock(sk);
1056
Joe Perchesba7a46f2014-11-11 10:59:17 -08001057 net_dbg_ratelimited("cork app bug 2\n");
David S. Millerdb8dac22008-03-06 16:22:02 -08001058 err = -EINVAL;
1059 goto out;
1060 }
1061 /*
1062 * Now cork the socket to pend data.
1063 */
David S. Millerb6f21b22011-03-12 02:09:18 -05001064 fl4 = &inet->cork.fl.u.ip4;
1065 fl4->daddr = daddr;
1066 fl4->saddr = saddr;
David S. Miller9cce96d2011-03-12 03:00:33 -05001067 fl4->fl4_dport = dport;
1068 fl4->fl4_sport = inet->inet_sport;
David S. Millerdb8dac22008-03-06 16:22:02 -08001069 up->pending = AF_INET;
1070
1071do_append_data:
1072 up->len += ulen;
Al Virof69e6d12014-11-24 13:23:40 -05001073 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
David S. Millerf5fca602011-05-08 17:24:10 -07001074 sizeof(struct udphdr), &ipc, &rt,
1075 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
David S. Millerdb8dac22008-03-06 16:22:02 -08001076 if (err)
1077 udp_flush_pending_frames(sk);
1078 else if (!corkreq)
1079 err = udp_push_pending_frames(sk);
1080 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1081 up->pending = 0;
1082 release_sock(sk);
1083
1084out:
1085 ip_rt_put(rt);
1086 if (free)
1087 kfree(ipc.opt);
1088 if (!err)
1089 return len;
1090 /*
1091 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1092 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1093 * we don't have a good statistic (IpOutDiscards but it can be too many
1094 * things). We could add another new stat but at least for now that
1095 * seems like overkill.
1096 */
1097 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
Eric Dumazet6aef70a2016-04-27 16:44:27 -07001098 UDP_INC_STATS(sock_net(sk),
1099 UDP_MIB_SNDBUFERRORS, is_udplite);
David S. Millerdb8dac22008-03-06 16:22:02 -08001100 }
1101 return err;
1102
1103do_confirm:
Julian Anastasov0dec8792017-02-06 23:14:16 +02001104 if (msg->msg_flags & MSG_PROBE)
1105 dst_confirm_neigh(&rt->dst, &fl4->daddr);
David S. Millerdb8dac22008-03-06 16:22:02 -08001106 if (!(msg->msg_flags&MSG_PROBE) || len)
1107 goto back_from_confirm;
1108 err = 0;
1109 goto out;
1110}
Eric Dumazetc482c562009-07-17 00:26:32 +00001111EXPORT_SYMBOL(udp_sendmsg);
David S. Millerdb8dac22008-03-06 16:22:02 -08001112
1113int udp_sendpage(struct sock *sk, struct page *page, int offset,
1114 size_t size, int flags)
1115{
David S. Millerf5fca602011-05-08 17:24:10 -07001116 struct inet_sock *inet = inet_sk(sk);
David S. Millerdb8dac22008-03-06 16:22:02 -08001117 struct udp_sock *up = udp_sk(sk);
1118 int ret;
1119
Shawn Landdend3f7d562013-11-24 22:36:28 -08001120 if (flags & MSG_SENDPAGE_NOTLAST)
1121 flags |= MSG_MORE;
1122
David S. Millerdb8dac22008-03-06 16:22:02 -08001123 if (!up->pending) {
1124 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1125
1126 /* Call udp_sendmsg to specify destination address which
1127 * sendpage interface can't pass.
1128 * This will succeed only when the socket is connected.
1129 */
Ying Xue1b784142015-03-02 15:37:48 +08001130 ret = udp_sendmsg(sk, &msg, 0);
David S. Millerdb8dac22008-03-06 16:22:02 -08001131 if (ret < 0)
1132 return ret;
1133 }
1134
1135 lock_sock(sk);
1136
1137 if (unlikely(!up->pending)) {
1138 release_sock(sk);
1139
Joe Perchesba7a46f2014-11-11 10:59:17 -08001140 net_dbg_ratelimited("udp cork app bug 3\n");
David S. Millerdb8dac22008-03-06 16:22:02 -08001141 return -EINVAL;
1142 }
1143
David S. Millerf5fca602011-05-08 17:24:10 -07001144 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1145 page, offset, size, flags);
David S. Millerdb8dac22008-03-06 16:22:02 -08001146 if (ret == -EOPNOTSUPP) {
1147 release_sock(sk);
1148 return sock_no_sendpage(sk->sk_socket, page, offset,
1149 size, flags);
1150 }
1151 if (ret < 0) {
1152 udp_flush_pending_frames(sk);
1153 goto out;
1154 }
1155
1156 up->len += size;
1157 if (!(up->corkflag || (flags&MSG_MORE)))
1158 ret = udp_push_pending_frames(sk);
1159 if (!ret)
1160 ret = size;
1161out:
1162 release_sock(sk);
1163 return ret;
1164}
1165
Paolo Abenib65ac442017-06-12 11:23:43 +02001166/* Copy as much information as possible into skb->dev_scratch to avoid
1167 * possibly multiple cache miss on dequeue();
1168 */
1169#if BITS_PER_LONG == 64
1170
1171/* we can store multiple info here: truesize, len and the bit needed to
1172 * compute skb_csum_unnecessary will be on cold cache lines at recvmsg
1173 * time.
1174 * skb->len can be stored on 16 bits since the udp header has been already
1175 * validated and pulled.
1176 */
1177struct udp_dev_scratch {
1178 u32 truesize;
1179 u16 len;
1180 bool is_linear;
1181 bool csum_unnecessary;
1182};
1183
1184static void udp_set_dev_scratch(struct sk_buff *skb)
1185{
1186 struct udp_dev_scratch *scratch;
1187
1188 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1189 scratch = (struct udp_dev_scratch *)&skb->dev_scratch;
1190 scratch->truesize = skb->truesize;
1191 scratch->len = skb->len;
1192 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1193 scratch->is_linear = !skb_is_nonlinear(skb);
1194}
1195
1196static int udp_skb_truesize(struct sk_buff *skb)
1197{
1198 return ((struct udp_dev_scratch *)&skb->dev_scratch)->truesize;
1199}
1200
1201static unsigned int udp_skb_len(struct sk_buff *skb)
1202{
1203 return ((struct udp_dev_scratch *)&skb->dev_scratch)->len;
1204}
1205
1206static bool udp_skb_csum_unnecessary(struct sk_buff *skb)
1207{
1208 return ((struct udp_dev_scratch *)&skb->dev_scratch)->csum_unnecessary;
1209}
1210
1211static bool udp_skb_is_linear(struct sk_buff *skb)
1212{
1213 return ((struct udp_dev_scratch *)&skb->dev_scratch)->is_linear;
1214}
1215
1216#else
1217static void udp_set_dev_scratch(struct sk_buff *skb)
1218{
1219 skb->dev_scratch = skb->truesize;
1220}
1221
1222static int udp_skb_truesize(struct sk_buff *skb)
1223{
1224 return skb->dev_scratch;
1225}
1226
1227static unsigned int udp_skb_len(struct sk_buff *skb)
1228{
1229 return skb->len;
1230}
1231
1232static bool udp_skb_csum_unnecessary(struct sk_buff *skb)
1233{
1234 return skb_csum_unnecessary(skb);
1235}
1236
1237static bool udp_skb_is_linear(struct sk_buff *skb)
1238{
1239 return !skb_is_nonlinear(skb);
1240}
1241#endif
1242
Paolo Abeni7c13f972016-11-04 11:28:59 +01001243/* fully reclaim rmem/fwd memory allocated for skb */
Paolo Abeni6dfb4362017-05-16 11:20:15 +02001244static void udp_rmem_release(struct sock *sk, int size, int partial,
1245 bool rx_queue_lock_held)
Paolo Abenif970bd92016-10-21 13:55:46 +02001246{
Eric Dumazet6b229cf2016-12-08 11:41:56 -08001247 struct udp_sock *up = udp_sk(sk);
Paolo Abeni2276f582017-05-16 11:20:14 +02001248 struct sk_buff_head *sk_queue;
Paolo Abenif970bd92016-10-21 13:55:46 +02001249 int amt;
1250
Eric Dumazet6b229cf2016-12-08 11:41:56 -08001251 if (likely(partial)) {
1252 up->forward_deficit += size;
1253 size = up->forward_deficit;
1254 if (size < (sk->sk_rcvbuf >> 2) &&
Paolo Abeni2276f582017-05-16 11:20:14 +02001255 !skb_queue_empty(&up->reader_queue))
Eric Dumazet6b229cf2016-12-08 11:41:56 -08001256 return;
1257 } else {
1258 size += up->forward_deficit;
1259 }
1260 up->forward_deficit = 0;
1261
Paolo Abeni6dfb4362017-05-16 11:20:15 +02001262 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1263 * if the called don't held it already
1264 */
Paolo Abeni2276f582017-05-16 11:20:14 +02001265 sk_queue = &sk->sk_receive_queue;
Paolo Abeni6dfb4362017-05-16 11:20:15 +02001266 if (!rx_queue_lock_held)
1267 spin_lock(&sk_queue->lock);
1268
Paolo Abeni2276f582017-05-16 11:20:14 +02001269
Paolo Abenif970bd92016-10-21 13:55:46 +02001270 sk->sk_forward_alloc += size;
1271 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1272 sk->sk_forward_alloc -= amt;
Paolo Abenif970bd92016-10-21 13:55:46 +02001273
1274 if (amt)
1275 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
Eric Dumazet02ab0d12016-12-08 11:41:57 -08001276
1277 atomic_sub(size, &sk->sk_rmem_alloc);
Paolo Abeni2276f582017-05-16 11:20:14 +02001278
1279 /* this can save us from acquiring the rx queue lock on next receive */
1280 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1281
Paolo Abeni6dfb4362017-05-16 11:20:15 +02001282 if (!rx_queue_lock_held)
1283 spin_unlock(&sk_queue->lock);
Paolo Abenif970bd92016-10-21 13:55:46 +02001284}
1285
Paolo Abeni2276f582017-05-16 11:20:14 +02001286/* Note: called with reader_queue.lock held.
Eric Dumazetc84d9492016-12-08 11:41:55 -08001287 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1288 * This avoids a cache line miss while receive_queue lock is held.
1289 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1290 */
Paolo Abeni7c13f972016-11-04 11:28:59 +01001291void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
Paolo Abenif970bd92016-10-21 13:55:46 +02001292{
Paolo Abenib65ac442017-06-12 11:23:43 +02001293 prefetch(&skb->data);
1294 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
Paolo Abenif970bd92016-10-21 13:55:46 +02001295}
Paolo Abeni7c13f972016-11-04 11:28:59 +01001296EXPORT_SYMBOL(udp_skb_destructor);
Paolo Abenif970bd92016-10-21 13:55:46 +02001297
Paolo Abeni6dfb4362017-05-16 11:20:15 +02001298/* as above, but the caller held the rx queue lock, too */
Colin Ian King64f51022017-05-17 09:50:36 +01001299static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
Paolo Abeni6dfb4362017-05-16 11:20:15 +02001300{
Paolo Abenib65ac442017-06-12 11:23:43 +02001301 prefetch(&skb->data);
1302 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
Paolo Abeni6dfb4362017-05-16 11:20:15 +02001303}
1304
Eric Dumazet4b272752016-12-08 11:41:54 -08001305/* Idea of busylocks is to let producers grab an extra spinlock
1306 * to relieve pressure on the receive_queue spinlock shared by consumer.
1307 * Under flood, this means that only one producer can be in line
1308 * trying to acquire the receive_queue spinlock.
1309 * These busylock can be allocated on a per cpu manner, instead of a
1310 * per socket one (that would consume a cache line per socket)
1311 */
1312static int udp_busylocks_log __read_mostly;
1313static spinlock_t *udp_busylocks __read_mostly;
1314
1315static spinlock_t *busylock_acquire(void *ptr)
1316{
1317 spinlock_t *busy;
1318
1319 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1320 spin_lock(busy);
1321 return busy;
1322}
1323
1324static void busylock_release(spinlock_t *busy)
1325{
1326 if (busy)
1327 spin_unlock(busy);
1328}
1329
Paolo Abenif970bd92016-10-21 13:55:46 +02001330int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1331{
1332 struct sk_buff_head *list = &sk->sk_receive_queue;
1333 int rmem, delta, amt, err = -ENOMEM;
Eric Dumazet4b272752016-12-08 11:41:54 -08001334 spinlock_t *busy = NULL;
Eric Dumazetc8c8b122016-12-07 09:19:33 -08001335 int size;
Paolo Abenif970bd92016-10-21 13:55:46 +02001336
1337 /* try to avoid the costly atomic add/sub pair when the receive
1338 * queue is full; always allow at least a packet
1339 */
1340 rmem = atomic_read(&sk->sk_rmem_alloc);
Paolo Abeni363dc732016-12-02 17:35:49 +01001341 if (rmem > sk->sk_rcvbuf)
Paolo Abenif970bd92016-10-21 13:55:46 +02001342 goto drop;
1343
Eric Dumazetc8c8b122016-12-07 09:19:33 -08001344 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1345 * having linear skbs :
1346 * - Reduce memory overhead and thus increase receive queue capacity
1347 * - Less cache line misses at copyout() time
1348 * - Less work at consume_skb() (less alien page frag freeing)
1349 */
Eric Dumazet4b272752016-12-08 11:41:54 -08001350 if (rmem > (sk->sk_rcvbuf >> 1)) {
Eric Dumazetc8c8b122016-12-07 09:19:33 -08001351 skb_condense(skb);
Eric Dumazet4b272752016-12-08 11:41:54 -08001352
1353 busy = busylock_acquire(sk);
1354 }
Eric Dumazetc8c8b122016-12-07 09:19:33 -08001355 size = skb->truesize;
Paolo Abenib65ac442017-06-12 11:23:43 +02001356 udp_set_dev_scratch(skb);
Eric Dumazetc8c8b122016-12-07 09:19:33 -08001357
Paolo Abenif970bd92016-10-21 13:55:46 +02001358 /* we drop only if the receive buf is full and the receive
1359 * queue contains some other skb
1360 */
1361 rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
Paolo Abeni363dc732016-12-02 17:35:49 +01001362 if (rmem > (size + sk->sk_rcvbuf))
Paolo Abenif970bd92016-10-21 13:55:46 +02001363 goto uncharge_drop;
1364
1365 spin_lock(&list->lock);
1366 if (size >= sk->sk_forward_alloc) {
1367 amt = sk_mem_pages(size);
1368 delta = amt << SK_MEM_QUANTUM_SHIFT;
1369 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1370 err = -ENOBUFS;
1371 spin_unlock(&list->lock);
1372 goto uncharge_drop;
1373 }
1374
1375 sk->sk_forward_alloc += delta;
1376 }
1377
1378 sk->sk_forward_alloc -= size;
1379
Paolo Abeni7c13f972016-11-04 11:28:59 +01001380 /* no need to setup a destructor, we will explicitly release the
1381 * forward allocated memory on dequeue
1382 */
Paolo Abenif970bd92016-10-21 13:55:46 +02001383 sock_skb_set_dropcount(sk, skb);
1384
1385 __skb_queue_tail(list, skb);
1386 spin_unlock(&list->lock);
1387
1388 if (!sock_flag(sk, SOCK_DEAD))
1389 sk->sk_data_ready(sk);
1390
Eric Dumazet4b272752016-12-08 11:41:54 -08001391 busylock_release(busy);
Paolo Abenif970bd92016-10-21 13:55:46 +02001392 return 0;
1393
1394uncharge_drop:
1395 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1396
1397drop:
1398 atomic_inc(&sk->sk_drops);
Eric Dumazet4b272752016-12-08 11:41:54 -08001399 busylock_release(busy);
Paolo Abenif970bd92016-10-21 13:55:46 +02001400 return err;
1401}
1402EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1403
Paolo Abenic915fe12016-11-15 16:37:53 +01001404void udp_destruct_sock(struct sock *sk)
Paolo Abenif970bd92016-10-21 13:55:46 +02001405{
1406 /* reclaim completely the forward allocated memory */
Paolo Abeni2276f582017-05-16 11:20:14 +02001407 struct udp_sock *up = udp_sk(sk);
Paolo Abeni7c13f972016-11-04 11:28:59 +01001408 unsigned int total = 0;
1409 struct sk_buff *skb;
1410
Paolo Abeni2276f582017-05-16 11:20:14 +02001411 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1412 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
Paolo Abeni7c13f972016-11-04 11:28:59 +01001413 total += skb->truesize;
1414 kfree_skb(skb);
1415 }
Paolo Abeni6dfb4362017-05-16 11:20:15 +02001416 udp_rmem_release(sk, total, 0, true);
Paolo Abeni7c13f972016-11-04 11:28:59 +01001417
Paolo Abenif970bd92016-10-21 13:55:46 +02001418 inet_sock_destruct(sk);
1419}
Paolo Abenic915fe12016-11-15 16:37:53 +01001420EXPORT_SYMBOL_GPL(udp_destruct_sock);
Paolo Abenif970bd92016-10-21 13:55:46 +02001421
1422int udp_init_sock(struct sock *sk)
1423{
Paolo Abeni2276f582017-05-16 11:20:14 +02001424 skb_queue_head_init(&udp_sk(sk)->reader_queue);
Paolo Abenif970bd92016-10-21 13:55:46 +02001425 sk->sk_destruct = udp_destruct_sock;
1426 return 0;
1427}
1428EXPORT_SYMBOL_GPL(udp_init_sock);
1429
1430void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1431{
1432 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1433 bool slow = lock_sock_fast(sk);
1434
1435 sk_peek_offset_bwd(sk, len);
1436 unlock_sock_fast(sk, slow);
1437 }
Paolo Abeni0a463c72017-06-12 11:23:42 +02001438
1439 consume_stateless_skb(skb);
Paolo Abenif970bd92016-10-21 13:55:46 +02001440}
1441EXPORT_SYMBOL_GPL(skb_consume_udp);
1442
Paolo Abeni2276f582017-05-16 11:20:14 +02001443static struct sk_buff *__first_packet_length(struct sock *sk,
1444 struct sk_buff_head *rcvq,
1445 int *total)
1446{
1447 struct sk_buff *skb;
1448
Paolo Abeni9bd780f2017-06-23 14:19:51 +02001449 while ((skb = skb_peek(rcvq)) != NULL) {
1450 if (udp_lib_checksum_complete(skb)) {
1451 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1452 IS_UDPLITE(sk));
1453 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1454 IS_UDPLITE(sk));
1455 atomic_inc(&sk->sk_drops);
1456 __skb_unlink(skb, rcvq);
1457 *total += skb->truesize;
1458 kfree_skb(skb);
1459 } else {
1460 /* the csum related bits could be changed, refresh
1461 * the scratch area
1462 */
1463 udp_set_dev_scratch(skb);
1464 break;
1465 }
Paolo Abeni2276f582017-05-16 11:20:14 +02001466 }
1467 return skb;
1468}
1469
Eric Dumazet85584672009-10-09 04:43:40 +00001470/**
1471 * first_packet_length - return length of first packet in receive queue
1472 * @sk: socket
1473 *
1474 * Drops all bad checksum frames, until a valid one is found.
Eric Dumazete83c6742016-08-23 13:59:33 -07001475 * Returns the length of found skb, or -1 if none is found.
Eric Dumazet85584672009-10-09 04:43:40 +00001476 */
Eric Dumazete83c6742016-08-23 13:59:33 -07001477static int first_packet_length(struct sock *sk)
Eric Dumazet85584672009-10-09 04:43:40 +00001478{
Paolo Abeni2276f582017-05-16 11:20:14 +02001479 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1480 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
Eric Dumazet85584672009-10-09 04:43:40 +00001481 struct sk_buff *skb;
Paolo Abeni7c13f972016-11-04 11:28:59 +01001482 int total = 0;
Eric Dumazete83c6742016-08-23 13:59:33 -07001483 int res;
Eric Dumazet85584672009-10-09 04:43:40 +00001484
Eric Dumazet85584672009-10-09 04:43:40 +00001485 spin_lock_bh(&rcvq->lock);
Paolo Abeni2276f582017-05-16 11:20:14 +02001486 skb = __first_packet_length(sk, rcvq, &total);
1487 if (!skb && !skb_queue_empty(sk_queue)) {
1488 spin_lock(&sk_queue->lock);
1489 skb_queue_splice_tail_init(sk_queue, rcvq);
1490 spin_unlock(&sk_queue->lock);
1491
1492 skb = __first_packet_length(sk, rcvq, &total);
Eric Dumazet85584672009-10-09 04:43:40 +00001493 }
Eric Dumazete83c6742016-08-23 13:59:33 -07001494 res = skb ? skb->len : -1;
Paolo Abeni7c13f972016-11-04 11:28:59 +01001495 if (total)
Paolo Abeni6dfb4362017-05-16 11:20:15 +02001496 udp_rmem_release(sk, total, 1, false);
Eric Dumazet85584672009-10-09 04:43:40 +00001497 spin_unlock_bh(&rcvq->lock);
Eric Dumazet85584672009-10-09 04:43:40 +00001498 return res;
1499}
1500
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501/*
1502 * IOCTL requests applicable to the UDP protocol
1503 */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001504
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1506{
Stephen Hemminger6516c652007-03-08 20:41:55 -08001507 switch (cmd) {
1508 case SIOCOUTQ:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 {
Eric Dumazet31e6d362009-06-17 19:05:41 -07001510 int amount = sk_wmem_alloc_get(sk);
1511
Stephen Hemminger6516c652007-03-08 20:41:55 -08001512 return put_user(amount, (int __user *)arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 }
Stephen Hemminger6516c652007-03-08 20:41:55 -08001514
1515 case SIOCINQ:
1516 {
Eric Dumazete83c6742016-08-23 13:59:33 -07001517 int amount = max_t(int, 0, first_packet_length(sk));
Stephen Hemminger6516c652007-03-08 20:41:55 -08001518
Stephen Hemminger6516c652007-03-08 20:41:55 -08001519 return put_user(amount, (int __user *)arg);
1520 }
1521
1522 default:
1523 return -ENOIOCTLCMD;
1524 }
1525
1526 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527}
Eric Dumazetc482c562009-07-17 00:26:32 +00001528EXPORT_SYMBOL(udp_ioctl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529
Paolo Abeni2276f582017-05-16 11:20:14 +02001530struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1531 int noblock, int *peeked, int *off, int *err)
1532{
1533 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1534 struct sk_buff_head *queue;
1535 struct sk_buff *last;
1536 long timeo;
1537 int error;
1538
1539 queue = &udp_sk(sk)->reader_queue;
1540 flags |= noblock ? MSG_DONTWAIT : 0;
1541 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1542 do {
1543 struct sk_buff *skb;
1544
1545 error = sock_error(sk);
1546 if (error)
1547 break;
1548
1549 error = -EAGAIN;
1550 *peeked = 0;
1551 do {
Paolo Abeni2276f582017-05-16 11:20:14 +02001552 spin_lock_bh(&queue->lock);
1553 skb = __skb_try_recv_from_queue(sk, queue, flags,
1554 udp_skb_destructor,
Andrey Vaginde321ed2017-05-17 11:39:05 -07001555 peeked, off, err,
Paolo Abeni2276f582017-05-16 11:20:14 +02001556 &last);
1557 if (skb) {
1558 spin_unlock_bh(&queue->lock);
Paolo Abeni2276f582017-05-16 11:20:14 +02001559 return skb;
1560 }
1561
1562 if (skb_queue_empty(sk_queue)) {
1563 spin_unlock_bh(&queue->lock);
1564 goto busy_check;
1565 }
1566
Paolo Abeni6dfb4362017-05-16 11:20:15 +02001567 /* refill the reader queue and walk it again
1568 * keep both queues locked to avoid re-acquiring
1569 * the sk_receive_queue lock if fwd memory scheduling
1570 * is needed.
1571 */
Paolo Abeni2276f582017-05-16 11:20:14 +02001572 spin_lock(&sk_queue->lock);
1573 skb_queue_splice_tail_init(sk_queue, queue);
Paolo Abeni2276f582017-05-16 11:20:14 +02001574
1575 skb = __skb_try_recv_from_queue(sk, queue, flags,
Paolo Abeni6dfb4362017-05-16 11:20:15 +02001576 udp_skb_dtor_locked,
Andrey Vaginde321ed2017-05-17 11:39:05 -07001577 peeked, off, err,
Paolo Abeni2276f582017-05-16 11:20:14 +02001578 &last);
Paolo Abeni6dfb4362017-05-16 11:20:15 +02001579 spin_unlock(&sk_queue->lock);
Paolo Abeni2276f582017-05-16 11:20:14 +02001580 spin_unlock_bh(&queue->lock);
Andrey Vaginde321ed2017-05-17 11:39:05 -07001581 if (skb)
Paolo Abeni2276f582017-05-16 11:20:14 +02001582 return skb;
Paolo Abeni2276f582017-05-16 11:20:14 +02001583
1584busy_check:
1585 if (!sk_can_busy_loop(sk))
1586 break;
1587
1588 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1589 } while (!skb_queue_empty(sk_queue));
1590
1591 /* sk_queue is empty, reader_queue may contain peeked packets */
1592 } while (timeo &&
1593 !__skb_wait_for_more_packets(sk, &error, &timeo,
1594 (struct sk_buff *)sk_queue));
1595
1596 *err = error;
1597 return NULL;
1598}
1599EXPORT_SYMBOL_GPL(__skb_recv_udp);
1600
Paolo Abenib65ac442017-06-12 11:23:43 +02001601static int copy_linear_skb(struct sk_buff *skb, int len, int off,
1602 struct iov_iter *to)
1603{
1604 int n, copy = len - off;
1605
1606 n = copy_to_iter(skb->data + off, copy, to);
1607 if (n == copy)
1608 return 0;
1609
1610 return -EFAULT;
1611}
1612
David S. Millerdb8dac22008-03-06 16:22:02 -08001613/*
1614 * This should be easy, if there is something there we
1615 * return it, otherwise we block.
1616 */
1617
Ying Xue1b784142015-03-02 15:37:48 +08001618int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1619 int flags, int *addr_len)
David S. Millerdb8dac22008-03-06 16:22:02 -08001620{
1621 struct inet_sock *inet = inet_sk(sk);
Steffen Hurrle342dfc32014-01-17 22:53:15 +01001622 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
David S. Millerdb8dac22008-03-06 16:22:02 -08001623 struct sk_buff *skb;
David S. Miller59c2cda2011-12-01 14:12:55 -05001624 unsigned int ulen, copied;
samanthakumar627d2d62016-04-05 12:41:16 -04001625 int peeked, peeking, off;
David S. Millerdb8dac22008-03-06 16:22:02 -08001626 int err;
1627 int is_udplite = IS_UDPLITE(sk);
Eric Dumazet197c9492015-12-30 08:51:12 -05001628 bool checksum_valid = false;
David S. Millerdb8dac22008-03-06 16:22:02 -08001629
David S. Millerdb8dac22008-03-06 16:22:02 -08001630 if (flags & MSG_ERRQUEUE)
Hannes Frederic Sowa85fbaa72013-11-23 00:46:12 +01001631 return ip_recv_error(sk, msg, len, addr_len);
David S. Millerdb8dac22008-03-06 16:22:02 -08001632
1633try_again:
samanthakumar627d2d62016-04-05 12:41:16 -04001634 peeking = off = sk_peek_offset(sk, flags);
Paolo Abeni7c13f972016-11-04 11:28:59 +01001635 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
David S. Millerdb8dac22008-03-06 16:22:02 -08001636 if (!skb)
samanthakumar627d2d62016-04-05 12:41:16 -04001637 return err;
David S. Millerdb8dac22008-03-06 16:22:02 -08001638
Paolo Abenib65ac442017-06-12 11:23:43 +02001639 ulen = udp_skb_len(skb);
David S. Miller59c2cda2011-12-01 14:12:55 -05001640 copied = len;
samanthakumar627d2d62016-04-05 12:41:16 -04001641 if (copied > ulen - off)
1642 copied = ulen - off;
David S. Miller59c2cda2011-12-01 14:12:55 -05001643 else if (copied < ulen)
David S. Millerdb8dac22008-03-06 16:22:02 -08001644 msg->msg_flags |= MSG_TRUNC;
1645
1646 /*
1647 * If checksum is needed at all, try to do it while copying the
1648 * data. If the data is truncated, or if we only want a partial
1649 * coverage checksum (UDP-Lite), do it before the copy.
1650 */
1651
Eric Dumazetd21dbdf2016-11-18 17:18:03 -08001652 if (copied < ulen || peeking ||
1653 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
Paolo Abenib65ac442017-06-12 11:23:43 +02001654 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1655 !__udp_lib_checksum_complete(skb);
Eric Dumazet197c9492015-12-30 08:51:12 -05001656 if (!checksum_valid)
David S. Millerdb8dac22008-03-06 16:22:02 -08001657 goto csum_copy_err;
1658 }
1659
Paolo Abenib65ac442017-06-12 11:23:43 +02001660 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1661 if (udp_skb_is_linear(skb))
1662 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1663 else
1664 err = skb_copy_datagram_msg(skb, off, msg, copied);
1665 } else {
samanthakumar627d2d62016-04-05 12:41:16 -04001666 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
David S. Millerdb8dac22008-03-06 16:22:02 -08001667
1668 if (err == -EINVAL)
1669 goto csum_copy_err;
1670 }
1671
Eric Dumazet22911fc2012-06-27 00:23:44 +00001672 if (unlikely(err)) {
Eric Dumazet979402b2012-09-05 23:34:44 +00001673 if (!peeked) {
1674 atomic_inc(&sk->sk_drops);
Eric Dumazet6aef70a2016-04-27 16:44:27 -07001675 UDP_INC_STATS(sock_net(sk),
1676 UDP_MIB_INERRORS, is_udplite);
Eric Dumazet979402b2012-09-05 23:34:44 +00001677 }
Paolo Abeni850cbad2016-10-21 13:55:47 +02001678 kfree_skb(skb);
samanthakumar627d2d62016-04-05 12:41:16 -04001679 return err;
Eric Dumazet22911fc2012-06-27 00:23:44 +00001680 }
David S. Millerdb8dac22008-03-06 16:22:02 -08001681
1682 if (!peeked)
Eric Dumazet6aef70a2016-04-27 16:44:27 -07001683 UDP_INC_STATS(sock_net(sk),
1684 UDP_MIB_INDATAGRAMS, is_udplite);
David S. Millerdb8dac22008-03-06 16:22:02 -08001685
Neil Horman3b885782009-10-12 13:26:31 -07001686 sock_recv_ts_and_drops(msg, sk, skb);
David S. Millerdb8dac22008-03-06 16:22:02 -08001687
1688 /* Copy the address. */
Eric Dumazetc482c562009-07-17 00:26:32 +00001689 if (sin) {
David S. Millerdb8dac22008-03-06 16:22:02 -08001690 sin->sin_family = AF_INET;
1691 sin->sin_port = udp_hdr(skb)->source;
1692 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1693 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
Hannes Frederic Sowabceaa902013-11-18 04:20:45 +01001694 *addr_len = sizeof(*sin);
David S. Millerdb8dac22008-03-06 16:22:02 -08001695 }
1696 if (inet->cmsg_flags)
Paolo Abeniad959032016-11-04 11:28:58 +01001697 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
David S. Millerdb8dac22008-03-06 16:22:02 -08001698
David S. Miller59c2cda2011-12-01 14:12:55 -05001699 err = copied;
David S. Millerdb8dac22008-03-06 16:22:02 -08001700 if (flags & MSG_TRUNC)
1701 err = ulen;
1702
Paolo Abeni850cbad2016-10-21 13:55:47 +02001703 skb_consume_udp(sk, skb, peeking ? -err : err);
David S. Millerdb8dac22008-03-06 16:22:02 -08001704 return err;
1705
1706csum_copy_err:
Paolo Abeni2276f582017-05-16 11:20:14 +02001707 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1708 udp_skb_destructor)) {
Eric Dumazet6aef70a2016-04-27 16:44:27 -07001709 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1710 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
Eric Dumazet6a5dc9e2013-04-29 08:39:56 +00001711 }
Paolo Abeni850cbad2016-10-21 13:55:47 +02001712 kfree_skb(skb);
David S. Millerdb8dac22008-03-06 16:22:02 -08001713
Eric Dumazetbeb39db2015-05-30 09:16:53 -07001714 /* starting over for a new packet, but check if we need to yield */
1715 cond_resched();
Xufeng Zhang9cfaa8d2011-06-21 10:43:40 +00001716 msg->msg_flags &= ~MSG_TRUNC;
David S. Millerdb8dac22008-03-06 16:22:02 -08001717 goto try_again;
1718}
1719
Eric Dumazet286c72d2016-10-20 09:39:40 -07001720int __udp_disconnect(struct sock *sk, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721{
1722 struct inet_sock *inet = inet_sk(sk);
1723 /*
1724 * 1003.1g - break association.
1725 */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001726
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 sk->sk_state = TCP_CLOSE;
Eric Dumazetc720c7e82009-10-15 06:30:45 +00001728 inet->inet_daddr = 0;
1729 inet->inet_dport = 0;
Tom Herbertbdeab992011-08-14 19:45:55 +00001730 sock_rps_reset_rxhash(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 sk->sk_bound_dev_if = 0;
1732 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1733 inet_reset_saddr(sk);
1734
1735 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1736 sk->sk_prot->unhash(sk);
Eric Dumazetc720c7e82009-10-15 06:30:45 +00001737 inet->inet_sport = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 }
1739 sk_dst_reset(sk);
1740 return 0;
1741}
Eric Dumazet286c72d2016-10-20 09:39:40 -07001742EXPORT_SYMBOL(__udp_disconnect);
1743
1744int udp_disconnect(struct sock *sk, int flags)
1745{
1746 lock_sock(sk);
1747 __udp_disconnect(sk, flags);
1748 release_sock(sk);
1749 return 0;
1750}
Eric Dumazetc482c562009-07-17 00:26:32 +00001751EXPORT_SYMBOL(udp_disconnect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752
Eric Dumazet645ca702008-10-29 01:41:45 -07001753void udp_lib_unhash(struct sock *sk)
1754{
Eric Dumazet723b4612008-11-25 13:55:15 -08001755 if (sk_hashed(sk)) {
1756 struct udp_table *udptable = sk->sk_prot->h.udp_table;
Eric Dumazet512615b2009-11-08 10:17:58 +00001757 struct udp_hslot *hslot, *hslot2;
1758
1759 hslot = udp_hashslot(udptable, sock_net(sk),
1760 udp_sk(sk)->udp_port_hash);
1761 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
Eric Dumazet645ca702008-10-29 01:41:45 -07001762
Eric Dumazet723b4612008-11-25 13:55:15 -08001763 spin_lock_bh(&hslot->lock);
Craig Galleke32ea7e2016-01-04 17:41:46 -05001764 if (rcu_access_pointer(sk->sk_reuseport_cb))
1765 reuseport_detach_sock(sk);
Eric Dumazetca065d02016-04-01 08:52:13 -07001766 if (sk_del_node_init_rcu(sk)) {
Eric Dumazetfdcc8aa92009-11-08 10:17:05 +00001767 hslot->count--;
Eric Dumazetc720c7e82009-10-15 06:30:45 +00001768 inet_sk(sk)->inet_num = 0;
Eric Dumazet723b4612008-11-25 13:55:15 -08001769 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
Eric Dumazet512615b2009-11-08 10:17:58 +00001770
1771 spin_lock(&hslot2->lock);
Eric Dumazetca065d02016-04-01 08:52:13 -07001772 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
Eric Dumazet512615b2009-11-08 10:17:58 +00001773 hslot2->count--;
1774 spin_unlock(&hslot2->lock);
Eric Dumazet723b4612008-11-25 13:55:15 -08001775 }
1776 spin_unlock_bh(&hslot->lock);
Eric Dumazet645ca702008-10-29 01:41:45 -07001777 }
Eric Dumazet645ca702008-10-29 01:41:45 -07001778}
1779EXPORT_SYMBOL(udp_lib_unhash);
1780
Eric Dumazet719f8352010-09-08 05:08:44 +00001781/*
1782 * inet_rcv_saddr was changed, we must rehash secondary hash
1783 */
1784void udp_lib_rehash(struct sock *sk, u16 newhash)
1785{
1786 if (sk_hashed(sk)) {
1787 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1788 struct udp_hslot *hslot, *hslot2, *nhslot2;
1789
1790 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1791 nhslot2 = udp_hashslot2(udptable, newhash);
1792 udp_sk(sk)->udp_portaddr_hash = newhash;
Craig Galleke32ea7e2016-01-04 17:41:46 -05001793
1794 if (hslot2 != nhslot2 ||
1795 rcu_access_pointer(sk->sk_reuseport_cb)) {
Eric Dumazet719f8352010-09-08 05:08:44 +00001796 hslot = udp_hashslot(udptable, sock_net(sk),
1797 udp_sk(sk)->udp_port_hash);
1798 /* we must lock primary chain too */
1799 spin_lock_bh(&hslot->lock);
Craig Galleke32ea7e2016-01-04 17:41:46 -05001800 if (rcu_access_pointer(sk->sk_reuseport_cb))
1801 reuseport_detach_sock(sk);
Eric Dumazet719f8352010-09-08 05:08:44 +00001802
Craig Galleke32ea7e2016-01-04 17:41:46 -05001803 if (hslot2 != nhslot2) {
1804 spin_lock(&hslot2->lock);
Eric Dumazetca065d02016-04-01 08:52:13 -07001805 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
Craig Galleke32ea7e2016-01-04 17:41:46 -05001806 hslot2->count--;
1807 spin_unlock(&hslot2->lock);
Eric Dumazet719f8352010-09-08 05:08:44 +00001808
Craig Galleke32ea7e2016-01-04 17:41:46 -05001809 spin_lock(&nhslot2->lock);
Eric Dumazetca065d02016-04-01 08:52:13 -07001810 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
Craig Galleke32ea7e2016-01-04 17:41:46 -05001811 &nhslot2->head);
1812 nhslot2->count++;
1813 spin_unlock(&nhslot2->lock);
1814 }
Eric Dumazet719f8352010-09-08 05:08:44 +00001815
1816 spin_unlock_bh(&hslot->lock);
1817 }
1818 }
1819}
1820EXPORT_SYMBOL(udp_lib_rehash);
1821
1822static void udp_v4_rehash(struct sock *sk)
1823{
1824 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1825 inet_sk(sk)->inet_rcv_saddr,
1826 inet_sk(sk)->inet_num);
1827 udp_lib_rehash(sk, new_hash);
1828}
1829
Paolo Abenia3f96c42017-05-17 14:52:16 +02001830static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
Herbert Xu93821772008-09-15 11:48:46 -07001831{
Tom Herbertfec5e652010-04-16 16:01:27 -07001832 int rc;
Herbert Xu93821772008-09-15 11:48:46 -07001833
Shawn Bohrer005ec972013-10-07 11:01:38 -05001834 if (inet_sk(sk)->inet_daddr) {
Tom Herbertbdeab992011-08-14 19:45:55 +00001835 sock_rps_save_rxhash(sk, skb);
Shawn Bohrer005ec972013-10-07 11:01:38 -05001836 sk_mark_napi_id(sk, skb);
Eric Dumazet2c8c56e2014-11-11 05:54:28 -08001837 sk_incoming_cpu_update(sk);
Eric Dumazete68b6e52016-11-16 09:10:42 -08001838 } else {
1839 sk_mark_napi_id_once(sk, skb);
Shawn Bohrer005ec972013-10-07 11:01:38 -05001840 }
Tom Herbertfec5e652010-04-16 16:01:27 -07001841
Paolo Abeni0a463c72017-06-12 11:23:42 +02001842 /* clear all pending head states while they are hot in the cache */
1843 skb_release_head_state(skb);
1844
Paolo Abeni850cbad2016-10-21 13:55:47 +02001845 rc = __udp_enqueue_schedule_skb(sk, skb);
Eric Dumazet766e90372009-10-14 20:40:11 -07001846 if (rc < 0) {
1847 int is_udplite = IS_UDPLITE(sk);
1848
Herbert Xu93821772008-09-15 11:48:46 -07001849 /* Note that an ENOMEM error is charged twice */
Eric Dumazet766e90372009-10-14 20:40:11 -07001850 if (rc == -ENOMEM)
Eric Dumazete61da9e2016-04-29 14:16:50 -07001851 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
Eric Dumazet02c22342016-04-27 16:44:30 -07001852 is_udplite);
Eric Dumazete61da9e2016-04-29 14:16:50 -07001853 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
Eric Dumazet766e90372009-10-14 20:40:11 -07001854 kfree_skb(skb);
Satoru Moriya296f7ea2011-06-17 11:58:39 +00001855 trace_udp_fail_queue_rcv_skb(rc, sk);
Eric Dumazet766e90372009-10-14 20:40:11 -07001856 return -1;
Herbert Xu93821772008-09-15 11:48:46 -07001857 }
1858
1859 return 0;
Herbert Xu93821772008-09-15 11:48:46 -07001860}
1861
Eric Dumazet447167b2012-04-11 23:05:28 +00001862static struct static_key udp_encap_needed __read_mostly;
1863void udp_encap_enable(void)
1864{
1865 if (!static_key_enabled(&udp_encap_needed))
1866 static_key_slow_inc(&udp_encap_needed);
1867}
1868EXPORT_SYMBOL(udp_encap_enable);
1869
David S. Millerdb8dac22008-03-06 16:22:02 -08001870/* returns:
1871 * -1: error
1872 * 0: success
1873 * >0: "udp encap" protocol resubmission
1874 *
1875 * Note that in the success and error cases, the skb is assumed to
1876 * have either been requeued or freed.
1877 */
Paolo Abenia3f96c42017-05-17 14:52:16 +02001878static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
David S. Millerdb8dac22008-03-06 16:22:02 -08001879{
1880 struct udp_sock *up = udp_sk(sk);
David S. Millerdb8dac22008-03-06 16:22:02 -08001881 int is_udplite = IS_UDPLITE(sk);
1882
1883 /*
1884 * Charge it to the socket, dropping if the queue is full.
1885 */
1886 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1887 goto drop;
1888 nf_reset(skb);
1889
Eric Dumazet447167b2012-04-11 23:05:28 +00001890 if (static_key_false(&udp_encap_needed) && up->encap_type) {
Eric Dumazet0ad92ad2011-11-01 12:56:59 +00001891 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1892
David S. Millerdb8dac22008-03-06 16:22:02 -08001893 /*
1894 * This is an encapsulation socket so pass the skb to
1895 * the socket's udp_encap_rcv() hook. Otherwise, just
1896 * fall through and pass this up the UDP socket.
1897 * up->encap_rcv() returns the following value:
1898 * =0 if skb was successfully passed to the encap
1899 * handler or was discarded by it.
1900 * >0 if skb should be passed on to UDP.
1901 * <0 if skb should be resubmitted as proto -N
1902 */
1903
1904 /* if we're overly short, let UDP handle it */
Eric Dumazet0ad92ad2011-11-01 12:56:59 +00001905 encap_rcv = ACCESS_ONCE(up->encap_rcv);
Hannes Frederic Sowae5aed002016-05-19 15:58:33 +02001906 if (encap_rcv) {
David S. Millerdb8dac22008-03-06 16:22:02 -08001907 int ret;
1908
Tom Herbert0a809662014-05-07 16:52:39 -07001909 /* Verify checksum before giving to encap */
1910 if (udp_lib_checksum_complete(skb))
1911 goto csum_error;
1912
Eric Dumazet0ad92ad2011-11-01 12:56:59 +00001913 ret = encap_rcv(sk, skb);
David S. Millerdb8dac22008-03-06 16:22:02 -08001914 if (ret <= 0) {
Eric Dumazet02c22342016-04-27 16:44:30 -07001915 __UDP_INC_STATS(sock_net(sk),
1916 UDP_MIB_INDATAGRAMS,
1917 is_udplite);
David S. Millerdb8dac22008-03-06 16:22:02 -08001918 return -ret;
1919 }
1920 }
1921
1922 /* FALLTHROUGH -- it's a UDP Packet */
1923 }
1924
1925 /*
1926 * UDP-Lite specific tests, ignored on UDP sockets
1927 */
1928 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1929
1930 /*
1931 * MIB statistics other than incrementing the error count are
1932 * disabled for the following two types of errors: these depend
1933 * on the application settings, not on the functioning of the
1934 * protocol stack as such.
1935 *
1936 * RFC 3828 here recommends (sec 3.3): "There should also be a
1937 * way ... to ... at least let the receiving application block
1938 * delivery of packets with coverage values less than a value
1939 * provided by the application."
1940 */
1941 if (up->pcrlen == 0) { /* full coverage was set */
Joe Perchesba7a46f2014-11-11 10:59:17 -08001942 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1943 UDP_SKB_CB(skb)->cscov, skb->len);
David S. Millerdb8dac22008-03-06 16:22:02 -08001944 goto drop;
1945 }
1946 /* The next case involves violating the min. coverage requested
1947 * by the receiver. This is subtle: if receiver wants x and x is
1948 * greater than the buffersize/MTU then receiver will complain
1949 * that it wants x while sender emits packets of smaller size y.
1950 * Therefore the above ...()->partial_cov statement is essential.
1951 */
1952 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
Joe Perchesba7a46f2014-11-11 10:59:17 -08001953 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1954 UDP_SKB_CB(skb)->cscov, up->pcrlen);
David S. Millerdb8dac22008-03-06 16:22:02 -08001955 goto drop;
1956 }
1957 }
1958
Paolo Abenidd99e422017-06-21 10:24:40 +02001959 prefetch(&sk->sk_rmem_alloc);
Eric Dumazetce25d662016-06-02 14:52:43 -07001960 if (rcu_access_pointer(sk->sk_filter) &&
1961 udp_lib_checksum_complete(skb))
samanthakumare6afc8a2016-04-05 12:41:15 -04001962 goto csum_error;
Eric Dumazetce25d662016-06-02 14:52:43 -07001963
Daniel Borkmannba66bbe2016-07-25 18:06:12 +02001964 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
Michal Kubečeka6127692016-07-08 17:52:33 +02001965 goto drop;
David S. Millerdb8dac22008-03-06 16:22:02 -08001966
samanthakumare6afc8a2016-04-05 12:41:15 -04001967 udp_csum_pull_header(skb);
David S. Millerdb8dac22008-03-06 16:22:02 -08001968
Shawn Bohrerfbf88662013-10-07 11:01:40 -05001969 ipv4_pktinfo_prepare(sk, skb);
Paolo Abeni850cbad2016-10-21 13:55:47 +02001970 return __udp_queue_rcv_skb(sk, skb);
David S. Millerdb8dac22008-03-06 16:22:02 -08001971
Eric Dumazet6a5dc9e2013-04-29 08:39:56 +00001972csum_error:
Eric Dumazet02c22342016-04-27 16:44:30 -07001973 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
David S. Millerdb8dac22008-03-06 16:22:02 -08001974drop:
Eric Dumazet02c22342016-04-27 16:44:30 -07001975 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
Eric Dumazet8edf19c2009-10-15 00:12:40 +00001976 atomic_inc(&sk->sk_drops);
David S. Millerdb8dac22008-03-06 16:22:02 -08001977 kfree_skb(skb);
1978 return -1;
1979}
1980
Eric Dumazet97502232013-12-11 14:46:51 -08001981/* For TCP sockets, sk_rx_dst is protected by socket lock
Eric Dumazete47eb5d2013-12-15 10:53:46 -08001982 * For UDP, we use xchg() to guard against concurrent changes.
Eric Dumazet97502232013-12-11 14:46:51 -08001983 */
1984static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
Shawn Bohrer421b3882013-10-07 11:01:39 -05001985{
Eric Dumazet97502232013-12-11 14:46:51 -08001986 struct dst_entry *old;
Shawn Bohrer421b3882013-10-07 11:01:39 -05001987
Wei Wangd24406c2017-06-17 10:42:25 -07001988 if (dst_hold_safe(dst)) {
1989 old = xchg(&sk->sk_rx_dst, dst);
1990 dst_release(old);
1991 }
Shawn Bohrer421b3882013-10-07 11:01:39 -05001992}
1993
David S. Millerdb8dac22008-03-06 16:22:02 -08001994/*
1995 * Multicasts and broadcasts go to each listener.
1996 *
Eric Dumazet1240d132009-11-08 10:18:44 +00001997 * Note: called only from the BH handler context.
David S. Millerdb8dac22008-03-06 16:22:02 -08001998 */
Pavel Emelyanove3163492008-06-16 17:12:11 -07001999static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
David S. Millerdb8dac22008-03-06 16:22:02 -08002000 struct udphdr *uh,
2001 __be32 saddr, __be32 daddr,
Rick Jones36cbb242014-11-06 10:37:54 -08002002 struct udp_table *udptable,
2003 int proto)
David S. Millerdb8dac22008-03-06 16:22:02 -08002004{
Eric Dumazetca065d02016-04-01 08:52:13 -07002005 struct sock *sk, *first = NULL;
David Held5cf3d462014-07-15 23:28:31 -04002006 unsigned short hnum = ntohs(uh->dest);
2007 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
David Held2dc41cf2014-07-15 23:28:32 -04002008 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
Eric Dumazetca065d02016-04-01 08:52:13 -07002009 unsigned int offset = offsetof(typeof(*sk), sk_node);
2010 int dif = skb->dev->ifindex;
2011 struct hlist_node *node;
2012 struct sk_buff *nskb;
David Held2dc41cf2014-07-15 23:28:32 -04002013
2014 if (use_hash2) {
2015 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
Pablo Neira73e2d5e2016-11-14 23:40:30 +01002016 udptable->mask;
2017 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udptable->mask;
David Held2dc41cf2014-07-15 23:28:32 -04002018start_lookup:
Pablo Neira73e2d5e2016-11-14 23:40:30 +01002019 hslot = &udptable->hash2[hash2];
David Held2dc41cf2014-07-15 23:28:32 -04002020 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2021 }
David S. Millerdb8dac22008-03-06 16:22:02 -08002022
Eric Dumazetca065d02016-04-01 08:52:13 -07002023 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2024 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2025 uh->source, saddr, dif, hnum))
2026 continue;
David S. Millerdb8dac22008-03-06 16:22:02 -08002027
Eric Dumazetca065d02016-04-01 08:52:13 -07002028 if (!first) {
2029 first = sk;
2030 continue;
2031 }
2032 nskb = skb_clone(skb, GFP_ATOMIC);
2033
2034 if (unlikely(!nskb)) {
2035 atomic_inc(&sk->sk_drops);
Eric Dumazet02c22342016-04-27 16:44:30 -07002036 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2037 IS_UDPLITE(sk));
2038 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2039 IS_UDPLITE(sk));
Eric Dumazetca065d02016-04-01 08:52:13 -07002040 continue;
2041 }
2042 if (udp_queue_rcv_skb(sk, nskb) > 0)
2043 consume_skb(nskb);
2044 }
Eric Dumazet1240d132009-11-08 10:18:44 +00002045
David Held2dc41cf2014-07-15 23:28:32 -04002046 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2047 if (use_hash2 && hash2 != hash2_any) {
2048 hash2 = hash2_any;
2049 goto start_lookup;
2050 }
2051
Eric Dumazetca065d02016-04-01 08:52:13 -07002052 if (first) {
2053 if (udp_queue_rcv_skb(first, skb) > 0)
2054 consume_skb(skb);
Eric Dumazet1240d132009-11-08 10:18:44 +00002055 } else {
Eric Dumazetca065d02016-04-01 08:52:13 -07002056 kfree_skb(skb);
Eric Dumazet02c22342016-04-27 16:44:30 -07002057 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2058 proto == IPPROTO_UDPLITE);
Eric Dumazet1240d132009-11-08 10:18:44 +00002059 }
David S. Millerdb8dac22008-03-06 16:22:02 -08002060 return 0;
2061}
2062
2063/* Initialize UDP checksum. If exited with zero value (success),
2064 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2065 * Otherwise, csum completion requires chacksumming packet body,
2066 * including udp header and folding it to skb->csum.
2067 */
2068static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2069 int proto)
2070{
David S. Millerdb8dac22008-03-06 16:22:02 -08002071 int err;
2072
2073 UDP_SKB_CB(skb)->partial_cov = 0;
2074 UDP_SKB_CB(skb)->cscov = skb->len;
2075
2076 if (proto == IPPROTO_UDPLITE) {
2077 err = udplite_checksum_init(skb, uh);
2078 if (err)
2079 return err;
2080 }
2081
Hannes Frederic Sowab46d9f62016-06-12 12:02:46 +02002082 /* Note, we are only interested in != 0 or == 0, thus the
2083 * force to int.
2084 */
2085 return (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2086 inet_compute_pseudo);
David S. Millerdb8dac22008-03-06 16:22:02 -08002087}
2088
2089/*
2090 * All we need to do is get the socket, and then do a checksum.
2091 */
2092
Eric Dumazet645ca702008-10-29 01:41:45 -07002093int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
David S. Millerdb8dac22008-03-06 16:22:02 -08002094 int proto)
2095{
2096 struct sock *sk;
Jesper Dangaard Brouer7b5e56f2009-02-05 15:05:45 -08002097 struct udphdr *uh;
David S. Millerdb8dac22008-03-06 16:22:02 -08002098 unsigned short ulen;
Eric Dumazetadf30902009-06-02 05:19:30 +00002099 struct rtable *rt = skb_rtable(skb);
Jesper Dangaard Brouer2783ef22009-02-06 01:59:12 -08002100 __be32 saddr, daddr;
Pavel Emelyanov02833282008-07-05 21:18:48 -07002101 struct net *net = dev_net(skb->dev);
David S. Millerdb8dac22008-03-06 16:22:02 -08002102
2103 /*
2104 * Validate the packet.
2105 */
2106 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2107 goto drop; /* No space for header. */
2108
Jesper Dangaard Brouer7b5e56f2009-02-05 15:05:45 -08002109 uh = udp_hdr(skb);
David S. Millerdb8dac22008-03-06 16:22:02 -08002110 ulen = ntohs(uh->len);
Bjørn Morkccc2d972010-05-06 03:44:34 +00002111 saddr = ip_hdr(skb)->saddr;
2112 daddr = ip_hdr(skb)->daddr;
2113
David S. Millerdb8dac22008-03-06 16:22:02 -08002114 if (ulen > skb->len)
2115 goto short_packet;
2116
2117 if (proto == IPPROTO_UDP) {
2118 /* UDP validates ulen. */
2119 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2120 goto short_packet;
2121 uh = udp_hdr(skb);
2122 }
2123
2124 if (udp4_csum_init(skb, uh, proto))
2125 goto csum_error;
2126
Eric Dumazet8afdd992013-12-10 18:07:23 -08002127 sk = skb_steal_sock(skb);
2128 if (sk) {
Eric Dumazet97502232013-12-11 14:46:51 -08002129 struct dst_entry *dst = skb_dst(skb);
Shawn Bohrer421b3882013-10-07 11:01:39 -05002130 int ret;
David S. Millerdb8dac22008-03-06 16:22:02 -08002131
Eric Dumazet97502232013-12-11 14:46:51 -08002132 if (unlikely(sk->sk_rx_dst != dst))
2133 udp_sk_rx_dst_set(sk, dst);
Shawn Bohrer421b3882013-10-07 11:01:39 -05002134
2135 ret = udp_queue_rcv_skb(sk, skb);
Eric Dumazet8afdd992013-12-10 18:07:23 -08002136 sock_put(sk);
Shawn Bohrer421b3882013-10-07 11:01:39 -05002137 /* a return value > 0 means to resubmit the input, but
2138 * it wants the return to be -protocol, or 0
2139 */
2140 if (ret > 0)
2141 return -ret;
2142 return 0;
Shawn Bohrer421b3882013-10-07 11:01:39 -05002143 }
David S. Millerdb8dac22008-03-06 16:22:02 -08002144
Fabian Frederickc18450a2014-11-04 20:48:41 +01002145 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2146 return __udp4_lib_mcast_deliver(net, skb, uh,
Rick Jones36cbb242014-11-06 10:37:54 -08002147 saddr, daddr, udptable, proto);
Fabian Frederickc18450a2014-11-04 20:48:41 +01002148
2149 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
Ian Morris00db4122015-04-03 09:17:27 +01002150 if (sk) {
Eliezer Tamira5b50472013-06-10 11:40:00 +03002151 int ret;
2152
Tom Herbert224d0192015-01-05 13:56:14 -08002153 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
Tom Herbert2abb7cd2014-08-31 15:12:43 -07002154 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
2155 inet_compute_pseudo);
2156
Eliezer Tamira5b50472013-06-10 11:40:00 +03002157 ret = udp_queue_rcv_skb(sk, skb);
David S. Millerdb8dac22008-03-06 16:22:02 -08002158
2159 /* a return value > 0 means to resubmit the input, but
2160 * it wants the return to be -protocol, or 0
2161 */
2162 if (ret > 0)
2163 return -ret;
2164 return 0;
2165 }
2166
2167 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2168 goto drop;
2169 nf_reset(skb);
2170
2171 /* No socket. Drop packet silently, if checksum is wrong */
2172 if (udp_lib_checksum_complete(skb))
2173 goto csum_error;
2174
Eric Dumazet02c22342016-04-27 16:44:30 -07002175 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
David S. Millerdb8dac22008-03-06 16:22:02 -08002176 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2177
2178 /*
2179 * Hmm. We got an UDP packet to a port to which we
2180 * don't wanna listen. Ignore it.
2181 */
2182 kfree_skb(skb);
2183 return 0;
2184
2185short_packet:
Joe Perchesba7a46f2014-11-11 10:59:17 -08002186 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2187 proto == IPPROTO_UDPLITE ? "Lite" : "",
2188 &saddr, ntohs(uh->source),
2189 ulen, skb->len,
2190 &daddr, ntohs(uh->dest));
David S. Millerdb8dac22008-03-06 16:22:02 -08002191 goto drop;
2192
2193csum_error:
2194 /*
2195 * RFC1122: OK. Discards the bad packet silently (as far as
2196 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2197 */
Joe Perchesba7a46f2014-11-11 10:59:17 -08002198 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2199 proto == IPPROTO_UDPLITE ? "Lite" : "",
2200 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2201 ulen);
Eric Dumazet02c22342016-04-27 16:44:30 -07002202 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
David S. Millerdb8dac22008-03-06 16:22:02 -08002203drop:
Eric Dumazet02c22342016-04-27 16:44:30 -07002204 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
David S. Millerdb8dac22008-03-06 16:22:02 -08002205 kfree_skb(skb);
2206 return 0;
2207}
2208
Shawn Bohrer421b3882013-10-07 11:01:39 -05002209/* We can only early demux multicast if there is a single matching socket.
2210 * If more than one socket found returns NULL
2211 */
2212static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2213 __be16 loc_port, __be32 loc_addr,
2214 __be16 rmt_port, __be32 rmt_addr,
2215 int dif)
2216{
2217 struct sock *sk, *result;
Shawn Bohrer421b3882013-10-07 11:01:39 -05002218 unsigned short hnum = ntohs(loc_port);
Eric Dumazetca065d02016-04-01 08:52:13 -07002219 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
Shawn Bohrer421b3882013-10-07 11:01:39 -05002220 struct udp_hslot *hslot = &udp_table.hash[slot];
2221
Eric Dumazet63c6f812014-06-12 16:13:06 -07002222 /* Do not bother scanning a too big list */
2223 if (hslot->count > 10)
2224 return NULL;
2225
Shawn Bohrer421b3882013-10-07 11:01:39 -05002226 result = NULL;
Eric Dumazetca065d02016-04-01 08:52:13 -07002227 sk_for_each_rcu(sk, &hslot->head) {
2228 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2229 rmt_port, rmt_addr, dif, hnum)) {
2230 if (result)
2231 return NULL;
Shawn Bohrer421b3882013-10-07 11:01:39 -05002232 result = sk;
Shawn Bohrer421b3882013-10-07 11:01:39 -05002233 }
2234 }
Shawn Bohrer421b3882013-10-07 11:01:39 -05002235
Shawn Bohrer421b3882013-10-07 11:01:39 -05002236 return result;
2237}
2238
2239/* For unicast we should only early demux connected sockets or we can
2240 * break forwarding setups. The chains here can be long so only check
2241 * if the first socket is an exact match and if not move on.
2242 */
2243static struct sock *__udp4_lib_demux_lookup(struct net *net,
2244 __be16 loc_port, __be32 loc_addr,
2245 __be16 rmt_port, __be32 rmt_addr,
2246 int dif)
2247{
Shawn Bohrer421b3882013-10-07 11:01:39 -05002248 unsigned short hnum = ntohs(loc_port);
2249 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
2250 unsigned int slot2 = hash2 & udp_table.mask;
2251 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
Joe Perchesc7228312014-05-13 20:30:07 -07002252 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
Shawn Bohrer421b3882013-10-07 11:01:39 -05002253 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
Eric Dumazetca065d02016-04-01 08:52:13 -07002254 struct sock *sk;
Shawn Bohrer421b3882013-10-07 11:01:39 -05002255
Eric Dumazetca065d02016-04-01 08:52:13 -07002256 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2257 if (INET_MATCH(sk, net, acookie, rmt_addr,
2258 loc_addr, ports, dif))
2259 return sk;
Shawn Bohrer421b3882013-10-07 11:01:39 -05002260 /* Only check first socket in chain */
2261 break;
2262 }
Eric Dumazetca065d02016-04-01 08:52:13 -07002263 return NULL;
Shawn Bohrer421b3882013-10-07 11:01:39 -05002264}
2265
2266void udp_v4_early_demux(struct sk_buff *skb)
2267{
Eric Dumazet610438b2013-12-11 08:10:05 -08002268 struct net *net = dev_net(skb->dev);
2269 const struct iphdr *iph;
2270 const struct udphdr *uh;
Eric Dumazetca065d02016-04-01 08:52:13 -07002271 struct sock *sk = NULL;
Shawn Bohrer421b3882013-10-07 11:01:39 -05002272 struct dst_entry *dst;
Shawn Bohrer421b3882013-10-07 11:01:39 -05002273 int dif = skb->dev->ifindex;
Shawn Bohrer6e540302015-06-03 16:27:38 -05002274 int ours;
Shawn Bohrer421b3882013-10-07 11:01:39 -05002275
2276 /* validate the packet */
2277 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2278 return;
2279
Eric Dumazet610438b2013-12-11 08:10:05 -08002280 iph = ip_hdr(skb);
2281 uh = udp_hdr(skb);
2282
Shawn Bohrer421b3882013-10-07 11:01:39 -05002283 if (skb->pkt_type == PACKET_BROADCAST ||
Shawn Bohrer6e540302015-06-03 16:27:38 -05002284 skb->pkt_type == PACKET_MULTICAST) {
2285 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
2286
2287 if (!in_dev)
2288 return;
2289
Paolo Abeniad0ea192016-03-22 09:19:38 +01002290 /* we are supposed to accept bcast packets */
2291 if (skb->pkt_type == PACKET_MULTICAST) {
2292 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2293 iph->protocol);
2294 if (!ours)
2295 return;
2296 }
2297
Shawn Bohrer421b3882013-10-07 11:01:39 -05002298 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2299 uh->source, iph->saddr, dif);
Shawn Bohrer6e540302015-06-03 16:27:38 -05002300 } else if (skb->pkt_type == PACKET_HOST) {
Shawn Bohrer421b3882013-10-07 11:01:39 -05002301 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2302 uh->source, iph->saddr, dif);
Shawn Bohrer6e540302015-06-03 16:27:38 -05002303 }
Shawn Bohrer421b3882013-10-07 11:01:39 -05002304
Eric Dumazetca065d02016-04-01 08:52:13 -07002305 if (!sk || !atomic_inc_not_zero_hint(&sk->sk_refcnt, 2))
Shawn Bohrer421b3882013-10-07 11:01:39 -05002306 return;
2307
2308 skb->sk = sk;
Alexander Duyck82eabd92014-09-04 13:32:11 -04002309 skb->destructor = sock_efree;
Eric Dumazet10e2eb82015-08-01 12:14:33 +02002310 dst = READ_ONCE(sk->sk_rx_dst);
Shawn Bohrer421b3882013-10-07 11:01:39 -05002311
2312 if (dst)
2313 dst = dst_check(dst, 0);
Eric Dumazet10e2eb82015-08-01 12:14:33 +02002314 if (dst) {
Wei Wangd24406c2017-06-17 10:42:25 -07002315 /* set noref for now.
2316 * any place which wants to hold dst has to call
2317 * dst_hold_safe()
2318 */
2319 skb_dst_set_noref(skb, dst);
Eric Dumazet10e2eb82015-08-01 12:14:33 +02002320 }
Shawn Bohrer421b3882013-10-07 11:01:39 -05002321}
2322
David S. Millerdb8dac22008-03-06 16:22:02 -08002323int udp_rcv(struct sk_buff *skb)
2324{
Eric Dumazet645ca702008-10-29 01:41:45 -07002325 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
David S. Millerdb8dac22008-03-06 16:22:02 -08002326}
2327
Brian Haley7d06b2e2008-06-14 17:04:49 -07002328void udp_destroy_sock(struct sock *sk)
David S. Millerdb8dac22008-03-06 16:22:02 -08002329{
Tom Parkin44046a52013-03-19 06:11:12 +00002330 struct udp_sock *up = udp_sk(sk);
Eric Dumazet8a74ad62010-05-26 19:20:18 +00002331 bool slow = lock_sock_fast(sk);
David S. Millerdb8dac22008-03-06 16:22:02 -08002332 udp_flush_pending_frames(sk);
Eric Dumazet8a74ad62010-05-26 19:20:18 +00002333 unlock_sock_fast(sk, slow);
Tom Parkin44046a52013-03-19 06:11:12 +00002334 if (static_key_false(&udp_encap_needed) && up->encap_type) {
2335 void (*encap_destroy)(struct sock *sk);
2336 encap_destroy = ACCESS_ONCE(up->encap_destroy);
2337 if (encap_destroy)
2338 encap_destroy(sk);
2339 }
David S. Millerdb8dac22008-03-06 16:22:02 -08002340}
2341
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342/*
2343 * Socket option code for UDP
2344 */
Gerrit Renker4c0a6cb2006-11-27 09:29:59 -08002345int udp_lib_setsockopt(struct sock *sk, int level, int optname,
David S. Millerb7058842009-09-30 16:12:20 -07002346 char __user *optval, unsigned int optlen,
Gerrit Renker4c0a6cb2006-11-27 09:29:59 -08002347 int (*push_pending_frames)(struct sock *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348{
2349 struct udp_sock *up = udp_sk(sk);
Tom Herbert1c194482014-05-23 08:47:32 -07002350 int val, valbool;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351 int err = 0;
Wang Chenb2bf1e22007-12-03 22:34:16 +11002352 int is_udplite = IS_UDPLITE(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353
Eric Dumazetc482c562009-07-17 00:26:32 +00002354 if (optlen < sizeof(int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 return -EINVAL;
2356
2357 if (get_user(val, (int __user *)optval))
2358 return -EFAULT;
2359
Tom Herbert1c194482014-05-23 08:47:32 -07002360 valbool = val ? 1 : 0;
2361
Stephen Hemminger6516c652007-03-08 20:41:55 -08002362 switch (optname) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363 case UDP_CORK:
2364 if (val != 0) {
2365 up->corkflag = 1;
2366 } else {
2367 up->corkflag = 0;
2368 lock_sock(sk);
Joe Perches4243cdc2014-11-11 21:59:20 -08002369 push_pending_frames(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 release_sock(sk);
2371 }
2372 break;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09002373
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 case UDP_ENCAP:
2375 switch (val) {
2376 case 0:
2377 case UDP_ENCAP_ESPINUDP:
2378 case UDP_ENCAP_ESPINUDP_NON_IKE:
James Chapman067b2072007-07-05 17:08:05 -07002379 up->encap_rcv = xfrm4_udp_encap_rcv;
2380 /* FALLTHROUGH */
James Chapman342f0232007-06-27 15:37:46 -07002381 case UDP_ENCAP_L2TPINUDP:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 up->encap_type = val;
Eric Dumazet447167b2012-04-11 23:05:28 +00002383 udp_encap_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 break;
2385 default:
2386 err = -ENOPROTOOPT;
2387 break;
2388 }
2389 break;
2390
Tom Herbert1c194482014-05-23 08:47:32 -07002391 case UDP_NO_CHECK6_TX:
2392 up->no_check6_tx = valbool;
2393 break;
2394
2395 case UDP_NO_CHECK6_RX:
2396 up->no_check6_rx = valbool;
2397 break;
2398
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08002399 /*
2400 * UDP-Lite's partial checksum coverage (RFC 3828).
2401 */
2402 /* The sender sets actual checksum coverage length via this option.
2403 * The case coverage > packet length is handled by send module. */
2404 case UDPLITE_SEND_CSCOV:
Wang Chenb2bf1e22007-12-03 22:34:16 +11002405 if (!is_udplite) /* Disable the option on UDP sockets */
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08002406 return -ENOPROTOOPT;
2407 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2408 val = 8;
Alexey Dobriyan4be929b2010-05-24 14:33:03 -07002409 else if (val > USHRT_MAX)
2410 val = USHRT_MAX;
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08002411 up->pcslen = val;
2412 up->pcflag |= UDPLITE_SEND_CC;
2413 break;
2414
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09002415 /* The receiver specifies a minimum checksum coverage value. To make
2416 * sense, this should be set to at least 8 (as done below). If zero is
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08002417 * used, this again means full checksum coverage. */
2418 case UDPLITE_RECV_CSCOV:
Wang Chenb2bf1e22007-12-03 22:34:16 +11002419 if (!is_udplite) /* Disable the option on UDP sockets */
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08002420 return -ENOPROTOOPT;
2421 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2422 val = 8;
Alexey Dobriyan4be929b2010-05-24 14:33:03 -07002423 else if (val > USHRT_MAX)
2424 val = USHRT_MAX;
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08002425 up->pcrlen = val;
2426 up->pcflag |= UDPLITE_RECV_CC;
2427 break;
2428
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429 default:
2430 err = -ENOPROTOOPT;
2431 break;
Stephen Hemminger6516c652007-03-08 20:41:55 -08002432 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433
2434 return err;
2435}
Eric Dumazetc482c562009-07-17 00:26:32 +00002436EXPORT_SYMBOL(udp_lib_setsockopt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437
David S. Millerdb8dac22008-03-06 16:22:02 -08002438int udp_setsockopt(struct sock *sk, int level, int optname,
David S. Millerb7058842009-09-30 16:12:20 -07002439 char __user *optval, unsigned int optlen)
David S. Millerdb8dac22008-03-06 16:22:02 -08002440{
2441 if (level == SOL_UDP || level == SOL_UDPLITE)
2442 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2443 udp_push_pending_frames);
2444 return ip_setsockopt(sk, level, optname, optval, optlen);
2445}
2446
2447#ifdef CONFIG_COMPAT
2448int compat_udp_setsockopt(struct sock *sk, int level, int optname,
David S. Millerb7058842009-09-30 16:12:20 -07002449 char __user *optval, unsigned int optlen)
David S. Millerdb8dac22008-03-06 16:22:02 -08002450{
2451 if (level == SOL_UDP || level == SOL_UDPLITE)
2452 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2453 udp_push_pending_frames);
2454 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2455}
2456#endif
2457
Gerrit Renker4c0a6cb2006-11-27 09:29:59 -08002458int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2459 char __user *optval, int __user *optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460{
2461 struct udp_sock *up = udp_sk(sk);
2462 int val, len;
2463
Eric Dumazetc482c562009-07-17 00:26:32 +00002464 if (get_user(len, optlen))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 return -EFAULT;
2466
2467 len = min_t(unsigned int, len, sizeof(int));
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09002468
Stephen Hemminger6516c652007-03-08 20:41:55 -08002469 if (len < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 return -EINVAL;
2471
Stephen Hemminger6516c652007-03-08 20:41:55 -08002472 switch (optname) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473 case UDP_CORK:
2474 val = up->corkflag;
2475 break;
2476
2477 case UDP_ENCAP:
2478 val = up->encap_type;
2479 break;
2480
Tom Herbert1c194482014-05-23 08:47:32 -07002481 case UDP_NO_CHECK6_TX:
2482 val = up->no_check6_tx;
2483 break;
2484
2485 case UDP_NO_CHECK6_RX:
2486 val = up->no_check6_rx;
2487 break;
2488
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08002489 /* The following two cannot be changed on UDP sockets, the return is
2490 * always 0 (which corresponds to the full checksum coverage of UDP). */
2491 case UDPLITE_SEND_CSCOV:
2492 val = up->pcslen;
2493 break;
2494
2495 case UDPLITE_RECV_CSCOV:
2496 val = up->pcrlen;
2497 break;
2498
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499 default:
2500 return -ENOPROTOOPT;
Stephen Hemminger6516c652007-03-08 20:41:55 -08002501 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502
Stephen Hemminger6516c652007-03-08 20:41:55 -08002503 if (put_user(len, optlen))
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09002504 return -EFAULT;
Eric Dumazetc482c562009-07-17 00:26:32 +00002505 if (copy_to_user(optval, &val, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 return -EFAULT;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09002507 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508}
Eric Dumazetc482c562009-07-17 00:26:32 +00002509EXPORT_SYMBOL(udp_lib_getsockopt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510
David S. Millerdb8dac22008-03-06 16:22:02 -08002511int udp_getsockopt(struct sock *sk, int level, int optname,
2512 char __user *optval, int __user *optlen)
2513{
2514 if (level == SOL_UDP || level == SOL_UDPLITE)
2515 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2516 return ip_getsockopt(sk, level, optname, optval, optlen);
2517}
2518
2519#ifdef CONFIG_COMPAT
2520int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2521 char __user *optval, int __user *optlen)
2522{
2523 if (level == SOL_UDP || level == SOL_UDPLITE)
2524 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2525 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2526}
2527#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528/**
2529 * udp_poll - wait for a UDP event.
2530 * @file - file struct
2531 * @sock - socket
2532 * @wait - poll table
2533 *
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09002534 * This is same as datagram poll, except for the special case of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535 * blocking sockets. If application is using a blocking fd
2536 * and a packet with checksum error is in the queue;
2537 * then it could get return from select indicating data available
2538 * but then block when reading it. Add special case code
2539 * to work around these arguably broken applications.
2540 */
2541unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2542{
2543 unsigned int mask = datagram_poll(file, sock, wait);
2544 struct sock *sk = sock->sk;
Gerrit Renkerba4e58e2006-11-27 11:10:57 -08002545
Paolo Abeni2276f582017-05-16 11:20:14 +02002546 if (!skb_queue_empty(&udp_sk(sk)->reader_queue))
2547 mask |= POLLIN | POLLRDNORM;
2548
David Majnemerc3f1dba2013-05-31 13:15:38 +00002549 sock_rps_record_flow(sk);
2550
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 /* Check for false positives due to checksum errors */
Eric Dumazet85584672009-10-09 04:43:40 +00002552 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
Eric Dumazete83c6742016-08-23 13:59:33 -07002553 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
Eric Dumazet85584672009-10-09 04:43:40 +00002554 mask &= ~(POLLIN | POLLRDNORM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555
2556 return mask;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09002557
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558}
Eric Dumazetc482c562009-07-17 00:26:32 +00002559EXPORT_SYMBOL(udp_poll);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560
David Ahern5d77dca2016-08-23 21:06:33 -07002561int udp_abort(struct sock *sk, int err)
2562{
2563 lock_sock(sk);
2564
2565 sk->sk_err = err;
2566 sk->sk_error_report(sk);
Eric Dumazet286c72d2016-10-20 09:39:40 -07002567 __udp_disconnect(sk, 0);
David Ahern5d77dca2016-08-23 21:06:33 -07002568
2569 release_sock(sk);
2570
2571 return 0;
2572}
2573EXPORT_SYMBOL_GPL(udp_abort);
2574
David S. Millerdb8dac22008-03-06 16:22:02 -08002575struct proto udp_prot = {
2576 .name = "UDP",
2577 .owner = THIS_MODULE,
2578 .close = udp_lib_close,
2579 .connect = ip4_datagram_connect,
2580 .disconnect = udp_disconnect,
2581 .ioctl = udp_ioctl,
Paolo Abeni850cbad2016-10-21 13:55:47 +02002582 .init = udp_init_sock,
David S. Millerdb8dac22008-03-06 16:22:02 -08002583 .destroy = udp_destroy_sock,
2584 .setsockopt = udp_setsockopt,
2585 .getsockopt = udp_getsockopt,
2586 .sendmsg = udp_sendmsg,
2587 .recvmsg = udp_recvmsg,
2588 .sendpage = udp_sendpage,
Steffen Klassert8141ed92013-01-21 02:00:03 +00002589 .release_cb = ip4_datagram_release_cb,
David S. Millerdb8dac22008-03-06 16:22:02 -08002590 .hash = udp_lib_hash,
2591 .unhash = udp_lib_unhash,
Eric Dumazet719f8352010-09-08 05:08:44 +00002592 .rehash = udp_v4_rehash,
David S. Millerdb8dac22008-03-06 16:22:02 -08002593 .get_port = udp_v4_get_port,
2594 .memory_allocated = &udp_memory_allocated,
2595 .sysctl_mem = sysctl_udp_mem,
2596 .sysctl_wmem = &sysctl_udp_wmem_min,
2597 .sysctl_rmem = &sysctl_udp_rmem_min,
2598 .obj_size = sizeof(struct udp_sock),
Eric Dumazet645ca702008-10-29 01:41:45 -07002599 .h.udp_table = &udp_table,
David S. Millerdb8dac22008-03-06 16:22:02 -08002600#ifdef CONFIG_COMPAT
2601 .compat_setsockopt = compat_udp_setsockopt,
2602 .compat_getsockopt = compat_udp_getsockopt,
2603#endif
David Ahern5d77dca2016-08-23 21:06:33 -07002604 .diag_destroy = udp_abort,
David S. Millerdb8dac22008-03-06 16:22:02 -08002605};
Eric Dumazetc482c562009-07-17 00:26:32 +00002606EXPORT_SYMBOL(udp_prot);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607
2608/* ------------------------------------------------------------------------ */
2609#ifdef CONFIG_PROC_FS
2610
Eric Dumazet645ca702008-10-29 01:41:45 -07002611static struct sock *udp_get_first(struct seq_file *seq, int start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612{
2613 struct sock *sk;
2614 struct udp_iter_state *state = seq->private;
Denis V. Lunev6f191ef2008-03-28 18:23:33 -07002615 struct net *net = seq_file_net(seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002617 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2618 ++state->bucket) {
Eric Dumazet645ca702008-10-29 01:41:45 -07002619 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002620
Eric Dumazetca065d02016-04-01 08:52:13 -07002621 if (hlist_empty(&hslot->head))
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002622 continue;
2623
Eric Dumazet645ca702008-10-29 01:41:45 -07002624 spin_lock_bh(&hslot->lock);
Eric Dumazetca065d02016-04-01 08:52:13 -07002625 sk_for_each(sk, &hslot->head) {
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +09002626 if (!net_eq(sock_net(sk), net))
Daniel Lezcanoa91275e2008-03-21 04:11:58 -07002627 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 if (sk->sk_family == state->family)
2629 goto found;
2630 }
Eric Dumazet645ca702008-10-29 01:41:45 -07002631 spin_unlock_bh(&hslot->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 }
2633 sk = NULL;
2634found:
2635 return sk;
2636}
2637
2638static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2639{
2640 struct udp_iter_state *state = seq->private;
Denis V. Lunev6f191ef2008-03-28 18:23:33 -07002641 struct net *net = seq_file_net(seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642
2643 do {
Eric Dumazetca065d02016-04-01 08:52:13 -07002644 sk = sk_next(sk);
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +09002645 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646
Eric Dumazet645ca702008-10-29 01:41:45 -07002647 if (!sk) {
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002648 if (state->bucket <= state->udp_table->mask)
Vitaly Mayatskikh30842f22009-03-23 15:22:33 -07002649 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
Eric Dumazet645ca702008-10-29 01:41:45 -07002650 return udp_get_first(seq, state->bucket + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651 }
2652 return sk;
2653}
2654
2655static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2656{
Eric Dumazet645ca702008-10-29 01:41:45 -07002657 struct sock *sk = udp_get_first(seq, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658
2659 if (sk)
Stephen Hemminger6516c652007-03-08 20:41:55 -08002660 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661 --pos;
2662 return pos ? NULL : sk;
2663}
2664
2665static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2666{
Vitaly Mayatskikh30842f22009-03-23 15:22:33 -07002667 struct udp_iter_state *state = seq->private;
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002668 state->bucket = MAX_UDP_PORTS;
Vitaly Mayatskikh30842f22009-03-23 15:22:33 -07002669
YOSHIFUJI Hideakib50660f2008-03-31 19:38:15 -07002670 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671}
2672
2673static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2674{
2675 struct sock *sk;
2676
YOSHIFUJI Hideakib50660f2008-03-31 19:38:15 -07002677 if (v == SEQ_START_TOKEN)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678 sk = udp_get_idx(seq, 0);
2679 else
2680 sk = udp_get_next(seq, v);
2681
2682 ++*pos;
2683 return sk;
2684}
2685
2686static void udp_seq_stop(struct seq_file *seq, void *v)
2687{
Eric Dumazet645ca702008-10-29 01:41:45 -07002688 struct udp_iter_state *state = seq->private;
2689
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002690 if (state->bucket <= state->udp_table->mask)
Eric Dumazet645ca702008-10-29 01:41:45 -07002691 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692}
2693
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00002694int udp_seq_open(struct inode *inode, struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695{
Al Virod9dda782013-03-31 18:16:14 -04002696 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
Denis V. Luneva2be75c2008-03-28 18:25:06 -07002697 struct udp_iter_state *s;
2698 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699
Denis V. Luneva2be75c2008-03-28 18:25:06 -07002700 err = seq_open_net(inode, file, &afinfo->seq_ops,
2701 sizeof(struct udp_iter_state));
2702 if (err < 0)
2703 return err;
Daniel Lezcanoa91275e2008-03-21 04:11:58 -07002704
Denis V. Luneva2be75c2008-03-28 18:25:06 -07002705 s = ((struct seq_file *)file->private_data)->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 s->family = afinfo->family;
Eric Dumazet645ca702008-10-29 01:41:45 -07002707 s->udp_table = afinfo->udp_table;
Denis V. Luneva2be75c2008-03-28 18:25:06 -07002708 return err;
Daniel Lezcanoa91275e2008-03-21 04:11:58 -07002709}
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00002710EXPORT_SYMBOL(udp_seq_open);
Daniel Lezcanoa91275e2008-03-21 04:11:58 -07002711
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712/* ------------------------------------------------------------------------ */
Daniel Lezcano0c96d8c2008-03-21 04:14:17 -07002713int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714{
2715 struct proc_dir_entry *p;
2716 int rc = 0;
2717
Denis V. Lunevdda61922008-03-28 18:24:26 -07002718 afinfo->seq_ops.start = udp_seq_start;
2719 afinfo->seq_ops.next = udp_seq_next;
2720 afinfo->seq_ops.stop = udp_seq_stop;
2721
Denis V. Lunev84841c32008-05-02 04:10:08 -07002722 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00002723 afinfo->seq_fops, afinfo);
Denis V. Lunev84841c32008-05-02 04:10:08 -07002724 if (!p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725 rc = -ENOMEM;
2726 return rc;
2727}
Eric Dumazetc482c562009-07-17 00:26:32 +00002728EXPORT_SYMBOL(udp_proc_register);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729
Daniel Lezcano0c96d8c2008-03-21 04:14:17 -07002730void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731{
Gao fengece31ff2013-02-18 01:34:56 +00002732 remove_proc_entry(afinfo->name, net->proc_net);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733}
Eric Dumazetc482c562009-07-17 00:26:32 +00002734EXPORT_SYMBOL(udp_proc_unregister);
David S. Millerdb8dac22008-03-06 16:22:02 -08002735
2736/* ------------------------------------------------------------------------ */
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002737static void udp4_format_sock(struct sock *sp, struct seq_file *f,
Tetsuo Handa652586d2013-11-14 14:31:57 -08002738 int bucket)
David S. Millerdb8dac22008-03-06 16:22:02 -08002739{
2740 struct inet_sock *inet = inet_sk(sp);
Eric Dumazetc720c7e82009-10-15 06:30:45 +00002741 __be32 dest = inet->inet_daddr;
2742 __be32 src = inet->inet_rcv_saddr;
2743 __u16 destp = ntohs(inet->inet_dport);
2744 __u16 srcp = ntohs(inet->inet_sport);
David S. Millerdb8dac22008-03-06 16:22:02 -08002745
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002746 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
Tetsuo Handa652586d2013-11-14 14:31:57 -08002747 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
David S. Millerdb8dac22008-03-06 16:22:02 -08002748 bucket, src, srcp, dest, destp, sp->sk_state,
Eric Dumazet31e6d362009-06-17 19:05:41 -07002749 sk_wmem_alloc_get(sp),
2750 sk_rmem_alloc_get(sp),
Eric W. Biedermana7cb5a42012-05-24 01:10:10 -06002751 0, 0L, 0,
2752 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2753 0, sock_i_ino(sp),
Eric Dumazetcb61cb92008-06-17 21:04:56 -07002754 atomic_read(&sp->sk_refcnt), sp,
Tetsuo Handa652586d2013-11-14 14:31:57 -08002755 atomic_read(&sp->sk_drops));
David S. Millerdb8dac22008-03-06 16:22:02 -08002756}
2757
2758int udp4_seq_show(struct seq_file *seq, void *v)
2759{
Tetsuo Handa652586d2013-11-14 14:31:57 -08002760 seq_setwidth(seq, 127);
David S. Millerdb8dac22008-03-06 16:22:02 -08002761 if (v == SEQ_START_TOKEN)
Tetsuo Handa652586d2013-11-14 14:31:57 -08002762 seq_puts(seq, " sl local_address rem_address st tx_queue "
David S. Millerdb8dac22008-03-06 16:22:02 -08002763 "rx_queue tr tm->when retrnsmt uid timeout "
Eric Dumazetcb61cb92008-06-17 21:04:56 -07002764 "inode ref pointer drops");
David S. Millerdb8dac22008-03-06 16:22:02 -08002765 else {
David S. Millerdb8dac22008-03-06 16:22:02 -08002766 struct udp_iter_state *state = seq->private;
2767
Tetsuo Handa652586d2013-11-14 14:31:57 -08002768 udp4_format_sock(v, seq, state->bucket);
David S. Millerdb8dac22008-03-06 16:22:02 -08002769 }
Tetsuo Handa652586d2013-11-14 14:31:57 -08002770 seq_pad(seq, '\n');
David S. Millerdb8dac22008-03-06 16:22:02 -08002771 return 0;
2772}
2773
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00002774static const struct file_operations udp_afinfo_seq_fops = {
2775 .owner = THIS_MODULE,
2776 .open = udp_seq_open,
2777 .read = seq_read,
2778 .llseek = seq_lseek,
2779 .release = seq_release_net
2780};
2781
David S. Millerdb8dac22008-03-06 16:22:02 -08002782/* ------------------------------------------------------------------------ */
David S. Millerdb8dac22008-03-06 16:22:02 -08002783static struct udp_seq_afinfo udp4_seq_afinfo = {
David S. Millerdb8dac22008-03-06 16:22:02 -08002784 .name = "udp",
2785 .family = AF_INET,
Eric Dumazet645ca702008-10-29 01:41:45 -07002786 .udp_table = &udp_table,
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00002787 .seq_fops = &udp_afinfo_seq_fops,
Denis V. Lunevdda61922008-03-28 18:24:26 -07002788 .seq_ops = {
2789 .show = udp4_seq_show,
2790 },
David S. Millerdb8dac22008-03-06 16:22:02 -08002791};
2792
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00002793static int __net_init udp4_proc_init_net(struct net *net)
Pavel Emelyanov15439fe2008-03-24 14:53:49 -07002794{
2795 return udp_proc_register(net, &udp4_seq_afinfo);
2796}
2797
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00002798static void __net_exit udp4_proc_exit_net(struct net *net)
Pavel Emelyanov15439fe2008-03-24 14:53:49 -07002799{
2800 udp_proc_unregister(net, &udp4_seq_afinfo);
2801}
2802
2803static struct pernet_operations udp4_net_ops = {
2804 .init = udp4_proc_init_net,
2805 .exit = udp4_proc_exit_net,
2806};
2807
David S. Millerdb8dac22008-03-06 16:22:02 -08002808int __init udp4_proc_init(void)
2809{
Pavel Emelyanov15439fe2008-03-24 14:53:49 -07002810 return register_pernet_subsys(&udp4_net_ops);
David S. Millerdb8dac22008-03-06 16:22:02 -08002811}
2812
2813void udp4_proc_exit(void)
2814{
Pavel Emelyanov15439fe2008-03-24 14:53:49 -07002815 unregister_pernet_subsys(&udp4_net_ops);
David S. Millerdb8dac22008-03-06 16:22:02 -08002816}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817#endif /* CONFIG_PROC_FS */
2818
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002819static __initdata unsigned long uhash_entries;
2820static int __init set_uhash_entries(char *str)
Eric Dumazet645ca702008-10-29 01:41:45 -07002821{
Eldad Zack413c27d2012-05-19 14:13:18 +00002822 ssize_t ret;
2823
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002824 if (!str)
2825 return 0;
Eldad Zack413c27d2012-05-19 14:13:18 +00002826
2827 ret = kstrtoul(str, 0, &uhash_entries);
2828 if (ret)
2829 return 0;
2830
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002831 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2832 uhash_entries = UDP_HTABLE_SIZE_MIN;
2833 return 1;
2834}
2835__setup("uhash_entries=", set_uhash_entries);
Eric Dumazet645ca702008-10-29 01:41:45 -07002836
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002837void __init udp_table_init(struct udp_table *table, const char *name)
2838{
2839 unsigned int i;
2840
Tim Bird31fe62b2012-05-23 13:33:35 +00002841 table->hash = alloc_large_system_hash(name,
2842 2 * sizeof(struct udp_hslot),
2843 uhash_entries,
2844 21, /* one slot per 2 MB */
2845 0,
2846 &table->log,
2847 &table->mask,
2848 UDP_HTABLE_SIZE_MIN,
2849 64 * 1024);
2850
Eric Dumazet512615b2009-11-08 10:17:58 +00002851 table->hash2 = table->hash + (table->mask + 1);
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002852 for (i = 0; i <= table->mask; i++) {
Eric Dumazetca065d02016-04-01 08:52:13 -07002853 INIT_HLIST_HEAD(&table->hash[i].head);
Eric Dumazetfdcc8aa92009-11-08 10:17:05 +00002854 table->hash[i].count = 0;
Eric Dumazet645ca702008-10-29 01:41:45 -07002855 spin_lock_init(&table->hash[i].lock);
2856 }
Eric Dumazet512615b2009-11-08 10:17:58 +00002857 for (i = 0; i <= table->mask; i++) {
Eric Dumazetca065d02016-04-01 08:52:13 -07002858 INIT_HLIST_HEAD(&table->hash2[i].head);
Eric Dumazet512615b2009-11-08 10:17:58 +00002859 table->hash2[i].count = 0;
2860 spin_lock_init(&table->hash2[i].lock);
2861 }
Eric Dumazet645ca702008-10-29 01:41:45 -07002862}
2863
Tom Herbert723b8e42015-02-24 09:17:31 -08002864u32 udp_flow_hashrnd(void)
2865{
2866 static u32 hashrnd __read_mostly;
2867
2868 net_get_random_once(&hashrnd, sizeof(hashrnd));
2869
2870 return hashrnd;
2871}
2872EXPORT_SYMBOL(udp_flow_hashrnd);
2873
Hideo Aoki95766ff2007-12-31 00:29:24 -08002874void __init udp_init(void)
2875{
Eric Dumazetf03d78d2011-07-07 00:27:05 -07002876 unsigned long limit;
Eric Dumazet4b272752016-12-08 11:41:54 -08002877 unsigned int i;
Hideo Aoki95766ff2007-12-31 00:29:24 -08002878
Eric Dumazetf86dcc52009-10-07 00:37:59 +00002879 udp_table_init(&udp_table, "UDP");
Eric Dumazetf03d78d2011-07-07 00:27:05 -07002880 limit = nr_free_buffer_pages() / 8;
Hideo Aoki95766ff2007-12-31 00:29:24 -08002881 limit = max(limit, 128UL);
2882 sysctl_udp_mem[0] = limit / 4 * 3;
2883 sysctl_udp_mem[1] = limit;
2884 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2885
2886 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2887 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
Eric Dumazet4b272752016-12-08 11:41:54 -08002888
2889 /* 16 spinlocks per cpu */
2890 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
2891 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
2892 GFP_KERNEL);
2893 if (!udp_busylocks)
2894 panic("UDP: failed to alloc udp_busylocks\n");
2895 for (i = 0; i < (1U << udp_busylocks_log); i++)
2896 spin_lock_init(udp_busylocks + i);
Hideo Aoki95766ff2007-12-31 00:29:24 -08002897}