blob: 61474ea02152dc0886901044b4fd210ee5d9596d [file] [log] [blame]
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -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 * Definitions for a generic INET TIMEWAIT sock
7 *
8 * From code originally in net/tcp.h
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
14 */
15#ifndef _INET_TIMEWAIT_SOCK_
16#define _INET_TIMEWAIT_SOCK_
17
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -070018
Vegard Nossum9e337b02008-10-18 17:37:51 +020019#include <linux/kmemcheck.h>
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -070020#include <linux/list.h>
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070021#include <linux/timer.h>
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -070022#include <linux/types.h>
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070023#include <linux/workqueue.h>
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -070024
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -020025#include <net/inet_sock.h>
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -070026#include <net/sock.h>
27#include <net/tcp_states.h>
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -080028#include <net/timewait_sock.h>
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -070029
Arun Sharma600634972011-07-26 16:09:06 -070030#include <linux/atomic.h>
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -070031
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070032struct inet_hashinfo;
33
34#define INET_TWDR_RECYCLE_SLOTS_LOG 5
35#define INET_TWDR_RECYCLE_SLOTS (1 << INET_TWDR_RECYCLE_SLOTS_LOG)
36
37/*
38 * If time > 4sec, it is "slow" path, no recycling is required,
39 * so that we select tick to get range about 4 seconds.
40 */
41#if HZ <= 16 || HZ > 4096
42# error Unsupported: HZ <= 16 or HZ > 4096
43#elif HZ <= 32
44# define INET_TWDR_RECYCLE_TICK (5 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
45#elif HZ <= 64
46# define INET_TWDR_RECYCLE_TICK (6 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
47#elif HZ <= 128
48# define INET_TWDR_RECYCLE_TICK (7 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
49#elif HZ <= 256
50# define INET_TWDR_RECYCLE_TICK (8 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
51#elif HZ <= 512
52# define INET_TWDR_RECYCLE_TICK (9 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
53#elif HZ <= 1024
54# define INET_TWDR_RECYCLE_TICK (10 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
55#elif HZ <= 2048
56# define INET_TWDR_RECYCLE_TICK (11 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
57#else
58# define INET_TWDR_RECYCLE_TICK (12 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
59#endif
60
Eric Dumazet96f817f2013-10-03 14:27:25 -070061static inline u32 inet_tw_time_stamp(void)
62{
63 return jiffies;
64}
65
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070066/* TIME_WAIT reaping mechanism. */
67#define INET_TWDR_TWKILL_SLOTS 8 /* Please keep this a power of 2. */
68
69#define INET_TWDR_TWKILL_QUOTA 100
70
71struct inet_timewait_death_row {
72 /* Short-time timewait calendar */
73 int twcal_hand;
Eric Dumazet187f5f842007-03-05 13:32:48 -080074 unsigned long twcal_jiffie;
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070075 struct timer_list twcal_timer;
76 struct hlist_head twcal_row[INET_TWDR_RECYCLE_SLOTS];
77
78 spinlock_t death_lock;
79 int tw_count;
80 int period;
81 u32 thread_slots;
82 struct work_struct twkill_work;
83 struct timer_list tw_timer;
84 int slot;
85 struct hlist_head cells[INET_TWDR_TWKILL_SLOTS];
86 struct inet_hashinfo *hashinfo;
87 int sysctl_tw_recycle;
88 int sysctl_max_tw_buckets;
89};
90
Joe Perches1fd51152013-09-21 10:22:41 -070091void inet_twdr_hangman(unsigned long data);
92void inet_twdr_twkill_work(struct work_struct *work);
93void inet_twdr_twcal_tick(unsigned long data);
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -070094
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -070095struct inet_bind_bucket;
96
97/*
98 * This is a TIME_WAIT sock. It works around the memory consumption
99 * problems of sockets in such a state on heavily loaded servers, but
100 * without violating the protocol specification.
101 */
102struct inet_timewait_sock {
103 /*
104 * Now struct sock also uses sock_common, so please just
105 * don't add nothing before this first member (__tw_common) --acme
106 */
107 struct sock_common __tw_common;
108#define tw_family __tw_common.skc_family
109#define tw_state __tw_common.skc_state
110#define tw_reuse __tw_common.skc_reuse
111#define tw_bound_dev_if __tw_common.skc_bound_dev_if
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800112#define tw_node __tw_common.skc_nulls_node
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700113#define tw_bind_node __tw_common.skc_bind_node
114#define tw_refcnt __tw_common.skc_refcnt
Eric Dumazet81c3d542005-10-03 14:13:38 -0700115#define tw_hash __tw_common.skc_hash
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700116#define tw_prot __tw_common.skc_prot
Eric W. Biederman07feaeb2007-09-12 11:58:02 +0200117#define tw_net __tw_common.skc_net
Eric Dumazet68835ab2010-11-30 19:04:07 +0000118#define tw_daddr __tw_common.skc_daddr
Eric Dumazetefe42082013-10-03 15:42:29 -0700119#define tw_v6_daddr __tw_common.skc_v6_daddr
Eric Dumazet68835ab2010-11-30 19:04:07 +0000120#define tw_rcv_saddr __tw_common.skc_rcv_saddr
Eric Dumazetefe42082013-10-03 15:42:29 -0700121#define tw_v6_rcv_saddr __tw_common.skc_v6_rcv_saddr
Eric Dumazetce43b032012-11-30 09:49:27 +0000122#define tw_dport __tw_common.skc_dport
123#define tw_num __tw_common.skc_num
Eric Dumazetce43b032012-11-30 09:49:27 +0000124
Arnaldo Carvalho de Meload8bb782008-02-03 04:08:26 -0800125 int tw_timeout;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700126 volatile unsigned char tw_substate;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700127 unsigned char tw_rcv_wscale;
Eric Dumazet68835ab2010-11-30 19:04:07 +0000128
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700129 /* Socket demultiplex comparisons on incoming packets. */
Eric Dumazet68835ab2010-11-30 19:04:07 +0000130 /* these three are in inet_sock */
Al Viro23f33c22006-09-27 18:43:50 -0700131 __be16 tw_sport;
Vegard Nossum9e337b02008-10-18 17:37:51 +0200132 kmemcheck_bitfield_begin(flags);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700133 /* And these are ours. */
Eric Dumazetabf90cc2009-10-18 22:48:51 +0000134 unsigned int tw_ipv6only : 1,
135 tw_transparent : 1,
Florent Fourcot1d13a962014-01-16 17:21:22 +0100136 tw_flowlabel : 20,
137 tw_pad : 2, /* 2 bits hole */
138 tw_tos : 8;
Vegard Nossum9e337b02008-10-18 17:37:51 +0200139 kmemcheck_bitfield_end(flags);
Eric Dumazet96f817f2013-10-03 14:27:25 -0700140 u32 tw_ttd;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700141 struct inet_bind_bucket *tw_tb;
142 struct hlist_node tw_death_node;
143};
Eric Dumazetb903d322011-10-27 00:44:35 -0400144#define tw_tclass tw_tos
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700145
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700146static inline int inet_twsk_dead_hashed(const struct inet_timewait_sock *tw)
147{
Akinobu Mitada753be2006-04-28 15:21:23 -0700148 return !hlist_unhashed(&tw->tw_death_node);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700149}
150
151static inline void inet_twsk_dead_node_init(struct inet_timewait_sock *tw)
152{
153 tw->tw_death_node.pprev = NULL;
154}
155
156static inline void __inet_twsk_del_dead_node(struct inet_timewait_sock *tw)
157{
158 __hlist_del(&tw->tw_death_node);
159 inet_twsk_dead_node_init(tw);
160}
161
162static inline int inet_twsk_del_dead_node(struct inet_timewait_sock *tw)
163{
164 if (inet_twsk_dead_hashed(tw)) {
165 __inet_twsk_del_dead_node(tw);
166 return 1;
167 }
168 return 0;
169}
170
171#define inet_twsk_for_each(tw, node, head) \
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800172 hlist_nulls_for_each_entry(tw, node, head, tw_node)
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700173
Sasha Levinb67bfe02013-02-27 17:06:00 -0800174#define inet_twsk_for_each_inmate(tw, jail) \
175 hlist_for_each_entry(tw, jail, tw_death_node)
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700176
Sasha Levinb67bfe02013-02-27 17:06:00 -0800177#define inet_twsk_for_each_inmate_safe(tw, safe, jail) \
178 hlist_for_each_entry_safe(tw, safe, jail, tw_death_node)
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700179
180static inline struct inet_timewait_sock *inet_twsk(const struct sock *sk)
181{
182 return (struct inet_timewait_sock *)sk;
183}
184
Eric Dumazet05dbc7b2013-10-03 00:22:02 -0700185void inet_twsk_free(struct inet_timewait_sock *tw);
Joe Perches1fd51152013-09-21 10:22:41 -0700186void inet_twsk_put(struct inet_timewait_sock *tw);
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700187
Joe Perches1fd51152013-09-21 10:22:41 -0700188int inet_twsk_unhash(struct inet_timewait_sock *tw);
Eric Dumazet13475a32009-12-02 22:31:19 +0000189
Joe Perches1fd51152013-09-21 10:22:41 -0700190int inet_twsk_bind_unhash(struct inet_timewait_sock *tw,
191 struct inet_hashinfo *hashinfo);
Eric Dumazet3cdaeda2009-12-04 03:47:42 +0000192
Joe Perches1fd51152013-09-21 10:22:41 -0700193struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk,
194 const int state);
Arnaldo Carvalho de Meloc6762702005-08-09 20:09:59 -0700195
Joe Perches1fd51152013-09-21 10:22:41 -0700196void __inet_twsk_hashdance(struct inet_timewait_sock *tw, struct sock *sk,
197 struct inet_hashinfo *hashinfo);
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700198
Joe Perches1fd51152013-09-21 10:22:41 -0700199void inet_twsk_schedule(struct inet_timewait_sock *tw,
200 struct inet_timewait_death_row *twdr,
201 const int timeo, const int timewait_len);
202void inet_twsk_deschedule(struct inet_timewait_sock *tw,
203 struct inet_timewait_death_row *twdr);
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900204
Joe Perches1fd51152013-09-21 10:22:41 -0700205void inet_twsk_purge(struct inet_hashinfo *hashinfo,
206 struct inet_timewait_death_row *twdr, int family);
Daniel Lezcanod3154922008-09-08 13:17:27 -0700207
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900208static inline
209struct net *twsk_net(const struct inet_timewait_sock *twsk)
210{
Eric Dumazetf943cbe2011-12-14 04:58:33 +0000211 return read_pnet(&twsk->tw_net);
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900212}
213
214static inline
Denis V. Lunevf5aa23f2008-03-26 00:48:17 -0700215void twsk_net_set(struct inet_timewait_sock *twsk, struct net *net)
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900216{
Eric Dumazetf943cbe2011-12-14 04:58:33 +0000217 write_pnet(&twsk->tw_net, net);
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900218}
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700219#endif /* _INET_TIMEWAIT_SOCK_ */