blob: a066fdd50da6c5041a12bb4983eeb4b9c8b9eee1 [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 Meloeeb2b852005-10-10 21:25:23 -070021#include <linux/module.h>
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070022#include <linux/timer.h>
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -070023#include <linux/types.h>
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070024#include <linux/workqueue.h>
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -070025
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -020026#include <net/inet_sock.h>
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -070027#include <net/sock.h>
28#include <net/tcp_states.h>
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -080029#include <net/timewait_sock.h>
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -070030
31#include <asm/atomic.h>
32
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070033struct inet_hashinfo;
34
35#define INET_TWDR_RECYCLE_SLOTS_LOG 5
36#define INET_TWDR_RECYCLE_SLOTS (1 << INET_TWDR_RECYCLE_SLOTS_LOG)
37
38/*
39 * If time > 4sec, it is "slow" path, no recycling is required,
40 * so that we select tick to get range about 4 seconds.
41 */
42#if HZ <= 16 || HZ > 4096
43# error Unsupported: HZ <= 16 or HZ > 4096
44#elif HZ <= 32
45# define INET_TWDR_RECYCLE_TICK (5 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
46#elif HZ <= 64
47# define INET_TWDR_RECYCLE_TICK (6 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
48#elif HZ <= 128
49# define INET_TWDR_RECYCLE_TICK (7 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
50#elif HZ <= 256
51# define INET_TWDR_RECYCLE_TICK (8 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
52#elif HZ <= 512
53# define INET_TWDR_RECYCLE_TICK (9 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
54#elif HZ <= 1024
55# define INET_TWDR_RECYCLE_TICK (10 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
56#elif HZ <= 2048
57# define INET_TWDR_RECYCLE_TICK (11 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
58#else
59# define INET_TWDR_RECYCLE_TICK (12 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
60#endif
61
62/* TIME_WAIT reaping mechanism. */
63#define INET_TWDR_TWKILL_SLOTS 8 /* Please keep this a power of 2. */
64
65#define INET_TWDR_TWKILL_QUOTA 100
66
67struct inet_timewait_death_row {
68 /* Short-time timewait calendar */
69 int twcal_hand;
Eric Dumazet187f5f842007-03-05 13:32:48 -080070 unsigned long twcal_jiffie;
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070071 struct timer_list twcal_timer;
72 struct hlist_head twcal_row[INET_TWDR_RECYCLE_SLOTS];
73
74 spinlock_t death_lock;
75 int tw_count;
76 int period;
77 u32 thread_slots;
78 struct work_struct twkill_work;
79 struct timer_list tw_timer;
80 int slot;
81 struct hlist_head cells[INET_TWDR_TWKILL_SLOTS];
82 struct inet_hashinfo *hashinfo;
83 int sysctl_tw_recycle;
84 int sysctl_max_tw_buckets;
85};
86
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -070087extern void inet_twdr_hangman(unsigned long data);
David Howells65f27f32006-11-22 14:55:48 +000088extern void inet_twdr_twkill_work(struct work_struct *work);
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -070089extern void inet_twdr_twcal_tick(unsigned long data);
90
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -070091#if (BITS_PER_LONG == 64)
92#define INET_TIMEWAIT_ADDRCMP_ALIGN_BYTES 8
93#else
94#define INET_TIMEWAIT_ADDRCMP_ALIGN_BYTES 4
95#endif
96
97struct inet_bind_bucket;
98
99/*
100 * This is a TIME_WAIT sock. It works around the memory consumption
101 * problems of sockets in such a state on heavily loaded servers, but
102 * without violating the protocol specification.
103 */
104struct inet_timewait_sock {
105 /*
106 * Now struct sock also uses sock_common, so please just
107 * don't add nothing before this first member (__tw_common) --acme
108 */
109 struct sock_common __tw_common;
110#define tw_family __tw_common.skc_family
111#define tw_state __tw_common.skc_state
112#define tw_reuse __tw_common.skc_reuse
113#define tw_bound_dev_if __tw_common.skc_bound_dev_if
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800114#define tw_node __tw_common.skc_nulls_node
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700115#define tw_bind_node __tw_common.skc_bind_node
116#define tw_refcnt __tw_common.skc_refcnt
Eric Dumazet81c3d542005-10-03 14:13:38 -0700117#define tw_hash __tw_common.skc_hash
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700118#define tw_prot __tw_common.skc_prot
Eric W. Biederman07feaeb2007-09-12 11:58:02 +0200119#define tw_net __tw_common.skc_net
Arnaldo Carvalho de Meload8bb782008-02-03 04:08:26 -0800120 int tw_timeout;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700121 volatile unsigned char tw_substate;
122 /* 3 bits hole, try to pack */
123 unsigned char tw_rcv_wscale;
124 /* Socket demultiplex comparisons on incoming packets. */
125 /* these five are in inet_sock */
Al Viro23f33c22006-09-27 18:43:50 -0700126 __be16 tw_sport;
127 __be32 tw_daddr __attribute__((aligned(INET_TIMEWAIT_ADDRCMP_ALIGN_BYTES)));
128 __be32 tw_rcv_saddr;
129 __be16 tw_dport;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700130 __u16 tw_num;
Vegard Nossum9e337b02008-10-18 17:37:51 +0200131 kmemcheck_bitfield_begin(flags);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700132 /* And these are ours. */
Eric Dumazetabf90cc2009-10-18 22:48:51 +0000133 unsigned int tw_ipv6only : 1,
134 tw_transparent : 1,
135 tw_pad : 14, /* 14 bits hole */
136 tw_ipv6_offset : 16;
Vegard Nossum9e337b02008-10-18 17:37:51 +0200137 kmemcheck_bitfield_end(flags);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700138 unsigned long tw_ttd;
139 struct inet_bind_bucket *tw_tb;
140 struct hlist_node tw_death_node;
141};
142
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800143static inline void inet_twsk_add_node_rcu(struct inet_timewait_sock *tw,
144 struct hlist_nulls_head *list)
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700145{
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800146 hlist_nulls_add_head_rcu(&tw->tw_node, list);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700147}
148
149static inline void inet_twsk_add_bind_node(struct inet_timewait_sock *tw,
150 struct hlist_head *list)
151{
152 hlist_add_head(&tw->tw_bind_node, list);
153}
154
155static inline int inet_twsk_dead_hashed(const struct inet_timewait_sock *tw)
156{
Akinobu Mitada753be2006-04-28 15:21:23 -0700157 return !hlist_unhashed(&tw->tw_death_node);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700158}
159
160static inline void inet_twsk_dead_node_init(struct inet_timewait_sock *tw)
161{
162 tw->tw_death_node.pprev = NULL;
163}
164
165static inline void __inet_twsk_del_dead_node(struct inet_timewait_sock *tw)
166{
167 __hlist_del(&tw->tw_death_node);
168 inet_twsk_dead_node_init(tw);
169}
170
171static inline int inet_twsk_del_dead_node(struct inet_timewait_sock *tw)
172{
173 if (inet_twsk_dead_hashed(tw)) {
174 __inet_twsk_del_dead_node(tw);
175 return 1;
176 }
177 return 0;
178}
179
180#define inet_twsk_for_each(tw, node, head) \
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800181 hlist_nulls_for_each_entry(tw, node, head, tw_node)
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700182
183#define inet_twsk_for_each_inmate(tw, node, jail) \
184 hlist_for_each_entry(tw, node, jail, tw_death_node)
185
186#define inet_twsk_for_each_inmate_safe(tw, node, safe, jail) \
187 hlist_for_each_entry_safe(tw, node, safe, jail, tw_death_node)
188
189static inline struct inet_timewait_sock *inet_twsk(const struct sock *sk)
190{
191 return (struct inet_timewait_sock *)sk;
192}
193
Al Viro82103232006-09-27 18:44:10 -0700194static inline __be32 inet_rcv_saddr(const struct sock *sk)
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700195{
196 return likely(sk->sk_state != TCP_TIME_WAIT) ?
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000197 inet_sk(sk)->inet_rcv_saddr : inet_twsk(sk)->tw_rcv_saddr;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700198}
199
Pavel Emelyanov7054fb92007-12-20 15:32:54 -0800200extern void inet_twsk_put(struct inet_timewait_sock *tw);
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700201
Eric Dumazet13475a32009-12-02 22:31:19 +0000202extern int inet_twsk_unhash(struct inet_timewait_sock *tw);
203
Eric Dumazet3cdaeda2009-12-04 03:47:42 +0000204extern int inet_twsk_bind_unhash(struct inet_timewait_sock *tw,
205 struct inet_hashinfo *hashinfo);
206
Arnaldo Carvalho de Meloc6762702005-08-09 20:09:59 -0700207extern struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk,
208 const int state);
209
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700210extern void __inet_twsk_hashdance(struct inet_timewait_sock *tw,
211 struct sock *sk,
212 struct inet_hashinfo *hashinfo);
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700213
214extern void inet_twsk_schedule(struct inet_timewait_sock *tw,
215 struct inet_timewait_death_row *twdr,
216 const int timeo, const int timewait_len);
217extern void inet_twsk_deschedule(struct inet_timewait_sock *tw,
218 struct inet_timewait_death_row *twdr);
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900219
Eric W. Biedermanb099ce22009-12-03 02:29:09 +0000220extern void inet_twsk_purge(struct inet_hashinfo *hashinfo,
Daniel Lezcanod3154922008-09-08 13:17:27 -0700221 struct inet_timewait_death_row *twdr, int family);
222
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900223static inline
224struct net *twsk_net(const struct inet_timewait_sock *twsk)
225{
226#ifdef CONFIG_NET_NS
Paul E. McKenney7ec75c52010-04-27 06:22:01 +0000227 return rcu_dereference_raw(twsk->tw_net); /* protected by locking, */
228 /* reference counting, */
229 /* initialization, or RCU. */
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900230#else
231 return &init_net;
232#endif
233}
234
235static inline
Denis V. Lunevf5aa23f2008-03-26 00:48:17 -0700236void twsk_net_set(struct inet_timewait_sock *twsk, struct net *net)
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900237{
238#ifdef CONFIG_NET_NS
Eric W. Biedermanb099ce22009-12-03 02:29:09 +0000239 rcu_assign_pointer(twsk->tw_net, net);
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900240#endif
241}
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700242#endif /* _INET_TIMEWAIT_SOCK_ */