blob: e8c25b98120523711f115867b247995e1db80a57 [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
61/* TIME_WAIT reaping mechanism. */
62#define INET_TWDR_TWKILL_SLOTS 8 /* Please keep this a power of 2. */
63
64#define INET_TWDR_TWKILL_QUOTA 100
65
66struct inet_timewait_death_row {
67 /* Short-time timewait calendar */
68 int twcal_hand;
Eric Dumazet187f5f842007-03-05 13:32:48 -080069 unsigned long twcal_jiffie;
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070070 struct timer_list twcal_timer;
71 struct hlist_head twcal_row[INET_TWDR_RECYCLE_SLOTS];
72
73 spinlock_t death_lock;
74 int tw_count;
75 int period;
76 u32 thread_slots;
77 struct work_struct twkill_work;
78 struct timer_list tw_timer;
79 int slot;
80 struct hlist_head cells[INET_TWDR_TWKILL_SLOTS];
81 struct inet_hashinfo *hashinfo;
82 int sysctl_tw_recycle;
83 int sysctl_max_tw_buckets;
84};
85
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -070086extern void inet_twdr_hangman(unsigned long data);
David Howells65f27f32006-11-22 14:55:48 +000087extern void inet_twdr_twkill_work(struct work_struct *work);
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -070088extern void inet_twdr_twcal_tick(unsigned long data);
89
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -070090struct inet_bind_bucket;
91
92/*
93 * This is a TIME_WAIT sock. It works around the memory consumption
94 * problems of sockets in such a state on heavily loaded servers, but
95 * without violating the protocol specification.
96 */
97struct inet_timewait_sock {
98 /*
99 * Now struct sock also uses sock_common, so please just
100 * don't add nothing before this first member (__tw_common) --acme
101 */
102 struct sock_common __tw_common;
103#define tw_family __tw_common.skc_family
104#define tw_state __tw_common.skc_state
105#define tw_reuse __tw_common.skc_reuse
106#define tw_bound_dev_if __tw_common.skc_bound_dev_if
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800107#define tw_node __tw_common.skc_nulls_node
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700108#define tw_bind_node __tw_common.skc_bind_node
109#define tw_refcnt __tw_common.skc_refcnt
Eric Dumazet81c3d542005-10-03 14:13:38 -0700110#define tw_hash __tw_common.skc_hash
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700111#define tw_prot __tw_common.skc_prot
Eric W. Biederman07feaeb2007-09-12 11:58:02 +0200112#define tw_net __tw_common.skc_net
Eric Dumazet68835ab2010-11-30 19:04:07 +0000113#define tw_daddr __tw_common.skc_daddr
114#define tw_rcv_saddr __tw_common.skc_rcv_saddr
Arnaldo Carvalho de Meload8bb782008-02-03 04:08:26 -0800115 int tw_timeout;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700116 volatile unsigned char tw_substate;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700117 unsigned char tw_rcv_wscale;
Eric Dumazet68835ab2010-11-30 19:04:07 +0000118
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700119 /* Socket demultiplex comparisons on incoming packets. */
Eric Dumazet68835ab2010-11-30 19:04:07 +0000120 /* these three are in inet_sock */
Al Viro23f33c22006-09-27 18:43:50 -0700121 __be16 tw_sport;
Al Viro23f33c22006-09-27 18:43:50 -0700122 __be16 tw_dport;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700123 __u16 tw_num;
Vegard Nossum9e337b02008-10-18 17:37:51 +0200124 kmemcheck_bitfield_begin(flags);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700125 /* And these are ours. */
Eric Dumazetabf90cc2009-10-18 22:48:51 +0000126 unsigned int tw_ipv6only : 1,
127 tw_transparent : 1,
Eric Dumazet66b13d92011-10-24 03:06:21 -0400128 tw_pad : 6, /* 6 bits hole */
129 tw_tos : 8,
Eric Dumazetabf90cc2009-10-18 22:48:51 +0000130 tw_ipv6_offset : 16;
Vegard Nossum9e337b02008-10-18 17:37:51 +0200131 kmemcheck_bitfield_end(flags);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700132 unsigned long tw_ttd;
133 struct inet_bind_bucket *tw_tb;
134 struct hlist_node tw_death_node;
135};
Eric Dumazetb903d322011-10-27 00:44:35 -0400136#define tw_tclass tw_tos
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700137
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800138static inline void inet_twsk_add_node_rcu(struct inet_timewait_sock *tw,
139 struct hlist_nulls_head *list)
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700140{
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800141 hlist_nulls_add_head_rcu(&tw->tw_node, list);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700142}
143
144static inline void inet_twsk_add_bind_node(struct inet_timewait_sock *tw,
145 struct hlist_head *list)
146{
147 hlist_add_head(&tw->tw_bind_node, list);
148}
149
150static inline int inet_twsk_dead_hashed(const struct inet_timewait_sock *tw)
151{
Akinobu Mitada753be2006-04-28 15:21:23 -0700152 return !hlist_unhashed(&tw->tw_death_node);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700153}
154
155static inline void inet_twsk_dead_node_init(struct inet_timewait_sock *tw)
156{
157 tw->tw_death_node.pprev = NULL;
158}
159
160static inline void __inet_twsk_del_dead_node(struct inet_timewait_sock *tw)
161{
162 __hlist_del(&tw->tw_death_node);
163 inet_twsk_dead_node_init(tw);
164}
165
166static inline int inet_twsk_del_dead_node(struct inet_timewait_sock *tw)
167{
168 if (inet_twsk_dead_hashed(tw)) {
169 __inet_twsk_del_dead_node(tw);
170 return 1;
171 }
172 return 0;
173}
174
175#define inet_twsk_for_each(tw, node, head) \
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800176 hlist_nulls_for_each_entry(tw, node, head, tw_node)
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700177
178#define inet_twsk_for_each_inmate(tw, node, jail) \
179 hlist_for_each_entry(tw, node, jail, tw_death_node)
180
181#define inet_twsk_for_each_inmate_safe(tw, node, safe, jail) \
182 hlist_for_each_entry_safe(tw, node, safe, jail, tw_death_node)
183
184static inline struct inet_timewait_sock *inet_twsk(const struct sock *sk)
185{
186 return (struct inet_timewait_sock *)sk;
187}
188
Eric Dumazet68835ab2010-11-30 19:04:07 +0000189static inline __be32 sk_rcv_saddr(const struct sock *sk)
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700190{
Eric Dumazet68835ab2010-11-30 19:04:07 +0000191/* both inet_sk() and inet_twsk() store rcv_saddr in skc_rcv_saddr */
192 return sk->__sk_common.skc_rcv_saddr;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700193}
194
Pavel Emelyanov7054fb92007-12-20 15:32:54 -0800195extern void inet_twsk_put(struct inet_timewait_sock *tw);
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700196
Eric Dumazet13475a32009-12-02 22:31:19 +0000197extern int inet_twsk_unhash(struct inet_timewait_sock *tw);
198
Eric Dumazet3cdaeda2009-12-04 03:47:42 +0000199extern int inet_twsk_bind_unhash(struct inet_timewait_sock *tw,
200 struct inet_hashinfo *hashinfo);
201
Arnaldo Carvalho de Meloc6762702005-08-09 20:09:59 -0700202extern struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk,
203 const int state);
204
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700205extern void __inet_twsk_hashdance(struct inet_timewait_sock *tw,
206 struct sock *sk,
207 struct inet_hashinfo *hashinfo);
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700208
209extern void inet_twsk_schedule(struct inet_timewait_sock *tw,
210 struct inet_timewait_death_row *twdr,
211 const int timeo, const int timewait_len);
212extern void inet_twsk_deschedule(struct inet_timewait_sock *tw,
213 struct inet_timewait_death_row *twdr);
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900214
Eric W. Biedermanb099ce22009-12-03 02:29:09 +0000215extern void inet_twsk_purge(struct inet_hashinfo *hashinfo,
Daniel Lezcanod3154922008-09-08 13:17:27 -0700216 struct inet_timewait_death_row *twdr, int family);
217
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900218static inline
219struct net *twsk_net(const struct inet_timewait_sock *twsk)
220{
221#ifdef CONFIG_NET_NS
Paul E. McKenney7ec75c52010-04-27 06:22:01 +0000222 return rcu_dereference_raw(twsk->tw_net); /* protected by locking, */
223 /* reference counting, */
224 /* initialization, or RCU. */
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900225#else
226 return &init_net;
227#endif
228}
229
230static inline
Denis V. Lunevf5aa23f2008-03-26 00:48:17 -0700231void twsk_net_set(struct inet_timewait_sock *twsk, struct net *net)
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900232{
233#ifdef CONFIG_NET_NS
Eric W. Biedermanb099ce22009-12-03 02:29:09 +0000234 rcu_assign_pointer(twsk->tw_net, net);
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900235#endif
236}
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700237#endif /* _INET_TIMEWAIT_SOCK_ */