blob: 17404b5388a75302c4951d5f469fd96814ad7709 [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 -070091struct inet_bind_bucket;
92
93/*
94 * This is a TIME_WAIT sock. It works around the memory consumption
95 * problems of sockets in such a state on heavily loaded servers, but
96 * without violating the protocol specification.
97 */
98struct inet_timewait_sock {
99 /*
100 * Now struct sock also uses sock_common, so please just
101 * don't add nothing before this first member (__tw_common) --acme
102 */
103 struct sock_common __tw_common;
104#define tw_family __tw_common.skc_family
105#define tw_state __tw_common.skc_state
106#define tw_reuse __tw_common.skc_reuse
107#define tw_bound_dev_if __tw_common.skc_bound_dev_if
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800108#define tw_node __tw_common.skc_nulls_node
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700109#define tw_bind_node __tw_common.skc_bind_node
110#define tw_refcnt __tw_common.skc_refcnt
Eric Dumazet81c3d542005-10-03 14:13:38 -0700111#define tw_hash __tw_common.skc_hash
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700112#define tw_prot __tw_common.skc_prot
Eric W. Biederman07feaeb2007-09-12 11:58:02 +0200113#define tw_net __tw_common.skc_net
Eric Dumazet68835ab2010-11-30 19:04:07 +0000114#define tw_daddr __tw_common.skc_daddr
115#define tw_rcv_saddr __tw_common.skc_rcv_saddr
Arnaldo Carvalho de Meload8bb782008-02-03 04:08:26 -0800116 int tw_timeout;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700117 volatile unsigned char tw_substate;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700118 unsigned char tw_rcv_wscale;
Eric Dumazet68835ab2010-11-30 19:04:07 +0000119
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700120 /* Socket demultiplex comparisons on incoming packets. */
Eric Dumazet68835ab2010-11-30 19:04:07 +0000121 /* these three are in inet_sock */
Al Viro23f33c22006-09-27 18:43:50 -0700122 __be16 tw_sport;
Al Viro23f33c22006-09-27 18:43:50 -0700123 __be16 tw_dport;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700124 __u16 tw_num;
Vegard Nossum9e337b02008-10-18 17:37:51 +0200125 kmemcheck_bitfield_begin(flags);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700126 /* And these are ours. */
Eric Dumazetabf90cc2009-10-18 22:48:51 +0000127 unsigned int tw_ipv6only : 1,
128 tw_transparent : 1,
129 tw_pad : 14, /* 14 bits hole */
130 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};
136
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800137static inline void inet_twsk_add_node_rcu(struct inet_timewait_sock *tw,
138 struct hlist_nulls_head *list)
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700139{
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800140 hlist_nulls_add_head_rcu(&tw->tw_node, list);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700141}
142
143static inline void inet_twsk_add_bind_node(struct inet_timewait_sock *tw,
144 struct hlist_head *list)
145{
146 hlist_add_head(&tw->tw_bind_node, list);
147}
148
149static inline int inet_twsk_dead_hashed(const struct inet_timewait_sock *tw)
150{
Akinobu Mitada753be2006-04-28 15:21:23 -0700151 return !hlist_unhashed(&tw->tw_death_node);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700152}
153
154static inline void inet_twsk_dead_node_init(struct inet_timewait_sock *tw)
155{
156 tw->tw_death_node.pprev = NULL;
157}
158
159static inline void __inet_twsk_del_dead_node(struct inet_timewait_sock *tw)
160{
161 __hlist_del(&tw->tw_death_node);
162 inet_twsk_dead_node_init(tw);
163}
164
165static inline int inet_twsk_del_dead_node(struct inet_timewait_sock *tw)
166{
167 if (inet_twsk_dead_hashed(tw)) {
168 __inet_twsk_del_dead_node(tw);
169 return 1;
170 }
171 return 0;
172}
173
174#define inet_twsk_for_each(tw, node, head) \
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800175 hlist_nulls_for_each_entry(tw, node, head, tw_node)
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700176
177#define inet_twsk_for_each_inmate(tw, node, jail) \
178 hlist_for_each_entry(tw, node, jail, tw_death_node)
179
180#define inet_twsk_for_each_inmate_safe(tw, node, safe, jail) \
181 hlist_for_each_entry_safe(tw, node, safe, jail, tw_death_node)
182
183static inline struct inet_timewait_sock *inet_twsk(const struct sock *sk)
184{
185 return (struct inet_timewait_sock *)sk;
186}
187
Eric Dumazet68835ab2010-11-30 19:04:07 +0000188static inline __be32 sk_rcv_saddr(const struct sock *sk)
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700189{
Eric Dumazet68835ab2010-11-30 19:04:07 +0000190/* both inet_sk() and inet_twsk() store rcv_saddr in skc_rcv_saddr */
191 return sk->__sk_common.skc_rcv_saddr;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700192}
193
Pavel Emelyanov7054fb92007-12-20 15:32:54 -0800194extern void inet_twsk_put(struct inet_timewait_sock *tw);
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700195
Eric Dumazet13475a32009-12-02 22:31:19 +0000196extern int inet_twsk_unhash(struct inet_timewait_sock *tw);
197
Eric Dumazet3cdaeda2009-12-04 03:47:42 +0000198extern int inet_twsk_bind_unhash(struct inet_timewait_sock *tw,
199 struct inet_hashinfo *hashinfo);
200
Arnaldo Carvalho de Meloc6762702005-08-09 20:09:59 -0700201extern struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk,
202 const int state);
203
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700204extern void __inet_twsk_hashdance(struct inet_timewait_sock *tw,
205 struct sock *sk,
206 struct inet_hashinfo *hashinfo);
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700207
208extern void inet_twsk_schedule(struct inet_timewait_sock *tw,
209 struct inet_timewait_death_row *twdr,
210 const int timeo, const int timewait_len);
211extern void inet_twsk_deschedule(struct inet_timewait_sock *tw,
212 struct inet_timewait_death_row *twdr);
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900213
Eric W. Biedermanb099ce22009-12-03 02:29:09 +0000214extern void inet_twsk_purge(struct inet_hashinfo *hashinfo,
Daniel Lezcanod3154922008-09-08 13:17:27 -0700215 struct inet_timewait_death_row *twdr, int family);
216
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900217static inline
218struct net *twsk_net(const struct inet_timewait_sock *twsk)
219{
220#ifdef CONFIG_NET_NS
Paul E. McKenney7ec75c52010-04-27 06:22:01 +0000221 return rcu_dereference_raw(twsk->tw_net); /* protected by locking, */
222 /* reference counting, */
223 /* initialization, or RCU. */
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900224#else
225 return &init_net;
226#endif
227}
228
229static inline
Denis V. Lunevf5aa23f2008-03-26 00:48:17 -0700230void twsk_net_set(struct inet_timewait_sock *twsk, struct net *net)
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900231{
232#ifdef CONFIG_NET_NS
Eric W. Biedermanb099ce22009-12-03 02:29:09 +0000233 rcu_assign_pointer(twsk->tw_net, net);
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900234#endif
235}
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700236#endif /* _INET_TIMEWAIT_SOCK_ */