xref: /linux/include/net/inet_timewait_sock.h (revision 54a8a2220c936a47840c9a3d74910c5a56fae2ed)
1 /*
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 
18 #include <linux/config.h>
19 
20 #include <linux/ip.h>
21 #include <linux/list.h>
22 #include <linux/timer.h>
23 #include <linux/types.h>
24 #include <linux/workqueue.h>
25 
26 #include <net/sock.h>
27 #include <net/tcp_states.h>
28 
29 #include <asm/atomic.h>
30 
31 struct inet_hashinfo;
32 
33 #define INET_TWDR_RECYCLE_SLOTS_LOG	5
34 #define INET_TWDR_RECYCLE_SLOTS		(1 << INET_TWDR_RECYCLE_SLOTS_LOG)
35 
36 /*
37  * If time > 4sec, it is "slow" path, no recycling is required,
38  * so that we select tick to get range about 4 seconds.
39  */
40 #if HZ <= 16 || HZ > 4096
41 # error Unsupported: HZ <= 16 or HZ > 4096
42 #elif HZ <= 32
43 # define INET_TWDR_RECYCLE_TICK (5 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
44 #elif HZ <= 64
45 # define INET_TWDR_RECYCLE_TICK (6 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
46 #elif HZ <= 128
47 # define INET_TWDR_RECYCLE_TICK (7 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
48 #elif HZ <= 256
49 # define INET_TWDR_RECYCLE_TICK (8 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
50 #elif HZ <= 512
51 # define INET_TWDR_RECYCLE_TICK (9 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
52 #elif HZ <= 1024
53 # define INET_TWDR_RECYCLE_TICK (10 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
54 #elif HZ <= 2048
55 # define INET_TWDR_RECYCLE_TICK (11 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
56 #else
57 # define INET_TWDR_RECYCLE_TICK (12 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG)
58 #endif
59 
60 /* TIME_WAIT reaping mechanism. */
61 #define INET_TWDR_TWKILL_SLOTS	8 /* Please keep this a power of 2. */
62 
63 #define INET_TWDR_TWKILL_QUOTA 100
64 
65 struct inet_timewait_death_row {
66 	/* Short-time timewait calendar */
67 	int			twcal_hand;
68 	int			twcal_jiffie;
69 	struct timer_list	twcal_timer;
70 	struct hlist_head	twcal_row[INET_TWDR_RECYCLE_SLOTS];
71 
72 	spinlock_t		death_lock;
73 	int			tw_count;
74 	int			period;
75 	u32			thread_slots;
76 	struct work_struct	twkill_work;
77 	struct timer_list	tw_timer;
78 	int			slot;
79 	struct hlist_head	cells[INET_TWDR_TWKILL_SLOTS];
80 	struct inet_hashinfo 	*hashinfo;
81 	int			sysctl_tw_recycle;
82 	int			sysctl_max_tw_buckets;
83 };
84 
85 extern void inet_twdr_hangman(unsigned long data);
86 extern void inet_twdr_twkill_work(void *data);
87 extern void inet_twdr_twcal_tick(unsigned long data);
88 
89 #if (BITS_PER_LONG == 64)
90 #define INET_TIMEWAIT_ADDRCMP_ALIGN_BYTES 8
91 #else
92 #define INET_TIMEWAIT_ADDRCMP_ALIGN_BYTES 4
93 #endif
94 
95 struct 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  */
102 struct 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
112 #define tw_node			__tw_common.skc_node
113 #define tw_bind_node		__tw_common.skc_bind_node
114 #define tw_refcnt		__tw_common.skc_refcnt
115 #define tw_hash			__tw_common.skc_hash
116 #define tw_prot			__tw_common.skc_prot
117 	volatile unsigned char	tw_substate;
118 	/* 3 bits hole, try to pack */
119 	unsigned char		tw_rcv_wscale;
120 	/* Socket demultiplex comparisons on incoming packets. */
121 	/* these five are in inet_sock */
122 	__u16			tw_sport;
123 	__u32			tw_daddr __attribute__((aligned(INET_TIMEWAIT_ADDRCMP_ALIGN_BYTES)));
124 	__u32			tw_rcv_saddr;
125 	__u16			tw_dport;
126 	__u16			tw_num;
127 	/* And these are ours. */
128 	__u8			tw_ipv6only:1;
129 	/* 31 bits hole, try to pack */
130 	int			tw_timeout;
131 	unsigned long		tw_ttd;
132 	struct inet_bind_bucket	*tw_tb;
133 	struct hlist_node	tw_death_node;
134 };
135 
136 static inline void inet_twsk_add_node(struct inet_timewait_sock *tw,
137 				      struct hlist_head *list)
138 {
139 	hlist_add_head(&tw->tw_node, list);
140 }
141 
142 static inline void inet_twsk_add_bind_node(struct inet_timewait_sock *tw,
143 					   struct hlist_head *list)
144 {
145 	hlist_add_head(&tw->tw_bind_node, list);
146 }
147 
148 static inline int inet_twsk_dead_hashed(const struct inet_timewait_sock *tw)
149 {
150 	return tw->tw_death_node.pprev != NULL;
151 }
152 
153 static inline void inet_twsk_dead_node_init(struct inet_timewait_sock *tw)
154 {
155 	tw->tw_death_node.pprev = NULL;
156 }
157 
158 static inline void __inet_twsk_del_dead_node(struct inet_timewait_sock *tw)
159 {
160 	__hlist_del(&tw->tw_death_node);
161 	inet_twsk_dead_node_init(tw);
162 }
163 
164 static inline int inet_twsk_del_dead_node(struct inet_timewait_sock *tw)
165 {
166 	if (inet_twsk_dead_hashed(tw)) {
167 		__inet_twsk_del_dead_node(tw);
168 		return 1;
169 	}
170 	return 0;
171 }
172 
173 #define inet_twsk_for_each(tw, node, head) \
174 	hlist_for_each_entry(tw, node, head, tw_node)
175 
176 #define inet_twsk_for_each_inmate(tw, node, jail) \
177 	hlist_for_each_entry(tw, node, jail, tw_death_node)
178 
179 #define inet_twsk_for_each_inmate_safe(tw, node, safe, jail) \
180 	hlist_for_each_entry_safe(tw, node, safe, jail, tw_death_node)
181 
182 static inline struct inet_timewait_sock *inet_twsk(const struct sock *sk)
183 {
184 	return (struct inet_timewait_sock *)sk;
185 }
186 
187 static inline u32 inet_rcv_saddr(const struct sock *sk)
188 {
189 	return likely(sk->sk_state != TCP_TIME_WAIT) ?
190 		inet_sk(sk)->rcv_saddr : inet_twsk(sk)->tw_rcv_saddr;
191 }
192 
193 static inline void inet_twsk_put(struct inet_timewait_sock *tw)
194 {
195 	if (atomic_dec_and_test(&tw->tw_refcnt)) {
196 #ifdef SOCK_REFCNT_DEBUG
197 		printk(KERN_DEBUG "%s timewait_sock %p released\n",
198 		       tw->tw_prot->name, tw);
199 #endif
200 		kmem_cache_free(tw->tw_prot->twsk_slab, tw);
201 	}
202 }
203 
204 extern struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk,
205 						  const int state);
206 
207 extern void __inet_twsk_kill(struct inet_timewait_sock *tw,
208 			     struct inet_hashinfo *hashinfo);
209 
210 extern void __inet_twsk_hashdance(struct inet_timewait_sock *tw,
211 				  struct sock *sk,
212 				  struct inet_hashinfo *hashinfo);
213 
214 extern void inet_twsk_schedule(struct inet_timewait_sock *tw,
215 			       struct inet_timewait_death_row *twdr,
216 			       const int timeo, const int timewait_len);
217 extern void inet_twsk_deschedule(struct inet_timewait_sock *tw,
218 				 struct inet_timewait_death_row *twdr);
219 #endif	/* _INET_TIMEWAIT_SOCK_ */
220