xref: /linux/include/net/inet_timewait_sock.h (revision d96caf61ace778b56ab189caaf2b2294101878a9)
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 
19 #include <linux/kmemcheck.h>
20 #include <linux/list.h>
21 #include <linux/timer.h>
22 #include <linux/types.h>
23 #include <linux/workqueue.h>
24 
25 #include <net/inet_sock.h>
26 #include <net/sock.h>
27 #include <net/tcp_states.h>
28 #include <net/timewait_sock.h>
29 
30 #include <linux/atomic.h>
31 
32 struct 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 static inline u32 inet_tw_time_stamp(void)
62 {
63 	return jiffies;
64 }
65 
66 /* 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 
71 struct inet_timewait_death_row {
72 	/* Short-time timewait calendar */
73 	int			twcal_hand;
74 	unsigned long		twcal_jiffie;
75 	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 
91 void inet_twdr_hangman(unsigned long data);
92 void inet_twdr_twkill_work(struct work_struct *work);
93 void inet_twdr_twcal_tick(unsigned long data);
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_nulls_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 #define tw_net			__tw_common.skc_net
118 #define tw_daddr        	__tw_common.skc_daddr
119 #define tw_rcv_saddr    	__tw_common.skc_rcv_saddr
120 #define tw_dport		__tw_common.skc_dport
121 #define tw_num			__tw_common.skc_num
122 
123 	int			tw_timeout;
124 	volatile unsigned char	tw_substate;
125 	unsigned char		tw_rcv_wscale;
126 
127 	/* Socket demultiplex comparisons on incoming packets. */
128 	/* these three are in inet_sock */
129 	__be16			tw_sport;
130 	kmemcheck_bitfield_begin(flags);
131 	/* And these are ours. */
132 	unsigned int		tw_ipv6only     : 1,
133 				tw_transparent  : 1,
134 				tw_pad		: 6,	/* 6 bits hole */
135 				tw_tos		: 8,
136 				tw_ipv6_offset  : 16;
137 	kmemcheck_bitfield_end(flags);
138 	u32			tw_ttd;
139 	struct inet_bind_bucket	*tw_tb;
140 	struct hlist_node	tw_death_node;
141 };
142 #define tw_tclass tw_tos
143 
144 static inline void inet_twsk_add_node_rcu(struct inet_timewait_sock *tw,
145 				      struct hlist_nulls_head *list)
146 {
147 	hlist_nulls_add_head_rcu(&tw->tw_node, list);
148 }
149 
150 static inline void inet_twsk_add_bind_node(struct inet_timewait_sock *tw,
151 					   struct hlist_head *list)
152 {
153 	hlist_add_head(&tw->tw_bind_node, list);
154 }
155 
156 static inline int inet_twsk_dead_hashed(const struct inet_timewait_sock *tw)
157 {
158 	return !hlist_unhashed(&tw->tw_death_node);
159 }
160 
161 static inline void inet_twsk_dead_node_init(struct inet_timewait_sock *tw)
162 {
163 	tw->tw_death_node.pprev = NULL;
164 }
165 
166 static inline void __inet_twsk_del_dead_node(struct inet_timewait_sock *tw)
167 {
168 	__hlist_del(&tw->tw_death_node);
169 	inet_twsk_dead_node_init(tw);
170 }
171 
172 static inline int inet_twsk_del_dead_node(struct inet_timewait_sock *tw)
173 {
174 	if (inet_twsk_dead_hashed(tw)) {
175 		__inet_twsk_del_dead_node(tw);
176 		return 1;
177 	}
178 	return 0;
179 }
180 
181 #define inet_twsk_for_each(tw, node, head) \
182 	hlist_nulls_for_each_entry(tw, node, head, tw_node)
183 
184 #define inet_twsk_for_each_inmate(tw, jail) \
185 	hlist_for_each_entry(tw, jail, tw_death_node)
186 
187 #define inet_twsk_for_each_inmate_safe(tw, safe, jail) \
188 	hlist_for_each_entry_safe(tw, safe, jail, tw_death_node)
189 
190 static inline struct inet_timewait_sock *inet_twsk(const struct sock *sk)
191 {
192 	return (struct inet_timewait_sock *)sk;
193 }
194 
195 void inet_twsk_put(struct inet_timewait_sock *tw);
196 
197 int inet_twsk_unhash(struct inet_timewait_sock *tw);
198 
199 int inet_twsk_bind_unhash(struct inet_timewait_sock *tw,
200 			  struct inet_hashinfo *hashinfo);
201 
202 struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk,
203 					   const int state);
204 
205 void __inet_twsk_hashdance(struct inet_timewait_sock *tw, struct sock *sk,
206 			   struct inet_hashinfo *hashinfo);
207 
208 void inet_twsk_schedule(struct inet_timewait_sock *tw,
209 			struct inet_timewait_death_row *twdr,
210 			const int timeo, const int timewait_len);
211 void inet_twsk_deschedule(struct inet_timewait_sock *tw,
212 			  struct inet_timewait_death_row *twdr);
213 
214 void inet_twsk_purge(struct inet_hashinfo *hashinfo,
215 		     struct inet_timewait_death_row *twdr, int family);
216 
217 static inline
218 struct net *twsk_net(const struct inet_timewait_sock *twsk)
219 {
220 	return read_pnet(&twsk->tw_net);
221 }
222 
223 static inline
224 void twsk_net_set(struct inet_timewait_sock *twsk, struct net *net)
225 {
226 	write_pnet(&twsk->tw_net, net);
227 }
228 #endif	/* _INET_TIMEWAIT_SOCK_ */
229