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/module.h>
22 #include <linux/timer.h>
23 #include <linux/types.h>
24 #include <linux/workqueue.h>
25 
26 #include <net/inet_sock.h>
27 #include <net/sock.h>
28 #include <net/tcp_states.h>
29 #include <net/timewait_sock.h>
30 
31 #include <linux/atomic.h>
32 
33 struct 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 
67 struct inet_timewait_death_row {
68 	/* Short-time timewait calendar */
69 	int			twcal_hand;
70 	unsigned long		twcal_jiffie;
71 	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 
87 extern void inet_twdr_hangman(unsigned long data);
88 extern void inet_twdr_twkill_work(struct work_struct *work);
89 extern void inet_twdr_twcal_tick(unsigned long data);
90 
91 struct 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  */
98 struct 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
108 #define tw_node			__tw_common.skc_nulls_node
109 #define tw_bind_node		__tw_common.skc_bind_node
110 #define tw_refcnt		__tw_common.skc_refcnt
111 #define tw_hash			__tw_common.skc_hash
112 #define tw_prot			__tw_common.skc_prot
113 #define tw_net			__tw_common.skc_net
114 #define tw_daddr        	__tw_common.skc_daddr
115 #define tw_rcv_saddr    	__tw_common.skc_rcv_saddr
116 	int			tw_timeout;
117 	volatile unsigned char	tw_substate;
118 	unsigned char		tw_rcv_wscale;
119 
120 	/* Socket demultiplex comparisons on incoming packets. */
121 	/* these three are in inet_sock */
122 	__be16			tw_sport;
123 	__be16			tw_dport;
124 	__u16			tw_num;
125 	kmemcheck_bitfield_begin(flags);
126 	/* And these are ours. */
127 	unsigned int		tw_ipv6only     : 1,
128 				tw_transparent  : 1,
129 				tw_pad		: 6,	/* 6 bits hole */
130 				tw_tos		: 8,
131 				tw_ipv6_offset  : 16;
132 	kmemcheck_bitfield_end(flags);
133 	unsigned long		tw_ttd;
134 	struct inet_bind_bucket	*tw_tb;
135 	struct hlist_node	tw_death_node;
136 };
137 
138 static inline void inet_twsk_add_node_rcu(struct inet_timewait_sock *tw,
139 				      struct hlist_nulls_head *list)
140 {
141 	hlist_nulls_add_head_rcu(&tw->tw_node, list);
142 }
143 
144 static 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 
150 static inline int inet_twsk_dead_hashed(const struct inet_timewait_sock *tw)
151 {
152 	return !hlist_unhashed(&tw->tw_death_node);
153 }
154 
155 static inline void inet_twsk_dead_node_init(struct inet_timewait_sock *tw)
156 {
157 	tw->tw_death_node.pprev = NULL;
158 }
159 
160 static 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 
166 static 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) \
176 	hlist_nulls_for_each_entry(tw, node, head, tw_node)
177 
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 
184 static inline struct inet_timewait_sock *inet_twsk(const struct sock *sk)
185 {
186 	return (struct inet_timewait_sock *)sk;
187 }
188 
189 static inline __be32 sk_rcv_saddr(const struct sock *sk)
190 {
191 /* both inet_sk() and inet_twsk() store rcv_saddr in skc_rcv_saddr */
192 	return sk->__sk_common.skc_rcv_saddr;
193 }
194 
195 extern void inet_twsk_put(struct inet_timewait_sock *tw);
196 
197 extern int inet_twsk_unhash(struct inet_timewait_sock *tw);
198 
199 extern int inet_twsk_bind_unhash(struct inet_timewait_sock *tw,
200 				 struct inet_hashinfo *hashinfo);
201 
202 extern struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk,
203 						  const int state);
204 
205 extern void __inet_twsk_hashdance(struct inet_timewait_sock *tw,
206 				  struct sock *sk,
207 				  struct inet_hashinfo *hashinfo);
208 
209 extern void inet_twsk_schedule(struct inet_timewait_sock *tw,
210 			       struct inet_timewait_death_row *twdr,
211 			       const int timeo, const int timewait_len);
212 extern void inet_twsk_deschedule(struct inet_timewait_sock *tw,
213 				 struct inet_timewait_death_row *twdr);
214 
215 extern void inet_twsk_purge(struct inet_hashinfo *hashinfo,
216 			    struct inet_timewait_death_row *twdr, int family);
217 
218 static inline
219 struct net *twsk_net(const struct inet_timewait_sock *twsk)
220 {
221 #ifdef CONFIG_NET_NS
222 	return rcu_dereference_raw(twsk->tw_net); /* protected by locking, */
223 						  /* reference counting, */
224 						  /* initialization, or RCU. */
225 #else
226 	return &init_net;
227 #endif
228 }
229 
230 static inline
231 void twsk_net_set(struct inet_timewait_sock *twsk, struct net *net)
232 {
233 #ifdef CONFIG_NET_NS
234 	rcu_assign_pointer(twsk->tw_net, net);
235 #endif
236 }
237 #endif	/* _INET_TIMEWAIT_SOCK_ */
238