Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 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 | |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 18 | |
Vegard Nossum | 9e337b0 | 2008-10-18 17:37:51 +0200 | [diff] [blame] | 19 | #include <linux/kmemcheck.h> |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 20 | #include <linux/list.h> |
Arnaldo Carvalho de Melo | 295ff7e | 2005-08-09 20:44:40 -0700 | [diff] [blame] | 21 | #include <linux/timer.h> |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 22 | #include <linux/types.h> |
Arnaldo Carvalho de Melo | 295ff7e | 2005-08-09 20:44:40 -0700 | [diff] [blame] | 23 | #include <linux/workqueue.h> |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 24 | |
Arnaldo Carvalho de Melo | 14c8502 | 2005-12-27 02:43:12 -0200 | [diff] [blame] | 25 | #include <net/inet_sock.h> |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 26 | #include <net/sock.h> |
| 27 | #include <net/tcp_states.h> |
Arnaldo Carvalho de Melo | 6d6ee43 | 2005-12-13 23:25:19 -0800 | [diff] [blame] | 28 | #include <net/timewait_sock.h> |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 29 | |
Arun Sharma | 60063497 | 2011-07-26 16:09:06 -0700 | [diff] [blame] | 30 | #include <linux/atomic.h> |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 31 | |
Arnaldo Carvalho de Melo | 295ff7e | 2005-08-09 20:44:40 -0700 | [diff] [blame] | 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 | |
Eric Dumazet | 96f817f | 2013-10-03 14:27:25 -0700 | [diff] [blame] | 61 | static inline u32 inet_tw_time_stamp(void) |
| 62 | { |
| 63 | return jiffies; |
| 64 | } |
| 65 | |
Arnaldo Carvalho de Melo | 295ff7e | 2005-08-09 20:44:40 -0700 | [diff] [blame] | 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; |
Eric Dumazet | 187f5f84 | 2007-03-05 13:32:48 -0800 | [diff] [blame] | 74 | unsigned long twcal_jiffie; |
Arnaldo Carvalho de Melo | 295ff7e | 2005-08-09 20:44:40 -0700 | [diff] [blame] | 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 | |
Joe Perches | 1fd5115 | 2013-09-21 10:22:41 -0700 | [diff] [blame] | 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); |
Arnaldo Carvalho de Melo | 696ab2d | 2005-08-09 20:45:03 -0700 | [diff] [blame] | 94 | |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 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 |
Eric Dumazet | 3ab5aee | 2008-11-16 19:40:17 -0800 | [diff] [blame] | 112 | #define tw_node __tw_common.skc_nulls_node |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 113 | #define tw_bind_node __tw_common.skc_bind_node |
| 114 | #define tw_refcnt __tw_common.skc_refcnt |
Eric Dumazet | 81c3d54 | 2005-10-03 14:13:38 -0700 | [diff] [blame] | 115 | #define tw_hash __tw_common.skc_hash |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 116 | #define tw_prot __tw_common.skc_prot |
Eric W. Biederman | 07feaeb | 2007-09-12 11:58:02 +0200 | [diff] [blame] | 117 | #define tw_net __tw_common.skc_net |
Eric Dumazet | 68835ab | 2010-11-30 19:04:07 +0000 | [diff] [blame] | 118 | #define tw_daddr __tw_common.skc_daddr |
Eric Dumazet | efe4208 | 2013-10-03 15:42:29 -0700 | [diff] [blame] | 119 | #define tw_v6_daddr __tw_common.skc_v6_daddr |
Eric Dumazet | 68835ab | 2010-11-30 19:04:07 +0000 | [diff] [blame] | 120 | #define tw_rcv_saddr __tw_common.skc_rcv_saddr |
Eric Dumazet | efe4208 | 2013-10-03 15:42:29 -0700 | [diff] [blame] | 121 | #define tw_v6_rcv_saddr __tw_common.skc_v6_rcv_saddr |
Eric Dumazet | ce43b03 | 2012-11-30 09:49:27 +0000 | [diff] [blame] | 122 | #define tw_dport __tw_common.skc_dport |
| 123 | #define tw_num __tw_common.skc_num |
Eric Dumazet | ce43b03 | 2012-11-30 09:49:27 +0000 | [diff] [blame] | 124 | |
Arnaldo Carvalho de Melo | ad8bb78 | 2008-02-03 04:08:26 -0800 | [diff] [blame] | 125 | int tw_timeout; |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 126 | volatile unsigned char tw_substate; |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 127 | unsigned char tw_rcv_wscale; |
Eric Dumazet | 68835ab | 2010-11-30 19:04:07 +0000 | [diff] [blame] | 128 | |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 129 | /* Socket demultiplex comparisons on incoming packets. */ |
Eric Dumazet | 68835ab | 2010-11-30 19:04:07 +0000 | [diff] [blame] | 130 | /* these three are in inet_sock */ |
Al Viro | 23f33c2 | 2006-09-27 18:43:50 -0700 | [diff] [blame] | 131 | __be16 tw_sport; |
Vegard Nossum | 9e337b0 | 2008-10-18 17:37:51 +0200 | [diff] [blame] | 132 | kmemcheck_bitfield_begin(flags); |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 133 | /* And these are ours. */ |
Eric Dumazet | abf90cc | 2009-10-18 22:48:51 +0000 | [diff] [blame] | 134 | unsigned int tw_ipv6only : 1, |
| 135 | tw_transparent : 1, |
Eric Dumazet | 66b13d9 | 2011-10-24 03:06:21 -0400 | [diff] [blame] | 136 | tw_pad : 6, /* 6 bits hole */ |
| 137 | tw_tos : 8, |
Randy Dunlap | a5bb202 | 2013-10-14 12:36:32 -0700 | [diff] [blame] | 138 | tw_pad2 : 16; /* 16 bits hole */ |
Vegard Nossum | 9e337b0 | 2008-10-18 17:37:51 +0200 | [diff] [blame] | 139 | kmemcheck_bitfield_end(flags); |
Eric Dumazet | 96f817f | 2013-10-03 14:27:25 -0700 | [diff] [blame] | 140 | u32 tw_ttd; |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 141 | struct inet_bind_bucket *tw_tb; |
| 142 | struct hlist_node tw_death_node; |
| 143 | }; |
Eric Dumazet | b903d32 | 2011-10-27 00:44:35 -0400 | [diff] [blame] | 144 | #define tw_tclass tw_tos |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 145 | |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 146 | static inline int inet_twsk_dead_hashed(const struct inet_timewait_sock *tw) |
| 147 | { |
Akinobu Mita | da753be | 2006-04-28 15:21:23 -0700 | [diff] [blame] | 148 | return !hlist_unhashed(&tw->tw_death_node); |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 149 | } |
| 150 | |
| 151 | static inline void inet_twsk_dead_node_init(struct inet_timewait_sock *tw) |
| 152 | { |
| 153 | tw->tw_death_node.pprev = NULL; |
| 154 | } |
| 155 | |
| 156 | static inline void __inet_twsk_del_dead_node(struct inet_timewait_sock *tw) |
| 157 | { |
| 158 | __hlist_del(&tw->tw_death_node); |
| 159 | inet_twsk_dead_node_init(tw); |
| 160 | } |
| 161 | |
| 162 | static inline int inet_twsk_del_dead_node(struct inet_timewait_sock *tw) |
| 163 | { |
| 164 | if (inet_twsk_dead_hashed(tw)) { |
| 165 | __inet_twsk_del_dead_node(tw); |
| 166 | return 1; |
| 167 | } |
| 168 | return 0; |
| 169 | } |
| 170 | |
| 171 | #define inet_twsk_for_each(tw, node, head) \ |
Eric Dumazet | 3ab5aee | 2008-11-16 19:40:17 -0800 | [diff] [blame] | 172 | hlist_nulls_for_each_entry(tw, node, head, tw_node) |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 173 | |
Sasha Levin | b67bfe0 | 2013-02-27 17:06:00 -0800 | [diff] [blame] | 174 | #define inet_twsk_for_each_inmate(tw, jail) \ |
| 175 | hlist_for_each_entry(tw, jail, tw_death_node) |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 176 | |
Sasha Levin | b67bfe0 | 2013-02-27 17:06:00 -0800 | [diff] [blame] | 177 | #define inet_twsk_for_each_inmate_safe(tw, safe, jail) \ |
| 178 | hlist_for_each_entry_safe(tw, safe, jail, tw_death_node) |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 179 | |
| 180 | static inline struct inet_timewait_sock *inet_twsk(const struct sock *sk) |
| 181 | { |
| 182 | return (struct inet_timewait_sock *)sk; |
| 183 | } |
| 184 | |
Eric Dumazet | 05dbc7b | 2013-10-03 00:22:02 -0700 | [diff] [blame] | 185 | void inet_twsk_free(struct inet_timewait_sock *tw); |
Joe Perches | 1fd5115 | 2013-09-21 10:22:41 -0700 | [diff] [blame] | 186 | void inet_twsk_put(struct inet_timewait_sock *tw); |
Arnaldo Carvalho de Melo | e48c414 | 2005-08-09 20:09:46 -0700 | [diff] [blame] | 187 | |
Joe Perches | 1fd5115 | 2013-09-21 10:22:41 -0700 | [diff] [blame] | 188 | int inet_twsk_unhash(struct inet_timewait_sock *tw); |
Eric Dumazet | 13475a3 | 2009-12-02 22:31:19 +0000 | [diff] [blame] | 189 | |
Joe Perches | 1fd5115 | 2013-09-21 10:22:41 -0700 | [diff] [blame] | 190 | int inet_twsk_bind_unhash(struct inet_timewait_sock *tw, |
| 191 | struct inet_hashinfo *hashinfo); |
Eric Dumazet | 3cdaeda | 2009-12-04 03:47:42 +0000 | [diff] [blame] | 192 | |
Joe Perches | 1fd5115 | 2013-09-21 10:22:41 -0700 | [diff] [blame] | 193 | struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk, |
| 194 | const int state); |
Arnaldo Carvalho de Melo | c676270 | 2005-08-09 20:09:59 -0700 | [diff] [blame] | 195 | |
Joe Perches | 1fd5115 | 2013-09-21 10:22:41 -0700 | [diff] [blame] | 196 | void __inet_twsk_hashdance(struct inet_timewait_sock *tw, struct sock *sk, |
| 197 | struct inet_hashinfo *hashinfo); |
Arnaldo Carvalho de Melo | 696ab2d | 2005-08-09 20:45:03 -0700 | [diff] [blame] | 198 | |
Joe Perches | 1fd5115 | 2013-09-21 10:22:41 -0700 | [diff] [blame] | 199 | void inet_twsk_schedule(struct inet_timewait_sock *tw, |
| 200 | struct inet_timewait_death_row *twdr, |
| 201 | const int timeo, const int timewait_len); |
| 202 | void inet_twsk_deschedule(struct inet_timewait_sock *tw, |
| 203 | struct inet_timewait_death_row *twdr); |
YOSHIFUJI Hideaki | 3b1e0a6 | 2008-03-26 02:26:21 +0900 | [diff] [blame] | 204 | |
Joe Perches | 1fd5115 | 2013-09-21 10:22:41 -0700 | [diff] [blame] | 205 | void inet_twsk_purge(struct inet_hashinfo *hashinfo, |
| 206 | struct inet_timewait_death_row *twdr, int family); |
Daniel Lezcano | d315492 | 2008-09-08 13:17:27 -0700 | [diff] [blame] | 207 | |
YOSHIFUJI Hideaki | 3b1e0a6 | 2008-03-26 02:26:21 +0900 | [diff] [blame] | 208 | static inline |
| 209 | struct net *twsk_net(const struct inet_timewait_sock *twsk) |
| 210 | { |
Eric Dumazet | f943cbe | 2011-12-14 04:58:33 +0000 | [diff] [blame] | 211 | return read_pnet(&twsk->tw_net); |
YOSHIFUJI Hideaki | 3b1e0a6 | 2008-03-26 02:26:21 +0900 | [diff] [blame] | 212 | } |
| 213 | |
| 214 | static inline |
Denis V. Lunev | f5aa23f | 2008-03-26 00:48:17 -0700 | [diff] [blame] | 215 | void twsk_net_set(struct inet_timewait_sock *twsk, struct net *net) |
YOSHIFUJI Hideaki | 3b1e0a6 | 2008-03-26 02:26:21 +0900 | [diff] [blame] | 216 | { |
Eric Dumazet | f943cbe | 2011-12-14 04:58:33 +0000 | [diff] [blame] | 217 | write_pnet(&twsk->tw_net, net); |
YOSHIFUJI Hideaki | 3b1e0a6 | 2008-03-26 02:26:21 +0900 | [diff] [blame] | 218 | } |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 219 | #endif /* _INET_TIMEWAIT_SOCK_ */ |