Arnaldo Carvalho de Melo | 304a161 | 2005-08-09 19:59:20 -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 | * Authors: Lotsa people, from code originally in tcp |
| 7 | * |
| 8 | * This program is free software; you can redistribute it and/or |
| 9 | * modify it under the terms of the GNU General Public License |
| 10 | * as published by the Free Software Foundation; either version |
| 11 | * 2 of the License, or (at your option) any later version. |
| 12 | */ |
| 13 | |
| 14 | #ifndef _INET_HASHTABLES_H |
| 15 | #define _INET_HASHTABLES_H |
| 16 | |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 17 | #include <linux/config.h> |
| 18 | |
Arnaldo Carvalho de Melo | 2d8c4ce | 2005-08-09 20:07:13 -0700 | [diff] [blame] | 19 | #include <linux/interrupt.h> |
Arnaldo Carvalho de Melo | 33b6223 | 2005-08-09 20:09:06 -0700 | [diff] [blame] | 20 | #include <linux/ipv6.h> |
Arnaldo Carvalho de Melo | 77d8bf9 | 2005-08-09 20:00:51 -0700 | [diff] [blame] | 21 | #include <linux/list.h> |
| 22 | #include <linux/slab.h> |
Arnaldo Carvalho de Melo | 33b6223 | 2005-08-09 20:09:06 -0700 | [diff] [blame] | 23 | #include <linux/socket.h> |
Arnaldo Carvalho de Melo | 77d8bf9 | 2005-08-09 20:00:51 -0700 | [diff] [blame] | 24 | #include <linux/spinlock.h> |
Arnaldo Carvalho de Melo | 304a161 | 2005-08-09 19:59:20 -0700 | [diff] [blame] | 25 | #include <linux/types.h> |
Arnaldo Carvalho de Melo | f3f05f7 | 2005-08-09 20:08:09 -0700 | [diff] [blame] | 26 | #include <linux/wait.h> |
Arnaldo Carvalho de Melo | 304a161 | 2005-08-09 19:59:20 -0700 | [diff] [blame] | 27 | |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 28 | #include <net/inet_connection_sock.h> |
Arnaldo Carvalho de Melo | 14c8502 | 2005-12-27 02:43:12 -0200 | [diff] [blame] | 29 | #include <net/inet_sock.h> |
Arnaldo Carvalho de Melo | 0a5578c | 2005-08-09 20:11:41 -0700 | [diff] [blame] | 30 | #include <net/route.h> |
Arnaldo Carvalho de Melo | 2d8c4ce | 2005-08-09 20:07:13 -0700 | [diff] [blame] | 31 | #include <net/sock.h> |
Arnaldo Carvalho de Melo | c752f07 | 2005-08-09 20:08:28 -0700 | [diff] [blame] | 32 | #include <net/tcp_states.h> |
Arnaldo Carvalho de Melo | 2d8c4ce | 2005-08-09 20:07:13 -0700 | [diff] [blame] | 33 | |
Arnaldo Carvalho de Melo | f3f05f7 | 2005-08-09 20:08:09 -0700 | [diff] [blame] | 34 | #include <asm/atomic.h> |
Arnaldo Carvalho de Melo | e48c414 | 2005-08-09 20:09:46 -0700 | [diff] [blame] | 35 | #include <asm/byteorder.h> |
Arnaldo Carvalho de Melo | f3f05f7 | 2005-08-09 20:08:09 -0700 | [diff] [blame] | 36 | |
Arnaldo Carvalho de Melo | 77d8bf9 | 2005-08-09 20:00:51 -0700 | [diff] [blame] | 37 | /* This is for all connections with a full identity, no wildcards. |
| 38 | * New scheme, half the table is for TIME_WAIT, the other half is |
| 39 | * for the rest. I'll experiment with dynamic table growth later. |
| 40 | */ |
| 41 | struct inet_ehash_bucket { |
| 42 | rwlock_t lock; |
| 43 | struct hlist_head chain; |
Eric Dumazet | 6d25536 | 2005-10-04 15:55:51 -0700 | [diff] [blame] | 44 | }; |
Arnaldo Carvalho de Melo | 77d8bf9 | 2005-08-09 20:00:51 -0700 | [diff] [blame] | 45 | |
| 46 | /* There are a few simple rules, which allow for local port reuse by |
| 47 | * an application. In essence: |
| 48 | * |
| 49 | * 1) Sockets bound to different interfaces may share a local port. |
| 50 | * Failing that, goto test 2. |
| 51 | * 2) If all sockets have sk->sk_reuse set, and none of them are in |
| 52 | * TCP_LISTEN state, the port may be shared. |
| 53 | * Failing that, goto test 3. |
| 54 | * 3) If all sockets are bound to a specific inet_sk(sk)->rcv_saddr local |
| 55 | * address, and none of them are the same, the port may be |
| 56 | * shared. |
| 57 | * Failing this, the port cannot be shared. |
| 58 | * |
| 59 | * The interesting point, is test #2. This is what an FTP server does |
| 60 | * all day. To optimize this case we use a specific flag bit defined |
| 61 | * below. As we add sockets to a bind bucket list, we perform a |
| 62 | * check of: (newsk->sk_reuse && (newsk->sk_state != TCP_LISTEN)) |
| 63 | * As long as all sockets added to a bind bucket pass this test, |
| 64 | * the flag bit will be set. |
| 65 | * The resulting situation is that tcp_v[46]_verify_bind() can just check |
| 66 | * for this flag bit, if it is set and the socket trying to bind has |
| 67 | * sk->sk_reuse set, we don't even have to walk the owners list at all, |
| 68 | * we return that it is ok to bind this socket to the requested local port. |
| 69 | * |
| 70 | * Sounds like a lot of work, but it is worth it. In a more naive |
| 71 | * implementation (ie. current FreeBSD etc.) the entire list of ports |
| 72 | * must be walked for each data port opened by an ftp server. Needless |
| 73 | * to say, this does not scale at all. With a couple thousand FTP |
| 74 | * users logged onto your box, isn't it nice to know that new data |
| 75 | * ports are created in O(1) time? I thought so. ;-) -DaveM |
| 76 | */ |
| 77 | struct inet_bind_bucket { |
| 78 | unsigned short port; |
| 79 | signed short fastreuse; |
| 80 | struct hlist_node node; |
| 81 | struct hlist_head owners; |
| 82 | }; |
| 83 | |
| 84 | #define inet_bind_bucket_for_each(tb, node, head) \ |
| 85 | hlist_for_each_entry(tb, node, head, node) |
| 86 | |
| 87 | struct inet_bind_hashbucket { |
| 88 | spinlock_t lock; |
| 89 | struct hlist_head chain; |
| 90 | }; |
| 91 | |
| 92 | /* This is for listening sockets, thus all sockets which possess wildcards. */ |
| 93 | #define INET_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */ |
| 94 | |
| 95 | struct inet_hashinfo { |
| 96 | /* This is for sockets with full identity only. Sockets here will |
| 97 | * always be without wildcards and will have the following invariant: |
| 98 | * |
| 99 | * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE |
| 100 | * |
| 101 | * First half of the table is for sockets not in TIME_WAIT, second half |
| 102 | * is for TIME_WAIT sockets only. |
| 103 | */ |
| 104 | struct inet_ehash_bucket *ehash; |
| 105 | |
| 106 | /* Ok, let's try this, I give up, we do need a local binding |
| 107 | * TCP hash as well as the others for fast bind/connect. |
| 108 | */ |
| 109 | struct inet_bind_hashbucket *bhash; |
| 110 | |
| 111 | int bhash_size; |
Eric Dumazet | 81c3d54 | 2005-10-03 14:13:38 -0700 | [diff] [blame] | 112 | unsigned int ehash_size; |
Arnaldo Carvalho de Melo | 77d8bf9 | 2005-08-09 20:00:51 -0700 | [diff] [blame] | 113 | |
| 114 | /* All sockets in TCP_LISTEN state will be in here. This is the only |
| 115 | * table where wildcard'd TCP sockets can exist. Hash function here |
| 116 | * is just local port number. |
| 117 | */ |
| 118 | struct hlist_head listening_hash[INET_LHTABLE_SIZE]; |
| 119 | |
| 120 | /* All the above members are written once at bootup and |
| 121 | * never written again _or_ are predominantly read-access. |
| 122 | * |
| 123 | * Now align to a new cache line as all the following members |
| 124 | * are often dirty. |
| 125 | */ |
| 126 | rwlock_t lhash_lock ____cacheline_aligned; |
| 127 | atomic_t lhash_users; |
| 128 | wait_queue_head_t lhash_wait; |
Arnaldo Carvalho de Melo | 2d8c4ce | 2005-08-09 20:07:13 -0700 | [diff] [blame] | 129 | kmem_cache_t *bind_bucket_cachep; |
Arnaldo Carvalho de Melo | 77d8bf9 | 2005-08-09 20:00:51 -0700 | [diff] [blame] | 130 | }; |
| 131 | |
Eric Dumazet | 81c3d54 | 2005-10-03 14:13:38 -0700 | [diff] [blame] | 132 | static inline struct inet_ehash_bucket *inet_ehash_bucket( |
| 133 | struct inet_hashinfo *hashinfo, |
| 134 | unsigned int hash) |
| 135 | { |
| 136 | return &hashinfo->ehash[hash & (hashinfo->ehash_size - 1)]; |
Arnaldo Carvalho de Melo | 304a161 | 2005-08-09 19:59:20 -0700 | [diff] [blame] | 137 | } |
| 138 | |
Arnaldo Carvalho de Melo | 77d8bf9 | 2005-08-09 20:00:51 -0700 | [diff] [blame] | 139 | extern struct inet_bind_bucket * |
| 140 | inet_bind_bucket_create(kmem_cache_t *cachep, |
| 141 | struct inet_bind_hashbucket *head, |
| 142 | const unsigned short snum); |
| 143 | extern void inet_bind_bucket_destroy(kmem_cache_t *cachep, |
| 144 | struct inet_bind_bucket *tb); |
| 145 | |
| 146 | static inline int inet_bhashfn(const __u16 lport, const int bhash_size) |
| 147 | { |
| 148 | return lport & (bhash_size - 1); |
| 149 | } |
| 150 | |
Arnaldo Carvalho de Melo | 2d8c4ce | 2005-08-09 20:07:13 -0700 | [diff] [blame] | 151 | extern void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb, |
| 152 | const unsigned short snum); |
| 153 | |
Arnaldo Carvalho de Melo | 77d8bf9 | 2005-08-09 20:00:51 -0700 | [diff] [blame] | 154 | /* These can have wildcards, don't try too hard. */ |
| 155 | static inline int inet_lhashfn(const unsigned short num) |
| 156 | { |
| 157 | return num & (INET_LHTABLE_SIZE - 1); |
| 158 | } |
| 159 | |
| 160 | static inline int inet_sk_listen_hashfn(const struct sock *sk) |
| 161 | { |
| 162 | return inet_lhashfn(inet_sk(sk)->num); |
| 163 | } |
| 164 | |
Arnaldo Carvalho de Melo | 2d8c4ce | 2005-08-09 20:07:13 -0700 | [diff] [blame] | 165 | /* Caller must disable local BH processing. */ |
| 166 | static inline void __inet_inherit_port(struct inet_hashinfo *table, |
| 167 | struct sock *sk, struct sock *child) |
| 168 | { |
| 169 | const int bhash = inet_bhashfn(inet_sk(child)->num, table->bhash_size); |
| 170 | struct inet_bind_hashbucket *head = &table->bhash[bhash]; |
| 171 | struct inet_bind_bucket *tb; |
| 172 | |
| 173 | spin_lock(&head->lock); |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 174 | tb = inet_csk(sk)->icsk_bind_hash; |
Arnaldo Carvalho de Melo | 2d8c4ce | 2005-08-09 20:07:13 -0700 | [diff] [blame] | 175 | sk_add_bind_node(child, &tb->owners); |
Arnaldo Carvalho de Melo | 463c84b | 2005-08-09 20:10:42 -0700 | [diff] [blame] | 176 | inet_csk(child)->icsk_bind_hash = tb; |
Arnaldo Carvalho de Melo | 2d8c4ce | 2005-08-09 20:07:13 -0700 | [diff] [blame] | 177 | spin_unlock(&head->lock); |
| 178 | } |
| 179 | |
| 180 | static inline void inet_inherit_port(struct inet_hashinfo *table, |
| 181 | struct sock *sk, struct sock *child) |
| 182 | { |
| 183 | local_bh_disable(); |
| 184 | __inet_inherit_port(table, sk, child); |
| 185 | local_bh_enable(); |
| 186 | } |
| 187 | |
| 188 | extern void inet_put_port(struct inet_hashinfo *table, struct sock *sk); |
| 189 | |
Arnaldo Carvalho de Melo | f3f05f7 | 2005-08-09 20:08:09 -0700 | [diff] [blame] | 190 | extern void inet_listen_wlock(struct inet_hashinfo *hashinfo); |
| 191 | |
| 192 | /* |
| 193 | * - We may sleep inside this lock. |
| 194 | * - If sleeping is not required (or called from BH), |
| 195 | * use plain read_(un)lock(&inet_hashinfo.lhash_lock). |
| 196 | */ |
| 197 | static inline void inet_listen_lock(struct inet_hashinfo *hashinfo) |
| 198 | { |
| 199 | /* read_lock synchronizes to candidates to writers */ |
| 200 | read_lock(&hashinfo->lhash_lock); |
| 201 | atomic_inc(&hashinfo->lhash_users); |
| 202 | read_unlock(&hashinfo->lhash_lock); |
| 203 | } |
| 204 | |
| 205 | static inline void inet_listen_unlock(struct inet_hashinfo *hashinfo) |
| 206 | { |
| 207 | if (atomic_dec_and_test(&hashinfo->lhash_users)) |
| 208 | wake_up(&hashinfo->lhash_wait); |
| 209 | } |
| 210 | |
| 211 | static inline void __inet_hash(struct inet_hashinfo *hashinfo, |
| 212 | struct sock *sk, const int listen_possible) |
| 213 | { |
| 214 | struct hlist_head *list; |
| 215 | rwlock_t *lock; |
| 216 | |
| 217 | BUG_TRAP(sk_unhashed(sk)); |
| 218 | if (listen_possible && sk->sk_state == TCP_LISTEN) { |
| 219 | list = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)]; |
| 220 | lock = &hashinfo->lhash_lock; |
| 221 | inet_listen_wlock(hashinfo); |
| 222 | } else { |
Eric Dumazet | 81c3d54 | 2005-10-03 14:13:38 -0700 | [diff] [blame] | 223 | struct inet_ehash_bucket *head; |
| 224 | sk->sk_hash = inet_sk_ehashfn(sk); |
| 225 | head = inet_ehash_bucket(hashinfo, sk->sk_hash); |
| 226 | list = &head->chain; |
| 227 | lock = &head->lock; |
Arnaldo Carvalho de Melo | f3f05f7 | 2005-08-09 20:08:09 -0700 | [diff] [blame] | 228 | write_lock(lock); |
| 229 | } |
| 230 | __sk_add_node(sk, list); |
| 231 | sock_prot_inc_use(sk->sk_prot); |
| 232 | write_unlock(lock); |
| 233 | if (listen_possible && sk->sk_state == TCP_LISTEN) |
| 234 | wake_up(&hashinfo->lhash_wait); |
| 235 | } |
Arnaldo Carvalho de Melo | 81849d1 | 2005-08-09 20:08:50 -0700 | [diff] [blame] | 236 | |
| 237 | static inline void inet_hash(struct inet_hashinfo *hashinfo, struct sock *sk) |
| 238 | { |
| 239 | if (sk->sk_state != TCP_CLOSE) { |
| 240 | local_bh_disable(); |
| 241 | __inet_hash(hashinfo, sk, 1); |
| 242 | local_bh_enable(); |
| 243 | } |
| 244 | } |
| 245 | |
| 246 | static inline void inet_unhash(struct inet_hashinfo *hashinfo, struct sock *sk) |
| 247 | { |
| 248 | rwlock_t *lock; |
| 249 | |
| 250 | if (sk_unhashed(sk)) |
| 251 | goto out; |
| 252 | |
| 253 | if (sk->sk_state == TCP_LISTEN) { |
| 254 | local_bh_disable(); |
| 255 | inet_listen_wlock(hashinfo); |
| 256 | lock = &hashinfo->lhash_lock; |
| 257 | } else { |
Eric Dumazet | 81c3d54 | 2005-10-03 14:13:38 -0700 | [diff] [blame] | 258 | lock = &inet_ehash_bucket(hashinfo, sk->sk_hash)->lock; |
| 259 | write_lock_bh(lock); |
Arnaldo Carvalho de Melo | 81849d1 | 2005-08-09 20:08:50 -0700 | [diff] [blame] | 260 | } |
| 261 | |
| 262 | if (__sk_del_node_init(sk)) |
| 263 | sock_prot_dec_use(sk->sk_prot); |
| 264 | write_unlock_bh(lock); |
| 265 | out: |
| 266 | if (sk->sk_state == TCP_LISTEN) |
| 267 | wake_up(&hashinfo->lhash_wait); |
| 268 | } |
Arnaldo Carvalho de Melo | 33b6223 | 2005-08-09 20:09:06 -0700 | [diff] [blame] | 269 | |
Arnaldo Carvalho de Melo | 0a5578c | 2005-08-09 20:11:41 -0700 | [diff] [blame] | 270 | static inline int inet_iif(const struct sk_buff *skb) |
| 271 | { |
| 272 | return ((struct rtable *)skb->dst)->rt_iif; |
| 273 | } |
| 274 | |
Arnaldo Carvalho de Melo | 33b6223 | 2005-08-09 20:09:06 -0700 | [diff] [blame] | 275 | extern struct sock *__inet_lookup_listener(const struct hlist_head *head, |
| 276 | const u32 daddr, |
| 277 | const unsigned short hnum, |
| 278 | const int dif); |
| 279 | |
| 280 | /* Optimize the common listener case. */ |
Arnaldo Carvalho de Melo | e48c414 | 2005-08-09 20:09:46 -0700 | [diff] [blame] | 281 | static inline struct sock * |
| 282 | inet_lookup_listener(struct inet_hashinfo *hashinfo, |
| 283 | const u32 daddr, |
| 284 | const unsigned short hnum, const int dif) |
Arnaldo Carvalho de Melo | 33b6223 | 2005-08-09 20:09:06 -0700 | [diff] [blame] | 285 | { |
| 286 | struct sock *sk = NULL; |
Arnaldo Carvalho de Melo | e48c414 | 2005-08-09 20:09:46 -0700 | [diff] [blame] | 287 | const struct hlist_head *head; |
Arnaldo Carvalho de Melo | 33b6223 | 2005-08-09 20:09:06 -0700 | [diff] [blame] | 288 | |
| 289 | read_lock(&hashinfo->lhash_lock); |
| 290 | head = &hashinfo->listening_hash[inet_lhashfn(hnum)]; |
| 291 | if (!hlist_empty(head)) { |
| 292 | const struct inet_sock *inet = inet_sk((sk = __sk_head(head))); |
| 293 | |
| 294 | if (inet->num == hnum && !sk->sk_node.next && |
| 295 | (!inet->rcv_saddr || inet->rcv_saddr == daddr) && |
| 296 | (sk->sk_family == PF_INET || !ipv6_only_sock(sk)) && |
| 297 | !sk->sk_bound_dev_if) |
| 298 | goto sherry_cache; |
| 299 | sk = __inet_lookup_listener(head, daddr, hnum, dif); |
| 300 | } |
| 301 | if (sk) { |
| 302 | sherry_cache: |
| 303 | sock_hold(sk); |
| 304 | } |
| 305 | read_unlock(&hashinfo->lhash_lock); |
| 306 | return sk; |
| 307 | } |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 308 | |
| 309 | /* Socket demux engine toys. */ |
| 310 | #ifdef __BIG_ENDIAN |
| 311 | #define INET_COMBINED_PORTS(__sport, __dport) \ |
| 312 | (((__u32)(__sport) << 16) | (__u32)(__dport)) |
| 313 | #else /* __LITTLE_ENDIAN */ |
| 314 | #define INET_COMBINED_PORTS(__sport, __dport) \ |
| 315 | (((__u32)(__dport) << 16) | (__u32)(__sport)) |
| 316 | #endif |
| 317 | |
| 318 | #if (BITS_PER_LONG == 64) |
| 319 | #ifdef __BIG_ENDIAN |
| 320 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ |
| 321 | const __u64 __name = (((__u64)(__saddr)) << 32) | ((__u64)(__daddr)); |
| 322 | #else /* __LITTLE_ENDIAN */ |
| 323 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ |
| 324 | const __u64 __name = (((__u64)(__daddr)) << 32) | ((__u64)(__saddr)); |
| 325 | #endif /* __BIG_ENDIAN */ |
Eric Dumazet | 81c3d54 | 2005-10-03 14:13:38 -0700 | [diff] [blame] | 326 | #define INET_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif)\ |
| 327 | (((__sk)->sk_hash == (__hash)) && \ |
| 328 | ((*((__u64 *)&(inet_sk(__sk)->daddr))) == (__cookie)) && \ |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 329 | ((*((__u32 *)&(inet_sk(__sk)->dport))) == (__ports)) && \ |
| 330 | (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) |
Eric Dumazet | 81c3d54 | 2005-10-03 14:13:38 -0700 | [diff] [blame] | 331 | #define INET_TW_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif)\ |
| 332 | (((__sk)->sk_hash == (__hash)) && \ |
| 333 | ((*((__u64 *)&(inet_twsk(__sk)->tw_daddr))) == (__cookie)) && \ |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 334 | ((*((__u32 *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \ |
| 335 | (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) |
| 336 | #else /* 32-bit arch */ |
| 337 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) |
Eric Dumazet | 81c3d54 | 2005-10-03 14:13:38 -0700 | [diff] [blame] | 338 | #define INET_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif) \ |
| 339 | (((__sk)->sk_hash == (__hash)) && \ |
| 340 | (inet_sk(__sk)->daddr == (__saddr)) && \ |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 341 | (inet_sk(__sk)->rcv_saddr == (__daddr)) && \ |
| 342 | ((*((__u32 *)&(inet_sk(__sk)->dport))) == (__ports)) && \ |
| 343 | (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) |
Eric Dumazet | 81c3d54 | 2005-10-03 14:13:38 -0700 | [diff] [blame] | 344 | #define INET_TW_MATCH(__sk, __hash,__cookie, __saddr, __daddr, __ports, __dif) \ |
| 345 | (((__sk)->sk_hash == (__hash)) && \ |
| 346 | (inet_twsk(__sk)->tw_daddr == (__saddr)) && \ |
Arnaldo Carvalho de Melo | 8feaf0c0 | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 347 | (inet_twsk(__sk)->tw_rcv_saddr == (__daddr)) && \ |
| 348 | ((*((__u32 *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \ |
| 349 | (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) |
| 350 | #endif /* 64-bit arch */ |
Arnaldo Carvalho de Melo | e48c414 | 2005-08-09 20:09:46 -0700 | [diff] [blame] | 351 | |
| 352 | /* |
| 353 | * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need |
| 354 | * not check it for lookups anymore, thanks Alexey. -DaveM |
| 355 | * |
| 356 | * Local BH must be disabled here. |
| 357 | */ |
| 358 | static inline struct sock * |
| 359 | __inet_lookup_established(struct inet_hashinfo *hashinfo, |
| 360 | const u32 saddr, const u16 sport, |
| 361 | const u32 daddr, const u16 hnum, |
| 362 | const int dif) |
| 363 | { |
| 364 | INET_ADDR_COOKIE(acookie, saddr, daddr) |
| 365 | const __u32 ports = INET_COMBINED_PORTS(sport, hnum); |
| 366 | struct sock *sk; |
| 367 | const struct hlist_node *node; |
| 368 | /* Optimize here for direct hit, only listening connections can |
| 369 | * have wildcards anyways. |
| 370 | */ |
Eric Dumazet | 81c3d54 | 2005-10-03 14:13:38 -0700 | [diff] [blame] | 371 | unsigned int hash = inet_ehashfn(daddr, hnum, saddr, sport); |
| 372 | struct inet_ehash_bucket *head = inet_ehash_bucket(hashinfo, hash); |
Arnaldo Carvalho de Melo | e48c414 | 2005-08-09 20:09:46 -0700 | [diff] [blame] | 373 | |
Eric Dumazet | 81c3d54 | 2005-10-03 14:13:38 -0700 | [diff] [blame] | 374 | prefetch(head->chain.first); |
Arnaldo Carvalho de Melo | e48c414 | 2005-08-09 20:09:46 -0700 | [diff] [blame] | 375 | read_lock(&head->lock); |
| 376 | sk_for_each(sk, node, &head->chain) { |
Eric Dumazet | 81c3d54 | 2005-10-03 14:13:38 -0700 | [diff] [blame] | 377 | if (INET_MATCH(sk, hash, acookie, saddr, daddr, ports, dif)) |
Arnaldo Carvalho de Melo | e48c414 | 2005-08-09 20:09:46 -0700 | [diff] [blame] | 378 | goto hit; /* You sunk my battleship! */ |
| 379 | } |
| 380 | |
| 381 | /* Must check for a TIME_WAIT'er before going to listener hash. */ |
| 382 | sk_for_each(sk, node, &(head + hashinfo->ehash_size)->chain) { |
Eric Dumazet | 81c3d54 | 2005-10-03 14:13:38 -0700 | [diff] [blame] | 383 | if (INET_TW_MATCH(sk, hash, acookie, saddr, daddr, ports, dif)) |
Arnaldo Carvalho de Melo | e48c414 | 2005-08-09 20:09:46 -0700 | [diff] [blame] | 384 | goto hit; |
| 385 | } |
| 386 | sk = NULL; |
| 387 | out: |
| 388 | read_unlock(&head->lock); |
| 389 | return sk; |
| 390 | hit: |
| 391 | sock_hold(sk); |
| 392 | goto out; |
| 393 | } |
| 394 | |
| 395 | static inline struct sock *__inet_lookup(struct inet_hashinfo *hashinfo, |
| 396 | const u32 saddr, const u16 sport, |
| 397 | const u32 daddr, const u16 hnum, |
| 398 | const int dif) |
| 399 | { |
| 400 | struct sock *sk = __inet_lookup_established(hashinfo, saddr, sport, daddr, |
| 401 | hnum, dif); |
| 402 | return sk ? : inet_lookup_listener(hashinfo, daddr, hnum, dif); |
| 403 | } |
| 404 | |
| 405 | static inline struct sock *inet_lookup(struct inet_hashinfo *hashinfo, |
| 406 | const u32 saddr, const u16 sport, |
| 407 | const u32 daddr, const u16 dport, |
| 408 | const int dif) |
| 409 | { |
| 410 | struct sock *sk; |
| 411 | |
| 412 | local_bh_disable(); |
| 413 | sk = __inet_lookup(hashinfo, saddr, sport, daddr, ntohs(dport), dif); |
| 414 | local_bh_enable(); |
| 415 | |
| 416 | return sk; |
| 417 | } |
Arnaldo Carvalho de Melo | a7f5e7f | 2005-12-13 23:25:31 -0800 | [diff] [blame] | 418 | |
| 419 | extern int inet_hash_connect(struct inet_timewait_death_row *death_row, |
| 420 | struct sock *sk); |
Arnaldo Carvalho de Melo | 304a161 | 2005-08-09 19:59:20 -0700 | [diff] [blame] | 421 | #endif /* _INET_HASHTABLES_H */ |