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.
6 * Authors: Lotsa people, from code originally in tcp
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.
14 #ifndef _INET_HASHTABLES_H
15 #define _INET_HASHTABLES_H
18 #include <linux/interrupt.h>
20 #include <linux/ipv6.h>
21 #include <linux/list.h>
22 #include <linux/slab.h>
23 #include <linux/socket.h>
24 #include <linux/spinlock.h>
25 #include <linux/types.h>
26 #include <linux/wait.h>
27 #include <linux/vmalloc.h>
29 #include <net/inet_connection_sock.h>
30 #include <net/inet_sock.h>
32 #include <net/route.h>
33 #include <net/tcp_states.h>
34 #include <net/netns/hash.h>
36 #include <asm/atomic.h>
37 #include <asm/byteorder.h>
39 /* This is for all connections with a full identity, no wildcards.
40 * One chain is dedicated to TIME_WAIT sockets.
41 * I'll experiment with dynamic table growth later.
43 struct inet_ehash_bucket {
44 struct hlist_nulls_head chain;
45 struct hlist_nulls_head twchain;
48 /* There are a few simple rules, which allow for local port reuse by
49 * an application. In essence:
51 * 1) Sockets bound to different interfaces may share a local port.
52 * Failing that, goto test 2.
53 * 2) If all sockets have sk->sk_reuse set, and none of them are in
54 * TCP_LISTEN state, the port may be shared.
55 * Failing that, goto test 3.
56 * 3) If all sockets are bound to a specific inet_sk(sk)->rcv_saddr local
57 * address, and none of them are the same, the port may be
59 * Failing this, the port cannot be shared.
61 * The interesting point, is test #2. This is what an FTP server does
62 * all day. To optimize this case we use a specific flag bit defined
63 * below. As we add sockets to a bind bucket list, we perform a
64 * check of: (newsk->sk_reuse && (newsk->sk_state != TCP_LISTEN))
65 * As long as all sockets added to a bind bucket pass this test,
66 * the flag bit will be set.
67 * The resulting situation is that tcp_v[46]_verify_bind() can just check
68 * for this flag bit, if it is set and the socket trying to bind has
69 * sk->sk_reuse set, we don't even have to walk the owners list at all,
70 * we return that it is ok to bind this socket to the requested local port.
72 * Sounds like a lot of work, but it is worth it. In a more naive
73 * implementation (ie. current FreeBSD etc.) the entire list of ports
74 * must be walked for each data port opened by an ftp server. Needless
75 * to say, this does not scale at all. With a couple thousand FTP
76 * users logged onto your box, isn't it nice to know that new data
77 * ports are created in O(1) time? I thought so. ;-) -DaveM
79 struct inet_bind_bucket {
84 signed short fastreuse;
85 struct hlist_node node;
86 struct hlist_head owners;
89 static inline struct net *ib_net(struct inet_bind_bucket *ib)
91 return read_pnet(&ib->ib_net);
94 #define inet_bind_bucket_for_each(tb, node, head) \
95 hlist_for_each_entry(tb, node, head, node)
97 struct inet_bind_hashbucket {
99 struct hlist_head chain;
102 struct inet_listen_hashbucket {
104 struct hlist_head head;
107 /* This is for listening sockets, thus all sockets which possess wildcards. */
108 #define INET_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */
110 struct inet_hashinfo {
111 /* This is for sockets with full identity only. Sockets here will
112 * always be without wildcards and will have the following invariant:
114 * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
116 * TIME_WAIT sockets use a separate chain (twchain).
118 struct inet_ehash_bucket *ehash;
119 spinlock_t *ehash_locks;
120 unsigned int ehash_size;
121 unsigned int ehash_locks_mask;
123 /* Ok, let's try this, I give up, we do need a local binding
124 * TCP hash as well as the others for fast bind/connect.
126 struct inet_bind_hashbucket *bhash;
128 unsigned int bhash_size;
129 /* Note : 4 bytes padding on 64 bit arches */
131 struct kmem_cache *bind_bucket_cachep;
133 /* All the above members are written once at bootup and
134 * never written again _or_ are predominantly read-access.
136 * Now align to a new cache line as all the following members
137 * might be often dirty.
139 /* All sockets in TCP_LISTEN state will be in here. This is the only
140 * table where wildcard'd TCP sockets can exist. Hash function here
141 * is just local port number.
143 struct inet_listen_hashbucket listening_hash[INET_LHTABLE_SIZE]
144 ____cacheline_aligned_in_smp;
148 static inline struct inet_ehash_bucket *inet_ehash_bucket(
149 struct inet_hashinfo *hashinfo,
152 return &hashinfo->ehash[hash & (hashinfo->ehash_size - 1)];
155 static inline spinlock_t *inet_ehash_lockp(
156 struct inet_hashinfo *hashinfo,
159 return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
162 static inline int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo)
164 unsigned int i, size = 256;
165 #if defined(CONFIG_PROVE_LOCKING)
166 unsigned int nr_pcpus = 2;
168 unsigned int nr_pcpus = num_possible_cpus();
178 if (sizeof(rwlock_t) != 0) {
180 if (size * sizeof(spinlock_t) > PAGE_SIZE)
181 hashinfo->ehash_locks = vmalloc(size * sizeof(spinlock_t));
184 hashinfo->ehash_locks = kmalloc(size * sizeof(spinlock_t),
186 if (!hashinfo->ehash_locks)
188 for (i = 0; i < size; i++)
189 spin_lock_init(&hashinfo->ehash_locks[i]);
191 hashinfo->ehash_locks_mask = size - 1;
195 static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
197 if (hashinfo->ehash_locks) {
199 unsigned int size = (hashinfo->ehash_locks_mask + 1) *
201 if (size > PAGE_SIZE)
202 vfree(hashinfo->ehash_locks);
205 kfree(hashinfo->ehash_locks);
206 hashinfo->ehash_locks = NULL;
210 extern struct inet_bind_bucket *
211 inet_bind_bucket_create(struct kmem_cache *cachep,
213 struct inet_bind_hashbucket *head,
214 const unsigned short snum);
215 extern void inet_bind_bucket_destroy(struct kmem_cache *cachep,
216 struct inet_bind_bucket *tb);
218 static inline int inet_bhashfn(struct net *net,
219 const __u16 lport, const int bhash_size)
221 return (lport + net_hash_mix(net)) & (bhash_size - 1);
224 extern void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
225 const unsigned short snum);
227 /* These can have wildcards, don't try too hard. */
228 static inline int inet_lhashfn(struct net *net, const unsigned short num)
230 return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1);
233 static inline int inet_sk_listen_hashfn(const struct sock *sk)
235 return inet_lhashfn(sock_net(sk), inet_sk(sk)->num);
238 /* Caller must disable local BH processing. */
239 extern void __inet_inherit_port(struct sock *sk, struct sock *child);
241 extern void inet_put_port(struct sock *sk);
243 void inet_hashinfo_init(struct inet_hashinfo *h);
245 extern void __inet_hash_nolisten(struct sock *sk);
246 extern void inet_hash(struct sock *sk);
247 extern void inet_unhash(struct sock *sk);
249 extern struct sock *__inet_lookup_listener(struct net *net,
250 struct inet_hashinfo *hashinfo,
252 const unsigned short hnum,
255 static inline struct sock *inet_lookup_listener(struct net *net,
256 struct inet_hashinfo *hashinfo,
257 __be32 daddr, __be16 dport, int dif)
259 return __inet_lookup_listener(net, hashinfo, daddr, ntohs(dport), dif);
262 /* Socket demux engine toys. */
263 /* What happens here is ugly; there's a pair of adjacent fields in
264 struct inet_sock; __be16 dport followed by __u16 num. We want to
265 search by pair, so we combine the keys into a single 32bit value
266 and compare with 32bit value read from &...->dport. Let's at least
267 make sure that it's not mixed with anything else...
268 On 64bit targets we combine comparisons with pair of adjacent __be32
269 fields in the same way.
271 typedef __u32 __bitwise __portpair;
273 #define INET_COMBINED_PORTS(__sport, __dport) \
274 ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
275 #else /* __LITTLE_ENDIAN */
276 #define INET_COMBINED_PORTS(__sport, __dport) \
277 ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
280 #if (BITS_PER_LONG == 64)
281 typedef __u64 __bitwise __addrpair;
283 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
284 const __addrpair __name = (__force __addrpair) ( \
285 (((__force __u64)(__be32)(__saddr)) << 32) | \
286 ((__force __u64)(__be32)(__daddr)));
287 #else /* __LITTLE_ENDIAN */
288 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
289 const __addrpair __name = (__force __addrpair) ( \
290 (((__force __u64)(__be32)(__daddr)) << 32) | \
291 ((__force __u64)(__be32)(__saddr)));
292 #endif /* __BIG_ENDIAN */
293 #define INET_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
294 (((__sk)->sk_hash == (__hash)) && sock_net((__sk)) == (__net) && \
295 ((*((__addrpair *)&(inet_sk(__sk)->daddr))) == (__cookie)) && \
296 ((*((__portpair *)&(inet_sk(__sk)->dport))) == (__ports)) && \
297 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
298 #define INET_TW_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
299 (((__sk)->sk_hash == (__hash)) && sock_net((__sk)) == (__net) && \
300 ((*((__addrpair *)&(inet_twsk(__sk)->tw_daddr))) == (__cookie)) && \
301 ((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \
302 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
303 #else /* 32-bit arch */
304 #define INET_ADDR_COOKIE(__name, __saddr, __daddr)
305 #define INET_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif) \
306 (((__sk)->sk_hash == (__hash)) && sock_net((__sk)) == (__net) && \
307 (inet_sk(__sk)->daddr == (__saddr)) && \
308 (inet_sk(__sk)->rcv_saddr == (__daddr)) && \
309 ((*((__portpair *)&(inet_sk(__sk)->dport))) == (__ports)) && \
310 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
311 #define INET_TW_MATCH(__sk, __net, __hash,__cookie, __saddr, __daddr, __ports, __dif) \
312 (((__sk)->sk_hash == (__hash)) && sock_net((__sk)) == (__net) && \
313 (inet_twsk(__sk)->tw_daddr == (__saddr)) && \
314 (inet_twsk(__sk)->tw_rcv_saddr == (__daddr)) && \
315 ((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \
316 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
317 #endif /* 64-bit arch */
320 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
321 * not check it for lookups anymore, thanks Alexey. -DaveM
323 * Local BH must be disabled here.
325 extern struct sock * __inet_lookup_established(struct net *net,
326 struct inet_hashinfo *hashinfo,
327 const __be32 saddr, const __be16 sport,
328 const __be32 daddr, const u16 hnum, const int dif);
330 static inline struct sock *
331 inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
332 const __be32 saddr, const __be16 sport,
333 const __be32 daddr, const __be16 dport,
336 return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
340 static inline struct sock *__inet_lookup(struct net *net,
341 struct inet_hashinfo *hashinfo,
342 const __be32 saddr, const __be16 sport,
343 const __be32 daddr, const __be16 dport,
346 u16 hnum = ntohs(dport);
347 struct sock *sk = __inet_lookup_established(net, hashinfo,
348 saddr, sport, daddr, hnum, dif);
350 return sk ? : __inet_lookup_listener(net, hashinfo, daddr, hnum, dif);
353 static inline struct sock *inet_lookup(struct net *net,
354 struct inet_hashinfo *hashinfo,
355 const __be32 saddr, const __be16 sport,
356 const __be32 daddr, const __be16 dport,
362 sk = __inet_lookup(net, hashinfo, saddr, sport, daddr, dport, dif);
368 static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
374 const struct iphdr *iph = ip_hdr(skb);
376 if (unlikely(sk = skb_steal_sock(skb)))
379 return __inet_lookup(dev_net(skb->dst->dev), hashinfo,
381 iph->daddr, dport, inet_iif(skb));
384 extern int __inet_hash_connect(struct inet_timewait_death_row *death_row,
385 struct sock *sk, u32 port_offset,
386 int (*check_established)(struct inet_timewait_death_row *,
387 struct sock *, __u16, struct inet_timewait_sock **),
388 void (*hash)(struct sock *sk));
389 extern int inet_hash_connect(struct inet_timewait_death_row *death_row,
391 #endif /* _INET_HASHTABLES_H */