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 <linux/atomic.h>
37 #include <asm/byteorder.h>
39 /* This is for all connections with a full identity, no wildcards.
40 * The 'e' prefix stands for Establish, but we really put all sockets
43 struct inet_ehash_bucket {
44 struct hlist_nulls_head chain;
47 /* There are a few simple rules, which allow for local port reuse by
48 * an application. In essence:
50 * 1) Sockets bound to different interfaces may share a local port.
51 * Failing that, goto test 2.
52 * 2) If all sockets have sk->sk_reuse set, and none of them are in
53 * TCP_LISTEN state, the port may be shared.
54 * Failing that, goto test 3.
55 * 3) If all sockets are bound to a specific inet_sk(sk)->rcv_saddr local
56 * address, and none of them are the same, the port may be
58 * Failing this, the port cannot be shared.
60 * The interesting point, is test #2. This is what an FTP server does
61 * all day. To optimize this case we use a specific flag bit defined
62 * below. As we add sockets to a bind bucket list, we perform a
63 * check of: (newsk->sk_reuse && (newsk->sk_state != TCP_LISTEN))
64 * As long as all sockets added to a bind bucket pass this test,
65 * the flag bit will be set.
66 * The resulting situation is that tcp_v[46]_verify_bind() can just check
67 * for this flag bit, if it is set and the socket trying to bind has
68 * sk->sk_reuse set, we don't even have to walk the owners list at all,
69 * we return that it is ok to bind this socket to the requested local port.
71 * Sounds like a lot of work, but it is worth it. In a more naive
72 * implementation (ie. current FreeBSD etc.) the entire list of ports
73 * must be walked for each data port opened by an ftp server. Needless
74 * to say, this does not scale at all. With a couple thousand FTP
75 * users logged onto your box, isn't it nice to know that new data
76 * ports are created in O(1) time? I thought so. ;-) -DaveM
78 struct inet_bind_bucket {
83 signed char fastreuse;
84 signed char fastreuseport;
87 struct hlist_node node;
88 struct hlist_head owners;
91 static inline struct net *ib_net(struct inet_bind_bucket *ib)
93 return read_pnet(&ib->ib_net);
96 #define inet_bind_bucket_for_each(tb, head) \
97 hlist_for_each_entry(tb, head, node)
99 struct inet_bind_hashbucket {
101 struct hlist_head chain;
105 * Sockets can be hashed in established or listening table
106 * We must use different 'nulls' end-of-chain value for listening
107 * hash table, or we might find a socket that was closed and
108 * reallocated/inserted into established hash table
110 #define LISTENING_NULLS_BASE (1U << 29)
111 struct inet_listen_hashbucket {
113 struct hlist_nulls_head head;
116 /* This is for listening sockets, thus all sockets which possess wildcards. */
117 #define INET_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */
119 struct inet_hashinfo {
120 /* This is for sockets with full identity only. Sockets here will
121 * always be without wildcards and will have the following invariant:
123 * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
126 struct inet_ehash_bucket *ehash;
127 spinlock_t *ehash_locks;
128 unsigned int ehash_mask;
129 unsigned int ehash_locks_mask;
131 /* Ok, let's try this, I give up, we do need a local binding
132 * TCP hash as well as the others for fast bind/connect.
134 struct inet_bind_hashbucket *bhash;
136 unsigned int bhash_size;
137 /* 4 bytes hole on 64 bit */
139 struct kmem_cache *bind_bucket_cachep;
141 /* All the above members are written once at bootup and
142 * never written again _or_ are predominantly read-access.
144 * Now align to a new cache line as all the following members
145 * might be often dirty.
147 /* All sockets in TCP_LISTEN state will be in here. This is the only
148 * table where wildcard'd TCP sockets can exist. Hash function here
149 * is just local port number.
151 struct inet_listen_hashbucket listening_hash[INET_LHTABLE_SIZE]
152 ____cacheline_aligned_in_smp;
157 static inline struct inet_ehash_bucket *inet_ehash_bucket(
158 struct inet_hashinfo *hashinfo,
161 return &hashinfo->ehash[hash & hashinfo->ehash_mask];
164 static inline spinlock_t *inet_ehash_lockp(
165 struct inet_hashinfo *hashinfo,
168 return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
171 static inline int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo)
173 unsigned int i, size = 256;
174 #if defined(CONFIG_PROVE_LOCKING)
175 unsigned int nr_pcpus = 2;
177 unsigned int nr_pcpus = num_possible_cpus();
187 if (sizeof(spinlock_t) != 0) {
189 if (size * sizeof(spinlock_t) > PAGE_SIZE)
190 hashinfo->ehash_locks = vmalloc(size * sizeof(spinlock_t));
193 hashinfo->ehash_locks = kmalloc(size * sizeof(spinlock_t),
195 if (!hashinfo->ehash_locks)
197 for (i = 0; i < size; i++)
198 spin_lock_init(&hashinfo->ehash_locks[i]);
200 hashinfo->ehash_locks_mask = size - 1;
204 static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
206 if (hashinfo->ehash_locks) {
208 unsigned int size = (hashinfo->ehash_locks_mask + 1) *
210 if (size > PAGE_SIZE)
211 vfree(hashinfo->ehash_locks);
214 kfree(hashinfo->ehash_locks);
215 hashinfo->ehash_locks = NULL;
219 struct inet_bind_bucket *
220 inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net,
221 struct inet_bind_hashbucket *head,
222 const unsigned short snum);
223 void inet_bind_bucket_destroy(struct kmem_cache *cachep,
224 struct inet_bind_bucket *tb);
226 static inline int inet_bhashfn(struct net *net, const __u16 lport,
227 const int bhash_size)
229 return (lport + net_hash_mix(net)) & (bhash_size - 1);
232 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
233 const unsigned short snum);
235 /* These can have wildcards, don't try too hard. */
236 static inline int inet_lhashfn(struct net *net, const unsigned short num)
238 return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1);
241 static inline int inet_sk_listen_hashfn(const struct sock *sk)
243 return inet_lhashfn(sock_net(sk), inet_sk(sk)->inet_num);
246 /* Caller must disable local BH processing. */
247 int __inet_inherit_port(struct sock *sk, struct sock *child);
249 void inet_put_port(struct sock *sk);
251 void inet_hashinfo_init(struct inet_hashinfo *h);
253 int __inet_hash_nolisten(struct sock *sk, struct inet_timewait_sock *tw);
254 void inet_hash(struct sock *sk);
255 void inet_unhash(struct sock *sk);
257 struct sock *__inet_lookup_listener(struct net *net,
258 struct inet_hashinfo *hashinfo,
259 const __be32 saddr, const __be16 sport,
261 const unsigned short hnum,
264 static inline struct sock *inet_lookup_listener(struct net *net,
265 struct inet_hashinfo *hashinfo,
266 __be32 saddr, __be16 sport,
267 __be32 daddr, __be16 dport, int dif)
269 return __inet_lookup_listener(net, hashinfo, saddr, sport,
270 daddr, ntohs(dport), dif);
273 /* Socket demux engine toys. */
274 /* What happens here is ugly; there's a pair of adjacent fields in
275 struct inet_sock; __be16 dport followed by __u16 num. We want to
276 search by pair, so we combine the keys into a single 32bit value
277 and compare with 32bit value read from &...->dport. Let's at least
278 make sure that it's not mixed with anything else...
279 On 64bit targets we combine comparisons with pair of adjacent __be32
280 fields in the same way.
283 #define INET_COMBINED_PORTS(__sport, __dport) \
284 ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
285 #else /* __LITTLE_ENDIAN */
286 #define INET_COMBINED_PORTS(__sport, __dport) \
287 ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
290 #if (BITS_PER_LONG == 64)
292 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
293 const __addrpair __name = (__force __addrpair) ( \
294 (((__force __u64)(__be32)(__saddr)) << 32) | \
295 ((__force __u64)(__be32)(__daddr)))
296 #else /* __LITTLE_ENDIAN */
297 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
298 const __addrpair __name = (__force __addrpair) ( \
299 (((__force __u64)(__be32)(__daddr)) << 32) | \
300 ((__force __u64)(__be32)(__saddr)))
301 #endif /* __BIG_ENDIAN */
302 #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif) \
303 (((__sk)->sk_portpair == (__ports)) && \
304 ((__sk)->sk_addrpair == (__cookie)) && \
305 (!(__sk)->sk_bound_dev_if || \
306 ((__sk)->sk_bound_dev_if == (__dif))) && \
307 net_eq(sock_net(__sk), (__net)))
308 #else /* 32-bit arch */
309 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
310 const int __name __deprecated __attribute__((unused))
312 #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif) \
313 (((__sk)->sk_portpair == (__ports)) && \
314 ((__sk)->sk_daddr == (__saddr)) && \
315 ((__sk)->sk_rcv_saddr == (__daddr)) && \
316 (!(__sk)->sk_bound_dev_if || \
317 ((__sk)->sk_bound_dev_if == (__dif))) && \
318 net_eq(sock_net(__sk), (__net)))
319 #endif /* 64-bit arch */
322 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
323 * not check it for lookups anymore, thanks Alexey. -DaveM
325 * Local BH must be disabled here.
327 struct sock *__inet_lookup_established(struct net *net,
328 struct inet_hashinfo *hashinfo,
329 const __be32 saddr, const __be16 sport,
330 const __be32 daddr, const u16 hnum,
333 static inline struct sock *
334 inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
335 const __be32 saddr, const __be16 sport,
336 const __be32 daddr, const __be16 dport,
339 return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
343 static inline struct sock *__inet_lookup(struct net *net,
344 struct inet_hashinfo *hashinfo,
345 const __be32 saddr, const __be16 sport,
346 const __be32 daddr, const __be16 dport,
349 u16 hnum = ntohs(dport);
350 struct sock *sk = __inet_lookup_established(net, hashinfo,
351 saddr, sport, daddr, hnum, dif);
353 return sk ? : __inet_lookup_listener(net, hashinfo, saddr, sport,
357 static inline struct sock *inet_lookup(struct net *net,
358 struct inet_hashinfo *hashinfo,
359 const __be32 saddr, const __be16 sport,
360 const __be32 daddr, const __be16 dport,
366 sk = __inet_lookup(net, hashinfo, saddr, sport, daddr, dport, dif);
372 static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
377 struct sock *sk = skb_steal_sock(skb);
378 const struct iphdr *iph = ip_hdr(skb);
383 return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo,
385 iph->daddr, dport, inet_iif(skb));
388 int __inet_hash_connect(struct inet_timewait_death_row *death_row,
389 struct sock *sk, u32 port_offset,
390 int (*check_established)(struct inet_timewait_death_row *,
391 struct sock *, __u16,
392 struct inet_timewait_sock **),
393 int (*hash)(struct sock *sk,
394 struct inet_timewait_sock *twp));
396 int inet_hash_connect(struct inet_timewait_death_row *death_row,
398 #endif /* _INET_HASHTABLES_H */