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 * 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 char fastreuse;
85 signed char fastreuseport;
88 struct hlist_node node;
89 struct hlist_head owners;
92 static inline struct net *ib_net(struct inet_bind_bucket *ib)
94 return read_pnet(&ib->ib_net);
97 #define inet_bind_bucket_for_each(tb, head) \
98 hlist_for_each_entry(tb, head, node)
100 struct inet_bind_hashbucket {
102 struct hlist_head chain;
106 * Sockets can be hashed in established or listening table
107 * We must use different 'nulls' end-of-chain value for listening
108 * hash table, or we might find a socket that was closed and
109 * reallocated/inserted into established hash table
111 #define LISTENING_NULLS_BASE (1U << 29)
112 struct inet_listen_hashbucket {
114 struct hlist_nulls_head head;
117 /* This is for listening sockets, thus all sockets which possess wildcards. */
118 #define INET_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */
120 struct inet_hashinfo {
121 /* This is for sockets with full identity only. Sockets here will
122 * always be without wildcards and will have the following invariant:
124 * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
126 * TIME_WAIT sockets use a separate chain (twchain).
128 struct inet_ehash_bucket *ehash;
129 spinlock_t *ehash_locks;
130 unsigned int ehash_mask;
131 unsigned int ehash_locks_mask;
133 /* Ok, let's try this, I give up, we do need a local binding
134 * TCP hash as well as the others for fast bind/connect.
136 struct inet_bind_hashbucket *bhash;
138 unsigned int bhash_size;
139 /* 4 bytes hole on 64 bit */
141 struct kmem_cache *bind_bucket_cachep;
143 /* All the above members are written once at bootup and
144 * never written again _or_ are predominantly read-access.
146 * Now align to a new cache line as all the following members
147 * might be often dirty.
149 /* All sockets in TCP_LISTEN state will be in here. This is the only
150 * table where wildcard'd TCP sockets can exist. Hash function here
151 * is just local port number.
153 struct inet_listen_hashbucket listening_hash[INET_LHTABLE_SIZE]
154 ____cacheline_aligned_in_smp;
159 static inline struct inet_ehash_bucket *inet_ehash_bucket(
160 struct inet_hashinfo *hashinfo,
163 return &hashinfo->ehash[hash & hashinfo->ehash_mask];
166 static inline spinlock_t *inet_ehash_lockp(
167 struct inet_hashinfo *hashinfo,
170 return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
173 static inline int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo)
175 unsigned int i, size = 256;
176 #if defined(CONFIG_PROVE_LOCKING)
177 unsigned int nr_pcpus = 2;
179 unsigned int nr_pcpus = num_possible_cpus();
189 if (sizeof(spinlock_t) != 0) {
191 if (size * sizeof(spinlock_t) > PAGE_SIZE)
192 hashinfo->ehash_locks = vmalloc(size * sizeof(spinlock_t));
195 hashinfo->ehash_locks = kmalloc(size * sizeof(spinlock_t),
197 if (!hashinfo->ehash_locks)
199 for (i = 0; i < size; i++)
200 spin_lock_init(&hashinfo->ehash_locks[i]);
202 hashinfo->ehash_locks_mask = size - 1;
206 static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
208 if (hashinfo->ehash_locks) {
210 unsigned int size = (hashinfo->ehash_locks_mask + 1) *
212 if (size > PAGE_SIZE)
213 vfree(hashinfo->ehash_locks);
216 kfree(hashinfo->ehash_locks);
217 hashinfo->ehash_locks = NULL;
221 struct inet_bind_bucket *
222 inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net,
223 struct inet_bind_hashbucket *head,
224 const unsigned short snum);
225 void inet_bind_bucket_destroy(struct kmem_cache *cachep,
226 struct inet_bind_bucket *tb);
228 static inline int inet_bhashfn(struct net *net, const __u16 lport,
229 const int bhash_size)
231 return (lport + net_hash_mix(net)) & (bhash_size - 1);
234 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
235 const unsigned short snum);
237 /* These can have wildcards, don't try too hard. */
238 static inline int inet_lhashfn(struct net *net, const unsigned short num)
240 return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1);
243 static inline int inet_sk_listen_hashfn(const struct sock *sk)
245 return inet_lhashfn(sock_net(sk), inet_sk(sk)->inet_num);
248 /* Caller must disable local BH processing. */
249 int __inet_inherit_port(struct sock *sk, struct sock *child);
251 void inet_put_port(struct sock *sk);
253 void inet_hashinfo_init(struct inet_hashinfo *h);
255 int __inet_hash_nolisten(struct sock *sk, struct inet_timewait_sock *tw);
256 void inet_hash(struct sock *sk);
257 void inet_unhash(struct sock *sk);
259 struct sock *__inet_lookup_listener(struct net *net,
260 struct inet_hashinfo *hashinfo,
261 const __be32 saddr, const __be16 sport,
263 const unsigned short hnum,
266 static inline struct sock *inet_lookup_listener(struct net *net,
267 struct inet_hashinfo *hashinfo,
268 __be32 saddr, __be16 sport,
269 __be32 daddr, __be16 dport, int dif)
271 return __inet_lookup_listener(net, hashinfo, saddr, sport,
272 daddr, ntohs(dport), dif);
275 /* Socket demux engine toys. */
276 /* What happens here is ugly; there's a pair of adjacent fields in
277 struct inet_sock; __be16 dport followed by __u16 num. We want to
278 search by pair, so we combine the keys into a single 32bit value
279 and compare with 32bit value read from &...->dport. Let's at least
280 make sure that it's not mixed with anything else...
281 On 64bit targets we combine comparisons with pair of adjacent __be32
282 fields in the same way.
285 #define INET_COMBINED_PORTS(__sport, __dport) \
286 ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
287 #else /* __LITTLE_ENDIAN */
288 #define INET_COMBINED_PORTS(__sport, __dport) \
289 ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
292 #if (BITS_PER_LONG == 64)
294 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
295 const __addrpair __name = (__force __addrpair) ( \
296 (((__force __u64)(__be32)(__saddr)) << 32) | \
297 ((__force __u64)(__be32)(__daddr)));
298 #else /* __LITTLE_ENDIAN */
299 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
300 const __addrpair __name = (__force __addrpair) ( \
301 (((__force __u64)(__be32)(__daddr)) << 32) | \
302 ((__force __u64)(__be32)(__saddr)));
303 #endif /* __BIG_ENDIAN */
304 #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif) \
305 ((inet_sk(__sk)->inet_portpair == (__ports)) && \
306 (inet_sk(__sk)->inet_addrpair == (__cookie)) && \
307 (!(__sk)->sk_bound_dev_if || \
308 ((__sk)->sk_bound_dev_if == (__dif))) && \
309 net_eq(sock_net(__sk), (__net)))
310 #define INET_TW_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif)\
311 ((inet_twsk(__sk)->tw_portpair == (__ports)) && \
312 (inet_twsk(__sk)->tw_addrpair == (__cookie)) && \
313 (!(__sk)->sk_bound_dev_if || \
314 ((__sk)->sk_bound_dev_if == (__dif))) && \
315 net_eq(sock_net(__sk), (__net)))
316 #else /* 32-bit arch */
317 #define INET_ADDR_COOKIE(__name, __saddr, __daddr)
318 #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif) \
319 ((inet_sk(__sk)->inet_portpair == (__ports)) && \
320 (inet_sk(__sk)->inet_daddr == (__saddr)) && \
321 (inet_sk(__sk)->inet_rcv_saddr == (__daddr)) && \
322 (!(__sk)->sk_bound_dev_if || \
323 ((__sk)->sk_bound_dev_if == (__dif))) && \
324 net_eq(sock_net(__sk), (__net)))
325 #define INET_TW_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif) \
326 ((inet_twsk(__sk)->tw_portpair == (__ports)) && \
327 (inet_twsk(__sk)->tw_daddr == (__saddr)) && \
328 (inet_twsk(__sk)->tw_rcv_saddr == (__daddr)) && \
329 (!(__sk)->sk_bound_dev_if || \
330 ((__sk)->sk_bound_dev_if == (__dif))) && \
331 net_eq(sock_net(__sk), (__net)))
332 #endif /* 64-bit arch */
335 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
336 * not check it for lookups anymore, thanks Alexey. -DaveM
338 * Local BH must be disabled here.
340 struct sock *__inet_lookup_established(struct net *net,
341 struct inet_hashinfo *hashinfo,
342 const __be32 saddr, const __be16 sport,
343 const __be32 daddr, const u16 hnum,
346 static inline struct sock *
347 inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
348 const __be32 saddr, const __be16 sport,
349 const __be32 daddr, const __be16 dport,
352 return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
356 static inline struct sock *__inet_lookup(struct net *net,
357 struct inet_hashinfo *hashinfo,
358 const __be32 saddr, const __be16 sport,
359 const __be32 daddr, const __be16 dport,
362 u16 hnum = ntohs(dport);
363 struct sock *sk = __inet_lookup_established(net, hashinfo,
364 saddr, sport, daddr, hnum, dif);
366 return sk ? : __inet_lookup_listener(net, hashinfo, saddr, sport,
370 static inline struct sock *inet_lookup(struct net *net,
371 struct inet_hashinfo *hashinfo,
372 const __be32 saddr, const __be16 sport,
373 const __be32 daddr, const __be16 dport,
379 sk = __inet_lookup(net, hashinfo, saddr, sport, daddr, dport, dif);
385 static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
390 struct sock *sk = skb_steal_sock(skb);
391 const struct iphdr *iph = ip_hdr(skb);
396 return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo,
398 iph->daddr, dport, inet_iif(skb));
401 int __inet_hash_connect(struct inet_timewait_death_row *death_row,
402 struct sock *sk, u32 port_offset,
403 int (*check_established)(struct inet_timewait_death_row *,
404 struct sock *, __u16,
405 struct inet_timewait_sock **),
406 int (*hash)(struct sock *sk,
407 struct inet_timewait_sock *twp));
409 int inet_hash_connect(struct inet_timewait_death_row *death_row,
411 #endif /* _INET_HASHTABLES_H */