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
17
18 #include <linux/interrupt.h>
19 #include <linux/ip.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>
28
29 #include <net/inet_connection_sock.h>
30 #include <net/inet_sock.h>
31 #include <net/sock.h>
32 #include <net/route.h>
33 #include <net/tcp_states.h>
34 #include <net/netns/hash.h>
35
36 #include <asm/atomic.h>
37 #include <asm/byteorder.h>
38
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.
42 */
43 struct inet_ehash_bucket {
44 struct hlist_nulls_head chain;
45 struct hlist_nulls_head twchain;
46 };
47
48 /* There are a few simple rules, which allow for local port reuse by
49 * an application. In essence:
50 *
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
58 * shared.
59 * Failing this, the port cannot be shared.
60 *
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.
71 *
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
78 */
79 struct inet_bind_bucket {
80 #ifdef CONFIG_NET_NS
81 struct net *ib_net;
82 #endif
83 unsigned short port;
84 signed short fastreuse;
85 struct hlist_node node;
86 struct hlist_head owners;
87 };
88
ib_net(struct inet_bind_bucket * ib)89 static inline struct net *ib_net(struct inet_bind_bucket *ib)
90 {
91 return read_pnet(&ib->ib_net);
92 }
93
94 #define inet_bind_bucket_for_each(tb, node, head) \
95 hlist_for_each_entry(tb, node, head, node)
96
97 struct inet_bind_hashbucket {
98 spinlock_t lock;
99 struct hlist_head chain;
100 };
101
102 /*
103 * Sockets can be hashed in established or listening table
104 * We must use different 'nulls' end-of-chain value for listening
105 * hash table, or we might find a socket that was closed and
106 * reallocated/inserted into established hash table
107 */
108 #define LISTENING_NULLS_BASE (1U << 29)
109 struct inet_listen_hashbucket {
110 spinlock_t lock;
111 struct hlist_nulls_head head;
112 };
113
114 /* This is for listening sockets, thus all sockets which possess wildcards. */
115 #define INET_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */
116
117 struct inet_hashinfo {
118 /* This is for sockets with full identity only. Sockets here will
119 * always be without wildcards and will have the following invariant:
120 *
121 * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
122 *
123 * TIME_WAIT sockets use a separate chain (twchain).
124 */
125 struct inet_ehash_bucket *ehash;
126 spinlock_t *ehash_locks;
127 unsigned int ehash_size;
128 unsigned int ehash_locks_mask;
129
130 /* Ok, let's try this, I give up, we do need a local binding
131 * TCP hash as well as the others for fast bind/connect.
132 */
133 struct inet_bind_hashbucket *bhash;
134
135 unsigned int bhash_size;
136 /* Note : 4 bytes padding on 64 bit arches */
137
138 struct kmem_cache *bind_bucket_cachep;
139
140 /* All the above members are written once at bootup and
141 * never written again _or_ are predominantly read-access.
142 *
143 * Now align to a new cache line as all the following members
144 * might be often dirty.
145 */
146 /* All sockets in TCP_LISTEN state will be in here. This is the only
147 * table where wildcard'd TCP sockets can exist. Hash function here
148 * is just local port number.
149 */
150 struct inet_listen_hashbucket listening_hash[INET_LHTABLE_SIZE]
151 ____cacheline_aligned_in_smp;
152
153 };
154
inet_ehash_bucket(struct inet_hashinfo * hashinfo,unsigned int hash)155 static inline struct inet_ehash_bucket *inet_ehash_bucket(
156 struct inet_hashinfo *hashinfo,
157 unsigned int hash)
158 {
159 return &hashinfo->ehash[hash & (hashinfo->ehash_size - 1)];
160 }
161
inet_ehash_lockp(struct inet_hashinfo * hashinfo,unsigned int hash)162 static inline spinlock_t *inet_ehash_lockp(
163 struct inet_hashinfo *hashinfo,
164 unsigned int hash)
165 {
166 return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
167 }
168
inet_ehash_locks_alloc(struct inet_hashinfo * hashinfo)169 static inline int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo)
170 {
171 unsigned int i, size = 256;
172 #if defined(CONFIG_PROVE_LOCKING)
173 unsigned int nr_pcpus = 2;
174 #else
175 unsigned int nr_pcpus = num_possible_cpus();
176 #endif
177 if (nr_pcpus >= 4)
178 size = 512;
179 if (nr_pcpus >= 8)
180 size = 1024;
181 if (nr_pcpus >= 16)
182 size = 2048;
183 if (nr_pcpus >= 32)
184 size = 4096;
185 if (sizeof(spinlock_t) != 0) {
186 #ifdef CONFIG_NUMA
187 if (size * sizeof(spinlock_t) > PAGE_SIZE)
188 hashinfo->ehash_locks = vmalloc(size * sizeof(spinlock_t));
189 else
190 #endif
191 hashinfo->ehash_locks = kmalloc(size * sizeof(spinlock_t),
192 GFP_KERNEL);
193 if (!hashinfo->ehash_locks)
194 return ENOMEM;
195 for (i = 0; i < size; i++)
196 spin_lock_init(&hashinfo->ehash_locks[i]);
197 }
198 hashinfo->ehash_locks_mask = size - 1;
199 return 0;
200 }
201
inet_ehash_locks_free(struct inet_hashinfo * hashinfo)202 static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
203 {
204 if (hashinfo->ehash_locks) {
205 #ifdef CONFIG_NUMA
206 unsigned int size = (hashinfo->ehash_locks_mask + 1) *
207 sizeof(spinlock_t);
208 if (size > PAGE_SIZE)
209 vfree(hashinfo->ehash_locks);
210 else
211 #endif
212 kfree(hashinfo->ehash_locks);
213 hashinfo->ehash_locks = NULL;
214 }
215 }
216
217 extern struct inet_bind_bucket *
218 inet_bind_bucket_create(struct kmem_cache *cachep,
219 struct net *net,
220 struct inet_bind_hashbucket *head,
221 const unsigned short snum);
222 extern void inet_bind_bucket_destroy(struct kmem_cache *cachep,
223 struct inet_bind_bucket *tb);
224
inet_bhashfn(struct net * net,const __u16 lport,const int bhash_size)225 static inline int inet_bhashfn(struct net *net,
226 const __u16 lport, const int bhash_size)
227 {
228 return (lport + net_hash_mix(net)) & (bhash_size - 1);
229 }
230
231 extern void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
232 const unsigned short snum);
233
234 /* These can have wildcards, don't try too hard. */
inet_lhashfn(struct net * net,const unsigned short num)235 static inline int inet_lhashfn(struct net *net, const unsigned short num)
236 {
237 return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1);
238 }
239
inet_sk_listen_hashfn(const struct sock * sk)240 static inline int inet_sk_listen_hashfn(const struct sock *sk)
241 {
242 return inet_lhashfn(sock_net(sk), inet_sk(sk)->num);
243 }
244
245 /* Caller must disable local BH processing. */
246 extern void __inet_inherit_port(struct sock *sk, struct sock *child);
247
248 extern void inet_put_port(struct sock *sk);
249
250 void inet_hashinfo_init(struct inet_hashinfo *h);
251
252 extern void __inet_hash_nolisten(struct sock *sk);
253 extern void inet_hash(struct sock *sk);
254 extern void inet_unhash(struct sock *sk);
255
256 extern struct sock *__inet_lookup_listener(struct net *net,
257 struct inet_hashinfo *hashinfo,
258 const __be32 daddr,
259 const unsigned short hnum,
260 const int dif);
261
inet_lookup_listener(struct net * net,struct inet_hashinfo * hashinfo,__be32 daddr,__be16 dport,int dif)262 static inline struct sock *inet_lookup_listener(struct net *net,
263 struct inet_hashinfo *hashinfo,
264 __be32 daddr, __be16 dport, int dif)
265 {
266 return __inet_lookup_listener(net, hashinfo, daddr, ntohs(dport), dif);
267 }
268
269 /* Socket demux engine toys. */
270 /* What happens here is ugly; there's a pair of adjacent fields in
271 struct inet_sock; __be16 dport followed by __u16 num. We want to
272 search by pair, so we combine the keys into a single 32bit value
273 and compare with 32bit value read from &...->dport. Let's at least
274 make sure that it's not mixed with anything else...
275 On 64bit targets we combine comparisons with pair of adjacent __be32
276 fields in the same way.
277 */
278 typedef __u32 __bitwise __portpair;
279 #ifdef __BIG_ENDIAN
280 #define INET_COMBINED_PORTS(__sport, __dport) \
281 ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
282 #else /* __LITTLE_ENDIAN */
283 #define INET_COMBINED_PORTS(__sport, __dport) \
284 ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
285 #endif
286
287 #if (BITS_PER_LONG == 64)
288 typedef __u64 __bitwise __addrpair;
289 #ifdef __BIG_ENDIAN
290 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
291 const __addrpair __name = (__force __addrpair) ( \
292 (((__force __u64)(__be32)(__saddr)) << 32) | \
293 ((__force __u64)(__be32)(__daddr)));
294 #else /* __LITTLE_ENDIAN */
295 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
296 const __addrpair __name = (__force __addrpair) ( \
297 (((__force __u64)(__be32)(__daddr)) << 32) | \
298 ((__force __u64)(__be32)(__saddr)));
299 #endif /* __BIG_ENDIAN */
300 #define INET_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
301 (((__sk)->sk_hash == (__hash)) && net_eq(sock_net(__sk), (__net)) && \
302 ((*((__addrpair *)&(inet_sk(__sk)->daddr))) == (__cookie)) && \
303 ((*((__portpair *)&(inet_sk(__sk)->dport))) == (__ports)) && \
304 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
305 #define INET_TW_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
306 (((__sk)->sk_hash == (__hash)) && net_eq(sock_net(__sk), (__net)) && \
307 ((*((__addrpair *)&(inet_twsk(__sk)->tw_daddr))) == (__cookie)) && \
308 ((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \
309 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
310 #else /* 32-bit arch */
311 #define INET_ADDR_COOKIE(__name, __saddr, __daddr)
312 #define INET_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif) \
313 (((__sk)->sk_hash == (__hash)) && net_eq(sock_net(__sk), (__net)) && \
314 (inet_sk(__sk)->daddr == (__saddr)) && \
315 (inet_sk(__sk)->rcv_saddr == (__daddr)) && \
316 ((*((__portpair *)&(inet_sk(__sk)->dport))) == (__ports)) && \
317 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
318 #define INET_TW_MATCH(__sk, __net, __hash,__cookie, __saddr, __daddr, __ports, __dif) \
319 (((__sk)->sk_hash == (__hash)) && net_eq(sock_net(__sk), (__net)) && \
320 (inet_twsk(__sk)->tw_daddr == (__saddr)) && \
321 (inet_twsk(__sk)->tw_rcv_saddr == (__daddr)) && \
322 ((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \
323 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
324 #endif /* 64-bit arch */
325
326 /*
327 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
328 * not check it for lookups anymore, thanks Alexey. -DaveM
329 *
330 * Local BH must be disabled here.
331 */
332 extern struct sock * __inet_lookup_established(struct net *net,
333 struct inet_hashinfo *hashinfo,
334 const __be32 saddr, const __be16 sport,
335 const __be32 daddr, const u16 hnum, const int dif);
336
337 static inline struct sock *
inet_lookup_established(struct net * net,struct inet_hashinfo * hashinfo,const __be32 saddr,const __be16 sport,const __be32 daddr,const __be16 dport,const int dif)338 inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
339 const __be32 saddr, const __be16 sport,
340 const __be32 daddr, const __be16 dport,
341 const int dif)
342 {
343 return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
344 ntohs(dport), dif);
345 }
346
__inet_lookup(struct net * net,struct inet_hashinfo * hashinfo,const __be32 saddr,const __be16 sport,const __be32 daddr,const __be16 dport,const int dif)347 static inline struct sock *__inet_lookup(struct net *net,
348 struct inet_hashinfo *hashinfo,
349 const __be32 saddr, const __be16 sport,
350 const __be32 daddr, const __be16 dport,
351 const int dif)
352 {
353 u16 hnum = ntohs(dport);
354 struct sock *sk = __inet_lookup_established(net, hashinfo,
355 saddr, sport, daddr, hnum, dif);
356
357 return sk ? : __inet_lookup_listener(net, hashinfo, daddr, hnum, dif);
358 }
359
inet_lookup(struct net * net,struct inet_hashinfo * hashinfo,const __be32 saddr,const __be16 sport,const __be32 daddr,const __be16 dport,const int dif)360 static inline struct sock *inet_lookup(struct net *net,
361 struct inet_hashinfo *hashinfo,
362 const __be32 saddr, const __be16 sport,
363 const __be32 daddr, const __be16 dport,
364 const int dif)
365 {
366 struct sock *sk;
367
368 local_bh_disable();
369 sk = __inet_lookup(net, hashinfo, saddr, sport, daddr, dport, dif);
370 local_bh_enable();
371
372 return sk;
373 }
374
__inet_lookup_skb(struct inet_hashinfo * hashinfo,struct sk_buff * skb,const __be16 sport,const __be16 dport)375 static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
376 struct sk_buff *skb,
377 const __be16 sport,
378 const __be16 dport)
379 {
380 struct sock *sk;
381 const struct iphdr *iph = ip_hdr(skb);
382
383 if (unlikely(sk = skb_steal_sock(skb)))
384 return sk;
385 else
386 return __inet_lookup(dev_net(skb->dst->dev), hashinfo,
387 iph->saddr, sport,
388 iph->daddr, dport, inet_iif(skb));
389 }
390
391 extern int __inet_hash_connect(struct inet_timewait_death_row *death_row,
392 struct sock *sk, u32 port_offset,
393 int (*check_established)(struct inet_timewait_death_row *,
394 struct sock *, __u16, struct inet_timewait_sock **),
395 void (*hash)(struct sock *sk));
396 extern int inet_hash_connect(struct inet_timewait_death_row *death_row,
397 struct sock *sk);
398 #endif /* _INET_HASHTABLES_H */
399