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>
28 #include <net/inet_connection_sock.h>
29 #include <net/inet_sock.h>
31 #include <net/route.h>
32 #include <net/tcp_states.h>
33 #include <net/netns/hash.h>
35 #include <linux/refcount.h>
36 #include <asm/byteorder.h>
38 /* This is for all connections with a full identity, no wildcards.
39 * The 'e' prefix stands for Establish, but we really put all sockets
42 struct inet_ehash_bucket {
43 struct hlist_nulls_head chain;
46 /* There are a few simple rules, which allow for local port reuse by
47 * an application. In essence:
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
57 * Failing this, the port cannot be shared.
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.
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
77 #define FASTREUSEPORT_ANY 1
78 #define FASTREUSEPORT_STRICT 2
80 struct inet_bind_bucket {
81 possible_net_t ib_net;
83 signed char fastreuse;
84 signed char fastreuseport;
86 #if IS_ENABLED(CONFIG_IPV6)
87 struct in6_addr fast_v6_rcv_saddr;
89 __be32 fast_rcv_saddr;
90 unsigned short fast_sk_family;
92 struct hlist_node node;
93 struct hlist_head owners;
96 static inline struct net *ib_net(struct inet_bind_bucket *ib)
98 return read_pnet(&ib->ib_net);
101 #define inet_bind_bucket_for_each(tb, head) \
102 hlist_for_each_entry(tb, head, node)
104 struct inet_bind_hashbucket {
106 struct hlist_head chain;
109 /* Sockets can be hashed in established or listening table.
110 * We must use different 'nulls' end-of-chain value for all hash buckets :
111 * A socket might transition from ESTABLISH to LISTEN state without
112 * RCU grace period. A lookup in ehash table needs to handle this case.
114 #define LISTENING_NULLS_BASE (1U << 29)
115 struct inet_listen_hashbucket {
119 struct hlist_head head;
120 struct hlist_nulls_head nulls_head;
124 /* This is for listening sockets, thus all sockets which possess wildcards. */
125 #define INET_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */
127 struct inet_hashinfo {
128 /* This is for sockets with full identity only. Sockets here will
129 * always be without wildcards and will have the following invariant:
131 * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
134 struct inet_ehash_bucket *ehash;
135 spinlock_t *ehash_locks;
136 unsigned int ehash_mask;
137 unsigned int ehash_locks_mask;
139 /* Ok, let's try this, I give up, we do need a local binding
140 * TCP hash as well as the others for fast bind/connect.
142 struct kmem_cache *bind_bucket_cachep;
143 struct inet_bind_hashbucket *bhash;
144 unsigned int bhash_size;
146 /* The 2nd listener table hashed by local port and address */
147 unsigned int lhash2_mask;
148 struct inet_listen_hashbucket *lhash2;
150 /* All the above members are written once at bootup and
151 * never written again _or_ are predominantly read-access.
153 * Now align to a new cache line as all the following members
154 * might be often dirty.
156 /* All sockets in TCP_LISTEN state will be in listening_hash.
157 * This is the only table where wildcard'd TCP sockets can
158 * exist. listening_hash is only hashed by local port number.
159 * If lhash2 is initialized, the same socket will also be hashed
160 * to lhash2 by port and address.
162 struct inet_listen_hashbucket listening_hash[INET_LHTABLE_SIZE]
163 ____cacheline_aligned_in_smp;
166 #define inet_lhash2_for_each_icsk_rcu(__icsk, list) \
167 hlist_for_each_entry_rcu(__icsk, list, icsk_listen_portaddr_node)
169 static inline struct inet_listen_hashbucket *
170 inet_lhash2_bucket(struct inet_hashinfo *h, u32 hash)
172 return &h->lhash2[hash & h->lhash2_mask];
175 static inline struct inet_ehash_bucket *inet_ehash_bucket(
176 struct inet_hashinfo *hashinfo,
179 return &hashinfo->ehash[hash & hashinfo->ehash_mask];
182 static inline spinlock_t *inet_ehash_lockp(
183 struct inet_hashinfo *hashinfo,
186 return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
189 int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo);
191 static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
193 kvfree(hashinfo->ehash_locks);
194 hashinfo->ehash_locks = NULL;
197 struct inet_bind_bucket *
198 inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net,
199 struct inet_bind_hashbucket *head,
200 const unsigned short snum);
201 void inet_bind_bucket_destroy(struct kmem_cache *cachep,
202 struct inet_bind_bucket *tb);
204 static inline u32 inet_bhashfn(const struct net *net, const __u16 lport,
205 const u32 bhash_size)
207 return (lport + net_hash_mix(net)) & (bhash_size - 1);
210 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
211 const unsigned short snum);
213 /* These can have wildcards, don't try too hard. */
214 static inline u32 inet_lhashfn(const struct net *net, const unsigned short num)
216 return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1);
219 static inline int inet_sk_listen_hashfn(const struct sock *sk)
221 return inet_lhashfn(sock_net(sk), inet_sk(sk)->inet_num);
224 /* Caller must disable local BH processing. */
225 int __inet_inherit_port(const struct sock *sk, struct sock *child);
227 void inet_put_port(struct sock *sk);
229 void inet_hashinfo_init(struct inet_hashinfo *h);
230 void inet_hashinfo2_init(struct inet_hashinfo *h, const char *name,
231 unsigned long numentries, int scale,
232 unsigned long low_limit,
233 unsigned long high_limit);
235 bool inet_ehash_insert(struct sock *sk, struct sock *osk, bool *found_dup_sk);
236 bool inet_ehash_nolisten(struct sock *sk, struct sock *osk,
238 int __inet_hash(struct sock *sk, struct sock *osk);
239 int inet_hash(struct sock *sk);
240 void inet_unhash(struct sock *sk);
242 struct sock *__inet_lookup_listener(struct net *net,
243 struct inet_hashinfo *hashinfo,
244 struct sk_buff *skb, int doff,
245 const __be32 saddr, const __be16 sport,
247 const unsigned short hnum,
248 const int dif, const int sdif);
250 static inline struct sock *inet_lookup_listener(struct net *net,
251 struct inet_hashinfo *hashinfo,
252 struct sk_buff *skb, int doff,
253 __be32 saddr, __be16 sport,
254 __be32 daddr, __be16 dport, int dif, int sdif)
256 return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, sport,
257 daddr, ntohs(dport), dif, sdif);
260 /* Socket demux engine toys. */
261 /* What happens here is ugly; there's a pair of adjacent fields in
262 struct inet_sock; __be16 dport followed by __u16 num. We want to
263 search by pair, so we combine the keys into a single 32bit value
264 and compare with 32bit value read from &...->dport. Let's at least
265 make sure that it's not mixed with anything else...
266 On 64bit targets we combine comparisons with pair of adjacent __be32
267 fields in the same way.
270 #define INET_COMBINED_PORTS(__sport, __dport) \
271 ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
272 #else /* __LITTLE_ENDIAN */
273 #define INET_COMBINED_PORTS(__sport, __dport) \
274 ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
277 #if (BITS_PER_LONG == 64)
279 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
280 const __addrpair __name = (__force __addrpair) ( \
281 (((__force __u64)(__be32)(__saddr)) << 32) | \
282 ((__force __u64)(__be32)(__daddr)))
283 #else /* __LITTLE_ENDIAN */
284 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
285 const __addrpair __name = (__force __addrpair) ( \
286 (((__force __u64)(__be32)(__daddr)) << 32) | \
287 ((__force __u64)(__be32)(__saddr)))
288 #endif /* __BIG_ENDIAN */
289 #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif, __sdif) \
290 (((__sk)->sk_portpair == (__ports)) && \
291 ((__sk)->sk_addrpair == (__cookie)) && \
292 (!(__sk)->sk_bound_dev_if || \
293 ((__sk)->sk_bound_dev_if == (__dif)) || \
294 ((__sk)->sk_bound_dev_if == (__sdif))) && \
295 net_eq(sock_net(__sk), (__net)))
296 #else /* 32-bit arch */
297 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
298 const int __name __deprecated __attribute__((unused))
300 #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif, __sdif) \
301 (((__sk)->sk_portpair == (__ports)) && \
302 ((__sk)->sk_daddr == (__saddr)) && \
303 ((__sk)->sk_rcv_saddr == (__daddr)) && \
304 (!(__sk)->sk_bound_dev_if || \
305 ((__sk)->sk_bound_dev_if == (__dif)) || \
306 ((__sk)->sk_bound_dev_if == (__sdif))) && \
307 net_eq(sock_net(__sk), (__net)))
308 #endif /* 64-bit arch */
310 /* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
311 * not check it for lookups anymore, thanks Alexey. -DaveM
313 struct sock *__inet_lookup_established(struct net *net,
314 struct inet_hashinfo *hashinfo,
315 const __be32 saddr, const __be16 sport,
316 const __be32 daddr, const u16 hnum,
317 const int dif, const int sdif);
319 static inline struct sock *
320 inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
321 const __be32 saddr, const __be16 sport,
322 const __be32 daddr, const __be16 dport,
325 return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
326 ntohs(dport), dif, 0);
329 static inline struct sock *__inet_lookup(struct net *net,
330 struct inet_hashinfo *hashinfo,
331 struct sk_buff *skb, int doff,
332 const __be32 saddr, const __be16 sport,
333 const __be32 daddr, const __be16 dport,
334 const int dif, const int sdif,
337 u16 hnum = ntohs(dport);
340 sk = __inet_lookup_established(net, hashinfo, saddr, sport,
341 daddr, hnum, dif, sdif);
346 return __inet_lookup_listener(net, hashinfo, skb, doff, saddr,
347 sport, daddr, hnum, dif, sdif);
350 static inline struct sock *inet_lookup(struct net *net,
351 struct inet_hashinfo *hashinfo,
352 struct sk_buff *skb, int doff,
353 const __be32 saddr, const __be16 sport,
354 const __be32 daddr, const __be16 dport,
360 sk = __inet_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
361 dport, dif, 0, &refcounted);
363 if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
368 static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
376 struct sock *sk = skb_steal_sock(skb);
377 const struct iphdr *iph = ip_hdr(skb);
383 return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo, skb,
384 doff, iph->saddr, sport,
385 iph->daddr, dport, inet_iif(skb), sdif,
389 u32 inet6_ehashfn(const struct net *net,
390 const struct in6_addr *laddr, const u16 lport,
391 const struct in6_addr *faddr, const __be16 fport);
393 static inline void sk_daddr_set(struct sock *sk, __be32 addr)
395 sk->sk_daddr = addr; /* alias of inet_daddr */
396 #if IS_ENABLED(CONFIG_IPV6)
397 ipv6_addr_set_v4mapped(addr, &sk->sk_v6_daddr);
401 static inline void sk_rcv_saddr_set(struct sock *sk, __be32 addr)
403 sk->sk_rcv_saddr = addr; /* alias of inet_rcv_saddr */
404 #if IS_ENABLED(CONFIG_IPV6)
405 ipv6_addr_set_v4mapped(addr, &sk->sk_v6_rcv_saddr);
409 int __inet_hash_connect(struct inet_timewait_death_row *death_row,
410 struct sock *sk, u32 port_offset,
411 int (*check_established)(struct inet_timewait_death_row *,
412 struct sock *, __u16,
413 struct inet_timewait_sock **));
415 int inet_hash_connect(struct inet_timewait_death_row *death_row,
417 #endif /* _INET_HASHTABLES_H */