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 {
118 struct hlist_head head;
119 struct hlist_nulls_head nulls_head;
123 /* This is for listening sockets, thus all sockets which possess wildcards. */
124 #define INET_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */
126 struct inet_hashinfo {
127 /* This is for sockets with full identity only. Sockets here will
128 * always be without wildcards and will have the following invariant:
130 * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
133 struct inet_ehash_bucket *ehash;
134 spinlock_t *ehash_locks;
135 unsigned int ehash_mask;
136 unsigned int ehash_locks_mask;
138 /* Ok, let's try this, I give up, we do need a local binding
139 * TCP hash as well as the others for fast bind/connect.
141 struct inet_bind_hashbucket *bhash;
143 unsigned int bhash_size;
144 /* 4 bytes hole on 64 bit */
146 struct kmem_cache *bind_bucket_cachep;
148 /* All the above members are written once at bootup and
149 * never written again _or_ are predominantly read-access.
151 * Now align to a new cache line as all the following members
152 * might be often dirty.
154 /* All sockets in TCP_LISTEN state will be in here. This is the only
155 * table where wildcard'd TCP sockets can exist. Hash function here
156 * is just local port number.
158 struct inet_listen_hashbucket listening_hash[INET_LHTABLE_SIZE]
159 ____cacheline_aligned_in_smp;
162 static inline struct inet_ehash_bucket *inet_ehash_bucket(
163 struct inet_hashinfo *hashinfo,
166 return &hashinfo->ehash[hash & hashinfo->ehash_mask];
169 static inline spinlock_t *inet_ehash_lockp(
170 struct inet_hashinfo *hashinfo,
173 return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
176 int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo);
178 static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
180 kvfree(hashinfo->ehash_locks);
181 hashinfo->ehash_locks = NULL;
184 struct inet_bind_bucket *
185 inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net,
186 struct inet_bind_hashbucket *head,
187 const unsigned short snum);
188 void inet_bind_bucket_destroy(struct kmem_cache *cachep,
189 struct inet_bind_bucket *tb);
191 static inline u32 inet_bhashfn(const struct net *net, const __u16 lport,
192 const u32 bhash_size)
194 return (lport + net_hash_mix(net)) & (bhash_size - 1);
197 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
198 const unsigned short snum);
200 /* These can have wildcards, don't try too hard. */
201 static inline u32 inet_lhashfn(const struct net *net, const unsigned short num)
203 return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1);
206 static inline int inet_sk_listen_hashfn(const struct sock *sk)
208 return inet_lhashfn(sock_net(sk), inet_sk(sk)->inet_num);
211 /* Caller must disable local BH processing. */
212 int __inet_inherit_port(const struct sock *sk, struct sock *child);
214 void inet_put_port(struct sock *sk);
216 void inet_hashinfo_init(struct inet_hashinfo *h);
218 bool inet_ehash_insert(struct sock *sk, struct sock *osk, bool *found_dup_sk);
219 bool inet_ehash_nolisten(struct sock *sk, struct sock *osk,
221 int __inet_hash(struct sock *sk, struct sock *osk);
222 int inet_hash(struct sock *sk);
223 void inet_unhash(struct sock *sk);
225 struct sock *__inet_lookup_listener(struct net *net,
226 struct inet_hashinfo *hashinfo,
227 struct sk_buff *skb, int doff,
228 const __be32 saddr, const __be16 sport,
230 const unsigned short hnum,
231 const int dif, const int sdif);
233 static inline struct sock *inet_lookup_listener(struct net *net,
234 struct inet_hashinfo *hashinfo,
235 struct sk_buff *skb, int doff,
236 __be32 saddr, __be16 sport,
237 __be32 daddr, __be16 dport, int dif, int sdif)
239 return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, sport,
240 daddr, ntohs(dport), dif, sdif);
243 /* Socket demux engine toys. */
244 /* What happens here is ugly; there's a pair of adjacent fields in
245 struct inet_sock; __be16 dport followed by __u16 num. We want to
246 search by pair, so we combine the keys into a single 32bit value
247 and compare with 32bit value read from &...->dport. Let's at least
248 make sure that it's not mixed with anything else...
249 On 64bit targets we combine comparisons with pair of adjacent __be32
250 fields in the same way.
253 #define INET_COMBINED_PORTS(__sport, __dport) \
254 ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
255 #else /* __LITTLE_ENDIAN */
256 #define INET_COMBINED_PORTS(__sport, __dport) \
257 ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
260 #if (BITS_PER_LONG == 64)
262 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
263 const __addrpair __name = (__force __addrpair) ( \
264 (((__force __u64)(__be32)(__saddr)) << 32) | \
265 ((__force __u64)(__be32)(__daddr)))
266 #else /* __LITTLE_ENDIAN */
267 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
268 const __addrpair __name = (__force __addrpair) ( \
269 (((__force __u64)(__be32)(__daddr)) << 32) | \
270 ((__force __u64)(__be32)(__saddr)))
271 #endif /* __BIG_ENDIAN */
272 #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif, __sdif) \
273 (((__sk)->sk_portpair == (__ports)) && \
274 ((__sk)->sk_addrpair == (__cookie)) && \
275 (!(__sk)->sk_bound_dev_if || \
276 ((__sk)->sk_bound_dev_if == (__dif)) || \
277 ((__sk)->sk_bound_dev_if == (__sdif))) && \
278 net_eq(sock_net(__sk), (__net)))
279 #else /* 32-bit arch */
280 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
281 const int __name __deprecated __attribute__((unused))
283 #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif, __sdif) \
284 (((__sk)->sk_portpair == (__ports)) && \
285 ((__sk)->sk_daddr == (__saddr)) && \
286 ((__sk)->sk_rcv_saddr == (__daddr)) && \
287 (!(__sk)->sk_bound_dev_if || \
288 ((__sk)->sk_bound_dev_if == (__dif)) || \
289 ((__sk)->sk_bound_dev_if == (__sdif))) && \
290 net_eq(sock_net(__sk), (__net)))
291 #endif /* 64-bit arch */
293 /* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
294 * not check it for lookups anymore, thanks Alexey. -DaveM
296 struct sock *__inet_lookup_established(struct net *net,
297 struct inet_hashinfo *hashinfo,
298 const __be32 saddr, const __be16 sport,
299 const __be32 daddr, const u16 hnum,
300 const int dif, const int sdif);
302 static inline struct sock *
303 inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
304 const __be32 saddr, const __be16 sport,
305 const __be32 daddr, const __be16 dport,
308 return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
309 ntohs(dport), dif, 0);
312 static inline struct sock *__inet_lookup(struct net *net,
313 struct inet_hashinfo *hashinfo,
314 struct sk_buff *skb, int doff,
315 const __be32 saddr, const __be16 sport,
316 const __be32 daddr, const __be16 dport,
317 const int dif, const int sdif,
320 u16 hnum = ntohs(dport);
323 sk = __inet_lookup_established(net, hashinfo, saddr, sport,
324 daddr, hnum, dif, sdif);
329 return __inet_lookup_listener(net, hashinfo, skb, doff, saddr,
330 sport, daddr, hnum, dif, sdif);
333 static inline struct sock *inet_lookup(struct net *net,
334 struct inet_hashinfo *hashinfo,
335 struct sk_buff *skb, int doff,
336 const __be32 saddr, const __be16 sport,
337 const __be32 daddr, const __be16 dport,
343 sk = __inet_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
344 dport, dif, 0, &refcounted);
346 if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
351 static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
359 struct sock *sk = skb_steal_sock(skb);
360 const struct iphdr *iph = ip_hdr(skb);
366 return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo, skb,
367 doff, iph->saddr, sport,
368 iph->daddr, dport, inet_iif(skb), sdif,
372 u32 inet6_ehashfn(const struct net *net,
373 const struct in6_addr *laddr, const u16 lport,
374 const struct in6_addr *faddr, const __be16 fport);
376 static inline void sk_daddr_set(struct sock *sk, __be32 addr)
378 sk->sk_daddr = addr; /* alias of inet_daddr */
379 #if IS_ENABLED(CONFIG_IPV6)
380 ipv6_addr_set_v4mapped(addr, &sk->sk_v6_daddr);
384 static inline void sk_rcv_saddr_set(struct sock *sk, __be32 addr)
386 sk->sk_rcv_saddr = addr; /* alias of inet_rcv_saddr */
387 #if IS_ENABLED(CONFIG_IPV6)
388 ipv6_addr_set_v4mapped(addr, &sk->sk_v6_rcv_saddr);
392 int __inet_hash_connect(struct inet_timewait_death_row *death_row,
393 struct sock *sk, u64 port_offset,
394 int (*check_established)(struct inet_timewait_death_row *,
395 struct sock *, __u16,
396 struct inet_timewait_sock **));
398 int inet_hash_connect(struct inet_timewait_death_row *death_row,
400 #endif /* _INET_HASHTABLES_H */