1 /* SPDX-License-Identifier: GPL-2.0-or-later */
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * Authors: Lotsa people, from code originally in tcp
10 #ifndef _INET_HASHTABLES_H
11 #define _INET_HASHTABLES_H
14 #include <linux/interrupt.h>
16 #include <linux/ipv6.h>
17 #include <linux/list.h>
18 #include <linux/slab.h>
19 #include <linux/socket.h>
20 #include <linux/spinlock.h>
21 #include <linux/types.h>
22 #include <linux/wait.h>
24 #include <net/inet_connection_sock.h>
25 #include <net/inet_sock.h>
28 #include <net/route.h>
29 #include <net/tcp_states.h>
30 #include <net/netns/hash.h>
32 #include <linux/refcount.h>
33 #include <asm/byteorder.h>
35 /* This is for all connections with a full identity, no wildcards.
36 * The 'e' prefix stands for Establish, but we really put all sockets
39 struct inet_ehash_bucket {
40 struct hlist_nulls_head chain;
43 /* There are a few simple rules, which allow for local port reuse by
44 * an application. In essence:
46 * 1) Sockets bound to different interfaces may share a local port.
47 * Failing that, goto test 2.
48 * 2) If all sockets have sk->sk_reuse set, and none of them are in
49 * TCP_LISTEN state, the port may be shared.
50 * Failing that, goto test 3.
51 * 3) If all sockets are bound to a specific inet_sk(sk)->rcv_saddr local
52 * address, and none of them are the same, the port may be
54 * Failing this, the port cannot be shared.
56 * The interesting point, is test #2. This is what an FTP server does
57 * all day. To optimize this case we use a specific flag bit defined
58 * below. As we add sockets to a bind bucket list, we perform a
59 * check of: (newsk->sk_reuse && (newsk->sk_state != TCP_LISTEN))
60 * As long as all sockets added to a bind bucket pass this test,
61 * the flag bit will be set.
62 * The resulting situation is that tcp_v[46]_verify_bind() can just check
63 * for this flag bit, if it is set and the socket trying to bind has
64 * sk->sk_reuse set, we don't even have to walk the owners list at all,
65 * we return that it is ok to bind this socket to the requested local port.
67 * Sounds like a lot of work, but it is worth it. In a more naive
68 * implementation (ie. current FreeBSD etc.) the entire list of ports
69 * must be walked for each data port opened by an ftp server. Needless
70 * to say, this does not scale at all. With a couple thousand FTP
71 * users logged onto your box, isn't it nice to know that new data
72 * ports are created in O(1) time? I thought so. ;-) -DaveM
74 #define FASTREUSEPORT_ANY 1
75 #define FASTREUSEPORT_STRICT 2
77 struct inet_bind_bucket {
78 possible_net_t ib_net;
81 signed char fastreuse;
82 signed char fastreuseport;
84 #if IS_ENABLED(CONFIG_IPV6)
85 struct in6_addr fast_v6_rcv_saddr;
87 __be32 fast_rcv_saddr;
88 unsigned short fast_sk_family;
90 struct hlist_node node;
91 struct hlist_head owners;
94 struct inet_bind2_bucket {
95 possible_net_t ib_net;
99 #if IS_ENABLED(CONFIG_IPV6)
100 struct in6_addr v6_rcv_saddr;
104 /* Node in the bhash2 inet_bind_hashbucket chain */
105 struct hlist_node node;
106 /* List of sockets hashed to this bucket */
107 struct hlist_head owners;
110 static inline struct net *ib_net(const struct inet_bind_bucket *ib)
112 return read_pnet(&ib->ib_net);
115 static inline struct net *ib2_net(const struct inet_bind2_bucket *ib)
117 return read_pnet(&ib->ib_net);
120 #define inet_bind_bucket_for_each(tb, head) \
121 hlist_for_each_entry(tb, head, node)
123 struct inet_bind_hashbucket {
125 struct hlist_head chain;
128 /* Sockets can be hashed in established or listening table.
129 * We must use different 'nulls' end-of-chain value for all hash buckets :
130 * A socket might transition from ESTABLISH to LISTEN state without
131 * RCU grace period. A lookup in ehash table needs to handle this case.
133 #define LISTENING_NULLS_BASE (1U << 29)
134 struct inet_listen_hashbucket {
136 struct hlist_nulls_head nulls_head;
139 /* This is for listening sockets, thus all sockets which possess wildcards. */
140 #define INET_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */
142 struct inet_hashinfo {
143 /* This is for sockets with full identity only. Sockets here will
144 * always be without wildcards and will have the following invariant:
146 * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
149 struct inet_ehash_bucket *ehash;
150 spinlock_t *ehash_locks;
151 unsigned int ehash_mask;
152 unsigned int ehash_locks_mask;
154 /* Ok, let's try this, I give up, we do need a local binding
155 * TCP hash as well as the others for fast bind/connect.
157 struct kmem_cache *bind_bucket_cachep;
158 /* This bind table is hashed by local port */
159 struct inet_bind_hashbucket *bhash;
160 struct kmem_cache *bind2_bucket_cachep;
161 /* This bind table is hashed by local port and sk->sk_rcv_saddr (ipv4)
162 * or sk->sk_v6_rcv_saddr (ipv6). This 2nd bind table is used
163 * primarily for expediting bind conflict resolution.
165 struct inet_bind_hashbucket *bhash2;
166 unsigned int bhash_size;
168 /* The 2nd listener table hashed by local port and address */
169 unsigned int lhash2_mask;
170 struct inet_listen_hashbucket *lhash2;
173 static inline struct inet_hashinfo *tcp_or_dccp_get_hashinfo(const struct sock *sk)
175 #if IS_ENABLED(CONFIG_IP_DCCP)
176 return sk->sk_prot->h.hashinfo ? :
177 sock_net(sk)->ipv4.tcp_death_row.hashinfo;
179 return sock_net(sk)->ipv4.tcp_death_row.hashinfo;
183 static inline struct inet_listen_hashbucket *
184 inet_lhash2_bucket(struct inet_hashinfo *h, u32 hash)
186 return &h->lhash2[hash & h->lhash2_mask];
189 static inline struct inet_ehash_bucket *inet_ehash_bucket(
190 struct inet_hashinfo *hashinfo,
193 return &hashinfo->ehash[hash & hashinfo->ehash_mask];
196 static inline spinlock_t *inet_ehash_lockp(
197 struct inet_hashinfo *hashinfo,
200 return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
203 int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo);
205 static inline void inet_hashinfo2_free_mod(struct inet_hashinfo *h)
211 static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
213 kvfree(hashinfo->ehash_locks);
214 hashinfo->ehash_locks = NULL;
217 struct inet_bind_bucket *
218 inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net,
219 struct inet_bind_hashbucket *head,
220 const unsigned short snum, int l3mdev);
221 void inet_bind_bucket_destroy(struct kmem_cache *cachep,
222 struct inet_bind_bucket *tb);
224 bool inet_bind_bucket_match(const struct inet_bind_bucket *tb,
225 const struct net *net, unsigned short port,
228 struct inet_bind2_bucket *
229 inet_bind2_bucket_create(struct kmem_cache *cachep, struct net *net,
230 struct inet_bind_hashbucket *head,
231 unsigned short port, int l3mdev,
232 const struct sock *sk);
234 void inet_bind2_bucket_destroy(struct kmem_cache *cachep,
235 struct inet_bind2_bucket *tb);
237 struct inet_bind2_bucket *
238 inet_bind2_bucket_find(const struct inet_bind_hashbucket *head,
239 const struct net *net,
240 unsigned short port, int l3mdev,
241 const struct sock *sk);
243 bool inet_bind2_bucket_match_addr_any(const struct inet_bind2_bucket *tb,
244 const struct net *net, unsigned short port,
245 int l3mdev, const struct sock *sk);
247 static inline u32 inet_bhashfn(const struct net *net, const __u16 lport,
248 const u32 bhash_size)
250 return (lport + net_hash_mix(net)) & (bhash_size - 1);
253 static inline struct inet_bind_hashbucket *
254 inet_bhashfn_portaddr(const struct inet_hashinfo *hinfo, const struct sock *sk,
255 const struct net *net, unsigned short port)
259 #if IS_ENABLED(CONFIG_IPV6)
260 if (sk->sk_family == AF_INET6)
261 hash = ipv6_portaddr_hash(net, &sk->sk_v6_rcv_saddr, port);
264 hash = ipv4_portaddr_hash(net, sk->sk_rcv_saddr, port);
265 return &hinfo->bhash2[hash & (hinfo->bhash_size - 1)];
268 struct inet_bind_hashbucket *
269 inet_bhash2_addr_any_hashbucket(const struct sock *sk, const struct net *net, int port);
271 /* This should be called whenever a socket's sk_rcv_saddr (ipv4) or
272 * sk_v6_rcv_saddr (ipv6) changes after it has been binded. The socket's
273 * rcv_saddr field should already have been updated when this is called.
275 int inet_bhash2_update_saddr(struct inet_bind_hashbucket *prev_saddr, struct sock *sk);
277 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
278 struct inet_bind2_bucket *tb2, unsigned short port);
280 /* Caller must disable local BH processing. */
281 int __inet_inherit_port(const struct sock *sk, struct sock *child);
283 void inet_put_port(struct sock *sk);
285 void inet_hashinfo2_init(struct inet_hashinfo *h, const char *name,
286 unsigned long numentries, int scale,
287 unsigned long low_limit,
288 unsigned long high_limit);
289 int inet_hashinfo2_init_mod(struct inet_hashinfo *h);
291 bool inet_ehash_insert(struct sock *sk, struct sock *osk, bool *found_dup_sk);
292 bool inet_ehash_nolisten(struct sock *sk, struct sock *osk,
294 int __inet_hash(struct sock *sk, struct sock *osk);
295 int inet_hash(struct sock *sk);
296 void inet_unhash(struct sock *sk);
298 struct sock *__inet_lookup_listener(struct net *net,
299 struct inet_hashinfo *hashinfo,
300 struct sk_buff *skb, int doff,
301 const __be32 saddr, const __be16 sport,
303 const unsigned short hnum,
304 const int dif, const int sdif);
306 static inline struct sock *inet_lookup_listener(struct net *net,
307 struct inet_hashinfo *hashinfo,
308 struct sk_buff *skb, int doff,
309 __be32 saddr, __be16 sport,
310 __be32 daddr, __be16 dport, int dif, int sdif)
312 return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, sport,
313 daddr, ntohs(dport), dif, sdif);
316 /* Socket demux engine toys. */
317 /* What happens here is ugly; there's a pair of adjacent fields in
318 struct inet_sock; __be16 dport followed by __u16 num. We want to
319 search by pair, so we combine the keys into a single 32bit value
320 and compare with 32bit value read from &...->dport. Let's at least
321 make sure that it's not mixed with anything else...
322 On 64bit targets we combine comparisons with pair of adjacent __be32
323 fields in the same way.
326 #define INET_COMBINED_PORTS(__sport, __dport) \
327 ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
328 #else /* __LITTLE_ENDIAN */
329 #define INET_COMBINED_PORTS(__sport, __dport) \
330 ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
334 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
335 const __addrpair __name = (__force __addrpair) ( \
336 (((__force __u64)(__be32)(__saddr)) << 32) | \
337 ((__force __u64)(__be32)(__daddr)))
338 #else /* __LITTLE_ENDIAN */
339 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
340 const __addrpair __name = (__force __addrpair) ( \
341 (((__force __u64)(__be32)(__daddr)) << 32) | \
342 ((__force __u64)(__be32)(__saddr)))
343 #endif /* __BIG_ENDIAN */
345 static inline bool inet_match(struct net *net, const struct sock *sk,
346 const __addrpair cookie, const __portpair ports,
349 if (!net_eq(sock_net(sk), net) ||
350 sk->sk_portpair != ports ||
351 sk->sk_addrpair != cookie)
354 /* READ_ONCE() paired with WRITE_ONCE() in sock_bindtoindex_locked() */
355 return inet_sk_bound_dev_eq(net, READ_ONCE(sk->sk_bound_dev_if), dif,
359 /* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
360 * not check it for lookups anymore, thanks Alexey. -DaveM
362 struct sock *__inet_lookup_established(struct net *net,
363 struct inet_hashinfo *hashinfo,
364 const __be32 saddr, const __be16 sport,
365 const __be32 daddr, const u16 hnum,
366 const int dif, const int sdif);
368 static inline struct sock *
369 inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
370 const __be32 saddr, const __be16 sport,
371 const __be32 daddr, const __be16 dport,
374 return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
375 ntohs(dport), dif, 0);
378 static inline struct sock *__inet_lookup(struct net *net,
379 struct inet_hashinfo *hashinfo,
380 struct sk_buff *skb, int doff,
381 const __be32 saddr, const __be16 sport,
382 const __be32 daddr, const __be16 dport,
383 const int dif, const int sdif,
386 u16 hnum = ntohs(dport);
389 sk = __inet_lookup_established(net, hashinfo, saddr, sport,
390 daddr, hnum, dif, sdif);
395 return __inet_lookup_listener(net, hashinfo, skb, doff, saddr,
396 sport, daddr, hnum, dif, sdif);
399 static inline struct sock *inet_lookup(struct net *net,
400 struct inet_hashinfo *hashinfo,
401 struct sk_buff *skb, int doff,
402 const __be32 saddr, const __be16 sport,
403 const __be32 daddr, const __be16 dport,
409 sk = __inet_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
410 dport, dif, 0, &refcounted);
412 if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
417 static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
425 struct sock *sk = skb_steal_sock(skb, refcounted);
426 const struct iphdr *iph = ip_hdr(skb);
431 return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo, skb,
432 doff, iph->saddr, sport,
433 iph->daddr, dport, inet_iif(skb), sdif,
437 u32 inet6_ehashfn(const struct net *net,
438 const struct in6_addr *laddr, const u16 lport,
439 const struct in6_addr *faddr, const __be16 fport);
441 static inline void sk_daddr_set(struct sock *sk, __be32 addr)
443 sk->sk_daddr = addr; /* alias of inet_daddr */
444 #if IS_ENABLED(CONFIG_IPV6)
445 ipv6_addr_set_v4mapped(addr, &sk->sk_v6_daddr);
449 static inline void sk_rcv_saddr_set(struct sock *sk, __be32 addr)
451 sk->sk_rcv_saddr = addr; /* alias of inet_rcv_saddr */
452 #if IS_ENABLED(CONFIG_IPV6)
453 ipv6_addr_set_v4mapped(addr, &sk->sk_v6_rcv_saddr);
457 int __inet_hash_connect(struct inet_timewait_death_row *death_row,
458 struct sock *sk, u64 port_offset,
459 int (*check_established)(struct inet_timewait_death_row *,
460 struct sock *, __u16,
461 struct inet_timewait_sock **));
463 int inet_hash_connect(struct inet_timewait_death_row *death_row,
465 #endif /* _INET_HASHTABLES_H */