Linux Audio

Check our new training course

Loading...
v3.1
 
  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 <linux/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 */
 43struct 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 */
 
 
 
 79struct inet_bind_bucket {
 80#ifdef CONFIG_NET_NS
 81	struct net		*ib_net;
 
 
 
 
 
 
 82#endif
 
 
 
 
 
 
 
 
 
 
 83	unsigned short		port;
 84	signed short		fastreuse;
 85	int			num_owners;
 
 
 
 
 
 
 86	struct hlist_node	node;
 
 
 87	struct hlist_head	owners;
 88};
 89
 90static inline struct net *ib_net(struct inet_bind_bucket *ib)
 
 
 
 
 
 91{
 92	return read_pnet(&ib->ib_net);
 93}
 94
 95#define inet_bind_bucket_for_each(tb, pos, head) \
 96	hlist_for_each_entry(tb, pos, head, node)
 97
 98struct inet_bind_hashbucket {
 99	spinlock_t		lock;
100	struct hlist_head	chain;
101};
102
103/*
104 * Sockets can be hashed in established or listening table
105 * We must use different 'nulls' end-of-chain value for listening
106 * hash table, or we might find a socket that was closed and
107 * reallocated/inserted into established hash table
108 */
109#define LISTENING_NULLS_BASE (1U << 29)
110struct inet_listen_hashbucket {
111	spinlock_t		lock;
112	struct hlist_nulls_head	head;
113};
114
115/* This is for listening sockets, thus all sockets which possess wildcards. */
116#define INET_LHTABLE_SIZE	32	/* Yes, really, this is all you need. */
117
118struct inet_hashinfo {
119	/* This is for sockets with full identity only.  Sockets here will
120	 * always be without wildcards and will have the following invariant:
121	 *
122	 *          TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
123	 *
124	 * TIME_WAIT sockets use a separate chain (twchain).
125	 */
126	struct inet_ehash_bucket	*ehash;
127	spinlock_t			*ehash_locks;
128	unsigned int			ehash_mask;
129	unsigned int			ehash_locks_mask;
130
131	/* Ok, let's try this, I give up, we do need a local binding
132	 * TCP hash as well as the others for fast bind/connect.
133	 */
 
 
134	struct inet_bind_hashbucket	*bhash;
135
 
 
 
 
 
136	unsigned int			bhash_size;
137	/* 4 bytes hole on 64 bit */
138
139	struct kmem_cache		*bind_bucket_cachep;
140
141	/* All the above members are written once at bootup and
142	 * never written again _or_ are predominantly read-access.
143	 *
144	 * Now align to a new cache line as all the following members
145	 * might be often dirty.
146	 */
147	/* All sockets in TCP_LISTEN state will be in here.  This is the only
148	 * table where wildcard'd TCP sockets can exist.  Hash function here
149	 * is just local port number.
150	 */
151	struct inet_listen_hashbucket	listening_hash[INET_LHTABLE_SIZE]
152					____cacheline_aligned_in_smp;
 
 
 
 
153
154	atomic_t			bsockets;
155};
 
 
 
156
157static inline struct inet_ehash_bucket *inet_ehash_bucket(
158	struct inet_hashinfo *hashinfo,
159	unsigned int hash)
160{
161	return &hashinfo->ehash[hash & hashinfo->ehash_mask];
162}
163
164static inline spinlock_t *inet_ehash_lockp(
165	struct inet_hashinfo *hashinfo,
166	unsigned int hash)
167{
168	return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
169}
170
171static inline int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo)
 
 
172{
173	unsigned int i, size = 256;
174#if defined(CONFIG_PROVE_LOCKING)
175	unsigned int nr_pcpus = 2;
176#else
177	unsigned int nr_pcpus = num_possible_cpus();
178#endif
179	if (nr_pcpus >= 4)
180		size = 512;
181	if (nr_pcpus >= 8)
182		size = 1024;
183	if (nr_pcpus >= 16)
184		size = 2048;
185	if (nr_pcpus >= 32)
186		size = 4096;
187	if (sizeof(spinlock_t) != 0) {
188#ifdef CONFIG_NUMA
189		if (size * sizeof(spinlock_t) > PAGE_SIZE)
190			hashinfo->ehash_locks = vmalloc(size * sizeof(spinlock_t));
191		else
192#endif
193		hashinfo->ehash_locks =	kmalloc(size * sizeof(spinlock_t),
194						GFP_KERNEL);
195		if (!hashinfo->ehash_locks)
196			return ENOMEM;
197		for (i = 0; i < size; i++)
198			spin_lock_init(&hashinfo->ehash_locks[i]);
199	}
200	hashinfo->ehash_locks_mask = size - 1;
201	return 0;
202}
203
204static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
205{
206	if (hashinfo->ehash_locks) {
207#ifdef CONFIG_NUMA
208		unsigned int size = (hashinfo->ehash_locks_mask + 1) *
209							sizeof(spinlock_t);
210		if (size > PAGE_SIZE)
211			vfree(hashinfo->ehash_locks);
212		else
213#endif
214		kfree(hashinfo->ehash_locks);
215		hashinfo->ehash_locks = NULL;
216	}
217}
218
219extern struct inet_bind_bucket *
220		    inet_bind_bucket_create(struct kmem_cache *cachep,
221					    struct net *net,
222					    struct inet_bind_hashbucket *head,
223					    const unsigned short snum);
224extern void inet_bind_bucket_destroy(struct kmem_cache *cachep,
225				     struct inet_bind_bucket *tb);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
226
227static inline int inet_bhashfn(struct net *net,
228		const __u16 lport, const int bhash_size)
229{
230	return (lport + net_hash_mix(net)) & (bhash_size - 1);
231}
232
233extern void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
234			   const unsigned short snum);
235
236/* These can have wildcards, don't try too hard. */
237static inline int inet_lhashfn(struct net *net, const unsigned short num)
238{
239	return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1);
 
 
 
 
 
 
240}
241
242static inline int inet_sk_listen_hashfn(const struct sock *sk)
243{
244	return inet_lhashfn(sock_net(sk), inet_sk(sk)->inet_num);
245}
246
247/* Caller must disable local BH processing. */
248extern int __inet_inherit_port(struct sock *sk, struct sock *child);
 
 
 
 
249
250extern void inet_put_port(struct sock *sk);
 
251
252void inet_hashinfo_init(struct inet_hashinfo *h);
 
253
254extern int __inet_hash_nolisten(struct sock *sk, struct inet_timewait_sock *tw);
255extern void inet_hash(struct sock *sk);
256extern void inet_unhash(struct sock *sk);
257
258extern struct sock *__inet_lookup_listener(struct net *net,
259					   struct inet_hashinfo *hashinfo,
260					   const __be32 daddr,
261					   const unsigned short hnum,
262					   const int dif);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
263
264static inline struct sock *inet_lookup_listener(struct net *net,
265		struct inet_hashinfo *hashinfo,
266		__be32 daddr, __be16 dport, int dif)
 
 
267{
268	return __inet_lookup_listener(net, hashinfo, daddr, ntohs(dport), dif);
 
269}
270
271/* Socket demux engine toys. */
272/* What happens here is ugly; there's a pair of adjacent fields in
273   struct inet_sock; __be16 dport followed by __u16 num.  We want to
274   search by pair, so we combine the keys into a single 32bit value
275   and compare with 32bit value read from &...->dport.  Let's at least
276   make sure that it's not mixed with anything else...
277   On 64bit targets we combine comparisons with pair of adjacent __be32
278   fields in the same way.
279*/
280typedef __u32 __bitwise __portpair;
281#ifdef __BIG_ENDIAN
282#define INET_COMBINED_PORTS(__sport, __dport) \
283	((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
284#else /* __LITTLE_ENDIAN */
285#define INET_COMBINED_PORTS(__sport, __dport) \
286	((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
287#endif
288
289#if (BITS_PER_LONG == 64)
290typedef __u64 __bitwise __addrpair;
291#ifdef __BIG_ENDIAN
292#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
293	const __addrpair __name = (__force __addrpair) ( \
294				   (((__force __u64)(__be32)(__saddr)) << 32) | \
295				   ((__force __u64)(__be32)(__daddr)));
296#else /* __LITTLE_ENDIAN */
297#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
298	const __addrpair __name = (__force __addrpair) ( \
299				   (((__force __u64)(__be32)(__daddr)) << 32) | \
300				   ((__force __u64)(__be32)(__saddr)));
301#endif /* __BIG_ENDIAN */
302#define INET_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
303	(((__sk)->sk_hash == (__hash)) && net_eq(sock_net(__sk), (__net)) &&	\
304	 ((*((__addrpair *)&(inet_sk(__sk)->inet_daddr))) == (__cookie))  &&	\
305	 ((*((__portpair *)&(inet_sk(__sk)->inet_dport))) == (__ports))   &&	\
306	 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
307#define INET_TW_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
308	(((__sk)->sk_hash == (__hash)) && net_eq(sock_net(__sk), (__net)) &&	\
309	 ((*((__addrpair *)&(inet_twsk(__sk)->tw_daddr))) == (__cookie)) &&	\
310	 ((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) &&	\
311	 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
312#else /* 32-bit arch */
313#define INET_ADDR_COOKIE(__name, __saddr, __daddr)
314#define INET_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif)	\
315	(((__sk)->sk_hash == (__hash)) && net_eq(sock_net(__sk), (__net))	&&	\
316	 (inet_sk(__sk)->inet_daddr	== (__saddr))		&&	\
317	 (inet_sk(__sk)->inet_rcv_saddr	== (__daddr))		&&	\
318	 ((*((__portpair *)&(inet_sk(__sk)->inet_dport))) == (__ports))	&&	\
319	 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
320#define INET_TW_MATCH(__sk, __net, __hash,__cookie, __saddr, __daddr, __ports, __dif)	\
321	(((__sk)->sk_hash == (__hash)) && net_eq(sock_net(__sk), (__net))	&&	\
322	 (inet_twsk(__sk)->tw_daddr	== (__saddr))		&&	\
323	 (inet_twsk(__sk)->tw_rcv_saddr	== (__daddr))		&&	\
324	 ((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) &&	\
325	 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
326#endif /* 64-bit arch */
327
328/*
329 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
 
 
 
 
 
 
 
 
 
 
 
 
 
330 * not check it for lookups anymore, thanks Alexey. -DaveM
331 *
332 * Local BH must be disabled here.
333 */
334extern struct sock * __inet_lookup_established(struct net *net,
335		struct inet_hashinfo *hashinfo,
336		const __be32 saddr, const __be16 sport,
337		const __be32 daddr, const u16 hnum, const int dif);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
338
339static inline struct sock *
340	inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
341				const __be32 saddr, const __be16 sport,
342				const __be32 daddr, const __be16 dport,
343				const int dif)
344{
345	return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
346					 ntohs(dport), dif);
347}
348
349static inline struct sock *__inet_lookup(struct net *net,
350					 struct inet_hashinfo *hashinfo,
 
351					 const __be32 saddr, const __be16 sport,
352					 const __be32 daddr, const __be16 dport,
353					 const int dif)
 
354{
355	u16 hnum = ntohs(dport);
356	struct sock *sk = __inet_lookup_established(net, hashinfo,
357				saddr, sport, daddr, hnum, dif);
358
359	return sk ? : __inet_lookup_listener(net, hashinfo, daddr, hnum, dif);
 
 
 
 
 
 
 
360}
361
362static inline struct sock *inet_lookup(struct net *net,
363				       struct inet_hashinfo *hashinfo,
 
364				       const __be32 saddr, const __be16 sport,
365				       const __be32 daddr, const __be16 dport,
366				       const int dif)
367{
368	struct sock *sk;
 
369
370	local_bh_disable();
371	sk = __inet_lookup(net, hashinfo, saddr, sport, daddr, dport, dif);
372	local_bh_enable();
373
 
 
374	return sk;
375}
376
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
377static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
378					     struct sk_buff *skb,
 
379					     const __be16 sport,
380					     const __be16 dport)
 
 
381{
382	struct sock *sk;
383	const struct iphdr *iph = ip_hdr(skb);
 
384
385	if (unlikely(sk = skb_steal_sock(skb)))
 
 
 
 
386		return sk;
387	else
388		return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo,
389				     iph->saddr, sport,
390				     iph->daddr, dport, inet_iif(skb));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
391}
392
393extern int __inet_hash_connect(struct inet_timewait_death_row *death_row,
394		struct sock *sk,
395		u32 port_offset,
396		int (*check_established)(struct inet_timewait_death_row *,
397			struct sock *, __u16, struct inet_timewait_sock **),
398		int (*hash)(struct sock *sk, struct inet_timewait_sock *twp));
399
400extern int inet_hash_connect(struct inet_timewait_death_row *death_row,
401			     struct sock *sk);
402#endif /* _INET_HASHTABLES_H */
v6.13.7
  1/* SPDX-License-Identifier: GPL-2.0-or-later */
  2/*
  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.
  6 *
  7 * Authors:	Lotsa people, from code originally in tcp
 
 
 
 
 
  8 */
  9
 10#ifndef _INET_HASHTABLES_H
 11#define _INET_HASHTABLES_H
 12
 13
 14#include <linux/interrupt.h>
 15#include <linux/ip.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>
 
 23
 24#include <net/inet_connection_sock.h>
 25#include <net/inet_sock.h>
 26#include <net/ip.h>
 27#include <net/sock.h>
 28#include <net/route.h>
 29#include <net/tcp_states.h>
 30#include <net/netns/hash.h>
 31
 32#include <linux/refcount.h>
 33#include <asm/byteorder.h>
 34
 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
 37 * but LISTEN ones.
 38 */
 39struct inet_ehash_bucket {
 40	struct hlist_nulls_head chain;
 
 41};
 42
 43/* There are a few simple rules, which allow for local port reuse by
 44 * an application.  In essence:
 45 *
 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
 53 *	   shared.
 54 *	   Failing this, the port cannot be shared.
 55 *
 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.
 66 *
 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
 73 */
 74#define FASTREUSEPORT_ANY	1
 75#define FASTREUSEPORT_STRICT	2
 76
 77struct inet_bind_bucket {
 78	possible_net_t		ib_net;
 79	int			l3mdev;
 80	unsigned short		port;
 81	signed char		fastreuse;
 82	signed char		fastreuseport;
 83	kuid_t			fastuid;
 84#if IS_ENABLED(CONFIG_IPV6)
 85	struct in6_addr		fast_v6_rcv_saddr;
 86#endif
 87	__be32			fast_rcv_saddr;
 88	unsigned short		fast_sk_family;
 89	bool			fast_ipv6_only;
 90	struct hlist_node	node;
 91	struct hlist_head	bhash2;
 92};
 93
 94struct inet_bind2_bucket {
 95	possible_net_t		ib_net;
 96	int			l3mdev;
 97	unsigned short		port;
 98#if IS_ENABLED(CONFIG_IPV6)
 99	unsigned short		addr_type;
100	struct in6_addr		v6_rcv_saddr;
101#define rcv_saddr		v6_rcv_saddr.s6_addr32[3]
102#else
103	__be32			rcv_saddr;
104#endif
105	/* Node in the bhash2 inet_bind_hashbucket chain */
106	struct hlist_node	node;
107	struct hlist_node	bhash_node;
108	/* List of sockets hashed to this bucket */
109	struct hlist_head	owners;
110};
111
112static inline struct net *ib_net(const struct inet_bind_bucket *ib)
113{
114	return read_pnet(&ib->ib_net);
115}
116
117static inline struct net *ib2_net(const struct inet_bind2_bucket *ib)
118{
119	return read_pnet(&ib->ib_net);
120}
121
122#define inet_bind_bucket_for_each(tb, head) \
123	hlist_for_each_entry(tb, head, node)
124
125struct inet_bind_hashbucket {
126	spinlock_t		lock;
127	struct hlist_head	chain;
128};
129
130/* Sockets can be hashed in established or listening table.
131 * We must use different 'nulls' end-of-chain value for all hash buckets :
132 * A socket might transition from ESTABLISH to LISTEN state without
133 * RCU grace period. A lookup in ehash table needs to handle this case.
 
134 */
135#define LISTENING_NULLS_BASE (1U << 29)
136struct inet_listen_hashbucket {
137	spinlock_t		lock;
138	struct hlist_nulls_head	nulls_head;
139};
140
141/* This is for listening sockets, thus all sockets which possess wildcards. */
142#define INET_LHTABLE_SIZE	32	/* Yes, really, this is all you need. */
143
144struct inet_hashinfo {
145	/* This is for sockets with full identity only.  Sockets here will
146	 * always be without wildcards and will have the following invariant:
147	 *
148	 *          TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
149	 *
 
150	 */
151	struct inet_ehash_bucket	*ehash;
152	spinlock_t			*ehash_locks;
153	unsigned int			ehash_mask;
154	unsigned int			ehash_locks_mask;
155
156	/* Ok, let's try this, I give up, we do need a local binding
157	 * TCP hash as well as the others for fast bind/connect.
158	 */
159	struct kmem_cache		*bind_bucket_cachep;
160	/* This bind table is hashed by local port */
161	struct inet_bind_hashbucket	*bhash;
162	struct kmem_cache		*bind2_bucket_cachep;
163	/* This bind table is hashed by local port and sk->sk_rcv_saddr (ipv4)
164	 * or sk->sk_v6_rcv_saddr (ipv6). This 2nd bind table is used
165	 * primarily for expediting bind conflict resolution.
166	 */
167	struct inet_bind_hashbucket	*bhash2;
168	unsigned int			bhash_size;
 
 
 
169
170	/* The 2nd listener table hashed by local port and address */
171	unsigned int			lhash2_mask;
172	struct inet_listen_hashbucket	*lhash2;
173
174	bool				pernet;
175} ____cacheline_aligned_in_smp;
176
177static inline struct inet_hashinfo *tcp_or_dccp_get_hashinfo(const struct sock *sk)
178{
179#if IS_ENABLED(CONFIG_IP_DCCP)
180	return sk->sk_prot->h.hashinfo ? :
181		sock_net(sk)->ipv4.tcp_death_row.hashinfo;
182#else
183	return sock_net(sk)->ipv4.tcp_death_row.hashinfo;
184#endif
185}
186
187static inline struct inet_listen_hashbucket *
188inet_lhash2_bucket(struct inet_hashinfo *h, u32 hash)
189{
190	return &h->lhash2[hash & h->lhash2_mask];
191}
192
193static inline struct inet_ehash_bucket *inet_ehash_bucket(
194	struct inet_hashinfo *hashinfo,
195	unsigned int hash)
196{
197	return &hashinfo->ehash[hash & hashinfo->ehash_mask];
198}
199
200static inline spinlock_t *inet_ehash_lockp(
201	struct inet_hashinfo *hashinfo,
202	unsigned int hash)
203{
204	return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
205}
206
207int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo);
208
209static inline void inet_hashinfo2_free_mod(struct inet_hashinfo *h)
210{
211	kfree(h->lhash2);
212	h->lhash2 = NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
213}
214
215static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
216{
217	kvfree(hashinfo->ehash_locks);
218	hashinfo->ehash_locks = NULL;
 
 
 
 
 
 
 
 
 
219}
220
221struct inet_hashinfo *inet_pernet_hashinfo_alloc(struct inet_hashinfo *hashinfo,
222						 unsigned int ehash_entries);
223void inet_pernet_hashinfo_free(struct inet_hashinfo *hashinfo);
224
225struct inet_bind_bucket *
226inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net,
227			struct inet_bind_hashbucket *head,
228			const unsigned short snum, int l3mdev);
229void inet_bind_bucket_destroy(struct kmem_cache *cachep,
230			      struct inet_bind_bucket *tb);
231
232bool inet_bind_bucket_match(const struct inet_bind_bucket *tb,
233			    const struct net *net, unsigned short port,
234			    int l3mdev);
235
236struct inet_bind2_bucket *
237inet_bind2_bucket_create(struct kmem_cache *cachep, struct net *net,
238			 struct inet_bind_hashbucket *head,
239			 struct inet_bind_bucket *tb,
240			 const struct sock *sk);
241
242void inet_bind2_bucket_destroy(struct kmem_cache *cachep,
243			       struct inet_bind2_bucket *tb);
244
245struct inet_bind2_bucket *
246inet_bind2_bucket_find(const struct inet_bind_hashbucket *head,
247		       const struct net *net,
248		       unsigned short port, int l3mdev,
249		       const struct sock *sk);
250
251bool inet_bind2_bucket_match_addr_any(const struct inet_bind2_bucket *tb,
252				      const struct net *net, unsigned short port,
253				      int l3mdev, const struct sock *sk);
254
255static inline u32 inet_bhashfn(const struct net *net, const __u16 lport,
256			       const u32 bhash_size)
257{
258	return (lport + net_hash_mix(net)) & (bhash_size - 1);
259}
260
261static inline struct inet_bind_hashbucket *
262inet_bhashfn_portaddr(const struct inet_hashinfo *hinfo, const struct sock *sk,
263		      const struct net *net, unsigned short port)
264{
265	u32 hash;
266
267#if IS_ENABLED(CONFIG_IPV6)
268	if (sk->sk_family == AF_INET6)
269		hash = ipv6_portaddr_hash(net, &sk->sk_v6_rcv_saddr, port);
270	else
271#endif
272		hash = ipv4_portaddr_hash(net, sk->sk_rcv_saddr, port);
273	return &hinfo->bhash2[hash & (hinfo->bhash_size - 1)];
274}
275
276struct inet_bind_hashbucket *
277inet_bhash2_addr_any_hashbucket(const struct sock *sk, const struct net *net, int port);
 
 
278
279/* This should be called whenever a socket's sk_rcv_saddr (ipv4) or
280 * sk_v6_rcv_saddr (ipv6) changes after it has been binded. The socket's
281 * rcv_saddr field should already have been updated when this is called.
282 */
283int inet_bhash2_update_saddr(struct sock *sk, void *saddr, int family);
284void inet_bhash2_reset_saddr(struct sock *sk);
285
286void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
287		    struct inet_bind2_bucket *tb2, unsigned short port);
288
289/* Caller must disable local BH processing. */
290int __inet_inherit_port(const struct sock *sk, struct sock *child);
291
292void inet_put_port(struct sock *sk);
 
 
293
294void inet_hashinfo2_init(struct inet_hashinfo *h, const char *name,
295			 unsigned long numentries, int scale,
296			 unsigned long low_limit,
297			 unsigned long high_limit);
298int inet_hashinfo2_init_mod(struct inet_hashinfo *h);
299
300bool inet_ehash_insert(struct sock *sk, struct sock *osk, bool *found_dup_sk);
301bool inet_ehash_nolisten(struct sock *sk, struct sock *osk,
302			 bool *found_dup_sk);
303int __inet_hash(struct sock *sk, struct sock *osk);
304int inet_hash(struct sock *sk);
305void inet_unhash(struct sock *sk);
306
307struct sock *__inet_lookup_listener(const struct net *net,
308				    struct inet_hashinfo *hashinfo,
309				    struct sk_buff *skb, int doff,
310				    const __be32 saddr, const __be16 sport,
311				    const __be32 daddr,
312				    const unsigned short hnum,
313				    const int dif, const int sdif);
314
315static inline struct sock *inet_lookup_listener(struct net *net,
316		struct inet_hashinfo *hashinfo,
317		struct sk_buff *skb, int doff,
318		__be32 saddr, __be16 sport,
319		__be32 daddr, __be16 dport, int dif, int sdif)
320{
321	return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, sport,
322				      daddr, ntohs(dport), dif, sdif);
323}
324
325/* Socket demux engine toys. */
326/* What happens here is ugly; there's a pair of adjacent fields in
327   struct inet_sock; __be16 dport followed by __u16 num.  We want to
328   search by pair, so we combine the keys into a single 32bit value
329   and compare with 32bit value read from &...->dport.  Let's at least
330   make sure that it's not mixed with anything else...
331   On 64bit targets we combine comparisons with pair of adjacent __be32
332   fields in the same way.
333*/
 
334#ifdef __BIG_ENDIAN
335#define INET_COMBINED_PORTS(__sport, __dport) \
336	((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
337#else /* __LITTLE_ENDIAN */
338#define INET_COMBINED_PORTS(__sport, __dport) \
339	((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
340#endif
341
 
 
342#ifdef __BIG_ENDIAN
343#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
344	const __addrpair __name = (__force __addrpair) ( \
345				   (((__force __u64)(__be32)(__saddr)) << 32) | \
346				   ((__force __u64)(__be32)(__daddr)))
347#else /* __LITTLE_ENDIAN */
348#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
349	const __addrpair __name = (__force __addrpair) ( \
350				   (((__force __u64)(__be32)(__daddr)) << 32) | \
351				   ((__force __u64)(__be32)(__saddr)))
352#endif /* __BIG_ENDIAN */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
353
354static inline bool inet_match(const struct net *net, const struct sock *sk,
355			      const __addrpair cookie, const __portpair ports,
356			      int dif, int sdif)
357{
358	if (!net_eq(sock_net(sk), net) ||
359	    sk->sk_portpair != ports ||
360	    sk->sk_addrpair != cookie)
361	        return false;
362
363	/* READ_ONCE() paired with WRITE_ONCE() in sock_bindtoindex_locked() */
364	return inet_sk_bound_dev_eq(net, READ_ONCE(sk->sk_bound_dev_if), dif,
365				    sdif);
366}
367
368/* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
369 * not check it for lookups anymore, thanks Alexey. -DaveM
 
 
370 */
371struct sock *__inet_lookup_established(const struct net *net,
372				       struct inet_hashinfo *hashinfo,
373				       const __be32 saddr, const __be16 sport,
374				       const __be32 daddr, const u16 hnum,
375				       const int dif, const int sdif);
376
377typedef u32 (inet_ehashfn_t)(const struct net *net,
378			      const __be32 laddr, const __u16 lport,
379			      const __be32 faddr, const __be16 fport);
380
381inet_ehashfn_t inet_ehashfn;
382
383INDIRECT_CALLABLE_DECLARE(inet_ehashfn_t udp_ehashfn);
384
385struct sock *inet_lookup_reuseport(const struct net *net, struct sock *sk,
386				   struct sk_buff *skb, int doff,
387				   __be32 saddr, __be16 sport,
388				   __be32 daddr, unsigned short hnum,
389				   inet_ehashfn_t *ehashfn);
390
391struct sock *inet_lookup_run_sk_lookup(const struct net *net,
392				       int protocol,
393				       struct sk_buff *skb, int doff,
394				       __be32 saddr, __be16 sport,
395				       __be32 daddr, u16 hnum, const int dif,
396				       inet_ehashfn_t *ehashfn);
397
398static inline struct sock *
399	inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
400				const __be32 saddr, const __be16 sport,
401				const __be32 daddr, const __be16 dport,
402				const int dif)
403{
404	return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
405					 ntohs(dport), dif, 0);
406}
407
408static inline struct sock *__inet_lookup(struct net *net,
409					 struct inet_hashinfo *hashinfo,
410					 struct sk_buff *skb, int doff,
411					 const __be32 saddr, const __be16 sport,
412					 const __be32 daddr, const __be16 dport,
413					 const int dif, const int sdif,
414					 bool *refcounted)
415{
416	u16 hnum = ntohs(dport);
417	struct sock *sk;
 
418
419	sk = __inet_lookup_established(net, hashinfo, saddr, sport,
420				       daddr, hnum, dif, sdif);
421	*refcounted = true;
422	if (sk)
423		return sk;
424	*refcounted = false;
425	return __inet_lookup_listener(net, hashinfo, skb, doff, saddr,
426				      sport, daddr, hnum, dif, sdif);
427}
428
429static inline struct sock *inet_lookup(struct net *net,
430				       struct inet_hashinfo *hashinfo,
431				       struct sk_buff *skb, int doff,
432				       const __be32 saddr, const __be16 sport,
433				       const __be32 daddr, const __be16 dport,
434				       const int dif)
435{
436	struct sock *sk;
437	bool refcounted;
438
439	sk = __inet_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
440			   dport, dif, 0, &refcounted);
 
441
442	if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
443		sk = NULL;
444	return sk;
445}
446
447static inline
448struct sock *inet_steal_sock(struct net *net, struct sk_buff *skb, int doff,
449			     const __be32 saddr, const __be16 sport,
450			     const __be32 daddr, const __be16 dport,
451			     bool *refcounted, inet_ehashfn_t *ehashfn)
452{
453	struct sock *sk, *reuse_sk;
454	bool prefetched;
455
456	sk = skb_steal_sock(skb, refcounted, &prefetched);
457	if (!sk)
458		return NULL;
459
460	if (!prefetched || !sk_fullsock(sk))
461		return sk;
462
463	if (sk->sk_protocol == IPPROTO_TCP) {
464		if (sk->sk_state != TCP_LISTEN)
465			return sk;
466	} else if (sk->sk_protocol == IPPROTO_UDP) {
467		if (sk->sk_state != TCP_CLOSE)
468			return sk;
469	} else {
470		return sk;
471	}
472
473	reuse_sk = inet_lookup_reuseport(net, sk, skb, doff,
474					 saddr, sport, daddr, ntohs(dport),
475					 ehashfn);
476	if (!reuse_sk)
477		return sk;
478
479	/* We've chosen a new reuseport sock which is never refcounted. This
480	 * implies that sk also isn't refcounted.
481	 */
482	WARN_ON_ONCE(*refcounted);
483
484	return reuse_sk;
485}
486
487static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
488					     struct sk_buff *skb,
489					     int doff,
490					     const __be16 sport,
491					     const __be16 dport,
492					     const int sdif,
493					     bool *refcounted)
494{
495	struct net *net = dev_net(skb_dst(skb)->dev);
496	const struct iphdr *iph = ip_hdr(skb);
497	struct sock *sk;
498
499	sk = inet_steal_sock(net, skb, doff, iph->saddr, sport, iph->daddr, dport,
500			     refcounted, inet_ehashfn);
501	if (IS_ERR(sk))
502		return NULL;
503	if (sk)
504		return sk;
505
506	return __inet_lookup(net, hashinfo, skb,
507			     doff, iph->saddr, sport,
508			     iph->daddr, dport, inet_iif(skb), sdif,
509			     refcounted);
510}
511
512static inline void sk_daddr_set(struct sock *sk, __be32 addr)
513{
514	sk->sk_daddr = addr; /* alias of inet_daddr */
515#if IS_ENABLED(CONFIG_IPV6)
516	ipv6_addr_set_v4mapped(addr, &sk->sk_v6_daddr);
517#endif
518}
519
520static inline void sk_rcv_saddr_set(struct sock *sk, __be32 addr)
521{
522	sk->sk_rcv_saddr = addr; /* alias of inet_rcv_saddr */
523#if IS_ENABLED(CONFIG_IPV6)
524	ipv6_addr_set_v4mapped(addr, &sk->sk_v6_rcv_saddr);
525#endif
526}
527
528int __inet_hash_connect(struct inet_timewait_death_row *death_row,
529			struct sock *sk, u64 port_offset,
530			int (*check_established)(struct inet_timewait_death_row *,
531						 struct sock *, __u16,
532						 struct inet_timewait_sock **));
 
533
534int inet_hash_connect(struct inet_timewait_death_row *death_row,
535		      struct sock *sk);
536#endif /* _INET_HASHTABLES_H */