Linux Audio

Check our new training course

Loading...
v5.4
  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/sock.h>
 27#include <net/route.h>
 28#include <net/tcp_states.h>
 29#include <net/netns/hash.h>
 30
 31#include <linux/refcount.h>
 32#include <asm/byteorder.h>
 33
 34/* This is for all connections with a full identity, no wildcards.
 35 * The 'e' prefix stands for Establish, but we really put all sockets
 36 * but LISTEN ones.
 37 */
 38struct inet_ehash_bucket {
 39	struct hlist_nulls_head chain;
 40};
 41
 42/* There are a few simple rules, which allow for local port reuse by
 43 * an application.  In essence:
 44 *
 45 *	1) Sockets bound to different interfaces may share a local port.
 46 *	   Failing that, goto test 2.
 47 *	2) If all sockets have sk->sk_reuse set, and none of them are in
 48 *	   TCP_LISTEN state, the port may be shared.
 49 *	   Failing that, goto test 3.
 50 *	3) If all sockets are bound to a specific inet_sk(sk)->rcv_saddr local
 51 *	   address, and none of them are the same, the port may be
 52 *	   shared.
 53 *	   Failing this, the port cannot be shared.
 54 *
 55 * The interesting point, is test #2.  This is what an FTP server does
 56 * all day.  To optimize this case we use a specific flag bit defined
 57 * below.  As we add sockets to a bind bucket list, we perform a
 58 * check of: (newsk->sk_reuse && (newsk->sk_state != TCP_LISTEN))
 59 * As long as all sockets added to a bind bucket pass this test,
 60 * the flag bit will be set.
 61 * The resulting situation is that tcp_v[46]_verify_bind() can just check
 62 * for this flag bit, if it is set and the socket trying to bind has
 63 * sk->sk_reuse set, we don't even have to walk the owners list at all,
 64 * we return that it is ok to bind this socket to the requested local port.
 65 *
 66 * Sounds like a lot of work, but it is worth it.  In a more naive
 67 * implementation (ie. current FreeBSD etc.) the entire list of ports
 68 * must be walked for each data port opened by an ftp server.  Needless
 69 * to say, this does not scale at all.  With a couple thousand FTP
 70 * users logged onto your box, isn't it nice to know that new data
 71 * ports are created in O(1) time?  I thought so. ;-)	-DaveM
 72 */
 73#define FASTREUSEPORT_ANY	1
 74#define FASTREUSEPORT_STRICT	2
 75
 76struct inet_bind_bucket {
 77	possible_net_t		ib_net;
 78	int			l3mdev;
 79	unsigned short		port;
 80	signed char		fastreuse;
 81	signed char		fastreuseport;
 82	kuid_t			fastuid;
 83#if IS_ENABLED(CONFIG_IPV6)
 84	struct in6_addr		fast_v6_rcv_saddr;
 85#endif
 86	__be32			fast_rcv_saddr;
 87	unsigned short		fast_sk_family;
 88	bool			fast_ipv6_only;
 89	struct hlist_node	node;
 90	struct hlist_head	owners;
 91};
 92
 93static inline struct net *ib_net(struct inet_bind_bucket *ib)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 94{
 95	return read_pnet(&ib->ib_net);
 96}
 97
 98#define inet_bind_bucket_for_each(tb, head) \
 99	hlist_for_each_entry(tb, head, node)
100
101struct inet_bind_hashbucket {
102	spinlock_t		lock;
103	struct hlist_head	chain;
104};
105
106/*
107 * Sockets can be hashed in established or listening table
 
 
108 */
 
109struct inet_listen_hashbucket {
110	spinlock_t		lock;
111	unsigned int		count;
112	struct hlist_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	 */
125	struct inet_ehash_bucket	*ehash;
126	spinlock_t			*ehash_locks;
127	unsigned int			ehash_mask;
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 kmem_cache		*bind_bucket_cachep;
 
134	struct inet_bind_hashbucket	*bhash;
 
 
 
 
 
 
135	unsigned int			bhash_size;
136
137	/* The 2nd listener table hashed by local port and address */
138	unsigned int			lhash2_mask;
139	struct inet_listen_hashbucket	*lhash2;
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 listening_hash.
148	 * This is the only table where wildcard'd TCP sockets can
149	 * exist.  listening_hash is only hashed by local port number.
150	 * If lhash2 is initialized, the same socket will also be hashed
151	 * to lhash2 by port and address.
152	 */
153	struct inet_listen_hashbucket	listening_hash[INET_LHTABLE_SIZE]
154					____cacheline_aligned_in_smp;
155};
156
157#define inet_lhash2_for_each_icsk_rcu(__icsk, list) \
158	hlist_for_each_entry_rcu(__icsk, list, icsk_listen_portaddr_node)
 
 
 
 
 
 
 
159
160static inline struct inet_listen_hashbucket *
161inet_lhash2_bucket(struct inet_hashinfo *h, u32 hash)
162{
163	return &h->lhash2[hash & h->lhash2_mask];
164}
165
166static inline struct inet_ehash_bucket *inet_ehash_bucket(
167	struct inet_hashinfo *hashinfo,
168	unsigned int hash)
169{
170	return &hashinfo->ehash[hash & hashinfo->ehash_mask];
171}
172
173static inline spinlock_t *inet_ehash_lockp(
174	struct inet_hashinfo *hashinfo,
175	unsigned int hash)
176{
177	return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
178}
179
180int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo);
181
 
 
 
 
 
 
182static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
183{
184	kvfree(hashinfo->ehash_locks);
185	hashinfo->ehash_locks = NULL;
186}
187
188static inline bool inet_sk_bound_dev_eq(struct net *net, int bound_dev_if,
189					int dif, int sdif)
190{
191#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
192	return inet_bound_dev_eq(!!net->ipv4.sysctl_tcp_l3mdev_accept,
193				 bound_dev_if, dif, sdif);
194#else
195	return inet_bound_dev_eq(true, bound_dev_if, dif, sdif);
196#endif
197}
198
199struct inet_bind_bucket *
200inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net,
201			struct inet_bind_hashbucket *head,
202			const unsigned short snum, int l3mdev);
203void inet_bind_bucket_destroy(struct kmem_cache *cachep,
204			      struct inet_bind_bucket *tb);
205
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
206static inline u32 inet_bhashfn(const struct net *net, const __u16 lport,
207			       const u32 bhash_size)
208{
209	return (lport + net_hash_mix(net)) & (bhash_size - 1);
210}
211
212void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
213		    const unsigned short snum);
214
215/* These can have wildcards, don't try too hard. */
216static inline u32 inet_lhashfn(const struct net *net, const unsigned short num)
217{
218	return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1);
219}
220
221static inline int inet_sk_listen_hashfn(const struct sock *sk)
222{
223	return inet_lhashfn(sock_net(sk), inet_sk(sk)->inet_num);
 
 
 
 
224}
225
 
 
 
 
 
 
 
 
 
 
 
 
 
226/* Caller must disable local BH processing. */
227int __inet_inherit_port(const struct sock *sk, struct sock *child);
228
229void inet_put_port(struct sock *sk);
230
231void inet_hashinfo_init(struct inet_hashinfo *h);
232void inet_hashinfo2_init(struct inet_hashinfo *h, const char *name,
233			 unsigned long numentries, int scale,
234			 unsigned long low_limit,
235			 unsigned long high_limit);
236int inet_hashinfo2_init_mod(struct inet_hashinfo *h);
237
238bool inet_ehash_insert(struct sock *sk, struct sock *osk);
239bool inet_ehash_nolisten(struct sock *sk, struct sock *osk);
 
240int __inet_hash(struct sock *sk, struct sock *osk);
241int inet_hash(struct sock *sk);
242void inet_unhash(struct sock *sk);
243
244struct sock *__inet_lookup_listener(struct net *net,
245				    struct inet_hashinfo *hashinfo,
246				    struct sk_buff *skb, int doff,
247				    const __be32 saddr, const __be16 sport,
248				    const __be32 daddr,
249				    const unsigned short hnum,
250				    const int dif, const int sdif);
251
252static inline struct sock *inet_lookup_listener(struct net *net,
253		struct inet_hashinfo *hashinfo,
254		struct sk_buff *skb, int doff,
255		__be32 saddr, __be16 sport,
256		__be32 daddr, __be16 dport, int dif, int sdif)
257{
258	return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, sport,
259				      daddr, ntohs(dport), dif, sdif);
260}
261
262/* Socket demux engine toys. */
263/* What happens here is ugly; there's a pair of adjacent fields in
264   struct inet_sock; __be16 dport followed by __u16 num.  We want to
265   search by pair, so we combine the keys into a single 32bit value
266   and compare with 32bit value read from &...->dport.  Let's at least
267   make sure that it's not mixed with anything else...
268   On 64bit targets we combine comparisons with pair of adjacent __be32
269   fields in the same way.
270*/
271#ifdef __BIG_ENDIAN
272#define INET_COMBINED_PORTS(__sport, __dport) \
273	((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
274#else /* __LITTLE_ENDIAN */
275#define INET_COMBINED_PORTS(__sport, __dport) \
276	((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
277#endif
278
279#if (BITS_PER_LONG == 64)
280#ifdef __BIG_ENDIAN
281#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
282	const __addrpair __name = (__force __addrpair) ( \
283				   (((__force __u64)(__be32)(__saddr)) << 32) | \
284				   ((__force __u64)(__be32)(__daddr)))
285#else /* __LITTLE_ENDIAN */
286#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
287	const __addrpair __name = (__force __addrpair) ( \
288				   (((__force __u64)(__be32)(__daddr)) << 32) | \
289				   ((__force __u64)(__be32)(__saddr)))
290#endif /* __BIG_ENDIAN */
291#define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif, __sdif) \
292	(((__sk)->sk_portpair == (__ports))			&&	\
293	 ((__sk)->sk_addrpair == (__cookie))			&&	\
294	 (((__sk)->sk_bound_dev_if == (__dif))			||	\
295	  ((__sk)->sk_bound_dev_if == (__sdif)))		&&	\
296	 net_eq(sock_net(__sk), (__net)))
297#else /* 32-bit arch */
298#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
299	const int __name __deprecated __attribute__((unused))
300
301#define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif, __sdif) \
302	(((__sk)->sk_portpair == (__ports))		&&		\
303	 ((__sk)->sk_daddr	== (__saddr))		&&		\
304	 ((__sk)->sk_rcv_saddr	== (__daddr))		&&		\
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 */
 
 
 
 
 
309
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
312 */
313struct 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);
318
319static 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,
323				const int dif)
324{
325	return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
326					 ntohs(dport), dif, 0);
327}
328
329static 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,
335					 bool *refcounted)
336{
337	u16 hnum = ntohs(dport);
338	struct sock *sk;
339
340	sk = __inet_lookup_established(net, hashinfo, saddr, sport,
341				       daddr, hnum, dif, sdif);
342	*refcounted = true;
343	if (sk)
344		return sk;
345	*refcounted = false;
346	return __inet_lookup_listener(net, hashinfo, skb, doff, saddr,
347				      sport, daddr, hnum, dif, sdif);
348}
349
350static 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,
355				       const int dif)
356{
357	struct sock *sk;
358	bool refcounted;
359
360	sk = __inet_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
361			   dport, dif, 0, &refcounted);
362
363	if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
364		sk = NULL;
365	return sk;
366}
367
368static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
369					     struct sk_buff *skb,
370					     int doff,
371					     const __be16 sport,
372					     const __be16 dport,
373					     const int sdif,
374					     bool *refcounted)
375{
376	struct sock *sk = skb_steal_sock(skb);
377	const struct iphdr *iph = ip_hdr(skb);
378
379	*refcounted = true;
380	if (sk)
381		return sk;
382
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,
386			     refcounted);
387}
388
389u32 inet6_ehashfn(const struct net *net,
390		  const struct in6_addr *laddr, const u16 lport,
391		  const struct in6_addr *faddr, const __be16 fport);
392
393static inline void sk_daddr_set(struct sock *sk, __be32 addr)
394{
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);
398#endif
399}
400
401static inline void sk_rcv_saddr_set(struct sock *sk, __be32 addr)
402{
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);
406#endif
407}
408
409int __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 **));
414
415int inet_hash_connect(struct inet_timewait_death_row *death_row,
416		      struct sock *sk);
417#endif /* _INET_HASHTABLES_H */
v6.2
  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	owners;
 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		family;
100#endif
101	union {
102#if IS_ENABLED(CONFIG_IPV6)
103		struct in6_addr		v6_rcv_saddr;
104#endif
105		__be32			rcv_saddr;
106	};
107	/* Node in the bhash2 inet_bind_hashbucket chain */
108	struct hlist_node	node;
109	/* List of sockets hashed to this bucket */
110	struct hlist_head	owners;
111	/* bhash has twsk in owners, but bhash2 has twsk in
112	 * deathrow not to add a member in struct sock_common.
113	 */
114	struct hlist_head	deathrow;
115};
116
117static inline struct net *ib_net(const struct inet_bind_bucket *ib)
118{
119	return read_pnet(&ib->ib_net);
120}
121
122static inline struct net *ib2_net(const struct inet_bind2_bucket *ib)
123{
124	return read_pnet(&ib->ib_net);
125}
126
127#define inet_bind_bucket_for_each(tb, head) \
128	hlist_for_each_entry(tb, head, node)
129
130struct inet_bind_hashbucket {
131	spinlock_t		lock;
132	struct hlist_head	chain;
133};
134
135/* Sockets can be hashed in established or listening table.
136 * We must use different 'nulls' end-of-chain value for all hash buckets :
137 * A socket might transition from ESTABLISH to LISTEN state without
138 * RCU grace period. A lookup in ehash table needs to handle this case.
139 */
140#define LISTENING_NULLS_BASE (1U << 29)
141struct inet_listen_hashbucket {
142	spinlock_t		lock;
143	struct hlist_nulls_head	nulls_head;
 
144};
145
146/* This is for listening sockets, thus all sockets which possess wildcards. */
147#define INET_LHTABLE_SIZE	32	/* Yes, really, this is all you need. */
148
149struct inet_hashinfo {
150	/* This is for sockets with full identity only.  Sockets here will
151	 * always be without wildcards and will have the following invariant:
152	 *
153	 *          TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
154	 *
155	 */
156	struct inet_ehash_bucket	*ehash;
157	spinlock_t			*ehash_locks;
158	unsigned int			ehash_mask;
159	unsigned int			ehash_locks_mask;
160
161	/* Ok, let's try this, I give up, we do need a local binding
162	 * TCP hash as well as the others for fast bind/connect.
163	 */
164	struct kmem_cache		*bind_bucket_cachep;
165	/* This bind table is hashed by local port */
166	struct inet_bind_hashbucket	*bhash;
167	struct kmem_cache		*bind2_bucket_cachep;
168	/* This bind table is hashed by local port and sk->sk_rcv_saddr (ipv4)
169	 * or sk->sk_v6_rcv_saddr (ipv6). This 2nd bind table is used
170	 * primarily for expediting bind conflict resolution.
171	 */
172	struct inet_bind_hashbucket	*bhash2;
173	unsigned int			bhash_size;
174
175	/* The 2nd listener table hashed by local port and address */
176	unsigned int			lhash2_mask;
177	struct inet_listen_hashbucket	*lhash2;
178
179	bool				pernet;
 
 
 
 
 
 
 
 
 
 
 
 
 
180};
181
182static inline struct inet_hashinfo *tcp_or_dccp_get_hashinfo(const struct sock *sk)
183{
184#if IS_ENABLED(CONFIG_IP_DCCP)
185	return sk->sk_prot->h.hashinfo ? :
186		sock_net(sk)->ipv4.tcp_death_row.hashinfo;
187#else
188	return sock_net(sk)->ipv4.tcp_death_row.hashinfo;
189#endif
190}
191
192static inline struct inet_listen_hashbucket *
193inet_lhash2_bucket(struct inet_hashinfo *h, u32 hash)
194{
195	return &h->lhash2[hash & h->lhash2_mask];
196}
197
198static inline struct inet_ehash_bucket *inet_ehash_bucket(
199	struct inet_hashinfo *hashinfo,
200	unsigned int hash)
201{
202	return &hashinfo->ehash[hash & hashinfo->ehash_mask];
203}
204
205static inline spinlock_t *inet_ehash_lockp(
206	struct inet_hashinfo *hashinfo,
207	unsigned int hash)
208{
209	return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
210}
211
212int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo);
213
214static inline void inet_hashinfo2_free_mod(struct inet_hashinfo *h)
215{
216	kfree(h->lhash2);
217	h->lhash2 = NULL;
218}
219
220static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
221{
222	kvfree(hashinfo->ehash_locks);
223	hashinfo->ehash_locks = NULL;
224}
225
226struct inet_hashinfo *inet_pernet_hashinfo_alloc(struct inet_hashinfo *hashinfo,
227						 unsigned int ehash_entries);
228void inet_pernet_hashinfo_free(struct inet_hashinfo *hashinfo);
 
 
 
 
 
 
 
229
230struct inet_bind_bucket *
231inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net,
232			struct inet_bind_hashbucket *head,
233			const unsigned short snum, int l3mdev);
234void inet_bind_bucket_destroy(struct kmem_cache *cachep,
235			      struct inet_bind_bucket *tb);
236
237bool inet_bind_bucket_match(const struct inet_bind_bucket *tb,
238			    const struct net *net, unsigned short port,
239			    int l3mdev);
240
241struct inet_bind2_bucket *
242inet_bind2_bucket_create(struct kmem_cache *cachep, struct net *net,
243			 struct inet_bind_hashbucket *head,
244			 unsigned short port, int l3mdev,
245			 const struct sock *sk);
246
247void inet_bind2_bucket_destroy(struct kmem_cache *cachep,
248			       struct inet_bind2_bucket *tb);
249
250struct inet_bind2_bucket *
251inet_bind2_bucket_find(const struct inet_bind_hashbucket *head,
252		       const struct net *net,
253		       unsigned short port, int l3mdev,
254		       const struct sock *sk);
255
256bool inet_bind2_bucket_match_addr_any(const struct inet_bind2_bucket *tb,
257				      const struct net *net, unsigned short port,
258				      int l3mdev, const struct sock *sk);
259
260static inline u32 inet_bhashfn(const struct net *net, const __u16 lport,
261			       const u32 bhash_size)
262{
263	return (lport + net_hash_mix(net)) & (bhash_size - 1);
264}
265
266static inline struct inet_bind_hashbucket *
267inet_bhashfn_portaddr(const struct inet_hashinfo *hinfo, const struct sock *sk,
268		      const struct net *net, unsigned short port)
 
 
269{
270	u32 hash;
 
271
272#if IS_ENABLED(CONFIG_IPV6)
273	if (sk->sk_family == AF_INET6)
274		hash = ipv6_portaddr_hash(net, &sk->sk_v6_rcv_saddr, port);
275	else
276#endif
277		hash = ipv4_portaddr_hash(net, sk->sk_rcv_saddr, port);
278	return &hinfo->bhash2[hash & (hinfo->bhash_size - 1)];
279}
280
281struct inet_bind_hashbucket *
282inet_bhash2_addr_any_hashbucket(const struct sock *sk, const struct net *net, int port);
283
284/* This should be called whenever a socket's sk_rcv_saddr (ipv4) or
285 * sk_v6_rcv_saddr (ipv6) changes after it has been binded. The socket's
286 * rcv_saddr field should already have been updated when this is called.
287 */
288int inet_bhash2_update_saddr(struct sock *sk, void *saddr, int family);
289void inet_bhash2_reset_saddr(struct sock *sk);
290
291void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
292		    struct inet_bind2_bucket *tb2, unsigned short port);
293
294/* Caller must disable local BH processing. */
295int __inet_inherit_port(const struct sock *sk, struct sock *child);
296
297void inet_put_port(struct sock *sk);
298
 
299void inet_hashinfo2_init(struct inet_hashinfo *h, const char *name,
300			 unsigned long numentries, int scale,
301			 unsigned long low_limit,
302			 unsigned long high_limit);
303int inet_hashinfo2_init_mod(struct inet_hashinfo *h);
304
305bool inet_ehash_insert(struct sock *sk, struct sock *osk, bool *found_dup_sk);
306bool inet_ehash_nolisten(struct sock *sk, struct sock *osk,
307			 bool *found_dup_sk);
308int __inet_hash(struct sock *sk, struct sock *osk);
309int inet_hash(struct sock *sk);
310void inet_unhash(struct sock *sk);
311
312struct sock *__inet_lookup_listener(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,
317				    const unsigned short hnum,
318				    const int dif, const int sdif);
319
320static inline struct sock *inet_lookup_listener(struct net *net,
321		struct inet_hashinfo *hashinfo,
322		struct sk_buff *skb, int doff,
323		__be32 saddr, __be16 sport,
324		__be32 daddr, __be16 dport, int dif, int sdif)
325{
326	return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, sport,
327				      daddr, ntohs(dport), dif, sdif);
328}
329
330/* Socket demux engine toys. */
331/* What happens here is ugly; there's a pair of adjacent fields in
332   struct inet_sock; __be16 dport followed by __u16 num.  We want to
333   search by pair, so we combine the keys into a single 32bit value
334   and compare with 32bit value read from &...->dport.  Let's at least
335   make sure that it's not mixed with anything else...
336   On 64bit targets we combine comparisons with pair of adjacent __be32
337   fields in the same way.
338*/
339#ifdef __BIG_ENDIAN
340#define INET_COMBINED_PORTS(__sport, __dport) \
341	((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
342#else /* __LITTLE_ENDIAN */
343#define INET_COMBINED_PORTS(__sport, __dport) \
344	((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
345#endif
346
 
347#ifdef __BIG_ENDIAN
348#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
349	const __addrpair __name = (__force __addrpair) ( \
350				   (((__force __u64)(__be32)(__saddr)) << 32) | \
351				   ((__force __u64)(__be32)(__daddr)))
352#else /* __LITTLE_ENDIAN */
353#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
354	const __addrpair __name = (__force __addrpair) ( \
355				   (((__force __u64)(__be32)(__daddr)) << 32) | \
356				   ((__force __u64)(__be32)(__saddr)))
357#endif /* __BIG_ENDIAN */
 
 
 
 
 
 
 
 
 
358
359static inline bool inet_match(struct net *net, const struct sock *sk,
360			      const __addrpair cookie, const __portpair ports,
361			      int dif, int sdif)
362{
363	if (!net_eq(sock_net(sk), net) ||
364	    sk->sk_portpair != ports ||
365	    sk->sk_addrpair != cookie)
366	        return false;
367
368	/* READ_ONCE() paired with WRITE_ONCE() in sock_bindtoindex_locked() */
369	return inet_sk_bound_dev_eq(net, READ_ONCE(sk->sk_bound_dev_if), dif,
370				    sdif);
371}
372
373/* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
374 * not check it for lookups anymore, thanks Alexey. -DaveM
375 */
376struct sock *__inet_lookup_established(struct net *net,
377				       struct inet_hashinfo *hashinfo,
378				       const __be32 saddr, const __be16 sport,
379				       const __be32 daddr, const u16 hnum,
380				       const int dif, const int sdif);
381
382static inline struct sock *
383	inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
384				const __be32 saddr, const __be16 sport,
385				const __be32 daddr, const __be16 dport,
386				const int dif)
387{
388	return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
389					 ntohs(dport), dif, 0);
390}
391
392static inline struct sock *__inet_lookup(struct net *net,
393					 struct inet_hashinfo *hashinfo,
394					 struct sk_buff *skb, int doff,
395					 const __be32 saddr, const __be16 sport,
396					 const __be32 daddr, const __be16 dport,
397					 const int dif, const int sdif,
398					 bool *refcounted)
399{
400	u16 hnum = ntohs(dport);
401	struct sock *sk;
402
403	sk = __inet_lookup_established(net, hashinfo, saddr, sport,
404				       daddr, hnum, dif, sdif);
405	*refcounted = true;
406	if (sk)
407		return sk;
408	*refcounted = false;
409	return __inet_lookup_listener(net, hashinfo, skb, doff, saddr,
410				      sport, daddr, hnum, dif, sdif);
411}
412
413static inline struct sock *inet_lookup(struct net *net,
414				       struct inet_hashinfo *hashinfo,
415				       struct sk_buff *skb, int doff,
416				       const __be32 saddr, const __be16 sport,
417				       const __be32 daddr, const __be16 dport,
418				       const int dif)
419{
420	struct sock *sk;
421	bool refcounted;
422
423	sk = __inet_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
424			   dport, dif, 0, &refcounted);
425
426	if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
427		sk = NULL;
428	return sk;
429}
430
431static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
432					     struct sk_buff *skb,
433					     int doff,
434					     const __be16 sport,
435					     const __be16 dport,
436					     const int sdif,
437					     bool *refcounted)
438{
439	struct sock *sk = skb_steal_sock(skb, refcounted);
440	const struct iphdr *iph = ip_hdr(skb);
441
 
442	if (sk)
443		return sk;
444
445	return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo, skb,
446			     doff, iph->saddr, sport,
447			     iph->daddr, dport, inet_iif(skb), sdif,
448			     refcounted);
449}
450
451u32 inet6_ehashfn(const struct net *net,
452		  const struct in6_addr *laddr, const u16 lport,
453		  const struct in6_addr *faddr, const __be16 fport);
454
455static inline void sk_daddr_set(struct sock *sk, __be32 addr)
456{
457	sk->sk_daddr = addr; /* alias of inet_daddr */
458#if IS_ENABLED(CONFIG_IPV6)
459	ipv6_addr_set_v4mapped(addr, &sk->sk_v6_daddr);
460#endif
461}
462
463static inline void sk_rcv_saddr_set(struct sock *sk, __be32 addr)
464{
465	sk->sk_rcv_saddr = addr; /* alias of inet_rcv_saddr */
466#if IS_ENABLED(CONFIG_IPV6)
467	ipv6_addr_set_v4mapped(addr, &sk->sk_v6_rcv_saddr);
468#endif
469}
470
471int __inet_hash_connect(struct inet_timewait_death_row *death_row,
472			struct sock *sk, u64 port_offset,
473			int (*check_established)(struct inet_timewait_death_row *,
474						 struct sock *, __u16,
475						 struct inet_timewait_sock **));
476
477int inet_hash_connect(struct inet_timewait_death_row *death_row,
478		      struct sock *sk);
479#endif /* _INET_HASHTABLES_H */