xref: /openbmc/linux/include/net/inet_hashtables.h (revision e2c75e76)
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 
28 #include <net/inet_connection_sock.h>
29 #include <net/inet_sock.h>
30 #include <net/sock.h>
31 #include <net/route.h>
32 #include <net/tcp_states.h>
33 #include <net/netns/hash.h>
34 
35 #include <linux/refcount.h>
36 #include <asm/byteorder.h>
37 
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
40  * but LISTEN ones.
41  */
42 struct inet_ehash_bucket {
43 	struct hlist_nulls_head chain;
44 };
45 
46 /* There are a few simple rules, which allow for local port reuse by
47  * an application.  In essence:
48  *
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
56  *	   shared.
57  *	   Failing this, the port cannot be shared.
58  *
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.
69  *
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
76  */
77 #define FASTREUSEPORT_ANY	1
78 #define FASTREUSEPORT_STRICT	2
79 
80 struct inet_bind_bucket {
81 	possible_net_t		ib_net;
82 	unsigned short		port;
83 	signed char		fastreuse;
84 	signed char		fastreuseport;
85 	kuid_t			fastuid;
86 #if IS_ENABLED(CONFIG_IPV6)
87 	struct in6_addr		fast_v6_rcv_saddr;
88 #endif
89 	__be32			fast_rcv_saddr;
90 	unsigned short		fast_sk_family;
91 	bool			fast_ipv6_only;
92 	struct hlist_node	node;
93 	struct hlist_head	owners;
94 };
95 
96 static inline struct net *ib_net(struct inet_bind_bucket *ib)
97 {
98 	return read_pnet(&ib->ib_net);
99 }
100 
101 #define inet_bind_bucket_for_each(tb, head) \
102 	hlist_for_each_entry(tb, head, node)
103 
104 struct inet_bind_hashbucket {
105 	spinlock_t		lock;
106 	struct hlist_head	chain;
107 };
108 
109 /*
110  * Sockets can be hashed in established or listening table
111  */
112 struct inet_listen_hashbucket {
113 	spinlock_t		lock;
114 	unsigned int		count;
115 	struct hlist_head	head;
116 };
117 
118 /* This is for listening sockets, thus all sockets which possess wildcards. */
119 #define INET_LHTABLE_SIZE	32	/* Yes, really, this is all you need. */
120 
121 struct inet_hashinfo {
122 	/* This is for sockets with full identity only.  Sockets here will
123 	 * always be without wildcards and will have the following invariant:
124 	 *
125 	 *          TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
126 	 *
127 	 */
128 	struct inet_ehash_bucket	*ehash;
129 	spinlock_t			*ehash_locks;
130 	unsigned int			ehash_mask;
131 	unsigned int			ehash_locks_mask;
132 
133 	/* Ok, let's try this, I give up, we do need a local binding
134 	 * TCP hash as well as the others for fast bind/connect.
135 	 */
136 	struct kmem_cache		*bind_bucket_cachep;
137 	struct inet_bind_hashbucket	*bhash;
138 	unsigned int			bhash_size;
139 
140 	/* The 2nd listener table hashed by local port and address */
141 	unsigned int			lhash2_mask;
142 	struct inet_listen_hashbucket	*lhash2;
143 
144 	/* All the above members are written once at bootup and
145 	 * never written again _or_ are predominantly read-access.
146 	 *
147 	 * Now align to a new cache line as all the following members
148 	 * might be often dirty.
149 	 */
150 	/* All sockets in TCP_LISTEN state will be in listening_hash.
151 	 * This is the only table where wildcard'd TCP sockets can
152 	 * exist.  listening_hash is only hashed by local port number.
153 	 * If lhash2 is initialized, the same socket will also be hashed
154 	 * to lhash2 by port and address.
155 	 */
156 	struct inet_listen_hashbucket	listening_hash[INET_LHTABLE_SIZE]
157 					____cacheline_aligned_in_smp;
158 };
159 
160 #define inet_lhash2_for_each_icsk_rcu(__icsk, list) \
161 	hlist_for_each_entry_rcu(__icsk, list, icsk_listen_portaddr_node)
162 
163 static inline struct inet_listen_hashbucket *
164 inet_lhash2_bucket(struct inet_hashinfo *h, u32 hash)
165 {
166 	return &h->lhash2[hash & h->lhash2_mask];
167 }
168 
169 static inline struct inet_ehash_bucket *inet_ehash_bucket(
170 	struct inet_hashinfo *hashinfo,
171 	unsigned int hash)
172 {
173 	return &hashinfo->ehash[hash & hashinfo->ehash_mask];
174 }
175 
176 static inline spinlock_t *inet_ehash_lockp(
177 	struct inet_hashinfo *hashinfo,
178 	unsigned int hash)
179 {
180 	return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
181 }
182 
183 int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo);
184 
185 static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
186 {
187 	kvfree(hashinfo->ehash_locks);
188 	hashinfo->ehash_locks = NULL;
189 }
190 
191 struct inet_bind_bucket *
192 inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net,
193 			struct inet_bind_hashbucket *head,
194 			const unsigned short snum);
195 void inet_bind_bucket_destroy(struct kmem_cache *cachep,
196 			      struct inet_bind_bucket *tb);
197 
198 static inline u32 inet_bhashfn(const struct net *net, const __u16 lport,
199 			       const u32 bhash_size)
200 {
201 	return (lport + net_hash_mix(net)) & (bhash_size - 1);
202 }
203 
204 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
205 		    const unsigned short snum);
206 
207 /* These can have wildcards, don't try too hard. */
208 static inline u32 inet_lhashfn(const struct net *net, const unsigned short num)
209 {
210 	return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1);
211 }
212 
213 static inline int inet_sk_listen_hashfn(const struct sock *sk)
214 {
215 	return inet_lhashfn(sock_net(sk), inet_sk(sk)->inet_num);
216 }
217 
218 /* Caller must disable local BH processing. */
219 int __inet_inherit_port(const struct sock *sk, struct sock *child);
220 
221 void inet_put_port(struct sock *sk);
222 
223 void inet_hashinfo_init(struct inet_hashinfo *h);
224 void inet_hashinfo2_init(struct inet_hashinfo *h, const char *name,
225 			 unsigned long numentries, int scale,
226 			 unsigned long low_limit,
227 			 unsigned long high_limit);
228 
229 bool inet_ehash_insert(struct sock *sk, struct sock *osk);
230 bool inet_ehash_nolisten(struct sock *sk, struct sock *osk);
231 int __inet_hash(struct sock *sk, struct sock *osk);
232 int inet_hash(struct sock *sk);
233 void inet_unhash(struct sock *sk);
234 
235 struct sock *__inet_lookup_listener(struct net *net,
236 				    struct inet_hashinfo *hashinfo,
237 				    struct sk_buff *skb, int doff,
238 				    const __be32 saddr, const __be16 sport,
239 				    const __be32 daddr,
240 				    const unsigned short hnum,
241 				    const int dif, const int sdif);
242 
243 static inline struct sock *inet_lookup_listener(struct net *net,
244 		struct inet_hashinfo *hashinfo,
245 		struct sk_buff *skb, int doff,
246 		__be32 saddr, __be16 sport,
247 		__be32 daddr, __be16 dport, int dif, int sdif)
248 {
249 	return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, sport,
250 				      daddr, ntohs(dport), dif, sdif);
251 }
252 
253 /* Socket demux engine toys. */
254 /* What happens here is ugly; there's a pair of adjacent fields in
255    struct inet_sock; __be16 dport followed by __u16 num.  We want to
256    search by pair, so we combine the keys into a single 32bit value
257    and compare with 32bit value read from &...->dport.  Let's at least
258    make sure that it's not mixed with anything else...
259    On 64bit targets we combine comparisons with pair of adjacent __be32
260    fields in the same way.
261 */
262 #ifdef __BIG_ENDIAN
263 #define INET_COMBINED_PORTS(__sport, __dport) \
264 	((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
265 #else /* __LITTLE_ENDIAN */
266 #define INET_COMBINED_PORTS(__sport, __dport) \
267 	((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
268 #endif
269 
270 #if (BITS_PER_LONG == 64)
271 #ifdef __BIG_ENDIAN
272 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
273 	const __addrpair __name = (__force __addrpair) ( \
274 				   (((__force __u64)(__be32)(__saddr)) << 32) | \
275 				   ((__force __u64)(__be32)(__daddr)))
276 #else /* __LITTLE_ENDIAN */
277 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
278 	const __addrpair __name = (__force __addrpair) ( \
279 				   (((__force __u64)(__be32)(__daddr)) << 32) | \
280 				   ((__force __u64)(__be32)(__saddr)))
281 #endif /* __BIG_ENDIAN */
282 #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif, __sdif) \
283 	(((__sk)->sk_portpair == (__ports))			&&	\
284 	 ((__sk)->sk_addrpair == (__cookie))			&&	\
285 	 (!(__sk)->sk_bound_dev_if	||				\
286 	   ((__sk)->sk_bound_dev_if == (__dif))			||	\
287 	   ((__sk)->sk_bound_dev_if == (__sdif)))		&&	\
288 	 net_eq(sock_net(__sk), (__net)))
289 #else /* 32-bit arch */
290 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
291 	const int __name __deprecated __attribute__((unused))
292 
293 #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif, __sdif) \
294 	(((__sk)->sk_portpair == (__ports))		&&		\
295 	 ((__sk)->sk_daddr	== (__saddr))		&&		\
296 	 ((__sk)->sk_rcv_saddr	== (__daddr))		&&		\
297 	 (!(__sk)->sk_bound_dev_if	||				\
298 	   ((__sk)->sk_bound_dev_if == (__dif))		||		\
299 	   ((__sk)->sk_bound_dev_if == (__sdif)))	&&		\
300 	 net_eq(sock_net(__sk), (__net)))
301 #endif /* 64-bit arch */
302 
303 /* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
304  * not check it for lookups anymore, thanks Alexey. -DaveM
305  */
306 struct sock *__inet_lookup_established(struct net *net,
307 				       struct inet_hashinfo *hashinfo,
308 				       const __be32 saddr, const __be16 sport,
309 				       const __be32 daddr, const u16 hnum,
310 				       const int dif, const int sdif);
311 
312 static inline struct sock *
313 	inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
314 				const __be32 saddr, const __be16 sport,
315 				const __be32 daddr, const __be16 dport,
316 				const int dif)
317 {
318 	return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
319 					 ntohs(dport), dif, 0);
320 }
321 
322 static inline struct sock *__inet_lookup(struct net *net,
323 					 struct inet_hashinfo *hashinfo,
324 					 struct sk_buff *skb, int doff,
325 					 const __be32 saddr, const __be16 sport,
326 					 const __be32 daddr, const __be16 dport,
327 					 const int dif, const int sdif,
328 					 bool *refcounted)
329 {
330 	u16 hnum = ntohs(dport);
331 	struct sock *sk;
332 
333 	sk = __inet_lookup_established(net, hashinfo, saddr, sport,
334 				       daddr, hnum, dif, sdif);
335 	*refcounted = true;
336 	if (sk)
337 		return sk;
338 	*refcounted = false;
339 	return __inet_lookup_listener(net, hashinfo, skb, doff, saddr,
340 				      sport, daddr, hnum, dif, sdif);
341 }
342 
343 static inline struct sock *inet_lookup(struct net *net,
344 				       struct inet_hashinfo *hashinfo,
345 				       struct sk_buff *skb, int doff,
346 				       const __be32 saddr, const __be16 sport,
347 				       const __be32 daddr, const __be16 dport,
348 				       const int dif)
349 {
350 	struct sock *sk;
351 	bool refcounted;
352 
353 	sk = __inet_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
354 			   dport, dif, 0, &refcounted);
355 
356 	if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
357 		sk = NULL;
358 	return sk;
359 }
360 
361 static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
362 					     struct sk_buff *skb,
363 					     int doff,
364 					     const __be16 sport,
365 					     const __be16 dport,
366 					     const int sdif,
367 					     bool *refcounted)
368 {
369 	struct sock *sk = skb_steal_sock(skb);
370 	const struct iphdr *iph = ip_hdr(skb);
371 
372 	*refcounted = true;
373 	if (sk)
374 		return sk;
375 
376 	return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo, skb,
377 			     doff, iph->saddr, sport,
378 			     iph->daddr, dport, inet_iif(skb), sdif,
379 			     refcounted);
380 }
381 
382 u32 inet6_ehashfn(const struct net *net,
383 		  const struct in6_addr *laddr, const u16 lport,
384 		  const struct in6_addr *faddr, const __be16 fport);
385 
386 static inline void sk_daddr_set(struct sock *sk, __be32 addr)
387 {
388 	sk->sk_daddr = addr; /* alias of inet_daddr */
389 #if IS_ENABLED(CONFIG_IPV6)
390 	ipv6_addr_set_v4mapped(addr, &sk->sk_v6_daddr);
391 #endif
392 }
393 
394 static inline void sk_rcv_saddr_set(struct sock *sk, __be32 addr)
395 {
396 	sk->sk_rcv_saddr = addr; /* alias of inet_rcv_saddr */
397 #if IS_ENABLED(CONFIG_IPV6)
398 	ipv6_addr_set_v4mapped(addr, &sk->sk_v6_rcv_saddr);
399 #endif
400 }
401 
402 int __inet_hash_connect(struct inet_timewait_death_row *death_row,
403 			struct sock *sk, u32 port_offset,
404 			int (*check_established)(struct inet_timewait_death_row *,
405 						 struct sock *, __u16,
406 						 struct inet_timewait_sock **));
407 
408 int inet_hash_connect(struct inet_timewait_death_row *death_row,
409 		      struct sock *sk);
410 #endif /* _INET_HASHTABLES_H */
411