xref: /openbmc/linux/include/net/inet_hashtables.h (revision 63705da3)
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  */
38 struct 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 
76 struct 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 
93 static 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 
101 struct inet_bind_hashbucket {
102 	spinlock_t		lock;
103 	struct hlist_head	chain;
104 };
105 
106 /* Sockets can be hashed in established or listening table.
107  * We must use different 'nulls' end-of-chain value for all hash buckets :
108  * A socket might transition from ESTABLISH to LISTEN state without
109  * RCU grace period. A lookup in ehash table needs to handle this case.
110  */
111 #define LISTENING_NULLS_BASE (1U << 29)
112 struct inet_listen_hashbucket {
113 	spinlock_t		lock;
114 	unsigned int		count;
115 	union {
116 		struct hlist_head	head;
117 		struct hlist_nulls_head	nulls_head;
118 	};
119 };
120 
121 /* This is for listening sockets, thus all sockets which possess wildcards. */
122 #define INET_LHTABLE_SIZE	32	/* Yes, really, this is all you need. */
123 
124 struct inet_hashinfo {
125 	/* This is for sockets with full identity only.  Sockets here will
126 	 * always be without wildcards and will have the following invariant:
127 	 *
128 	 *          TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
129 	 *
130 	 */
131 	struct inet_ehash_bucket	*ehash;
132 	spinlock_t			*ehash_locks;
133 	unsigned int			ehash_mask;
134 	unsigned int			ehash_locks_mask;
135 
136 	/* Ok, let's try this, I give up, we do need a local binding
137 	 * TCP hash as well as the others for fast bind/connect.
138 	 */
139 	struct kmem_cache		*bind_bucket_cachep;
140 	struct inet_bind_hashbucket	*bhash;
141 	unsigned int			bhash_size;
142 
143 	/* The 2nd listener table hashed by local port and address */
144 	unsigned int			lhash2_mask;
145 	struct inet_listen_hashbucket	*lhash2;
146 
147 	/* All the above members are written once at bootup and
148 	 * never written again _or_ are predominantly read-access.
149 	 *
150 	 * Now align to a new cache line as all the following members
151 	 * might be often dirty.
152 	 */
153 	/* All sockets in TCP_LISTEN state will be in listening_hash.
154 	 * This is the only table where wildcard'd TCP sockets can
155 	 * exist.  listening_hash is only hashed by local port number.
156 	 * If lhash2 is initialized, the same socket will also be hashed
157 	 * to lhash2 by port and address.
158 	 */
159 	struct inet_listen_hashbucket	listening_hash[INET_LHTABLE_SIZE]
160 					____cacheline_aligned_in_smp;
161 };
162 
163 #define inet_lhash2_for_each_icsk_continue(__icsk) \
164 	hlist_for_each_entry_continue(__icsk, icsk_listen_portaddr_node)
165 
166 #define inet_lhash2_for_each_icsk(__icsk, list) \
167 	hlist_for_each_entry(__icsk, list, icsk_listen_portaddr_node)
168 
169 #define inet_lhash2_for_each_icsk_rcu(__icsk, list) \
170 	hlist_for_each_entry_rcu(__icsk, list, icsk_listen_portaddr_node)
171 
172 static inline struct inet_listen_hashbucket *
173 inet_lhash2_bucket(struct inet_hashinfo *h, u32 hash)
174 {
175 	return &h->lhash2[hash & h->lhash2_mask];
176 }
177 
178 static inline struct inet_ehash_bucket *inet_ehash_bucket(
179 	struct inet_hashinfo *hashinfo,
180 	unsigned int hash)
181 {
182 	return &hashinfo->ehash[hash & hashinfo->ehash_mask];
183 }
184 
185 static inline spinlock_t *inet_ehash_lockp(
186 	struct inet_hashinfo *hashinfo,
187 	unsigned int hash)
188 {
189 	return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
190 }
191 
192 int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo);
193 
194 static inline void inet_hashinfo2_free_mod(struct inet_hashinfo *h)
195 {
196 	kfree(h->lhash2);
197 	h->lhash2 = NULL;
198 }
199 
200 static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
201 {
202 	kvfree(hashinfo->ehash_locks);
203 	hashinfo->ehash_locks = NULL;
204 }
205 
206 static inline bool inet_sk_bound_dev_eq(struct net *net, int bound_dev_if,
207 					int dif, int sdif)
208 {
209 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
210 	return inet_bound_dev_eq(!!net->ipv4.sysctl_tcp_l3mdev_accept,
211 				 bound_dev_if, dif, sdif);
212 #else
213 	return inet_bound_dev_eq(true, bound_dev_if, dif, sdif);
214 #endif
215 }
216 
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);
223 
224 static inline u32 inet_bhashfn(const struct net *net, const __u16 lport,
225 			       const u32 bhash_size)
226 {
227 	return (lport + net_hash_mix(net)) & (bhash_size - 1);
228 }
229 
230 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
231 		    const unsigned short snum);
232 
233 /* These can have wildcards, don't try too hard. */
234 static inline u32 inet_lhashfn(const struct net *net, const unsigned short num)
235 {
236 	return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1);
237 }
238 
239 static inline int inet_sk_listen_hashfn(const struct sock *sk)
240 {
241 	return inet_lhashfn(sock_net(sk), inet_sk(sk)->inet_num);
242 }
243 
244 /* Caller must disable local BH processing. */
245 int __inet_inherit_port(const struct sock *sk, struct sock *child);
246 
247 void inet_put_port(struct sock *sk);
248 
249 void inet_hashinfo_init(struct inet_hashinfo *h);
250 void inet_hashinfo2_init(struct inet_hashinfo *h, const char *name,
251 			 unsigned long numentries, int scale,
252 			 unsigned long low_limit,
253 			 unsigned long high_limit);
254 int inet_hashinfo2_init_mod(struct inet_hashinfo *h);
255 
256 bool inet_ehash_insert(struct sock *sk, struct sock *osk, bool *found_dup_sk);
257 bool inet_ehash_nolisten(struct sock *sk, struct sock *osk,
258 			 bool *found_dup_sk);
259 int __inet_hash(struct sock *sk, struct sock *osk);
260 int inet_hash(struct sock *sk);
261 void inet_unhash(struct sock *sk);
262 
263 struct sock *__inet_lookup_listener(struct net *net,
264 				    struct inet_hashinfo *hashinfo,
265 				    struct sk_buff *skb, int doff,
266 				    const __be32 saddr, const __be16 sport,
267 				    const __be32 daddr,
268 				    const unsigned short hnum,
269 				    const int dif, const int sdif);
270 
271 static inline struct sock *inet_lookup_listener(struct net *net,
272 		struct inet_hashinfo *hashinfo,
273 		struct sk_buff *skb, int doff,
274 		__be32 saddr, __be16 sport,
275 		__be32 daddr, __be16 dport, int dif, int sdif)
276 {
277 	return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, sport,
278 				      daddr, ntohs(dport), dif, sdif);
279 }
280 
281 /* Socket demux engine toys. */
282 /* What happens here is ugly; there's a pair of adjacent fields in
283    struct inet_sock; __be16 dport followed by __u16 num.  We want to
284    search by pair, so we combine the keys into a single 32bit value
285    and compare with 32bit value read from &...->dport.  Let's at least
286    make sure that it's not mixed with anything else...
287    On 64bit targets we combine comparisons with pair of adjacent __be32
288    fields in the same way.
289 */
290 #ifdef __BIG_ENDIAN
291 #define INET_COMBINED_PORTS(__sport, __dport) \
292 	((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
293 #else /* __LITTLE_ENDIAN */
294 #define INET_COMBINED_PORTS(__sport, __dport) \
295 	((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
296 #endif
297 
298 #if (BITS_PER_LONG == 64)
299 #ifdef __BIG_ENDIAN
300 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
301 	const __addrpair __name = (__force __addrpair) ( \
302 				   (((__force __u64)(__be32)(__saddr)) << 32) | \
303 				   ((__force __u64)(__be32)(__daddr)))
304 #else /* __LITTLE_ENDIAN */
305 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
306 	const __addrpair __name = (__force __addrpair) ( \
307 				   (((__force __u64)(__be32)(__daddr)) << 32) | \
308 				   ((__force __u64)(__be32)(__saddr)))
309 #endif /* __BIG_ENDIAN */
310 #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif, __sdif) \
311 	(((__sk)->sk_portpair == (__ports))			&&	\
312 	 ((__sk)->sk_addrpair == (__cookie))			&&	\
313 	 (((__sk)->sk_bound_dev_if == (__dif))			||	\
314 	  ((__sk)->sk_bound_dev_if == (__sdif)))		&&	\
315 	 net_eq(sock_net(__sk), (__net)))
316 #else /* 32-bit arch */
317 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
318 	const int __name __deprecated __attribute__((unused))
319 
320 #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif, __sdif) \
321 	(((__sk)->sk_portpair == (__ports))		&&		\
322 	 ((__sk)->sk_daddr	== (__saddr))		&&		\
323 	 ((__sk)->sk_rcv_saddr	== (__daddr))		&&		\
324 	 (((__sk)->sk_bound_dev_if == (__dif))		||		\
325 	  ((__sk)->sk_bound_dev_if == (__sdif)))	&&		\
326 	 net_eq(sock_net(__sk), (__net)))
327 #endif /* 64-bit arch */
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 struct sock *__inet_lookup_established(struct net *net,
333 				       struct inet_hashinfo *hashinfo,
334 				       const __be32 saddr, const __be16 sport,
335 				       const __be32 daddr, const u16 hnum,
336 				       const int dif, const int sdif);
337 
338 static inline struct sock *
339 	inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
340 				const __be32 saddr, const __be16 sport,
341 				const __be32 daddr, const __be16 dport,
342 				const int dif)
343 {
344 	return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
345 					 ntohs(dport), dif, 0);
346 }
347 
348 static inline struct sock *__inet_lookup(struct net *net,
349 					 struct inet_hashinfo *hashinfo,
350 					 struct sk_buff *skb, int doff,
351 					 const __be32 saddr, const __be16 sport,
352 					 const __be32 daddr, const __be16 dport,
353 					 const int dif, const int sdif,
354 					 bool *refcounted)
355 {
356 	u16 hnum = ntohs(dport);
357 	struct sock *sk;
358 
359 	sk = __inet_lookup_established(net, hashinfo, saddr, sport,
360 				       daddr, hnum, dif, sdif);
361 	*refcounted = true;
362 	if (sk)
363 		return sk;
364 	*refcounted = false;
365 	return __inet_lookup_listener(net, hashinfo, skb, doff, saddr,
366 				      sport, daddr, hnum, dif, sdif);
367 }
368 
369 static inline struct sock *inet_lookup(struct net *net,
370 				       struct inet_hashinfo *hashinfo,
371 				       struct sk_buff *skb, int doff,
372 				       const __be32 saddr, const __be16 sport,
373 				       const __be32 daddr, const __be16 dport,
374 				       const int dif)
375 {
376 	struct sock *sk;
377 	bool refcounted;
378 
379 	sk = __inet_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
380 			   dport, dif, 0, &refcounted);
381 
382 	if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
383 		sk = NULL;
384 	return sk;
385 }
386 
387 static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
388 					     struct sk_buff *skb,
389 					     int doff,
390 					     const __be16 sport,
391 					     const __be16 dport,
392 					     const int sdif,
393 					     bool *refcounted)
394 {
395 	struct sock *sk = skb_steal_sock(skb, refcounted);
396 	const struct iphdr *iph = ip_hdr(skb);
397 
398 	if (sk)
399 		return sk;
400 
401 	return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo, skb,
402 			     doff, iph->saddr, sport,
403 			     iph->daddr, dport, inet_iif(skb), sdif,
404 			     refcounted);
405 }
406 
407 u32 inet6_ehashfn(const struct net *net,
408 		  const struct in6_addr *laddr, const u16 lport,
409 		  const struct in6_addr *faddr, const __be16 fport);
410 
411 static inline void sk_daddr_set(struct sock *sk, __be32 addr)
412 {
413 	sk->sk_daddr = addr; /* alias of inet_daddr */
414 #if IS_ENABLED(CONFIG_IPV6)
415 	ipv6_addr_set_v4mapped(addr, &sk->sk_v6_daddr);
416 #endif
417 }
418 
419 static inline void sk_rcv_saddr_set(struct sock *sk, __be32 addr)
420 {
421 	sk->sk_rcv_saddr = addr; /* alias of inet_rcv_saddr */
422 #if IS_ENABLED(CONFIG_IPV6)
423 	ipv6_addr_set_v4mapped(addr, &sk->sk_v6_rcv_saddr);
424 #endif
425 }
426 
427 int __inet_hash_connect(struct inet_timewait_death_row *death_row,
428 			struct sock *sk, u32 port_offset,
429 			int (*check_established)(struct inet_timewait_death_row *,
430 						 struct sock *, __u16,
431 						 struct inet_timewait_sock **));
432 
433 int inet_hash_connect(struct inet_timewait_death_row *death_row,
434 		      struct sock *sk);
435 #endif /* _INET_HASHTABLES_H */
436