xref: /linux/include/net/inet_hashtables.h (revision dec1c62e91ba268ab2a6e339d4d7a59287d5eba1)
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 struct inet_bind2_bucket {
94 	possible_net_t		ib_net;
95 	int			l3mdev;
96 	unsigned short		port;
97 	union {
98 #if IS_ENABLED(CONFIG_IPV6)
99 		struct in6_addr		v6_rcv_saddr;
100 #endif
101 		__be32			rcv_saddr;
102 	};
103 	/* Node in the inet2_bind_hashbucket chain */
104 	struct hlist_node	node;
105 	/* List of sockets hashed to this bucket */
106 	struct hlist_head	owners;
107 };
108 
109 static inline struct net *ib_net(struct inet_bind_bucket *ib)
110 {
111 	return read_pnet(&ib->ib_net);
112 }
113 
114 static inline struct net *ib2_net(struct inet_bind2_bucket *ib)
115 {
116 	return read_pnet(&ib->ib_net);
117 }
118 
119 #define inet_bind_bucket_for_each(tb, head) \
120 	hlist_for_each_entry(tb, head, node)
121 
122 struct inet_bind_hashbucket {
123 	spinlock_t		lock;
124 	struct hlist_head	chain;
125 };
126 
127 /* This is synchronized using the inet_bind_hashbucket's spinlock.
128  * Instead of having separate spinlocks, the inet_bind2_hashbucket can share
129  * the inet_bind_hashbucket's given that in every case where the bhash2 table
130  * is useful, a lookup in the bhash table also occurs.
131  */
132 struct inet_bind2_hashbucket {
133 	struct hlist_head	chain;
134 };
135 
136 /* Sockets can be hashed in established or listening table.
137  * We must use different 'nulls' end-of-chain value for all hash buckets :
138  * A socket might transition from ESTABLISH to LISTEN state without
139  * RCU grace period. A lookup in ehash table needs to handle this case.
140  */
141 #define LISTENING_NULLS_BASE (1U << 29)
142 struct inet_listen_hashbucket {
143 	spinlock_t		lock;
144 	struct hlist_nulls_head	nulls_head;
145 };
146 
147 /* This is for listening sockets, thus all sockets which possess wildcards. */
148 #define INET_LHTABLE_SIZE	32	/* Yes, really, this is all you need. */
149 
150 struct inet_hashinfo {
151 	/* This is for sockets with full identity only.  Sockets here will
152 	 * always be without wildcards and will have the following invariant:
153 	 *
154 	 *          TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
155 	 *
156 	 */
157 	struct inet_ehash_bucket	*ehash;
158 	spinlock_t			*ehash_locks;
159 	unsigned int			ehash_mask;
160 	unsigned int			ehash_locks_mask;
161 
162 	/* Ok, let's try this, I give up, we do need a local binding
163 	 * TCP hash as well as the others for fast bind/connect.
164 	 */
165 	struct kmem_cache		*bind_bucket_cachep;
166 	struct inet_bind_hashbucket	*bhash;
167 	/* The 2nd binding table hashed by port and address.
168 	 * This is used primarily for expediting the resolution of bind
169 	 * conflicts.
170 	 */
171 	struct kmem_cache		*bind2_bucket_cachep;
172 	struct inet_bind2_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 
180 static inline struct inet_listen_hashbucket *
181 inet_lhash2_bucket(struct inet_hashinfo *h, u32 hash)
182 {
183 	return &h->lhash2[hash & h->lhash2_mask];
184 }
185 
186 static inline struct inet_ehash_bucket *inet_ehash_bucket(
187 	struct inet_hashinfo *hashinfo,
188 	unsigned int hash)
189 {
190 	return &hashinfo->ehash[hash & hashinfo->ehash_mask];
191 }
192 
193 static inline spinlock_t *inet_ehash_lockp(
194 	struct inet_hashinfo *hashinfo,
195 	unsigned int hash)
196 {
197 	return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
198 }
199 
200 int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo);
201 
202 static inline void inet_hashinfo2_free_mod(struct inet_hashinfo *h)
203 {
204 	kfree(h->lhash2);
205 	h->lhash2 = NULL;
206 }
207 
208 static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
209 {
210 	kvfree(hashinfo->ehash_locks);
211 	hashinfo->ehash_locks = NULL;
212 }
213 
214 static inline bool inet_sk_bound_dev_eq(struct net *net, int bound_dev_if,
215 					int dif, int sdif)
216 {
217 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
218 	return inet_bound_dev_eq(!!net->ipv4.sysctl_tcp_l3mdev_accept,
219 				 bound_dev_if, dif, sdif);
220 #else
221 	return inet_bound_dev_eq(true, bound_dev_if, dif, sdif);
222 #endif
223 }
224 
225 struct inet_bind_bucket *
226 inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net,
227 			struct inet_bind_hashbucket *head,
228 			const unsigned short snum, int l3mdev);
229 void inet_bind_bucket_destroy(struct kmem_cache *cachep,
230 			      struct inet_bind_bucket *tb);
231 
232 static inline bool check_bind_bucket_match(struct inet_bind_bucket *tb,
233 					   struct net *net,
234 					   const unsigned short port,
235 					   int l3mdev)
236 {
237 	return net_eq(ib_net(tb), net) && tb->port == port &&
238 		tb->l3mdev == l3mdev;
239 }
240 
241 struct inet_bind2_bucket *
242 inet_bind2_bucket_create(struct kmem_cache *cachep, struct net *net,
243 			 struct inet_bind2_hashbucket *head,
244 			 const unsigned short port, int l3mdev,
245 			 const struct sock *sk);
246 
247 void inet_bind2_bucket_destroy(struct kmem_cache *cachep,
248 			       struct inet_bind2_bucket *tb);
249 
250 struct inet_bind2_bucket *
251 inet_bind2_bucket_find(struct inet_hashinfo *hinfo, struct net *net,
252 		       const unsigned short port, int l3mdev,
253 		       struct sock *sk,
254 		       struct inet_bind2_hashbucket **head);
255 
256 bool check_bind2_bucket_match_nulladdr(struct inet_bind2_bucket *tb,
257 				       struct net *net,
258 				       const unsigned short port,
259 				       int l3mdev,
260 				       const struct sock *sk);
261 
262 static inline u32 inet_bhashfn(const struct net *net, const __u16 lport,
263 			       const u32 bhash_size)
264 {
265 	return (lport + net_hash_mix(net)) & (bhash_size - 1);
266 }
267 
268 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
269 		    struct inet_bind2_bucket *tb2, const unsigned short snum);
270 
271 /* Caller must disable local BH processing. */
272 int __inet_inherit_port(const struct sock *sk, struct sock *child);
273 
274 void inet_put_port(struct sock *sk);
275 
276 void inet_hashinfo2_init(struct inet_hashinfo *h, const char *name,
277 			 unsigned long numentries, int scale,
278 			 unsigned long low_limit,
279 			 unsigned long high_limit);
280 int inet_hashinfo2_init_mod(struct inet_hashinfo *h);
281 
282 bool inet_ehash_insert(struct sock *sk, struct sock *osk, bool *found_dup_sk);
283 bool inet_ehash_nolisten(struct sock *sk, struct sock *osk,
284 			 bool *found_dup_sk);
285 int __inet_hash(struct sock *sk, struct sock *osk);
286 int inet_hash(struct sock *sk);
287 void inet_unhash(struct sock *sk);
288 
289 struct sock *__inet_lookup_listener(struct net *net,
290 				    struct inet_hashinfo *hashinfo,
291 				    struct sk_buff *skb, int doff,
292 				    const __be32 saddr, const __be16 sport,
293 				    const __be32 daddr,
294 				    const unsigned short hnum,
295 				    const int dif, const int sdif);
296 
297 static inline struct sock *inet_lookup_listener(struct net *net,
298 		struct inet_hashinfo *hashinfo,
299 		struct sk_buff *skb, int doff,
300 		__be32 saddr, __be16 sport,
301 		__be32 daddr, __be16 dport, int dif, int sdif)
302 {
303 	return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, sport,
304 				      daddr, ntohs(dport), dif, sdif);
305 }
306 
307 /* Socket demux engine toys. */
308 /* What happens here is ugly; there's a pair of adjacent fields in
309    struct inet_sock; __be16 dport followed by __u16 num.  We want to
310    search by pair, so we combine the keys into a single 32bit value
311    and compare with 32bit value read from &...->dport.  Let's at least
312    make sure that it's not mixed with anything else...
313    On 64bit targets we combine comparisons with pair of adjacent __be32
314    fields in the same way.
315 */
316 #ifdef __BIG_ENDIAN
317 #define INET_COMBINED_PORTS(__sport, __dport) \
318 	((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
319 #else /* __LITTLE_ENDIAN */
320 #define INET_COMBINED_PORTS(__sport, __dport) \
321 	((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
322 #endif
323 
324 #ifdef __BIG_ENDIAN
325 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
326 	const __addrpair __name = (__force __addrpair) ( \
327 				   (((__force __u64)(__be32)(__saddr)) << 32) | \
328 				   ((__force __u64)(__be32)(__daddr)))
329 #else /* __LITTLE_ENDIAN */
330 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
331 	const __addrpair __name = (__force __addrpair) ( \
332 				   (((__force __u64)(__be32)(__daddr)) << 32) | \
333 				   ((__force __u64)(__be32)(__saddr)))
334 #endif /* __BIG_ENDIAN */
335 
336 static inline bool inet_match(struct net *net, const struct sock *sk,
337 			      const __addrpair cookie, const __portpair ports,
338 			      int dif, int sdif)
339 {
340 	int bound_dev_if;
341 
342 	if (!net_eq(sock_net(sk), net) ||
343 	    sk->sk_portpair != ports ||
344 	    sk->sk_addrpair != cookie)
345 	        return false;
346 
347 	/* Paired with WRITE_ONCE() from sock_bindtoindex_locked() */
348 	bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
349 	return bound_dev_if == dif || bound_dev_if == sdif;
350 }
351 
352 /* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
353  * not check it for lookups anymore, thanks Alexey. -DaveM
354  */
355 struct sock *__inet_lookup_established(struct net *net,
356 				       struct inet_hashinfo *hashinfo,
357 				       const __be32 saddr, const __be16 sport,
358 				       const __be32 daddr, const u16 hnum,
359 				       const int dif, const int sdif);
360 
361 static inline struct sock *
362 	inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
363 				const __be32 saddr, const __be16 sport,
364 				const __be32 daddr, const __be16 dport,
365 				const int dif)
366 {
367 	return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
368 					 ntohs(dport), dif, 0);
369 }
370 
371 static inline struct sock *__inet_lookup(struct net *net,
372 					 struct inet_hashinfo *hashinfo,
373 					 struct sk_buff *skb, int doff,
374 					 const __be32 saddr, const __be16 sport,
375 					 const __be32 daddr, const __be16 dport,
376 					 const int dif, const int sdif,
377 					 bool *refcounted)
378 {
379 	u16 hnum = ntohs(dport);
380 	struct sock *sk;
381 
382 	sk = __inet_lookup_established(net, hashinfo, saddr, sport,
383 				       daddr, hnum, dif, sdif);
384 	*refcounted = true;
385 	if (sk)
386 		return sk;
387 	*refcounted = false;
388 	return __inet_lookup_listener(net, hashinfo, skb, doff, saddr,
389 				      sport, daddr, hnum, dif, sdif);
390 }
391 
392 static 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)
398 {
399 	struct sock *sk;
400 	bool refcounted;
401 
402 	sk = __inet_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
403 			   dport, dif, 0, &refcounted);
404 
405 	if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
406 		sk = NULL;
407 	return sk;
408 }
409 
410 static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
411 					     struct sk_buff *skb,
412 					     int doff,
413 					     const __be16 sport,
414 					     const __be16 dport,
415 					     const int sdif,
416 					     bool *refcounted)
417 {
418 	struct sock *sk = skb_steal_sock(skb, refcounted);
419 	const struct iphdr *iph = ip_hdr(skb);
420 
421 	if (sk)
422 		return sk;
423 
424 	return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo, skb,
425 			     doff, iph->saddr, sport,
426 			     iph->daddr, dport, inet_iif(skb), sdif,
427 			     refcounted);
428 }
429 
430 u32 inet6_ehashfn(const struct net *net,
431 		  const struct in6_addr *laddr, const u16 lport,
432 		  const struct in6_addr *faddr, const __be16 fport);
433 
434 static inline void sk_daddr_set(struct sock *sk, __be32 addr)
435 {
436 	sk->sk_daddr = addr; /* alias of inet_daddr */
437 #if IS_ENABLED(CONFIG_IPV6)
438 	ipv6_addr_set_v4mapped(addr, &sk->sk_v6_daddr);
439 #endif
440 }
441 
442 static inline void sk_rcv_saddr_set(struct sock *sk, __be32 addr)
443 {
444 	sk->sk_rcv_saddr = addr; /* alias of inet_rcv_saddr */
445 #if IS_ENABLED(CONFIG_IPV6)
446 	ipv6_addr_set_v4mapped(addr, &sk->sk_v6_rcv_saddr);
447 #endif
448 }
449 
450 int __inet_hash_connect(struct inet_timewait_death_row *death_row,
451 			struct sock *sk, u64 port_offset,
452 			int (*check_established)(struct inet_timewait_death_row *,
453 						 struct sock *, __u16,
454 						 struct inet_timewait_sock **));
455 
456 int inet_hash_connect(struct inet_timewait_death_row *death_row,
457 		      struct sock *sk);
458 #endif /* _INET_HASHTABLES_H */
459