xref: /linux/include/net/inet_hashtables.h (revision 320fefa9e2edc67011e235ea1d50f0d00ddfe004)
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  */
39 struct 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 
77 struct 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 
94 struct 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 };
112 
113 static inline struct net *ib_net(const struct inet_bind_bucket *ib)
114 {
115 	return read_pnet(&ib->ib_net);
116 }
117 
118 static inline struct net *ib2_net(const struct inet_bind2_bucket *ib)
119 {
120 	return read_pnet(&ib->ib_net);
121 }
122 
123 #define inet_bind_bucket_for_each(tb, head) \
124 	hlist_for_each_entry(tb, head, node)
125 
126 struct inet_bind_hashbucket {
127 	spinlock_t		lock;
128 	struct hlist_head	chain;
129 };
130 
131 /* Sockets can be hashed in established or listening table.
132  * We must use different 'nulls' end-of-chain value for all hash buckets :
133  * A socket might transition from ESTABLISH to LISTEN state without
134  * RCU grace period. A lookup in ehash table needs to handle this case.
135  */
136 #define LISTENING_NULLS_BASE (1U << 29)
137 struct inet_listen_hashbucket {
138 	spinlock_t		lock;
139 	struct hlist_nulls_head	nulls_head;
140 };
141 
142 /* This is for listening sockets, thus all sockets which possess wildcards. */
143 #define INET_LHTABLE_SIZE	32	/* Yes, really, this is all you need. */
144 
145 struct inet_hashinfo {
146 	/* This is for sockets with full identity only.  Sockets here will
147 	 * always be without wildcards and will have the following invariant:
148 	 *
149 	 *          TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
150 	 *
151 	 */
152 	struct inet_ehash_bucket	*ehash;
153 	spinlock_t			*ehash_locks;
154 	unsigned int			ehash_mask;
155 	unsigned int			ehash_locks_mask;
156 
157 	/* Ok, let's try this, I give up, we do need a local binding
158 	 * TCP hash as well as the others for fast bind/connect.
159 	 */
160 	struct kmem_cache		*bind_bucket_cachep;
161 	/* This bind table is hashed by local port */
162 	struct inet_bind_hashbucket	*bhash;
163 	struct kmem_cache		*bind2_bucket_cachep;
164 	/* This bind table is hashed by local port and sk->sk_rcv_saddr (ipv4)
165 	 * or sk->sk_v6_rcv_saddr (ipv6). This 2nd bind table is used
166 	 * primarily for expediting bind conflict resolution.
167 	 */
168 	struct inet_bind_hashbucket	*bhash2;
169 	unsigned int			bhash_size;
170 
171 	/* The 2nd listener table hashed by local port and address */
172 	unsigned int			lhash2_mask;
173 	struct inet_listen_hashbucket	*lhash2;
174 
175 	bool				pernet;
176 };
177 
178 static inline struct inet_hashinfo *tcp_or_dccp_get_hashinfo(const struct sock *sk)
179 {
180 #if IS_ENABLED(CONFIG_IP_DCCP)
181 	return sk->sk_prot->h.hashinfo ? :
182 		sock_net(sk)->ipv4.tcp_death_row.hashinfo;
183 #else
184 	return sock_net(sk)->ipv4.tcp_death_row.hashinfo;
185 #endif
186 }
187 
188 static inline struct inet_listen_hashbucket *
189 inet_lhash2_bucket(struct inet_hashinfo *h, u32 hash)
190 {
191 	return &h->lhash2[hash & h->lhash2_mask];
192 }
193 
194 static inline struct inet_ehash_bucket *inet_ehash_bucket(
195 	struct inet_hashinfo *hashinfo,
196 	unsigned int hash)
197 {
198 	return &hashinfo->ehash[hash & hashinfo->ehash_mask];
199 }
200 
201 static inline spinlock_t *inet_ehash_lockp(
202 	struct inet_hashinfo *hashinfo,
203 	unsigned int hash)
204 {
205 	return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
206 }
207 
208 int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo);
209 
210 static inline void inet_hashinfo2_free_mod(struct inet_hashinfo *h)
211 {
212 	kfree(h->lhash2);
213 	h->lhash2 = NULL;
214 }
215 
216 static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
217 {
218 	kvfree(hashinfo->ehash_locks);
219 	hashinfo->ehash_locks = NULL;
220 }
221 
222 struct inet_hashinfo *inet_pernet_hashinfo_alloc(struct inet_hashinfo *hashinfo,
223 						 unsigned int ehash_entries);
224 void inet_pernet_hashinfo_free(struct inet_hashinfo *hashinfo);
225 
226 struct inet_bind_bucket *
227 inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net,
228 			struct inet_bind_hashbucket *head,
229 			const unsigned short snum, int l3mdev);
230 void inet_bind_bucket_destroy(struct kmem_cache *cachep,
231 			      struct inet_bind_bucket *tb);
232 
233 bool inet_bind_bucket_match(const struct inet_bind_bucket *tb,
234 			    const struct net *net, unsigned short port,
235 			    int l3mdev);
236 
237 struct inet_bind2_bucket *
238 inet_bind2_bucket_create(struct kmem_cache *cachep, struct net *net,
239 			 struct inet_bind_hashbucket *head,
240 			 unsigned short port, int l3mdev,
241 			 const struct sock *sk);
242 
243 void inet_bind2_bucket_destroy(struct kmem_cache *cachep,
244 			       struct inet_bind2_bucket *tb);
245 
246 struct inet_bind2_bucket *
247 inet_bind2_bucket_find(const struct inet_bind_hashbucket *head,
248 		       const struct net *net,
249 		       unsigned short port, int l3mdev,
250 		       const struct sock *sk);
251 
252 bool inet_bind2_bucket_match_addr_any(const struct inet_bind2_bucket *tb,
253 				      const struct net *net, unsigned short port,
254 				      int l3mdev, const struct sock *sk);
255 
256 static inline u32 inet_bhashfn(const struct net *net, const __u16 lport,
257 			       const u32 bhash_size)
258 {
259 	return (lport + net_hash_mix(net)) & (bhash_size - 1);
260 }
261 
262 static inline struct inet_bind_hashbucket *
263 inet_bhashfn_portaddr(const struct inet_hashinfo *hinfo, const struct sock *sk,
264 		      const struct net *net, unsigned short port)
265 {
266 	u32 hash;
267 
268 #if IS_ENABLED(CONFIG_IPV6)
269 	if (sk->sk_family == AF_INET6)
270 		hash = ipv6_portaddr_hash(net, &sk->sk_v6_rcv_saddr, port);
271 	else
272 #endif
273 		hash = ipv4_portaddr_hash(net, sk->sk_rcv_saddr, port);
274 	return &hinfo->bhash2[hash & (hinfo->bhash_size - 1)];
275 }
276 
277 struct inet_bind_hashbucket *
278 inet_bhash2_addr_any_hashbucket(const struct sock *sk, const struct net *net, int port);
279 
280 /* This should be called whenever a socket's sk_rcv_saddr (ipv4) or
281  * sk_v6_rcv_saddr (ipv6) changes after it has been binded. The socket's
282  * rcv_saddr field should already have been updated when this is called.
283  */
284 int inet_bhash2_update_saddr(struct sock *sk, void *saddr, int family);
285 void inet_bhash2_reset_saddr(struct sock *sk);
286 
287 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
288 		    struct inet_bind2_bucket *tb2, unsigned short port);
289 
290 /* Caller must disable local BH processing. */
291 int __inet_inherit_port(const struct sock *sk, struct sock *child);
292 
293 void inet_put_port(struct sock *sk);
294 
295 void inet_hashinfo2_init(struct inet_hashinfo *h, const char *name,
296 			 unsigned long numentries, int scale,
297 			 unsigned long low_limit,
298 			 unsigned long high_limit);
299 int inet_hashinfo2_init_mod(struct inet_hashinfo *h);
300 
301 bool inet_ehash_insert(struct sock *sk, struct sock *osk, bool *found_dup_sk);
302 bool inet_ehash_nolisten(struct sock *sk, struct sock *osk,
303 			 bool *found_dup_sk);
304 int __inet_hash(struct sock *sk, struct sock *osk);
305 int inet_hash(struct sock *sk);
306 void inet_unhash(struct sock *sk);
307 
308 struct sock *__inet_lookup_listener(struct net *net,
309 				    struct inet_hashinfo *hashinfo,
310 				    struct sk_buff *skb, int doff,
311 				    const __be32 saddr, const __be16 sport,
312 				    const __be32 daddr,
313 				    const unsigned short hnum,
314 				    const int dif, const int sdif);
315 
316 static inline struct sock *inet_lookup_listener(struct net *net,
317 		struct inet_hashinfo *hashinfo,
318 		struct sk_buff *skb, int doff,
319 		__be32 saddr, __be16 sport,
320 		__be32 daddr, __be16 dport, int dif, int sdif)
321 {
322 	return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, sport,
323 				      daddr, ntohs(dport), dif, sdif);
324 }
325 
326 /* Socket demux engine toys. */
327 /* What happens here is ugly; there's a pair of adjacent fields in
328    struct inet_sock; __be16 dport followed by __u16 num.  We want to
329    search by pair, so we combine the keys into a single 32bit value
330    and compare with 32bit value read from &...->dport.  Let's at least
331    make sure that it's not mixed with anything else...
332    On 64bit targets we combine comparisons with pair of adjacent __be32
333    fields in the same way.
334 */
335 #ifdef __BIG_ENDIAN
336 #define INET_COMBINED_PORTS(__sport, __dport) \
337 	((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
338 #else /* __LITTLE_ENDIAN */
339 #define INET_COMBINED_PORTS(__sport, __dport) \
340 	((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
341 #endif
342 
343 #ifdef __BIG_ENDIAN
344 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
345 	const __addrpair __name = (__force __addrpair) ( \
346 				   (((__force __u64)(__be32)(__saddr)) << 32) | \
347 				   ((__force __u64)(__be32)(__daddr)))
348 #else /* __LITTLE_ENDIAN */
349 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
350 	const __addrpair __name = (__force __addrpair) ( \
351 				   (((__force __u64)(__be32)(__daddr)) << 32) | \
352 				   ((__force __u64)(__be32)(__saddr)))
353 #endif /* __BIG_ENDIAN */
354 
355 static inline bool inet_match(struct net *net, const struct sock *sk,
356 			      const __addrpair cookie, const __portpair ports,
357 			      int dif, int sdif)
358 {
359 	if (!net_eq(sock_net(sk), net) ||
360 	    sk->sk_portpair != ports ||
361 	    sk->sk_addrpair != cookie)
362 	        return false;
363 
364 	/* READ_ONCE() paired with WRITE_ONCE() in sock_bindtoindex_locked() */
365 	return inet_sk_bound_dev_eq(net, READ_ONCE(sk->sk_bound_dev_if), dif,
366 				    sdif);
367 }
368 
369 /* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
370  * not check it for lookups anymore, thanks Alexey. -DaveM
371  */
372 struct sock *__inet_lookup_established(struct net *net,
373 				       struct inet_hashinfo *hashinfo,
374 				       const __be32 saddr, const __be16 sport,
375 				       const __be32 daddr, const u16 hnum,
376 				       const int dif, const int sdif);
377 
378 static inline struct sock *
379 	inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
380 				const __be32 saddr, const __be16 sport,
381 				const __be32 daddr, const __be16 dport,
382 				const int dif)
383 {
384 	return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
385 					 ntohs(dport), dif, 0);
386 }
387 
388 static inline struct sock *__inet_lookup(struct net *net,
389 					 struct inet_hashinfo *hashinfo,
390 					 struct sk_buff *skb, int doff,
391 					 const __be32 saddr, const __be16 sport,
392 					 const __be32 daddr, const __be16 dport,
393 					 const int dif, const int sdif,
394 					 bool *refcounted)
395 {
396 	u16 hnum = ntohs(dport);
397 	struct sock *sk;
398 
399 	sk = __inet_lookup_established(net, hashinfo, saddr, sport,
400 				       daddr, hnum, dif, sdif);
401 	*refcounted = true;
402 	if (sk)
403 		return sk;
404 	*refcounted = false;
405 	return __inet_lookup_listener(net, hashinfo, skb, doff, saddr,
406 				      sport, daddr, hnum, dif, sdif);
407 }
408 
409 static inline struct sock *inet_lookup(struct net *net,
410 				       struct inet_hashinfo *hashinfo,
411 				       struct sk_buff *skb, int doff,
412 				       const __be32 saddr, const __be16 sport,
413 				       const __be32 daddr, const __be16 dport,
414 				       const int dif)
415 {
416 	struct sock *sk;
417 	bool refcounted;
418 
419 	sk = __inet_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
420 			   dport, dif, 0, &refcounted);
421 
422 	if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
423 		sk = NULL;
424 	return sk;
425 }
426 
427 static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
428 					     struct sk_buff *skb,
429 					     int doff,
430 					     const __be16 sport,
431 					     const __be16 dport,
432 					     const int sdif,
433 					     bool *refcounted)
434 {
435 	struct sock *sk = skb_steal_sock(skb, refcounted);
436 	const struct iphdr *iph = ip_hdr(skb);
437 
438 	if (sk)
439 		return sk;
440 
441 	return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo, skb,
442 			     doff, iph->saddr, sport,
443 			     iph->daddr, dport, inet_iif(skb), sdif,
444 			     refcounted);
445 }
446 
447 u32 inet6_ehashfn(const struct net *net,
448 		  const struct in6_addr *laddr, const u16 lport,
449 		  const struct in6_addr *faddr, const __be16 fport);
450 
451 static inline void sk_daddr_set(struct sock *sk, __be32 addr)
452 {
453 	sk->sk_daddr = addr; /* alias of inet_daddr */
454 #if IS_ENABLED(CONFIG_IPV6)
455 	ipv6_addr_set_v4mapped(addr, &sk->sk_v6_daddr);
456 #endif
457 }
458 
459 static inline void sk_rcv_saddr_set(struct sock *sk, __be32 addr)
460 {
461 	sk->sk_rcv_saddr = addr; /* alias of inet_rcv_saddr */
462 #if IS_ENABLED(CONFIG_IPV6)
463 	ipv6_addr_set_v4mapped(addr, &sk->sk_v6_rcv_saddr);
464 #endif
465 }
466 
467 int __inet_hash_connect(struct inet_timewait_death_row *death_row,
468 			struct sock *sk, u64 port_offset,
469 			int (*check_established)(struct inet_timewait_death_row *,
470 						 struct sock *, __u16,
471 						 struct inet_timewait_sock **));
472 
473 int inet_hash_connect(struct inet_timewait_death_row *death_row,
474 		      struct sock *sk);
475 #endif /* _INET_HASHTABLES_H */
476