xref: /linux/include/net/inet_hashtables.h (revision 21ab2078ff37ad9f81c080244ed51d80463d58d9)
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 	union {
99 #if IS_ENABLED(CONFIG_IPV6)
100 		struct in6_addr		v6_rcv_saddr;
101 #endif
102 		__be32			rcv_saddr;
103 	};
104 	/* Node in the bhash2 inet_bind_hashbucket chain */
105 	struct hlist_node	node;
106 	/* List of sockets hashed to this bucket */
107 	struct hlist_head	owners;
108 };
109 
110 static inline struct net *ib_net(const struct inet_bind_bucket *ib)
111 {
112 	return read_pnet(&ib->ib_net);
113 }
114 
115 static inline struct net *ib2_net(const struct inet_bind2_bucket *ib)
116 {
117 	return read_pnet(&ib->ib_net);
118 }
119 
120 #define inet_bind_bucket_for_each(tb, head) \
121 	hlist_for_each_entry(tb, head, node)
122 
123 struct inet_bind_hashbucket {
124 	spinlock_t		lock;
125 	struct hlist_head	chain;
126 };
127 
128 /* Sockets can be hashed in established or listening table.
129  * We must use different 'nulls' end-of-chain value for all hash buckets :
130  * A socket might transition from ESTABLISH to LISTEN state without
131  * RCU grace period. A lookup in ehash table needs to handle this case.
132  */
133 #define LISTENING_NULLS_BASE (1U << 29)
134 struct inet_listen_hashbucket {
135 	spinlock_t		lock;
136 	struct hlist_nulls_head	nulls_head;
137 };
138 
139 /* This is for listening sockets, thus all sockets which possess wildcards. */
140 #define INET_LHTABLE_SIZE	32	/* Yes, really, this is all you need. */
141 
142 struct inet_hashinfo {
143 	/* This is for sockets with full identity only.  Sockets here will
144 	 * always be without wildcards and will have the following invariant:
145 	 *
146 	 *          TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
147 	 *
148 	 */
149 	struct inet_ehash_bucket	*ehash;
150 	spinlock_t			*ehash_locks;
151 	unsigned int			ehash_mask;
152 	unsigned int			ehash_locks_mask;
153 
154 	/* Ok, let's try this, I give up, we do need a local binding
155 	 * TCP hash as well as the others for fast bind/connect.
156 	 */
157 	struct kmem_cache		*bind_bucket_cachep;
158 	/* This bind table is hashed by local port */
159 	struct inet_bind_hashbucket	*bhash;
160 	struct kmem_cache		*bind2_bucket_cachep;
161 	/* This bind table is hashed by local port and sk->sk_rcv_saddr (ipv4)
162 	 * or sk->sk_v6_rcv_saddr (ipv6). This 2nd bind table is used
163 	 * primarily for expediting bind conflict resolution.
164 	 */
165 	struct inet_bind_hashbucket	*bhash2;
166 	unsigned int			bhash_size;
167 
168 	/* The 2nd listener table hashed by local port and address */
169 	unsigned int			lhash2_mask;
170 	struct inet_listen_hashbucket	*lhash2;
171 
172 	bool				pernet;
173 };
174 
175 static inline struct inet_hashinfo *tcp_or_dccp_get_hashinfo(const struct sock *sk)
176 {
177 #if IS_ENABLED(CONFIG_IP_DCCP)
178 	return sk->sk_prot->h.hashinfo ? :
179 		sock_net(sk)->ipv4.tcp_death_row.hashinfo;
180 #else
181 	return sock_net(sk)->ipv4.tcp_death_row.hashinfo;
182 #endif
183 }
184 
185 static inline struct inet_listen_hashbucket *
186 inet_lhash2_bucket(struct inet_hashinfo *h, u32 hash)
187 {
188 	return &h->lhash2[hash & h->lhash2_mask];
189 }
190 
191 static inline struct inet_ehash_bucket *inet_ehash_bucket(
192 	struct inet_hashinfo *hashinfo,
193 	unsigned int hash)
194 {
195 	return &hashinfo->ehash[hash & hashinfo->ehash_mask];
196 }
197 
198 static inline spinlock_t *inet_ehash_lockp(
199 	struct inet_hashinfo *hashinfo,
200 	unsigned int hash)
201 {
202 	return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
203 }
204 
205 int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo);
206 
207 static inline void inet_hashinfo2_free_mod(struct inet_hashinfo *h)
208 {
209 	kfree(h->lhash2);
210 	h->lhash2 = NULL;
211 }
212 
213 static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
214 {
215 	kvfree(hashinfo->ehash_locks);
216 	hashinfo->ehash_locks = NULL;
217 }
218 
219 struct inet_hashinfo *inet_pernet_hashinfo_alloc(struct inet_hashinfo *hashinfo,
220 						 unsigned int ehash_entries);
221 void inet_pernet_hashinfo_free(struct inet_hashinfo *hashinfo);
222 
223 struct inet_bind_bucket *
224 inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net,
225 			struct inet_bind_hashbucket *head,
226 			const unsigned short snum, int l3mdev);
227 void inet_bind_bucket_destroy(struct kmem_cache *cachep,
228 			      struct inet_bind_bucket *tb);
229 
230 bool inet_bind_bucket_match(const struct inet_bind_bucket *tb,
231 			    const struct net *net, unsigned short port,
232 			    int l3mdev);
233 
234 struct inet_bind2_bucket *
235 inet_bind2_bucket_create(struct kmem_cache *cachep, struct net *net,
236 			 struct inet_bind_hashbucket *head,
237 			 unsigned short port, int l3mdev,
238 			 const struct sock *sk);
239 
240 void inet_bind2_bucket_destroy(struct kmem_cache *cachep,
241 			       struct inet_bind2_bucket *tb);
242 
243 struct inet_bind2_bucket *
244 inet_bind2_bucket_find(const struct inet_bind_hashbucket *head,
245 		       const struct net *net,
246 		       unsigned short port, int l3mdev,
247 		       const struct sock *sk);
248 
249 bool inet_bind2_bucket_match_addr_any(const struct inet_bind2_bucket *tb,
250 				      const struct net *net, unsigned short port,
251 				      int l3mdev, const struct sock *sk);
252 
253 static inline u32 inet_bhashfn(const struct net *net, const __u16 lport,
254 			       const u32 bhash_size)
255 {
256 	return (lport + net_hash_mix(net)) & (bhash_size - 1);
257 }
258 
259 static inline struct inet_bind_hashbucket *
260 inet_bhashfn_portaddr(const struct inet_hashinfo *hinfo, const struct sock *sk,
261 		      const struct net *net, unsigned short port)
262 {
263 	u32 hash;
264 
265 #if IS_ENABLED(CONFIG_IPV6)
266 	if (sk->sk_family == AF_INET6)
267 		hash = ipv6_portaddr_hash(net, &sk->sk_v6_rcv_saddr, port);
268 	else
269 #endif
270 		hash = ipv4_portaddr_hash(net, sk->sk_rcv_saddr, port);
271 	return &hinfo->bhash2[hash & (hinfo->bhash_size - 1)];
272 }
273 
274 struct inet_bind_hashbucket *
275 inet_bhash2_addr_any_hashbucket(const struct sock *sk, const struct net *net, int port);
276 
277 /* This should be called whenever a socket's sk_rcv_saddr (ipv4) or
278  * sk_v6_rcv_saddr (ipv6) changes after it has been binded. The socket's
279  * rcv_saddr field should already have been updated when this is called.
280  */
281 int inet_bhash2_update_saddr(struct inet_bind_hashbucket *prev_saddr, struct sock *sk);
282 
283 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
284 		    struct inet_bind2_bucket *tb2, unsigned short port);
285 
286 /* Caller must disable local BH processing. */
287 int __inet_inherit_port(const struct sock *sk, struct sock *child);
288 
289 void inet_put_port(struct sock *sk);
290 
291 void inet_hashinfo2_init(struct inet_hashinfo *h, const char *name,
292 			 unsigned long numentries, int scale,
293 			 unsigned long low_limit,
294 			 unsigned long high_limit);
295 int inet_hashinfo2_init_mod(struct inet_hashinfo *h);
296 
297 bool inet_ehash_insert(struct sock *sk, struct sock *osk, bool *found_dup_sk);
298 bool inet_ehash_nolisten(struct sock *sk, struct sock *osk,
299 			 bool *found_dup_sk);
300 int __inet_hash(struct sock *sk, struct sock *osk);
301 int inet_hash(struct sock *sk);
302 void inet_unhash(struct sock *sk);
303 
304 struct sock *__inet_lookup_listener(struct net *net,
305 				    struct inet_hashinfo *hashinfo,
306 				    struct sk_buff *skb, int doff,
307 				    const __be32 saddr, const __be16 sport,
308 				    const __be32 daddr,
309 				    const unsigned short hnum,
310 				    const int dif, const int sdif);
311 
312 static inline struct sock *inet_lookup_listener(struct net *net,
313 		struct inet_hashinfo *hashinfo,
314 		struct sk_buff *skb, int doff,
315 		__be32 saddr, __be16 sport,
316 		__be32 daddr, __be16 dport, int dif, int sdif)
317 {
318 	return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, sport,
319 				      daddr, ntohs(dport), dif, sdif);
320 }
321 
322 /* Socket demux engine toys. */
323 /* What happens here is ugly; there's a pair of adjacent fields in
324    struct inet_sock; __be16 dport followed by __u16 num.  We want to
325    search by pair, so we combine the keys into a single 32bit value
326    and compare with 32bit value read from &...->dport.  Let's at least
327    make sure that it's not mixed with anything else...
328    On 64bit targets we combine comparisons with pair of adjacent __be32
329    fields in the same way.
330 */
331 #ifdef __BIG_ENDIAN
332 #define INET_COMBINED_PORTS(__sport, __dport) \
333 	((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
334 #else /* __LITTLE_ENDIAN */
335 #define INET_COMBINED_PORTS(__sport, __dport) \
336 	((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
337 #endif
338 
339 #ifdef __BIG_ENDIAN
340 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
341 	const __addrpair __name = (__force __addrpair) ( \
342 				   (((__force __u64)(__be32)(__saddr)) << 32) | \
343 				   ((__force __u64)(__be32)(__daddr)))
344 #else /* __LITTLE_ENDIAN */
345 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
346 	const __addrpair __name = (__force __addrpair) ( \
347 				   (((__force __u64)(__be32)(__daddr)) << 32) | \
348 				   ((__force __u64)(__be32)(__saddr)))
349 #endif /* __BIG_ENDIAN */
350 
351 static inline bool inet_match(struct net *net, const struct sock *sk,
352 			      const __addrpair cookie, const __portpair ports,
353 			      int dif, int sdif)
354 {
355 	if (!net_eq(sock_net(sk), net) ||
356 	    sk->sk_portpair != ports ||
357 	    sk->sk_addrpair != cookie)
358 	        return false;
359 
360 	/* READ_ONCE() paired with WRITE_ONCE() in sock_bindtoindex_locked() */
361 	return inet_sk_bound_dev_eq(net, READ_ONCE(sk->sk_bound_dev_if), dif,
362 				    sdif);
363 }
364 
365 /* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
366  * not check it for lookups anymore, thanks Alexey. -DaveM
367  */
368 struct sock *__inet_lookup_established(struct net *net,
369 				       struct inet_hashinfo *hashinfo,
370 				       const __be32 saddr, const __be16 sport,
371 				       const __be32 daddr, const u16 hnum,
372 				       const int dif, const int sdif);
373 
374 static inline struct sock *
375 	inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
376 				const __be32 saddr, const __be16 sport,
377 				const __be32 daddr, const __be16 dport,
378 				const int dif)
379 {
380 	return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
381 					 ntohs(dport), dif, 0);
382 }
383 
384 static inline struct sock *__inet_lookup(struct net *net,
385 					 struct inet_hashinfo *hashinfo,
386 					 struct sk_buff *skb, int doff,
387 					 const __be32 saddr, const __be16 sport,
388 					 const __be32 daddr, const __be16 dport,
389 					 const int dif, const int sdif,
390 					 bool *refcounted)
391 {
392 	u16 hnum = ntohs(dport);
393 	struct sock *sk;
394 
395 	sk = __inet_lookup_established(net, hashinfo, saddr, sport,
396 				       daddr, hnum, dif, sdif);
397 	*refcounted = true;
398 	if (sk)
399 		return sk;
400 	*refcounted = false;
401 	return __inet_lookup_listener(net, hashinfo, skb, doff, saddr,
402 				      sport, daddr, hnum, dif, sdif);
403 }
404 
405 static inline struct sock *inet_lookup(struct net *net,
406 				       struct inet_hashinfo *hashinfo,
407 				       struct sk_buff *skb, int doff,
408 				       const __be32 saddr, const __be16 sport,
409 				       const __be32 daddr, const __be16 dport,
410 				       const int dif)
411 {
412 	struct sock *sk;
413 	bool refcounted;
414 
415 	sk = __inet_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
416 			   dport, dif, 0, &refcounted);
417 
418 	if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
419 		sk = NULL;
420 	return sk;
421 }
422 
423 static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
424 					     struct sk_buff *skb,
425 					     int doff,
426 					     const __be16 sport,
427 					     const __be16 dport,
428 					     const int sdif,
429 					     bool *refcounted)
430 {
431 	struct sock *sk = skb_steal_sock(skb, refcounted);
432 	const struct iphdr *iph = ip_hdr(skb);
433 
434 	if (sk)
435 		return sk;
436 
437 	return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo, skb,
438 			     doff, iph->saddr, sport,
439 			     iph->daddr, dport, inet_iif(skb), sdif,
440 			     refcounted);
441 }
442 
443 u32 inet6_ehashfn(const struct net *net,
444 		  const struct in6_addr *laddr, const u16 lport,
445 		  const struct in6_addr *faddr, const __be16 fport);
446 
447 static inline void sk_daddr_set(struct sock *sk, __be32 addr)
448 {
449 	sk->sk_daddr = addr; /* alias of inet_daddr */
450 #if IS_ENABLED(CONFIG_IPV6)
451 	ipv6_addr_set_v4mapped(addr, &sk->sk_v6_daddr);
452 #endif
453 }
454 
455 static inline void sk_rcv_saddr_set(struct sock *sk, __be32 addr)
456 {
457 	sk->sk_rcv_saddr = addr; /* alias of inet_rcv_saddr */
458 #if IS_ENABLED(CONFIG_IPV6)
459 	ipv6_addr_set_v4mapped(addr, &sk->sk_v6_rcv_saddr);
460 #endif
461 }
462 
463 int __inet_hash_connect(struct inet_timewait_death_row *death_row,
464 			struct sock *sk, u64 port_offset,
465 			int (*check_established)(struct inet_timewait_death_row *,
466 						 struct sock *, __u16,
467 						 struct inet_timewait_sock **));
468 
469 int inet_hash_connect(struct inet_timewait_death_row *death_row,
470 		      struct sock *sk);
471 #endif /* _INET_HASHTABLES_H */
472