xref: /linux/include/net/inet_hashtables.h (revision 5ce42b5de461c3154f61a023b191dd6b77ee66c0)
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	bhash2;
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		addr_type;
100 	struct in6_addr		v6_rcv_saddr;
101 #define rcv_saddr		v6_rcv_saddr.s6_addr32[3]
102 #else
103 	__be32			rcv_saddr;
104 #endif
105 	/* Node in the bhash2 inet_bind_hashbucket chain */
106 	struct hlist_node	node;
107 	struct hlist_node	bhash_node;
108 	/* List of sockets hashed to this bucket */
109 	struct hlist_head	owners;
110 };
111 
112 static inline struct net *ib_net(const struct inet_bind_bucket *ib)
113 {
114 	return read_pnet(&ib->ib_net);
115 }
116 
117 static inline struct net *ib2_net(const struct inet_bind2_bucket *ib)
118 {
119 	return read_pnet(&ib->ib_net);
120 }
121 
122 #define inet_bind_bucket_for_each(tb, head) \
123 	hlist_for_each_entry(tb, head, node)
124 
125 struct inet_bind_hashbucket {
126 	spinlock_t		lock;
127 	struct hlist_head	chain;
128 };
129 
130 /* Sockets can be hashed in established or listening table.
131  * We must use different 'nulls' end-of-chain value for all hash buckets :
132  * A socket might transition from ESTABLISH to LISTEN state without
133  * RCU grace period. A lookup in ehash table needs to handle this case.
134  */
135 #define LISTENING_NULLS_BASE (1U << 29)
136 struct inet_listen_hashbucket {
137 	spinlock_t		lock;
138 	struct hlist_nulls_head	nulls_head;
139 };
140 
141 /* This is for listening sockets, thus all sockets which possess wildcards. */
142 #define INET_LHTABLE_SIZE	32	/* Yes, really, this is all you need. */
143 
144 struct inet_hashinfo {
145 	/* This is for sockets with full identity only.  Sockets here will
146 	 * always be without wildcards and will have the following invariant:
147 	 *
148 	 *          TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
149 	 *
150 	 */
151 	struct inet_ehash_bucket	*ehash;
152 	spinlock_t			*ehash_locks;
153 	unsigned int			ehash_mask;
154 	unsigned int			ehash_locks_mask;
155 
156 	/* Ok, let's try this, I give up, we do need a local binding
157 	 * TCP hash as well as the others for fast bind/connect.
158 	 */
159 	struct kmem_cache		*bind_bucket_cachep;
160 	/* This bind table is hashed by local port */
161 	struct inet_bind_hashbucket	*bhash;
162 	struct kmem_cache		*bind2_bucket_cachep;
163 	/* This bind table is hashed by local port and sk->sk_rcv_saddr (ipv4)
164 	 * or sk->sk_v6_rcv_saddr (ipv6). This 2nd bind table is used
165 	 * primarily for expediting bind conflict resolution.
166 	 */
167 	struct inet_bind_hashbucket	*bhash2;
168 	unsigned int			bhash_size;
169 
170 	/* The 2nd listener table hashed by local port and address */
171 	unsigned int			lhash2_mask;
172 	struct inet_listen_hashbucket	*lhash2;
173 
174 	bool				pernet;
175 } ____cacheline_aligned_in_smp;
176 
177 static inline struct inet_hashinfo *tcp_or_dccp_get_hashinfo(const struct sock *sk)
178 {
179 #if IS_ENABLED(CONFIG_IP_DCCP)
180 	return sk->sk_prot->h.hashinfo ? :
181 		sock_net(sk)->ipv4.tcp_death_row.hashinfo;
182 #else
183 	return sock_net(sk)->ipv4.tcp_death_row.hashinfo;
184 #endif
185 }
186 
187 static inline struct inet_listen_hashbucket *
188 inet_lhash2_bucket(struct inet_hashinfo *h, u32 hash)
189 {
190 	return &h->lhash2[hash & h->lhash2_mask];
191 }
192 
193 static inline struct inet_ehash_bucket *inet_ehash_bucket(
194 	struct inet_hashinfo *hashinfo,
195 	unsigned int hash)
196 {
197 	return &hashinfo->ehash[hash & hashinfo->ehash_mask];
198 }
199 
200 static inline spinlock_t *inet_ehash_lockp(
201 	struct inet_hashinfo *hashinfo,
202 	unsigned int hash)
203 {
204 	return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
205 }
206 
207 int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo);
208 
209 static inline void inet_hashinfo2_free_mod(struct inet_hashinfo *h)
210 {
211 	kfree(h->lhash2);
212 	h->lhash2 = NULL;
213 }
214 
215 static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
216 {
217 	kvfree(hashinfo->ehash_locks);
218 	hashinfo->ehash_locks = NULL;
219 }
220 
221 struct inet_hashinfo *inet_pernet_hashinfo_alloc(struct inet_hashinfo *hashinfo,
222 						 unsigned int ehash_entries);
223 void inet_pernet_hashinfo_free(struct inet_hashinfo *hashinfo);
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 bool inet_bind_bucket_match(const struct inet_bind_bucket *tb,
233 			    const struct net *net, unsigned short port,
234 			    int l3mdev);
235 
236 struct inet_bind2_bucket *
237 inet_bind2_bucket_create(struct kmem_cache *cachep, struct net *net,
238 			 struct inet_bind_hashbucket *head,
239 			 struct inet_bind_bucket *tb,
240 			 const struct sock *sk);
241 
242 void inet_bind2_bucket_destroy(struct kmem_cache *cachep,
243 			       struct inet_bind2_bucket *tb);
244 
245 struct inet_bind2_bucket *
246 inet_bind2_bucket_find(const struct inet_bind_hashbucket *head,
247 		       const struct net *net,
248 		       unsigned short port, int l3mdev,
249 		       const struct sock *sk);
250 
251 bool inet_bind2_bucket_match_addr_any(const struct inet_bind2_bucket *tb,
252 				      const struct net *net, unsigned short port,
253 				      int l3mdev, const struct sock *sk);
254 
255 static inline u32 inet_bhashfn(const struct net *net, const __u16 lport,
256 			       const u32 bhash_size)
257 {
258 	return (lport + net_hash_mix(net)) & (bhash_size - 1);
259 }
260 
261 static inline struct inet_bind_hashbucket *
262 inet_bhashfn_portaddr(const struct inet_hashinfo *hinfo, const struct sock *sk,
263 		      const struct net *net, unsigned short port)
264 {
265 	u32 hash;
266 
267 #if IS_ENABLED(CONFIG_IPV6)
268 	if (sk->sk_family == AF_INET6)
269 		hash = ipv6_portaddr_hash(net, &sk->sk_v6_rcv_saddr, port);
270 	else
271 #endif
272 		hash = ipv4_portaddr_hash(net, sk->sk_rcv_saddr, port);
273 	return &hinfo->bhash2[hash & (hinfo->bhash_size - 1)];
274 }
275 
276 struct inet_bind_hashbucket *
277 inet_bhash2_addr_any_hashbucket(const struct sock *sk, const struct net *net, int port);
278 
279 /* This should be called whenever a socket's sk_rcv_saddr (ipv4) or
280  * sk_v6_rcv_saddr (ipv6) changes after it has been binded. The socket's
281  * rcv_saddr field should already have been updated when this is called.
282  */
283 int inet_bhash2_update_saddr(struct sock *sk, void *saddr, int family);
284 void inet_bhash2_reset_saddr(struct sock *sk);
285 
286 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
287 		    struct inet_bind2_bucket *tb2, unsigned short port);
288 
289 /* Caller must disable local BH processing. */
290 int __inet_inherit_port(const struct sock *sk, struct sock *child);
291 
292 void inet_put_port(struct sock *sk);
293 
294 void inet_hashinfo2_init(struct inet_hashinfo *h, const char *name,
295 			 unsigned long numentries, int scale,
296 			 unsigned long low_limit,
297 			 unsigned long high_limit);
298 int inet_hashinfo2_init_mod(struct inet_hashinfo *h);
299 
300 bool inet_ehash_insert(struct sock *sk, struct sock *osk, bool *found_dup_sk);
301 bool inet_ehash_nolisten(struct sock *sk, struct sock *osk,
302 			 bool *found_dup_sk);
303 int __inet_hash(struct sock *sk, struct sock *osk);
304 int inet_hash(struct sock *sk);
305 void inet_unhash(struct sock *sk);
306 
307 struct sock *__inet_lookup_listener(const struct net *net,
308 				    struct inet_hashinfo *hashinfo,
309 				    struct sk_buff *skb, int doff,
310 				    const __be32 saddr, const __be16 sport,
311 				    const __be32 daddr,
312 				    const unsigned short hnum,
313 				    const int dif, const int sdif);
314 
315 static inline struct sock *inet_lookup_listener(struct net *net,
316 		struct inet_hashinfo *hashinfo,
317 		struct sk_buff *skb, int doff,
318 		__be32 saddr, __be16 sport,
319 		__be32 daddr, __be16 dport, int dif, int sdif)
320 {
321 	return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, sport,
322 				      daddr, ntohs(dport), dif, sdif);
323 }
324 
325 /* Socket demux engine toys. */
326 /* What happens here is ugly; there's a pair of adjacent fields in
327    struct inet_sock; __be16 dport followed by __u16 num.  We want to
328    search by pair, so we combine the keys into a single 32bit value
329    and compare with 32bit value read from &...->dport.  Let's at least
330    make sure that it's not mixed with anything else...
331    On 64bit targets we combine comparisons with pair of adjacent __be32
332    fields in the same way.
333 */
334 #ifdef __BIG_ENDIAN
335 #define INET_COMBINED_PORTS(__sport, __dport) \
336 	((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
337 #else /* __LITTLE_ENDIAN */
338 #define INET_COMBINED_PORTS(__sport, __dport) \
339 	((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
340 #endif
341 
342 #ifdef __BIG_ENDIAN
343 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
344 	const __addrpair __name = (__force __addrpair) ( \
345 				   (((__force __u64)(__be32)(__saddr)) << 32) | \
346 				   ((__force __u64)(__be32)(__daddr)))
347 #else /* __LITTLE_ENDIAN */
348 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
349 	const __addrpair __name = (__force __addrpair) ( \
350 				   (((__force __u64)(__be32)(__daddr)) << 32) | \
351 				   ((__force __u64)(__be32)(__saddr)))
352 #endif /* __BIG_ENDIAN */
353 
354 static inline bool inet_match(const struct net *net, const struct sock *sk,
355 			      const __addrpair cookie, const __portpair ports,
356 			      int dif, int sdif)
357 {
358 	if (!net_eq(sock_net(sk), net) ||
359 	    sk->sk_portpair != ports ||
360 	    sk->sk_addrpair != cookie)
361 	        return false;
362 
363 	/* READ_ONCE() paired with WRITE_ONCE() in sock_bindtoindex_locked() */
364 	return inet_sk_bound_dev_eq(net, READ_ONCE(sk->sk_bound_dev_if), dif,
365 				    sdif);
366 }
367 
368 /* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
369  * not check it for lookups anymore, thanks Alexey. -DaveM
370  */
371 struct sock *__inet_lookup_established(const struct net *net,
372 				       struct inet_hashinfo *hashinfo,
373 				       const __be32 saddr, const __be16 sport,
374 				       const __be32 daddr, const u16 hnum,
375 				       const int dif, const int sdif);
376 
377 typedef u32 (inet_ehashfn_t)(const struct net *net,
378 			      const __be32 laddr, const __u16 lport,
379 			      const __be32 faddr, const __be16 fport);
380 
381 inet_ehashfn_t inet_ehashfn;
382 
383 INDIRECT_CALLABLE_DECLARE(inet_ehashfn_t udp_ehashfn);
384 
385 struct sock *inet_lookup_reuseport(const struct net *net, struct sock *sk,
386 				   struct sk_buff *skb, int doff,
387 				   __be32 saddr, __be16 sport,
388 				   __be32 daddr, unsigned short hnum,
389 				   inet_ehashfn_t *ehashfn);
390 
391 struct sock *inet_lookup_run_sk_lookup(const struct net *net,
392 				       int protocol,
393 				       struct sk_buff *skb, int doff,
394 				       __be32 saddr, __be16 sport,
395 				       __be32 daddr, u16 hnum, const int dif,
396 				       inet_ehashfn_t *ehashfn);
397 
398 static inline struct sock *
399 	inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
400 				const __be32 saddr, const __be16 sport,
401 				const __be32 daddr, const __be16 dport,
402 				const int dif)
403 {
404 	return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
405 					 ntohs(dport), dif, 0);
406 }
407 
408 static inline struct sock *__inet_lookup(struct net *net,
409 					 struct inet_hashinfo *hashinfo,
410 					 struct sk_buff *skb, int doff,
411 					 const __be32 saddr, const __be16 sport,
412 					 const __be32 daddr, const __be16 dport,
413 					 const int dif, const int sdif,
414 					 bool *refcounted)
415 {
416 	u16 hnum = ntohs(dport);
417 	struct sock *sk;
418 
419 	sk = __inet_lookup_established(net, hashinfo, saddr, sport,
420 				       daddr, hnum, dif, sdif);
421 	*refcounted = true;
422 	if (sk)
423 		return sk;
424 	*refcounted = false;
425 	return __inet_lookup_listener(net, hashinfo, skb, doff, saddr,
426 				      sport, daddr, hnum, dif, sdif);
427 }
428 
429 static inline struct sock *inet_lookup(struct net *net,
430 				       struct inet_hashinfo *hashinfo,
431 				       struct sk_buff *skb, int doff,
432 				       const __be32 saddr, const __be16 sport,
433 				       const __be32 daddr, const __be16 dport,
434 				       const int dif)
435 {
436 	struct sock *sk;
437 	bool refcounted;
438 
439 	sk = __inet_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
440 			   dport, dif, 0, &refcounted);
441 
442 	if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
443 		sk = NULL;
444 	return sk;
445 }
446 
447 static inline
448 struct sock *inet_steal_sock(struct net *net, struct sk_buff *skb, int doff,
449 			     const __be32 saddr, const __be16 sport,
450 			     const __be32 daddr, const __be16 dport,
451 			     bool *refcounted, inet_ehashfn_t *ehashfn)
452 {
453 	struct sock *sk, *reuse_sk;
454 	bool prefetched;
455 
456 	sk = skb_steal_sock(skb, refcounted, &prefetched);
457 	if (!sk)
458 		return NULL;
459 
460 	if (!prefetched || !sk_fullsock(sk))
461 		return sk;
462 
463 	if (sk->sk_protocol == IPPROTO_TCP) {
464 		if (sk->sk_state != TCP_LISTEN)
465 			return sk;
466 	} else if (sk->sk_protocol == IPPROTO_UDP) {
467 		if (sk->sk_state != TCP_CLOSE)
468 			return sk;
469 	} else {
470 		return sk;
471 	}
472 
473 	reuse_sk = inet_lookup_reuseport(net, sk, skb, doff,
474 					 saddr, sport, daddr, ntohs(dport),
475 					 ehashfn);
476 	if (!reuse_sk)
477 		return sk;
478 
479 	/* We've chosen a new reuseport sock which is never refcounted. This
480 	 * implies that sk also isn't refcounted.
481 	 */
482 	WARN_ON_ONCE(*refcounted);
483 
484 	return reuse_sk;
485 }
486 
487 static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
488 					     struct sk_buff *skb,
489 					     int doff,
490 					     const __be16 sport,
491 					     const __be16 dport,
492 					     const int sdif,
493 					     bool *refcounted)
494 {
495 	struct net *net = dev_net(skb_dst(skb)->dev);
496 	const struct iphdr *iph = ip_hdr(skb);
497 	struct sock *sk;
498 
499 	sk = inet_steal_sock(net, skb, doff, iph->saddr, sport, iph->daddr, dport,
500 			     refcounted, inet_ehashfn);
501 	if (IS_ERR(sk))
502 		return NULL;
503 	if (sk)
504 		return sk;
505 
506 	return __inet_lookup(net, hashinfo, skb,
507 			     doff, iph->saddr, sport,
508 			     iph->daddr, dport, inet_iif(skb), sdif,
509 			     refcounted);
510 }
511 
512 static inline void sk_daddr_set(struct sock *sk, __be32 addr)
513 {
514 	sk->sk_daddr = addr; /* alias of inet_daddr */
515 #if IS_ENABLED(CONFIG_IPV6)
516 	ipv6_addr_set_v4mapped(addr, &sk->sk_v6_daddr);
517 #endif
518 }
519 
520 static inline void sk_rcv_saddr_set(struct sock *sk, __be32 addr)
521 {
522 	sk->sk_rcv_saddr = addr; /* alias of inet_rcv_saddr */
523 #if IS_ENABLED(CONFIG_IPV6)
524 	ipv6_addr_set_v4mapped(addr, &sk->sk_v6_rcv_saddr);
525 #endif
526 }
527 
528 int __inet_hash_connect(struct inet_timewait_death_row *death_row,
529 			struct sock *sk, u64 port_offset,
530 			int (*check_established)(struct inet_timewait_death_row *,
531 						 struct sock *, __u16,
532 						 struct inet_timewait_sock **));
533 
534 int inet_hash_connect(struct inet_timewait_death_row *death_row,
535 		      struct sock *sk);
536 #endif /* _INET_HASHTABLES_H */
537