inet_hashtables.h

  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 */
 39struct 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
 77struct 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	struct rcu_head		rcu;
 93};
 94
 95struct inet_bind2_bucket {
 96	possible_net_t		ib_net;
 97	int			l3mdev;
 98	unsigned short		port;
 99#if IS_ENABLED(CONFIG_IPV6)
100	unsigned short		addr_type;
101	struct in6_addr		v6_rcv_saddr;
102#define rcv_saddr		v6_rcv_saddr.s6_addr32[3]
103#else
104	__be32			rcv_saddr;
105#endif
106	/* Node in the bhash2 inet_bind_hashbucket chain */
107	struct hlist_node	node;
108	struct hlist_node	bhash_node;
109	/* List of sockets hashed to this bucket */
110	struct hlist_head	owners;
111	signed char		fastreuse;
112	signed char		fastreuseport;
113};
114
115static inline struct net *ib_net(const struct inet_bind_bucket *ib)
116{
117	return read_pnet(&ib->ib_net);
118}
119
120static inline struct net *ib2_net(const struct inet_bind2_bucket *ib)
121{
122	return read_pnet(&ib->ib_net);
123}
124
125#define inet_bind_bucket_for_each(tb, head) \
126	hlist_for_each_entry(tb, head, node)
127
128struct inet_bind_hashbucket {
129	spinlock_t		lock;
130	struct hlist_head	chain;
131};
132
133/* Sockets can be hashed in established or listening table.
134 * We must use different 'nulls' end-of-chain value for all hash buckets :
135 * A socket might transition from ESTABLISH to LISTEN state without
136 * RCU grace period. A lookup in ehash table needs to handle this case.
137 */
138#define LISTENING_NULLS_BASE (1U << 29)
139struct inet_listen_hashbucket {
140	spinlock_t		lock;
141	struct hlist_nulls_head	nulls_head;
142};
143
144/* This is for listening sockets, thus all sockets which possess wildcards. */
145#define INET_LHTABLE_SIZE	32	/* Yes, really, this is all you need. */
146
147struct inet_hashinfo {
148	/* This is for sockets with full identity only.  Sockets here will
149	 * always be without wildcards and will have the following invariant:
150	 *
151	 *          TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
152	 *
153	 */
154	struct inet_ehash_bucket	*ehash;
155	spinlock_t			*ehash_locks;
156	unsigned int			ehash_mask;
157	unsigned int			ehash_locks_mask;
158
159	/* Ok, let's try this, I give up, we do need a local binding
160	 * TCP hash as well as the others for fast bind/connect.
161	 */
162	struct kmem_cache		*bind_bucket_cachep;
163	/* This bind table is hashed by local port */
164	struct inet_bind_hashbucket	*bhash;
165	struct kmem_cache		*bind2_bucket_cachep;
166	/* This bind table is hashed by local port and sk->sk_rcv_saddr (ipv4)
167	 * or sk->sk_v6_rcv_saddr (ipv6). This 2nd bind table is used
168	 * primarily for expediting bind conflict resolution.
169	 */
170	struct inet_bind_hashbucket	*bhash2;
171	unsigned int			bhash_size;
172
173	/* The 2nd listener table hashed by local port and address */
174	unsigned int			lhash2_mask;
175	struct inet_listen_hashbucket	*lhash2;
176
177	bool				pernet;
178} ____cacheline_aligned_in_smp;
179
180static inline struct inet_hashinfo *tcp_get_hashinfo(const struct sock *sk)
181{
182	return sock_net(sk)->ipv4.tcp_death_row.hashinfo;
183}
184
185static inline struct inet_listen_hashbucket *
186inet_lhash2_bucket(struct inet_hashinfo *h, u32 hash)
187{
188	return &h->lhash2[hash & h->lhash2_mask];
189}
190
191static 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
198static 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
205int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo);
206
207static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
208{
209	kvfree(hashinfo->ehash_locks);
210	hashinfo->ehash_locks = NULL;
211}
212
213struct inet_hashinfo *inet_pernet_hashinfo_alloc(struct inet_hashinfo *hashinfo,
214						 unsigned int ehash_entries);
215void inet_pernet_hashinfo_free(struct inet_hashinfo *hashinfo);
216
217struct inet_bind_bucket *
218inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net,
219			struct inet_bind_hashbucket *head,
220			const unsigned short snum, int l3mdev);
221void inet_bind_bucket_destroy(struct inet_bind_bucket *tb);
222
223bool inet_bind_bucket_match(const struct inet_bind_bucket *tb,
224			    const struct net *net, unsigned short port,
225			    int l3mdev);
226
227struct inet_bind2_bucket *
228inet_bind2_bucket_create(struct kmem_cache *cachep, struct net *net,
229			 struct inet_bind_hashbucket *head,
230			 struct inet_bind_bucket *tb,
231			 const struct sock *sk);
232
233void inet_bind2_bucket_destroy(struct kmem_cache *cachep,
234			       struct inet_bind2_bucket *tb);
235
236struct inet_bind2_bucket *
237inet_bind2_bucket_find(const struct inet_bind_hashbucket *head,
238		       const struct net *net,
239		       unsigned short port, int l3mdev,
240		       const struct sock *sk);
241
242bool inet_bind2_bucket_match_addr_any(const struct inet_bind2_bucket *tb,
243				      const struct net *net, unsigned short port,
244				      int l3mdev, const struct sock *sk);
245
246static inline u32 inet_bhashfn(const struct net *net, const __u16 lport,
247			       const u32 bhash_size)
248{
249	return (lport + net_hash_mix(net)) & (bhash_size - 1);
250}
251
252static inline struct inet_bind_hashbucket *
253inet_bhashfn_portaddr(const struct inet_hashinfo *hinfo, const struct sock *sk,
254		      const struct net *net, unsigned short port)
255{
256	u32 hash;
257
258#if IS_ENABLED(CONFIG_IPV6)
259	if (sk->sk_family == AF_INET6)
260		hash = ipv6_portaddr_hash(net, &sk->sk_v6_rcv_saddr, port);
261	else
262#endif
263		hash = ipv4_portaddr_hash(net, sk->sk_rcv_saddr, port);
264	return &hinfo->bhash2[hash & (hinfo->bhash_size - 1)];
265}
266
267static inline bool inet_use_hash2_on_bind(const struct sock *sk)
268{
269#if IS_ENABLED(CONFIG_IPV6)
270	if (sk->sk_family == AF_INET6) {
271		if (ipv6_addr_any(&sk->sk_v6_rcv_saddr))
272			return false;
273
274		if (!ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
275			return true;
276	}
277#endif
278	return sk->sk_rcv_saddr != htonl(INADDR_ANY);
279}
280
281struct inet_bind_hashbucket *
282inet_bhash2_addr_any_hashbucket(const struct sock *sk, const struct net *net, int port);
283
284/* This should be called whenever a socket's sk_rcv_saddr (ipv4) or
285 * sk_v6_rcv_saddr (ipv6) changes after it has been binded. The socket's
286 * rcv_saddr field should already have been updated when this is called.
287 */
288int inet_bhash2_update_saddr(struct sock *sk, void *saddr, int family);
289void inet_bhash2_reset_saddr(struct sock *sk);
290
291void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
292		    struct inet_bind2_bucket *tb2, unsigned short port);
293
294/* Caller must disable local BH processing. */
295int __inet_inherit_port(const struct sock *sk, struct sock *child);
296
297void inet_put_port(struct sock *sk);
298
299void inet_hashinfo2_init(struct inet_hashinfo *h, const char *name,
300			 unsigned long numentries, int scale,
301			 unsigned long low_limit,
302			 unsigned long high_limit);
303int inet_hashinfo2_init_mod(struct inet_hashinfo *h);
304
305bool inet_ehash_insert(struct sock *sk, struct sock *osk, bool *found_dup_sk);
306bool inet_ehash_nolisten(struct sock *sk, struct sock *osk,
307			 bool *found_dup_sk);
308int inet_hash(struct sock *sk);
309void inet_unhash(struct sock *sk);
310
311struct sock *__inet_lookup_listener(const struct net *net,
312				    struct sk_buff *skb, int doff,
313				    const __be32 saddr, const __be16 sport,
314				    const __be32 daddr,
315				    const unsigned short hnum,
316				    const int dif, const int sdif);
317
318static inline struct sock *inet_lookup_listener(struct net *net,
319						struct sk_buff *skb, int doff,
320						__be32 saddr, __be16 sport,
321						__be32 daddr, __be16 dport,
322						int dif, int sdif)
323{
324	return __inet_lookup_listener(net, skb, doff, saddr, sport,
325				      daddr, ntohs(dport), dif, sdif);
326}
327
328/* Socket demux engine toys. */
329/* What happens here is ugly; there's a pair of adjacent fields in
330   struct inet_sock; __be16 dport followed by __u16 num.  We want to
331   search by pair, so we combine the keys into a single 32bit value
332   and compare with 32bit value read from &...->dport.  Let's at least
333   make sure that it's not mixed with anything else...
334   On 64bit targets we combine comparisons with pair of adjacent __be32
335   fields in the same way.
336*/
337#ifdef __BIG_ENDIAN
338#define INET_COMBINED_PORTS(__sport, __dport) \
339	((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
340#else /* __LITTLE_ENDIAN */
341#define INET_COMBINED_PORTS(__sport, __dport) \
342	((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
343#endif
344
345#ifdef __BIG_ENDIAN
346#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
347	const __addrpair __name = (__force __addrpair) ( \
348				   (((__force __u64)(__be32)(__saddr)) << 32) | \
349				   ((__force __u64)(__be32)(__daddr)))
350#else /* __LITTLE_ENDIAN */
351#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
352	const __addrpair __name = (__force __addrpair) ( \
353				   (((__force __u64)(__be32)(__daddr)) << 32) | \
354				   ((__force __u64)(__be32)(__saddr)))
355#endif /* __BIG_ENDIAN */
356
357static inline bool inet_match(const struct net *net, const struct sock *sk,
358			      const __addrpair cookie, const __portpair ports,
359			      int dif, int sdif)
360{
361	if (!net_eq(sock_net(sk), net) ||
362	    READ_ONCE(sk->sk_portpair) != ports ||
363	    sk->sk_addrpair != cookie)
364	        return false;
365
366	/* READ_ONCE() paired with WRITE_ONCE() in sock_bindtoindex_locked() */
367	return inet_sk_bound_dev_eq(net, READ_ONCE(sk->sk_bound_dev_if), dif,
368				    sdif);
369}
370
371/* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
372 * not check it for lookups anymore, thanks Alexey. -DaveM
373 */
374struct sock *__inet_lookup_established(const struct net *net,
375				       const __be32 saddr, const __be16 sport,
376				       const __be32 daddr, const u16 hnum,
377				       const int dif, const int sdif);
378
379typedef u32 (inet_ehashfn_t)(const struct net *net,
380			      const __be32 laddr, const __u16 lport,
381			      const __be32 faddr, const __be16 fport);
382
383inet_ehashfn_t inet_ehashfn;
384
385INDIRECT_CALLABLE_DECLARE(inet_ehashfn_t udp_ehashfn);
386
387struct sock *inet_lookup_reuseport(const struct net *net, struct sock *sk,
388				   struct sk_buff *skb, int doff,
389				   __be32 saddr, __be16 sport,
390				   __be32 daddr, unsigned short hnum,
391				   inet_ehashfn_t *ehashfn);
392
393struct sock *inet_lookup_run_sk_lookup(const struct net *net,
394				       int protocol,
395				       struct sk_buff *skb, int doff,
396				       __be32 saddr, __be16 sport,
397				       __be32 daddr, u16 hnum, const int dif,
398				       inet_ehashfn_t *ehashfn);
399
400static inline struct sock *inet_lookup_established(struct net *net,
401						   const __be32 saddr, const __be16 sport,
402						   const __be32 daddr, const __be16 dport,
403						   const int dif)
404{
405	return __inet_lookup_established(net, saddr, sport, daddr,
406					 ntohs(dport), dif, 0);
407}
408
409static inline struct sock *__inet_lookup(struct net *net,
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, 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, skb, doff, saddr,
426				      sport, daddr, hnum, dif, sdif);
427}
428
429static inline struct sock *inet_lookup(struct net *net,
430				       struct sk_buff *skb, int doff,
431				       const __be32 saddr, const __be16 sport,
432				       const __be32 daddr, const __be16 dport,
433				       const int dif)
434{
435	struct sock *sk;
436	bool refcounted;
437
438	sk = __inet_lookup(net, skb, doff, saddr, sport, daddr,
439			   dport, dif, 0, &refcounted);
440
441	if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
442		sk = NULL;
443	return sk;
444}
445
446static inline
447struct sock *inet_steal_sock(struct net *net, struct sk_buff *skb, int doff,
448			     const __be32 saddr, const __be16 sport,
449			     const __be32 daddr, const __be16 dport,
450			     bool *refcounted, inet_ehashfn_t *ehashfn)
451{
452	struct sock *sk, *reuse_sk;
453	bool prefetched;
454
455	sk = skb_steal_sock(skb, refcounted, &prefetched);
456	if (!sk)
457		return NULL;
458
459	if (!prefetched || !sk_fullsock(sk))
460		return sk;
461
462	if (sk->sk_protocol == IPPROTO_TCP) {
463		if (sk->sk_state != TCP_LISTEN)
464			return sk;
465	} else if (sk->sk_protocol == IPPROTO_UDP) {
466		if (sk->sk_state != TCP_CLOSE)
467			return sk;
468	} else {
469		return sk;
470	}
471
472	reuse_sk = inet_lookup_reuseport(net, sk, skb, doff,
473					 saddr, sport, daddr, ntohs(dport),
474					 ehashfn);
475	if (!reuse_sk)
476		return sk;
477
478	/* We've chosen a new reuseport sock which is never refcounted. This
479	 * implies that sk also isn't refcounted.
480	 */
481	WARN_ON_ONCE(*refcounted);
482
483	return reuse_sk;
484}
485
486static inline struct sock *__inet_lookup_skb(struct sk_buff *skb,
487					     int doff,
488					     const __be16 sport,
489					     const __be16 dport,
490					     const int sdif,
491					     bool *refcounted)
492{
493	struct net *net = skb_dst_dev_net_rcu(skb);
494	const struct iphdr *iph = ip_hdr(skb);
495	struct sock *sk;
496
497	sk = inet_steal_sock(net, skb, doff, iph->saddr, sport, iph->daddr, dport,
498			     refcounted, inet_ehashfn);
499	if (IS_ERR(sk))
500		return NULL;
501	if (sk)
502		return sk;
503
504	return __inet_lookup(net, skb, doff, iph->saddr, sport,
505			     iph->daddr, dport, inet_iif(skb), sdif,
506			     refcounted);
507}
508
509static inline void sk_daddr_set(struct sock *sk, __be32 addr)
510{
511	sk->sk_daddr = addr; /* alias of inet_daddr */
512#if IS_ENABLED(CONFIG_IPV6)
513	ipv6_addr_set_v4mapped(addr, &sk->sk_v6_daddr);
514#endif
515}
516
517static inline void sk_rcv_saddr_set(struct sock *sk, __be32 addr)
518{
519	sk->sk_rcv_saddr = addr; /* alias of inet_rcv_saddr */
520#if IS_ENABLED(CONFIG_IPV6)
521	ipv6_addr_set_v4mapped(addr, &sk->sk_v6_rcv_saddr);
522#endif
523}
524
525int __inet_hash_connect(struct inet_timewait_death_row *death_row,
526			struct sock *sk, u64 port_offset,
527			u32 hash_port0,
528			int (*check_established)(struct inet_timewait_death_row *,
529						 struct sock *, __u16,
530						 struct inet_timewait_sock **,
531						 bool rcu_lookup,
532						 u32 hash));
533
534int inet_hash_connect(struct inet_timewait_death_row *death_row,
535		      struct sock *sk);
536#endif /* _INET_HASHTABLES_H */