xref: /openbmc/linux/include/net/inet_hashtables.h (revision e9088629)
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 static inline struct net *ib_net(struct inet_bind_bucket *ib)
94 {
95 	return read_pnet(&ib->ib_net);
96 }
97 
98 #define inet_bind_bucket_for_each(tb, head) \
99 	hlist_for_each_entry(tb, head, node)
100 
101 struct inet_bind_hashbucket {
102 	spinlock_t		lock;
103 	struct hlist_head	chain;
104 };
105 
106 /* Sockets can be hashed in established or listening table.
107  * We must use different 'nulls' end-of-chain value for all hash buckets :
108  * A socket might transition from ESTABLISH to LISTEN state without
109  * RCU grace period. A lookup in ehash table needs to handle this case.
110  */
111 #define LISTENING_NULLS_BASE (1U << 29)
112 struct inet_listen_hashbucket {
113 	spinlock_t		lock;
114 	struct hlist_nulls_head	nulls_head;
115 };
116 
117 /* This is for listening sockets, thus all sockets which possess wildcards. */
118 #define INET_LHTABLE_SIZE	32	/* Yes, really, this is all you need. */
119 
120 struct inet_hashinfo {
121 	/* This is for sockets with full identity only.  Sockets here will
122 	 * always be without wildcards and will have the following invariant:
123 	 *
124 	 *          TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
125 	 *
126 	 */
127 	struct inet_ehash_bucket	*ehash;
128 	spinlock_t			*ehash_locks;
129 	unsigned int			ehash_mask;
130 	unsigned int			ehash_locks_mask;
131 
132 	/* Ok, let's try this, I give up, we do need a local binding
133 	 * TCP hash as well as the others for fast bind/connect.
134 	 */
135 	struct kmem_cache		*bind_bucket_cachep;
136 	struct inet_bind_hashbucket	*bhash;
137 	unsigned int			bhash_size;
138 
139 	/* The 2nd listener table hashed by local port and address */
140 	unsigned int			lhash2_mask;
141 	struct inet_listen_hashbucket	*lhash2;
142 };
143 
144 static inline struct inet_listen_hashbucket *
145 inet_lhash2_bucket(struct inet_hashinfo *h, u32 hash)
146 {
147 	return &h->lhash2[hash & h->lhash2_mask];
148 }
149 
150 static inline struct inet_ehash_bucket *inet_ehash_bucket(
151 	struct inet_hashinfo *hashinfo,
152 	unsigned int hash)
153 {
154 	return &hashinfo->ehash[hash & hashinfo->ehash_mask];
155 }
156 
157 static inline spinlock_t *inet_ehash_lockp(
158 	struct inet_hashinfo *hashinfo,
159 	unsigned int hash)
160 {
161 	return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
162 }
163 
164 int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo);
165 
166 static inline void inet_hashinfo2_free_mod(struct inet_hashinfo *h)
167 {
168 	kfree(h->lhash2);
169 	h->lhash2 = NULL;
170 }
171 
172 static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
173 {
174 	kvfree(hashinfo->ehash_locks);
175 	hashinfo->ehash_locks = NULL;
176 }
177 
178 static inline bool inet_sk_bound_dev_eq(struct net *net, int bound_dev_if,
179 					int dif, int sdif)
180 {
181 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
182 	return inet_bound_dev_eq(!!READ_ONCE(net->ipv4.sysctl_tcp_l3mdev_accept),
183 				 bound_dev_if, dif, sdif);
184 #else
185 	return inet_bound_dev_eq(true, bound_dev_if, dif, sdif);
186 #endif
187 }
188 
189 struct inet_bind_bucket *
190 inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net,
191 			struct inet_bind_hashbucket *head,
192 			const unsigned short snum, int l3mdev);
193 void inet_bind_bucket_destroy(struct kmem_cache *cachep,
194 			      struct inet_bind_bucket *tb);
195 
196 static inline u32 inet_bhashfn(const struct net *net, const __u16 lport,
197 			       const u32 bhash_size)
198 {
199 	return (lport + net_hash_mix(net)) & (bhash_size - 1);
200 }
201 
202 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
203 		    const unsigned short snum);
204 
205 /* Caller must disable local BH processing. */
206 int __inet_inherit_port(const struct sock *sk, struct sock *child);
207 
208 void inet_put_port(struct sock *sk);
209 
210 void inet_hashinfo2_init(struct inet_hashinfo *h, const char *name,
211 			 unsigned long numentries, int scale,
212 			 unsigned long low_limit,
213 			 unsigned long high_limit);
214 int inet_hashinfo2_init_mod(struct inet_hashinfo *h);
215 
216 bool inet_ehash_insert(struct sock *sk, struct sock *osk, bool *found_dup_sk);
217 bool inet_ehash_nolisten(struct sock *sk, struct sock *osk,
218 			 bool *found_dup_sk);
219 int __inet_hash(struct sock *sk, struct sock *osk);
220 int inet_hash(struct sock *sk);
221 void inet_unhash(struct sock *sk);
222 
223 struct sock *__inet_lookup_listener(struct net *net,
224 				    struct inet_hashinfo *hashinfo,
225 				    struct sk_buff *skb, int doff,
226 				    const __be32 saddr, const __be16 sport,
227 				    const __be32 daddr,
228 				    const unsigned short hnum,
229 				    const int dif, const int sdif);
230 
231 static inline struct sock *inet_lookup_listener(struct net *net,
232 		struct inet_hashinfo *hashinfo,
233 		struct sk_buff *skb, int doff,
234 		__be32 saddr, __be16 sport,
235 		__be32 daddr, __be16 dport, int dif, int sdif)
236 {
237 	return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, sport,
238 				      daddr, ntohs(dport), dif, sdif);
239 }
240 
241 /* Socket demux engine toys. */
242 /* What happens here is ugly; there's a pair of adjacent fields in
243    struct inet_sock; __be16 dport followed by __u16 num.  We want to
244    search by pair, so we combine the keys into a single 32bit value
245    and compare with 32bit value read from &...->dport.  Let's at least
246    make sure that it's not mixed with anything else...
247    On 64bit targets we combine comparisons with pair of adjacent __be32
248    fields in the same way.
249 */
250 #ifdef __BIG_ENDIAN
251 #define INET_COMBINED_PORTS(__sport, __dport) \
252 	((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
253 #else /* __LITTLE_ENDIAN */
254 #define INET_COMBINED_PORTS(__sport, __dport) \
255 	((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
256 #endif
257 
258 #ifdef __BIG_ENDIAN
259 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
260 	const __addrpair __name = (__force __addrpair) ( \
261 				   (((__force __u64)(__be32)(__saddr)) << 32) | \
262 				   ((__force __u64)(__be32)(__daddr)))
263 #else /* __LITTLE_ENDIAN */
264 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
265 	const __addrpair __name = (__force __addrpair) ( \
266 				   (((__force __u64)(__be32)(__daddr)) << 32) | \
267 				   ((__force __u64)(__be32)(__saddr)))
268 #endif /* __BIG_ENDIAN */
269 
270 static inline bool inet_match(struct net *net, const struct sock *sk,
271 			      const __addrpair cookie, const __portpair ports,
272 			      int dif, int sdif)
273 {
274 	int bound_dev_if;
275 
276 	if (!net_eq(sock_net(sk), net) ||
277 	    sk->sk_portpair != ports ||
278 	    sk->sk_addrpair != cookie)
279 	        return false;
280 
281 	/* Paired with WRITE_ONCE() from sock_bindtoindex_locked() */
282 	bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
283 	return bound_dev_if == dif || bound_dev_if == sdif;
284 }
285 
286 /* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
287  * not check it for lookups anymore, thanks Alexey. -DaveM
288  */
289 struct sock *__inet_lookup_established(struct net *net,
290 				       struct inet_hashinfo *hashinfo,
291 				       const __be32 saddr, const __be16 sport,
292 				       const __be32 daddr, const u16 hnum,
293 				       const int dif, const int sdif);
294 
295 static inline struct sock *
296 	inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
297 				const __be32 saddr, const __be16 sport,
298 				const __be32 daddr, const __be16 dport,
299 				const int dif)
300 {
301 	return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
302 					 ntohs(dport), dif, 0);
303 }
304 
305 static inline struct sock *__inet_lookup(struct net *net,
306 					 struct inet_hashinfo *hashinfo,
307 					 struct sk_buff *skb, int doff,
308 					 const __be32 saddr, const __be16 sport,
309 					 const __be32 daddr, const __be16 dport,
310 					 const int dif, const int sdif,
311 					 bool *refcounted)
312 {
313 	u16 hnum = ntohs(dport);
314 	struct sock *sk;
315 
316 	sk = __inet_lookup_established(net, hashinfo, saddr, sport,
317 				       daddr, hnum, dif, sdif);
318 	*refcounted = true;
319 	if (sk)
320 		return sk;
321 	*refcounted = false;
322 	return __inet_lookup_listener(net, hashinfo, skb, doff, saddr,
323 				      sport, daddr, hnum, dif, sdif);
324 }
325 
326 static inline struct sock *inet_lookup(struct net *net,
327 				       struct inet_hashinfo *hashinfo,
328 				       struct sk_buff *skb, int doff,
329 				       const __be32 saddr, const __be16 sport,
330 				       const __be32 daddr, const __be16 dport,
331 				       const int dif)
332 {
333 	struct sock *sk;
334 	bool refcounted;
335 
336 	sk = __inet_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
337 			   dport, dif, 0, &refcounted);
338 
339 	if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
340 		sk = NULL;
341 	return sk;
342 }
343 
344 static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
345 					     struct sk_buff *skb,
346 					     int doff,
347 					     const __be16 sport,
348 					     const __be16 dport,
349 					     const int sdif,
350 					     bool *refcounted)
351 {
352 	struct sock *sk = skb_steal_sock(skb, refcounted);
353 	const struct iphdr *iph = ip_hdr(skb);
354 
355 	if (sk)
356 		return sk;
357 
358 	return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo, skb,
359 			     doff, iph->saddr, sport,
360 			     iph->daddr, dport, inet_iif(skb), sdif,
361 			     refcounted);
362 }
363 
364 u32 inet6_ehashfn(const struct net *net,
365 		  const struct in6_addr *laddr, const u16 lport,
366 		  const struct in6_addr *faddr, const __be16 fport);
367 
368 static inline void sk_daddr_set(struct sock *sk, __be32 addr)
369 {
370 	sk->sk_daddr = addr; /* alias of inet_daddr */
371 #if IS_ENABLED(CONFIG_IPV6)
372 	ipv6_addr_set_v4mapped(addr, &sk->sk_v6_daddr);
373 #endif
374 }
375 
376 static inline void sk_rcv_saddr_set(struct sock *sk, __be32 addr)
377 {
378 	sk->sk_rcv_saddr = addr; /* alias of inet_rcv_saddr */
379 #if IS_ENABLED(CONFIG_IPV6)
380 	ipv6_addr_set_v4mapped(addr, &sk->sk_v6_rcv_saddr);
381 #endif
382 }
383 
384 int __inet_hash_connect(struct inet_timewait_death_row *death_row,
385 			struct sock *sk, u64 port_offset,
386 			int (*check_established)(struct inet_timewait_death_row *,
387 						 struct sock *, __u16,
388 						 struct inet_timewait_sock **));
389 
390 int inet_hash_connect(struct inet_timewait_death_row *death_row,
391 		      struct sock *sk);
392 #endif /* _INET_HASHTABLES_H */
393