xref: /linux/net/ipv6/inet6_hashtables.c (revision 908fc4c2)
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  *		Generic INET6 transport hashtables
8  *
9  * Authors:	Lotsa people, from code originally in tcp, generalised here
10  *		by Arnaldo Carvalho de Melo <acme@mandriva.com>
11  */
12 
13 #include <linux/module.h>
14 #include <linux/random.h>
15 
16 #include <net/addrconf.h>
17 #include <net/inet_connection_sock.h>
18 #include <net/inet_hashtables.h>
19 #include <net/inet6_hashtables.h>
20 #include <net/secure_seq.h>
21 #include <net/ip.h>
22 #include <net/sock_reuseport.h>
23 
24 extern struct inet_hashinfo tcp_hashinfo;
25 
26 u32 inet6_ehashfn(const struct net *net,
27 		  const struct in6_addr *laddr, const u16 lport,
28 		  const struct in6_addr *faddr, const __be16 fport)
29 {
30 	static u32 inet6_ehash_secret __read_mostly;
31 	static u32 ipv6_hash_secret __read_mostly;
32 
33 	u32 lhash, fhash;
34 
35 	net_get_random_once(&inet6_ehash_secret, sizeof(inet6_ehash_secret));
36 	net_get_random_once(&ipv6_hash_secret, sizeof(ipv6_hash_secret));
37 
38 	lhash = (__force u32)laddr->s6_addr32[3];
39 	fhash = __ipv6_addr_jhash(faddr, ipv6_hash_secret);
40 
41 	return __inet6_ehashfn(lhash, lport, fhash, fport,
42 			       inet6_ehash_secret + net_hash_mix(net));
43 }
44 
45 /*
46  * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
47  * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
48  *
49  * The sockhash lock must be held as a reader here.
50  */
51 struct sock *__inet6_lookup_established(struct net *net,
52 					struct inet_hashinfo *hashinfo,
53 					   const struct in6_addr *saddr,
54 					   const __be16 sport,
55 					   const struct in6_addr *daddr,
56 					   const u16 hnum,
57 					   const int dif, const int sdif)
58 {
59 	struct sock *sk;
60 	const struct hlist_nulls_node *node;
61 	const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
62 	/* Optimize here for direct hit, only listening connections can
63 	 * have wildcards anyways.
64 	 */
65 	unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
66 	unsigned int slot = hash & hashinfo->ehash_mask;
67 	struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
68 
69 
70 begin:
71 	sk_nulls_for_each_rcu(sk, node, &head->chain) {
72 		if (sk->sk_hash != hash)
73 			continue;
74 		if (!inet6_match(net, sk, saddr, daddr, ports, dif, sdif))
75 			continue;
76 		if (unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
77 			goto out;
78 
79 		if (unlikely(!inet6_match(net, sk, saddr, daddr, ports, dif, sdif))) {
80 			sock_gen_put(sk);
81 			goto begin;
82 		}
83 		goto found;
84 	}
85 	if (get_nulls_value(node) != slot)
86 		goto begin;
87 out:
88 	sk = NULL;
89 found:
90 	return sk;
91 }
92 EXPORT_SYMBOL(__inet6_lookup_established);
93 
94 static inline int compute_score(struct sock *sk, struct net *net,
95 				const unsigned short hnum,
96 				const struct in6_addr *daddr,
97 				const int dif, const int sdif)
98 {
99 	int score = -1;
100 
101 	if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum &&
102 	    sk->sk_family == PF_INET6) {
103 		if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
104 			return -1;
105 
106 		if (!inet_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
107 			return -1;
108 
109 		score =  sk->sk_bound_dev_if ? 2 : 1;
110 		if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
111 			score++;
112 	}
113 	return score;
114 }
115 
116 static inline struct sock *lookup_reuseport(struct net *net, struct sock *sk,
117 					    struct sk_buff *skb, int doff,
118 					    const struct in6_addr *saddr,
119 					    __be16 sport,
120 					    const struct in6_addr *daddr,
121 					    unsigned short hnum)
122 {
123 	struct sock *reuse_sk = NULL;
124 	u32 phash;
125 
126 	if (sk->sk_reuseport) {
127 		phash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
128 		reuse_sk = reuseport_select_sock(sk, phash, skb, doff);
129 	}
130 	return reuse_sk;
131 }
132 
133 /* called with rcu_read_lock() */
134 static struct sock *inet6_lhash2_lookup(struct net *net,
135 		struct inet_listen_hashbucket *ilb2,
136 		struct sk_buff *skb, int doff,
137 		const struct in6_addr *saddr,
138 		const __be16 sport, const struct in6_addr *daddr,
139 		const unsigned short hnum, const int dif, const int sdif)
140 {
141 	struct sock *sk, *result = NULL;
142 	struct hlist_nulls_node *node;
143 	int score, hiscore = 0;
144 
145 	sk_nulls_for_each_rcu(sk, node, &ilb2->nulls_head) {
146 		score = compute_score(sk, net, hnum, daddr, dif, sdif);
147 		if (score > hiscore) {
148 			result = lookup_reuseport(net, sk, skb, doff,
149 						  saddr, sport, daddr, hnum);
150 			if (result)
151 				return result;
152 
153 			result = sk;
154 			hiscore = score;
155 		}
156 	}
157 
158 	return result;
159 }
160 
161 static inline struct sock *inet6_lookup_run_bpf(struct net *net,
162 						struct inet_hashinfo *hashinfo,
163 						struct sk_buff *skb, int doff,
164 						const struct in6_addr *saddr,
165 						const __be16 sport,
166 						const struct in6_addr *daddr,
167 						const u16 hnum, const int dif)
168 {
169 	struct sock *sk, *reuse_sk;
170 	bool no_reuseport;
171 
172 	if (hashinfo != &tcp_hashinfo)
173 		return NULL; /* only TCP is supported */
174 
175 	no_reuseport = bpf_sk_lookup_run_v6(net, IPPROTO_TCP, saddr, sport,
176 					    daddr, hnum, dif, &sk);
177 	if (no_reuseport || IS_ERR_OR_NULL(sk))
178 		return sk;
179 
180 	reuse_sk = lookup_reuseport(net, sk, skb, doff, saddr, sport, daddr, hnum);
181 	if (reuse_sk)
182 		sk = reuse_sk;
183 	return sk;
184 }
185 
186 struct sock *inet6_lookup_listener(struct net *net,
187 		struct inet_hashinfo *hashinfo,
188 		struct sk_buff *skb, int doff,
189 		const struct in6_addr *saddr,
190 		const __be16 sport, const struct in6_addr *daddr,
191 		const unsigned short hnum, const int dif, const int sdif)
192 {
193 	struct inet_listen_hashbucket *ilb2;
194 	struct sock *result = NULL;
195 	unsigned int hash2;
196 
197 	/* Lookup redirect from BPF */
198 	if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
199 		result = inet6_lookup_run_bpf(net, hashinfo, skb, doff,
200 					      saddr, sport, daddr, hnum, dif);
201 		if (result)
202 			goto done;
203 	}
204 
205 	hash2 = ipv6_portaddr_hash(net, daddr, hnum);
206 	ilb2 = inet_lhash2_bucket(hashinfo, hash2);
207 
208 	result = inet6_lhash2_lookup(net, ilb2, skb, doff,
209 				     saddr, sport, daddr, hnum,
210 				     dif, sdif);
211 	if (result)
212 		goto done;
213 
214 	/* Lookup lhash2 with in6addr_any */
215 	hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
216 	ilb2 = inet_lhash2_bucket(hashinfo, hash2);
217 
218 	result = inet6_lhash2_lookup(net, ilb2, skb, doff,
219 				     saddr, sport, &in6addr_any, hnum,
220 				     dif, sdif);
221 done:
222 	if (IS_ERR(result))
223 		return NULL;
224 	return result;
225 }
226 EXPORT_SYMBOL_GPL(inet6_lookup_listener);
227 
228 struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
229 			  struct sk_buff *skb, int doff,
230 			  const struct in6_addr *saddr, const __be16 sport,
231 			  const struct in6_addr *daddr, const __be16 dport,
232 			  const int dif)
233 {
234 	struct sock *sk;
235 	bool refcounted;
236 
237 	sk = __inet6_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
238 			    ntohs(dport), dif, 0, &refcounted);
239 	if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
240 		sk = NULL;
241 	return sk;
242 }
243 EXPORT_SYMBOL_GPL(inet6_lookup);
244 
245 static int __inet6_check_established(struct inet_timewait_death_row *death_row,
246 				     struct sock *sk, const __u16 lport,
247 				     struct inet_timewait_sock **twp)
248 {
249 	struct inet_hashinfo *hinfo = death_row->hashinfo;
250 	struct inet_sock *inet = inet_sk(sk);
251 	const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr;
252 	const struct in6_addr *saddr = &sk->sk_v6_daddr;
253 	const int dif = sk->sk_bound_dev_if;
254 	struct net *net = sock_net(sk);
255 	const int sdif = l3mdev_master_ifindex_by_index(net, dif);
256 	const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
257 	const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
258 						inet->inet_dport);
259 	struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
260 	spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
261 	struct sock *sk2;
262 	const struct hlist_nulls_node *node;
263 	struct inet_timewait_sock *tw = NULL;
264 
265 	spin_lock(lock);
266 
267 	sk_nulls_for_each(sk2, node, &head->chain) {
268 		if (sk2->sk_hash != hash)
269 			continue;
270 
271 		if (likely(inet6_match(net, sk2, saddr, daddr, ports,
272 				       dif, sdif))) {
273 			if (sk2->sk_state == TCP_TIME_WAIT) {
274 				tw = inet_twsk(sk2);
275 				if (twsk_unique(sk, sk2, twp))
276 					break;
277 			}
278 			goto not_unique;
279 		}
280 	}
281 
282 	/* Must record num and sport now. Otherwise we will see
283 	 * in hash table socket with a funny identity.
284 	 */
285 	inet->inet_num = lport;
286 	inet->inet_sport = htons(lport);
287 	sk->sk_hash = hash;
288 	WARN_ON(!sk_unhashed(sk));
289 	__sk_nulls_add_node_rcu(sk, &head->chain);
290 	if (tw) {
291 		sk_nulls_del_node_init_rcu((struct sock *)tw);
292 		__NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED);
293 	}
294 	spin_unlock(lock);
295 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
296 
297 	if (twp) {
298 		*twp = tw;
299 	} else if (tw) {
300 		/* Silly. Should hash-dance instead... */
301 		inet_twsk_deschedule_put(tw);
302 	}
303 	return 0;
304 
305 not_unique:
306 	spin_unlock(lock);
307 	return -EADDRNOTAVAIL;
308 }
309 
310 static u64 inet6_sk_port_offset(const struct sock *sk)
311 {
312 	const struct inet_sock *inet = inet_sk(sk);
313 
314 	return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32,
315 					  sk->sk_v6_daddr.s6_addr32,
316 					  inet->inet_dport);
317 }
318 
319 int inet6_hash_connect(struct inet_timewait_death_row *death_row,
320 		       struct sock *sk)
321 {
322 	u64 port_offset = 0;
323 
324 	if (!inet_sk(sk)->inet_num)
325 		port_offset = inet6_sk_port_offset(sk);
326 	return __inet_hash_connect(death_row, sk, port_offset,
327 				   __inet6_check_established);
328 }
329 EXPORT_SYMBOL_GPL(inet6_hash_connect);
330 
331 int inet6_hash(struct sock *sk)
332 {
333 	int err = 0;
334 
335 	if (sk->sk_state != TCP_CLOSE)
336 		err = __inet_hash(sk, NULL);
337 
338 	return err;
339 }
340 EXPORT_SYMBOL_GPL(inet6_hash);
341