xref: /linux/net/ipv6/inet6_hashtables.c (revision 447e140e66fd226350b3ce86cffc965eaae4c856)
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/hotdata.h>
18 #include <net/inet_connection_sock.h>
19 #include <net/inet_hashtables.h>
20 #include <net/inet6_hashtables.h>
21 #include <net/secure_seq.h>
22 #include <net/ip.h>
23 #include <net/sock_reuseport.h>
24 #include <net/tcp.h>
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 	u32 lhash, fhash;
31 
32 	net_get_random_once(&inet6_ehash_secret, sizeof(inet6_ehash_secret));
33 	net_get_random_once(&tcp_ipv6_hash_secret, sizeof(tcp_ipv6_hash_secret));
34 
35 	lhash = (__force u32)laddr->s6_addr32[3];
36 	fhash = __ipv6_addr_jhash(faddr, tcp_ipv6_hash_secret);
37 
38 	return __inet6_ehashfn(lhash, lport, fhash, fport,
39 			       inet6_ehash_secret + net_hash_mix(net));
40 }
41 EXPORT_SYMBOL_GPL(inet6_ehashfn);
42 
43 /*
44  * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
45  * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
46  *
47  * The sockhash lock must be held as a reader here.
48  */
49 struct sock *__inet6_lookup_established(struct net *net,
50 					struct inet_hashinfo *hashinfo,
51 					   const struct in6_addr *saddr,
52 					   const __be16 sport,
53 					   const struct in6_addr *daddr,
54 					   const u16 hnum,
55 					   const int dif, const int sdif)
56 {
57 	struct sock *sk;
58 	const struct hlist_nulls_node *node;
59 	const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
60 	/* Optimize here for direct hit, only listening connections can
61 	 * have wildcards anyways.
62 	 */
63 	unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
64 	unsigned int slot = hash & hashinfo->ehash_mask;
65 	struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
66 
67 
68 begin:
69 	sk_nulls_for_each_rcu(sk, node, &head->chain) {
70 		if (sk->sk_hash != hash)
71 			continue;
72 		if (!inet6_match(net, sk, saddr, daddr, ports, dif, sdif))
73 			continue;
74 		if (unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
75 			goto out;
76 
77 		if (unlikely(!inet6_match(net, sk, saddr, daddr, ports, dif, sdif))) {
78 			sock_gen_put(sk);
79 			goto begin;
80 		}
81 		goto found;
82 	}
83 	if (get_nulls_value(node) != slot)
84 		goto begin;
85 out:
86 	sk = NULL;
87 found:
88 	return sk;
89 }
90 EXPORT_SYMBOL(__inet6_lookup_established);
91 
92 static inline int compute_score(struct sock *sk, struct net *net,
93 				const unsigned short hnum,
94 				const struct in6_addr *daddr,
95 				const int dif, const int sdif)
96 {
97 	int score = -1;
98 
99 	if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum &&
100 	    sk->sk_family == PF_INET6) {
101 		if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
102 			return -1;
103 
104 		if (!inet_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
105 			return -1;
106 
107 		score =  sk->sk_bound_dev_if ? 2 : 1;
108 		if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
109 			score++;
110 	}
111 	return score;
112 }
113 
114 /**
115  * inet6_lookup_reuseport() - execute reuseport logic on AF_INET6 socket if necessary.
116  * @net: network namespace.
117  * @sk: AF_INET6 socket, must be in TCP_LISTEN state for TCP or TCP_CLOSE for UDP.
118  * @skb: context for a potential SK_REUSEPORT program.
119  * @doff: header offset.
120  * @saddr: source address.
121  * @sport: source port.
122  * @daddr: destination address.
123  * @hnum: destination port in host byte order.
124  * @ehashfn: hash function used to generate the fallback hash.
125  *
126  * Return: NULL if sk doesn't have SO_REUSEPORT set, otherwise a pointer to
127  *         the selected sock or an error.
128  */
129 struct sock *inet6_lookup_reuseport(struct net *net, struct sock *sk,
130 				    struct sk_buff *skb, int doff,
131 				    const struct in6_addr *saddr,
132 				    __be16 sport,
133 				    const struct in6_addr *daddr,
134 				    unsigned short hnum,
135 				    inet6_ehashfn_t *ehashfn)
136 {
137 	struct sock *reuse_sk = NULL;
138 	u32 phash;
139 
140 	if (sk->sk_reuseport) {
141 		phash = INDIRECT_CALL_INET(ehashfn, udp6_ehashfn, inet6_ehashfn,
142 					   net, daddr, hnum, saddr, sport);
143 		reuse_sk = reuseport_select_sock(sk, phash, skb, doff);
144 	}
145 	return reuse_sk;
146 }
147 EXPORT_SYMBOL_GPL(inet6_lookup_reuseport);
148 
149 /* called with rcu_read_lock() */
150 static struct sock *inet6_lhash2_lookup(struct net *net,
151 		struct inet_listen_hashbucket *ilb2,
152 		struct sk_buff *skb, int doff,
153 		const struct in6_addr *saddr,
154 		const __be16 sport, const struct in6_addr *daddr,
155 		const unsigned short hnum, const int dif, const int sdif)
156 {
157 	struct sock *sk, *result = NULL;
158 	struct hlist_nulls_node *node;
159 	int score, hiscore = 0;
160 
161 	sk_nulls_for_each_rcu(sk, node, &ilb2->nulls_head) {
162 		score = compute_score(sk, net, hnum, daddr, dif, sdif);
163 		if (score > hiscore) {
164 			result = inet6_lookup_reuseport(net, sk, skb, doff,
165 							saddr, sport, daddr, hnum, inet6_ehashfn);
166 			if (result)
167 				return result;
168 
169 			result = sk;
170 			hiscore = score;
171 		}
172 	}
173 
174 	return result;
175 }
176 
177 struct sock *inet6_lookup_run_sk_lookup(struct net *net,
178 					int protocol,
179 					struct sk_buff *skb, int doff,
180 					const struct in6_addr *saddr,
181 					const __be16 sport,
182 					const struct in6_addr *daddr,
183 					const u16 hnum, const int dif,
184 					inet6_ehashfn_t *ehashfn)
185 {
186 	struct sock *sk, *reuse_sk;
187 	bool no_reuseport;
188 
189 	no_reuseport = bpf_sk_lookup_run_v6(net, protocol, saddr, sport,
190 					    daddr, hnum, dif, &sk);
191 	if (no_reuseport || IS_ERR_OR_NULL(sk))
192 		return sk;
193 
194 	reuse_sk = inet6_lookup_reuseport(net, sk, skb, doff,
195 					  saddr, sport, daddr, hnum, ehashfn);
196 	if (reuse_sk)
197 		sk = reuse_sk;
198 	return sk;
199 }
200 EXPORT_SYMBOL_GPL(inet6_lookup_run_sk_lookup);
201 
202 struct sock *inet6_lookup_listener(struct net *net,
203 		struct inet_hashinfo *hashinfo,
204 		struct sk_buff *skb, int doff,
205 		const struct in6_addr *saddr,
206 		const __be16 sport, const struct in6_addr *daddr,
207 		const unsigned short hnum, const int dif, const int sdif)
208 {
209 	struct inet_listen_hashbucket *ilb2;
210 	struct sock *result = NULL;
211 	unsigned int hash2;
212 
213 	/* Lookup redirect from BPF */
214 	if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
215 	    hashinfo == net->ipv4.tcp_death_row.hashinfo) {
216 		result = inet6_lookup_run_sk_lookup(net, IPPROTO_TCP, skb, doff,
217 						    saddr, sport, daddr, hnum, dif,
218 						    inet6_ehashfn);
219 		if (result)
220 			goto done;
221 	}
222 
223 	hash2 = ipv6_portaddr_hash(net, daddr, hnum);
224 	ilb2 = inet_lhash2_bucket(hashinfo, hash2);
225 
226 	result = inet6_lhash2_lookup(net, ilb2, skb, doff,
227 				     saddr, sport, daddr, hnum,
228 				     dif, sdif);
229 	if (result)
230 		goto done;
231 
232 	/* Lookup lhash2 with in6addr_any */
233 	hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
234 	ilb2 = inet_lhash2_bucket(hashinfo, hash2);
235 
236 	result = inet6_lhash2_lookup(net, ilb2, skb, doff,
237 				     saddr, sport, &in6addr_any, hnum,
238 				     dif, sdif);
239 done:
240 	if (IS_ERR(result))
241 		return NULL;
242 	return result;
243 }
244 EXPORT_SYMBOL_GPL(inet6_lookup_listener);
245 
246 struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
247 			  struct sk_buff *skb, int doff,
248 			  const struct in6_addr *saddr, const __be16 sport,
249 			  const struct in6_addr *daddr, const __be16 dport,
250 			  const int dif)
251 {
252 	struct sock *sk;
253 	bool refcounted;
254 
255 	sk = __inet6_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
256 			    ntohs(dport), dif, 0, &refcounted);
257 	if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
258 		sk = NULL;
259 	return sk;
260 }
261 EXPORT_SYMBOL_GPL(inet6_lookup);
262 
263 static int __inet6_check_established(struct inet_timewait_death_row *death_row,
264 				     struct sock *sk, const __u16 lport,
265 				     struct inet_timewait_sock **twp)
266 {
267 	struct inet_hashinfo *hinfo = death_row->hashinfo;
268 	struct inet_sock *inet = inet_sk(sk);
269 	const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr;
270 	const struct in6_addr *saddr = &sk->sk_v6_daddr;
271 	const int dif = sk->sk_bound_dev_if;
272 	struct net *net = sock_net(sk);
273 	const int sdif = l3mdev_master_ifindex_by_index(net, dif);
274 	const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
275 	const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
276 						inet->inet_dport);
277 	struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
278 	spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
279 	struct sock *sk2;
280 	const struct hlist_nulls_node *node;
281 	struct inet_timewait_sock *tw = NULL;
282 
283 	spin_lock(lock);
284 
285 	sk_nulls_for_each(sk2, node, &head->chain) {
286 		if (sk2->sk_hash != hash)
287 			continue;
288 
289 		if (likely(inet6_match(net, sk2, saddr, daddr, ports,
290 				       dif, sdif))) {
291 			if (sk2->sk_state == TCP_TIME_WAIT) {
292 				tw = inet_twsk(sk2);
293 				if (sk->sk_protocol == IPPROTO_TCP &&
294 				    tcp_twsk_unique(sk, sk2, twp))
295 					break;
296 			}
297 			goto not_unique;
298 		}
299 	}
300 
301 	/* Must record num and sport now. Otherwise we will see
302 	 * in hash table socket with a funny identity.
303 	 */
304 	inet->inet_num = lport;
305 	inet->inet_sport = htons(lport);
306 	sk->sk_hash = hash;
307 	WARN_ON(!sk_unhashed(sk));
308 	__sk_nulls_add_node_rcu(sk, &head->chain);
309 	if (tw) {
310 		sk_nulls_del_node_init_rcu((struct sock *)tw);
311 		__NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED);
312 	}
313 	spin_unlock(lock);
314 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
315 
316 	if (twp) {
317 		*twp = tw;
318 	} else if (tw) {
319 		/* Silly. Should hash-dance instead... */
320 		inet_twsk_deschedule_put(tw);
321 	}
322 	return 0;
323 
324 not_unique:
325 	spin_unlock(lock);
326 	return -EADDRNOTAVAIL;
327 }
328 
329 static u64 inet6_sk_port_offset(const struct sock *sk)
330 {
331 	const struct inet_sock *inet = inet_sk(sk);
332 
333 	return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32,
334 					  sk->sk_v6_daddr.s6_addr32,
335 					  inet->inet_dport);
336 }
337 
338 int inet6_hash_connect(struct inet_timewait_death_row *death_row,
339 		       struct sock *sk)
340 {
341 	u64 port_offset = 0;
342 
343 	if (!inet_sk(sk)->inet_num)
344 		port_offset = inet6_sk_port_offset(sk);
345 	return __inet_hash_connect(death_row, sk, port_offset,
346 				   __inet6_check_established);
347 }
348 EXPORT_SYMBOL_GPL(inet6_hash_connect);
349 
350 int inet6_hash(struct sock *sk)
351 {
352 	int err = 0;
353 
354 	if (sk->sk_state != TCP_CLOSE)
355 		err = __inet_hash(sk, NULL);
356 
357 	return err;
358 }
359 EXPORT_SYMBOL_GPL(inet6_hash);
360