xref: /linux/net/ipv6/inet6_hashtables.c (revision e80a48bade619ec5a92230b3d4ae84bfc2746822)
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 u32 inet6_ehashfn(const struct net *net,
25 		  const struct in6_addr *laddr, const u16 lport,
26 		  const struct in6_addr *faddr, const __be16 fport)
27 {
28 	static u32 inet6_ehash_secret __read_mostly;
29 	static u32 ipv6_hash_secret __read_mostly;
30 
31 	u32 lhash, fhash;
32 
33 	net_get_random_once(&inet6_ehash_secret, sizeof(inet6_ehash_secret));
34 	net_get_random_once(&ipv6_hash_secret, sizeof(ipv6_hash_secret));
35 
36 	lhash = (__force u32)laddr->s6_addr32[3];
37 	fhash = __ipv6_addr_jhash(faddr, ipv6_hash_secret);
38 
39 	return __inet6_ehashfn(lhash, lport, fhash, fport,
40 			       inet6_ehash_secret + net_hash_mix(net));
41 }
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 static inline struct sock *lookup_reuseport(struct net *net, struct sock *sk,
115 					    struct sk_buff *skb, int doff,
116 					    const struct in6_addr *saddr,
117 					    __be16 sport,
118 					    const struct in6_addr *daddr,
119 					    unsigned short hnum)
120 {
121 	struct sock *reuse_sk = NULL;
122 	u32 phash;
123 
124 	if (sk->sk_reuseport) {
125 		phash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
126 		reuse_sk = reuseport_select_sock(sk, phash, skb, doff);
127 	}
128 	return reuse_sk;
129 }
130 
131 /* called with rcu_read_lock() */
132 static struct sock *inet6_lhash2_lookup(struct net *net,
133 		struct inet_listen_hashbucket *ilb2,
134 		struct sk_buff *skb, int doff,
135 		const struct in6_addr *saddr,
136 		const __be16 sport, const struct in6_addr *daddr,
137 		const unsigned short hnum, const int dif, const int sdif)
138 {
139 	struct sock *sk, *result = NULL;
140 	struct hlist_nulls_node *node;
141 	int score, hiscore = 0;
142 
143 	sk_nulls_for_each_rcu(sk, node, &ilb2->nulls_head) {
144 		score = compute_score(sk, net, hnum, daddr, dif, sdif);
145 		if (score > hiscore) {
146 			result = lookup_reuseport(net, sk, skb, doff,
147 						  saddr, sport, daddr, hnum);
148 			if (result)
149 				return result;
150 
151 			result = sk;
152 			hiscore = score;
153 		}
154 	}
155 
156 	return result;
157 }
158 
159 static inline struct sock *inet6_lookup_run_bpf(struct net *net,
160 						struct inet_hashinfo *hashinfo,
161 						struct sk_buff *skb, int doff,
162 						const struct in6_addr *saddr,
163 						const __be16 sport,
164 						const struct in6_addr *daddr,
165 						const u16 hnum, const int dif)
166 {
167 	struct sock *sk, *reuse_sk;
168 	bool no_reuseport;
169 
170 	if (hashinfo != net->ipv4.tcp_death_row.hashinfo)
171 		return NULL; /* only TCP is supported */
172 
173 	no_reuseport = bpf_sk_lookup_run_v6(net, IPPROTO_TCP, saddr, sport,
174 					    daddr, hnum, dif, &sk);
175 	if (no_reuseport || IS_ERR_OR_NULL(sk))
176 		return sk;
177 
178 	reuse_sk = lookup_reuseport(net, sk, skb, doff, saddr, sport, daddr, hnum);
179 	if (reuse_sk)
180 		sk = reuse_sk;
181 	return sk;
182 }
183 
184 struct sock *inet6_lookup_listener(struct net *net,
185 		struct inet_hashinfo *hashinfo,
186 		struct sk_buff *skb, int doff,
187 		const struct in6_addr *saddr,
188 		const __be16 sport, const struct in6_addr *daddr,
189 		const unsigned short hnum, const int dif, const int sdif)
190 {
191 	struct inet_listen_hashbucket *ilb2;
192 	struct sock *result = NULL;
193 	unsigned int hash2;
194 
195 	/* Lookup redirect from BPF */
196 	if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
197 		result = inet6_lookup_run_bpf(net, hashinfo, skb, doff,
198 					      saddr, sport, daddr, hnum, dif);
199 		if (result)
200 			goto done;
201 	}
202 
203 	hash2 = ipv6_portaddr_hash(net, daddr, hnum);
204 	ilb2 = inet_lhash2_bucket(hashinfo, hash2);
205 
206 	result = inet6_lhash2_lookup(net, ilb2, skb, doff,
207 				     saddr, sport, daddr, hnum,
208 				     dif, sdif);
209 	if (result)
210 		goto done;
211 
212 	/* Lookup lhash2 with in6addr_any */
213 	hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
214 	ilb2 = inet_lhash2_bucket(hashinfo, hash2);
215 
216 	result = inet6_lhash2_lookup(net, ilb2, skb, doff,
217 				     saddr, sport, &in6addr_any, hnum,
218 				     dif, sdif);
219 done:
220 	if (IS_ERR(result))
221 		return NULL;
222 	return result;
223 }
224 EXPORT_SYMBOL_GPL(inet6_lookup_listener);
225 
226 struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
227 			  struct sk_buff *skb, int doff,
228 			  const struct in6_addr *saddr, const __be16 sport,
229 			  const struct in6_addr *daddr, const __be16 dport,
230 			  const int dif)
231 {
232 	struct sock *sk;
233 	bool refcounted;
234 
235 	sk = __inet6_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
236 			    ntohs(dport), dif, 0, &refcounted);
237 	if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
238 		sk = NULL;
239 	return sk;
240 }
241 EXPORT_SYMBOL_GPL(inet6_lookup);
242 
243 static int __inet6_check_established(struct inet_timewait_death_row *death_row,
244 				     struct sock *sk, const __u16 lport,
245 				     struct inet_timewait_sock **twp)
246 {
247 	struct inet_hashinfo *hinfo = death_row->hashinfo;
248 	struct inet_sock *inet = inet_sk(sk);
249 	const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr;
250 	const struct in6_addr *saddr = &sk->sk_v6_daddr;
251 	const int dif = sk->sk_bound_dev_if;
252 	struct net *net = sock_net(sk);
253 	const int sdif = l3mdev_master_ifindex_by_index(net, dif);
254 	const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
255 	const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
256 						inet->inet_dport);
257 	struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
258 	spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
259 	struct sock *sk2;
260 	const struct hlist_nulls_node *node;
261 	struct inet_timewait_sock *tw = NULL;
262 
263 	spin_lock(lock);
264 
265 	sk_nulls_for_each(sk2, node, &head->chain) {
266 		if (sk2->sk_hash != hash)
267 			continue;
268 
269 		if (likely(inet6_match(net, sk2, saddr, daddr, ports,
270 				       dif, sdif))) {
271 			if (sk2->sk_state == TCP_TIME_WAIT) {
272 				tw = inet_twsk(sk2);
273 				if (twsk_unique(sk, sk2, twp))
274 					break;
275 			}
276 			goto not_unique;
277 		}
278 	}
279 
280 	/* Must record num and sport now. Otherwise we will see
281 	 * in hash table socket with a funny identity.
282 	 */
283 	inet->inet_num = lport;
284 	inet->inet_sport = htons(lport);
285 	sk->sk_hash = hash;
286 	WARN_ON(!sk_unhashed(sk));
287 	__sk_nulls_add_node_rcu(sk, &head->chain);
288 	if (tw) {
289 		sk_nulls_del_node_init_rcu((struct sock *)tw);
290 		__NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED);
291 	}
292 	spin_unlock(lock);
293 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
294 
295 	if (twp) {
296 		*twp = tw;
297 	} else if (tw) {
298 		/* Silly. Should hash-dance instead... */
299 		inet_twsk_deschedule_put(tw);
300 	}
301 	return 0;
302 
303 not_unique:
304 	spin_unlock(lock);
305 	return -EADDRNOTAVAIL;
306 }
307 
308 static u64 inet6_sk_port_offset(const struct sock *sk)
309 {
310 	const struct inet_sock *inet = inet_sk(sk);
311 
312 	return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32,
313 					  sk->sk_v6_daddr.s6_addr32,
314 					  inet->inet_dport);
315 }
316 
317 int inet6_hash_connect(struct inet_timewait_death_row *death_row,
318 		       struct sock *sk)
319 {
320 	u64 port_offset = 0;
321 
322 	if (!inet_sk(sk)->inet_num)
323 		port_offset = inet6_sk_port_offset(sk);
324 	return __inet_hash_connect(death_row, sk, port_offset,
325 				   __inet6_check_established);
326 }
327 EXPORT_SYMBOL_GPL(inet6_hash_connect);
328 
329 int inet6_hash(struct sock *sk)
330 {
331 	int err = 0;
332 
333 	if (sk->sk_state != TCP_CLOSE)
334 		err = __inet_hash(sk, NULL);
335 
336 	return err;
337 }
338 EXPORT_SYMBOL_GPL(inet6_hash);
339