xref: /openbmc/linux/net/ipv6/inet6_hashtables.c (revision a5b2c10c)
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(sk, net, 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(sk, net, 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, bool exact_dif)
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 = 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 	bool exact_dif = inet6_exact_dif_match(net, skb);
142 	struct inet_connection_sock *icsk;
143 	struct sock *sk, *result = NULL;
144 	int score, hiscore = 0;
145 
146 	inet_lhash2_for_each_icsk_rcu(icsk, &ilb2->head) {
147 		sk = (struct sock *)icsk;
148 		score = compute_score(sk, net, hnum, daddr, dif, sdif,
149 				      exact_dif);
150 		if (score > hiscore) {
151 			result = lookup_reuseport(net, sk, skb, doff,
152 						  saddr, sport, daddr, hnum);
153 			if (result)
154 				return result;
155 
156 			result = sk;
157 			hiscore = score;
158 		}
159 	}
160 
161 	return result;
162 }
163 
164 static inline struct sock *inet6_lookup_run_bpf(struct net *net,
165 						struct inet_hashinfo *hashinfo,
166 						struct sk_buff *skb, int doff,
167 						const struct in6_addr *saddr,
168 						const __be16 sport,
169 						const struct in6_addr *daddr,
170 						const u16 hnum)
171 {
172 	struct sock *sk, *reuse_sk;
173 	bool no_reuseport;
174 
175 	if (hashinfo != &tcp_hashinfo)
176 		return NULL; /* only TCP is supported */
177 
178 	no_reuseport = bpf_sk_lookup_run_v6(net, IPPROTO_TCP,
179 					    saddr, sport, daddr, hnum, &sk);
180 	if (no_reuseport || IS_ERR_OR_NULL(sk))
181 		return sk;
182 
183 	reuse_sk = lookup_reuseport(net, sk, skb, doff, saddr, sport, daddr, hnum);
184 	if (reuse_sk)
185 		sk = reuse_sk;
186 	return sk;
187 }
188 
189 struct sock *inet6_lookup_listener(struct net *net,
190 		struct inet_hashinfo *hashinfo,
191 		struct sk_buff *skb, int doff,
192 		const struct in6_addr *saddr,
193 		const __be16 sport, const struct in6_addr *daddr,
194 		const unsigned short hnum, const int dif, const int sdif)
195 {
196 	struct inet_listen_hashbucket *ilb2;
197 	struct sock *result = NULL;
198 	unsigned int hash2;
199 
200 	/* Lookup redirect from BPF */
201 	if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
202 		result = inet6_lookup_run_bpf(net, hashinfo, skb, doff,
203 					      saddr, sport, daddr, hnum);
204 		if (result)
205 			goto done;
206 	}
207 
208 	hash2 = ipv6_portaddr_hash(net, daddr, hnum);
209 	ilb2 = inet_lhash2_bucket(hashinfo, hash2);
210 
211 	result = inet6_lhash2_lookup(net, ilb2, skb, doff,
212 				     saddr, sport, daddr, hnum,
213 				     dif, sdif);
214 	if (result)
215 		goto done;
216 
217 	/* Lookup lhash2 with in6addr_any */
218 	hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
219 	ilb2 = inet_lhash2_bucket(hashinfo, hash2);
220 
221 	result = inet6_lhash2_lookup(net, ilb2, skb, doff,
222 				     saddr, sport, &in6addr_any, hnum,
223 				     dif, sdif);
224 done:
225 	if (IS_ERR(result))
226 		return NULL;
227 	return result;
228 }
229 EXPORT_SYMBOL_GPL(inet6_lookup_listener);
230 
231 struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
232 			  struct sk_buff *skb, int doff,
233 			  const struct in6_addr *saddr, const __be16 sport,
234 			  const struct in6_addr *daddr, const __be16 dport,
235 			  const int dif)
236 {
237 	struct sock *sk;
238 	bool refcounted;
239 
240 	sk = __inet6_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
241 			    ntohs(dport), dif, 0, &refcounted);
242 	if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
243 		sk = NULL;
244 	return sk;
245 }
246 EXPORT_SYMBOL_GPL(inet6_lookup);
247 
248 static int __inet6_check_established(struct inet_timewait_death_row *death_row,
249 				     struct sock *sk, const __u16 lport,
250 				     struct inet_timewait_sock **twp)
251 {
252 	struct inet_hashinfo *hinfo = death_row->hashinfo;
253 	struct inet_sock *inet = inet_sk(sk);
254 	const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr;
255 	const struct in6_addr *saddr = &sk->sk_v6_daddr;
256 	const int dif = sk->sk_bound_dev_if;
257 	struct net *net = sock_net(sk);
258 	const int sdif = l3mdev_master_ifindex_by_index(net, dif);
259 	const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
260 	const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
261 						inet->inet_dport);
262 	struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
263 	spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
264 	struct sock *sk2;
265 	const struct hlist_nulls_node *node;
266 	struct inet_timewait_sock *tw = NULL;
267 
268 	spin_lock(lock);
269 
270 	sk_nulls_for_each(sk2, node, &head->chain) {
271 		if (sk2->sk_hash != hash)
272 			continue;
273 
274 		if (likely(INET6_MATCH(sk2, net, saddr, daddr, ports,
275 				       dif, sdif))) {
276 			if (sk2->sk_state == TCP_TIME_WAIT) {
277 				tw = inet_twsk(sk2);
278 				if (twsk_unique(sk, sk2, twp))
279 					break;
280 			}
281 			goto not_unique;
282 		}
283 	}
284 
285 	/* Must record num and sport now. Otherwise we will see
286 	 * in hash table socket with a funny identity.
287 	 */
288 	inet->inet_num = lport;
289 	inet->inet_sport = htons(lport);
290 	sk->sk_hash = hash;
291 	WARN_ON(!sk_unhashed(sk));
292 	__sk_nulls_add_node_rcu(sk, &head->chain);
293 	if (tw) {
294 		sk_nulls_del_node_init_rcu((struct sock *)tw);
295 		__NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED);
296 	}
297 	spin_unlock(lock);
298 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
299 
300 	if (twp) {
301 		*twp = tw;
302 	} else if (tw) {
303 		/* Silly. Should hash-dance instead... */
304 		inet_twsk_deschedule_put(tw);
305 	}
306 	return 0;
307 
308 not_unique:
309 	spin_unlock(lock);
310 	return -EADDRNOTAVAIL;
311 }
312 
313 static u32 inet6_sk_port_offset(const struct sock *sk)
314 {
315 	const struct inet_sock *inet = inet_sk(sk);
316 
317 	return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32,
318 					  sk->sk_v6_daddr.s6_addr32,
319 					  inet->inet_dport);
320 }
321 
322 int inet6_hash_connect(struct inet_timewait_death_row *death_row,
323 		       struct sock *sk)
324 {
325 	u32 port_offset = 0;
326 
327 	if (!inet_sk(sk)->inet_num)
328 		port_offset = inet6_sk_port_offset(sk);
329 	return __inet_hash_connect(death_row, sk, port_offset,
330 				   __inet6_check_established);
331 }
332 EXPORT_SYMBOL_GPL(inet6_hash_connect);
333 
334 int inet6_hash(struct sock *sk)
335 {
336 	int err = 0;
337 
338 	if (sk->sk_state != TCP_CLOSE) {
339 		local_bh_disable();
340 		err = __inet_hash(sk, NULL);
341 		local_bh_enable();
342 	}
343 
344 	return err;
345 }
346 EXPORT_SYMBOL_GPL(inet6_hash);
347