1 // SPDX-License-Identifier: GPL-2.0-or-later
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.
7 * Generic INET6 transport hashtables
9 * Authors: Lotsa people, from code originally in tcp, generalised here
10 * by Arnaldo Carvalho de Melo <acme@mandriva.com>
13 #include <linux/module.h>
14 #include <linux/random.h>
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>
22 #include <net/sock_reuseport.h>
24 extern struct inet_hashinfo tcp_hashinfo;
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)
30 static u32 inet6_ehash_secret __read_mostly;
31 static u32 ipv6_hash_secret __read_mostly;
35 net_get_random_once(&inet6_ehash_secret, sizeof(inet6_ehash_secret));
36 net_get_random_once(&ipv6_hash_secret, sizeof(ipv6_hash_secret));
38 lhash = (__force u32)laddr->s6_addr32[3];
39 fhash = __ipv6_addr_jhash(faddr, ipv6_hash_secret);
41 return __inet6_ehashfn(lhash, lport, fhash, fport,
42 inet6_ehash_secret + net_hash_mix(net));
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
49 * The sockhash lock must be held as a reader here.
51 struct sock *__inet6_lookup_established(struct net *net,
52 struct inet_hashinfo *hashinfo,
53 const struct in6_addr *saddr,
55 const struct in6_addr *daddr,
57 const int dif, const int sdif)
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.
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];
71 sk_nulls_for_each_rcu(sk, node, &head->chain) {
72 if (sk->sk_hash != hash)
74 if (!inet6_match(net, sk, saddr, daddr, ports, dif, sdif))
76 if (unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
79 if (unlikely(!inet6_match(net, sk, saddr, daddr, ports, dif, sdif))) {
85 if (get_nulls_value(node) != slot)
92 EXPORT_SYMBOL(__inet6_lookup_established);
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)
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))
106 if (!inet_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
109 score = sk->sk_bound_dev_if ? 2 : 1;
110 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
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,
120 const struct in6_addr *daddr,
123 struct sock *reuse_sk = NULL;
126 if (sk->sk_reuseport) {
127 phash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
128 reuse_sk = reuseport_select_sock(sk, phash, skb, doff);
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)
141 struct sock *sk, *result = NULL;
142 struct hlist_nulls_node *node;
143 int score, hiscore = 0;
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);
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,
166 const struct in6_addr *daddr,
167 const u16 hnum, const int dif)
169 struct sock *sk, *reuse_sk;
172 if (hashinfo != &tcp_hashinfo)
173 return NULL; /* only TCP is supported */
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))
180 reuse_sk = lookup_reuseport(net, sk, skb, doff, saddr, sport, daddr, hnum);
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)
193 struct inet_listen_hashbucket *ilb2;
194 struct sock *result = NULL;
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);
205 hash2 = ipv6_portaddr_hash(net, daddr, hnum);
206 ilb2 = inet_lhash2_bucket(hashinfo, hash2);
208 result = inet6_lhash2_lookup(net, ilb2, skb, doff,
209 saddr, sport, daddr, hnum,
214 /* Lookup lhash2 with in6addr_any */
215 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
216 ilb2 = inet_lhash2_bucket(hashinfo, hash2);
218 result = inet6_lhash2_lookup(net, ilb2, skb, doff,
219 saddr, sport, &in6addr_any, hnum,
226 EXPORT_SYMBOL_GPL(inet6_lookup_listener);
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,
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))
243 EXPORT_SYMBOL_GPL(inet6_lookup);
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)
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,
259 struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
260 spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
262 const struct hlist_nulls_node *node;
263 struct inet_timewait_sock *tw = NULL;
267 sk_nulls_for_each(sk2, node, &head->chain) {
268 if (sk2->sk_hash != hash)
271 if (likely(inet6_match(net, sk2, saddr, daddr, ports,
273 if (sk2->sk_state == TCP_TIME_WAIT) {
275 if (twsk_unique(sk, sk2, twp))
282 /* Must record num and sport now. Otherwise we will see
283 * in hash table socket with a funny identity.
285 inet->inet_num = lport;
286 inet->inet_sport = htons(lport);
288 WARN_ON(!sk_unhashed(sk));
289 __sk_nulls_add_node_rcu(sk, &head->chain);
291 sk_nulls_del_node_init_rcu((struct sock *)tw);
292 __NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED);
295 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
300 /* Silly. Should hash-dance instead... */
301 inet_twsk_deschedule_put(tw);
307 return -EADDRNOTAVAIL;
310 static u64 inet6_sk_port_offset(const struct sock *sk)
312 const struct inet_sock *inet = inet_sk(sk);
314 return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32,
315 sk->sk_v6_daddr.s6_addr32,
319 int inet6_hash_connect(struct inet_timewait_death_row *death_row,
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);
329 EXPORT_SYMBOL_GPL(inet6_hash_connect);
331 int inet6_hash(struct sock *sk)
335 if (sk->sk_state != TCP_CLOSE)
336 err = __inet_hash(sk, NULL);
340 EXPORT_SYMBOL_GPL(inet6_hash);