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
26u32 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(a: faddr, tcp_ipv6_hash_secret);
37
38 return lport + __inet6_ehashfn(lhash, lport: 0, fhash, fport,
39 inet6_ehash_secret + net_hash_mix(net));
40}
41EXPORT_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 */
49struct sock *__inet6_lookup_established(const struct net *net,
50 const struct in6_addr *saddr,
51 const __be16 sport,
52 const struct in6_addr *daddr,
53 const u16 hnum,
54 const int dif, const int sdif)
55{
56 const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
57 const struct hlist_nulls_node *node;
58 struct inet_ehash_bucket *head;
59 struct inet_hashinfo *hashinfo;
60 unsigned int hash, slot;
61 struct sock *sk;
62
63 hashinfo = net->ipv4.tcp_death_row.hashinfo;
64 hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
65 slot = hash & hashinfo->ehash_mask;
66 head = &hashinfo->ehash[slot];
67begin:
68 sk_nulls_for_each_rcu(sk, node, &head->chain) {
69 if (sk->sk_hash != hash)
70 continue;
71 if (!inet6_match(net, sk, saddr, daddr, ports, dif, sdif))
72 continue;
73 if (unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
74 goto out;
75
76 if (unlikely(!inet6_match(net, sk, saddr, daddr, ports, dif, sdif))) {
77 sock_gen_put(sk);
78 goto begin;
79 }
80 goto found;
81 }
82 if (get_nulls_value(ptr: node) != slot)
83 goto begin;
84out:
85 sk = NULL;
86found:
87 return sk;
88}
89EXPORT_SYMBOL(__inet6_lookup_established);
90
91static inline int compute_score(struct sock *sk, const struct net *net,
92 const unsigned short hnum,
93 const struct in6_addr *daddr,
94 const int dif, const int sdif)
95{
96 int score = -1;
97
98 if (net_eq(net1: sock_net(sk), net2: net) && inet_sk(sk)->inet_num == hnum &&
99 sk->sk_family == PF_INET6) {
100 if (!ipv6_addr_equal(a1: &sk->sk_v6_rcv_saddr, a2: daddr))
101 return -1;
102
103 if (!inet_sk_bound_dev_eq(net, bound_dev_if: sk->sk_bound_dev_if, dif, sdif))
104 return -1;
105
106 score = sk->sk_bound_dev_if ? 2 : 1;
107 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
108 score++;
109 }
110 return score;
111}
112
113/**
114 * inet6_lookup_reuseport() - execute reuseport logic on AF_INET6 socket if necessary.
115 * @net: network namespace.
116 * @sk: AF_INET6 socket, must be in TCP_LISTEN state for TCP or TCP_CLOSE for UDP.
117 * @skb: context for a potential SK_REUSEPORT program.
118 * @doff: header offset.
119 * @saddr: source address.
120 * @sport: source port.
121 * @daddr: destination address.
122 * @hnum: destination port in host byte order.
123 * @ehashfn: hash function used to generate the fallback hash.
124 *
125 * Return: NULL if sk doesn't have SO_REUSEPORT set, otherwise a pointer to
126 * the selected sock or an error.
127 */
128struct sock *inet6_lookup_reuseport(const struct net *net, struct sock *sk,
129 struct sk_buff *skb, int doff,
130 const struct in6_addr *saddr,
131 __be16 sport,
132 const struct in6_addr *daddr,
133 unsigned short hnum,
134 inet6_ehashfn_t *ehashfn)
135{
136 struct sock *reuse_sk = NULL;
137 u32 phash;
138
139 if (sk->sk_reuseport) {
140 phash = INDIRECT_CALL_INET(ehashfn, udp6_ehashfn, inet6_ehashfn,
141 net, daddr, hnum, saddr, sport);
142 reuse_sk = reuseport_select_sock(sk, hash: phash, skb, hdr_len: doff);
143 }
144 return reuse_sk;
145}
146EXPORT_SYMBOL_GPL(inet6_lookup_reuseport);
147
148/* called with rcu_read_lock() */
149static struct sock *inet6_lhash2_lookup(const struct net *net,
150 struct inet_listen_hashbucket *ilb2,
151 struct sk_buff *skb, int doff,
152 const struct in6_addr *saddr,
153 const __be16 sport, const struct in6_addr *daddr,
154 const unsigned short hnum, const int dif, const int sdif)
155{
156 struct sock *sk, *result = NULL;
157 struct hlist_nulls_node *node;
158 int score, hiscore = 0;
159
160 sk_nulls_for_each_rcu(sk, node, &ilb2->nulls_head) {
161 score = compute_score(sk, net, hnum, daddr, dif, sdif);
162 if (score > hiscore) {
163 result = inet6_lookup_reuseport(net, sk, skb, doff,
164 saddr, sport, daddr, hnum, inet6_ehashfn);
165 if (result)
166 return result;
167
168 result = sk;
169 hiscore = score;
170 }
171 }
172
173 return result;
174}
175
176struct sock *inet6_lookup_run_sk_lookup(const struct net *net,
177 int protocol,
178 struct sk_buff *skb, int doff,
179 const struct in6_addr *saddr,
180 const __be16 sport,
181 const struct in6_addr *daddr,
182 const u16 hnum, const int dif,
183 inet6_ehashfn_t *ehashfn)
184{
185 struct sock *sk, *reuse_sk;
186 bool no_reuseport;
187
188 no_reuseport = bpf_sk_lookup_run_v6(net, protocol, saddr, sport,
189 daddr, dport: hnum, ifindex: dif, psk: &sk);
190 if (no_reuseport || IS_ERR_OR_NULL(ptr: sk))
191 return sk;
192
193 reuse_sk = inet6_lookup_reuseport(net, sk, skb, doff,
194 saddr, sport, daddr, hnum, ehashfn);
195 if (reuse_sk)
196 sk = reuse_sk;
197 return sk;
198}
199EXPORT_SYMBOL_GPL(inet6_lookup_run_sk_lookup);
200
201struct sock *inet6_lookup_listener(const struct net *net,
202 struct sk_buff *skb, int doff,
203 const struct in6_addr *saddr,
204 const __be16 sport,
205 const struct in6_addr *daddr,
206 const unsigned short hnum,
207 const int dif, const int sdif)
208{
209 struct inet_listen_hashbucket *ilb2;
210 struct inet_hashinfo *hashinfo;
211 struct sock *result = NULL;
212 unsigned int hash2;
213
214 /* Lookup redirect from BPF */
215 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
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 hashinfo = net->ipv4.tcp_death_row.hashinfo;
224 hash2 = ipv6_portaddr_hash(net, addr6: daddr, port: hnum);
225 ilb2 = inet_lhash2_bucket(h: hashinfo, hash: hash2);
226
227 result = inet6_lhash2_lookup(net, ilb2, skb, doff,
228 saddr, sport, daddr, hnum,
229 dif, sdif);
230 if (result)
231 goto done;
232
233 /* Lookup lhash2 with in6addr_any */
234 hash2 = ipv6_portaddr_hash(net, addr6: &in6addr_any, port: hnum);
235 ilb2 = inet_lhash2_bucket(h: hashinfo, hash: hash2);
236
237 result = inet6_lhash2_lookup(net, ilb2, skb, doff,
238 saddr, sport, daddr: &in6addr_any, hnum,
239 dif, sdif);
240done:
241 if (IS_ERR(ptr: result))
242 return NULL;
243 return result;
244}
245EXPORT_SYMBOL_GPL(inet6_lookup_listener);
246
247struct sock *inet6_lookup(const struct net *net,
248 struct sk_buff *skb, int doff,
249 const struct in6_addr *saddr, const __be16 sport,
250 const struct in6_addr *daddr, const __be16 dport,
251 const int dif)
252{
253 struct sock *sk;
254 bool refcounted;
255
256 sk = __inet6_lookup(net, skb, doff, saddr, sport, daddr,
257 ntohs(dport), dif, sdif: 0, refcounted: &refcounted);
258 if (sk && !refcounted && !refcount_inc_not_zero(r: &sk->sk_refcnt))
259 sk = NULL;
260 return sk;
261}
262EXPORT_SYMBOL_GPL(inet6_lookup);
263
264static int __inet6_check_established(struct inet_timewait_death_row *death_row,
265 struct sock *sk, const __u16 lport,
266 struct inet_timewait_sock **twp,
267 bool rcu_lookup,
268 u32 hash)
269{
270 struct inet_hashinfo *hinfo = death_row->hashinfo;
271 struct inet_sock *inet = inet_sk(sk);
272 const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr;
273 const struct in6_addr *saddr = &sk->sk_v6_daddr;
274 const int dif = sk->sk_bound_dev_if;
275 struct net *net = sock_net(sk);
276 const int sdif = l3mdev_master_ifindex_by_index(net, ifindex: dif);
277 const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
278 struct inet_ehash_bucket *head = inet_ehash_bucket(hashinfo: hinfo, hash);
279 struct inet_timewait_sock *tw = NULL;
280 const struct hlist_nulls_node *node;
281 struct sock *sk2;
282 spinlock_t *lock;
283
284 if (rcu_lookup) {
285 sk_nulls_for_each(sk2, node, &head->chain) {
286 if (sk2->sk_hash != hash ||
287 !inet6_match(net, sk: sk2, saddr, daddr,
288 ports, dif, sdif))
289 continue;
290 if (sk2->sk_state == TCP_TIME_WAIT)
291 break;
292 return -EADDRNOTAVAIL;
293 }
294 return 0;
295 }
296
297 lock = inet_ehash_lockp(hashinfo: hinfo, hash);
298 spin_lock(lock);
299
300 sk_nulls_for_each(sk2, node, &head->chain) {
301 if (sk2->sk_hash != hash)
302 continue;
303
304 if (likely(inet6_match(net, sk2, saddr, daddr, ports,
305 dif, sdif))) {
306 if (sk2->sk_state == TCP_TIME_WAIT) {
307 tw = inet_twsk(sk: sk2);
308 if (tcp_twsk_unique(sk, sktw: sk2, twp))
309 break;
310 }
311 goto not_unique;
312 }
313 }
314
315 /* Must record num and sport now. Otherwise we will see
316 * in hash table socket with a funny identity.
317 */
318 inet->inet_num = lport;
319 inet->inet_sport = htons(lport);
320 sk->sk_hash = hash;
321 WARN_ON(!sk_unhashed(sk));
322 __sk_nulls_add_node_rcu(sk, list: &head->chain);
323 if (tw) {
324 sk_nulls_del_node_init_rcu(sk: (struct sock *)tw);
325 __NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED);
326 }
327 spin_unlock(lock);
328 sock_prot_inuse_add(net: sock_net(sk), prot: sk->sk_prot, val: 1);
329
330 if (twp) {
331 *twp = tw;
332 } else if (tw) {
333 /* Silly. Should hash-dance instead... */
334 inet_twsk_deschedule_put(tw);
335 }
336 return 0;
337
338not_unique:
339 spin_unlock(lock);
340 return -EADDRNOTAVAIL;
341}
342
343static u64 inet6_sk_port_offset(const struct sock *sk)
344{
345 const struct inet_sock *inet = inet_sk(sk);
346
347 return secure_ipv6_port_ephemeral(saddr: sk->sk_v6_rcv_saddr.s6_addr32,
348 daddr: sk->sk_v6_daddr.s6_addr32,
349 dport: inet->inet_dport);
350}
351
352int inet6_hash_connect(struct inet_timewait_death_row *death_row,
353 struct sock *sk)
354{
355 const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr;
356 const struct in6_addr *saddr = &sk->sk_v6_daddr;
357 const struct inet_sock *inet = inet_sk(sk);
358 const struct net *net = sock_net(sk);
359 u64 port_offset = 0;
360 u32 hash_port0;
361
362 if (!inet_sk(sk)->inet_num)
363 port_offset = inet6_sk_port_offset(sk);
364
365 hash_port0 = inet6_ehashfn(net, daddr, 0, saddr, inet->inet_dport);
366
367 return __inet_hash_connect(death_row, sk, port_offset, hash_port0,
368 check_established: __inet6_check_established);
369}
370EXPORT_SYMBOL_GPL(inet6_hash_connect);
371

source code of linux/net/ipv6/inet6_hashtables.c