Commit | Line | Data |
---|---|---|
2874c5fd | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
5324a040 ACM |
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 | * | |
d8313f5c | 9 | * Authors: Lotsa people, from code originally in tcp, generalised here |
67ba4152 | 10 | * by Arnaldo Carvalho de Melo <acme@mandriva.com> |
5324a040 ACM |
11 | */ |
12 | ||
5324a040 | 13 | #include <linux/module.h> |
d8313f5c | 14 | #include <linux/random.h> |
5324a040 | 15 | |
c125e80b | 16 | #include <net/addrconf.h> |
5af674bb | 17 | #include <net/hotdata.h> |
5324a040 ACM |
18 | #include <net/inet_connection_sock.h> |
19 | #include <net/inet_hashtables.h> | |
20 | #include <net/inet6_hashtables.h> | |
6e5714ea | 21 | #include <net/secure_seq.h> |
d8313f5c | 22 | #include <net/ip.h> |
c125e80b | 23 | #include <net/sock_reuseport.h> |
383eed2d | 24 | #include <net/tcp.h> |
5324a040 | 25 | |
d1e559d0 ED |
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) | |
b50026b5 | 29 | { |
1bbdceef HFS |
30 | u32 lhash, fhash; |
31 | ||
32 | net_get_random_once(&inet6_ehash_secret, sizeof(inet6_ehash_secret)); | |
df51b845 | 33 | net_get_random_once(&tcp_ipv6_hash_secret, sizeof(tcp_ipv6_hash_secret)); |
1bbdceef HFS |
34 | |
35 | lhash = (__force u32)laddr->s6_addr32[3]; | |
df51b845 | 36 | fhash = __ipv6_addr_jhash(faddr, tcp_ipv6_hash_secret); |
1bbdceef | 37 | |
9544d60a ED |
38 | return lport + __inet6_ehashfn(lhash, 0, fhash, fport, |
39 | inet6_ehash_secret + net_hash_mix(net)); | |
b50026b5 | 40 | } |
0f495f76 | 41 | EXPORT_SYMBOL_GPL(inet6_ehashfn); |
b50026b5 | 42 | |
b1a7ffcb DV |
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 | */ | |
10b2a44c | 49 | struct sock *__inet6_lookup_established(const struct net *net, |
d86e0dac | 50 | struct inet_hashinfo *hashinfo, |
b1a7ffcb | 51 | const struct in6_addr *saddr, |
d2ecd9cc | 52 | const __be16 sport, |
b1a7ffcb DV |
53 | const struct in6_addr *daddr, |
54 | const u16 hnum, | |
4297a0ef | 55 | const int dif, const int sdif) |
b1a7ffcb DV |
56 | { |
57 | struct sock *sk; | |
3ab5aee7 | 58 | const struct hlist_nulls_node *node; |
4f765d84 | 59 | const __portpair ports = INET_COMBINED_PORTS(sport, hnum); |
b1a7ffcb DV |
60 | /* Optimize here for direct hit, only listening connections can |
61 | * have wildcards anyways. | |
62 | */ | |
33de014c | 63 | unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport); |
f373b53b | 64 | unsigned int slot = hash & hashinfo->ehash_mask; |
3ab5aee7 | 65 | struct inet_ehash_bucket *head = &hashinfo->ehash[slot]; |
b1a7ffcb | 66 | |
3ab5aee7 | 67 | |
3ab5aee7 ED |
68 | begin: |
69 | sk_nulls_for_each_rcu(sk, node, &head->chain) { | |
ce43b03e ED |
70 | if (sk->sk_hash != hash) |
71 | continue; | |
5d368f03 | 72 | if (!inet6_match(net, sk, saddr, daddr, ports, dif, sdif)) |
efe4208f | 73 | continue; |
41c6d650 | 74 | if (unlikely(!refcount_inc_not_zero(&sk->sk_refcnt))) |
05dbc7b5 ED |
75 | goto out; |
76 | ||
5d368f03 | 77 | if (unlikely(!inet6_match(net, sk, saddr, daddr, ports, dif, sdif))) { |
efe4208f ED |
78 | sock_gen_put(sk); |
79 | goto begin; | |
3ab5aee7 | 80 | } |
efe4208f | 81 | goto found; |
b1a7ffcb | 82 | } |
3ab5aee7 ED |
83 | if (get_nulls_value(node) != slot) |
84 | goto begin; | |
3ab5aee7 | 85 | out: |
05dbc7b5 ED |
86 | sk = NULL; |
87 | found: | |
b1a7ffcb DV |
88 | return sk; |
89 | } | |
90 | EXPORT_SYMBOL(__inet6_lookup_established); | |
91 | ||
a2dc7bee | 92 | static inline int compute_score(struct sock *sk, const struct net *net, |
c25eb3bf ED |
93 | const unsigned short hnum, |
94 | const struct in6_addr *daddr, | |
3f7d820b | 95 | const int dif, const int sdif) |
c25eb3bf ED |
96 | { |
97 | int score = -1; | |
98 | ||
c720c7e8 | 99 | if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum && |
c25eb3bf | 100 | sk->sk_family == PF_INET6) { |
0ee58dad PO |
101 | if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) |
102 | return -1; | |
c25eb3bf | 103 | |
0ee58dad | 104 | if (!inet_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif)) |
e7819058 | 105 | return -1; |
4297a0ef | 106 | |
8d6c414c | 107 | score = sk->sk_bound_dev_if ? 2 : 1; |
7170a977 | 108 | if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) |
70da268b | 109 | score++; |
c25eb3bf ED |
110 | } |
111 | return score; | |
112 | } | |
113 | ||
2a617763 LB |
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 | */ | |
10b2a44c | 129 | struct sock *inet6_lookup_reuseport(const struct net *net, struct sock *sk, |
ce796e60 LB |
130 | struct sk_buff *skb, int doff, |
131 | const struct in6_addr *saddr, | |
132 | __be16 sport, | |
133 | const struct in6_addr *daddr, | |
0f495f76 LB |
134 | unsigned short hnum, |
135 | inet6_ehashfn_t *ehashfn) | |
5df65312 JS |
136 | { |
137 | struct sock *reuse_sk = NULL; | |
138 | u32 phash; | |
139 | ||
140 | if (sk->sk_reuseport) { | |
0f495f76 LB |
141 | phash = INDIRECT_CALL_INET(ehashfn, udp6_ehashfn, inet6_ehashfn, |
142 | net, daddr, hnum, saddr, sport); | |
5df65312 JS |
143 | reuse_sk = reuseport_select_sock(sk, phash, skb, doff); |
144 | } | |
145 | return reuse_sk; | |
146 | } | |
ce796e60 | 147 | EXPORT_SYMBOL_GPL(inet6_lookup_reuseport); |
5df65312 | 148 | |
3b24d854 | 149 | /* called with rcu_read_lock() */ |
10b2a44c | 150 | static struct sock *inet6_lhash2_lookup(const struct net *net, |
61b7c691 MKL |
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 | { | |
61b7c691 | 157 | struct sock *sk, *result = NULL; |
cae3873c | 158 | struct hlist_nulls_node *node; |
61b7c691 | 159 | int score, hiscore = 0; |
61b7c691 | 160 | |
cae3873c | 161 | sk_nulls_for_each_rcu(sk, node, &ilb2->nulls_head) { |
3f7d820b | 162 | score = compute_score(sk, net, hnum, daddr, dif, sdif); |
61b7c691 | 163 | if (score > hiscore) { |
ce796e60 | 164 | result = inet6_lookup_reuseport(net, sk, skb, doff, |
0f495f76 | 165 | saddr, sport, daddr, hnum, inet6_ehashfn); |
5df65312 JS |
166 | if (result) |
167 | return result; | |
168 | ||
61b7c691 MKL |
169 | result = sk; |
170 | hiscore = score; | |
171 | } | |
172 | } | |
173 | ||
174 | return result; | |
175 | } | |
176 | ||
10b2a44c | 177 | struct sock *inet6_lookup_run_sk_lookup(const struct net *net, |
6c886db2 LB |
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) | |
1122702f JS |
185 | { |
186 | struct sock *sk, *reuse_sk; | |
187 | bool no_reuseport; | |
188 | ||
6c886db2 | 189 | no_reuseport = bpf_sk_lookup_run_v6(net, protocol, saddr, sport, |
f8931565 | 190 | daddr, hnum, dif, &sk); |
1122702f JS |
191 | if (no_reuseport || IS_ERR_OR_NULL(sk)) |
192 | return sk; | |
193 | ||
0f495f76 | 194 | reuse_sk = inet6_lookup_reuseport(net, sk, skb, doff, |
6c886db2 | 195 | saddr, sport, daddr, hnum, ehashfn); |
1122702f JS |
196 | if (reuse_sk) |
197 | sk = reuse_sk; | |
198 | return sk; | |
199 | } | |
6c886db2 | 200 | EXPORT_SYMBOL_GPL(inet6_lookup_run_sk_lookup); |
1122702f | 201 | |
10b2a44c | 202 | struct sock *inet6_lookup_listener(const struct net *net, |
a583636a CG |
203 | struct inet_hashinfo *hashinfo, |
204 | struct sk_buff *skb, int doff, | |
205 | const struct in6_addr *saddr, | |
5ba24953 | 206 | const __be16 sport, const struct in6_addr *daddr, |
4297a0ef | 207 | const unsigned short hnum, const int dif, const int sdif) |
5324a040 | 208 | { |
61b7c691 | 209 | struct inet_listen_hashbucket *ilb2; |
0ee58dad | 210 | struct sock *result = NULL; |
61b7c691 | 211 | unsigned int hash2; |
61b7c691 | 212 | |
1122702f | 213 | /* Lookup redirect from BPF */ |
6c886db2 LB |
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); | |
1122702f JS |
219 | if (result) |
220 | goto done; | |
221 | } | |
222 | ||
61b7c691 MKL |
223 | hash2 = ipv6_portaddr_hash(net, daddr, hnum); |
224 | ilb2 = inet_lhash2_bucket(hashinfo, hash2); | |
61b7c691 MKL |
225 | |
226 | result = inet6_lhash2_lookup(net, ilb2, skb, doff, | |
227 | saddr, sport, daddr, hnum, | |
228 | dif, sdif); | |
229 | if (result) | |
8217ca65 | 230 | goto done; |
61b7c691 MKL |
231 | |
232 | /* Lookup lhash2 with in6addr_any */ | |
61b7c691 MKL |
233 | hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum); |
234 | ilb2 = inet_lhash2_bucket(hashinfo, hash2); | |
61b7c691 | 235 | |
8217ca65 | 236 | result = inet6_lhash2_lookup(net, ilb2, skb, doff, |
0ee58dad | 237 | saddr, sport, &in6addr_any, hnum, |
8217ca65 | 238 | dif, sdif); |
8217ca65 | 239 | done: |
26f8113c | 240 | if (IS_ERR(result)) |
8217ca65 | 241 | return NULL; |
5324a040 ACM |
242 | return result; |
243 | } | |
5324a040 ACM |
244 | EXPORT_SYMBOL_GPL(inet6_lookup_listener); |
245 | ||
10b2a44c ED |
246 | struct sock *inet6_lookup(const struct net *net, |
247 | struct inet_hashinfo *hashinfo, | |
a583636a | 248 | struct sk_buff *skb, int doff, |
d2ecd9cc AV |
249 | const struct in6_addr *saddr, const __be16 sport, |
250 | const struct in6_addr *daddr, const __be16 dport, | |
5324a040 ACM |
251 | const int dif) |
252 | { | |
253 | struct sock *sk; | |
3b24d854 | 254 | bool refcounted; |
5324a040 | 255 | |
a583636a | 256 | sk = __inet6_lookup(net, hashinfo, skb, doff, saddr, sport, daddr, |
4297a0ef | 257 | ntohs(dport), dif, 0, &refcounted); |
41c6d650 | 258 | if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt)) |
3b24d854 | 259 | sk = NULL; |
5324a040 ACM |
260 | return sk; |
261 | } | |
5324a040 | 262 | EXPORT_SYMBOL_GPL(inet6_lookup); |
d8313f5c ACM |
263 | |
264 | static int __inet6_check_established(struct inet_timewait_death_row *death_row, | |
265 | struct sock *sk, const __u16 lport, | |
86c2bc29 | 266 | struct inet_timewait_sock **twp, |
d4438ce6 ED |
267 | bool rcu_lookup, |
268 | u32 hash) | |
d8313f5c ACM |
269 | { |
270 | struct inet_hashinfo *hinfo = death_row->hashinfo; | |
3759fa9c | 271 | struct inet_sock *inet = inet_sk(sk); |
efe4208f ED |
272 | const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr; |
273 | const struct in6_addr *saddr = &sk->sk_v6_daddr; | |
d8313f5c | 274 | const int dif = sk->sk_bound_dev_if; |
33de014c | 275 | struct net *net = sock_net(sk); |
4297a0ef DA |
276 | const int sdif = l3mdev_master_ifindex_by_index(net, dif); |
277 | const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport); | |
d8313f5c | 278 | struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash); |
05dbc7b5 | 279 | struct inet_timewait_sock *tw = NULL; |
ae9d5b19 ED |
280 | const struct hlist_nulls_node *node; |
281 | struct sock *sk2; | |
282 | spinlock_t *lock; | |
283 | ||
86c2bc29 ED |
284 | if (rcu_lookup) { |
285 | sk_nulls_for_each(sk2, node, &head->chain) { | |
286 | if (sk2->sk_hash != hash || | |
287 | !inet6_match(net, 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; | |
ae9d5b19 | 295 | } |
d8313f5c | 296 | |
ae9d5b19 | 297 | lock = inet_ehash_lockp(hinfo, hash); |
9db66bdc | 298 | spin_lock(lock); |
d8313f5c | 299 | |
05dbc7b5 | 300 | sk_nulls_for_each(sk2, node, &head->chain) { |
ce43b03e ED |
301 | if (sk2->sk_hash != hash) |
302 | continue; | |
d8313f5c | 303 | |
5d368f03 | 304 | if (likely(inet6_match(net, sk2, saddr, daddr, ports, |
4297a0ef | 305 | dif, sdif))) { |
efe4208f | 306 | if (sk2->sk_state == TCP_TIME_WAIT) { |
05dbc7b5 | 307 | tw = inet_twsk(sk2); |
383eed2d ED |
308 | if (sk->sk_protocol == IPPROTO_TCP && |
309 | tcp_twsk_unique(sk, sk2, twp)) | |
efe4208f | 310 | break; |
05dbc7b5 | 311 | } |
d8313f5c | 312 | goto not_unique; |
efe4208f | 313 | } |
d8313f5c ACM |
314 | } |
315 | ||
3759fa9c | 316 | /* Must record num and sport now. Otherwise we will see |
efe4208f ED |
317 | * in hash table socket with a funny identity. |
318 | */ | |
c720c7e8 ED |
319 | inet->inet_num = lport; |
320 | inet->inet_sport = htons(lport); | |
13475a30 | 321 | sk->sk_hash = hash; |
547b792c | 322 | WARN_ON(!sk_unhashed(sk)); |
3ab5aee7 | 323 | __sk_nulls_add_node_rcu(sk, &head->chain); |
13475a30 | 324 | if (tw) { |
fc01538f | 325 | sk_nulls_del_node_init_rcu((struct sock *)tw); |
02a1d6e7 | 326 | __NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED); |
13475a30 | 327 | } |
9db66bdc | 328 | spin_unlock(lock); |
c29a0bc4 | 329 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); |
d8313f5c | 330 | |
13475a30 | 331 | if (twp) { |
d8313f5c | 332 | *twp = tw; |
13475a30 | 333 | } else if (tw) { |
d8313f5c | 334 | /* Silly. Should hash-dance instead... */ |
dbe7faa4 | 335 | inet_twsk_deschedule_put(tw); |
d8313f5c ACM |
336 | } |
337 | return 0; | |
338 | ||
339 | not_unique: | |
9db66bdc | 340 | spin_unlock(lock); |
d8313f5c ACM |
341 | return -EADDRNOTAVAIL; |
342 | } | |
343 | ||
b2d05756 | 344 | static u64 inet6_sk_port_offset(const struct sock *sk) |
d8313f5c ACM |
345 | { |
346 | const struct inet_sock *inet = inet_sk(sk); | |
efe4208f ED |
347 | |
348 | return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32, | |
349 | sk->sk_v6_daddr.s6_addr32, | |
c720c7e8 | 350 | inet->inet_dport); |
d8313f5c ACM |
351 | } |
352 | ||
353 | int inet6_hash_connect(struct inet_timewait_death_row *death_row, | |
354 | struct sock *sk) | |
355 | { | |
d4438ce6 ED |
356 | const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr; |
357 | const struct in6_addr *saddr = &sk->sk_v6_daddr; | |
358 | const struct inet_sock *inet = inet_sk(sk); | |
359 | const struct net *net = sock_net(sk); | |
b2d05756 | 360 | u64 port_offset = 0; |
d4438ce6 | 361 | u32 hash_port0; |
e2baad9e ED |
362 | |
363 | if (!inet_sk(sk)->inet_num) | |
364 | port_offset = inet6_sk_port_offset(sk); | |
d4438ce6 ED |
365 | |
366 | hash_port0 = inet6_ehashfn(net, daddr, 0, saddr, inet->inet_dport); | |
367 | ||
368 | return __inet_hash_connect(death_row, sk, port_offset, hash_port0, | |
b4d6444e | 369 | __inet6_check_established); |
d8313f5c | 370 | } |
d8313f5c | 371 | EXPORT_SYMBOL_GPL(inet6_hash_connect); |
496611d7 CG |
372 | |
373 | int inet6_hash(struct sock *sk) | |
374 | { | |
e4cabca5 CG |
375 | int err = 0; |
376 | ||
4f9bf2a2 | 377 | if (sk->sk_state != TCP_CLOSE) |
fe38d2a1 | 378 | err = __inet_hash(sk, NULL); |
496611d7 | 379 | |
e4cabca5 | 380 | return err; |
496611d7 CG |
381 | } |
382 | EXPORT_SYMBOL_GPL(inet6_hash); |