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> |
5324a040 ACM |
17 | #include <net/inet_connection_sock.h> |
18 | #include <net/inet_hashtables.h> | |
19 | #include <net/inet6_hashtables.h> | |
6e5714ea | 20 | #include <net/secure_seq.h> |
d8313f5c | 21 | #include <net/ip.h> |
c125e80b | 22 | #include <net/sock_reuseport.h> |
5324a040 | 23 | |
d1e559d0 ED |
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) | |
b50026b5 | 27 | { |
1bbdceef HFS |
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 | ||
b50026b5 | 39 | return __inet6_ehashfn(lhash, lport, fhash, fport, |
1bbdceef | 40 | inet6_ehash_secret + net_hash_mix(net)); |
b50026b5 HFS |
41 | } |
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 | */ | |
d86e0dac PE |
49 | struct sock *__inet6_lookup_established(struct net *net, |
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 | ||
42b16b3f | 92 | static inline int compute_score(struct sock *sk, 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 | ||
5df65312 JS |
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 | ||
3b24d854 | 131 | /* called with rcu_read_lock() */ |
61b7c691 MKL |
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 | { | |
61b7c691 | 139 | struct sock *sk, *result = NULL; |
cae3873c | 140 | struct hlist_nulls_node *node; |
61b7c691 | 141 | int score, hiscore = 0; |
61b7c691 | 142 | |
cae3873c | 143 | sk_nulls_for_each_rcu(sk, node, &ilb2->nulls_head) { |
3f7d820b | 144 | score = compute_score(sk, net, hnum, daddr, dif, sdif); |
61b7c691 | 145 | if (score > hiscore) { |
5df65312 JS |
146 | result = lookup_reuseport(net, sk, skb, doff, |
147 | saddr, sport, daddr, hnum); | |
148 | if (result) | |
149 | return result; | |
150 | ||
61b7c691 MKL |
151 | result = sk; |
152 | hiscore = score; | |
153 | } | |
154 | } | |
155 | ||
156 | return result; | |
157 | } | |
158 | ||
1122702f JS |
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, | |
f8931565 | 165 | const u16 hnum, const int dif) |
1122702f JS |
166 | { |
167 | struct sock *sk, *reuse_sk; | |
168 | bool no_reuseport; | |
169 | ||
4461568a | 170 | if (hashinfo != net->ipv4.tcp_death_row.hashinfo) |
1122702f JS |
171 | return NULL; /* only TCP is supported */ |
172 | ||
f8931565 MP |
173 | no_reuseport = bpf_sk_lookup_run_v6(net, IPPROTO_TCP, saddr, sport, |
174 | daddr, hnum, dif, &sk); | |
1122702f JS |
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 | ||
d86e0dac | 184 | struct sock *inet6_lookup_listener(struct net *net, |
a583636a CG |
185 | struct inet_hashinfo *hashinfo, |
186 | struct sk_buff *skb, int doff, | |
187 | const struct in6_addr *saddr, | |
5ba24953 | 188 | const __be16 sport, const struct in6_addr *daddr, |
4297a0ef | 189 | const unsigned short hnum, const int dif, const int sdif) |
5324a040 | 190 | { |
61b7c691 | 191 | struct inet_listen_hashbucket *ilb2; |
0ee58dad | 192 | struct sock *result = NULL; |
61b7c691 | 193 | unsigned int hash2; |
61b7c691 | 194 | |
1122702f JS |
195 | /* Lookup redirect from BPF */ |
196 | if (static_branch_unlikely(&bpf_sk_lookup_enabled)) { | |
197 | result = inet6_lookup_run_bpf(net, hashinfo, skb, doff, | |
f8931565 | 198 | saddr, sport, daddr, hnum, dif); |
1122702f JS |
199 | if (result) |
200 | goto done; | |
201 | } | |
202 | ||
61b7c691 MKL |
203 | hash2 = ipv6_portaddr_hash(net, daddr, hnum); |
204 | ilb2 = inet_lhash2_bucket(hashinfo, hash2); | |
61b7c691 MKL |
205 | |
206 | result = inet6_lhash2_lookup(net, ilb2, skb, doff, | |
207 | saddr, sport, daddr, hnum, | |
208 | dif, sdif); | |
209 | if (result) | |
8217ca65 | 210 | goto done; |
61b7c691 MKL |
211 | |
212 | /* Lookup lhash2 with in6addr_any */ | |
61b7c691 MKL |
213 | hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum); |
214 | ilb2 = inet_lhash2_bucket(hashinfo, hash2); | |
61b7c691 | 215 | |
8217ca65 | 216 | result = inet6_lhash2_lookup(net, ilb2, skb, doff, |
0ee58dad | 217 | saddr, sport, &in6addr_any, hnum, |
8217ca65 | 218 | dif, sdif); |
8217ca65 | 219 | done: |
26f8113c | 220 | if (IS_ERR(result)) |
8217ca65 | 221 | return NULL; |
5324a040 ACM |
222 | return result; |
223 | } | |
5324a040 ACM |
224 | EXPORT_SYMBOL_GPL(inet6_lookup_listener); |
225 | ||
d86e0dac | 226 | struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo, |
a583636a | 227 | struct sk_buff *skb, int doff, |
d2ecd9cc AV |
228 | const struct in6_addr *saddr, const __be16 sport, |
229 | const struct in6_addr *daddr, const __be16 dport, | |
5324a040 ACM |
230 | const int dif) |
231 | { | |
232 | struct sock *sk; | |
3b24d854 | 233 | bool refcounted; |
5324a040 | 234 | |
a583636a | 235 | sk = __inet6_lookup(net, hashinfo, skb, doff, saddr, sport, daddr, |
4297a0ef | 236 | ntohs(dport), dif, 0, &refcounted); |
41c6d650 | 237 | if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt)) |
3b24d854 | 238 | sk = NULL; |
5324a040 ACM |
239 | return sk; |
240 | } | |
5324a040 | 241 | EXPORT_SYMBOL_GPL(inet6_lookup); |
d8313f5c ACM |
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; | |
3759fa9c | 248 | struct inet_sock *inet = inet_sk(sk); |
efe4208f ED |
249 | const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr; |
250 | const struct in6_addr *saddr = &sk->sk_v6_daddr; | |
d8313f5c | 251 | const int dif = sk->sk_bound_dev_if; |
33de014c | 252 | struct net *net = sock_net(sk); |
4297a0ef DA |
253 | const int sdif = l3mdev_master_ifindex_by_index(net, dif); |
254 | const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport); | |
33de014c | 255 | const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr, |
c720c7e8 | 256 | inet->inet_dport); |
d8313f5c | 257 | struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash); |
9db66bdc | 258 | spinlock_t *lock = inet_ehash_lockp(hinfo, hash); |
d8313f5c | 259 | struct sock *sk2; |
3ab5aee7 | 260 | const struct hlist_nulls_node *node; |
05dbc7b5 | 261 | struct inet_timewait_sock *tw = NULL; |
d8313f5c | 262 | |
9db66bdc | 263 | spin_lock(lock); |
d8313f5c | 264 | |
05dbc7b5 | 265 | sk_nulls_for_each(sk2, node, &head->chain) { |
ce43b03e ED |
266 | if (sk2->sk_hash != hash) |
267 | continue; | |
d8313f5c | 268 | |
5d368f03 | 269 | if (likely(inet6_match(net, sk2, saddr, daddr, ports, |
4297a0ef | 270 | dif, sdif))) { |
efe4208f | 271 | if (sk2->sk_state == TCP_TIME_WAIT) { |
05dbc7b5 ED |
272 | tw = inet_twsk(sk2); |
273 | if (twsk_unique(sk, sk2, twp)) | |
efe4208f | 274 | break; |
05dbc7b5 | 275 | } |
d8313f5c | 276 | goto not_unique; |
efe4208f | 277 | } |
d8313f5c ACM |
278 | } |
279 | ||
3759fa9c | 280 | /* Must record num and sport now. Otherwise we will see |
efe4208f ED |
281 | * in hash table socket with a funny identity. |
282 | */ | |
c720c7e8 ED |
283 | inet->inet_num = lport; |
284 | inet->inet_sport = htons(lport); | |
13475a30 | 285 | sk->sk_hash = hash; |
547b792c | 286 | WARN_ON(!sk_unhashed(sk)); |
3ab5aee7 | 287 | __sk_nulls_add_node_rcu(sk, &head->chain); |
13475a30 | 288 | if (tw) { |
fc01538f | 289 | sk_nulls_del_node_init_rcu((struct sock *)tw); |
02a1d6e7 | 290 | __NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED); |
13475a30 | 291 | } |
9db66bdc | 292 | spin_unlock(lock); |
c29a0bc4 | 293 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); |
d8313f5c | 294 | |
13475a30 | 295 | if (twp) { |
d8313f5c | 296 | *twp = tw; |
13475a30 | 297 | } else if (tw) { |
d8313f5c | 298 | /* Silly. Should hash-dance instead... */ |
dbe7faa4 | 299 | inet_twsk_deschedule_put(tw); |
d8313f5c ACM |
300 | } |
301 | return 0; | |
302 | ||
303 | not_unique: | |
9db66bdc | 304 | spin_unlock(lock); |
d8313f5c ACM |
305 | return -EADDRNOTAVAIL; |
306 | } | |
307 | ||
b2d05756 | 308 | static u64 inet6_sk_port_offset(const struct sock *sk) |
d8313f5c ACM |
309 | { |
310 | const struct inet_sock *inet = inet_sk(sk); | |
efe4208f ED |
311 | |
312 | return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32, | |
313 | sk->sk_v6_daddr.s6_addr32, | |
c720c7e8 | 314 | inet->inet_dport); |
d8313f5c ACM |
315 | } |
316 | ||
317 | int inet6_hash_connect(struct inet_timewait_death_row *death_row, | |
318 | struct sock *sk) | |
319 | { | |
b2d05756 | 320 | u64 port_offset = 0; |
e2baad9e ED |
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, | |
b4d6444e | 325 | __inet6_check_established); |
d8313f5c | 326 | } |
d8313f5c | 327 | EXPORT_SYMBOL_GPL(inet6_hash_connect); |
496611d7 CG |
328 | |
329 | int inet6_hash(struct sock *sk) | |
330 | { | |
e4cabca5 CG |
331 | int err = 0; |
332 | ||
4f9bf2a2 | 333 | if (sk->sk_state != TCP_CLOSE) |
fe38d2a1 | 334 | err = __inet_hash(sk, NULL); |
496611d7 | 335 | |
e4cabca5 | 336 | return err; |
496611d7 CG |
337 | } |
338 | EXPORT_SYMBOL_GPL(inet6_hash); |