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 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) 27 { 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 39 return __inet6_ehashfn(lhash, lport, fhash, fport, 40 inet6_ehash_secret + net_hash_mix(net)); 41 } 42 EXPORT_SYMBOL_GPL(inet6_ehashfn); 43 44 /* 45 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so 46 * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM 47 * 48 * The sockhash lock must be held as a reader here. 49 */ 50 struct sock *__inet6_lookup_established(struct net *net, 51 struct inet_hashinfo *hashinfo, 52 const struct in6_addr *saddr, 53 const __be16 sport, 54 const struct in6_addr *daddr, 55 const u16 hnum, 56 const int dif, const int sdif) 57 { 58 struct sock *sk; 59 const struct hlist_nulls_node *node; 60 const __portpair ports = INET_COMBINED_PORTS(sport, hnum); 61 /* Optimize here for direct hit, only listening connections can 62 * have wildcards anyways. 63 */ 64 unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport); 65 unsigned int slot = hash & hashinfo->ehash_mask; 66 struct inet_ehash_bucket *head = &hashinfo->ehash[slot]; 67 68 69 begin: 70 sk_nulls_for_each_rcu(sk, node, &head->chain) { 71 if (sk->sk_hash != hash) 72 continue; 73 if (!inet6_match(net, sk, saddr, daddr, ports, dif, sdif)) 74 continue; 75 if (unlikely(!refcount_inc_not_zero(&sk->sk_refcnt))) 76 goto out; 77 78 if (unlikely(!inet6_match(net, sk, saddr, daddr, ports, dif, sdif))) { 79 sock_gen_put(sk); 80 goto begin; 81 } 82 goto found; 83 } 84 if (get_nulls_value(node) != slot) 85 goto begin; 86 out: 87 sk = NULL; 88 found: 89 return sk; 90 } 91 EXPORT_SYMBOL(__inet6_lookup_established); 92 93 static inline int compute_score(struct sock *sk, struct net *net, 94 const unsigned short hnum, 95 const struct in6_addr *daddr, 96 const int dif, const int sdif) 97 { 98 int score = -1; 99 100 if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum && 101 sk->sk_family == PF_INET6) { 102 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 103 return -1; 104 105 if (!inet_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif)) 106 return -1; 107 108 score = sk->sk_bound_dev_if ? 2 : 1; 109 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) 110 score++; 111 } 112 return score; 113 } 114 115 INDIRECT_CALLABLE_DECLARE(inet6_ehashfn_t udp6_ehashfn); 116 117 /** 118 * inet6_lookup_reuseport() - execute reuseport logic on AF_INET6 socket if necessary. 119 * @net: network namespace. 120 * @sk: AF_INET6 socket, must be in TCP_LISTEN state for TCP or TCP_CLOSE for UDP. 121 * @skb: context for a potential SK_REUSEPORT program. 122 * @doff: header offset. 123 * @saddr: source address. 124 * @sport: source port. 125 * @daddr: destination address. 126 * @hnum: destination port in host byte order. 127 * @ehashfn: hash function used to generate the fallback hash. 128 * 129 * Return: NULL if sk doesn't have SO_REUSEPORT set, otherwise a pointer to 130 * the selected sock or an error. 131 */ 132 struct sock *inet6_lookup_reuseport(struct net *net, struct sock *sk, 133 struct sk_buff *skb, int doff, 134 const struct in6_addr *saddr, 135 __be16 sport, 136 const struct in6_addr *daddr, 137 unsigned short hnum, 138 inet6_ehashfn_t *ehashfn) 139 { 140 struct sock *reuse_sk = NULL; 141 u32 phash; 142 143 if (sk->sk_reuseport) { 144 phash = INDIRECT_CALL_INET(ehashfn, udp6_ehashfn, inet6_ehashfn, 145 net, daddr, hnum, saddr, sport); 146 reuse_sk = reuseport_select_sock(sk, phash, skb, doff); 147 } 148 return reuse_sk; 149 } 150 EXPORT_SYMBOL_GPL(inet6_lookup_reuseport); 151 152 /* called with rcu_read_lock() */ 153 static struct sock *inet6_lhash2_lookup(struct net *net, 154 struct inet_listen_hashbucket *ilb2, 155 struct sk_buff *skb, int doff, 156 const struct in6_addr *saddr, 157 const __be16 sport, const struct in6_addr *daddr, 158 const unsigned short hnum, const int dif, const int sdif) 159 { 160 struct sock *sk, *result = NULL; 161 struct hlist_nulls_node *node; 162 int score, hiscore = 0; 163 164 sk_nulls_for_each_rcu(sk, node, &ilb2->nulls_head) { 165 score = compute_score(sk, net, hnum, daddr, dif, sdif); 166 if (score > hiscore) { 167 result = inet6_lookup_reuseport(net, sk, skb, doff, 168 saddr, sport, daddr, hnum, inet6_ehashfn); 169 if (result) 170 return result; 171 172 result = sk; 173 hiscore = score; 174 } 175 } 176 177 return result; 178 } 179 180 struct sock *inet6_lookup_run_sk_lookup(struct net *net, 181 int protocol, 182 struct sk_buff *skb, int doff, 183 const struct in6_addr *saddr, 184 const __be16 sport, 185 const struct in6_addr *daddr, 186 const u16 hnum, const int dif, 187 inet6_ehashfn_t *ehashfn) 188 { 189 struct sock *sk, *reuse_sk; 190 bool no_reuseport; 191 192 no_reuseport = bpf_sk_lookup_run_v6(net, protocol, saddr, sport, 193 daddr, hnum, dif, &sk); 194 if (no_reuseport || IS_ERR_OR_NULL(sk)) 195 return sk; 196 197 reuse_sk = inet6_lookup_reuseport(net, sk, skb, doff, 198 saddr, sport, daddr, hnum, ehashfn); 199 if (reuse_sk) 200 sk = reuse_sk; 201 return sk; 202 } 203 EXPORT_SYMBOL_GPL(inet6_lookup_run_sk_lookup); 204 205 struct sock *inet6_lookup_listener(struct net *net, 206 struct inet_hashinfo *hashinfo, 207 struct sk_buff *skb, int doff, 208 const struct in6_addr *saddr, 209 const __be16 sport, const struct in6_addr *daddr, 210 const unsigned short hnum, const int dif, const int sdif) 211 { 212 struct inet_listen_hashbucket *ilb2; 213 struct sock *result = NULL; 214 unsigned int hash2; 215 216 /* Lookup redirect from BPF */ 217 if (static_branch_unlikely(&bpf_sk_lookup_enabled) && 218 hashinfo == net->ipv4.tcp_death_row.hashinfo) { 219 result = inet6_lookup_run_sk_lookup(net, IPPROTO_TCP, skb, doff, 220 saddr, sport, daddr, hnum, dif, 221 inet6_ehashfn); 222 if (result) 223 goto done; 224 } 225 226 hash2 = ipv6_portaddr_hash(net, daddr, hnum); 227 ilb2 = inet_lhash2_bucket(hashinfo, hash2); 228 229 result = inet6_lhash2_lookup(net, ilb2, skb, doff, 230 saddr, sport, daddr, hnum, 231 dif, sdif); 232 if (result) 233 goto done; 234 235 /* Lookup lhash2 with in6addr_any */ 236 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum); 237 ilb2 = inet_lhash2_bucket(hashinfo, hash2); 238 239 result = inet6_lhash2_lookup(net, ilb2, skb, doff, 240 saddr, sport, &in6addr_any, hnum, 241 dif, sdif); 242 done: 243 if (IS_ERR(result)) 244 return NULL; 245 return result; 246 } 247 EXPORT_SYMBOL_GPL(inet6_lookup_listener); 248 249 struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo, 250 struct sk_buff *skb, int doff, 251 const struct in6_addr *saddr, const __be16 sport, 252 const struct in6_addr *daddr, const __be16 dport, 253 const int dif) 254 { 255 struct sock *sk; 256 bool refcounted; 257 258 sk = __inet6_lookup(net, hashinfo, skb, doff, saddr, sport, daddr, 259 ntohs(dport), dif, 0, &refcounted); 260 if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt)) 261 sk = NULL; 262 return sk; 263 } 264 EXPORT_SYMBOL_GPL(inet6_lookup); 265 266 static int __inet6_check_established(struct inet_timewait_death_row *death_row, 267 struct sock *sk, const __u16 lport, 268 struct inet_timewait_sock **twp) 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, dif); 277 const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport); 278 const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr, 279 inet->inet_dport); 280 struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash); 281 spinlock_t *lock = inet_ehash_lockp(hinfo, hash); 282 struct sock *sk2; 283 const struct hlist_nulls_node *node; 284 struct inet_timewait_sock *tw = NULL; 285 286 spin_lock(lock); 287 288 sk_nulls_for_each(sk2, node, &head->chain) { 289 if (sk2->sk_hash != hash) 290 continue; 291 292 if (likely(inet6_match(net, sk2, saddr, daddr, ports, 293 dif, sdif))) { 294 if (sk2->sk_state == TCP_TIME_WAIT) { 295 tw = inet_twsk(sk2); 296 if (twsk_unique(sk, sk2, twp)) 297 break; 298 } 299 goto not_unique; 300 } 301 } 302 303 /* Must record num and sport now. Otherwise we will see 304 * in hash table socket with a funny identity. 305 */ 306 inet->inet_num = lport; 307 inet->inet_sport = htons(lport); 308 sk->sk_hash = hash; 309 WARN_ON(!sk_unhashed(sk)); 310 __sk_nulls_add_node_rcu(sk, &head->chain); 311 if (tw) { 312 sk_nulls_del_node_init_rcu((struct sock *)tw); 313 __NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED); 314 } 315 spin_unlock(lock); 316 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 317 318 if (twp) { 319 *twp = tw; 320 } else if (tw) { 321 /* Silly. Should hash-dance instead... */ 322 inet_twsk_deschedule_put(tw); 323 } 324 return 0; 325 326 not_unique: 327 spin_unlock(lock); 328 return -EADDRNOTAVAIL; 329 } 330 331 static u64 inet6_sk_port_offset(const struct sock *sk) 332 { 333 const struct inet_sock *inet = inet_sk(sk); 334 335 return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32, 336 sk->sk_v6_daddr.s6_addr32, 337 inet->inet_dport); 338 } 339 340 int inet6_hash_connect(struct inet_timewait_death_row *death_row, 341 struct sock *sk) 342 { 343 u64 port_offset = 0; 344 345 if (!inet_sk(sk)->inet_num) 346 port_offset = inet6_sk_port_offset(sk); 347 return __inet_hash_connect(death_row, sk, port_offset, 348 __inet6_check_established); 349 } 350 EXPORT_SYMBOL_GPL(inet6_hash_connect); 351 352 int inet6_hash(struct sock *sk) 353 { 354 int err = 0; 355 356 if (sk->sk_state != TCP_CLOSE) 357 err = __inet_hash(sk, NULL); 358 359 return err; 360 } 361 EXPORT_SYMBOL_GPL(inet6_hash); 362