1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * Generic INET6 transport hashtables 7 * 8 * Authors: Lotsa people, from code originally in tcp, generalised here 9 * by Arnaldo Carvalho de Melo <acme@mandriva.com> 10 * 11 * This program is free software; you can redistribute it and/or 12 * modify it under the terms of the GNU General Public License 13 * as published by the Free Software Foundation; either version 14 * 2 of the License, or (at your option) any later version. 15 */ 16 17 #include <linux/module.h> 18 #include <linux/random.h> 19 20 #include <net/addrconf.h> 21 #include <net/inet_connection_sock.h> 22 #include <net/inet_hashtables.h> 23 #include <net/inet6_hashtables.h> 24 #include <net/secure_seq.h> 25 #include <net/ip.h> 26 #include <net/sock_reuseport.h> 27 28 u32 inet6_ehashfn(const struct net *net, 29 const struct in6_addr *laddr, const u16 lport, 30 const struct in6_addr *faddr, const __be16 fport) 31 { 32 static u32 inet6_ehash_secret __read_mostly; 33 static u32 ipv6_hash_secret __read_mostly; 34 35 u32 lhash, fhash; 36 37 net_get_random_once(&inet6_ehash_secret, sizeof(inet6_ehash_secret)); 38 net_get_random_once(&ipv6_hash_secret, sizeof(ipv6_hash_secret)); 39 40 lhash = (__force u32)laddr->s6_addr32[3]; 41 fhash = __ipv6_addr_jhash(faddr, ipv6_hash_secret); 42 43 return __inet6_ehashfn(lhash, lport, fhash, fport, 44 inet6_ehash_secret + net_hash_mix(net)); 45 } 46 47 /* 48 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so 49 * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM 50 * 51 * The sockhash lock must be held as a reader here. 52 */ 53 struct sock *__inet6_lookup_established(struct net *net, 54 struct inet_hashinfo *hashinfo, 55 const struct in6_addr *saddr, 56 const __be16 sport, 57 const struct in6_addr *daddr, 58 const u16 hnum, 59 const int dif, const int sdif) 60 { 61 struct sock *sk; 62 const struct hlist_nulls_node *node; 63 const __portpair ports = INET_COMBINED_PORTS(sport, hnum); 64 /* Optimize here for direct hit, only listening connections can 65 * have wildcards anyways. 66 */ 67 unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport); 68 unsigned int slot = hash & hashinfo->ehash_mask; 69 struct inet_ehash_bucket *head = &hashinfo->ehash[slot]; 70 71 72 begin: 73 sk_nulls_for_each_rcu(sk, node, &head->chain) { 74 if (sk->sk_hash != hash) 75 continue; 76 if (!INET6_MATCH(sk, net, saddr, daddr, ports, dif, sdif)) 77 continue; 78 if (unlikely(!refcount_inc_not_zero(&sk->sk_refcnt))) 79 goto out; 80 81 if (unlikely(!INET6_MATCH(sk, net, saddr, daddr, ports, dif, sdif))) { 82 sock_gen_put(sk); 83 goto begin; 84 } 85 goto found; 86 } 87 if (get_nulls_value(node) != slot) 88 goto begin; 89 out: 90 sk = NULL; 91 found: 92 return sk; 93 } 94 EXPORT_SYMBOL(__inet6_lookup_established); 95 96 static inline int compute_score(struct sock *sk, struct net *net, 97 const unsigned short hnum, 98 const struct in6_addr *daddr, 99 const int dif, const int sdif, bool exact_dif) 100 { 101 int score = -1; 102 103 if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum && 104 sk->sk_family == PF_INET6) { 105 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 106 return -1; 107 108 if (!inet_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif)) 109 return -1; 110 111 score = 1; 112 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 113 score++; 114 } 115 return score; 116 } 117 118 /* called with rcu_read_lock() */ 119 static struct sock *inet6_lhash2_lookup(struct net *net, 120 struct inet_listen_hashbucket *ilb2, 121 struct sk_buff *skb, int doff, 122 const struct in6_addr *saddr, 123 const __be16 sport, const struct in6_addr *daddr, 124 const unsigned short hnum, const int dif, const int sdif) 125 { 126 bool exact_dif = inet6_exact_dif_match(net, skb); 127 struct inet_connection_sock *icsk; 128 struct sock *sk, *result = NULL; 129 int score, hiscore = 0; 130 u32 phash = 0; 131 132 inet_lhash2_for_each_icsk_rcu(icsk, &ilb2->head) { 133 sk = (struct sock *)icsk; 134 score = compute_score(sk, net, hnum, daddr, dif, sdif, 135 exact_dif); 136 if (score > hiscore) { 137 if (sk->sk_reuseport) { 138 phash = inet6_ehashfn(net, daddr, hnum, 139 saddr, sport); 140 result = reuseport_select_sock(sk, phash, 141 skb, doff); 142 if (result) 143 return result; 144 } 145 result = sk; 146 hiscore = score; 147 } 148 } 149 150 return result; 151 } 152 153 struct sock *inet6_lookup_listener(struct net *net, 154 struct inet_hashinfo *hashinfo, 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 inet_listen_hashbucket *ilb2; 161 struct sock *result = NULL; 162 unsigned int hash2; 163 164 hash2 = ipv6_portaddr_hash(net, daddr, hnum); 165 ilb2 = inet_lhash2_bucket(hashinfo, hash2); 166 167 result = inet6_lhash2_lookup(net, ilb2, skb, doff, 168 saddr, sport, daddr, hnum, 169 dif, sdif); 170 if (result) 171 goto done; 172 173 /* Lookup lhash2 with in6addr_any */ 174 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum); 175 ilb2 = inet_lhash2_bucket(hashinfo, hash2); 176 177 result = inet6_lhash2_lookup(net, ilb2, skb, doff, 178 saddr, sport, &in6addr_any, hnum, 179 dif, sdif); 180 done: 181 if (unlikely(IS_ERR(result))) 182 return NULL; 183 return result; 184 } 185 EXPORT_SYMBOL_GPL(inet6_lookup_listener); 186 187 struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo, 188 struct sk_buff *skb, int doff, 189 const struct in6_addr *saddr, const __be16 sport, 190 const struct in6_addr *daddr, const __be16 dport, 191 const int dif) 192 { 193 struct sock *sk; 194 bool refcounted; 195 196 sk = __inet6_lookup(net, hashinfo, skb, doff, saddr, sport, daddr, 197 ntohs(dport), dif, 0, &refcounted); 198 if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt)) 199 sk = NULL; 200 return sk; 201 } 202 EXPORT_SYMBOL_GPL(inet6_lookup); 203 204 static int __inet6_check_established(struct inet_timewait_death_row *death_row, 205 struct sock *sk, const __u16 lport, 206 struct inet_timewait_sock **twp) 207 { 208 struct inet_hashinfo *hinfo = death_row->hashinfo; 209 struct inet_sock *inet = inet_sk(sk); 210 const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr; 211 const struct in6_addr *saddr = &sk->sk_v6_daddr; 212 const int dif = sk->sk_bound_dev_if; 213 struct net *net = sock_net(sk); 214 const int sdif = l3mdev_master_ifindex_by_index(net, dif); 215 const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport); 216 const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr, 217 inet->inet_dport); 218 struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash); 219 spinlock_t *lock = inet_ehash_lockp(hinfo, hash); 220 struct sock *sk2; 221 const struct hlist_nulls_node *node; 222 struct inet_timewait_sock *tw = NULL; 223 224 spin_lock(lock); 225 226 sk_nulls_for_each(sk2, node, &head->chain) { 227 if (sk2->sk_hash != hash) 228 continue; 229 230 if (likely(INET6_MATCH(sk2, net, saddr, daddr, ports, 231 dif, sdif))) { 232 if (sk2->sk_state == TCP_TIME_WAIT) { 233 tw = inet_twsk(sk2); 234 if (twsk_unique(sk, sk2, twp)) 235 break; 236 } 237 goto not_unique; 238 } 239 } 240 241 /* Must record num and sport now. Otherwise we will see 242 * in hash table socket with a funny identity. 243 */ 244 inet->inet_num = lport; 245 inet->inet_sport = htons(lport); 246 sk->sk_hash = hash; 247 WARN_ON(!sk_unhashed(sk)); 248 __sk_nulls_add_node_rcu(sk, &head->chain); 249 if (tw) { 250 sk_nulls_del_node_init_rcu((struct sock *)tw); 251 __NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED); 252 } 253 spin_unlock(lock); 254 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 255 256 if (twp) { 257 *twp = tw; 258 } else if (tw) { 259 /* Silly. Should hash-dance instead... */ 260 inet_twsk_deschedule_put(tw); 261 } 262 return 0; 263 264 not_unique: 265 spin_unlock(lock); 266 return -EADDRNOTAVAIL; 267 } 268 269 static u32 inet6_sk_port_offset(const struct sock *sk) 270 { 271 const struct inet_sock *inet = inet_sk(sk); 272 273 return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32, 274 sk->sk_v6_daddr.s6_addr32, 275 inet->inet_dport); 276 } 277 278 int inet6_hash_connect(struct inet_timewait_death_row *death_row, 279 struct sock *sk) 280 { 281 u32 port_offset = 0; 282 283 if (!inet_sk(sk)->inet_num) 284 port_offset = inet6_sk_port_offset(sk); 285 return __inet_hash_connect(death_row, sk, port_offset, 286 __inet6_check_established); 287 } 288 EXPORT_SYMBOL_GPL(inet6_hash_connect); 289 290 int inet6_hash(struct sock *sk) 291 { 292 int err = 0; 293 294 if (sk->sk_state != TCP_CLOSE) { 295 local_bh_disable(); 296 err = __inet_hash(sk, NULL); 297 local_bh_enable(); 298 } 299 300 return err; 301 } 302 EXPORT_SYMBOL_GPL(inet6_hash); 303