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) { 105 106 score = 1; 107 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) { 108 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 109 return -1; 110 score++; 111 } 112 if (sk->sk_bound_dev_if || exact_dif) { 113 bool dev_match = (sk->sk_bound_dev_if == dif || 114 sk->sk_bound_dev_if == sdif); 115 116 if (exact_dif && !dev_match) 117 return -1; 118 if (sk->sk_bound_dev_if && dev_match) 119 score++; 120 } 121 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 122 score++; 123 } 124 return score; 125 } 126 127 /* called with rcu_read_lock() */ 128 struct sock *inet6_lookup_listener(struct net *net, 129 struct inet_hashinfo *hashinfo, 130 struct sk_buff *skb, int doff, 131 const struct in6_addr *saddr, 132 const __be16 sport, const struct in6_addr *daddr, 133 const unsigned short hnum, const int dif, const int sdif) 134 { 135 unsigned int hash = inet_lhashfn(net, hnum); 136 struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash]; 137 int score, hiscore = 0, matches = 0, reuseport = 0; 138 bool exact_dif = inet6_exact_dif_match(net, skb); 139 struct sock *sk, *result = NULL; 140 u32 phash = 0; 141 142 sk_for_each(sk, &ilb->head) { 143 score = compute_score(sk, net, hnum, daddr, dif, sdif, exact_dif); 144 if (score > hiscore) { 145 reuseport = sk->sk_reuseport; 146 if (reuseport) { 147 phash = inet6_ehashfn(net, daddr, hnum, 148 saddr, sport); 149 result = reuseport_select_sock(sk, phash, 150 skb, doff); 151 if (result) 152 return result; 153 matches = 1; 154 } 155 result = sk; 156 hiscore = score; 157 } else if (score == hiscore && reuseport) { 158 matches++; 159 if (reciprocal_scale(phash, matches) == 0) 160 result = sk; 161 phash = next_pseudo_random32(phash); 162 } 163 } 164 return result; 165 } 166 EXPORT_SYMBOL_GPL(inet6_lookup_listener); 167 168 struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo, 169 struct sk_buff *skb, int doff, 170 const struct in6_addr *saddr, const __be16 sport, 171 const struct in6_addr *daddr, const __be16 dport, 172 const int dif) 173 { 174 struct sock *sk; 175 bool refcounted; 176 177 sk = __inet6_lookup(net, hashinfo, skb, doff, saddr, sport, daddr, 178 ntohs(dport), dif, 0, &refcounted); 179 if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt)) 180 sk = NULL; 181 return sk; 182 } 183 EXPORT_SYMBOL_GPL(inet6_lookup); 184 185 static int __inet6_check_established(struct inet_timewait_death_row *death_row, 186 struct sock *sk, const __u16 lport, 187 struct inet_timewait_sock **twp) 188 { 189 struct inet_hashinfo *hinfo = death_row->hashinfo; 190 struct inet_sock *inet = inet_sk(sk); 191 const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr; 192 const struct in6_addr *saddr = &sk->sk_v6_daddr; 193 const int dif = sk->sk_bound_dev_if; 194 struct net *net = sock_net(sk); 195 const int sdif = l3mdev_master_ifindex_by_index(net, dif); 196 const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport); 197 const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr, 198 inet->inet_dport); 199 struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash); 200 spinlock_t *lock = inet_ehash_lockp(hinfo, hash); 201 struct sock *sk2; 202 const struct hlist_nulls_node *node; 203 struct inet_timewait_sock *tw = NULL; 204 205 spin_lock(lock); 206 207 sk_nulls_for_each(sk2, node, &head->chain) { 208 if (sk2->sk_hash != hash) 209 continue; 210 211 if (likely(INET6_MATCH(sk2, net, saddr, daddr, ports, 212 dif, sdif))) { 213 if (sk2->sk_state == TCP_TIME_WAIT) { 214 tw = inet_twsk(sk2); 215 if (twsk_unique(sk, sk2, twp)) 216 break; 217 } 218 goto not_unique; 219 } 220 } 221 222 /* Must record num and sport now. Otherwise we will see 223 * in hash table socket with a funny identity. 224 */ 225 inet->inet_num = lport; 226 inet->inet_sport = htons(lport); 227 sk->sk_hash = hash; 228 WARN_ON(!sk_unhashed(sk)); 229 __sk_nulls_add_node_rcu(sk, &head->chain); 230 if (tw) { 231 sk_nulls_del_node_init_rcu((struct sock *)tw); 232 __NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED); 233 } 234 spin_unlock(lock); 235 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 236 237 if (twp) { 238 *twp = tw; 239 } else if (tw) { 240 /* Silly. Should hash-dance instead... */ 241 inet_twsk_deschedule_put(tw); 242 } 243 return 0; 244 245 not_unique: 246 spin_unlock(lock); 247 return -EADDRNOTAVAIL; 248 } 249 250 static u32 inet6_sk_port_offset(const struct sock *sk) 251 { 252 const struct inet_sock *inet = inet_sk(sk); 253 254 return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32, 255 sk->sk_v6_daddr.s6_addr32, 256 inet->inet_dport); 257 } 258 259 int inet6_hash_connect(struct inet_timewait_death_row *death_row, 260 struct sock *sk) 261 { 262 u32 port_offset = 0; 263 264 if (!inet_sk(sk)->inet_num) 265 port_offset = inet6_sk_port_offset(sk); 266 return __inet_hash_connect(death_row, sk, port_offset, 267 __inet6_check_established); 268 } 269 EXPORT_SYMBOL_GPL(inet6_hash_connect); 270 271 int inet6_hash(struct sock *sk) 272 { 273 int err = 0; 274 275 if (sk->sk_state != TCP_CLOSE) { 276 local_bh_disable(); 277 err = __inet_hash(sk, NULL); 278 local_bh_enable(); 279 } 280 281 return err; 282 } 283 EXPORT_SYMBOL_GPL(inet6_hash); 284