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 bool dev_match; 103 104 if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum && 105 sk->sk_family == PF_INET6) { 106 107 score = 1; 108 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) { 109 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 110 return -1; 111 score++; 112 } 113 dev_match = inet_sk_bound_dev_eq(net, sk->sk_bound_dev_if, 114 dif, sdif); 115 if (!dev_match) 116 return -1; 117 score++; 118 119 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 120 score++; 121 } 122 return score; 123 } 124 125 /* called with rcu_read_lock() */ 126 static struct sock *inet6_lhash2_lookup(struct net *net, 127 struct inet_listen_hashbucket *ilb2, 128 struct sk_buff *skb, int doff, 129 const struct in6_addr *saddr, 130 const __be16 sport, const struct in6_addr *daddr, 131 const unsigned short hnum, const int dif, const int sdif) 132 { 133 bool exact_dif = inet6_exact_dif_match(net, skb); 134 struct inet_connection_sock *icsk; 135 struct sock *sk, *result = NULL; 136 int score, hiscore = 0; 137 u32 phash = 0; 138 139 inet_lhash2_for_each_icsk_rcu(icsk, &ilb2->head) { 140 sk = (struct sock *)icsk; 141 score = compute_score(sk, net, hnum, daddr, dif, sdif, 142 exact_dif); 143 if (score > hiscore) { 144 if (sk->sk_reuseport) { 145 phash = inet6_ehashfn(net, daddr, hnum, 146 saddr, sport); 147 result = reuseport_select_sock(sk, phash, 148 skb, doff); 149 if (result) 150 return result; 151 } 152 result = sk; 153 hiscore = score; 154 } 155 } 156 157 return result; 158 } 159 160 struct sock *inet6_lookup_listener(struct net *net, 161 struct inet_hashinfo *hashinfo, 162 struct sk_buff *skb, int doff, 163 const struct in6_addr *saddr, 164 const __be16 sport, const struct in6_addr *daddr, 165 const unsigned short hnum, const int dif, const int sdif) 166 { 167 unsigned int hash = inet_lhashfn(net, hnum); 168 struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash]; 169 bool exact_dif = inet6_exact_dif_match(net, skb); 170 struct inet_listen_hashbucket *ilb2; 171 struct sock *sk, *result = NULL; 172 int score, hiscore = 0; 173 unsigned int hash2; 174 u32 phash = 0; 175 176 if (ilb->count <= 10 || !hashinfo->lhash2) 177 goto port_lookup; 178 179 /* Too many sk in the ilb bucket (which is hashed by port alone). 180 * Try lhash2 (which is hashed by port and addr) instead. 181 */ 182 183 hash2 = ipv6_portaddr_hash(net, daddr, hnum); 184 ilb2 = inet_lhash2_bucket(hashinfo, hash2); 185 if (ilb2->count > ilb->count) 186 goto port_lookup; 187 188 result = inet6_lhash2_lookup(net, ilb2, skb, doff, 189 saddr, sport, daddr, hnum, 190 dif, sdif); 191 if (result) 192 goto done; 193 194 /* Lookup lhash2 with in6addr_any */ 195 196 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum); 197 ilb2 = inet_lhash2_bucket(hashinfo, hash2); 198 if (ilb2->count > ilb->count) 199 goto port_lookup; 200 201 result = inet6_lhash2_lookup(net, ilb2, skb, doff, 202 saddr, sport, daddr, hnum, 203 dif, sdif); 204 goto done; 205 206 port_lookup: 207 sk_for_each(sk, &ilb->head) { 208 score = compute_score(sk, net, hnum, daddr, dif, sdif, exact_dif); 209 if (score > hiscore) { 210 if (sk->sk_reuseport) { 211 phash = inet6_ehashfn(net, daddr, hnum, 212 saddr, sport); 213 result = reuseport_select_sock(sk, phash, 214 skb, doff); 215 if (result) 216 goto done; 217 } 218 result = sk; 219 hiscore = score; 220 } 221 } 222 done: 223 if (unlikely(IS_ERR(result))) 224 return NULL; 225 return result; 226 } 227 EXPORT_SYMBOL_GPL(inet6_lookup_listener); 228 229 struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo, 230 struct sk_buff *skb, int doff, 231 const struct in6_addr *saddr, const __be16 sport, 232 const struct in6_addr *daddr, const __be16 dport, 233 const int dif) 234 { 235 struct sock *sk; 236 bool refcounted; 237 238 sk = __inet6_lookup(net, hashinfo, skb, doff, saddr, sport, daddr, 239 ntohs(dport), dif, 0, &refcounted); 240 if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt)) 241 sk = NULL; 242 return sk; 243 } 244 EXPORT_SYMBOL_GPL(inet6_lookup); 245 246 static int __inet6_check_established(struct inet_timewait_death_row *death_row, 247 struct sock *sk, const __u16 lport, 248 struct inet_timewait_sock **twp) 249 { 250 struct inet_hashinfo *hinfo = death_row->hashinfo; 251 struct inet_sock *inet = inet_sk(sk); 252 const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr; 253 const struct in6_addr *saddr = &sk->sk_v6_daddr; 254 const int dif = sk->sk_bound_dev_if; 255 struct net *net = sock_net(sk); 256 const int sdif = l3mdev_master_ifindex_by_index(net, dif); 257 const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport); 258 const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr, 259 inet->inet_dport); 260 struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash); 261 spinlock_t *lock = inet_ehash_lockp(hinfo, hash); 262 struct sock *sk2; 263 const struct hlist_nulls_node *node; 264 struct inet_timewait_sock *tw = NULL; 265 266 spin_lock(lock); 267 268 sk_nulls_for_each(sk2, node, &head->chain) { 269 if (sk2->sk_hash != hash) 270 continue; 271 272 if (likely(INET6_MATCH(sk2, net, saddr, daddr, ports, 273 dif, sdif))) { 274 if (sk2->sk_state == TCP_TIME_WAIT) { 275 tw = inet_twsk(sk2); 276 if (twsk_unique(sk, sk2, twp)) 277 break; 278 } 279 goto not_unique; 280 } 281 } 282 283 /* Must record num and sport now. Otherwise we will see 284 * in hash table socket with a funny identity. 285 */ 286 inet->inet_num = lport; 287 inet->inet_sport = htons(lport); 288 sk->sk_hash = hash; 289 WARN_ON(!sk_unhashed(sk)); 290 __sk_nulls_add_node_rcu(sk, &head->chain); 291 if (tw) { 292 sk_nulls_del_node_init_rcu((struct sock *)tw); 293 __NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED); 294 } 295 spin_unlock(lock); 296 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 297 298 if (twp) { 299 *twp = tw; 300 } else if (tw) { 301 /* Silly. Should hash-dance instead... */ 302 inet_twsk_deschedule_put(tw); 303 } 304 return 0; 305 306 not_unique: 307 spin_unlock(lock); 308 return -EADDRNOTAVAIL; 309 } 310 311 static u32 inet6_sk_port_offset(const struct sock *sk) 312 { 313 const struct inet_sock *inet = inet_sk(sk); 314 315 return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32, 316 sk->sk_v6_daddr.s6_addr32, 317 inet->inet_dport); 318 } 319 320 int inet6_hash_connect(struct inet_timewait_death_row *death_row, 321 struct sock *sk) 322 { 323 u32 port_offset = 0; 324 325 if (!inet_sk(sk)->inet_num) 326 port_offset = inet6_sk_port_offset(sk); 327 return __inet_hash_connect(death_row, sk, port_offset, 328 __inet6_check_established); 329 } 330 EXPORT_SYMBOL_GPL(inet6_hash_connect); 331 332 int inet6_hash(struct sock *sk) 333 { 334 int err = 0; 335 336 if (sk->sk_state != TCP_CLOSE) { 337 local_bh_disable(); 338 err = __inet_hash(sk, NULL); 339 local_bh_enable(); 340 } 341 342 return err; 343 } 344 EXPORT_SYMBOL_GPL(inet6_hash); 345