net/ipv6/inet6_hashtables.c

   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)
  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))
  77                         continue;
  78                 if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt)))
  79                         goto out;
  80
  81                 if (unlikely(!INET6_MATCH(sk, net, saddr, daddr, ports, dif))) {
  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)
 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) {
 113                         if (sk->sk_bound_dev_if != dif)
 114                                 return -1;
 115                         score++;
 116                 }
 117                 if (sk->sk_incoming_cpu == raw_smp_processor_id())
 118                         score++;
 119         }
 120         return score;
 121 }
 122
 123 /* called with rcu_read_lock() */
 124 struct sock *inet6_lookup_listener(struct net *net,
 125                 struct inet_hashinfo *hashinfo,
 126                 struct sk_buff *skb, int doff,
 127                 const struct in6_addr *saddr,
 128                 const __be16 sport, const struct in6_addr *daddr,
 129                 const unsigned short hnum, const int dif)
 130 {
 131         unsigned int hash = inet_lhashfn(net, hnum);
 132         struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
 133         int score, hiscore = 0, matches = 0, reuseport = 0;
 134         struct sock *sk, *result = NULL;
 135         u32 phash = 0;
 136
 137         sk_for_each(sk, &ilb->head) {
 138                 score = compute_score(sk, net, hnum, daddr, dif);
 139                 if (score > hiscore) {
 140                         reuseport = sk->sk_reuseport;
 141                         if (reuseport) {
 142                                 phash = inet6_ehashfn(net, daddr, hnum,
 143                                                       saddr, sport);
 144                                 result = reuseport_select_sock(sk, phash,
 145                                                                skb, doff);
 146                                 if (result)
 147                                         return result;
 148                                 matches = 1;
 149                         }
 150                         result = sk;
 151                         hiscore = score;
 152                 } else if (score == hiscore && reuseport) {
 153                         matches++;
 154                         if (reciprocal_scale(phash, matches) == 0)
 155                                 result = sk;
 156                         phash = next_pseudo_random32(phash);
 157                 }
 158         }
 159         return result;
 160 }
 161 EXPORT_SYMBOL_GPL(inet6_lookup_listener);
 162
 163 struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
 164                           struct sk_buff *skb, int doff,
 165                           const struct in6_addr *saddr, const __be16 sport,
 166                           const struct in6_addr *daddr, const __be16 dport,
 167                           const int dif)
 168 {
 169         struct sock *sk;
 170         bool refcounted;
 171
 172         sk = __inet6_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
 173                             ntohs(dport), dif, &refcounted);
 174         if (sk && !refcounted && !atomic_inc_not_zero(&sk->sk_refcnt))
 175                 sk = NULL;
 176         return sk;
 177 }
 178 EXPORT_SYMBOL_GPL(inet6_lookup);
 179
 180 static int __inet6_check_established(struct inet_timewait_death_row *death_row,
 181                                      struct sock *sk, const __u16 lport,
 182                                      struct inet_timewait_sock **twp)
 183 {
 184         struct inet_hashinfo *hinfo = death_row->hashinfo;
 185         struct inet_sock *inet = inet_sk(sk);
 186         const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr;
 187         const struct in6_addr *saddr = &sk->sk_v6_daddr;
 188         const int dif = sk->sk_bound_dev_if;
 189         const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
 190         struct net *net = sock_net(sk);
 191         const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
 192                                                 inet->inet_dport);
 193         struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
 194         spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
 195         struct sock *sk2;
 196         const struct hlist_nulls_node *node;
 197         struct inet_timewait_sock *tw = NULL;
 198
 199         spin_lock(lock);
 200
 201         sk_nulls_for_each(sk2, node, &head->chain) {
 202                 if (sk2->sk_hash != hash)
 203                         continue;
 204
 205                 if (likely(INET6_MATCH(sk2, net, saddr, daddr, ports, dif))) {
 206                         if (sk2->sk_state == TCP_TIME_WAIT) {
 207                                 tw = inet_twsk(sk2);
 208                                 if (twsk_unique(sk, sk2, twp))
 209                                         break;
 210                         }
 211                         goto not_unique;
 212                 }
 213         }
 214
 215         /* Must record num and sport now. Otherwise we will see
 216          * in hash table socket with a funny identity.
 217          */
 218         inet->inet_num = lport;
 219         inet->inet_sport = htons(lport);
 220         sk->sk_hash = hash;
 221         WARN_ON(!sk_unhashed(sk));
 222         __sk_nulls_add_node_rcu(sk, &head->chain);
 223         if (tw) {
 224                 sk_nulls_del_node_init_rcu((struct sock *)tw);
 225                 __NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED);
 226         }
 227         spin_unlock(lock);
 228         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 229
 230         if (twp) {
 231                 *twp = tw;
 232         } else if (tw) {
 233                 /* Silly. Should hash-dance instead... */
 234                 inet_twsk_deschedule_put(tw);
 235         }
 236         return 0;
 237
 238 not_unique:
 239         spin_unlock(lock);
 240         return -EADDRNOTAVAIL;
 241 }
 242
 243 static u32 inet6_sk_port_offset(const struct sock *sk)
 244 {
 245         const struct inet_sock *inet = inet_sk(sk);
 246
 247         return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32,
 248                                           sk->sk_v6_daddr.s6_addr32,
 249                                           inet->inet_dport);
 250 }
 251
 252 int inet6_hash_connect(struct inet_timewait_death_row *death_row,
 253                        struct sock *sk)
 254 {
 255         u32 port_offset = 0;
 256
 257         if (!inet_sk(sk)->inet_num)
 258                 port_offset = inet6_sk_port_offset(sk);
 259         return __inet_hash_connect(death_row, sk, port_offset,
 260                                    __inet6_check_established);
 261 }
 262 EXPORT_SYMBOL_GPL(inet6_hash_connect);
 263
 264 int inet6_hash(struct sock *sk)
 265 {
 266         if (sk->sk_state != TCP_CLOSE) {
 267                 local_bh_disable();
 268                 __inet_hash(sk, NULL, ipv6_rcv_saddr_equal);
 269                 local_bh_enable();
 270         }
 271
 272         return 0;
 273 }
 274 EXPORT_SYMBOL_GPL(inet6_hash);
 275
 276 /* match_wildcard == true:  IPV6_ADDR_ANY equals to any IPv6 addresses if IPv6
 277  *                          only, and any IPv4 addresses if not IPv6 only
 278  * match_wildcard == false: addresses must be exactly the same, i.e.
 279  *                          IPV6_ADDR_ANY only equals to IPV6_ADDR_ANY,
 280  *                          and 0.0.0.0 equals to 0.0.0.0 only
 281  */
 282 int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2,
 283                          bool match_wildcard)
 284 {
 285         const struct in6_addr *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2);
 286         int sk2_ipv6only = inet_v6_ipv6only(sk2);
 287         int addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
 288         int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
 289
 290         /* if both are mapped, treat as IPv4 */
 291         if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) {
 292                 if (!sk2_ipv6only) {
 293                         if (sk->sk_rcv_saddr == sk2->sk_rcv_saddr)
 294                                 return 1;
 295                         if (!sk->sk_rcv_saddr || !sk2->sk_rcv_saddr)
 296                                 return match_wildcard;
 297                 }
 298                 return 0;
 299         }
 300
 301         if (addr_type == IPV6_ADDR_ANY && addr_type2 == IPV6_ADDR_ANY)
 302                 return 1;
 303
 304         if (addr_type2 == IPV6_ADDR_ANY && match_wildcard &&
 305             !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
 306                 return 1;
 307
 308         if (addr_type == IPV6_ADDR_ANY && match_wildcard &&
 309             !(ipv6_only_sock(sk) && addr_type2 == IPV6_ADDR_MAPPED))
 310                 return 1;
 311
 312         if (sk2_rcv_saddr6 &&
 313             ipv6_addr_equal(&sk->sk_v6_rcv_saddr, sk2_rcv_saddr6))
 314                 return 1;
 315
 316         return 0;
 317 }
 318 EXPORT_SYMBOL_GPL(ipv6_rcv_saddr_equal);