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/inet_connection_sock.h>
  21 #include <net/inet_hashtables.h>
  22 #include <net/inet6_hashtables.h>
  23 #include <net/ip.h>
  24
  25 void __inet6_hash(struct inet_hashinfo *hashinfo,
  26                                 struct sock *sk)
  27 {
  28         struct hlist_head *list;
  29         rwlock_t *lock;
  30
  31         BUG_TRAP(sk_unhashed(sk));
  32
  33         if (sk->sk_state == TCP_LISTEN) {
  34                 list = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
  35                 lock = &hashinfo->lhash_lock;
  36                 inet_listen_wlock(hashinfo);
  37         } else {
  38                 unsigned int hash;
  39                 sk->sk_hash = hash = inet6_sk_ehashfn(sk);
  40                 list = &inet_ehash_bucket(hashinfo, hash)->chain;
  41                 lock = inet_ehash_lockp(hashinfo, hash);
  42                 write_lock(lock);
  43         }
  44
  45         __sk_add_node(sk, list);
  46         sock_prot_inuse_add(sk->sk_prot, 1);
  47         write_unlock(lock);
  48 }
  49 EXPORT_SYMBOL(__inet6_hash);
  50
  51 /*
  52  * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
  53  * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
  54  *
  55  * The sockhash lock must be held as a reader here.
  56  */
  57 struct sock *__inet6_lookup_established(struct inet_hashinfo *hashinfo,
  58                                            const struct in6_addr *saddr,
  59                                            const __be16 sport,
  60                                            const struct in6_addr *daddr,
  61                                            const u16 hnum,
  62                                            const int dif)
  63 {
  64         struct sock *sk;
  65         const struct hlist_node *node;
  66         const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
  67         /* Optimize here for direct hit, only listening connections can
  68          * have wildcards anyways.
  69          */
  70         unsigned int hash = inet6_ehashfn(daddr, hnum, saddr, sport);
  71         struct inet_ehash_bucket *head = inet_ehash_bucket(hashinfo, hash);
  72         rwlock_t *lock = inet_ehash_lockp(hashinfo, hash);
  73
  74         prefetch(head->chain.first);
  75         read_lock(lock);
  76         sk_for_each(sk, node, &head->chain) {
  77                 /* For IPV6 do the cheaper port and family tests first. */
  78                 if (INET6_MATCH(sk, hash, saddr, daddr, ports, dif))
  79                         goto hit; /* You sunk my battleship! */
  80         }
  81         /* Must check for a TIME_WAIT'er before going to listener hash. */
  82         sk_for_each(sk, node, &head->twchain) {
  83                 const struct inet_timewait_sock *tw = inet_twsk(sk);
  84
  85                 if(*((__portpair *)&(tw->tw_dport))     == ports        &&
  86                    sk->sk_family                == PF_INET6) {
  87                         const struct inet6_timewait_sock *tw6 = inet6_twsk(sk);
  88
  89                         if (ipv6_addr_equal(&tw6->tw_v6_daddr, saddr)   &&
  90                             ipv6_addr_equal(&tw6->tw_v6_rcv_saddr, daddr)       &&
  91                             (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif))
  92                                 goto hit;
  93                 }
  94         }
  95         read_unlock(lock);
  96         return NULL;
  97
  98 hit:
  99         sock_hold(sk);
 100         read_unlock(lock);
 101         return sk;
 102 }
 103 EXPORT_SYMBOL(__inet6_lookup_established);
 104
 105 struct sock *inet6_lookup_listener(struct inet_hashinfo *hashinfo,
 106                                    const struct in6_addr *daddr,
 107                                    const unsigned short hnum, const int dif)
 108 {
 109         struct sock *sk;
 110         const struct hlist_node *node;
 111         struct sock *result = NULL;
 112         int score, hiscore = 0;
 113
 114         read_lock(&hashinfo->lhash_lock);
 115         sk_for_each(sk, node, &hashinfo->listening_hash[inet_lhashfn(hnum)]) {
 116                 if (inet_sk(sk)->num == hnum && sk->sk_family == PF_INET6) {
 117                         const struct ipv6_pinfo *np = inet6_sk(sk);
 118
 119                         score = 1;
 120                         if (!ipv6_addr_any(&np->rcv_saddr)) {
 121                                 if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
 122                                         continue;
 123                                 score++;
 124                         }
 125                         if (sk->sk_bound_dev_if) {
 126                                 if (sk->sk_bound_dev_if != dif)
 127                                         continue;
 128                                 score++;
 129                         }
 130                         if (score == 3) {
 131                                 result = sk;
 132                                 break;
 133                         }
 134                         if (score > hiscore) {
 135                                 hiscore = score;
 136                                 result = sk;
 137                         }
 138                 }
 139         }
 140         if (result)
 141                 sock_hold(result);
 142         read_unlock(&hashinfo->lhash_lock);
 143         return result;
 144 }
 145
 146 EXPORT_SYMBOL_GPL(inet6_lookup_listener);
 147
 148 struct sock *inet6_lookup(struct inet_hashinfo *hashinfo,
 149                           const struct in6_addr *saddr, const __be16 sport,
 150                           const struct in6_addr *daddr, const __be16 dport,
 151                           const int dif)
 152 {
 153         struct sock *sk;
 154
 155         local_bh_disable();
 156         sk = __inet6_lookup(hashinfo, saddr, sport, daddr, ntohs(dport), dif);
 157         local_bh_enable();
 158
 159         return sk;
 160 }
 161
 162 EXPORT_SYMBOL_GPL(inet6_lookup);
 163
 164 static int __inet6_check_established(struct inet_timewait_death_row *death_row,
 165                                      struct sock *sk, const __u16 lport,
 166                                      struct inet_timewait_sock **twp)
 167 {
 168         struct inet_hashinfo *hinfo = death_row->hashinfo;
 169         struct inet_sock *inet = inet_sk(sk);
 170         const struct ipv6_pinfo *np = inet6_sk(sk);
 171         const struct in6_addr *daddr = &np->rcv_saddr;
 172         const struct in6_addr *saddr = &np->daddr;
 173         const int dif = sk->sk_bound_dev_if;
 174         const __portpair ports = INET_COMBINED_PORTS(inet->dport, lport);
 175         const unsigned int hash = inet6_ehashfn(daddr, lport, saddr,
 176                                                 inet->dport);
 177         struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
 178         rwlock_t *lock = inet_ehash_lockp(hinfo, hash);
 179         struct sock *sk2;
 180         const struct hlist_node *node;
 181         struct inet_timewait_sock *tw;
 182
 183         prefetch(head->chain.first);
 184         write_lock(lock);
 185
 186         /* Check TIME-WAIT sockets first. */
 187         sk_for_each(sk2, node, &head->twchain) {
 188                 const struct inet6_timewait_sock *tw6 = inet6_twsk(sk2);
 189
 190                 tw = inet_twsk(sk2);
 191
 192                 if(*((__portpair *)&(tw->tw_dport)) == ports             &&
 193                    sk2->sk_family              == PF_INET6       &&
 194                    ipv6_addr_equal(&tw6->tw_v6_daddr, saddr)     &&
 195                    ipv6_addr_equal(&tw6->tw_v6_rcv_saddr, daddr) &&
 196                    (!sk2->sk_bound_dev_if || sk2->sk_bound_dev_if == dif)) {
 197                         if (twsk_unique(sk, sk2, twp))
 198                                 goto unique;
 199                         else
 200                                 goto not_unique;
 201                 }
 202         }
 203         tw = NULL;
 204
 205         /* And established part... */
 206         sk_for_each(sk2, node, &head->chain) {
 207                 if (INET6_MATCH(sk2, hash, saddr, daddr, ports, dif))
 208                         goto not_unique;
 209         }
 210
 211 unique:
 212         /* Must record num and sport now. Otherwise we will see
 213          * in hash table socket with a funny identity. */
 214         inet->num = lport;
 215         inet->sport = htons(lport);
 216         BUG_TRAP(sk_unhashed(sk));
 217         __sk_add_node(sk, &head->chain);
 218         sk->sk_hash = hash;
 219         sock_prot_inuse_add(sk->sk_prot, 1);
 220         write_unlock(lock);
 221
 222         if (twp != NULL) {
 223                 *twp = tw;
 224                 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
 225         } else if (tw != NULL) {
 226                 /* Silly. Should hash-dance instead... */
 227                 inet_twsk_deschedule(tw, death_row);
 228                 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
 229
 230                 inet_twsk_put(tw);
 231         }
 232         return 0;
 233
 234 not_unique:
 235         write_unlock(lock);
 236         return -EADDRNOTAVAIL;
 237 }
 238
 239 static inline u32 inet6_sk_port_offset(const struct sock *sk)
 240 {
 241         const struct inet_sock *inet = inet_sk(sk);
 242         const struct ipv6_pinfo *np = inet6_sk(sk);
 243         return secure_ipv6_port_ephemeral(np->rcv_saddr.s6_addr32,
 244                                           np->daddr.s6_addr32,
 245                                           inet->dport);
 246 }
 247
 248 int inet6_hash_connect(struct inet_timewait_death_row *death_row,
 249                        struct sock *sk)
 250 {
 251         struct inet_hashinfo *hinfo = death_row->hashinfo;
 252         const unsigned short snum = inet_sk(sk)->num;
 253         struct inet_bind_hashbucket *head;
 254         struct inet_bind_bucket *tb;
 255         int ret;
 256
 257         if (snum == 0) {
 258                 int i, port, low, high, remaining;
 259                 static u32 hint;
 260                 const u32 offset = hint + inet6_sk_port_offset(sk);
 261                 struct hlist_node *node;
 262                 struct inet_timewait_sock *tw = NULL;
 263
 264                 inet_get_local_port_range(&low, &high);
 265                 remaining = (high - low) + 1;
 266
 267                 local_bh_disable();
 268                 for (i = 1; i <= remaining; i++) {
 269                         port = low + (i + offset) % remaining;
 270                         head = &hinfo->bhash[inet_bhashfn(port, hinfo->bhash_size)];
 271                         spin_lock(&head->lock);
 272
 273                         /* Does not bother with rcv_saddr checks,
 274                          * because the established check is already
 275                          * unique enough.
 276                          */
 277                         inet_bind_bucket_for_each(tb, node, &head->chain) {
 278                                 if (tb->port == port) {
 279                                         BUG_TRAP(!hlist_empty(&tb->owners));
 280                                         if (tb->fastreuse >= 0)
 281                                                 goto next_port;
 282                                         if (!__inet6_check_established(death_row,
 283                                                                        sk, port,
 284                                                                        &tw))
 285                                                 goto ok;
 286                                         goto next_port;
 287                                 }
 288                         }
 289
 290                         tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
 291                                                      head, port);
 292                         if (!tb) {
 293                                 spin_unlock(&head->lock);
 294                                 break;
 295                         }
 296                         tb->fastreuse = -1;
 297                         goto ok;
 298
 299                 next_port:
 300                         spin_unlock(&head->lock);
 301                 }
 302                 local_bh_enable();
 303
 304                 return -EADDRNOTAVAIL;
 305
 306 ok:
 307                 hint += i;
 308
 309                 /* Head lock still held and bh's disabled */
 310                 inet_bind_hash(sk, tb, port);
 311                 if (sk_unhashed(sk)) {
 312                         inet_sk(sk)->sport = htons(port);
 313                         __inet6_hash(hinfo, sk);
 314                 }
 315                 spin_unlock(&head->lock);
 316
 317                 if (tw) {
 318                         inet_twsk_deschedule(tw, death_row);
 319                         inet_twsk_put(tw);
 320                 }
 321
 322                 ret = 0;
 323                 goto out;
 324         }
 325
 326         head = &hinfo->bhash[inet_bhashfn(snum, hinfo->bhash_size)];
 327         tb   = inet_csk(sk)->icsk_bind_hash;
 328         spin_lock_bh(&head->lock);
 329
 330         if (sk_head(&tb->owners) == sk && sk->sk_bind_node.next == NULL) {
 331                 __inet6_hash(hinfo, sk);
 332                 spin_unlock_bh(&head->lock);
 333                 return 0;
 334         } else {
 335                 spin_unlock(&head->lock);
 336                 /* No definite answer... Walk to established hash table */
 337                 ret = __inet6_check_established(death_row, sk, snum, NULL);
 338 out:
 339                 local_bh_enable();
 340                 return ret;
 341         }
 342 }
 343
 344 EXPORT_SYMBOL_GPL(inet6_hash_connect);