SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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                 if (INET6_TW_MATCH(sk, hash, saddr, daddr, ports, dif))
  84                         goto hit;
  85         }
  86         read_unlock(lock);
  87         return NULL;
  88
  89 hit:
  90         sock_hold(sk);
  91         read_unlock(lock);
  92         return sk;
  93 }
  94 EXPORT_SYMBOL(__inet6_lookup_established);
  95
  96 struct sock *inet6_lookup_listener(struct inet_hashinfo *hashinfo,
  97                                    const struct in6_addr *daddr,
  98                                    const unsigned short hnum, const int dif)
  99 {
 100         struct sock *sk;
 101         const struct hlist_node *node;
 102         struct sock *result = NULL;
 103         int score, hiscore = 0;
 104
 105         read_lock(&hashinfo->lhash_lock);
 106         sk_for_each(sk, node, &hashinfo->listening_hash[inet_lhashfn(hnum)]) {
 107                 if (inet_sk(sk)->num == hnum && sk->sk_family == PF_INET6) {
 108                         const struct ipv6_pinfo *np = inet6_sk(sk);
 109
 110                         score = 1;
 111                         if (!ipv6_addr_any(&np->rcv_saddr)) {
 112                                 if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
 113                                         continue;
 114                                 score++;
 115                         }
 116                         if (sk->sk_bound_dev_if) {
 117                                 if (sk->sk_bound_dev_if != dif)
 118                                         continue;
 119                                 score++;
 120                         }
 121                         if (score == 3) {
 122                                 result = sk;
 123                                 break;
 124                         }
 125                         if (score > hiscore) {
 126                                 hiscore = score;
 127                                 result = sk;
 128                         }
 129                 }
 130         }
 131         if (result)
 132                 sock_hold(result);
 133         read_unlock(&hashinfo->lhash_lock);
 134         return result;
 135 }
 136
 137 EXPORT_SYMBOL_GPL(inet6_lookup_listener);
 138
 139 struct sock *inet6_lookup(struct inet_hashinfo *hashinfo,
 140                           const struct in6_addr *saddr, const __be16 sport,
 141                           const struct in6_addr *daddr, const __be16 dport,
 142                           const int dif)
 143 {
 144         struct sock *sk;
 145
 146         local_bh_disable();
 147         sk = __inet6_lookup(hashinfo, saddr, sport, daddr, ntohs(dport), dif);
 148         local_bh_enable();
 149
 150         return sk;
 151 }
 152
 153 EXPORT_SYMBOL_GPL(inet6_lookup);
 154
 155 static int __inet6_check_established(struct inet_timewait_death_row *death_row,
 156                                      struct sock *sk, const __u16 lport,
 157                                      struct inet_timewait_sock **twp)
 158 {
 159         struct inet_hashinfo *hinfo = death_row->hashinfo;
 160         struct inet_sock *inet = inet_sk(sk);
 161         const struct ipv6_pinfo *np = inet6_sk(sk);
 162         const struct in6_addr *daddr = &np->rcv_saddr;
 163         const struct in6_addr *saddr = &np->daddr;
 164         const int dif = sk->sk_bound_dev_if;
 165         const __portpair ports = INET_COMBINED_PORTS(inet->dport, lport);
 166         const unsigned int hash = inet6_ehashfn(daddr, lport, saddr,
 167                                                 inet->dport);
 168         struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
 169         rwlock_t *lock = inet_ehash_lockp(hinfo, hash);
 170         struct sock *sk2;
 171         const struct hlist_node *node;
 172         struct inet_timewait_sock *tw;
 173
 174         prefetch(head->chain.first);
 175         write_lock(lock);
 176
 177         /* Check TIME-WAIT sockets first. */
 178         sk_for_each(sk2, node, &head->twchain) {
 179                 tw = inet_twsk(sk2);
 180
 181                 if (INET6_TW_MATCH(sk2, hash, saddr, daddr, ports, dif)) {
 182                         if (twsk_unique(sk, sk2, twp))
 183                                 goto unique;
 184                         else
 185                                 goto not_unique;
 186                 }
 187         }
 188         tw = NULL;
 189
 190         /* And established part... */
 191         sk_for_each(sk2, node, &head->chain) {
 192                 if (INET6_MATCH(sk2, hash, saddr, daddr, ports, dif))
 193                         goto not_unique;
 194         }
 195
 196 unique:
 197         /* Must record num and sport now. Otherwise we will see
 198          * in hash table socket with a funny identity. */
 199         inet->num = lport;
 200         inet->sport = htons(lport);
 201         BUG_TRAP(sk_unhashed(sk));
 202         __sk_add_node(sk, &head->chain);
 203         sk->sk_hash = hash;
 204         sock_prot_inuse_add(sk->sk_prot, 1);
 205         write_unlock(lock);
 206
 207         if (twp != NULL) {
 208                 *twp = tw;
 209                 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
 210         } else if (tw != NULL) {
 211                 /* Silly. Should hash-dance instead... */
 212                 inet_twsk_deschedule(tw, death_row);
 213                 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
 214
 215                 inet_twsk_put(tw);
 216         }
 217         return 0;
 218
 219 not_unique:
 220         write_unlock(lock);
 221         return -EADDRNOTAVAIL;
 222 }
 223
 224 static inline u32 inet6_sk_port_offset(const struct sock *sk)
 225 {
 226         const struct inet_sock *inet = inet_sk(sk);
 227         const struct ipv6_pinfo *np = inet6_sk(sk);
 228         return secure_ipv6_port_ephemeral(np->rcv_saddr.s6_addr32,
 229                                           np->daddr.s6_addr32,
 230                                           inet->dport);
 231 }
 232
 233 int inet6_hash_connect(struct inet_timewait_death_row *death_row,
 234                        struct sock *sk)
 235 {
 236         struct inet_hashinfo *hinfo = death_row->hashinfo;
 237         const unsigned short snum = inet_sk(sk)->num;
 238         struct inet_bind_hashbucket *head;
 239         struct inet_bind_bucket *tb;
 240         int ret;
 241
 242         if (snum == 0) {
 243                 int i, port, low, high, remaining;
 244                 static u32 hint;
 245                 const u32 offset = hint + inet6_sk_port_offset(sk);
 246                 struct hlist_node *node;
 247                 struct inet_timewait_sock *tw = NULL;
 248
 249                 inet_get_local_port_range(&low, &high);
 250                 remaining = (high - low) + 1;
 251
 252                 local_bh_disable();
 253                 for (i = 1; i <= remaining; i++) {
 254                         port = low + (i + offset) % remaining;
 255                         head = &hinfo->bhash[inet_bhashfn(port, hinfo->bhash_size)];
 256                         spin_lock(&head->lock);
 257
 258                         /* Does not bother with rcv_saddr checks,
 259                          * because the established check is already
 260                          * unique enough.
 261                          */
 262                         inet_bind_bucket_for_each(tb, node, &head->chain) {
 263                                 if (tb->port == port) {
 264                                         BUG_TRAP(!hlist_empty(&tb->owners));
 265                                         if (tb->fastreuse >= 0)
 266                                                 goto next_port;
 267                                         if (!__inet6_check_established(death_row,
 268                                                                        sk, port,
 269                                                                        &tw))
 270                                                 goto ok;
 271                                         goto next_port;
 272                                 }
 273                         }
 274
 275                         tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
 276                                                      head, port);
 277                         if (!tb) {
 278                                 spin_unlock(&head->lock);
 279                                 break;
 280                         }
 281                         tb->fastreuse = -1;
 282                         goto ok;
 283
 284                 next_port:
 285                         spin_unlock(&head->lock);
 286                 }
 287                 local_bh_enable();
 288
 289                 return -EADDRNOTAVAIL;
 290
 291 ok:
 292                 hint += i;
 293
 294                 /* Head lock still held and bh's disabled */
 295                 inet_bind_hash(sk, tb, port);
 296                 if (sk_unhashed(sk)) {
 297                         inet_sk(sk)->sport = htons(port);
 298                         __inet6_hash(hinfo, sk);
 299                 }
 300                 spin_unlock(&head->lock);
 301
 302                 if (tw) {
 303                         inet_twsk_deschedule(tw, death_row);
 304                         inet_twsk_put(tw);
 305                 }
 306
 307                 ret = 0;
 308                 goto out;
 309         }
 310
 311         head = &hinfo->bhash[inet_bhashfn(snum, hinfo->bhash_size)];
 312         tb   = inet_csk(sk)->icsk_bind_hash;
 313         spin_lock_bh(&head->lock);
 314
 315         if (sk_head(&tb->owners) == sk && sk->sk_bind_node.next == NULL) {
 316                 __inet6_hash(hinfo, sk);
 317                 spin_unlock_bh(&head->lock);
 318                 return 0;
 319         } else {
 320                 spin_unlock(&head->lock);
 321                 /* No definite answer... Walk to established hash table */
 322                 ret = __inet6_check_established(death_row, sk, snum, NULL);
 323 out:
 324                 local_bh_enable();
 325                 return ret;
 326         }
 327 }
 328
 329 EXPORT_SYMBOL_GPL(inet6_hash_connect);