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                 hash &= (hashinfo->ehash_size - 1);
  41                 list = &hashinfo->ehash[hash].chain;
  42                 lock = &hashinfo->ehash[hash].lock;
  43                 write_lock(lock);
  44         }
  45
  46         __sk_add_node(sk, list);
  47         sock_prot_inc_use(sk->sk_prot);
  48         write_unlock(lock);
  49 }
  50 EXPORT_SYMBOL(__inet6_hash);
  51
  52 /*
  53  * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
  54  * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
  55  *
  56  * The sockhash lock must be held as a reader here.
  57  */
  58 struct sock *__inet6_lookup_established(struct inet_hashinfo *hashinfo,
  59                                            const struct in6_addr *saddr,
  60                                            const __be16 sport,
  61                                            const struct in6_addr *daddr,
  62                                            const u16 hnum,
  63                                            const int dif)
  64 {
  65         struct sock *sk;
  66         const struct hlist_node *node;
  67         const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
  68         /* Optimize here for direct hit, only listening connections can
  69          * have wildcards anyways.
  70          */
  71         unsigned int hash = inet6_ehashfn(daddr, hnum, saddr, sport);
  72         struct inet_ehash_bucket *head = inet_ehash_bucket(hashinfo, hash);
  73
  74         prefetch(head->chain.first);
  75         read_lock(&head->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 + hashinfo->ehash_size)->chain) {
  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(&head->lock);
  96         return NULL;
  97
  98 hit:
  99         sock_hold(sk);
 100         read_unlock(&head->lock);
 101         return sk;
 102 }
 103 EXPORT_SYMBOL(__inet6_lookup_established);
 104
 105 /*
 106  *      Check if a given address matches for an inet socket
 107  *
 108  *      nxi:    the socket's nx_info if any
 109  *      addr:   to be verified address
 110  *      saddr:  socket addresses
 111  */
 112 static inline int inet6_addr_match(
 113         struct nx_info *nxi,
 114         const struct in6_addr *addr,
 115         const struct in6_addr *saddr)
 116 {
 117         if ((addr->s6_addr32[0] || addr->s6_addr32[1] || addr->s6_addr32[2] || addr->s6_addr32[3]) &&
 118                         memcmp(saddr,addr, sizeof(struct in6_addr)) == 0)
 119                 return 1;
 120         if (!(saddr->s6_addr32[0] || saddr->s6_addr32[1] || saddr->s6_addr32[2] || saddr->s6_addr32[3]))
 121                 return addr6_in_nx_info(nxi, addr);
 122         return 0;
 123 }
 124
 125 struct sock *inet6_lookup_listener(struct inet_hashinfo *hashinfo,
 126                                    const struct in6_addr *daddr,
 127                                    const unsigned short hnum, const int dif)
 128 {
 129         struct sock *sk;
 130         const struct hlist_node *node;
 131         struct sock *result = NULL;
 132         int score, hiscore = 0;
 133
 134         read_lock(&hashinfo->lhash_lock);
 135         sk_for_each(sk, node, &hashinfo->listening_hash[inet_lhashfn(hnum)]) {
 136                 if (inet_sk(sk)->num == hnum && sk->sk_family == PF_INET6) {
 137                         const struct ipv6_pinfo *np = inet6_sk(sk);
 138
 139                         score = 1;
 140                         if (!ipv6_addr_any(&np->rcv_saddr)) {
 141                                 if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
 142                                         continue;
 143                                 score++;
 144                         }
 145                         if (sk->sk_bound_dev_if) {
 146                                 if (sk->sk_bound_dev_if != dif)
 147                                         continue;
 148                                 score++;
 149                         }
 150                         if (!inet6_addr_match(sk->sk_nx_info, daddr, &(np->rcv_saddr))) {
 151                                 /* No, this address is not available for guest */
 152                                 continue;
 153                         }
 154                         if (score == 3) {
 155                                 result = sk;
 156                                 break;
 157                         }
 158                         if (score > hiscore) {
 159                                 hiscore = score;
 160                                 result = sk;
 161                         }
 162                 }
 163         }
 164         if (result)
 165                 sock_hold(result);
 166         read_unlock(&hashinfo->lhash_lock);
 167         return result;
 168 }
 169
 170 EXPORT_SYMBOL_GPL(inet6_lookup_listener);
 171
 172 struct sock *inet6_lookup(struct inet_hashinfo *hashinfo,
 173                           const struct in6_addr *saddr, const __be16 sport,
 174                           const struct in6_addr *daddr, const __be16 dport,
 175                           const int dif)
 176 {
 177         struct sock *sk;
 178
 179         local_bh_disable();
 180         sk = __inet6_lookup(hashinfo, saddr, sport, daddr, ntohs(dport), dif);
 181         local_bh_enable();
 182
 183         return sk;
 184 }
 185
 186 EXPORT_SYMBOL_GPL(inet6_lookup);
 187
 188 static int __inet6_check_established(struct inet_timewait_death_row *death_row,
 189                                      struct sock *sk, const __u16 lport,
 190                                      struct inet_timewait_sock **twp)
 191 {
 192         struct inet_hashinfo *hinfo = death_row->hashinfo;
 193         struct inet_sock *inet = inet_sk(sk);
 194         const struct ipv6_pinfo *np = inet6_sk(sk);
 195         const struct in6_addr *daddr = &np->rcv_saddr;
 196         const struct in6_addr *saddr = &np->daddr;
 197         const int dif = sk->sk_bound_dev_if;
 198         const __portpair ports = INET_COMBINED_PORTS(inet->dport, lport);
 199         const unsigned int hash = inet6_ehashfn(daddr, lport, saddr,
 200                                                 inet->dport);
 201         struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
 202         struct sock *sk2;
 203         const struct hlist_node *node;
 204         struct inet_timewait_sock *tw;
 205
 206         prefetch(head->chain.first);
 207         write_lock(&head->lock);
 208
 209         /* Check TIME-WAIT sockets first. */
 210         sk_for_each(sk2, node, &(head + hinfo->ehash_size)->chain) {
 211                 const struct inet6_timewait_sock *tw6 = inet6_twsk(sk2);
 212
 213                 tw = inet_twsk(sk2);
 214
 215                 if(*((__portpair *)&(tw->tw_dport)) == ports             &&
 216                    sk2->sk_family              == PF_INET6       &&
 217                    ipv6_addr_equal(&tw6->tw_v6_daddr, saddr)     &&
 218                    ipv6_addr_equal(&tw6->tw_v6_rcv_saddr, daddr) &&
 219                    sk2->sk_bound_dev_if == sk->sk_bound_dev_if) {
 220                         if (twsk_unique(sk, sk2, twp))
 221                                 goto unique;
 222                         else
 223                                 goto not_unique;
 224                 }
 225         }
 226         tw = NULL;
 227
 228         /* And established part... */
 229         sk_for_each(sk2, node, &head->chain) {
 230                 if (INET6_MATCH(sk2, hash, saddr, daddr, ports, dif))
 231                         goto not_unique;
 232         }
 233
 234 unique:
 235         /* Must record num and sport now. Otherwise we will see
 236          * in hash table socket with a funny identity. */
 237         inet->num = lport;
 238         inet->sport = htons(lport);
 239         BUG_TRAP(sk_unhashed(sk));
 240         __sk_add_node(sk, &head->chain);
 241         sk->sk_hash = hash;
 242         sock_prot_inc_use(sk->sk_prot);
 243         write_unlock(&head->lock);
 244
 245         if (twp != NULL) {
 246                 *twp = tw;
 247                 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
 248         } else if (tw != NULL) {
 249                 /* Silly. Should hash-dance instead... */
 250                 inet_twsk_deschedule(tw, death_row);
 251                 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
 252
 253                 inet_twsk_put(tw);
 254         }
 255         return 0;
 256
 257 not_unique:
 258         write_unlock(&head->lock);
 259         return -EADDRNOTAVAIL;
 260 }
 261
 262 static inline u32 inet6_sk_port_offset(const struct sock *sk)
 263 {
 264         const struct inet_sock *inet = inet_sk(sk);
 265         const struct ipv6_pinfo *np = inet6_sk(sk);
 266         return secure_ipv6_port_ephemeral(np->rcv_saddr.s6_addr32,
 267                                           np->daddr.s6_addr32,
 268                                           inet->dport);
 269 }
 270
 271 int inet6_hash_connect(struct inet_timewait_death_row *death_row,
 272                        struct sock *sk)
 273 {
 274         struct inet_hashinfo *hinfo = death_row->hashinfo;
 275         const unsigned short snum = inet_sk(sk)->num;
 276         struct inet_bind_hashbucket *head;
 277         struct inet_bind_bucket *tb;
 278         int ret;
 279
 280         if (snum == 0) {
 281                 const int low = sysctl_local_port_range[0];
 282                 const int high = sysctl_local_port_range[1];
 283                 const int range = high - low;
 284                 int i, port;
 285                 static u32 hint;
 286                 const u32 offset = hint + inet6_sk_port_offset(sk);
 287                 struct hlist_node *node;
 288                 struct inet_timewait_sock *tw = NULL;
 289
 290                 local_bh_disable();
 291                 for (i = 1; i <= range; i++) {
 292                         port = low + (i + offset) % range;
 293                         head = &hinfo->bhash[inet_bhashfn(port, hinfo->bhash_size)];
 294                         spin_lock(&head->lock);
 295
 296                         /* Does not bother with rcv_saddr checks,
 297                          * because the established check is already
 298                          * unique enough.
 299                          */
 300                         inet_bind_bucket_for_each(tb, node, &head->chain) {
 301                                 if (tb->port == port) {
 302                                         BUG_TRAP(!hlist_empty(&tb->owners));
 303                                         if (tb->fastreuse >= 0)
 304                                                 goto next_port;
 305                                         if (!__inet6_check_established(death_row,
 306                                                                        sk, port,
 307                                                                        &tw))
 308                                                 goto ok;
 309                                         goto next_port;
 310                                 }
 311                         }
 312
 313                         tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
 314                                                      head, port);
 315                         if (!tb) {
 316                                 spin_unlock(&head->lock);
 317                                 break;
 318                         }
 319                         tb->fastreuse = -1;
 320                         goto ok;
 321
 322                 next_port:
 323                         spin_unlock(&head->lock);
 324                 }
 325                 local_bh_enable();
 326
 327                 return -EADDRNOTAVAIL;
 328
 329 ok:
 330                 hint += i;
 331
 332                 /* Head lock still held and bh's disabled */
 333                 inet_bind_hash(sk, tb, port);
 334                 if (sk_unhashed(sk)) {
 335                         inet_sk(sk)->sport = htons(port);
 336                         __inet6_hash(hinfo, sk);
 337                 }
 338                 spin_unlock(&head->lock);
 339
 340                 if (tw) {
 341                         inet_twsk_deschedule(tw, death_row);
 342                         inet_twsk_put(tw);
 343                 }
 344
 345                 ret = 0;
 346                 goto out;
 347         }
 348
 349         head = &hinfo->bhash[inet_bhashfn(snum, hinfo->bhash_size)];
 350         tb   = inet_csk(sk)->icsk_bind_hash;
 351         spin_lock_bh(&head->lock);
 352
 353         if (sk_head(&tb->owners) == sk && sk->sk_bind_node.next == NULL) {
 354                 __inet6_hash(hinfo, sk);
 355                 spin_unlock_bh(&head->lock);
 356                 return 0;
 357         } else {
 358                 spin_unlock(&head->lock);
 359                 /* No definite answer... Walk to established hash table */
 360                 ret = __inet6_check_established(death_row, sk, snum, NULL);
 361 out:
 362                 local_bh_enable();
 363                 return ret;
 364         }
 365 }
 366
 367 EXPORT_SYMBOL_GPL(inet6_hash_connect);