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/config.h>
  18 #include <linux/module.h>
  19 #include <linux/random.h>
  20
  21 #include <net/inet_connection_sock.h>
  22 #include <net/inet_hashtables.h>
  23 #include <net/inet6_hashtables.h>
  24 #include <net/ip.h>
  25
  26 struct sock *inet6_lookup_listener(struct inet_hashinfo *hashinfo,
  27                                    const struct in6_addr *daddr,
  28                                    const unsigned short hnum, const int dif)
  29 {
  30         struct sock *sk;
  31         const struct hlist_node *node;
  32         struct sock *result = NULL;
  33         int score, hiscore = 0;
  34
  35         read_lock(&hashinfo->lhash_lock);
  36         sk_for_each(sk, node, &hashinfo->listening_hash[inet_lhashfn(hnum)]) {
  37                 if (inet_sk(sk)->num == hnum && sk->sk_family == PF_INET6) {
  38                         const struct ipv6_pinfo *np = inet6_sk(sk);
  39
  40                         score = 1;
  41                         if (!ipv6_addr_any(&np->rcv_saddr)) {
  42                                 if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
  43                                         continue;
  44                                 score++;
  45                         }
  46                         if (sk->sk_bound_dev_if) {
  47                                 if (sk->sk_bound_dev_if != dif)
  48                                         continue;
  49                                 score++;
  50                         }
  51                         if (score == 3) {
  52                                 result = sk;
  53                                 break;
  54                         }
  55                         if (score > hiscore) {
  56                                 hiscore = score;
  57                                 result = sk;
  58                         }
  59                 }
  60         }
  61         if (result)
  62                 sock_hold(result);
  63         read_unlock(&hashinfo->lhash_lock);
  64         return result;
  65 }
  66
  67 EXPORT_SYMBOL_GPL(inet6_lookup_listener);
  68
  69 struct sock *inet6_lookup(struct inet_hashinfo *hashinfo,
  70                           const struct in6_addr *saddr, const u16 sport,
  71                           const struct in6_addr *daddr, const u16 dport,
  72                           const int dif)
  73 {
  74         struct sock *sk;
  75
  76         local_bh_disable();
  77         sk = __inet6_lookup(hashinfo, saddr, sport, daddr, ntohs(dport), dif);
  78         local_bh_enable();
  79
  80         return sk;
  81 }
  82
  83 EXPORT_SYMBOL_GPL(inet6_lookup);
  84
  85 static int __inet6_check_established(struct inet_timewait_death_row *death_row,
  86                                      struct sock *sk, const __u16 lport,
  87                                      struct inet_timewait_sock **twp)
  88 {
  89         struct inet_hashinfo *hinfo = death_row->hashinfo;
  90         struct inet_sock *inet = inet_sk(sk);
  91         const struct ipv6_pinfo *np = inet6_sk(sk);
  92         const struct in6_addr *daddr = &np->rcv_saddr;
  93         const struct in6_addr *saddr = &np->daddr;
  94         const int dif = sk->sk_bound_dev_if;
  95         const u32 ports = INET_COMBINED_PORTS(inet->dport, lport);
  96         const unsigned int hash = inet6_ehashfn(daddr, inet->num, saddr,
  97                                                 inet->dport);
  98         struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
  99         struct sock *sk2;
 100         const struct hlist_node *node;
 101         struct inet_timewait_sock *tw;
 102
 103         prefetch(head->chain.first);
 104         write_lock(&head->lock);
 105
 106         /* Check TIME-WAIT sockets first. */
 107         sk_for_each(sk2, node, &(head + hinfo->ehash_size)->chain) {
 108                 const struct inet6_timewait_sock *tw6 = inet6_twsk(sk2);
 109
 110                 tw = inet_twsk(sk2);
 111
 112                 if(*((__u32 *)&(tw->tw_dport)) == ports          &&
 113                    sk2->sk_family              == PF_INET6       &&
 114                    ipv6_addr_equal(&tw6->tw_v6_daddr, saddr)     &&
 115                    ipv6_addr_equal(&tw6->tw_v6_rcv_saddr, daddr) &&
 116                    sk2->sk_bound_dev_if == sk->sk_bound_dev_if) {
 117                         if (twsk_unique(sk, sk2, twp))
 118                                 goto unique;
 119                         else
 120                                 goto not_unique;
 121                 }
 122         }
 123         tw = NULL;
 124
 125         /* And established part... */
 126         sk_for_each(sk2, node, &head->chain) {
 127                 if (INET6_MATCH(sk2, hash, saddr, daddr, ports, dif))
 128                         goto not_unique;
 129         }
 130
 131 unique:
 132         /* Must record num and sport now. Otherwise we will see
 133          * in hash table socket with a funny identity. */
 134         inet->num = lport;
 135         inet->sport = htons(lport);
 136         BUG_TRAP(sk_unhashed(sk));
 137         __sk_add_node(sk, &head->chain);
 138         sk->sk_hash = hash;
 139         sock_prot_inc_use(sk->sk_prot);
 140         write_unlock(&head->lock);
 141
 142         if (twp != NULL) {
 143                 *twp = tw;
 144                 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
 145         } else if (tw != NULL) {
 146                 /* Silly. Should hash-dance instead... */
 147                 inet_twsk_deschedule(tw, death_row);
 148                 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
 149
 150                 inet_twsk_put(tw);
 151         }
 152         return 0;
 153
 154 not_unique:
 155         write_unlock(&head->lock);
 156         return -EADDRNOTAVAIL;
 157 }
 158
 159 static inline u32 inet6_sk_port_offset(const struct sock *sk)
 160 {
 161         const struct inet_sock *inet = inet_sk(sk);
 162         const struct ipv6_pinfo *np = inet6_sk(sk);
 163         return secure_ipv6_port_ephemeral(np->rcv_saddr.s6_addr32,
 164                                           np->daddr.s6_addr32,
 165                                           inet->dport);
 166 }
 167
 168 int inet6_hash_connect(struct inet_timewait_death_row *death_row,
 169                        struct sock *sk)
 170 {
 171         struct inet_hashinfo *hinfo = death_row->hashinfo;
 172         const unsigned short snum = inet_sk(sk)->num;
 173         struct inet_bind_hashbucket *head;
 174         struct inet_bind_bucket *tb;
 175         int ret;
 176
 177         if (snum == 0) {
 178                 const int low = sysctl_local_port_range[0];
 179                 const int high = sysctl_local_port_range[1];
 180                 const int range = high - low;
 181                 int i, port;
 182                 static u32 hint;
 183                 const u32 offset = hint + inet6_sk_port_offset(sk);
 184                 struct hlist_node *node;
 185                 struct inet_timewait_sock *tw = NULL;
 186
 187                 local_bh_disable();
 188                 for (i = 1; i <= range; i++) {
 189                         port = low + (i + offset) % range;
 190                         head = &hinfo->bhash[inet_bhashfn(port, hinfo->bhash_size)];
 191                         spin_lock(&head->lock);
 192
 193                         /* Does not bother with rcv_saddr checks,
 194                          * because the established check is already
 195                          * unique enough.
 196                          */
 197                         inet_bind_bucket_for_each(tb, node, &head->chain) {
 198                                 if (tb->port == port) {
 199                                         BUG_TRAP(!hlist_empty(&tb->owners));
 200                                         if (tb->fastreuse >= 0)
 201                                                 goto next_port;
 202                                         if (!__inet6_check_established(death_row,
 203                                                                        sk, port,
 204                                                                        &tw))
 205                                                 goto ok;
 206                                         goto next_port;
 207                                 }
 208                         }
 209
 210                         tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
 211                                                      head, port);
 212                         if (!tb) {
 213                                 spin_unlock(&head->lock);
 214                                 break;
 215                         }
 216                         tb->fastreuse = -1;
 217                         goto ok;
 218
 219                 next_port:
 220                         spin_unlock(&head->lock);
 221                 }
 222                 local_bh_enable();
 223
 224                 return -EADDRNOTAVAIL;
 225
 226 ok:
 227                 hint += i;
 228
 229                 /* Head lock still held and bh's disabled */
 230                 inet_bind_hash(sk, tb, port);
 231                 if (sk_unhashed(sk)) {
 232                         inet_sk(sk)->sport = htons(port);
 233                         __inet6_hash(hinfo, sk);
 234                 }
 235                 spin_unlock(&head->lock);
 236
 237                 if (tw) {
 238                         inet_twsk_deschedule(tw, death_row);
 239                         inet_twsk_put(tw);
 240                 }
 241
 242                 ret = 0;
 243                 goto out;
 244         }
 245
 246         head = &hinfo->bhash[inet_bhashfn(snum, hinfo->bhash_size)];
 247         tb   = inet_csk(sk)->icsk_bind_hash;
 248         spin_lock_bh(&head->lock);
 249
 250         if (sk_head(&tb->owners) == sk && sk->sk_bind_node.next == NULL) {
 251                 __inet6_hash(hinfo, sk);
 252                 spin_unlock_bh(&head->lock);
 253                 return 0;
 254         } else {
 255                 spin_unlock(&head->lock);
 256                 /* No definite answer... Walk to established hash table */
 257                 ret = __inet6_check_established(death_row, sk, snum, NULL);
 258 out:
 259                 local_bh_enable();
 260                 return ret;
 261         }
 262 }
 263
 264 EXPORT_SYMBOL_GPL(inet6_hash_connect);