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/secure_seq.h>
  24 #include <net/ip.h>
  25
  26 u32 inet6_ehashfn(const struct net *net,
  27                   const struct in6_addr *laddr, const u16 lport,
  28                   const struct in6_addr *faddr, const __be16 fport)
  29 {
  30         static u32 inet6_ehash_secret __read_mostly;
  31         static u32 ipv6_hash_secret __read_mostly;
  32
  33         u32 lhash, fhash;
  34
  35         net_get_random_once(&inet6_ehash_secret, sizeof(inet6_ehash_secret));
  36         net_get_random_once(&ipv6_hash_secret, sizeof(ipv6_hash_secret));
  37
  38         lhash = (__force u32)laddr->s6_addr32[3];
  39         fhash = __ipv6_addr_jhash(faddr, ipv6_hash_secret);
  40
  41         return __inet6_ehashfn(lhash, lport, fhash, fport,
  42                                inet6_ehash_secret + net_hash_mix(net));
  43 }
  44
  45 /*
  46  * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
  47  * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
  48  *
  49  * The sockhash lock must be held as a reader here.
  50  */
  51 struct sock *__inet6_lookup_established(struct net *net,
  52                                         struct inet_hashinfo *hashinfo,
  53                                            const struct in6_addr *saddr,
  54                                            const __be16 sport,
  55                                            const struct in6_addr *daddr,
  56                                            const u16 hnum,
  57                                            const int dif)
  58 {
  59         struct sock *sk;
  60         const struct hlist_nulls_node *node;
  61         const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
  62         /* Optimize here for direct hit, only listening connections can
  63          * have wildcards anyways.
  64          */
  65         unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
  66         unsigned int slot = hash & hashinfo->ehash_mask;
  67         struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
  68
  69
  70         rcu_read_lock();
  71 begin:
  72         sk_nulls_for_each_rcu(sk, node, &head->chain) {
  73                 if (sk->sk_hash != hash)
  74                         continue;
  75                 if (!INET6_MATCH(sk, net, saddr, daddr, ports, dif))
  76                         continue;
  77                 if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt)))
  78                         goto out;
  79
  80                 if (unlikely(!INET6_MATCH(sk, net, saddr, daddr, ports, dif))) {
  81                         sock_gen_put(sk);
  82                         goto begin;
  83                 }
  84                 goto found;
  85         }
  86         if (get_nulls_value(node) != slot)
  87                 goto begin;
  88 out:
  89         sk = NULL;
  90 found:
  91         rcu_read_unlock();
  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         }
 118         return score;
 119 }
 120
 121 struct sock *inet6_lookup_listener(struct net *net,
 122                 struct inet_hashinfo *hashinfo, const struct in6_addr *saddr,
 123                 const __be16 sport, const struct in6_addr *daddr,
 124                 const unsigned short hnum, const int dif)
 125 {
 126         struct sock *sk;
 127         const struct hlist_nulls_node *node;
 128         struct sock *result;
 129         int score, hiscore, matches = 0, reuseport = 0;
 130         u32 phash = 0;
 131         unsigned int hash = inet_lhashfn(net, hnum);
 132         struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
 133
 134         rcu_read_lock();
 135 begin:
 136         result = NULL;
 137         hiscore = 0;
 138         sk_nulls_for_each(sk, node, &ilb->head) {
 139                 score = compute_score(sk, net, hnum, daddr, dif);
 140                 if (score > hiscore) {
 141                         hiscore = score;
 142                         result = sk;
 143                         reuseport = sk->sk_reuseport;
 144                         if (reuseport) {
 145                                 phash = inet6_ehashfn(net, daddr, hnum,
 146                                                       saddr, sport);
 147                                 matches = 1;
 148                         }
 149                 } else if (score == hiscore && reuseport) {
 150                         matches++;
 151                         if (reciprocal_scale(phash, matches) == 0)
 152                                 result = sk;
 153                         phash = next_pseudo_random32(phash);
 154                 }
 155         }
 156         /*
 157          * if the nulls value we got at the end of this lookup is
 158          * not the expected one, we must restart lookup.
 159          * We probably met an item that was moved to another chain.
 160          */
 161         if (get_nulls_value(node) != hash + LISTENING_NULLS_BASE)
 162                 goto begin;
 163         if (result) {
 164                 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
 165                         result = NULL;
 166                 else if (unlikely(compute_score(result, net, hnum, daddr,
 167                                   dif) < hiscore)) {
 168                         sock_put(result);
 169                         goto begin;
 170                 }
 171         }
 172         rcu_read_unlock();
 173         return result;
 174 }
 175 EXPORT_SYMBOL_GPL(inet6_lookup_listener);
 176
 177 struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
 178                           const struct in6_addr *saddr, const __be16 sport,
 179                           const struct in6_addr *daddr, const __be16 dport,
 180                           const int dif)
 181 {
 182         struct sock *sk;
 183
 184         local_bh_disable();
 185         sk = __inet6_lookup(net, hashinfo, saddr, sport, daddr, ntohs(dport), dif);
 186         local_bh_enable();
 187
 188         return sk;
 189 }
 190 EXPORT_SYMBOL_GPL(inet6_lookup);
 191
 192 static int __inet6_check_established(struct inet_timewait_death_row *death_row,
 193                                      struct sock *sk, const __u16 lport,
 194                                      struct inet_timewait_sock **twp)
 195 {
 196         struct inet_hashinfo *hinfo = death_row->hashinfo;
 197         struct inet_sock *inet = inet_sk(sk);
 198         const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr;
 199         const struct in6_addr *saddr = &sk->sk_v6_daddr;
 200         const int dif = sk->sk_bound_dev_if;
 201         const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
 202         struct net *net = sock_net(sk);
 203         const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
 204                                                 inet->inet_dport);
 205         struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
 206         spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
 207         struct sock *sk2;
 208         const struct hlist_nulls_node *node;
 209         struct inet_timewait_sock *tw = NULL;
 210         int twrefcnt = 0;
 211
 212         spin_lock(lock);
 213
 214         sk_nulls_for_each(sk2, node, &head->chain) {
 215                 if (sk2->sk_hash != hash)
 216                         continue;
 217
 218                 if (likely(INET6_MATCH(sk2, net, saddr, daddr, ports, dif))) {
 219                         if (sk2->sk_state == TCP_TIME_WAIT) {
 220                                 tw = inet_twsk(sk2);
 221                                 if (twsk_unique(sk, sk2, twp))
 222                                         break;
 223                         }
 224                         goto not_unique;
 225                 }
 226         }
 227
 228         /* Must record num and sport now. Otherwise we will see
 229          * in hash table socket with a funny identity.
 230          */
 231         inet->inet_num = lport;
 232         inet->inet_sport = htons(lport);
 233         sk->sk_hash = hash;
 234         WARN_ON(!sk_unhashed(sk));
 235         __sk_nulls_add_node_rcu(sk, &head->chain);
 236         if (tw) {
 237                 twrefcnt = inet_twsk_unhash(tw);
 238                 NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
 239         }
 240         spin_unlock(lock);
 241         if (twrefcnt)
 242                 inet_twsk_put(tw);
 243         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 244
 245         if (twp) {
 246                 *twp = tw;
 247         } else if (tw) {
 248                 /* Silly. Should hash-dance instead... */
 249                 inet_twsk_deschedule(tw);
 250
 251                 inet_twsk_put(tw);
 252         }
 253         return 0;
 254
 255 not_unique:
 256         spin_unlock(lock);
 257         return -EADDRNOTAVAIL;
 258 }
 259
 260 static u32 inet6_sk_port_offset(const struct sock *sk)
 261 {
 262         const struct inet_sock *inet = inet_sk(sk);
 263
 264         return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32,
 265                                           sk->sk_v6_daddr.s6_addr32,
 266                                           inet->inet_dport);
 267 }
 268
 269 int inet6_hash_connect(struct inet_timewait_death_row *death_row,
 270                        struct sock *sk)
 271 {
 272         u32 port_offset = 0;
 273
 274         if (!inet_sk(sk)->inet_num)
 275                 port_offset = inet6_sk_port_offset(sk);
 276         return __inet_hash_connect(death_row, sk, port_offset,
 277                                    __inet6_check_established);
 278 }
 279 EXPORT_SYMBOL_GPL(inet6_hash_connect);