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 int __inet6_hash(struct sock *sk, struct inet_timewait_sock *tw)
  27 {
  28         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
  29         int twrefcnt = 0;
  30
  31         WARN_ON(!sk_unhashed(sk));
  32
  33         if (sk->sk_state == TCP_LISTEN) {
  34                 struct inet_listen_hashbucket *ilb;
  35
  36                 ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
  37                 spin_lock(&ilb->lock);
  38                 __sk_nulls_add_node_rcu(sk, &ilb->head);
  39                 spin_unlock(&ilb->lock);
  40         } else {
  41                 unsigned int hash;
  42                 struct hlist_nulls_head *list;
  43                 spinlock_t *lock;
  44
  45                 sk->sk_hash = hash = inet6_sk_ehashfn(sk);
  46                 list = &inet_ehash_bucket(hashinfo, hash)->chain;
  47                 lock = inet_ehash_lockp(hashinfo, hash);
  48                 spin_lock(lock);
  49                 __sk_nulls_add_node_rcu(sk, list);
  50                 if (tw) {
  51                         WARN_ON(sk->sk_hash != tw->tw_hash);
  52                         twrefcnt = inet_twsk_unhash(tw);
  53                 }
  54                 spin_unlock(lock);
  55         }
  56
  57         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
  58         return twrefcnt;
  59 }
  60 EXPORT_SYMBOL(__inet6_hash);
  61
  62 /*
  63  * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
  64  * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
  65  *
  66  * The sockhash lock must be held as a reader here.
  67  */
  68 struct sock *__inet6_lookup_established(struct net *net,
  69                                         struct inet_hashinfo *hashinfo,
  70                                            const struct in6_addr *saddr,
  71                                            const __be16 sport,
  72                                            const struct in6_addr *daddr,
  73                                            const u16 hnum,
  74                                            const int dif)
  75 {
  76         struct sock *sk;
  77         const struct hlist_nulls_node *node;
  78         const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
  79         /* Optimize here for direct hit, only listening connections can
  80          * have wildcards anyways.
  81          */
  82         unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
  83         unsigned int slot = hash & hashinfo->ehash_mask;
  84         struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
  85
  86
  87         rcu_read_lock();
  88 begin:
  89         sk_nulls_for_each_rcu(sk, node, &head->chain) {
  90                 if (sk->sk_hash != hash)
  91                         continue;
  92                 if (!INET6_MATCH(sk, net, saddr, daddr, ports, dif))
  93                         continue;
  94                 if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt)))
  95                         goto out;
  96
  97                 if (unlikely(!INET6_MATCH(sk, net, saddr, daddr, ports, dif))) {
  98                         sock_gen_put(sk);
  99                         goto begin;
 100                 }
 101                 goto found;
 102         }
 103         if (get_nulls_value(node) != slot)
 104                 goto begin;
 105 out:
 106         sk = NULL;
 107 found:
 108         rcu_read_unlock();
 109         return sk;
 110 }
 111 EXPORT_SYMBOL(__inet6_lookup_established);
 112
 113 static inline int compute_score(struct sock *sk, struct net *net,
 114                                 const unsigned short hnum,
 115                                 const struct in6_addr *daddr,
 116                                 const int dif)
 117 {
 118         int score = -1;
 119
 120         if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum &&
 121             sk->sk_family == PF_INET6) {
 122
 123                 score = 1;
 124                 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
 125                         if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
 126                                 return -1;
 127                         score++;
 128                 }
 129                 if (sk->sk_bound_dev_if) {
 130                         if (sk->sk_bound_dev_if != dif)
 131                                 return -1;
 132                         score++;
 133                 }
 134         }
 135         return score;
 136 }
 137
 138 struct sock *inet6_lookup_listener(struct net *net,
 139                 struct inet_hashinfo *hashinfo, const struct in6_addr *saddr,
 140                 const __be16 sport, const struct in6_addr *daddr,
 141                 const unsigned short hnum, const int dif)
 142 {
 143         struct sock *sk;
 144         const struct hlist_nulls_node *node;
 145         struct sock *result;
 146         int score, hiscore, matches = 0, reuseport = 0;
 147         u32 phash = 0;
 148         unsigned int hash = inet_lhashfn(net, hnum);
 149         struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
 150
 151         rcu_read_lock();
 152 begin:
 153         result = NULL;
 154         hiscore = 0;
 155         sk_nulls_for_each(sk, node, &ilb->head) {
 156                 score = compute_score(sk, net, hnum, daddr, dif);
 157                 if (score > hiscore) {
 158                         hiscore = score;
 159                         result = sk;
 160                         reuseport = sk->sk_reuseport;
 161                         if (reuseport) {
 162                                 phash = inet6_ehashfn(net, daddr, hnum,
 163                                                       saddr, sport);
 164                                 matches = 1;
 165                         }
 166                 } else if (score == hiscore && reuseport) {
 167                         matches++;
 168                         if (((u64)phash * matches) >> 32 == 0)
 169                                 result = sk;
 170                         phash = next_pseudo_random32(phash);
 171                 }
 172         }
 173         /*
 174          * if the nulls value we got at the end of this lookup is
 175          * not the expected one, we must restart lookup.
 176          * We probably met an item that was moved to another chain.
 177          */
 178         if (get_nulls_value(node) != hash + LISTENING_NULLS_BASE)
 179                 goto begin;
 180         if (result) {
 181                 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
 182                         result = NULL;
 183                 else if (unlikely(compute_score(result, net, hnum, daddr,
 184                                   dif) < hiscore)) {
 185                         sock_put(result);
 186                         goto begin;
 187                 }
 188         }
 189         rcu_read_unlock();
 190         return result;
 191 }
 192
 193 EXPORT_SYMBOL_GPL(inet6_lookup_listener);
 194
 195 struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
 196                           const struct in6_addr *saddr, const __be16 sport,
 197                           const struct in6_addr *daddr, const __be16 dport,
 198                           const int dif)
 199 {
 200         struct sock *sk;
 201
 202         local_bh_disable();
 203         sk = __inet6_lookup(net, hashinfo, saddr, sport, daddr, ntohs(dport), dif);
 204         local_bh_enable();
 205
 206         return sk;
 207 }
 208
 209 EXPORT_SYMBOL_GPL(inet6_lookup);
 210
 211 static int __inet6_check_established(struct inet_timewait_death_row *death_row,
 212                                      struct sock *sk, const __u16 lport,
 213                                      struct inet_timewait_sock **twp)
 214 {
 215         struct inet_hashinfo *hinfo = death_row->hashinfo;
 216         struct inet_sock *inet = inet_sk(sk);
 217         const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr;
 218         const struct in6_addr *saddr = &sk->sk_v6_daddr;
 219         const int dif = sk->sk_bound_dev_if;
 220         const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
 221         struct net *net = sock_net(sk);
 222         const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
 223                                                 inet->inet_dport);
 224         struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
 225         spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
 226         struct sock *sk2;
 227         const struct hlist_nulls_node *node;
 228         struct inet_timewait_sock *tw = NULL;
 229         int twrefcnt = 0;
 230
 231         spin_lock(lock);
 232
 233         sk_nulls_for_each(sk2, node, &head->chain) {
 234                 if (sk2->sk_hash != hash)
 235                         continue;
 236
 237                 if (likely(INET6_MATCH(sk2, net, saddr, daddr, ports, dif))) {
 238                         if (sk2->sk_state == TCP_TIME_WAIT) {
 239                                 tw = inet_twsk(sk2);
 240                                 if (twsk_unique(sk, sk2, twp))
 241                                         break;
 242                         }
 243                         goto not_unique;
 244                 }
 245         }
 246
 247         /* Must record num and sport now. Otherwise we will see
 248          * in hash table socket with a funny identity.
 249          */
 250         inet->inet_num = lport;
 251         inet->inet_sport = htons(lport);
 252         sk->sk_hash = hash;
 253         WARN_ON(!sk_unhashed(sk));
 254         __sk_nulls_add_node_rcu(sk, &head->chain);
 255         if (tw) {
 256                 twrefcnt = inet_twsk_unhash(tw);
 257                 NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
 258         }
 259         spin_unlock(lock);
 260         if (twrefcnt)
 261                 inet_twsk_put(tw);
 262         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 263
 264         if (twp) {
 265                 *twp = tw;
 266         } else if (tw) {
 267                 /* Silly. Should hash-dance instead... */
 268                 inet_twsk_deschedule(tw, death_row);
 269
 270                 inet_twsk_put(tw);
 271         }
 272         return 0;
 273
 274 not_unique:
 275         spin_unlock(lock);
 276         return -EADDRNOTAVAIL;
 277 }
 278
 279 static inline u32 inet6_sk_port_offset(const struct sock *sk)
 280 {
 281         const struct inet_sock *inet = inet_sk(sk);
 282
 283         return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32,
 284                                           sk->sk_v6_daddr.s6_addr32,
 285                                           inet->inet_dport);
 286 }
 287
 288 int inet6_hash_connect(struct inet_timewait_death_row *death_row,
 289                        struct sock *sk)
 290 {
 291         return __inet_hash_connect(death_row, sk, inet6_sk_port_offset(sk),
 292                         __inet6_check_established, __inet6_hash);
 293 }
 294
 295 EXPORT_SYMBOL_GPL(inet6_hash_connect);