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 (sk->sk_state == TCP_TIME_WAIT) {
  93                         if (!INET6_TW_MATCH(sk, net, saddr, daddr, ports, dif))
  94                                 continue;
  95                 } else {
  96                         if (!INET6_MATCH(sk, net, saddr, daddr, ports, dif))
  97                                 continue;
  98                 }
  99                 if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt)))
 100                         goto out;
 101
 102                 if (sk->sk_state == TCP_TIME_WAIT) {
 103                         if (unlikely(!INET6_TW_MATCH(sk, net, saddr, daddr,
 104                                                      ports, dif))) {
 105                                 sock_gen_put(sk);
 106                                 goto begin;
 107                         }
 108                 } else {
 109                         if (unlikely(!INET6_MATCH(sk, net, saddr, daddr,
 110                                                   ports, dif))) {
 111                                 sock_put(sk);
 112                                 goto begin;
 113                         }
 114                 goto found;
 115                 }
 116         }
 117         if (get_nulls_value(node) != slot)
 118                 goto begin;
 119 out:
 120         sk = NULL;
 121 found:
 122         rcu_read_unlock();
 123         return sk;
 124 }
 125 EXPORT_SYMBOL(__inet6_lookup_established);
 126
 127 static inline int compute_score(struct sock *sk, struct net *net,
 128                                 const unsigned short hnum,
 129                                 const struct in6_addr *daddr,
 130                                 const int dif)
 131 {
 132         int score = -1;
 133
 134         if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum &&
 135             sk->sk_family == PF_INET6) {
 136                 const struct ipv6_pinfo *np = inet6_sk(sk);
 137
 138                 score = 1;
 139                 if (!ipv6_addr_any(&np->rcv_saddr)) {
 140                         if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
 141                                 return -1;
 142                         score++;
 143                 }
 144                 if (sk->sk_bound_dev_if) {
 145                         if (sk->sk_bound_dev_if != dif)
 146                                 return -1;
 147                         score++;
 148                 }
 149         }
 150         return score;
 151 }
 152
 153 struct sock *inet6_lookup_listener(struct net *net,
 154                 struct inet_hashinfo *hashinfo, const struct in6_addr *saddr,
 155                 const __be16 sport, const struct in6_addr *daddr,
 156                 const unsigned short hnum, const int dif)
 157 {
 158         struct sock *sk;
 159         const struct hlist_nulls_node *node;
 160         struct sock *result;
 161         int score, hiscore, matches = 0, reuseport = 0;
 162         u32 phash = 0;
 163         unsigned int hash = inet_lhashfn(net, hnum);
 164         struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
 165
 166         rcu_read_lock();
 167 begin:
 168         result = NULL;
 169         hiscore = 0;
 170         sk_nulls_for_each(sk, node, &ilb->head) {
 171                 score = compute_score(sk, net, hnum, daddr, dif);
 172                 if (score > hiscore) {
 173                         hiscore = score;
 174                         result = sk;
 175                         reuseport = sk->sk_reuseport;
 176                         if (reuseport) {
 177                                 phash = inet6_ehashfn(net, daddr, hnum,
 178                                                       saddr, sport);
 179                                 matches = 1;
 180                         }
 181                 } else if (score == hiscore && reuseport) {
 182                         matches++;
 183                         if (((u64)phash * matches) >> 32 == 0)
 184                                 result = sk;
 185                         phash = next_pseudo_random32(phash);
 186                 }
 187         }
 188         /*
 189          * if the nulls value we got at the end of this lookup is
 190          * not the expected one, we must restart lookup.
 191          * We probably met an item that was moved to another chain.
 192          */
 193         if (get_nulls_value(node) != hash + LISTENING_NULLS_BASE)
 194                 goto begin;
 195         if (result) {
 196                 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
 197                         result = NULL;
 198                 else if (unlikely(compute_score(result, net, hnum, daddr,
 199                                   dif) < hiscore)) {
 200                         sock_put(result);
 201                         goto begin;
 202                 }
 203         }
 204         rcu_read_unlock();
 205         return result;
 206 }
 207
 208 EXPORT_SYMBOL_GPL(inet6_lookup_listener);
 209
 210 struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
 211                           const struct in6_addr *saddr, const __be16 sport,
 212                           const struct in6_addr *daddr, const __be16 dport,
 213                           const int dif)
 214 {
 215         struct sock *sk;
 216
 217         local_bh_disable();
 218         sk = __inet6_lookup(net, hashinfo, saddr, sport, daddr, ntohs(dport), dif);
 219         local_bh_enable();
 220
 221         return sk;
 222 }
 223
 224 EXPORT_SYMBOL_GPL(inet6_lookup);
 225
 226 static int __inet6_check_established(struct inet_timewait_death_row *death_row,
 227                                      struct sock *sk, const __u16 lport,
 228                                      struct inet_timewait_sock **twp)
 229 {
 230         struct inet_hashinfo *hinfo = death_row->hashinfo;
 231         struct inet_sock *inet = inet_sk(sk);
 232         const struct ipv6_pinfo *np = inet6_sk(sk);
 233         const struct in6_addr *daddr = &np->rcv_saddr;
 234         const struct in6_addr *saddr = &np->daddr;
 235         const int dif = sk->sk_bound_dev_if;
 236         const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
 237         struct net *net = sock_net(sk);
 238         const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
 239                                                 inet->inet_dport);
 240         struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
 241         spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
 242         struct sock *sk2;
 243         const struct hlist_nulls_node *node;
 244         struct inet_timewait_sock *tw = NULL;
 245         int twrefcnt = 0;
 246
 247         spin_lock(lock);
 248
 249         sk_nulls_for_each(sk2, node, &head->chain) {
 250                 if (sk2->sk_hash != hash)
 251                         continue;
 252
 253                 if (sk2->sk_state == TCP_TIME_WAIT) {
 254                         if (likely(INET6_TW_MATCH(sk2, net, saddr, daddr,
 255                                                   ports, dif))) {
 256                                 tw = inet_twsk(sk2);
 257                                 if (twsk_unique(sk, sk2, twp))
 258                                         goto unique;
 259                                 else
 260                                         goto not_unique;
 261                         }
 262                 }
 263                 if (likely(INET6_MATCH(sk2, net, saddr, daddr, ports, dif)))
 264                         goto not_unique;
 265         }
 266
 267 unique:
 268         /* Must record num and sport now. Otherwise we will see
 269          * in hash table socket with a funny identity. */
 270         inet->inet_num = lport;
 271         inet->inet_sport = htons(lport);
 272         sk->sk_hash = hash;
 273         WARN_ON(!sk_unhashed(sk));
 274         __sk_nulls_add_node_rcu(sk, &head->chain);
 275         if (tw) {
 276                 twrefcnt = inet_twsk_unhash(tw);
 277                 NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
 278         }
 279         spin_unlock(lock);
 280         if (twrefcnt)
 281                 inet_twsk_put(tw);
 282         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 283
 284         if (twp) {
 285                 *twp = tw;
 286         } else if (tw) {
 287                 /* Silly. Should hash-dance instead... */
 288                 inet_twsk_deschedule(tw, death_row);
 289
 290                 inet_twsk_put(tw);
 291         }
 292         return 0;
 293
 294 not_unique:
 295         spin_unlock(lock);
 296         return -EADDRNOTAVAIL;
 297 }
 298
 299 static inline u32 inet6_sk_port_offset(const struct sock *sk)
 300 {
 301         const struct inet_sock *inet = inet_sk(sk);
 302         const struct ipv6_pinfo *np = inet6_sk(sk);
 303         return secure_ipv6_port_ephemeral(np->rcv_saddr.s6_addr32,
 304                                           np->daddr.s6_addr32,
 305                                           inet->inet_dport);
 306 }
 307
 308 int inet6_hash_connect(struct inet_timewait_death_row *death_row,
 309                        struct sock *sk)
 310 {
 311         return __inet_hash_connect(death_row, sk, inet6_sk_port_offset(sk),
 312                         __inet6_check_established, __inet6_hash);
 313 }
 314
 315 EXPORT_SYMBOL_GPL(inet6_hash_connect);