include/net/inet_hashtables.h

   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  * Authors:     Lotsa people, from code originally in tcp
   7  *
   8  *      This program is free software; you can redistribute it and/or
   9  *      modify it under the terms of the GNU General Public License
  10  *      as published by the Free Software Foundation; either version
  11  *      2 of the License, or (at your option) any later version.
  12  */
  13
  14 #ifndef _INET_HASHTABLES_H
  15 #define _INET_HASHTABLES_H
  16
  17 #include <linux/interrupt.h>
  18 #include <linux/ip.h>
  19 #include <linux/list.h>
  20 #include <linux/slab.h>
  21 #include <linux/spinlock.h>
  22 #include <linux/types.h>
  23 #include <linux/wait.h>
  24
  25 #include <net/sock.h>
  26 #include <net/tcp_states.h>
  27
  28 #include <asm/atomic.h>
  29
  30 /* This is for all connections with a full identity, no wildcards.
  31  * New scheme, half the table is for TIME_WAIT, the other half is
  32  * for the rest.  I'll experiment with dynamic table growth later.
  33  */
  34 struct inet_ehash_bucket {
  35         rwlock_t          lock;
  36         struct hlist_head chain;
  37 } __attribute__((__aligned__(8)));
  38
  39 /* There are a few simple rules, which allow for local port reuse by
  40  * an application.  In essence:
  41  *
  42  *      1) Sockets bound to different interfaces may share a local port.
  43  *         Failing that, goto test 2.
  44  *      2) If all sockets have sk->sk_reuse set, and none of them are in
  45  *         TCP_LISTEN state, the port may be shared.
  46  *         Failing that, goto test 3.
  47  *      3) If all sockets are bound to a specific inet_sk(sk)->rcv_saddr local
  48  *         address, and none of them are the same, the port may be
  49  *         shared.
  50  *         Failing this, the port cannot be shared.
  51  *
  52  * The interesting point, is test #2.  This is what an FTP server does
  53  * all day.  To optimize this case we use a specific flag bit defined
  54  * below.  As we add sockets to a bind bucket list, we perform a
  55  * check of: (newsk->sk_reuse && (newsk->sk_state != TCP_LISTEN))
  56  * As long as all sockets added to a bind bucket pass this test,
  57  * the flag bit will be set.
  58  * The resulting situation is that tcp_v[46]_verify_bind() can just check
  59  * for this flag bit, if it is set and the socket trying to bind has
  60  * sk->sk_reuse set, we don't even have to walk the owners list at all,
  61  * we return that it is ok to bind this socket to the requested local port.
  62  *
  63  * Sounds like a lot of work, but it is worth it.  In a more naive
  64  * implementation (ie. current FreeBSD etc.) the entire list of ports
  65  * must be walked for each data port opened by an ftp server.  Needless
  66  * to say, this does not scale at all.  With a couple thousand FTP
  67  * users logged onto your box, isn't it nice to know that new data
  68  * ports are created in O(1) time?  I thought so. ;-)   -DaveM
  69  */
  70 struct inet_bind_bucket {
  71         unsigned short          port;
  72         signed short            fastreuse;
  73         struct hlist_node       node;
  74         struct hlist_head       owners;
  75 };
  76
  77 #define inet_bind_bucket_for_each(tb, node, head) \
  78         hlist_for_each_entry(tb, node, head, node)
  79
  80 struct inet_bind_hashbucket {
  81         spinlock_t              lock;
  82         struct hlist_head       chain;
  83 };
  84
  85 /* This is for listening sockets, thus all sockets which possess wildcards. */
  86 #define INET_LHTABLE_SIZE       32      /* Yes, really, this is all you need. */
  87
  88 struct inet_hashinfo {
  89         /* This is for sockets with full identity only.  Sockets here will
  90          * always be without wildcards and will have the following invariant:
  91          *
  92          *          TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
  93          *
  94          * First half of the table is for sockets not in TIME_WAIT, second half
  95          * is for TIME_WAIT sockets only.
  96          */
  97         struct inet_ehash_bucket        *ehash;
  98
  99         /* Ok, let's try this, I give up, we do need a local binding
 100          * TCP hash as well as the others for fast bind/connect.
 101          */
 102         struct inet_bind_hashbucket     *bhash;
 103
 104         int                             bhash_size;
 105         int                             ehash_size;
 106
 107         /* All sockets in TCP_LISTEN state will be in here.  This is the only
 108          * table where wildcard'd TCP sockets can exist.  Hash function here
 109          * is just local port number.
 110          */
 111         struct hlist_head               listening_hash[INET_LHTABLE_SIZE];
 112
 113         /* All the above members are written once at bootup and
 114          * never written again _or_ are predominantly read-access.
 115          *
 116          * Now align to a new cache line as all the following members
 117          * are often dirty.
 118          */
 119         rwlock_t                        lhash_lock ____cacheline_aligned;
 120         atomic_t                        lhash_users;
 121         wait_queue_head_t               lhash_wait;
 122         spinlock_t                      portalloc_lock;
 123         kmem_cache_t                    *bind_bucket_cachep;
 124         int                             port_rover;
 125 };
 126
 127 static inline int inet_ehashfn(const __u32 laddr, const __u16 lport,
 128                                const __u32 faddr, const __u16 fport,
 129                                const int ehash_size)
 130 {
 131         int h = (laddr ^ lport) ^ (faddr ^ fport);
 132         h ^= h >> 16;
 133         h ^= h >> 8;
 134         return h & (ehash_size - 1);
 135 }
 136
 137 static inline int inet_sk_ehashfn(const struct sock *sk, const int ehash_size)
 138 {
 139         const struct inet_sock *inet = inet_sk(sk);
 140         const __u32 laddr = inet->rcv_saddr;
 141         const __u16 lport = inet->num;
 142         const __u32 faddr = inet->daddr;
 143         const __u16 fport = inet->dport;
 144
 145         return inet_ehashfn(laddr, lport, faddr, fport, ehash_size);
 146 }
 147
 148 extern struct inet_bind_bucket *
 149                     inet_bind_bucket_create(kmem_cache_t *cachep,
 150                                             struct inet_bind_hashbucket *head,
 151                                             const unsigned short snum);
 152 extern void inet_bind_bucket_destroy(kmem_cache_t *cachep,
 153                                      struct inet_bind_bucket *tb);
 154
 155 static inline int inet_bhashfn(const __u16 lport, const int bhash_size)
 156 {
 157         return lport & (bhash_size - 1);
 158 }
 159
 160 extern void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
 161                            const unsigned short snum);
 162
 163 /* These can have wildcards, don't try too hard. */
 164 static inline int inet_lhashfn(const unsigned short num)
 165 {
 166         return num & (INET_LHTABLE_SIZE - 1);
 167 }
 168
 169 static inline int inet_sk_listen_hashfn(const struct sock *sk)
 170 {
 171         return inet_lhashfn(inet_sk(sk)->num);
 172 }
 173
 174 /* Caller must disable local BH processing. */
 175 static inline void __inet_inherit_port(struct inet_hashinfo *table,
 176                                        struct sock *sk, struct sock *child)
 177 {
 178         const int bhash = inet_bhashfn(inet_sk(child)->num, table->bhash_size);
 179         struct inet_bind_hashbucket *head = &table->bhash[bhash];
 180         struct inet_bind_bucket *tb;
 181
 182         spin_lock(&head->lock);
 183         tb = inet_sk(sk)->bind_hash;
 184         sk_add_bind_node(child, &tb->owners);
 185         inet_sk(child)->bind_hash = tb;
 186         spin_unlock(&head->lock);
 187 }
 188
 189 static inline void inet_inherit_port(struct inet_hashinfo *table,
 190                                      struct sock *sk, struct sock *child)
 191 {
 192         local_bh_disable();
 193         __inet_inherit_port(table, sk, child);
 194         local_bh_enable();
 195 }
 196
 197 extern void inet_put_port(struct inet_hashinfo *table, struct sock *sk);
 198
 199 extern void inet_listen_wlock(struct inet_hashinfo *hashinfo);
 200
 201 /*
 202  * - We may sleep inside this lock.
 203  * - If sleeping is not required (or called from BH),
 204  *   use plain read_(un)lock(&inet_hashinfo.lhash_lock).
 205  */
 206 static inline void inet_listen_lock(struct inet_hashinfo *hashinfo)
 207 {
 208         /* read_lock synchronizes to candidates to writers */
 209         read_lock(&hashinfo->lhash_lock);
 210         atomic_inc(&hashinfo->lhash_users);
 211         read_unlock(&hashinfo->lhash_lock);
 212 }
 213
 214 static inline void inet_listen_unlock(struct inet_hashinfo *hashinfo)
 215 {
 216         if (atomic_dec_and_test(&hashinfo->lhash_users))
 217                 wake_up(&hashinfo->lhash_wait);
 218 }
 219
 220 static inline void __inet_hash(struct inet_hashinfo *hashinfo,
 221                                struct sock *sk, const int listen_possible)
 222 {
 223         struct hlist_head *list;
 224         rwlock_t *lock;
 225
 226         BUG_TRAP(sk_unhashed(sk));
 227         if (listen_possible && sk->sk_state == TCP_LISTEN) {
 228                 list = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
 229                 lock = &hashinfo->lhash_lock;
 230                 inet_listen_wlock(hashinfo);
 231         } else {
 232                 sk->sk_hashent = inet_sk_ehashfn(sk, hashinfo->ehash_size);
 233                 list = &hashinfo->ehash[sk->sk_hashent].chain;
 234                 lock = &hashinfo->ehash[sk->sk_hashent].lock;
 235                 write_lock(lock);
 236         }
 237         __sk_add_node(sk, list);
 238         sock_prot_inc_use(sk->sk_prot);
 239         write_unlock(lock);
 240         if (listen_possible && sk->sk_state == TCP_LISTEN)
 241                 wake_up(&hashinfo->lhash_wait);
 242 }
 243 #endif /* _INET_HASHTABLES_H */