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
  18 #include <linux/interrupt.h>
  19 #include <linux/ip.h>
  20 #include <linux/ipv6.h>
  21 #include <linux/list.h>
  22 #include <linux/slab.h>
  23 #include <linux/socket.h>
  24 #include <linux/spinlock.h>
  25 #include <linux/types.h>
  26 #include <linux/wait.h>
  27 #include <linux/vmalloc.h>
  28
  29 #include <net/inet_connection_sock.h>
  30 #include <net/inet_sock.h>
  31 #include <net/sock.h>
  32 #include <net/route.h>
  33 #include <net/tcp_states.h>
  34 #include <net/netns/hash.h>
  35
  36 #include <linux/atomic.h>
  37 #include <asm/byteorder.h>
  38
  39 /* This is for all connections with a full identity, no wildcards.
  40  * One chain is dedicated to TIME_WAIT sockets.
  41  * I'll experiment with dynamic table growth later.
  42  */
  43 struct inet_ehash_bucket {
  44         struct hlist_nulls_head chain;
  45         struct hlist_nulls_head twchain;
  46 };
  47
  48 /* There are a few simple rules, which allow for local port reuse by
  49  * an application.  In essence:
  50  *
  51  *      1) Sockets bound to different interfaces may share a local port.
  52  *         Failing that, goto test 2.
  53  *      2) If all sockets have sk->sk_reuse set, and none of them are in
  54  *         TCP_LISTEN state, the port may be shared.
  55  *         Failing that, goto test 3.
  56  *      3) If all sockets are bound to a specific inet_sk(sk)->rcv_saddr local
  57  *         address, and none of them are the same, the port may be
  58  *         shared.
  59  *         Failing this, the port cannot be shared.
  60  *
  61  * The interesting point, is test #2.  This is what an FTP server does
  62  * all day.  To optimize this case we use a specific flag bit defined
  63  * below.  As we add sockets to a bind bucket list, we perform a
  64  * check of: (newsk->sk_reuse && (newsk->sk_state != TCP_LISTEN))
  65  * As long as all sockets added to a bind bucket pass this test,
  66  * the flag bit will be set.
  67  * The resulting situation is that tcp_v[46]_verify_bind() can just check
  68  * for this flag bit, if it is set and the socket trying to bind has
  69  * sk->sk_reuse set, we don't even have to walk the owners list at all,
  70  * we return that it is ok to bind this socket to the requested local port.
  71  *
  72  * Sounds like a lot of work, but it is worth it.  In a more naive
  73  * implementation (ie. current FreeBSD etc.) the entire list of ports
  74  * must be walked for each data port opened by an ftp server.  Needless
  75  * to say, this does not scale at all.  With a couple thousand FTP
  76  * users logged onto your box, isn't it nice to know that new data
  77  * ports are created in O(1) time?  I thought so. ;-)   -DaveM
  78  */
  79 struct inet_bind_bucket {
  80 #ifdef CONFIG_NET_NS
  81         struct net              *ib_net;
  82 #endif
  83         unsigned short          port;
  84         signed short            fastreuse;
  85         int                     num_owners;
  86         struct hlist_node       node;
  87         struct hlist_head       owners;
  88 };
  89
  90 static inline struct net *ib_net(struct inet_bind_bucket *ib)
  91 {
  92         return read_pnet(&ib->ib_net);
  93 }
  94
  95 #define inet_bind_bucket_for_each(tb, pos, head) \
  96         hlist_for_each_entry(tb, pos, head, node)
  97
  98 struct inet_bind_hashbucket {
  99         spinlock_t              lock;
 100         struct hlist_head       chain;
 101 };
 102
 103 /*
 104  * Sockets can be hashed in established or listening table
 105  * We must use different 'nulls' end-of-chain value for listening
 106  * hash table, or we might find a socket that was closed and
 107  * reallocated/inserted into established hash table
 108  */
 109 #define LISTENING_NULLS_BASE (1U << 29)
 110 struct inet_listen_hashbucket {
 111         spinlock_t              lock;
 112         struct hlist_nulls_head head;
 113 };
 114
 115 /* This is for listening sockets, thus all sockets which possess wildcards. */
 116 #define INET_LHTABLE_SIZE       32      /* Yes, really, this is all you need. */
 117
 118 struct inet_hashinfo {
 119         /* This is for sockets with full identity only.  Sockets here will
 120          * always be without wildcards and will have the following invariant:
 121          *
 122          *          TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
 123          *
 124          * TIME_WAIT sockets use a separate chain (twchain).
 125          */
 126         struct inet_ehash_bucket        *ehash;
 127         spinlock_t                      *ehash_locks;
 128         unsigned int                    ehash_mask;
 129         unsigned int                    ehash_locks_mask;
 130
 131         /* Ok, let's try this, I give up, we do need a local binding
 132          * TCP hash as well as the others for fast bind/connect.
 133          */
 134         struct inet_bind_hashbucket     *bhash;
 135
 136         unsigned int                    bhash_size;
 137         /* 4 bytes hole on 64 bit */
 138
 139         struct kmem_cache               *bind_bucket_cachep;
 140
 141         /* All the above members are written once at bootup and
 142          * never written again _or_ are predominantly read-access.
 143          *
 144          * Now align to a new cache line as all the following members
 145          * might be often dirty.
 146          */
 147         /* All sockets in TCP_LISTEN state will be in here.  This is the only
 148          * table where wildcard'd TCP sockets can exist.  Hash function here
 149          * is just local port number.
 150          */
 151         struct inet_listen_hashbucket   listening_hash[INET_LHTABLE_SIZE]
 152                                         ____cacheline_aligned_in_smp;
 153
 154         atomic_t                        bsockets;
 155 };
 156
 157 static inline struct inet_ehash_bucket *inet_ehash_bucket(
 158         struct inet_hashinfo *hashinfo,
 159         unsigned int hash)
 160 {
 161         return &hashinfo->ehash[hash & hashinfo->ehash_mask];
 162 }
 163
 164 static inline spinlock_t *inet_ehash_lockp(
 165         struct inet_hashinfo *hashinfo,
 166         unsigned int hash)
 167 {
 168         return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
 169 }
 170
 171 static inline int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo)
 172 {
 173         unsigned int i, size = 256;
 174 #if defined(CONFIG_PROVE_LOCKING)
 175         unsigned int nr_pcpus = 2;
 176 #else
 177         unsigned int nr_pcpus = num_possible_cpus();
 178 #endif
 179         if (nr_pcpus >= 4)
 180                 size = 512;
 181         if (nr_pcpus >= 8)
 182                 size = 1024;
 183         if (nr_pcpus >= 16)
 184                 size = 2048;
 185         if (nr_pcpus >= 32)
 186                 size = 4096;
 187         if (sizeof(spinlock_t) != 0) {
 188 #ifdef CONFIG_NUMA
 189                 if (size * sizeof(spinlock_t) > PAGE_SIZE)
 190                         hashinfo->ehash_locks = vmalloc(size * sizeof(spinlock_t));
 191                 else
 192 #endif
 193                 hashinfo->ehash_locks = kmalloc(size * sizeof(spinlock_t),
 194                                                 GFP_KERNEL);
 195                 if (!hashinfo->ehash_locks)
 196                         return ENOMEM;
 197                 for (i = 0; i < size; i++)
 198                         spin_lock_init(&hashinfo->ehash_locks[i]);
 199         }
 200         hashinfo->ehash_locks_mask = size - 1;
 201         return 0;
 202 }
 203
 204 static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
 205 {
 206         if (hashinfo->ehash_locks) {
 207 #ifdef CONFIG_NUMA
 208                 unsigned int size = (hashinfo->ehash_locks_mask + 1) *
 209                                                         sizeof(spinlock_t);
 210                 if (size > PAGE_SIZE)
 211                         vfree(hashinfo->ehash_locks);
 212                 else
 213 #endif
 214                 kfree(hashinfo->ehash_locks);
 215                 hashinfo->ehash_locks = NULL;
 216         }
 217 }
 218
 219 extern struct inet_bind_bucket *
 220                     inet_bind_bucket_create(struct kmem_cache *cachep,
 221                                             struct net *net,
 222                                             struct inet_bind_hashbucket *head,
 223                                             const unsigned short snum);
 224 extern void inet_bind_bucket_destroy(struct kmem_cache *cachep,
 225                                      struct inet_bind_bucket *tb);
 226
 227 static inline int inet_bhashfn(struct net *net,
 228                 const __u16 lport, const int bhash_size)
 229 {
 230         return (lport + net_hash_mix(net)) & (bhash_size - 1);
 231 }
 232
 233 extern void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
 234                            const unsigned short snum);
 235
 236 /* These can have wildcards, don't try too hard. */
 237 static inline int inet_lhashfn(struct net *net, const unsigned short num)
 238 {
 239         return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1);
 240 }
 241
 242 static inline int inet_sk_listen_hashfn(const struct sock *sk)
 243 {
 244         return inet_lhashfn(sock_net(sk), inet_sk(sk)->inet_num);
 245 }
 246
 247 /* Caller must disable local BH processing. */
 248 extern int __inet_inherit_port(struct sock *sk, struct sock *child);
 249
 250 extern void inet_put_port(struct sock *sk);
 251
 252 void inet_hashinfo_init(struct inet_hashinfo *h);
 253
 254 extern int __inet_hash_nolisten(struct sock *sk, struct inet_timewait_sock *tw);
 255 extern void inet_hash(struct sock *sk);
 256 extern void inet_unhash(struct sock *sk);
 257
 258 extern struct sock *__inet_lookup_listener(struct net *net,
 259                                            struct inet_hashinfo *hashinfo,
 260                                            const __be32 daddr,
 261                                            const unsigned short hnum,
 262                                            const int dif);
 263
 264 static inline struct sock *inet_lookup_listener(struct net *net,
 265                 struct inet_hashinfo *hashinfo,
 266                 __be32 daddr, __be16 dport, int dif)
 267 {
 268         return __inet_lookup_listener(net, hashinfo, daddr, ntohs(dport), dif);
 269 }
 270
 271 /* Socket demux engine toys. */
 272 /* What happens here is ugly; there's a pair of adjacent fields in
 273    struct inet_sock; __be16 dport followed by __u16 num.  We want to
 274    search by pair, so we combine the keys into a single 32bit value
 275    and compare with 32bit value read from &...->dport.  Let's at least
 276    make sure that it's not mixed with anything else...
 277    On 64bit targets we combine comparisons with pair of adjacent __be32
 278    fields in the same way.
 279 */
 280 typedef __u32 __bitwise __portpair;
 281 #ifdef __BIG_ENDIAN
 282 #define INET_COMBINED_PORTS(__sport, __dport) \
 283         ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
 284 #else /* __LITTLE_ENDIAN */
 285 #define INET_COMBINED_PORTS(__sport, __dport) \
 286         ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
 287 #endif
 288
 289 #if (BITS_PER_LONG == 64)
 290 typedef __u64 __bitwise __addrpair;
 291 #ifdef __BIG_ENDIAN
 292 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
 293         const __addrpair __name = (__force __addrpair) ( \
 294                                    (((__force __u64)(__be32)(__saddr)) << 32) | \
 295                                    ((__force __u64)(__be32)(__daddr)));
 296 #else /* __LITTLE_ENDIAN */
 297 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
 298         const __addrpair __name = (__force __addrpair) ( \
 299                                    (((__force __u64)(__be32)(__daddr)) << 32) | \
 300                                    ((__force __u64)(__be32)(__saddr)));
 301 #endif /* __BIG_ENDIAN */
 302 #define INET_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
 303         (((__sk)->sk_hash == (__hash)) && net_eq(sock_net(__sk), (__net)) &&    \
 304          ((*((__addrpair *)&(inet_sk(__sk)->inet_daddr))) == (__cookie))  &&    \
 305          ((*((__portpair *)&(inet_sk(__sk)->inet_dport))) == (__ports))   &&    \
 306          (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
 307 #define INET_TW_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
 308         (((__sk)->sk_hash == (__hash)) && net_eq(sock_net(__sk), (__net)) &&    \
 309          ((*((__addrpair *)&(inet_twsk(__sk)->tw_daddr))) == (__cookie)) &&     \
 310          ((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) &&      \
 311          (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
 312 #else /* 32-bit arch */
 313 #define INET_ADDR_COOKIE(__name, __saddr, __daddr)
 314 #define INET_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif)     \
 315         (((__sk)->sk_hash == (__hash)) && net_eq(sock_net(__sk), (__net))       &&      \
 316          (inet_sk(__sk)->inet_daddr     == (__saddr))           &&      \
 317          (inet_sk(__sk)->inet_rcv_saddr == (__daddr))           &&      \
 318          ((*((__portpair *)&(inet_sk(__sk)->inet_dport))) == (__ports)) &&      \
 319          (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
 320 #define INET_TW_MATCH(__sk, __net, __hash,__cookie, __saddr, __daddr, __ports, __dif)   \
 321         (((__sk)->sk_hash == (__hash)) && net_eq(sock_net(__sk), (__net))       &&      \
 322          (inet_twsk(__sk)->tw_daddr     == (__saddr))           &&      \
 323          (inet_twsk(__sk)->tw_rcv_saddr == (__daddr))           &&      \
 324          ((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) &&      \
 325          (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
 326 #endif /* 64-bit arch */
 327
 328 /*
 329  * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
 330  * not check it for lookups anymore, thanks Alexey. -DaveM
 331  *
 332  * Local BH must be disabled here.
 333  */
 334 extern struct sock * __inet_lookup_established(struct net *net,
 335                 struct inet_hashinfo *hashinfo,
 336                 const __be32 saddr, const __be16 sport,
 337                 const __be32 daddr, const u16 hnum, const int dif);
 338
 339 static inline struct sock *
 340         inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
 341                                 const __be32 saddr, const __be16 sport,
 342                                 const __be32 daddr, const __be16 dport,
 343                                 const int dif)
 344 {
 345         return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
 346                                          ntohs(dport), dif);
 347 }
 348
 349 static inline struct sock *__inet_lookup(struct net *net,
 350                                          struct inet_hashinfo *hashinfo,
 351                                          const __be32 saddr, const __be16 sport,
 352                                          const __be32 daddr, const __be16 dport,
 353                                          const int dif)
 354 {
 355         u16 hnum = ntohs(dport);
 356         struct sock *sk = __inet_lookup_established(net, hashinfo,
 357                                 saddr, sport, daddr, hnum, dif);
 358
 359         return sk ? : __inet_lookup_listener(net, hashinfo, daddr, hnum, dif);
 360 }
 361
 362 static inline struct sock *inet_lookup(struct net *net,
 363                                        struct inet_hashinfo *hashinfo,
 364                                        const __be32 saddr, const __be16 sport,
 365                                        const __be32 daddr, const __be16 dport,
 366                                        const int dif)
 367 {
 368         struct sock *sk;
 369
 370         local_bh_disable();
 371         sk = __inet_lookup(net, hashinfo, saddr, sport, daddr, dport, dif);
 372         local_bh_enable();
 373
 374         return sk;
 375 }
 376
 377 static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
 378                                              struct sk_buff *skb,
 379                                              const __be16 sport,
 380                                              const __be16 dport)
 381 {
 382         struct sock *sk;
 383         const struct iphdr *iph = ip_hdr(skb);
 384
 385         if (unlikely(sk = skb_steal_sock(skb)))
 386                 return sk;
 387         else
 388                 return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo,
 389                                      iph->saddr, sport,
 390                                      iph->daddr, dport, inet_iif(skb));
 391 }
 392
 393 extern int __inet_hash_connect(struct inet_timewait_death_row *death_row,
 394                 struct sock *sk,
 395                 u32 port_offset,
 396                 int (*check_established)(struct inet_timewait_death_row *,
 397                         struct sock *, __u16, struct inet_timewait_sock **),
 398                 int (*hash)(struct sock *sk, struct inet_timewait_sock *twp));
 399
 400 extern int inet_hash_connect(struct inet_timewait_death_row *death_row,
 401                              struct sock *sk);
 402 #endif /* _INET_HASHTABLES_H */