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 char             fastreuse;
  85         signed char             fastreuseport;
  86         kuid_t                  fastuid;
  87         int                     num_owners;
  88         struct hlist_node       node;
  89         struct hlist_head       owners;
  90 };
  91
  92 static inline struct net *ib_net(struct inet_bind_bucket *ib)
  93 {
  94         return read_pnet(&ib->ib_net);
  95 }
  96
  97 #define inet_bind_bucket_for_each(tb, head) \
  98         hlist_for_each_entry(tb, head, node)
  99
 100 struct inet_bind_hashbucket {
 101         spinlock_t              lock;
 102         struct hlist_head       chain;
 103 };
 104
 105 /*
 106  * Sockets can be hashed in established or listening table
 107  * We must use different 'nulls' end-of-chain value for listening
 108  * hash table, or we might find a socket that was closed and
 109  * reallocated/inserted into established hash table
 110  */
 111 #define LISTENING_NULLS_BASE (1U << 29)
 112 struct inet_listen_hashbucket {
 113         spinlock_t              lock;
 114         struct hlist_nulls_head head;
 115 };
 116
 117 /* This is for listening sockets, thus all sockets which possess wildcards. */
 118 #define INET_LHTABLE_SIZE       32      /* Yes, really, this is all you need. */
 119
 120 struct inet_hashinfo {
 121         /* This is for sockets with full identity only.  Sockets here will
 122          * always be without wildcards and will have the following invariant:
 123          *
 124          *          TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
 125          *
 126          * TIME_WAIT sockets use a separate chain (twchain).
 127          */
 128         struct inet_ehash_bucket        *ehash;
 129         spinlock_t                      *ehash_locks;
 130         unsigned int                    ehash_mask;
 131         unsigned int                    ehash_locks_mask;
 132
 133         /* Ok, let's try this, I give up, we do need a local binding
 134          * TCP hash as well as the others for fast bind/connect.
 135          */
 136         struct inet_bind_hashbucket     *bhash;
 137
 138         unsigned int                    bhash_size;
 139         /* 4 bytes hole on 64 bit */
 140
 141         struct kmem_cache               *bind_bucket_cachep;
 142
 143         /* All the above members are written once at bootup and
 144          * never written again _or_ are predominantly read-access.
 145          *
 146          * Now align to a new cache line as all the following members
 147          * might be often dirty.
 148          */
 149         /* All sockets in TCP_LISTEN state will be in here.  This is the only
 150          * table where wildcard'd TCP sockets can exist.  Hash function here
 151          * is just local port number.
 152          */
 153         struct inet_listen_hashbucket   listening_hash[INET_LHTABLE_SIZE]
 154                                         ____cacheline_aligned_in_smp;
 155
 156         atomic_t                        bsockets;
 157 };
 158
 159 static inline struct inet_ehash_bucket *inet_ehash_bucket(
 160         struct inet_hashinfo *hashinfo,
 161         unsigned int hash)
 162 {
 163         return &hashinfo->ehash[hash & hashinfo->ehash_mask];
 164 }
 165
 166 static inline spinlock_t *inet_ehash_lockp(
 167         struct inet_hashinfo *hashinfo,
 168         unsigned int hash)
 169 {
 170         return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
 171 }
 172
 173 static inline int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo)
 174 {
 175         unsigned int i, size = 256;
 176 #if defined(CONFIG_PROVE_LOCKING)
 177         unsigned int nr_pcpus = 2;
 178 #else
 179         unsigned int nr_pcpus = num_possible_cpus();
 180 #endif
 181         if (nr_pcpus >= 4)
 182                 size = 512;
 183         if (nr_pcpus >= 8)
 184                 size = 1024;
 185         if (nr_pcpus >= 16)
 186                 size = 2048;
 187         if (nr_pcpus >= 32)
 188                 size = 4096;
 189         if (sizeof(spinlock_t) != 0) {
 190 #ifdef CONFIG_NUMA
 191                 if (size * sizeof(spinlock_t) > PAGE_SIZE)
 192                         hashinfo->ehash_locks = vmalloc(size * sizeof(spinlock_t));
 193                 else
 194 #endif
 195                 hashinfo->ehash_locks = kmalloc(size * sizeof(spinlock_t),
 196                                                 GFP_KERNEL);
 197                 if (!hashinfo->ehash_locks)
 198                         return ENOMEM;
 199                 for (i = 0; i < size; i++)
 200                         spin_lock_init(&hashinfo->ehash_locks[i]);
 201         }
 202         hashinfo->ehash_locks_mask = size - 1;
 203         return 0;
 204 }
 205
 206 static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
 207 {
 208         if (hashinfo->ehash_locks) {
 209 #ifdef CONFIG_NUMA
 210                 unsigned int size = (hashinfo->ehash_locks_mask + 1) *
 211                                                         sizeof(spinlock_t);
 212                 if (size > PAGE_SIZE)
 213                         vfree(hashinfo->ehash_locks);
 214                 else
 215 #endif
 216                 kfree(hashinfo->ehash_locks);
 217                 hashinfo->ehash_locks = NULL;
 218         }
 219 }
 220
 221 extern struct inet_bind_bucket *
 222                     inet_bind_bucket_create(struct kmem_cache *cachep,
 223                                             struct net *net,
 224                                             struct inet_bind_hashbucket *head,
 225                                             const unsigned short snum);
 226 extern void inet_bind_bucket_destroy(struct kmem_cache *cachep,
 227                                      struct inet_bind_bucket *tb);
 228
 229 static inline int inet_bhashfn(struct net *net,
 230                 const __u16 lport, const int bhash_size)
 231 {
 232         return (lport + net_hash_mix(net)) & (bhash_size - 1);
 233 }
 234
 235 extern void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
 236                            const unsigned short snum);
 237
 238 /* These can have wildcards, don't try too hard. */
 239 static inline int inet_lhashfn(struct net *net, const unsigned short num)
 240 {
 241         return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1);
 242 }
 243
 244 static inline int inet_sk_listen_hashfn(const struct sock *sk)
 245 {
 246         return inet_lhashfn(sock_net(sk), inet_sk(sk)->inet_num);
 247 }
 248
 249 /* Caller must disable local BH processing. */
 250 extern int __inet_inherit_port(struct sock *sk, struct sock *child);
 251
 252 extern void inet_put_port(struct sock *sk);
 253
 254 void inet_hashinfo_init(struct inet_hashinfo *h);
 255
 256 extern int __inet_hash_nolisten(struct sock *sk, struct inet_timewait_sock *tw);
 257 extern void inet_hash(struct sock *sk);
 258 extern void inet_unhash(struct sock *sk);
 259
 260 extern struct sock *__inet_lookup_listener(struct net *net,
 261                                            struct inet_hashinfo *hashinfo,
 262                                            const __be32 saddr,
 263                                            const __be16 sport,
 264                                            const __be32 daddr,
 265                                            const unsigned short hnum,
 266                                            const int dif);
 267
 268 static inline struct sock *inet_lookup_listener(struct net *net,
 269                 struct inet_hashinfo *hashinfo,
 270                 __be32 saddr, __be16 sport,
 271                 __be32 daddr, __be16 dport, int dif)
 272 {
 273         return __inet_lookup_listener(net, hashinfo, saddr, sport,
 274                                       daddr, ntohs(dport), dif);
 275 }
 276
 277 /* Socket demux engine toys. */
 278 /* What happens here is ugly; there's a pair of adjacent fields in
 279    struct inet_sock; __be16 dport followed by __u16 num.  We want to
 280    search by pair, so we combine the keys into a single 32bit value
 281    and compare with 32bit value read from &...->dport.  Let's at least
 282    make sure that it's not mixed with anything else...
 283    On 64bit targets we combine comparisons with pair of adjacent __be32
 284    fields in the same way.
 285 */
 286 #ifdef __BIG_ENDIAN
 287 #define INET_COMBINED_PORTS(__sport, __dport) \
 288         ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
 289 #else /* __LITTLE_ENDIAN */
 290 #define INET_COMBINED_PORTS(__sport, __dport) \
 291         ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
 292 #endif
 293
 294 #if (BITS_PER_LONG == 64)
 295 #ifdef __BIG_ENDIAN
 296 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
 297         const __addrpair __name = (__force __addrpair) ( \
 298                                    (((__force __u64)(__be32)(__saddr)) << 32) | \
 299                                    ((__force __u64)(__be32)(__daddr)));
 300 #else /* __LITTLE_ENDIAN */
 301 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
 302         const __addrpair __name = (__force __addrpair) ( \
 303                                    (((__force __u64)(__be32)(__daddr)) << 32) | \
 304                                    ((__force __u64)(__be32)(__saddr)));
 305 #endif /* __BIG_ENDIAN */
 306 #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif)     \
 307         ((inet_sk(__sk)->inet_portpair == (__ports))            &&      \
 308          (inet_sk(__sk)->inet_addrpair == (__cookie))           &&      \
 309          (!(__sk)->sk_bound_dev_if      ||                              \
 310            ((__sk)->sk_bound_dev_if == (__dif)))                &&      \
 311          net_eq(sock_net(__sk), (__net)))
 312 #define INET_TW_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif)\
 313         ((inet_twsk(__sk)->tw_portpair == (__ports))    &&              \
 314          (inet_twsk(__sk)->tw_addrpair == (__cookie))   &&              \
 315          (!(__sk)->sk_bound_dev_if      ||                              \
 316            ((__sk)->sk_bound_dev_if == (__dif)))        &&              \
 317          net_eq(sock_net(__sk), (__net)))
 318 #else /* 32-bit arch */
 319 #define INET_ADDR_COOKIE(__name, __saddr, __daddr)
 320 #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif) \
 321         ((inet_sk(__sk)->inet_portpair == (__ports))    &&              \
 322          (inet_sk(__sk)->inet_daddr     == (__saddr))   &&              \
 323          (inet_sk(__sk)->inet_rcv_saddr == (__daddr))   &&              \
 324          (!(__sk)->sk_bound_dev_if      ||                              \
 325            ((__sk)->sk_bound_dev_if == (__dif)))        &&              \
 326          net_eq(sock_net(__sk), (__net)))
 327 #define INET_TW_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif) \
 328         ((inet_twsk(__sk)->tw_portpair == (__ports))    &&              \
 329          (inet_twsk(__sk)->tw_daddr     == (__saddr))   &&              \
 330          (inet_twsk(__sk)->tw_rcv_saddr == (__daddr))   &&              \
 331          (!(__sk)->sk_bound_dev_if      ||                              \
 332            ((__sk)->sk_bound_dev_if == (__dif)))        &&              \
 333          net_eq(sock_net(__sk), (__net)))
 334 #endif /* 64-bit arch */
 335
 336 /*
 337  * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
 338  * not check it for lookups anymore, thanks Alexey. -DaveM
 339  *
 340  * Local BH must be disabled here.
 341  */
 342 extern struct sock * __inet_lookup_established(struct net *net,
 343                 struct inet_hashinfo *hashinfo,
 344                 const __be32 saddr, const __be16 sport,
 345                 const __be32 daddr, const u16 hnum, const int dif);
 346
 347 static inline struct sock *
 348         inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
 349                                 const __be32 saddr, const __be16 sport,
 350                                 const __be32 daddr, const __be16 dport,
 351                                 const int dif)
 352 {
 353         return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
 354                                          ntohs(dport), dif);
 355 }
 356
 357 static inline struct sock *__inet_lookup(struct net *net,
 358                                          struct inet_hashinfo *hashinfo,
 359                                          const __be32 saddr, const __be16 sport,
 360                                          const __be32 daddr, const __be16 dport,
 361                                          const int dif)
 362 {
 363         u16 hnum = ntohs(dport);
 364         struct sock *sk = __inet_lookup_established(net, hashinfo,
 365                                 saddr, sport, daddr, hnum, dif);
 366
 367         return sk ? : __inet_lookup_listener(net, hashinfo, saddr, sport,
 368                                              daddr, hnum, dif);
 369 }
 370
 371 static inline struct sock *inet_lookup(struct net *net,
 372                                        struct inet_hashinfo *hashinfo,
 373                                        const __be32 saddr, const __be16 sport,
 374                                        const __be32 daddr, const __be16 dport,
 375                                        const int dif)
 376 {
 377         struct sock *sk;
 378
 379         local_bh_disable();
 380         sk = __inet_lookup(net, hashinfo, saddr, sport, daddr, dport, dif);
 381         local_bh_enable();
 382
 383         return sk;
 384 }
 385
 386 static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
 387                                              struct sk_buff *skb,
 388                                              const __be16 sport,
 389                                              const __be16 dport)
 390 {
 391         struct sock *sk = skb_steal_sock(skb);
 392         const struct iphdr *iph = ip_hdr(skb);
 393
 394         if (sk)
 395                 return sk;
 396         else
 397                 return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo,
 398                                      iph->saddr, sport,
 399                                      iph->daddr, dport, inet_iif(skb));
 400 }
 401
 402 extern int __inet_hash_connect(struct inet_timewait_death_row *death_row,
 403                 struct sock *sk,
 404                 u32 port_offset,
 405                 int (*check_established)(struct inet_timewait_death_row *,
 406                         struct sock *, __u16, struct inet_timewait_sock **),
 407                 int (*hash)(struct sock *sk, struct inet_timewait_sock *twp));
 408
 409 extern int inet_hash_connect(struct inet_timewait_death_row *death_row,
 410                              struct sock *sk);
 411 #endif /* _INET_HASHTABLES_H */