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
  28 #include <net/inet_connection_sock.h>
  29 #include <net/inet_sock.h>
  30 #include <net/sock.h>
  31 #include <net/route.h>
  32 #include <net/tcp_states.h>
  33 #include <net/netns/hash.h>
  34
  35 #include <linux/atomic.h>
  36 #include <asm/byteorder.h>
  37
  38 /* This is for all connections with a full identity, no wildcards.
  39  * The 'e' prefix stands for Establish, but we really put all sockets
  40  * but LISTEN ones.
  41  */
  42 struct inet_ehash_bucket {
  43         struct hlist_nulls_head chain;
  44 };
  45
  46 /* There are a few simple rules, which allow for local port reuse by
  47  * an application.  In essence:
  48  *
  49  *      1) Sockets bound to different interfaces may share a local port.
  50  *         Failing that, goto test 2.
  51  *      2) If all sockets have sk->sk_reuse set, and none of them are in
  52  *         TCP_LISTEN state, the port may be shared.
  53  *         Failing that, goto test 3.
  54  *      3) If all sockets are bound to a specific inet_sk(sk)->rcv_saddr local
  55  *         address, and none of them are the same, the port may be
  56  *         shared.
  57  *         Failing this, the port cannot be shared.
  58  *
  59  * The interesting point, is test #2.  This is what an FTP server does
  60  * all day.  To optimize this case we use a specific flag bit defined
  61  * below.  As we add sockets to a bind bucket list, we perform a
  62  * check of: (newsk->sk_reuse && (newsk->sk_state != TCP_LISTEN))
  63  * As long as all sockets added to a bind bucket pass this test,
  64  * the flag bit will be set.
  65  * The resulting situation is that tcp_v[46]_verify_bind() can just check
  66  * for this flag bit, if it is set and the socket trying to bind has
  67  * sk->sk_reuse set, we don't even have to walk the owners list at all,
  68  * we return that it is ok to bind this socket to the requested local port.
  69  *
  70  * Sounds like a lot of work, but it is worth it.  In a more naive
  71  * implementation (ie. current FreeBSD etc.) the entire list of ports
  72  * must be walked for each data port opened by an ftp server.  Needless
  73  * to say, this does not scale at all.  With a couple thousand FTP
  74  * users logged onto your box, isn't it nice to know that new data
  75  * ports are created in O(1) time?  I thought so. ;-)   -DaveM
  76  */
  77 struct inet_bind_bucket {
  78         possible_net_t          ib_net;
  79         unsigned short          port;
  80         signed char             fastreuse;
  81         signed char             fastreuseport;
  82         kuid_t                  fastuid;
  83         int                     num_owners;
  84         struct hlist_node       node;
  85         struct hlist_head       owners;
  86 };
  87
  88 static inline struct net *ib_net(struct inet_bind_bucket *ib)
  89 {
  90         return read_pnet(&ib->ib_net);
  91 }
  92
  93 #define inet_bind_bucket_for_each(tb, head) \
  94         hlist_for_each_entry(tb, head, node)
  95
  96 struct inet_bind_hashbucket {
  97         spinlock_t              lock;
  98         struct hlist_head       chain;
  99 };
 100
 101 /*
 102  * Sockets can be hashed in established or listening table
 103  * We must use different 'nulls' end-of-chain value for listening
 104  * hash table, or we might find a socket that was closed and
 105  * reallocated/inserted into established hash table
 106  */
 107 #define LISTENING_NULLS_BASE (1U << 29)
 108 struct inet_listen_hashbucket {
 109         spinlock_t              lock;
 110         struct hlist_nulls_head head;
 111 };
 112
 113 /* This is for listening sockets, thus all sockets which possess wildcards. */
 114 #define INET_LHTABLE_SIZE       32      /* Yes, really, this is all you need. */
 115
 116 struct inet_hashinfo {
 117         /* This is for sockets with full identity only.  Sockets here will
 118          * always be without wildcards and will have the following invariant:
 119          *
 120          *          TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
 121          *
 122          */
 123         struct inet_ehash_bucket        *ehash;
 124         spinlock_t                      *ehash_locks;
 125         unsigned int                    ehash_mask;
 126         unsigned int                    ehash_locks_mask;
 127
 128         /* Ok, let's try this, I give up, we do need a local binding
 129          * TCP hash as well as the others for fast bind/connect.
 130          */
 131         struct inet_bind_hashbucket     *bhash;
 132
 133         unsigned int                    bhash_size;
 134         /* 4 bytes hole on 64 bit */
 135
 136         struct kmem_cache               *bind_bucket_cachep;
 137
 138         /* All the above members are written once at bootup and
 139          * never written again _or_ are predominantly read-access.
 140          *
 141          * Now align to a new cache line as all the following members
 142          * might be often dirty.
 143          */
 144         /* All sockets in TCP_LISTEN state will be in here.  This is the only
 145          * table where wildcard'd TCP sockets can exist.  Hash function here
 146          * is just local port number.
 147          */
 148         struct inet_listen_hashbucket   listening_hash[INET_LHTABLE_SIZE]
 149                                         ____cacheline_aligned_in_smp;
 150 };
 151
 152 static inline struct inet_ehash_bucket *inet_ehash_bucket(
 153         struct inet_hashinfo *hashinfo,
 154         unsigned int hash)
 155 {
 156         return &hashinfo->ehash[hash & hashinfo->ehash_mask];
 157 }
 158
 159 static inline spinlock_t *inet_ehash_lockp(
 160         struct inet_hashinfo *hashinfo,
 161         unsigned int hash)
 162 {
 163         return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
 164 }
 165
 166 int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo);
 167
 168 static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
 169 {
 170         kvfree(hashinfo->ehash_locks);
 171         hashinfo->ehash_locks = NULL;
 172 }
 173
 174 struct inet_bind_bucket *
 175 inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net,
 176                         struct inet_bind_hashbucket *head,
 177                         const unsigned short snum);
 178 void inet_bind_bucket_destroy(struct kmem_cache *cachep,
 179                               struct inet_bind_bucket *tb);
 180
 181 static inline u32 inet_bhashfn(const struct net *net, const __u16 lport,
 182                                const u32 bhash_size)
 183 {
 184         return (lport + net_hash_mix(net)) & (bhash_size - 1);
 185 }
 186
 187 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
 188                     const unsigned short snum);
 189
 190 /* These can have wildcards, don't try too hard. */
 191 static inline u32 inet_lhashfn(const struct net *net, const unsigned short num)
 192 {
 193         return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1);
 194 }
 195
 196 static inline int inet_sk_listen_hashfn(const struct sock *sk)
 197 {
 198         return inet_lhashfn(sock_net(sk), inet_sk(sk)->inet_num);
 199 }
 200
 201 /* Caller must disable local BH processing. */
 202 int __inet_inherit_port(const struct sock *sk, struct sock *child);
 203
 204 void inet_put_port(struct sock *sk);
 205
 206 void inet_hashinfo_init(struct inet_hashinfo *h);
 207
 208 bool inet_ehash_insert(struct sock *sk, struct sock *osk);
 209 bool inet_ehash_nolisten(struct sock *sk, struct sock *osk);
 210 int __inet_hash(struct sock *sk, struct sock *osk,
 211                 int (*saddr_same)(const struct sock *sk1,
 212                                   const struct sock *sk2,
 213                                   bool match_wildcard));
 214 int inet_hash(struct sock *sk);
 215 void inet_unhash(struct sock *sk);
 216
 217 struct sock *__inet_lookup_listener(struct net *net,
 218                                     struct inet_hashinfo *hashinfo,
 219                                     struct sk_buff *skb, int doff,
 220                                     const __be32 saddr, const __be16 sport,
 221                                     const __be32 daddr,
 222                                     const unsigned short hnum,
 223                                     const int dif);
 224
 225 static inline struct sock *inet_lookup_listener(struct net *net,
 226                 struct inet_hashinfo *hashinfo,
 227                 struct sk_buff *skb, int doff,
 228                 __be32 saddr, __be16 sport,
 229                 __be32 daddr, __be16 dport, int dif)
 230 {
 231         return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, sport,
 232                                       daddr, ntohs(dport), dif);
 233 }
 234
 235 /* Socket demux engine toys. */
 236 /* What happens here is ugly; there's a pair of adjacent fields in
 237    struct inet_sock; __be16 dport followed by __u16 num.  We want to
 238    search by pair, so we combine the keys into a single 32bit value
 239    and compare with 32bit value read from &...->dport.  Let's at least
 240    make sure that it's not mixed with anything else...
 241    On 64bit targets we combine comparisons with pair of adjacent __be32
 242    fields in the same way.
 243 */
 244 #ifdef __BIG_ENDIAN
 245 #define INET_COMBINED_PORTS(__sport, __dport) \
 246         ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
 247 #else /* __LITTLE_ENDIAN */
 248 #define INET_COMBINED_PORTS(__sport, __dport) \
 249         ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
 250 #endif
 251
 252 #if (BITS_PER_LONG == 64)
 253 #ifdef __BIG_ENDIAN
 254 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
 255         const __addrpair __name = (__force __addrpair) ( \
 256                                    (((__force __u64)(__be32)(__saddr)) << 32) | \
 257                                    ((__force __u64)(__be32)(__daddr)))
 258 #else /* __LITTLE_ENDIAN */
 259 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
 260         const __addrpair __name = (__force __addrpair) ( \
 261                                    (((__force __u64)(__be32)(__daddr)) << 32) | \
 262                                    ((__force __u64)(__be32)(__saddr)))
 263 #endif /* __BIG_ENDIAN */
 264 #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif)     \
 265         (((__sk)->sk_portpair == (__ports))                     &&      \
 266          ((__sk)->sk_addrpair == (__cookie))                    &&      \
 267          (!(__sk)->sk_bound_dev_if      ||                              \
 268            ((__sk)->sk_bound_dev_if == (__dif)))                &&      \
 269          net_eq(sock_net(__sk), (__net)))
 270 #else /* 32-bit arch */
 271 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
 272         const int __name __deprecated __attribute__((unused))
 273
 274 #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif) \
 275         (((__sk)->sk_portpair == (__ports))             &&              \
 276          ((__sk)->sk_daddr      == (__saddr))           &&              \
 277          ((__sk)->sk_rcv_saddr  == (__daddr))           &&              \
 278          (!(__sk)->sk_bound_dev_if      ||                              \
 279            ((__sk)->sk_bound_dev_if == (__dif)))        &&              \
 280          net_eq(sock_net(__sk), (__net)))
 281 #endif /* 64-bit arch */
 282
 283 /*
 284  * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
 285  * not check it for lookups anymore, thanks Alexey. -DaveM
 286  *
 287  * Local BH must be disabled here.
 288  */
 289 struct sock *__inet_lookup_established(struct net *net,
 290                                        struct inet_hashinfo *hashinfo,
 291                                        const __be32 saddr, const __be16 sport,
 292                                        const __be32 daddr, const u16 hnum,
 293                                        const int dif);
 294
 295 static inline struct sock *
 296         inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
 297                                 const __be32 saddr, const __be16 sport,
 298                                 const __be32 daddr, const __be16 dport,
 299                                 const int dif)
 300 {
 301         return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
 302                                          ntohs(dport), dif);
 303 }
 304
 305 static inline struct sock *__inet_lookup(struct net *net,
 306                                          struct inet_hashinfo *hashinfo,
 307                                          struct sk_buff *skb, int doff,
 308                                          const __be32 saddr, const __be16 sport,
 309                                          const __be32 daddr, const __be16 dport,
 310                                          const int dif)
 311 {
 312         u16 hnum = ntohs(dport);
 313         struct sock *sk = __inet_lookup_established(net, hashinfo,
 314                                 saddr, sport, daddr, hnum, dif);
 315
 316         return sk ? : __inet_lookup_listener(net, hashinfo, skb, doff, saddr,
 317                                              sport, daddr, hnum, dif);
 318 }
 319
 320 static inline struct sock *inet_lookup(struct net *net,
 321                                        struct inet_hashinfo *hashinfo,
 322                                        struct sk_buff *skb, int doff,
 323                                        const __be32 saddr, const __be16 sport,
 324                                        const __be32 daddr, const __be16 dport,
 325                                        const int dif)
 326 {
 327         struct sock *sk;
 328
 329         local_bh_disable();
 330         sk = __inet_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
 331                            dport, dif);
 332         local_bh_enable();
 333
 334         return sk;
 335 }
 336
 337 static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
 338                                              struct sk_buff *skb,
 339                                              int doff,
 340                                              const __be16 sport,
 341                                              const __be16 dport)
 342 {
 343         struct sock *sk = skb_steal_sock(skb);
 344         const struct iphdr *iph = ip_hdr(skb);
 345
 346         if (sk)
 347                 return sk;
 348         else
 349                 return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo, skb,
 350                                      doff, iph->saddr, sport,
 351                                      iph->daddr, dport, inet_iif(skb));
 352 }
 353
 354 u32 sk_ehashfn(const struct sock *sk);
 355 u32 inet6_ehashfn(const struct net *net,
 356                   const struct in6_addr *laddr, const u16 lport,
 357                   const struct in6_addr *faddr, const __be16 fport);
 358
 359 static inline void sk_daddr_set(struct sock *sk, __be32 addr)
 360 {
 361         sk->sk_daddr = addr; /* alias of inet_daddr */
 362 #if IS_ENABLED(CONFIG_IPV6)
 363         ipv6_addr_set_v4mapped(addr, &sk->sk_v6_daddr);
 364 #endif
 365 }
 366
 367 static inline void sk_rcv_saddr_set(struct sock *sk, __be32 addr)
 368 {
 369         sk->sk_rcv_saddr = addr; /* alias of inet_rcv_saddr */
 370 #if IS_ENABLED(CONFIG_IPV6)
 371         ipv6_addr_set_v4mapped(addr, &sk->sk_v6_rcv_saddr);
 372 #endif
 373 }
 374
 375 int __inet_hash_connect(struct inet_timewait_death_row *death_row,
 376                         struct sock *sk, u32 port_offset,
 377                         int (*check_established)(struct inet_timewait_death_row *,
 378                                                  struct sock *, __u16,
 379                                                  struct inet_timewait_sock **));
 380
 381 int inet_hash_connect(struct inet_timewait_death_row *death_row,
 382                       struct sock *sk);
 383 #endif /* _INET_HASHTABLES_H */