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  */
 104 struct inet_listen_hashbucket {
 105         spinlock_t              lock;
 106         struct hlist_head       head;
 107 };
 108
 109 /* This is for listening sockets, thus all sockets which possess wildcards. */
 110 #define INET_LHTABLE_SIZE       32      /* Yes, really, this is all you need. */
 111
 112 struct inet_hashinfo {
 113         /* This is for sockets with full identity only.  Sockets here will
 114          * always be without wildcards and will have the following invariant:
 115          *
 116          *          TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
 117          *
 118          */
 119         struct inet_ehash_bucket        *ehash;
 120         spinlock_t                      *ehash_locks;
 121         unsigned int                    ehash_mask;
 122         unsigned int                    ehash_locks_mask;
 123
 124         /* Ok, let's try this, I give up, we do need a local binding
 125          * TCP hash as well as the others for fast bind/connect.
 126          */
 127         struct inet_bind_hashbucket     *bhash;
 128
 129         unsigned int                    bhash_size;
 130         /* 4 bytes hole on 64 bit */
 131
 132         struct kmem_cache               *bind_bucket_cachep;
 133
 134         /* All the above members are written once at bootup and
 135          * never written again _or_ are predominantly read-access.
 136          *
 137          * Now align to a new cache line as all the following members
 138          * might be often dirty.
 139          */
 140         /* All sockets in TCP_LISTEN state will be in here.  This is the only
 141          * table where wildcard'd TCP sockets can exist.  Hash function here
 142          * is just local port number.
 143          */
 144         struct inet_listen_hashbucket   listening_hash[INET_LHTABLE_SIZE]
 145                                         ____cacheline_aligned_in_smp;
 146 };
 147
 148 static inline struct inet_ehash_bucket *inet_ehash_bucket(
 149         struct inet_hashinfo *hashinfo,
 150         unsigned int hash)
 151 {
 152         return &hashinfo->ehash[hash & hashinfo->ehash_mask];
 153 }
 154
 155 static inline spinlock_t *inet_ehash_lockp(
 156         struct inet_hashinfo *hashinfo,
 157         unsigned int hash)
 158 {
 159         return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
 160 }
 161
 162 int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo);
 163
 164 static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
 165 {
 166         kvfree(hashinfo->ehash_locks);
 167         hashinfo->ehash_locks = NULL;
 168 }
 169
 170 struct inet_bind_bucket *
 171 inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net,
 172                         struct inet_bind_hashbucket *head,
 173                         const unsigned short snum);
 174 void inet_bind_bucket_destroy(struct kmem_cache *cachep,
 175                               struct inet_bind_bucket *tb);
 176
 177 static inline u32 inet_bhashfn(const struct net *net, const __u16 lport,
 178                                const u32 bhash_size)
 179 {
 180         return (lport + net_hash_mix(net)) & (bhash_size - 1);
 181 }
 182
 183 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
 184                     const unsigned short snum);
 185
 186 /* These can have wildcards, don't try too hard. */
 187 static inline u32 inet_lhashfn(const struct net *net, const unsigned short num)
 188 {
 189         return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1);
 190 }
 191
 192 static inline int inet_sk_listen_hashfn(const struct sock *sk)
 193 {
 194         return inet_lhashfn(sock_net(sk), inet_sk(sk)->inet_num);
 195 }
 196
 197 /* Caller must disable local BH processing. */
 198 int __inet_inherit_port(const struct sock *sk, struct sock *child);
 199
 200 void inet_put_port(struct sock *sk);
 201
 202 void inet_hashinfo_init(struct inet_hashinfo *h);
 203
 204 bool inet_ehash_insert(struct sock *sk, struct sock *osk);
 205 bool inet_ehash_nolisten(struct sock *sk, struct sock *osk);
 206 int __inet_hash(struct sock *sk, struct sock *osk,
 207                 int (*saddr_same)(const struct sock *sk1,
 208                                   const struct sock *sk2,
 209                                   bool match_wildcard));
 210 int inet_hash(struct sock *sk);
 211 void inet_unhash(struct sock *sk);
 212
 213 struct sock *__inet_lookup_listener(struct net *net,
 214                                     struct inet_hashinfo *hashinfo,
 215                                     struct sk_buff *skb, int doff,
 216                                     const __be32 saddr, const __be16 sport,
 217                                     const __be32 daddr,
 218                                     const unsigned short hnum,
 219                                     const int dif);
 220
 221 static inline struct sock *inet_lookup_listener(struct net *net,
 222                 struct inet_hashinfo *hashinfo,
 223                 struct sk_buff *skb, int doff,
 224                 __be32 saddr, __be16 sport,
 225                 __be32 daddr, __be16 dport, int dif)
 226 {
 227         return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, sport,
 228                                       daddr, ntohs(dport), dif);
 229 }
 230
 231 /* Socket demux engine toys. */
 232 /* What happens here is ugly; there's a pair of adjacent fields in
 233    struct inet_sock; __be16 dport followed by __u16 num.  We want to
 234    search by pair, so we combine the keys into a single 32bit value
 235    and compare with 32bit value read from &...->dport.  Let's at least
 236    make sure that it's not mixed with anything else...
 237    On 64bit targets we combine comparisons with pair of adjacent __be32
 238    fields in the same way.
 239 */
 240 #ifdef __BIG_ENDIAN
 241 #define INET_COMBINED_PORTS(__sport, __dport) \
 242         ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
 243 #else /* __LITTLE_ENDIAN */
 244 #define INET_COMBINED_PORTS(__sport, __dport) \
 245         ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
 246 #endif
 247
 248 #if (BITS_PER_LONG == 64)
 249 #ifdef __BIG_ENDIAN
 250 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
 251         const __addrpair __name = (__force __addrpair) ( \
 252                                    (((__force __u64)(__be32)(__saddr)) << 32) | \
 253                                    ((__force __u64)(__be32)(__daddr)))
 254 #else /* __LITTLE_ENDIAN */
 255 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
 256         const __addrpair __name = (__force __addrpair) ( \
 257                                    (((__force __u64)(__be32)(__daddr)) << 32) | \
 258                                    ((__force __u64)(__be32)(__saddr)))
 259 #endif /* __BIG_ENDIAN */
 260 #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif)     \
 261         (((__sk)->sk_portpair == (__ports))                     &&      \
 262          ((__sk)->sk_addrpair == (__cookie))                    &&      \
 263          (!(__sk)->sk_bound_dev_if      ||                              \
 264            ((__sk)->sk_bound_dev_if == (__dif)))                &&      \
 265          net_eq(sock_net(__sk), (__net)))
 266 #else /* 32-bit arch */
 267 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
 268         const int __name __deprecated __attribute__((unused))
 269
 270 #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif) \
 271         (((__sk)->sk_portpair == (__ports))             &&              \
 272          ((__sk)->sk_daddr      == (__saddr))           &&              \
 273          ((__sk)->sk_rcv_saddr  == (__daddr))           &&              \
 274          (!(__sk)->sk_bound_dev_if      ||                              \
 275            ((__sk)->sk_bound_dev_if == (__dif)))        &&              \
 276          net_eq(sock_net(__sk), (__net)))
 277 #endif /* 64-bit arch */
 278
 279 /* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
 280  * not check it for lookups anymore, thanks Alexey. -DaveM
 281  */
 282 struct sock *__inet_lookup_established(struct net *net,
 283                                        struct inet_hashinfo *hashinfo,
 284                                        const __be32 saddr, const __be16 sport,
 285                                        const __be32 daddr, const u16 hnum,
 286                                        const int dif);
 287
 288 static inline struct sock *
 289         inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
 290                                 const __be32 saddr, const __be16 sport,
 291                                 const __be32 daddr, const __be16 dport,
 292                                 const int dif)
 293 {
 294         return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
 295                                          ntohs(dport), dif);
 296 }
 297
 298 static inline struct sock *__inet_lookup(struct net *net,
 299                                          struct inet_hashinfo *hashinfo,
 300                                          struct sk_buff *skb, int doff,
 301                                          const __be32 saddr, const __be16 sport,
 302                                          const __be32 daddr, const __be16 dport,
 303                                          const int dif,
 304                                          bool *refcounted)
 305 {
 306         u16 hnum = ntohs(dport);
 307         struct sock *sk;
 308
 309         sk = __inet_lookup_established(net, hashinfo, saddr, sport,
 310                                        daddr, hnum, dif);
 311         *refcounted = true;
 312         if (sk)
 313                 return sk;
 314         *refcounted = false;
 315         return __inet_lookup_listener(net, hashinfo, skb, doff, saddr,
 316                                       sport, daddr, hnum, dif);
 317 }
 318
 319 static inline struct sock *inet_lookup(struct net *net,
 320                                        struct inet_hashinfo *hashinfo,
 321                                        struct sk_buff *skb, int doff,
 322                                        const __be32 saddr, const __be16 sport,
 323                                        const __be32 daddr, const __be16 dport,
 324                                        const int dif)
 325 {
 326         struct sock *sk;
 327         bool refcounted;
 328
 329         sk = __inet_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
 330                            dport, dif, &refcounted);
 331
 332         if (sk && !refcounted && !atomic_inc_not_zero(&sk->sk_refcnt))
 333                 sk = NULL;
 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                                              bool *refcounted)
 343 {
 344         struct sock *sk = skb_steal_sock(skb);
 345         const struct iphdr *iph = ip_hdr(skb);
 346
 347         *refcounted = true;
 348         if (sk)
 349                 return sk;
 350
 351         return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo, skb,
 352                              doff, iph->saddr, sport,
 353                              iph->daddr, dport, inet_iif(skb),
 354                              refcounted);
 355 }
 356
 357 u32 sk_ehashfn(const struct sock *sk);
 358 u32 inet6_ehashfn(const struct net *net,
 359                   const struct in6_addr *laddr, const u16 lport,
 360                   const struct in6_addr *faddr, const __be16 fport);
 361
 362 static inline void sk_daddr_set(struct sock *sk, __be32 addr)
 363 {
 364         sk->sk_daddr = addr; /* alias of inet_daddr */
 365 #if IS_ENABLED(CONFIG_IPV6)
 366         ipv6_addr_set_v4mapped(addr, &sk->sk_v6_daddr);
 367 #endif
 368 }
 369
 370 static inline void sk_rcv_saddr_set(struct sock *sk, __be32 addr)
 371 {
 372         sk->sk_rcv_saddr = addr; /* alias of inet_rcv_saddr */
 373 #if IS_ENABLED(CONFIG_IPV6)
 374         ipv6_addr_set_v4mapped(addr, &sk->sk_v6_rcv_saddr);
 375 #endif
 376 }
 377
 378 int __inet_hash_connect(struct inet_timewait_death_row *death_row,
 379                         struct sock *sk, u32 port_offset,
 380                         int (*check_established)(struct inet_timewait_death_row *,
 381                                                  struct sock *, __u16,
 382                                                  struct inet_timewait_sock **));
 383
 384 int inet_hash_connect(struct inet_timewait_death_row *death_row,
 385                       struct sock *sk);
 386 #endif /* _INET_HASHTABLES_H */