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/refcount.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 #define FASTREUSEPORT_ANY       1
  78 #define FASTREUSEPORT_STRICT    2
  79
  80 struct inet_bind_bucket {
  81         possible_net_t          ib_net;
  82         unsigned short          port;
  83         signed char             fastreuse;
  84         signed char             fastreuseport;
  85         kuid_t                  fastuid;
  86 #if IS_ENABLED(CONFIG_IPV6)
  87         struct in6_addr         fast_v6_rcv_saddr;
  88 #endif
  89         __be32                  fast_rcv_saddr;
  90         unsigned short          fast_sk_family;
  91         bool                    fast_ipv6_only;
  92         struct hlist_node       node;
  93         struct hlist_head       owners;
  94 };
  95
  96 static inline struct net *ib_net(struct inet_bind_bucket *ib)
  97 {
  98         return read_pnet(&ib->ib_net);
  99 }
 100
 101 #define inet_bind_bucket_for_each(tb, head) \
 102         hlist_for_each_entry(tb, head, node)
 103
 104 struct inet_bind_hashbucket {
 105         spinlock_t              lock;
 106         struct hlist_head       chain;
 107 };
 108
 109 /* Sockets can be hashed in established or listening table.
 110  * We must use different 'nulls' end-of-chain value for all hash buckets :
 111  * A socket might transition from ESTABLISH to LISTEN state without
 112  * RCU grace period. A lookup in ehash table needs to handle this case.
 113  */
 114 #define LISTENING_NULLS_BASE (1U << 29)
 115 struct inet_listen_hashbucket {
 116         spinlock_t              lock;
 117         union {
 118                 struct hlist_head       head;
 119                 struct hlist_nulls_head nulls_head;
 120         };
 121 };
 122
 123 /* This is for listening sockets, thus all sockets which possess wildcards. */
 124 #define INET_LHTABLE_SIZE       32      /* Yes, really, this is all you need. */
 125
 126 struct inet_hashinfo {
 127         /* This is for sockets with full identity only.  Sockets here will
 128          * always be without wildcards and will have the following invariant:
 129          *
 130          *          TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
 131          *
 132          */
 133         struct inet_ehash_bucket        *ehash;
 134         spinlock_t                      *ehash_locks;
 135         unsigned int                    ehash_mask;
 136         unsigned int                    ehash_locks_mask;
 137
 138         /* Ok, let's try this, I give up, we do need a local binding
 139          * TCP hash as well as the others for fast bind/connect.
 140          */
 141         struct inet_bind_hashbucket     *bhash;
 142
 143         unsigned int                    bhash_size;
 144         /* 4 bytes hole on 64 bit */
 145
 146         struct kmem_cache               *bind_bucket_cachep;
 147
 148         /* All the above members are written once at bootup and
 149          * never written again _or_ are predominantly read-access.
 150          *
 151          * Now align to a new cache line as all the following members
 152          * might be often dirty.
 153          */
 154         /* All sockets in TCP_LISTEN state will be in here.  This is the only
 155          * table where wildcard'd TCP sockets can exist.  Hash function here
 156          * is just local port number.
 157          */
 158         struct inet_listen_hashbucket   listening_hash[INET_LHTABLE_SIZE]
 159                                         ____cacheline_aligned_in_smp;
 160 };
 161
 162 static inline struct inet_ehash_bucket *inet_ehash_bucket(
 163         struct inet_hashinfo *hashinfo,
 164         unsigned int hash)
 165 {
 166         return &hashinfo->ehash[hash & hashinfo->ehash_mask];
 167 }
 168
 169 static inline spinlock_t *inet_ehash_lockp(
 170         struct inet_hashinfo *hashinfo,
 171         unsigned int hash)
 172 {
 173         return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
 174 }
 175
 176 int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo);
 177
 178 static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
 179 {
 180         kvfree(hashinfo->ehash_locks);
 181         hashinfo->ehash_locks = NULL;
 182 }
 183
 184 struct inet_bind_bucket *
 185 inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net,
 186                         struct inet_bind_hashbucket *head,
 187                         const unsigned short snum);
 188 void inet_bind_bucket_destroy(struct kmem_cache *cachep,
 189                               struct inet_bind_bucket *tb);
 190
 191 static inline u32 inet_bhashfn(const struct net *net, const __u16 lport,
 192                                const u32 bhash_size)
 193 {
 194         return (lport + net_hash_mix(net)) & (bhash_size - 1);
 195 }
 196
 197 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
 198                     const unsigned short snum);
 199
 200 /* These can have wildcards, don't try too hard. */
 201 static inline u32 inet_lhashfn(const struct net *net, const unsigned short num)
 202 {
 203         return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1);
 204 }
 205
 206 static inline int inet_sk_listen_hashfn(const struct sock *sk)
 207 {
 208         return inet_lhashfn(sock_net(sk), inet_sk(sk)->inet_num);
 209 }
 210
 211 /* Caller must disable local BH processing. */
 212 int __inet_inherit_port(const struct sock *sk, struct sock *child);
 213
 214 void inet_put_port(struct sock *sk);
 215
 216 void inet_hashinfo_init(struct inet_hashinfo *h);
 217
 218 bool inet_ehash_insert(struct sock *sk, struct sock *osk);
 219 bool inet_ehash_nolisten(struct sock *sk, struct sock *osk);
 220 int __inet_hash(struct sock *sk, struct sock *osk);
 221 int inet_hash(struct sock *sk);
 222 void inet_unhash(struct sock *sk);
 223
 224 struct sock *__inet_lookup_listener(struct net *net,
 225                                     struct inet_hashinfo *hashinfo,
 226                                     struct sk_buff *skb, int doff,
 227                                     const __be32 saddr, const __be16 sport,
 228                                     const __be32 daddr,
 229                                     const unsigned short hnum,
 230                                     const int dif, const int sdif);
 231
 232 static inline struct sock *inet_lookup_listener(struct net *net,
 233                 struct inet_hashinfo *hashinfo,
 234                 struct sk_buff *skb, int doff,
 235                 __be32 saddr, __be16 sport,
 236                 __be32 daddr, __be16 dport, int dif, int sdif)
 237 {
 238         return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, sport,
 239                                       daddr, ntohs(dport), dif, sdif);
 240 }
 241
 242 /* Socket demux engine toys. */
 243 /* What happens here is ugly; there's a pair of adjacent fields in
 244    struct inet_sock; __be16 dport followed by __u16 num.  We want to
 245    search by pair, so we combine the keys into a single 32bit value
 246    and compare with 32bit value read from &...->dport.  Let's at least
 247    make sure that it's not mixed with anything else...
 248    On 64bit targets we combine comparisons with pair of adjacent __be32
 249    fields in the same way.
 250 */
 251 #ifdef __BIG_ENDIAN
 252 #define INET_COMBINED_PORTS(__sport, __dport) \
 253         ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
 254 #else /* __LITTLE_ENDIAN */
 255 #define INET_COMBINED_PORTS(__sport, __dport) \
 256         ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
 257 #endif
 258
 259 #if (BITS_PER_LONG == 64)
 260 #ifdef __BIG_ENDIAN
 261 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
 262         const __addrpair __name = (__force __addrpair) ( \
 263                                    (((__force __u64)(__be32)(__saddr)) << 32) | \
 264                                    ((__force __u64)(__be32)(__daddr)))
 265 #else /* __LITTLE_ENDIAN */
 266 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
 267         const __addrpair __name = (__force __addrpair) ( \
 268                                    (((__force __u64)(__be32)(__daddr)) << 32) | \
 269                                    ((__force __u64)(__be32)(__saddr)))
 270 #endif /* __BIG_ENDIAN */
 271 #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif, __sdif) \
 272         (((__sk)->sk_portpair == (__ports))                     &&      \
 273          ((__sk)->sk_addrpair == (__cookie))                    &&      \
 274          (!(__sk)->sk_bound_dev_if      ||                              \
 275            ((__sk)->sk_bound_dev_if == (__dif))                 ||      \
 276            ((__sk)->sk_bound_dev_if == (__sdif)))               &&      \
 277          net_eq(sock_net(__sk), (__net)))
 278 #else /* 32-bit arch */
 279 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
 280         const int __name __deprecated __attribute__((unused))
 281
 282 #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif, __sdif) \
 283         (((__sk)->sk_portpair == (__ports))             &&              \
 284          ((__sk)->sk_daddr      == (__saddr))           &&              \
 285          ((__sk)->sk_rcv_saddr  == (__daddr))           &&              \
 286          (!(__sk)->sk_bound_dev_if      ||                              \
 287            ((__sk)->sk_bound_dev_if == (__dif))         ||              \
 288            ((__sk)->sk_bound_dev_if == (__sdif)))       &&              \
 289          net_eq(sock_net(__sk), (__net)))
 290 #endif /* 64-bit arch */
 291
 292 /* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
 293  * not check it for lookups anymore, thanks Alexey. -DaveM
 294  */
 295 struct sock *__inet_lookup_established(struct net *net,
 296                                        struct inet_hashinfo *hashinfo,
 297                                        const __be32 saddr, const __be16 sport,
 298                                        const __be32 daddr, const u16 hnum,
 299                                        const int dif, const int sdif);
 300
 301 static inline struct sock *
 302         inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
 303                                 const __be32 saddr, const __be16 sport,
 304                                 const __be32 daddr, const __be16 dport,
 305                                 const int dif)
 306 {
 307         return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
 308                                          ntohs(dport), dif, 0);
 309 }
 310
 311 static inline struct sock *__inet_lookup(struct net *net,
 312                                          struct inet_hashinfo *hashinfo,
 313                                          struct sk_buff *skb, int doff,
 314                                          const __be32 saddr, const __be16 sport,
 315                                          const __be32 daddr, const __be16 dport,
 316                                          const int dif, const int sdif,
 317                                          bool *refcounted)
 318 {
 319         u16 hnum = ntohs(dport);
 320         struct sock *sk;
 321
 322         sk = __inet_lookup_established(net, hashinfo, saddr, sport,
 323                                        daddr, hnum, dif, sdif);
 324         *refcounted = true;
 325         if (sk)
 326                 return sk;
 327         *refcounted = false;
 328         return __inet_lookup_listener(net, hashinfo, skb, doff, saddr,
 329                                       sport, daddr, hnum, dif, sdif);
 330 }
 331
 332 static inline struct sock *inet_lookup(struct net *net,
 333                                        struct inet_hashinfo *hashinfo,
 334                                        struct sk_buff *skb, int doff,
 335                                        const __be32 saddr, const __be16 sport,
 336                                        const __be32 daddr, const __be16 dport,
 337                                        const int dif)
 338 {
 339         struct sock *sk;
 340         bool refcounted;
 341
 342         sk = __inet_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
 343                            dport, dif, 0, &refcounted);
 344
 345         if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
 346                 sk = NULL;
 347         return sk;
 348 }
 349
 350 static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
 351                                              struct sk_buff *skb,
 352                                              int doff,
 353                                              const __be16 sport,
 354                                              const __be16 dport,
 355                                              const int sdif,
 356                                              bool *refcounted)
 357 {
 358         struct sock *sk = skb_steal_sock(skb);
 359         const struct iphdr *iph = ip_hdr(skb);
 360
 361         *refcounted = true;
 362         if (sk)
 363                 return sk;
 364
 365         return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo, skb,
 366                              doff, iph->saddr, sport,
 367                              iph->daddr, dport, inet_iif(skb), sdif,
 368                              refcounted);
 369 }
 370
 371 u32 inet6_ehashfn(const struct net *net,
 372                   const struct in6_addr *laddr, const u16 lport,
 373                   const struct in6_addr *faddr, const __be16 fport);
 374
 375 static inline void sk_daddr_set(struct sock *sk, __be32 addr)
 376 {
 377         sk->sk_daddr = addr; /* alias of inet_daddr */
 378 #if IS_ENABLED(CONFIG_IPV6)
 379         ipv6_addr_set_v4mapped(addr, &sk->sk_v6_daddr);
 380 #endif
 381 }
 382
 383 static inline void sk_rcv_saddr_set(struct sock *sk, __be32 addr)
 384 {
 385         sk->sk_rcv_saddr = addr; /* alias of inet_rcv_saddr */
 386 #if IS_ENABLED(CONFIG_IPV6)
 387         ipv6_addr_set_v4mapped(addr, &sk->sk_v6_rcv_saddr);
 388 #endif
 389 }
 390
 391 int __inet_hash_connect(struct inet_timewait_death_row *death_row,
 392                         struct sock *sk, u32 port_offset,
 393                         int (*check_established)(struct inet_timewait_death_row *,
 394                                                  struct sock *, __u16,
 395                                                  struct inet_timewait_sock **));
 396
 397 int inet_hash_connect(struct inet_timewait_death_row *death_row,
 398                       struct sock *sk);
 399 #endif /* _INET_HASHTABLES_H */