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