1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Authors: Lotsa people, from code originally in tcp 8 */ 9 10 #ifndef _INET_HASHTABLES_H 11 #define _INET_HASHTABLES_H 12 13 14 #include <linux/interrupt.h> 15 #include <linux/ip.h> 16 #include <linux/ipv6.h> 17 #include <linux/list.h> 18 #include <linux/slab.h> 19 #include <linux/socket.h> 20 #include <linux/spinlock.h> 21 #include <linux/types.h> 22 #include <linux/wait.h> 23 24 #include <net/inet_connection_sock.h> 25 #include <net/inet_sock.h> 26 #include <net/sock.h> 27 #include <net/route.h> 28 #include <net/tcp_states.h> 29 #include <net/netns/hash.h> 30 31 #include <linux/refcount.h> 32 #include <asm/byteorder.h> 33 34 /* This is for all connections with a full identity, no wildcards. 35 * The 'e' prefix stands for Establish, but we really put all sockets 36 * but LISTEN ones. 37 */ 38 struct inet_ehash_bucket { 39 struct hlist_nulls_head chain; 40 }; 41 42 /* There are a few simple rules, which allow for local port reuse by 43 * an application. In essence: 44 * 45 * 1) Sockets bound to different interfaces may share a local port. 46 * Failing that, goto test 2. 47 * 2) If all sockets have sk->sk_reuse set, and none of them are in 48 * TCP_LISTEN state, the port may be shared. 49 * Failing that, goto test 3. 50 * 3) If all sockets are bound to a specific inet_sk(sk)->rcv_saddr local 51 * address, and none of them are the same, the port may be 52 * shared. 53 * Failing this, the port cannot be shared. 54 * 55 * The interesting point, is test #2. This is what an FTP server does 56 * all day. To optimize this case we use a specific flag bit defined 57 * below. As we add sockets to a bind bucket list, we perform a 58 * check of: (newsk->sk_reuse && (newsk->sk_state != TCP_LISTEN)) 59 * As long as all sockets added to a bind bucket pass this test, 60 * the flag bit will be set. 61 * The resulting situation is that tcp_v[46]_verify_bind() can just check 62 * for this flag bit, if it is set and the socket trying to bind has 63 * sk->sk_reuse set, we don't even have to walk the owners list at all, 64 * we return that it is ok to bind this socket to the requested local port. 65 * 66 * Sounds like a lot of work, but it is worth it. In a more naive 67 * implementation (ie. current FreeBSD etc.) the entire list of ports 68 * must be walked for each data port opened by an ftp server. Needless 69 * to say, this does not scale at all. With a couple thousand FTP 70 * users logged onto your box, isn't it nice to know that new data 71 * ports are created in O(1) time? I thought so. ;-) -DaveM 72 */ 73 #define FASTREUSEPORT_ANY 1 74 #define FASTREUSEPORT_STRICT 2 75 76 struct inet_bind_bucket { 77 possible_net_t ib_net; 78 int l3mdev; 79 unsigned short port; 80 signed char fastreuse; 81 signed char fastreuseport; 82 kuid_t fastuid; 83 #if IS_ENABLED(CONFIG_IPV6) 84 struct in6_addr fast_v6_rcv_saddr; 85 #endif 86 __be32 fast_rcv_saddr; 87 unsigned short fast_sk_family; 88 bool fast_ipv6_only; 89 struct hlist_node node; 90 struct hlist_head owners; 91 }; 92 93 static inline struct net *ib_net(struct inet_bind_bucket *ib) 94 { 95 return read_pnet(&ib->ib_net); 96 } 97 98 #define inet_bind_bucket_for_each(tb, head) \ 99 hlist_for_each_entry(tb, head, node) 100 101 struct inet_bind_hashbucket { 102 spinlock_t lock; 103 struct hlist_head chain; 104 }; 105 106 /* Sockets can be hashed in established or listening table. 107 * We must use different 'nulls' end-of-chain value for all hash buckets : 108 * A socket might transition from ESTABLISH to LISTEN state without 109 * RCU grace period. A lookup in ehash table needs to handle this case. 110 */ 111 #define LISTENING_NULLS_BASE (1U << 29) 112 struct inet_listen_hashbucket { 113 spinlock_t lock; 114 unsigned int count; 115 union { 116 struct hlist_head head; 117 struct hlist_nulls_head nulls_head; 118 }; 119 }; 120 121 /* This is for listening sockets, thus all sockets which possess wildcards. */ 122 #define INET_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */ 123 124 struct inet_hashinfo { 125 /* This is for sockets with full identity only. Sockets here will 126 * always be without wildcards and will have the following invariant: 127 * 128 * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE 129 * 130 */ 131 struct inet_ehash_bucket *ehash; 132 spinlock_t *ehash_locks; 133 unsigned int ehash_mask; 134 unsigned int ehash_locks_mask; 135 136 /* Ok, let's try this, I give up, we do need a local binding 137 * TCP hash as well as the others for fast bind/connect. 138 */ 139 struct kmem_cache *bind_bucket_cachep; 140 struct inet_bind_hashbucket *bhash; 141 unsigned int bhash_size; 142 143 /* The 2nd listener table hashed by local port and address */ 144 unsigned int lhash2_mask; 145 struct inet_listen_hashbucket *lhash2; 146 147 /* All the above members are written once at bootup and 148 * never written again _or_ are predominantly read-access. 149 * 150 * Now align to a new cache line as all the following members 151 * might be often dirty. 152 */ 153 /* All sockets in TCP_LISTEN state will be in listening_hash. 154 * This is the only table where wildcard'd TCP sockets can 155 * exist. listening_hash is only hashed by local port number. 156 * If lhash2 is initialized, the same socket will also be hashed 157 * to lhash2 by port and address. 158 */ 159 struct inet_listen_hashbucket listening_hash[INET_LHTABLE_SIZE] 160 ____cacheline_aligned_in_smp; 161 }; 162 163 #define inet_lhash2_for_each_icsk_rcu(__icsk, list) \ 164 hlist_for_each_entry_rcu(__icsk, list, icsk_listen_portaddr_node) 165 166 static inline struct inet_listen_hashbucket * 167 inet_lhash2_bucket(struct inet_hashinfo *h, u32 hash) 168 { 169 return &h->lhash2[hash & h->lhash2_mask]; 170 } 171 172 static inline struct inet_ehash_bucket *inet_ehash_bucket( 173 struct inet_hashinfo *hashinfo, 174 unsigned int hash) 175 { 176 return &hashinfo->ehash[hash & hashinfo->ehash_mask]; 177 } 178 179 static inline spinlock_t *inet_ehash_lockp( 180 struct inet_hashinfo *hashinfo, 181 unsigned int hash) 182 { 183 return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask]; 184 } 185 186 int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo); 187 188 static inline void inet_hashinfo2_free_mod(struct inet_hashinfo *h) 189 { 190 kfree(h->lhash2); 191 h->lhash2 = NULL; 192 } 193 194 static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo) 195 { 196 kvfree(hashinfo->ehash_locks); 197 hashinfo->ehash_locks = NULL; 198 } 199 200 static inline bool inet_sk_bound_dev_eq(struct net *net, int bound_dev_if, 201 int dif, int sdif) 202 { 203 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 204 return inet_bound_dev_eq(!!net->ipv4.sysctl_tcp_l3mdev_accept, 205 bound_dev_if, dif, sdif); 206 #else 207 return inet_bound_dev_eq(true, bound_dev_if, dif, sdif); 208 #endif 209 } 210 211 struct inet_bind_bucket * 212 inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net, 213 struct inet_bind_hashbucket *head, 214 const unsigned short snum, int l3mdev); 215 void inet_bind_bucket_destroy(struct kmem_cache *cachep, 216 struct inet_bind_bucket *tb); 217 218 static inline u32 inet_bhashfn(const struct net *net, const __u16 lport, 219 const u32 bhash_size) 220 { 221 return (lport + net_hash_mix(net)) & (bhash_size - 1); 222 } 223 224 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb, 225 const unsigned short snum); 226 227 /* These can have wildcards, don't try too hard. */ 228 static inline u32 inet_lhashfn(const struct net *net, const unsigned short num) 229 { 230 return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1); 231 } 232 233 static inline int inet_sk_listen_hashfn(const struct sock *sk) 234 { 235 return inet_lhashfn(sock_net(sk), inet_sk(sk)->inet_num); 236 } 237 238 /* Caller must disable local BH processing. */ 239 int __inet_inherit_port(const struct sock *sk, struct sock *child); 240 241 void inet_put_port(struct sock *sk); 242 243 void inet_hashinfo_init(struct inet_hashinfo *h); 244 void inet_hashinfo2_init(struct inet_hashinfo *h, const char *name, 245 unsigned long numentries, int scale, 246 unsigned long low_limit, 247 unsigned long high_limit); 248 int inet_hashinfo2_init_mod(struct inet_hashinfo *h); 249 250 bool inet_ehash_insert(struct sock *sk, struct sock *osk); 251 bool inet_ehash_nolisten(struct sock *sk, struct sock *osk); 252 int __inet_hash(struct sock *sk, struct sock *osk); 253 int inet_hash(struct sock *sk); 254 void inet_unhash(struct sock *sk); 255 256 struct sock *__inet_lookup_listener(struct net *net, 257 struct inet_hashinfo *hashinfo, 258 struct sk_buff *skb, int doff, 259 const __be32 saddr, const __be16 sport, 260 const __be32 daddr, 261 const unsigned short hnum, 262 const int dif, const int sdif); 263 264 static inline struct sock *inet_lookup_listener(struct net *net, 265 struct inet_hashinfo *hashinfo, 266 struct sk_buff *skb, int doff, 267 __be32 saddr, __be16 sport, 268 __be32 daddr, __be16 dport, int dif, int sdif) 269 { 270 return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, sport, 271 daddr, ntohs(dport), dif, sdif); 272 } 273 274 /* Socket demux engine toys. */ 275 /* What happens here is ugly; there's a pair of adjacent fields in 276 struct inet_sock; __be16 dport followed by __u16 num. We want to 277 search by pair, so we combine the keys into a single 32bit value 278 and compare with 32bit value read from &...->dport. Let's at least 279 make sure that it's not mixed with anything else... 280 On 64bit targets we combine comparisons with pair of adjacent __be32 281 fields in the same way. 282 */ 283 #ifdef __BIG_ENDIAN 284 #define INET_COMBINED_PORTS(__sport, __dport) \ 285 ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport))) 286 #else /* __LITTLE_ENDIAN */ 287 #define INET_COMBINED_PORTS(__sport, __dport) \ 288 ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport))) 289 #endif 290 291 #if (BITS_PER_LONG == 64) 292 #ifdef __BIG_ENDIAN 293 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ 294 const __addrpair __name = (__force __addrpair) ( \ 295 (((__force __u64)(__be32)(__saddr)) << 32) | \ 296 ((__force __u64)(__be32)(__daddr))) 297 #else /* __LITTLE_ENDIAN */ 298 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ 299 const __addrpair __name = (__force __addrpair) ( \ 300 (((__force __u64)(__be32)(__daddr)) << 32) | \ 301 ((__force __u64)(__be32)(__saddr))) 302 #endif /* __BIG_ENDIAN */ 303 #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif, __sdif) \ 304 (((__sk)->sk_portpair == (__ports)) && \ 305 ((__sk)->sk_addrpair == (__cookie)) && \ 306 (((__sk)->sk_bound_dev_if == (__dif)) || \ 307 ((__sk)->sk_bound_dev_if == (__sdif))) && \ 308 net_eq(sock_net(__sk), (__net))) 309 #else /* 32-bit arch */ 310 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ 311 const int __name __deprecated __attribute__((unused)) 312 313 #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif, __sdif) \ 314 (((__sk)->sk_portpair == (__ports)) && \ 315 ((__sk)->sk_daddr == (__saddr)) && \ 316 ((__sk)->sk_rcv_saddr == (__daddr)) && \ 317 (((__sk)->sk_bound_dev_if == (__dif)) || \ 318 ((__sk)->sk_bound_dev_if == (__sdif))) && \ 319 net_eq(sock_net(__sk), (__net))) 320 #endif /* 64-bit arch */ 321 322 /* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need 323 * not check it for lookups anymore, thanks Alexey. -DaveM 324 */ 325 struct sock *__inet_lookup_established(struct net *net, 326 struct inet_hashinfo *hashinfo, 327 const __be32 saddr, const __be16 sport, 328 const __be32 daddr, const u16 hnum, 329 const int dif, const int sdif); 330 331 static inline struct sock * 332 inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo, 333 const __be32 saddr, const __be16 sport, 334 const __be32 daddr, const __be16 dport, 335 const int dif) 336 { 337 return __inet_lookup_established(net, hashinfo, saddr, sport, daddr, 338 ntohs(dport), dif, 0); 339 } 340 341 static inline struct sock *__inet_lookup(struct net *net, 342 struct inet_hashinfo *hashinfo, 343 struct sk_buff *skb, int doff, 344 const __be32 saddr, const __be16 sport, 345 const __be32 daddr, const __be16 dport, 346 const int dif, const int sdif, 347 bool *refcounted) 348 { 349 u16 hnum = ntohs(dport); 350 struct sock *sk; 351 352 sk = __inet_lookup_established(net, hashinfo, saddr, sport, 353 daddr, hnum, dif, sdif); 354 *refcounted = true; 355 if (sk) 356 return sk; 357 *refcounted = false; 358 return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, 359 sport, daddr, hnum, dif, sdif); 360 } 361 362 static inline struct sock *inet_lookup(struct net *net, 363 struct inet_hashinfo *hashinfo, 364 struct sk_buff *skb, int doff, 365 const __be32 saddr, const __be16 sport, 366 const __be32 daddr, const __be16 dport, 367 const int dif) 368 { 369 struct sock *sk; 370 bool refcounted; 371 372 sk = __inet_lookup(net, hashinfo, skb, doff, saddr, sport, daddr, 373 dport, dif, 0, &refcounted); 374 375 if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt)) 376 sk = NULL; 377 return sk; 378 } 379 380 static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo, 381 struct sk_buff *skb, 382 int doff, 383 const __be16 sport, 384 const __be16 dport, 385 const int sdif, 386 bool *refcounted) 387 { 388 struct sock *sk = skb_steal_sock(skb, refcounted); 389 const struct iphdr *iph = ip_hdr(skb); 390 391 if (sk) 392 return sk; 393 394 return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo, skb, 395 doff, iph->saddr, sport, 396 iph->daddr, dport, inet_iif(skb), sdif, 397 refcounted); 398 } 399 400 u32 inet6_ehashfn(const struct net *net, 401 const struct in6_addr *laddr, const u16 lport, 402 const struct in6_addr *faddr, const __be16 fport); 403 404 static inline void sk_daddr_set(struct sock *sk, __be32 addr) 405 { 406 sk->sk_daddr = addr; /* alias of inet_daddr */ 407 #if IS_ENABLED(CONFIG_IPV6) 408 ipv6_addr_set_v4mapped(addr, &sk->sk_v6_daddr); 409 #endif 410 } 411 412 static inline void sk_rcv_saddr_set(struct sock *sk, __be32 addr) 413 { 414 sk->sk_rcv_saddr = addr; /* alias of inet_rcv_saddr */ 415 #if IS_ENABLED(CONFIG_IPV6) 416 ipv6_addr_set_v4mapped(addr, &sk->sk_v6_rcv_saddr); 417 #endif 418 } 419 420 int __inet_hash_connect(struct inet_timewait_death_row *death_row, 421 struct sock *sk, u32 port_offset, 422 int (*check_established)(struct inet_timewait_death_row *, 423 struct sock *, __u16, 424 struct inet_timewait_sock **)); 425 426 int inet_hash_connect(struct inet_timewait_death_row *death_row, 427 struct sock *sk); 428 #endif /* _INET_HASHTABLES_H */ 429