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 * Definitions for the TCP module. 7 * 8 * Version: @(#)tcp.h 1.0.5 05/23/93 9 * 10 * Authors: Ross Biro, <bir7@leland.Stanford.Edu> 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * 13 * This program is free software; you can redistribute it and/or 14 * modify it under the terms of the GNU General Public License 15 * as published by the Free Software Foundation; either version 16 * 2 of the License, or (at your option) any later version. 17 */ 18 #ifndef _TCP_H 19 #define _TCP_H 20 21 #define TCP_DEBUG 1 22 #define FASTRETRANS_DEBUG 1 23 24 /* Cancel timers, when they are not required. */ 25 #undef TCP_CLEAR_TIMERS 26 27 #include <linux/config.h> 28 #include <linux/list.h> 29 #include <linux/tcp.h> 30 #include <linux/slab.h> 31 #include <linux/cache.h> 32 #include <linux/percpu.h> 33 #include <net/checksum.h> 34 #include <net/sock.h> 35 #include <net/snmp.h> 36 #include <net/ip.h> 37 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE) 38 #include <linux/ipv6.h> 39 #endif 40 #include <linux/seq_file.h> 41 42 /* This is for all connections with a full identity, no wildcards. 43 * New scheme, half the table is for TIME_WAIT, the other half is 44 * for the rest. I'll experiment with dynamic table growth later. 45 */ 46 struct tcp_ehash_bucket { 47 rwlock_t lock; 48 struct hlist_head chain; 49 } __attribute__((__aligned__(8))); 50 51 /* This is for listening sockets, thus all sockets which possess wildcards. */ 52 #define TCP_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */ 53 54 /* There are a few simple rules, which allow for local port reuse by 55 * an application. In essence: 56 * 57 * 1) Sockets bound to different interfaces may share a local port. 58 * Failing that, goto test 2. 59 * 2) If all sockets have sk->sk_reuse set, and none of them are in 60 * TCP_LISTEN state, the port may be shared. 61 * Failing that, goto test 3. 62 * 3) If all sockets are bound to a specific inet_sk(sk)->rcv_saddr local 63 * address, and none of them are the same, the port may be 64 * shared. 65 * Failing this, the port cannot be shared. 66 * 67 * The interesting point, is test #2. This is what an FTP server does 68 * all day. To optimize this case we use a specific flag bit defined 69 * below. As we add sockets to a bind bucket list, we perform a 70 * check of: (newsk->sk_reuse && (newsk->sk_state != TCP_LISTEN)) 71 * As long as all sockets added to a bind bucket pass this test, 72 * the flag bit will be set. 73 * The resulting situation is that tcp_v[46]_verify_bind() can just check 74 * for this flag bit, if it is set and the socket trying to bind has 75 * sk->sk_reuse set, we don't even have to walk the owners list at all, 76 * we return that it is ok to bind this socket to the requested local port. 77 * 78 * Sounds like a lot of work, but it is worth it. In a more naive 79 * implementation (ie. current FreeBSD etc.) the entire list of ports 80 * must be walked for each data port opened by an ftp server. Needless 81 * to say, this does not scale at all. With a couple thousand FTP 82 * users logged onto your box, isn't it nice to know that new data 83 * ports are created in O(1) time? I thought so. ;-) -DaveM 84 */ 85 struct tcp_bind_bucket { 86 unsigned short port; 87 signed short fastreuse; 88 struct hlist_node node; 89 struct hlist_head owners; 90 }; 91 92 #define tb_for_each(tb, node, head) hlist_for_each_entry(tb, node, head, node) 93 94 struct tcp_bind_hashbucket { 95 spinlock_t lock; 96 struct hlist_head chain; 97 }; 98 99 static inline struct tcp_bind_bucket *__tb_head(struct tcp_bind_hashbucket *head) 100 { 101 return hlist_entry(head->chain.first, struct tcp_bind_bucket, node); 102 } 103 104 static inline struct tcp_bind_bucket *tb_head(struct tcp_bind_hashbucket *head) 105 { 106 return hlist_empty(&head->chain) ? NULL : __tb_head(head); 107 } 108 109 extern struct tcp_hashinfo { 110 /* This is for sockets with full identity only. Sockets here will 111 * always be without wildcards and will have the following invariant: 112 * 113 * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE 114 * 115 * First half of the table is for sockets not in TIME_WAIT, second half 116 * is for TIME_WAIT sockets only. 117 */ 118 struct tcp_ehash_bucket *__tcp_ehash; 119 120 /* Ok, let's try this, I give up, we do need a local binding 121 * TCP hash as well as the others for fast bind/connect. 122 */ 123 struct tcp_bind_hashbucket *__tcp_bhash; 124 125 int __tcp_bhash_size; 126 int __tcp_ehash_size; 127 128 /* All sockets in TCP_LISTEN state will be in here. This is the only 129 * table where wildcard'd TCP sockets can exist. Hash function here 130 * is just local port number. 131 */ 132 struct hlist_head __tcp_listening_hash[TCP_LHTABLE_SIZE]; 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 * are often dirty. 139 */ 140 rwlock_t __tcp_lhash_lock ____cacheline_aligned; 141 atomic_t __tcp_lhash_users; 142 wait_queue_head_t __tcp_lhash_wait; 143 spinlock_t __tcp_portalloc_lock; 144 } tcp_hashinfo; 145 146 #define tcp_ehash (tcp_hashinfo.__tcp_ehash) 147 #define tcp_bhash (tcp_hashinfo.__tcp_bhash) 148 #define tcp_ehash_size (tcp_hashinfo.__tcp_ehash_size) 149 #define tcp_bhash_size (tcp_hashinfo.__tcp_bhash_size) 150 #define tcp_listening_hash (tcp_hashinfo.__tcp_listening_hash) 151 #define tcp_lhash_lock (tcp_hashinfo.__tcp_lhash_lock) 152 #define tcp_lhash_users (tcp_hashinfo.__tcp_lhash_users) 153 #define tcp_lhash_wait (tcp_hashinfo.__tcp_lhash_wait) 154 #define tcp_portalloc_lock (tcp_hashinfo.__tcp_portalloc_lock) 155 156 extern kmem_cache_t *tcp_bucket_cachep; 157 extern struct tcp_bind_bucket *tcp_bucket_create(struct tcp_bind_hashbucket *head, 158 unsigned short snum); 159 extern void tcp_bucket_destroy(struct tcp_bind_bucket *tb); 160 extern void tcp_bucket_unlock(struct sock *sk); 161 extern int tcp_port_rover; 162 163 /* These are AF independent. */ 164 static __inline__ int tcp_bhashfn(__u16 lport) 165 { 166 return (lport & (tcp_bhash_size - 1)); 167 } 168 169 extern void tcp_bind_hash(struct sock *sk, struct tcp_bind_bucket *tb, 170 unsigned short snum); 171 172 #if (BITS_PER_LONG == 64) 173 #define TCP_ADDRCMP_ALIGN_BYTES 8 174 #else 175 #define TCP_ADDRCMP_ALIGN_BYTES 4 176 #endif 177 178 /* This is a TIME_WAIT bucket. It works around the memory consumption 179 * problems of sockets in such a state on heavily loaded servers, but 180 * without violating the protocol specification. 181 */ 182 struct tcp_tw_bucket { 183 /* 184 * Now struct sock also uses sock_common, so please just 185 * don't add nothing before this first member (__tw_common) --acme 186 */ 187 struct sock_common __tw_common; 188 #define tw_family __tw_common.skc_family 189 #define tw_state __tw_common.skc_state 190 #define tw_reuse __tw_common.skc_reuse 191 #define tw_bound_dev_if __tw_common.skc_bound_dev_if 192 #define tw_node __tw_common.skc_node 193 #define tw_bind_node __tw_common.skc_bind_node 194 #define tw_refcnt __tw_common.skc_refcnt 195 volatile unsigned char tw_substate; 196 unsigned char tw_rcv_wscale; 197 __u16 tw_sport; 198 /* Socket demultiplex comparisons on incoming packets. */ 199 /* these five are in inet_sock */ 200 __u32 tw_daddr 201 __attribute__((aligned(TCP_ADDRCMP_ALIGN_BYTES))); 202 __u32 tw_rcv_saddr; 203 __u16 tw_dport; 204 __u16 tw_num; 205 /* And these are ours. */ 206 int tw_hashent; 207 int tw_timeout; 208 __u32 tw_rcv_nxt; 209 __u32 tw_snd_nxt; 210 __u32 tw_rcv_wnd; 211 __u32 tw_ts_recent; 212 long tw_ts_recent_stamp; 213 unsigned long tw_ttd; 214 struct tcp_bind_bucket *tw_tb; 215 struct hlist_node tw_death_node; 216 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 217 struct in6_addr tw_v6_daddr; 218 struct in6_addr tw_v6_rcv_saddr; 219 int tw_v6_ipv6only; 220 #endif 221 }; 222 223 static __inline__ void tw_add_node(struct tcp_tw_bucket *tw, 224 struct hlist_head *list) 225 { 226 hlist_add_head(&tw->tw_node, list); 227 } 228 229 static __inline__ void tw_add_bind_node(struct tcp_tw_bucket *tw, 230 struct hlist_head *list) 231 { 232 hlist_add_head(&tw->tw_bind_node, list); 233 } 234 235 static inline int tw_dead_hashed(struct tcp_tw_bucket *tw) 236 { 237 return tw->tw_death_node.pprev != NULL; 238 } 239 240 static __inline__ void tw_dead_node_init(struct tcp_tw_bucket *tw) 241 { 242 tw->tw_death_node.pprev = NULL; 243 } 244 245 static __inline__ void __tw_del_dead_node(struct tcp_tw_bucket *tw) 246 { 247 __hlist_del(&tw->tw_death_node); 248 tw_dead_node_init(tw); 249 } 250 251 static __inline__ int tw_del_dead_node(struct tcp_tw_bucket *tw) 252 { 253 if (tw_dead_hashed(tw)) { 254 __tw_del_dead_node(tw); 255 return 1; 256 } 257 return 0; 258 } 259 260 #define tw_for_each(tw, node, head) \ 261 hlist_for_each_entry(tw, node, head, tw_node) 262 263 #define tw_for_each_inmate(tw, node, jail) \ 264 hlist_for_each_entry(tw, node, jail, tw_death_node) 265 266 #define tw_for_each_inmate_safe(tw, node, safe, jail) \ 267 hlist_for_each_entry_safe(tw, node, safe, jail, tw_death_node) 268 269 #define tcptw_sk(__sk) ((struct tcp_tw_bucket *)(__sk)) 270 271 static inline u32 tcp_v4_rcv_saddr(const struct sock *sk) 272 { 273 return likely(sk->sk_state != TCP_TIME_WAIT) ? 274 inet_sk(sk)->rcv_saddr : tcptw_sk(sk)->tw_rcv_saddr; 275 } 276 277 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 278 static inline struct in6_addr *__tcp_v6_rcv_saddr(const struct sock *sk) 279 { 280 return likely(sk->sk_state != TCP_TIME_WAIT) ? 281 &inet6_sk(sk)->rcv_saddr : &tcptw_sk(sk)->tw_v6_rcv_saddr; 282 } 283 284 static inline struct in6_addr *tcp_v6_rcv_saddr(const struct sock *sk) 285 { 286 return sk->sk_family == AF_INET6 ? __tcp_v6_rcv_saddr(sk) : NULL; 287 } 288 289 #define tcptw_sk_ipv6only(__sk) (tcptw_sk(__sk)->tw_v6_ipv6only) 290 291 static inline int tcp_v6_ipv6only(const struct sock *sk) 292 { 293 return likely(sk->sk_state != TCP_TIME_WAIT) ? 294 ipv6_only_sock(sk) : tcptw_sk_ipv6only(sk); 295 } 296 #else 297 # define __tcp_v6_rcv_saddr(__sk) NULL 298 # define tcp_v6_rcv_saddr(__sk) NULL 299 # define tcptw_sk_ipv6only(__sk) 0 300 # define tcp_v6_ipv6only(__sk) 0 301 #endif 302 303 extern kmem_cache_t *tcp_timewait_cachep; 304 305 static inline void tcp_tw_put(struct tcp_tw_bucket *tw) 306 { 307 if (atomic_dec_and_test(&tw->tw_refcnt)) { 308 #ifdef INET_REFCNT_DEBUG 309 printk(KERN_DEBUG "tw_bucket %p released\n", tw); 310 #endif 311 kmem_cache_free(tcp_timewait_cachep, tw); 312 } 313 } 314 315 extern atomic_t tcp_orphan_count; 316 extern int tcp_tw_count; 317 extern void tcp_time_wait(struct sock *sk, int state, int timeo); 318 extern void tcp_tw_deschedule(struct tcp_tw_bucket *tw); 319 320 321 /* Socket demux engine toys. */ 322 #ifdef __BIG_ENDIAN 323 #define TCP_COMBINED_PORTS(__sport, __dport) \ 324 (((__u32)(__sport)<<16) | (__u32)(__dport)) 325 #else /* __LITTLE_ENDIAN */ 326 #define TCP_COMBINED_PORTS(__sport, __dport) \ 327 (((__u32)(__dport)<<16) | (__u32)(__sport)) 328 #endif 329 330 #if (BITS_PER_LONG == 64) 331 #ifdef __BIG_ENDIAN 332 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) \ 333 __u64 __name = (((__u64)(__saddr))<<32)|((__u64)(__daddr)); 334 #else /* __LITTLE_ENDIAN */ 335 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) \ 336 __u64 __name = (((__u64)(__daddr))<<32)|((__u64)(__saddr)); 337 #endif /* __BIG_ENDIAN */ 338 #define TCP_IPV4_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\ 339 (((*((__u64 *)&(inet_sk(__sk)->daddr)))== (__cookie)) && \ 340 ((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \ 341 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) 342 #define TCP_IPV4_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\ 343 (((*((__u64 *)&(tcptw_sk(__sk)->tw_daddr))) == (__cookie)) && \ 344 ((*((__u32 *)&(tcptw_sk(__sk)->tw_dport))) == (__ports)) && \ 345 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) 346 #else /* 32-bit arch */ 347 #define TCP_V4_ADDR_COOKIE(__name, __saddr, __daddr) 348 #define TCP_IPV4_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\ 349 ((inet_sk(__sk)->daddr == (__saddr)) && \ 350 (inet_sk(__sk)->rcv_saddr == (__daddr)) && \ 351 ((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \ 352 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) 353 #define TCP_IPV4_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\ 354 ((tcptw_sk(__sk)->tw_daddr == (__saddr)) && \ 355 (tcptw_sk(__sk)->tw_rcv_saddr == (__daddr)) && \ 356 ((*((__u32 *)&(tcptw_sk(__sk)->tw_dport))) == (__ports)) && \ 357 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) 358 #endif /* 64-bit arch */ 359 360 #define TCP_IPV6_MATCH(__sk, __saddr, __daddr, __ports, __dif) \ 361 (((*((__u32 *)&(inet_sk(__sk)->dport)))== (__ports)) && \ 362 ((__sk)->sk_family == AF_INET6) && \ 363 ipv6_addr_equal(&inet6_sk(__sk)->daddr, (__saddr)) && \ 364 ipv6_addr_equal(&inet6_sk(__sk)->rcv_saddr, (__daddr)) && \ 365 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) 366 367 /* These can have wildcards, don't try too hard. */ 368 static __inline__ int tcp_lhashfn(unsigned short num) 369 { 370 return num & (TCP_LHTABLE_SIZE - 1); 371 } 372 373 static __inline__ int tcp_sk_listen_hashfn(struct sock *sk) 374 { 375 return tcp_lhashfn(inet_sk(sk)->num); 376 } 377 378 #define MAX_TCP_HEADER (128 + MAX_HEADER) 379 380 /* 381 * Never offer a window over 32767 without using window scaling. Some 382 * poor stacks do signed 16bit maths! 383 */ 384 #define MAX_TCP_WINDOW 32767U 385 386 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 387 #define TCP_MIN_MSS 88U 388 389 /* Minimal RCV_MSS. */ 390 #define TCP_MIN_RCVMSS 536U 391 392 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 393 #define TCP_FASTRETRANS_THRESH 3 394 395 /* Maximal reordering. */ 396 #define TCP_MAX_REORDERING 127 397 398 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 399 #define TCP_MAX_QUICKACKS 16U 400 401 /* urg_data states */ 402 #define TCP_URG_VALID 0x0100 403 #define TCP_URG_NOTYET 0x0200 404 #define TCP_URG_READ 0x0400 405 406 #define TCP_RETR1 3 /* 407 * This is how many retries it does before it 408 * tries to figure out if the gateway is 409 * down. Minimal RFC value is 3; it corresponds 410 * to ~3sec-8min depending on RTO. 411 */ 412 413 #define TCP_RETR2 15 /* 414 * This should take at least 415 * 90 minutes to time out. 416 * RFC1122 says that the limit is 100 sec. 417 * 15 is ~13-30min depending on RTO. 418 */ 419 420 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a 421 * connection: ~180sec is RFC minumum */ 422 423 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a 424 * connection: ~180sec is RFC minumum */ 425 426 427 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned 428 * socket. 7 is ~50sec-16min. 429 */ 430 431 432 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 433 * state, about 60 seconds */ 434 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 435 /* BSD style FIN_WAIT2 deadlock breaker. 436 * It used to be 3min, new value is 60sec, 437 * to combine FIN-WAIT-2 timeout with 438 * TIME-WAIT timer. 439 */ 440 441 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 442 #if HZ >= 100 443 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 444 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 445 #else 446 #define TCP_DELACK_MIN 4U 447 #define TCP_ATO_MIN 4U 448 #endif 449 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 450 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 451 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */ 452 453 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 454 * for local resources. 455 */ 456 457 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 458 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 459 #define TCP_KEEPALIVE_INTVL (75*HZ) 460 461 #define MAX_TCP_KEEPIDLE 32767 462 #define MAX_TCP_KEEPINTVL 32767 463 #define MAX_TCP_KEEPCNT 127 464 #define MAX_TCP_SYNCNT 127 465 466 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 467 #define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */ 468 469 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 470 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 471 * after this time. It should be equal 472 * (or greater than) TCP_TIMEWAIT_LEN 473 * to provide reliability equal to one 474 * provided by timewait state. 475 */ 476 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 477 * timestamps. It must be less than 478 * minimal timewait lifetime. 479 */ 480 481 #define TCP_TW_RECYCLE_SLOTS_LOG 5 482 #define TCP_TW_RECYCLE_SLOTS (1<<TCP_TW_RECYCLE_SLOTS_LOG) 483 484 /* If time > 4sec, it is "slow" path, no recycling is required, 485 so that we select tick to get range about 4 seconds. 486 */ 487 488 #if HZ <= 16 || HZ > 4096 489 # error Unsupported: HZ <= 16 or HZ > 4096 490 #elif HZ <= 32 491 # define TCP_TW_RECYCLE_TICK (5+2-TCP_TW_RECYCLE_SLOTS_LOG) 492 #elif HZ <= 64 493 # define TCP_TW_RECYCLE_TICK (6+2-TCP_TW_RECYCLE_SLOTS_LOG) 494 #elif HZ <= 128 495 # define TCP_TW_RECYCLE_TICK (7+2-TCP_TW_RECYCLE_SLOTS_LOG) 496 #elif HZ <= 256 497 # define TCP_TW_RECYCLE_TICK (8+2-TCP_TW_RECYCLE_SLOTS_LOG) 498 #elif HZ <= 512 499 # define TCP_TW_RECYCLE_TICK (9+2-TCP_TW_RECYCLE_SLOTS_LOG) 500 #elif HZ <= 1024 501 # define TCP_TW_RECYCLE_TICK (10+2-TCP_TW_RECYCLE_SLOTS_LOG) 502 #elif HZ <= 2048 503 # define TCP_TW_RECYCLE_TICK (11+2-TCP_TW_RECYCLE_SLOTS_LOG) 504 #else 505 # define TCP_TW_RECYCLE_TICK (12+2-TCP_TW_RECYCLE_SLOTS_LOG) 506 #endif 507 508 #define BICTCP_BETA_SCALE 1024 /* Scale factor beta calculation 509 * max_cwnd = snd_cwnd * beta 510 */ 511 #define BICTCP_MAX_INCREMENT 32 /* 512 * Limit on the amount of 513 * increment allowed during 514 * binary search. 515 */ 516 #define BICTCP_FUNC_OF_MIN_INCR 11 /* 517 * log(B/Smin)/log(B/(B-1))+1, 518 * Smin:min increment 519 * B:log factor 520 */ 521 #define BICTCP_B 4 /* 522 * In binary search, 523 * go to point (max+min)/N 524 */ 525 526 /* 527 * TCP option 528 */ 529 530 #define TCPOPT_NOP 1 /* Padding */ 531 #define TCPOPT_EOL 0 /* End of options */ 532 #define TCPOPT_MSS 2 /* Segment size negotiating */ 533 #define TCPOPT_WINDOW 3 /* Window scaling */ 534 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 535 #define TCPOPT_SACK 5 /* SACK Block */ 536 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 537 538 /* 539 * TCP option lengths 540 */ 541 542 #define TCPOLEN_MSS 4 543 #define TCPOLEN_WINDOW 3 544 #define TCPOLEN_SACK_PERM 2 545 #define TCPOLEN_TIMESTAMP 10 546 547 /* But this is what stacks really send out. */ 548 #define TCPOLEN_TSTAMP_ALIGNED 12 549 #define TCPOLEN_WSCALE_ALIGNED 4 550 #define TCPOLEN_SACKPERM_ALIGNED 4 551 #define TCPOLEN_SACK_BASE 2 552 #define TCPOLEN_SACK_BASE_ALIGNED 4 553 #define TCPOLEN_SACK_PERBLOCK 8 554 555 #define TCP_TIME_RETRANS 1 /* Retransmit timer */ 556 #define TCP_TIME_DACK 2 /* Delayed ack timer */ 557 #define TCP_TIME_PROBE0 3 /* Zero window probe timer */ 558 #define TCP_TIME_KEEPOPEN 4 /* Keepalive timer */ 559 560 /* Flags in tp->nonagle */ 561 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 562 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 563 #define TCP_NAGLE_PUSH 4 /* Cork is overriden for already queued data */ 564 565 /* sysctl variables for tcp */ 566 extern int sysctl_max_syn_backlog; 567 extern int sysctl_tcp_timestamps; 568 extern int sysctl_tcp_window_scaling; 569 extern int sysctl_tcp_sack; 570 extern int sysctl_tcp_fin_timeout; 571 extern int sysctl_tcp_tw_recycle; 572 extern int sysctl_tcp_keepalive_time; 573 extern int sysctl_tcp_keepalive_probes; 574 extern int sysctl_tcp_keepalive_intvl; 575 extern int sysctl_tcp_syn_retries; 576 extern int sysctl_tcp_synack_retries; 577 extern int sysctl_tcp_retries1; 578 extern int sysctl_tcp_retries2; 579 extern int sysctl_tcp_orphan_retries; 580 extern int sysctl_tcp_syncookies; 581 extern int sysctl_tcp_retrans_collapse; 582 extern int sysctl_tcp_stdurg; 583 extern int sysctl_tcp_rfc1337; 584 extern int sysctl_tcp_abort_on_overflow; 585 extern int sysctl_tcp_max_orphans; 586 extern int sysctl_tcp_max_tw_buckets; 587 extern int sysctl_tcp_fack; 588 extern int sysctl_tcp_reordering; 589 extern int sysctl_tcp_ecn; 590 extern int sysctl_tcp_dsack; 591 extern int sysctl_tcp_mem[3]; 592 extern int sysctl_tcp_wmem[3]; 593 extern int sysctl_tcp_rmem[3]; 594 extern int sysctl_tcp_app_win; 595 extern int sysctl_tcp_adv_win_scale; 596 extern int sysctl_tcp_tw_reuse; 597 extern int sysctl_tcp_frto; 598 extern int sysctl_tcp_low_latency; 599 extern int sysctl_tcp_westwood; 600 extern int sysctl_tcp_vegas_cong_avoid; 601 extern int sysctl_tcp_vegas_alpha; 602 extern int sysctl_tcp_vegas_beta; 603 extern int sysctl_tcp_vegas_gamma; 604 extern int sysctl_tcp_nometrics_save; 605 extern int sysctl_tcp_bic; 606 extern int sysctl_tcp_bic_fast_convergence; 607 extern int sysctl_tcp_bic_low_window; 608 extern int sysctl_tcp_bic_beta; 609 extern int sysctl_tcp_moderate_rcvbuf; 610 extern int sysctl_tcp_tso_win_divisor; 611 612 extern atomic_t tcp_memory_allocated; 613 extern atomic_t tcp_sockets_allocated; 614 extern int tcp_memory_pressure; 615 616 struct open_request; 617 618 struct or_calltable { 619 int family; 620 int (*rtx_syn_ack) (struct sock *sk, struct open_request *req, struct dst_entry*); 621 void (*send_ack) (struct sk_buff *skb, struct open_request *req); 622 void (*destructor) (struct open_request *req); 623 void (*send_reset) (struct sk_buff *skb); 624 }; 625 626 struct tcp_v4_open_req { 627 __u32 loc_addr; 628 __u32 rmt_addr; 629 struct ip_options *opt; 630 }; 631 632 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE) 633 struct tcp_v6_open_req { 634 struct in6_addr loc_addr; 635 struct in6_addr rmt_addr; 636 struct sk_buff *pktopts; 637 int iif; 638 }; 639 #endif 640 641 /* this structure is too big */ 642 struct open_request { 643 struct open_request *dl_next; /* Must be first member! */ 644 __u32 rcv_isn; 645 __u32 snt_isn; 646 __u16 rmt_port; 647 __u16 mss; 648 __u8 retrans; 649 __u8 __pad; 650 __u16 snd_wscale : 4, 651 rcv_wscale : 4, 652 tstamp_ok : 1, 653 sack_ok : 1, 654 wscale_ok : 1, 655 ecn_ok : 1, 656 acked : 1; 657 /* The following two fields can be easily recomputed I think -AK */ 658 __u32 window_clamp; /* window clamp at creation time */ 659 __u32 rcv_wnd; /* rcv_wnd offered first time */ 660 __u32 ts_recent; 661 unsigned long expires; 662 struct or_calltable *class; 663 struct sock *sk; 664 union { 665 struct tcp_v4_open_req v4_req; 666 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE) 667 struct tcp_v6_open_req v6_req; 668 #endif 669 } af; 670 }; 671 672 /* SLAB cache for open requests. */ 673 extern kmem_cache_t *tcp_openreq_cachep; 674 675 #define tcp_openreq_alloc() kmem_cache_alloc(tcp_openreq_cachep, SLAB_ATOMIC) 676 #define tcp_openreq_fastfree(req) kmem_cache_free(tcp_openreq_cachep, req) 677 678 static inline void tcp_openreq_free(struct open_request *req) 679 { 680 req->class->destructor(req); 681 tcp_openreq_fastfree(req); 682 } 683 684 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 685 #define TCP_INET_FAMILY(fam) ((fam) == AF_INET) 686 #else 687 #define TCP_INET_FAMILY(fam) 1 688 #endif 689 690 /* 691 * Pointers to address related TCP functions 692 * (i.e. things that depend on the address family) 693 */ 694 695 struct tcp_func { 696 int (*queue_xmit) (struct sk_buff *skb, 697 int ipfragok); 698 699 void (*send_check) (struct sock *sk, 700 struct tcphdr *th, 701 int len, 702 struct sk_buff *skb); 703 704 int (*rebuild_header) (struct sock *sk); 705 706 int (*conn_request) (struct sock *sk, 707 struct sk_buff *skb); 708 709 struct sock * (*syn_recv_sock) (struct sock *sk, 710 struct sk_buff *skb, 711 struct open_request *req, 712 struct dst_entry *dst); 713 714 int (*remember_stamp) (struct sock *sk); 715 716 __u16 net_header_len; 717 718 int (*setsockopt) (struct sock *sk, 719 int level, 720 int optname, 721 char __user *optval, 722 int optlen); 723 724 int (*getsockopt) (struct sock *sk, 725 int level, 726 int optname, 727 char __user *optval, 728 int __user *optlen); 729 730 731 void (*addr2sockaddr) (struct sock *sk, 732 struct sockaddr *); 733 734 int sockaddr_len; 735 }; 736 737 /* 738 * The next routines deal with comparing 32 bit unsigned ints 739 * and worry about wraparound (automatic with unsigned arithmetic). 740 */ 741 742 static inline int before(__u32 seq1, __u32 seq2) 743 { 744 return (__s32)(seq1-seq2) < 0; 745 } 746 747 static inline int after(__u32 seq1, __u32 seq2) 748 { 749 return (__s32)(seq2-seq1) < 0; 750 } 751 752 753 /* is s2<=s1<=s3 ? */ 754 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3) 755 { 756 return seq3 - seq2 >= seq1 - seq2; 757 } 758 759 760 extern struct proto tcp_prot; 761 762 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics); 763 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field) 764 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field) 765 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field) 766 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field) 767 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val) 768 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val) 769 770 extern void tcp_put_port(struct sock *sk); 771 extern void tcp_inherit_port(struct sock *sk, struct sock *child); 772 773 extern void tcp_v4_err(struct sk_buff *skb, u32); 774 775 extern void tcp_shutdown (struct sock *sk, int how); 776 777 extern int tcp_v4_rcv(struct sk_buff *skb); 778 779 extern int tcp_v4_remember_stamp(struct sock *sk); 780 781 extern int tcp_v4_tw_remember_stamp(struct tcp_tw_bucket *tw); 782 783 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, 784 struct msghdr *msg, size_t size); 785 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags); 786 787 extern int tcp_ioctl(struct sock *sk, 788 int cmd, 789 unsigned long arg); 790 791 extern int tcp_rcv_state_process(struct sock *sk, 792 struct sk_buff *skb, 793 struct tcphdr *th, 794 unsigned len); 795 796 extern int tcp_rcv_established(struct sock *sk, 797 struct sk_buff *skb, 798 struct tcphdr *th, 799 unsigned len); 800 801 extern void tcp_rcv_space_adjust(struct sock *sk); 802 803 enum tcp_ack_state_t 804 { 805 TCP_ACK_SCHED = 1, 806 TCP_ACK_TIMER = 2, 807 TCP_ACK_PUSHED= 4 808 }; 809 810 static inline void tcp_schedule_ack(struct tcp_sock *tp) 811 { 812 tp->ack.pending |= TCP_ACK_SCHED; 813 } 814 815 static inline int tcp_ack_scheduled(struct tcp_sock *tp) 816 { 817 return tp->ack.pending&TCP_ACK_SCHED; 818 } 819 820 static __inline__ void tcp_dec_quickack_mode(struct tcp_sock *tp) 821 { 822 if (tp->ack.quick && --tp->ack.quick == 0) { 823 /* Leaving quickack mode we deflate ATO. */ 824 tp->ack.ato = TCP_ATO_MIN; 825 } 826 } 827 828 extern void tcp_enter_quickack_mode(struct tcp_sock *tp); 829 830 static __inline__ void tcp_delack_init(struct tcp_sock *tp) 831 { 832 memset(&tp->ack, 0, sizeof(tp->ack)); 833 } 834 835 static inline void tcp_clear_options(struct tcp_options_received *rx_opt) 836 { 837 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0; 838 } 839 840 enum tcp_tw_status 841 { 842 TCP_TW_SUCCESS = 0, 843 TCP_TW_RST = 1, 844 TCP_TW_ACK = 2, 845 TCP_TW_SYN = 3 846 }; 847 848 849 extern enum tcp_tw_status tcp_timewait_state_process(struct tcp_tw_bucket *tw, 850 struct sk_buff *skb, 851 struct tcphdr *th, 852 unsigned len); 853 854 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb, 855 struct open_request *req, 856 struct open_request **prev); 857 extern int tcp_child_process(struct sock *parent, 858 struct sock *child, 859 struct sk_buff *skb); 860 extern void tcp_enter_frto(struct sock *sk); 861 extern void tcp_enter_loss(struct sock *sk, int how); 862 extern void tcp_clear_retrans(struct tcp_sock *tp); 863 extern void tcp_update_metrics(struct sock *sk); 864 865 extern void tcp_close(struct sock *sk, 866 long timeout); 867 extern struct sock * tcp_accept(struct sock *sk, int flags, int *err); 868 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait); 869 870 extern int tcp_getsockopt(struct sock *sk, int level, 871 int optname, 872 char __user *optval, 873 int __user *optlen); 874 extern int tcp_setsockopt(struct sock *sk, int level, 875 int optname, char __user *optval, 876 int optlen); 877 extern void tcp_set_keepalive(struct sock *sk, int val); 878 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, 879 struct msghdr *msg, 880 size_t len, int nonblock, 881 int flags, int *addr_len); 882 883 extern int tcp_listen_start(struct sock *sk); 884 885 extern void tcp_parse_options(struct sk_buff *skb, 886 struct tcp_options_received *opt_rx, 887 int estab); 888 889 /* 890 * TCP v4 functions exported for the inet6 API 891 */ 892 893 extern int tcp_v4_rebuild_header(struct sock *sk); 894 895 extern int tcp_v4_build_header(struct sock *sk, 896 struct sk_buff *skb); 897 898 extern void tcp_v4_send_check(struct sock *sk, 899 struct tcphdr *th, int len, 900 struct sk_buff *skb); 901 902 extern int tcp_v4_conn_request(struct sock *sk, 903 struct sk_buff *skb); 904 905 extern struct sock * tcp_create_openreq_child(struct sock *sk, 906 struct open_request *req, 907 struct sk_buff *skb); 908 909 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, 910 struct sk_buff *skb, 911 struct open_request *req, 912 struct dst_entry *dst); 913 914 extern int tcp_v4_do_rcv(struct sock *sk, 915 struct sk_buff *skb); 916 917 extern int tcp_v4_connect(struct sock *sk, 918 struct sockaddr *uaddr, 919 int addr_len); 920 921 extern int tcp_connect(struct sock *sk); 922 923 extern struct sk_buff * tcp_make_synack(struct sock *sk, 924 struct dst_entry *dst, 925 struct open_request *req); 926 927 extern int tcp_disconnect(struct sock *sk, int flags); 928 929 extern void tcp_unhash(struct sock *sk); 930 931 extern int tcp_v4_hash_connecting(struct sock *sk); 932 933 934 /* From syncookies.c */ 935 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 936 struct ip_options *opt); 937 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, 938 __u16 *mss); 939 940 /* tcp_output.c */ 941 942 extern int tcp_write_xmit(struct sock *, int nonagle); 943 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); 944 extern void tcp_xmit_retransmit_queue(struct sock *); 945 extern void tcp_simple_retransmit(struct sock *); 946 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32); 947 948 extern void tcp_send_probe0(struct sock *); 949 extern void tcp_send_partial(struct sock *); 950 extern int tcp_write_wakeup(struct sock *); 951 extern void tcp_send_fin(struct sock *sk); 952 extern void tcp_send_active_reset(struct sock *sk, int priority); 953 extern int tcp_send_synack(struct sock *); 954 extern void tcp_push_one(struct sock *, unsigned mss_now); 955 extern void tcp_send_ack(struct sock *sk); 956 extern void tcp_send_delayed_ack(struct sock *sk); 957 958 /* tcp_timer.c */ 959 extern void tcp_init_xmit_timers(struct sock *); 960 extern void tcp_clear_xmit_timers(struct sock *); 961 962 extern void tcp_delete_keepalive_timer(struct sock *); 963 extern void tcp_reset_keepalive_timer(struct sock *, unsigned long); 964 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 965 extern unsigned int tcp_current_mss(struct sock *sk, int large); 966 967 #ifdef TCP_DEBUG 968 extern const char tcp_timer_bug_msg[]; 969 #endif 970 971 /* tcp_diag.c */ 972 extern void tcp_get_info(struct sock *, struct tcp_info *); 973 974 /* Read 'sendfile()'-style from a TCP socket */ 975 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 976 unsigned int, size_t); 977 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 978 sk_read_actor_t recv_actor); 979 980 static inline void tcp_clear_xmit_timer(struct sock *sk, int what) 981 { 982 struct tcp_sock *tp = tcp_sk(sk); 983 984 switch (what) { 985 case TCP_TIME_RETRANS: 986 case TCP_TIME_PROBE0: 987 tp->pending = 0; 988 989 #ifdef TCP_CLEAR_TIMERS 990 sk_stop_timer(sk, &tp->retransmit_timer); 991 #endif 992 break; 993 case TCP_TIME_DACK: 994 tp->ack.blocked = 0; 995 tp->ack.pending = 0; 996 997 #ifdef TCP_CLEAR_TIMERS 998 sk_stop_timer(sk, &tp->delack_timer); 999 #endif 1000 break; 1001 default: 1002 #ifdef TCP_DEBUG 1003 printk(tcp_timer_bug_msg); 1004 #endif 1005 return; 1006 }; 1007 1008 } 1009 1010 /* 1011 * Reset the retransmission timer 1012 */ 1013 static inline void tcp_reset_xmit_timer(struct sock *sk, int what, unsigned long when) 1014 { 1015 struct tcp_sock *tp = tcp_sk(sk); 1016 1017 if (when > TCP_RTO_MAX) { 1018 #ifdef TCP_DEBUG 1019 printk(KERN_DEBUG "reset_xmit_timer sk=%p %d when=0x%lx, caller=%p\n", sk, what, when, current_text_addr()); 1020 #endif 1021 when = TCP_RTO_MAX; 1022 } 1023 1024 switch (what) { 1025 case TCP_TIME_RETRANS: 1026 case TCP_TIME_PROBE0: 1027 tp->pending = what; 1028 tp->timeout = jiffies+when; 1029 sk_reset_timer(sk, &tp->retransmit_timer, tp->timeout); 1030 break; 1031 1032 case TCP_TIME_DACK: 1033 tp->ack.pending |= TCP_ACK_TIMER; 1034 tp->ack.timeout = jiffies+when; 1035 sk_reset_timer(sk, &tp->delack_timer, tp->ack.timeout); 1036 break; 1037 1038 default: 1039 #ifdef TCP_DEBUG 1040 printk(tcp_timer_bug_msg); 1041 #endif 1042 return; 1043 }; 1044 } 1045 1046 /* Initialize RCV_MSS value. 1047 * RCV_MSS is an our guess about MSS used by the peer. 1048 * We haven't any direct information about the MSS. 1049 * It's better to underestimate the RCV_MSS rather than overestimate. 1050 * Overestimations make us ACKing less frequently than needed. 1051 * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss(). 1052 */ 1053 1054 static inline void tcp_initialize_rcv_mss(struct sock *sk) 1055 { 1056 struct tcp_sock *tp = tcp_sk(sk); 1057 unsigned int hint = min(tp->advmss, tp->mss_cache_std); 1058 1059 hint = min(hint, tp->rcv_wnd/2); 1060 hint = min(hint, TCP_MIN_RCVMSS); 1061 hint = max(hint, TCP_MIN_MSS); 1062 1063 tp->ack.rcv_mss = hint; 1064 } 1065 1066 static __inline__ void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 1067 { 1068 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 1069 ntohl(TCP_FLAG_ACK) | 1070 snd_wnd); 1071 } 1072 1073 static __inline__ void tcp_fast_path_on(struct tcp_sock *tp) 1074 { 1075 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 1076 } 1077 1078 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp) 1079 { 1080 if (skb_queue_len(&tp->out_of_order_queue) == 0 && 1081 tp->rcv_wnd && 1082 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 1083 !tp->urg_data) 1084 tcp_fast_path_on(tp); 1085 } 1086 1087 /* Compute the actual receive window we are currently advertising. 1088 * Rcv_nxt can be after the window if our peer push more data 1089 * than the offered window. 1090 */ 1091 static __inline__ u32 tcp_receive_window(const struct tcp_sock *tp) 1092 { 1093 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 1094 1095 if (win < 0) 1096 win = 0; 1097 return (u32) win; 1098 } 1099 1100 /* Choose a new window, without checks for shrinking, and without 1101 * scaling applied to the result. The caller does these things 1102 * if necessary. This is a "raw" window selection. 1103 */ 1104 extern u32 __tcp_select_window(struct sock *sk); 1105 1106 /* TCP timestamps are only 32-bits, this causes a slight 1107 * complication on 64-bit systems since we store a snapshot 1108 * of jiffies in the buffer control blocks below. We decidely 1109 * only use of the low 32-bits of jiffies and hide the ugly 1110 * casts with the following macro. 1111 */ 1112 #define tcp_time_stamp ((__u32)(jiffies)) 1113 1114 /* This is what the send packet queueing engine uses to pass 1115 * TCP per-packet control information to the transmission 1116 * code. We also store the host-order sequence numbers in 1117 * here too. This is 36 bytes on 32-bit architectures, 1118 * 40 bytes on 64-bit machines, if this grows please adjust 1119 * skbuff.h:skbuff->cb[xxx] size appropriately. 1120 */ 1121 struct tcp_skb_cb { 1122 union { 1123 struct inet_skb_parm h4; 1124 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE) 1125 struct inet6_skb_parm h6; 1126 #endif 1127 } header; /* For incoming frames */ 1128 __u32 seq; /* Starting sequence number */ 1129 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 1130 __u32 when; /* used to compute rtt's */ 1131 __u8 flags; /* TCP header flags. */ 1132 1133 /* NOTE: These must match up to the flags byte in a 1134 * real TCP header. 1135 */ 1136 #define TCPCB_FLAG_FIN 0x01 1137 #define TCPCB_FLAG_SYN 0x02 1138 #define TCPCB_FLAG_RST 0x04 1139 #define TCPCB_FLAG_PSH 0x08 1140 #define TCPCB_FLAG_ACK 0x10 1141 #define TCPCB_FLAG_URG 0x20 1142 #define TCPCB_FLAG_ECE 0x40 1143 #define TCPCB_FLAG_CWR 0x80 1144 1145 __u8 sacked; /* State flags for SACK/FACK. */ 1146 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 1147 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 1148 #define TCPCB_LOST 0x04 /* SKB is lost */ 1149 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 1150 1151 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 1152 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 1153 1154 #define TCPCB_URG 0x20 /* Urgent pointer advenced here */ 1155 1156 #define TCPCB_AT_TAIL (TCPCB_URG) 1157 1158 __u16 urg_ptr; /* Valid w/URG flags is set. */ 1159 __u32 ack_seq; /* Sequence number ACK'd */ 1160 }; 1161 1162 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 1163 1164 #include <net/tcp_ecn.h> 1165 1166 /* Due to TSO, an SKB can be composed of multiple actual 1167 * packets. To keep these tracked properly, we use this. 1168 */ 1169 static inline int tcp_skb_pcount(const struct sk_buff *skb) 1170 { 1171 return skb_shinfo(skb)->tso_segs; 1172 } 1173 1174 /* This is valid iff tcp_skb_pcount() > 1. */ 1175 static inline int tcp_skb_mss(const struct sk_buff *skb) 1176 { 1177 return skb_shinfo(skb)->tso_size; 1178 } 1179 1180 static inline void tcp_dec_pcount_approx(__u32 *count, 1181 const struct sk_buff *skb) 1182 { 1183 if (*count) { 1184 *count -= tcp_skb_pcount(skb); 1185 if ((int)*count < 0) 1186 *count = 0; 1187 } 1188 } 1189 1190 static inline void tcp_packets_out_inc(struct sock *sk, 1191 struct tcp_sock *tp, 1192 const struct sk_buff *skb) 1193 { 1194 int orig = tp->packets_out; 1195 1196 tp->packets_out += tcp_skb_pcount(skb); 1197 if (!orig) 1198 tcp_reset_xmit_timer(sk, TCP_TIME_RETRANS, tp->rto); 1199 } 1200 1201 static inline void tcp_packets_out_dec(struct tcp_sock *tp, 1202 const struct sk_buff *skb) 1203 { 1204 tp->packets_out -= tcp_skb_pcount(skb); 1205 } 1206 1207 /* This determines how many packets are "in the network" to the best 1208 * of our knowledge. In many cases it is conservative, but where 1209 * detailed information is available from the receiver (via SACK 1210 * blocks etc.) we can make more aggressive calculations. 1211 * 1212 * Use this for decisions involving congestion control, use just 1213 * tp->packets_out to determine if the send queue is empty or not. 1214 * 1215 * Read this equation as: 1216 * 1217 * "Packets sent once on transmission queue" MINUS 1218 * "Packets left network, but not honestly ACKed yet" PLUS 1219 * "Packets fast retransmitted" 1220 */ 1221 static __inline__ unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1222 { 1223 return (tp->packets_out - tp->left_out + tp->retrans_out); 1224 } 1225 1226 /* 1227 * Which congestion algorithim is in use on the connection. 1228 */ 1229 #define tcp_is_vegas(__tp) ((__tp)->adv_cong == TCP_VEGAS) 1230 #define tcp_is_westwood(__tp) ((__tp)->adv_cong == TCP_WESTWOOD) 1231 #define tcp_is_bic(__tp) ((__tp)->adv_cong == TCP_BIC) 1232 1233 /* Recalculate snd_ssthresh, we want to set it to: 1234 * 1235 * Reno: 1236 * one half the current congestion window, but no 1237 * less than two segments 1238 * 1239 * BIC: 1240 * behave like Reno until low_window is reached, 1241 * then increase congestion window slowly 1242 */ 1243 static inline __u32 tcp_recalc_ssthresh(struct tcp_sock *tp) 1244 { 1245 if (tcp_is_bic(tp)) { 1246 if (sysctl_tcp_bic_fast_convergence && 1247 tp->snd_cwnd < tp->bictcp.last_max_cwnd) 1248 tp->bictcp.last_max_cwnd = (tp->snd_cwnd * 1249 (BICTCP_BETA_SCALE 1250 + sysctl_tcp_bic_beta)) 1251 / (2 * BICTCP_BETA_SCALE); 1252 else 1253 tp->bictcp.last_max_cwnd = tp->snd_cwnd; 1254 1255 if (tp->snd_cwnd > sysctl_tcp_bic_low_window) 1256 return max((tp->snd_cwnd * sysctl_tcp_bic_beta) 1257 / BICTCP_BETA_SCALE, 2U); 1258 } 1259 1260 return max(tp->snd_cwnd >> 1U, 2U); 1261 } 1262 1263 /* Stop taking Vegas samples for now. */ 1264 #define tcp_vegas_disable(__tp) ((__tp)->vegas.doing_vegas_now = 0) 1265 1266 static inline void tcp_vegas_enable(struct tcp_sock *tp) 1267 { 1268 /* There are several situations when we must "re-start" Vegas: 1269 * 1270 * o when a connection is established 1271 * o after an RTO 1272 * o after fast recovery 1273 * o when we send a packet and there is no outstanding 1274 * unacknowledged data (restarting an idle connection) 1275 * 1276 * In these circumstances we cannot do a Vegas calculation at the 1277 * end of the first RTT, because any calculation we do is using 1278 * stale info -- both the saved cwnd and congestion feedback are 1279 * stale. 1280 * 1281 * Instead we must wait until the completion of an RTT during 1282 * which we actually receive ACKs. 1283 */ 1284 1285 /* Begin taking Vegas samples next time we send something. */ 1286 tp->vegas.doing_vegas_now = 1; 1287 1288 /* Set the beginning of the next send window. */ 1289 tp->vegas.beg_snd_nxt = tp->snd_nxt; 1290 1291 tp->vegas.cntRTT = 0; 1292 tp->vegas.minRTT = 0x7fffffff; 1293 } 1294 1295 /* Should we be taking Vegas samples right now? */ 1296 #define tcp_vegas_enabled(__tp) ((__tp)->vegas.doing_vegas_now) 1297 1298 extern void tcp_ca_init(struct tcp_sock *tp); 1299 1300 static inline void tcp_set_ca_state(struct tcp_sock *tp, u8 ca_state) 1301 { 1302 if (tcp_is_vegas(tp)) { 1303 if (ca_state == TCP_CA_Open) 1304 tcp_vegas_enable(tp); 1305 else 1306 tcp_vegas_disable(tp); 1307 } 1308 tp->ca_state = ca_state; 1309 } 1310 1311 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1312 * The exception is rate halving phase, when cwnd is decreasing towards 1313 * ssthresh. 1314 */ 1315 static inline __u32 tcp_current_ssthresh(struct tcp_sock *tp) 1316 { 1317 if ((1<<tp->ca_state)&(TCPF_CA_CWR|TCPF_CA_Recovery)) 1318 return tp->snd_ssthresh; 1319 else 1320 return max(tp->snd_ssthresh, 1321 ((tp->snd_cwnd >> 1) + 1322 (tp->snd_cwnd >> 2))); 1323 } 1324 1325 static inline void tcp_sync_left_out(struct tcp_sock *tp) 1326 { 1327 if (tp->rx_opt.sack_ok && 1328 (tp->sacked_out >= tp->packets_out - tp->lost_out)) 1329 tp->sacked_out = tp->packets_out - tp->lost_out; 1330 tp->left_out = tp->sacked_out + tp->lost_out; 1331 } 1332 1333 extern void tcp_cwnd_application_limited(struct sock *sk); 1334 1335 /* Congestion window validation. (RFC2861) */ 1336 1337 static inline void tcp_cwnd_validate(struct sock *sk, struct tcp_sock *tp) 1338 { 1339 __u32 packets_out = tp->packets_out; 1340 1341 if (packets_out >= tp->snd_cwnd) { 1342 /* Network is feed fully. */ 1343 tp->snd_cwnd_used = 0; 1344 tp->snd_cwnd_stamp = tcp_time_stamp; 1345 } else { 1346 /* Network starves. */ 1347 if (tp->packets_out > tp->snd_cwnd_used) 1348 tp->snd_cwnd_used = tp->packets_out; 1349 1350 if ((s32)(tcp_time_stamp - tp->snd_cwnd_stamp) >= tp->rto) 1351 tcp_cwnd_application_limited(sk); 1352 } 1353 } 1354 1355 /* Set slow start threshould and cwnd not falling to slow start */ 1356 static inline void __tcp_enter_cwr(struct tcp_sock *tp) 1357 { 1358 tp->undo_marker = 0; 1359 tp->snd_ssthresh = tcp_recalc_ssthresh(tp); 1360 tp->snd_cwnd = min(tp->snd_cwnd, 1361 tcp_packets_in_flight(tp) + 1U); 1362 tp->snd_cwnd_cnt = 0; 1363 tp->high_seq = tp->snd_nxt; 1364 tp->snd_cwnd_stamp = tcp_time_stamp; 1365 TCP_ECN_queue_cwr(tp); 1366 } 1367 1368 static inline void tcp_enter_cwr(struct tcp_sock *tp) 1369 { 1370 tp->prior_ssthresh = 0; 1371 if (tp->ca_state < TCP_CA_CWR) { 1372 __tcp_enter_cwr(tp); 1373 tcp_set_ca_state(tp, TCP_CA_CWR); 1374 } 1375 } 1376 1377 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst); 1378 1379 /* Slow start with delack produces 3 packets of burst, so that 1380 * it is safe "de facto". 1381 */ 1382 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 1383 { 1384 return 3; 1385 } 1386 1387 static __inline__ int tcp_minshall_check(const struct tcp_sock *tp) 1388 { 1389 return after(tp->snd_sml,tp->snd_una) && 1390 !after(tp->snd_sml, tp->snd_nxt); 1391 } 1392 1393 static __inline__ void tcp_minshall_update(struct tcp_sock *tp, int mss, 1394 const struct sk_buff *skb) 1395 { 1396 if (skb->len < mss) 1397 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 1398 } 1399 1400 /* Return 0, if packet can be sent now without violation Nagle's rules: 1401 1. It is full sized. 1402 2. Or it contains FIN. 1403 3. Or TCP_NODELAY was set. 1404 4. Or TCP_CORK is not set, and all sent packets are ACKed. 1405 With Minshall's modification: all sent small packets are ACKed. 1406 */ 1407 1408 static __inline__ int 1409 tcp_nagle_check(const struct tcp_sock *tp, const struct sk_buff *skb, 1410 unsigned mss_now, int nonagle) 1411 { 1412 return (skb->len < mss_now && 1413 !(TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN) && 1414 ((nonagle&TCP_NAGLE_CORK) || 1415 (!nonagle && 1416 tp->packets_out && 1417 tcp_minshall_check(tp)))); 1418 } 1419 1420 extern void tcp_set_skb_tso_segs(struct sk_buff *, unsigned int); 1421 1422 /* This checks if the data bearing packet SKB (usually sk->sk_send_head) 1423 * should be put on the wire right now. 1424 */ 1425 static __inline__ int tcp_snd_test(const struct tcp_sock *tp, 1426 struct sk_buff *skb, 1427 unsigned cur_mss, int nonagle) 1428 { 1429 int pkts = tcp_skb_pcount(skb); 1430 1431 if (!pkts) { 1432 tcp_set_skb_tso_segs(skb, tp->mss_cache_std); 1433 pkts = tcp_skb_pcount(skb); 1434 } 1435 1436 /* RFC 1122 - section 4.2.3.4 1437 * 1438 * We must queue if 1439 * 1440 * a) The right edge of this frame exceeds the window 1441 * b) There are packets in flight and we have a small segment 1442 * [SWS avoidance and Nagle algorithm] 1443 * (part of SWS is done on packetization) 1444 * Minshall version sounds: there are no _small_ 1445 * segments in flight. (tcp_nagle_check) 1446 * c) We have too many packets 'in flight' 1447 * 1448 * Don't use the nagle rule for urgent data (or 1449 * for the final FIN -DaveM). 1450 * 1451 * Also, Nagle rule does not apply to frames, which 1452 * sit in the middle of queue (they have no chances 1453 * to get new data) and if room at tail of skb is 1454 * not enough to save something seriously (<32 for now). 1455 */ 1456 1457 /* Don't be strict about the congestion window for the 1458 * final FIN frame. -DaveM 1459 */ 1460 return (((nonagle&TCP_NAGLE_PUSH) || tp->urg_mode 1461 || !tcp_nagle_check(tp, skb, cur_mss, nonagle)) && 1462 (((tcp_packets_in_flight(tp) + (pkts-1)) < tp->snd_cwnd) || 1463 (TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN)) && 1464 !after(TCP_SKB_CB(skb)->end_seq, tp->snd_una + tp->snd_wnd)); 1465 } 1466 1467 static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp) 1468 { 1469 if (!tp->packets_out && !tp->pending) 1470 tcp_reset_xmit_timer(sk, TCP_TIME_PROBE0, tp->rto); 1471 } 1472 1473 static __inline__ int tcp_skb_is_last(const struct sock *sk, 1474 const struct sk_buff *skb) 1475 { 1476 return skb->next == (struct sk_buff *)&sk->sk_write_queue; 1477 } 1478 1479 /* Push out any pending frames which were held back due to 1480 * TCP_CORK or attempt at coalescing tiny packets. 1481 * The socket must be locked by the caller. 1482 */ 1483 static __inline__ void __tcp_push_pending_frames(struct sock *sk, 1484 struct tcp_sock *tp, 1485 unsigned cur_mss, 1486 int nonagle) 1487 { 1488 struct sk_buff *skb = sk->sk_send_head; 1489 1490 if (skb) { 1491 if (!tcp_skb_is_last(sk, skb)) 1492 nonagle = TCP_NAGLE_PUSH; 1493 if (!tcp_snd_test(tp, skb, cur_mss, nonagle) || 1494 tcp_write_xmit(sk, nonagle)) 1495 tcp_check_probe_timer(sk, tp); 1496 } 1497 tcp_cwnd_validate(sk, tp); 1498 } 1499 1500 static __inline__ void tcp_push_pending_frames(struct sock *sk, 1501 struct tcp_sock *tp) 1502 { 1503 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle); 1504 } 1505 1506 static __inline__ int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp) 1507 { 1508 struct sk_buff *skb = sk->sk_send_head; 1509 1510 return (skb && 1511 tcp_snd_test(tp, skb, tcp_current_mss(sk, 1), 1512 tcp_skb_is_last(sk, skb) ? TCP_NAGLE_PUSH : tp->nonagle)); 1513 } 1514 1515 static __inline__ void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq) 1516 { 1517 tp->snd_wl1 = seq; 1518 } 1519 1520 static __inline__ void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq) 1521 { 1522 tp->snd_wl1 = seq; 1523 } 1524 1525 extern void tcp_destroy_sock(struct sock *sk); 1526 1527 1528 /* 1529 * Calculate(/check) TCP checksum 1530 */ 1531 static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len, 1532 unsigned long saddr, unsigned long daddr, 1533 unsigned long base) 1534 { 1535 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1536 } 1537 1538 static __inline__ int __tcp_checksum_complete(struct sk_buff *skb) 1539 { 1540 return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum)); 1541 } 1542 1543 static __inline__ int tcp_checksum_complete(struct sk_buff *skb) 1544 { 1545 return skb->ip_summed != CHECKSUM_UNNECESSARY && 1546 __tcp_checksum_complete(skb); 1547 } 1548 1549 /* Prequeue for VJ style copy to user, combined with checksumming. */ 1550 1551 static __inline__ void tcp_prequeue_init(struct tcp_sock *tp) 1552 { 1553 tp->ucopy.task = NULL; 1554 tp->ucopy.len = 0; 1555 tp->ucopy.memory = 0; 1556 skb_queue_head_init(&tp->ucopy.prequeue); 1557 } 1558 1559 /* Packet is added to VJ-style prequeue for processing in process 1560 * context, if a reader task is waiting. Apparently, this exciting 1561 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) 1562 * failed somewhere. Latency? Burstiness? Well, at least now we will 1563 * see, why it failed. 8)8) --ANK 1564 * 1565 * NOTE: is this not too big to inline? 1566 */ 1567 static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb) 1568 { 1569 struct tcp_sock *tp = tcp_sk(sk); 1570 1571 if (!sysctl_tcp_low_latency && tp->ucopy.task) { 1572 __skb_queue_tail(&tp->ucopy.prequeue, skb); 1573 tp->ucopy.memory += skb->truesize; 1574 if (tp->ucopy.memory > sk->sk_rcvbuf) { 1575 struct sk_buff *skb1; 1576 1577 BUG_ON(sock_owned_by_user(sk)); 1578 1579 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { 1580 sk->sk_backlog_rcv(sk, skb1); 1581 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED); 1582 } 1583 1584 tp->ucopy.memory = 0; 1585 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { 1586 wake_up_interruptible(sk->sk_sleep); 1587 if (!tcp_ack_scheduled(tp)) 1588 tcp_reset_xmit_timer(sk, TCP_TIME_DACK, (3*TCP_RTO_MIN)/4); 1589 } 1590 return 1; 1591 } 1592 return 0; 1593 } 1594 1595 1596 #undef STATE_TRACE 1597 1598 #ifdef STATE_TRACE 1599 static const char *statename[]={ 1600 "Unused","Established","Syn Sent","Syn Recv", 1601 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1602 "Close Wait","Last ACK","Listen","Closing" 1603 }; 1604 #endif 1605 1606 static __inline__ void tcp_set_state(struct sock *sk, int state) 1607 { 1608 int oldstate = sk->sk_state; 1609 1610 switch (state) { 1611 case TCP_ESTABLISHED: 1612 if (oldstate != TCP_ESTABLISHED) 1613 TCP_INC_STATS(TCP_MIB_CURRESTAB); 1614 break; 1615 1616 case TCP_CLOSE: 1617 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED) 1618 TCP_INC_STATS(TCP_MIB_ESTABRESETS); 1619 1620 sk->sk_prot->unhash(sk); 1621 if (tcp_sk(sk)->bind_hash && 1622 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) 1623 tcp_put_port(sk); 1624 /* fall through */ 1625 default: 1626 if (oldstate==TCP_ESTABLISHED) 1627 TCP_DEC_STATS(TCP_MIB_CURRESTAB); 1628 } 1629 1630 /* Change state AFTER socket is unhashed to avoid closed 1631 * socket sitting in hash tables. 1632 */ 1633 sk->sk_state = state; 1634 1635 #ifdef STATE_TRACE 1636 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]); 1637 #endif 1638 } 1639 1640 static __inline__ void tcp_done(struct sock *sk) 1641 { 1642 tcp_set_state(sk, TCP_CLOSE); 1643 tcp_clear_xmit_timers(sk); 1644 1645 sk->sk_shutdown = SHUTDOWN_MASK; 1646 1647 if (!sock_flag(sk, SOCK_DEAD)) 1648 sk->sk_state_change(sk); 1649 else 1650 tcp_destroy_sock(sk); 1651 } 1652 1653 static __inline__ void tcp_sack_reset(struct tcp_options_received *rx_opt) 1654 { 1655 rx_opt->dsack = 0; 1656 rx_opt->eff_sacks = 0; 1657 rx_opt->num_sacks = 0; 1658 } 1659 1660 static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_sock *tp, __u32 tstamp) 1661 { 1662 if (tp->rx_opt.tstamp_ok) { 1663 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | 1664 (TCPOPT_NOP << 16) | 1665 (TCPOPT_TIMESTAMP << 8) | 1666 TCPOLEN_TIMESTAMP); 1667 *ptr++ = htonl(tstamp); 1668 *ptr++ = htonl(tp->rx_opt.ts_recent); 1669 } 1670 if (tp->rx_opt.eff_sacks) { 1671 struct tcp_sack_block *sp = tp->rx_opt.dsack ? tp->duplicate_sack : tp->selective_acks; 1672 int this_sack; 1673 1674 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | 1675 (TCPOPT_NOP << 16) | 1676 (TCPOPT_SACK << 8) | 1677 (TCPOLEN_SACK_BASE + 1678 (tp->rx_opt.eff_sacks * TCPOLEN_SACK_PERBLOCK))); 1679 for(this_sack = 0; this_sack < tp->rx_opt.eff_sacks; this_sack++) { 1680 *ptr++ = htonl(sp[this_sack].start_seq); 1681 *ptr++ = htonl(sp[this_sack].end_seq); 1682 } 1683 if (tp->rx_opt.dsack) { 1684 tp->rx_opt.dsack = 0; 1685 tp->rx_opt.eff_sacks--; 1686 } 1687 } 1688 } 1689 1690 /* Construct a tcp options header for a SYN or SYN_ACK packet. 1691 * If this is every changed make sure to change the definition of 1692 * MAX_SYN_SIZE to match the new maximum number of options that you 1693 * can generate. 1694 */ 1695 static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack, 1696 int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent) 1697 { 1698 /* We always get an MSS option. 1699 * The option bytes which will be seen in normal data 1700 * packets should timestamps be used, must be in the MSS 1701 * advertised. But we subtract them from tp->mss_cache so 1702 * that calculations in tcp_sendmsg are simpler etc. 1703 * So account for this fact here if necessary. If we 1704 * don't do this correctly, as a receiver we won't 1705 * recognize data packets as being full sized when we 1706 * should, and thus we won't abide by the delayed ACK 1707 * rules correctly. 1708 * SACKs don't matter, we never delay an ACK when we 1709 * have any of those going out. 1710 */ 1711 *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss); 1712 if (ts) { 1713 if(sack) 1714 *ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) | 1715 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP); 1716 else 1717 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | 1718 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP); 1719 *ptr++ = htonl(tstamp); /* TSVAL */ 1720 *ptr++ = htonl(ts_recent); /* TSECR */ 1721 } else if(sack) 1722 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | 1723 (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM); 1724 if (offer_wscale) 1725 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale)); 1726 } 1727 1728 /* Determine a window scaling and initial window to offer. */ 1729 extern void tcp_select_initial_window(int __space, __u32 mss, 1730 __u32 *rcv_wnd, __u32 *window_clamp, 1731 int wscale_ok, __u8 *rcv_wscale); 1732 1733 static inline int tcp_win_from_space(int space) 1734 { 1735 return sysctl_tcp_adv_win_scale<=0 ? 1736 (space>>(-sysctl_tcp_adv_win_scale)) : 1737 space - (space>>sysctl_tcp_adv_win_scale); 1738 } 1739 1740 /* Note: caller must be prepared to deal with negative returns */ 1741 static inline int tcp_space(const struct sock *sk) 1742 { 1743 return tcp_win_from_space(sk->sk_rcvbuf - 1744 atomic_read(&sk->sk_rmem_alloc)); 1745 } 1746 1747 static inline int tcp_full_space(const struct sock *sk) 1748 { 1749 return tcp_win_from_space(sk->sk_rcvbuf); 1750 } 1751 1752 static inline void tcp_acceptq_queue(struct sock *sk, struct open_request *req, 1753 struct sock *child) 1754 { 1755 struct tcp_sock *tp = tcp_sk(sk); 1756 1757 req->sk = child; 1758 sk_acceptq_added(sk); 1759 1760 if (!tp->accept_queue_tail) { 1761 tp->accept_queue = req; 1762 } else { 1763 tp->accept_queue_tail->dl_next = req; 1764 } 1765 tp->accept_queue_tail = req; 1766 req->dl_next = NULL; 1767 } 1768 1769 struct tcp_listen_opt 1770 { 1771 u8 max_qlen_log; /* log_2 of maximal queued SYNs */ 1772 int qlen; 1773 int qlen_young; 1774 int clock_hand; 1775 u32 hash_rnd; 1776 struct open_request *syn_table[TCP_SYNQ_HSIZE]; 1777 }; 1778 1779 static inline void 1780 tcp_synq_removed(struct sock *sk, struct open_request *req) 1781 { 1782 struct tcp_listen_opt *lopt = tcp_sk(sk)->listen_opt; 1783 1784 if (--lopt->qlen == 0) 1785 tcp_delete_keepalive_timer(sk); 1786 if (req->retrans == 0) 1787 lopt->qlen_young--; 1788 } 1789 1790 static inline void tcp_synq_added(struct sock *sk) 1791 { 1792 struct tcp_listen_opt *lopt = tcp_sk(sk)->listen_opt; 1793 1794 if (lopt->qlen++ == 0) 1795 tcp_reset_keepalive_timer(sk, TCP_TIMEOUT_INIT); 1796 lopt->qlen_young++; 1797 } 1798 1799 static inline int tcp_synq_len(struct sock *sk) 1800 { 1801 return tcp_sk(sk)->listen_opt->qlen; 1802 } 1803 1804 static inline int tcp_synq_young(struct sock *sk) 1805 { 1806 return tcp_sk(sk)->listen_opt->qlen_young; 1807 } 1808 1809 static inline int tcp_synq_is_full(struct sock *sk) 1810 { 1811 return tcp_synq_len(sk) >> tcp_sk(sk)->listen_opt->max_qlen_log; 1812 } 1813 1814 static inline void tcp_synq_unlink(struct tcp_sock *tp, struct open_request *req, 1815 struct open_request **prev) 1816 { 1817 write_lock(&tp->syn_wait_lock); 1818 *prev = req->dl_next; 1819 write_unlock(&tp->syn_wait_lock); 1820 } 1821 1822 static inline void tcp_synq_drop(struct sock *sk, struct open_request *req, 1823 struct open_request **prev) 1824 { 1825 tcp_synq_unlink(tcp_sk(sk), req, prev); 1826 tcp_synq_removed(sk, req); 1827 tcp_openreq_free(req); 1828 } 1829 1830 static __inline__ void tcp_openreq_init(struct open_request *req, 1831 struct tcp_options_received *rx_opt, 1832 struct sk_buff *skb) 1833 { 1834 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 1835 req->rcv_isn = TCP_SKB_CB(skb)->seq; 1836 req->mss = rx_opt->mss_clamp; 1837 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 1838 req->tstamp_ok = rx_opt->tstamp_ok; 1839 req->sack_ok = rx_opt->sack_ok; 1840 req->snd_wscale = rx_opt->snd_wscale; 1841 req->wscale_ok = rx_opt->wscale_ok; 1842 req->acked = 0; 1843 req->ecn_ok = 0; 1844 req->rmt_port = skb->h.th->source; 1845 } 1846 1847 extern void tcp_enter_memory_pressure(void); 1848 1849 extern void tcp_listen_wlock(void); 1850 1851 /* - We may sleep inside this lock. 1852 * - If sleeping is not required (or called from BH), 1853 * use plain read_(un)lock(&tcp_lhash_lock). 1854 */ 1855 1856 static inline void tcp_listen_lock(void) 1857 { 1858 /* read_lock synchronizes to candidates to writers */ 1859 read_lock(&tcp_lhash_lock); 1860 atomic_inc(&tcp_lhash_users); 1861 read_unlock(&tcp_lhash_lock); 1862 } 1863 1864 static inline void tcp_listen_unlock(void) 1865 { 1866 if (atomic_dec_and_test(&tcp_lhash_users)) 1867 wake_up(&tcp_lhash_wait); 1868 } 1869 1870 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1871 { 1872 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1873 } 1874 1875 static inline int keepalive_time_when(const struct tcp_sock *tp) 1876 { 1877 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1878 } 1879 1880 static inline int tcp_fin_time(const struct tcp_sock *tp) 1881 { 1882 int fin_timeout = tp->linger2 ? : sysctl_tcp_fin_timeout; 1883 1884 if (fin_timeout < (tp->rto<<2) - (tp->rto>>1)) 1885 fin_timeout = (tp->rto<<2) - (tp->rto>>1); 1886 1887 return fin_timeout; 1888 } 1889 1890 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst) 1891 { 1892 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0) 1893 return 0; 1894 if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS) 1895 return 0; 1896 1897 /* RST segments are not recommended to carry timestamp, 1898 and, if they do, it is recommended to ignore PAWS because 1899 "their cleanup function should take precedence over timestamps." 1900 Certainly, it is mistake. It is necessary to understand the reasons 1901 of this constraint to relax it: if peer reboots, clock may go 1902 out-of-sync and half-open connections will not be reset. 1903 Actually, the problem would be not existing if all 1904 the implementations followed draft about maintaining clock 1905 via reboots. Linux-2.2 DOES NOT! 1906 1907 However, we can relax time bounds for RST segments to MSL. 1908 */ 1909 if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1910 return 0; 1911 return 1; 1912 } 1913 1914 static inline void tcp_v4_setup_caps(struct sock *sk, struct dst_entry *dst) 1915 { 1916 sk->sk_route_caps = dst->dev->features; 1917 if (sk->sk_route_caps & NETIF_F_TSO) { 1918 if (sock_flag(sk, SOCK_NO_LARGESEND) || dst->header_len) 1919 sk->sk_route_caps &= ~NETIF_F_TSO; 1920 } 1921 } 1922 1923 #define TCP_CHECK_TIMER(sk) do { } while (0) 1924 1925 static inline int tcp_use_frto(const struct sock *sk) 1926 { 1927 const struct tcp_sock *tp = tcp_sk(sk); 1928 1929 /* F-RTO must be activated in sysctl and there must be some 1930 * unsent new data, and the advertised window should allow 1931 * sending it. 1932 */ 1933 return (sysctl_tcp_frto && sk->sk_send_head && 1934 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq, 1935 tp->snd_una + tp->snd_wnd)); 1936 } 1937 1938 static inline void tcp_mib_init(void) 1939 { 1940 /* See RFC 2012 */ 1941 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1); 1942 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1943 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1944 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1); 1945 } 1946 1947 /* /proc */ 1948 enum tcp_seq_states { 1949 TCP_SEQ_STATE_LISTENING, 1950 TCP_SEQ_STATE_OPENREQ, 1951 TCP_SEQ_STATE_ESTABLISHED, 1952 TCP_SEQ_STATE_TIME_WAIT, 1953 }; 1954 1955 struct tcp_seq_afinfo { 1956 struct module *owner; 1957 char *name; 1958 sa_family_t family; 1959 int (*seq_show) (struct seq_file *m, void *v); 1960 struct file_operations *seq_fops; 1961 }; 1962 1963 struct tcp_iter_state { 1964 sa_family_t family; 1965 enum tcp_seq_states state; 1966 struct sock *syn_wait_sk; 1967 int bucket, sbucket, num, uid; 1968 struct seq_operations seq_ops; 1969 }; 1970 1971 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo); 1972 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo); 1973 1974 /* TCP Westwood functions and constants */ 1975 1976 #define TCP_WESTWOOD_INIT_RTT (20*HZ) /* maybe too conservative?! */ 1977 #define TCP_WESTWOOD_RTT_MIN (HZ/20) /* 50ms */ 1978 1979 static inline void tcp_westwood_update_rtt(struct tcp_sock *tp, __u32 rtt_seq) 1980 { 1981 if (tcp_is_westwood(tp)) 1982 tp->westwood.rtt = rtt_seq; 1983 } 1984 1985 static inline __u32 __tcp_westwood_bw_rttmin(const struct tcp_sock *tp) 1986 { 1987 return max((tp->westwood.bw_est) * (tp->westwood.rtt_min) / 1988 (__u32) (tp->mss_cache_std), 1989 2U); 1990 } 1991 1992 static inline __u32 tcp_westwood_bw_rttmin(const struct tcp_sock *tp) 1993 { 1994 return tcp_is_westwood(tp) ? __tcp_westwood_bw_rttmin(tp) : 0; 1995 } 1996 1997 static inline int tcp_westwood_ssthresh(struct tcp_sock *tp) 1998 { 1999 __u32 ssthresh = 0; 2000 2001 if (tcp_is_westwood(tp)) { 2002 ssthresh = __tcp_westwood_bw_rttmin(tp); 2003 if (ssthresh) 2004 tp->snd_ssthresh = ssthresh; 2005 } 2006 2007 return (ssthresh != 0); 2008 } 2009 2010 static inline int tcp_westwood_cwnd(struct tcp_sock *tp) 2011 { 2012 __u32 cwnd = 0; 2013 2014 if (tcp_is_westwood(tp)) { 2015 cwnd = __tcp_westwood_bw_rttmin(tp); 2016 if (cwnd) 2017 tp->snd_cwnd = cwnd; 2018 } 2019 2020 return (cwnd != 0); 2021 } 2022 #endif /* _TCP_H */ 2023