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 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 #include <linux/config.h> 25 #include <linux/list.h> 26 #include <linux/tcp.h> 27 #include <linux/slab.h> 28 #include <linux/cache.h> 29 #include <linux/percpu.h> 30 #include <linux/skbuff.h> 31 32 #include <net/inet_connection_sock.h> 33 #include <net/inet_timewait_sock.h> 34 #include <net/inet_hashtables.h> 35 #include <net/checksum.h> 36 #include <net/request_sock.h> 37 #include <net/sock.h> 38 #include <net/snmp.h> 39 #include <net/ip.h> 40 #include <net/tcp_states.h> 41 42 #include <linux/seq_file.h> 43 44 extern struct inet_hashinfo tcp_hashinfo; 45 46 extern atomic_t tcp_orphan_count; 47 extern void tcp_time_wait(struct sock *sk, int state, int timeo); 48 49 #define MAX_TCP_HEADER (128 + MAX_HEADER) 50 51 /* 52 * Never offer a window over 32767 without using window scaling. Some 53 * poor stacks do signed 16bit maths! 54 */ 55 #define MAX_TCP_WINDOW 32767U 56 57 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 58 #define TCP_MIN_MSS 88U 59 60 /* Minimal RCV_MSS. */ 61 #define TCP_MIN_RCVMSS 536U 62 63 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 64 #define TCP_FASTRETRANS_THRESH 3 65 66 /* Maximal reordering. */ 67 #define TCP_MAX_REORDERING 127 68 69 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 70 #define TCP_MAX_QUICKACKS 16U 71 72 /* urg_data states */ 73 #define TCP_URG_VALID 0x0100 74 #define TCP_URG_NOTYET 0x0200 75 #define TCP_URG_READ 0x0400 76 77 #define TCP_RETR1 3 /* 78 * This is how many retries it does before it 79 * tries to figure out if the gateway is 80 * down. Minimal RFC value is 3; it corresponds 81 * to ~3sec-8min depending on RTO. 82 */ 83 84 #define TCP_RETR2 15 /* 85 * This should take at least 86 * 90 minutes to time out. 87 * RFC1122 says that the limit is 100 sec. 88 * 15 is ~13-30min depending on RTO. 89 */ 90 91 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a 92 * connection: ~180sec is RFC minimum */ 93 94 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a 95 * connection: ~180sec is RFC minimum */ 96 97 98 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned 99 * socket. 7 is ~50sec-16min. 100 */ 101 102 103 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 104 * state, about 60 seconds */ 105 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 106 /* BSD style FIN_WAIT2 deadlock breaker. 107 * It used to be 3min, new value is 60sec, 108 * to combine FIN-WAIT-2 timeout with 109 * TIME-WAIT timer. 110 */ 111 112 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 113 #if HZ >= 100 114 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 115 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 116 #else 117 #define TCP_DELACK_MIN 4U 118 #define TCP_ATO_MIN 4U 119 #endif 120 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 121 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 122 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */ 123 124 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 125 * for local resources. 126 */ 127 128 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 129 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 130 #define TCP_KEEPALIVE_INTVL (75*HZ) 131 132 #define MAX_TCP_KEEPIDLE 32767 133 #define MAX_TCP_KEEPINTVL 32767 134 #define MAX_TCP_KEEPCNT 127 135 #define MAX_TCP_SYNCNT 127 136 137 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 138 #define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */ 139 140 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 141 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 142 * after this time. It should be equal 143 * (or greater than) TCP_TIMEWAIT_LEN 144 * to provide reliability equal to one 145 * provided by timewait state. 146 */ 147 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 148 * timestamps. It must be less than 149 * minimal timewait lifetime. 150 */ 151 /* 152 * TCP option 153 */ 154 155 #define TCPOPT_NOP 1 /* Padding */ 156 #define TCPOPT_EOL 0 /* End of options */ 157 #define TCPOPT_MSS 2 /* Segment size negotiating */ 158 #define TCPOPT_WINDOW 3 /* Window scaling */ 159 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 160 #define TCPOPT_SACK 5 /* SACK Block */ 161 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 162 163 /* 164 * TCP option lengths 165 */ 166 167 #define TCPOLEN_MSS 4 168 #define TCPOLEN_WINDOW 3 169 #define TCPOLEN_SACK_PERM 2 170 #define TCPOLEN_TIMESTAMP 10 171 172 /* But this is what stacks really send out. */ 173 #define TCPOLEN_TSTAMP_ALIGNED 12 174 #define TCPOLEN_WSCALE_ALIGNED 4 175 #define TCPOLEN_SACKPERM_ALIGNED 4 176 #define TCPOLEN_SACK_BASE 2 177 #define TCPOLEN_SACK_BASE_ALIGNED 4 178 #define TCPOLEN_SACK_PERBLOCK 8 179 180 /* Flags in tp->nonagle */ 181 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 182 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 183 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 184 185 extern struct inet_timewait_death_row tcp_death_row; 186 187 /* sysctl variables for tcp */ 188 extern int sysctl_tcp_timestamps; 189 extern int sysctl_tcp_window_scaling; 190 extern int sysctl_tcp_sack; 191 extern int sysctl_tcp_fin_timeout; 192 extern int sysctl_tcp_keepalive_time; 193 extern int sysctl_tcp_keepalive_probes; 194 extern int sysctl_tcp_keepalive_intvl; 195 extern int sysctl_tcp_syn_retries; 196 extern int sysctl_tcp_synack_retries; 197 extern int sysctl_tcp_retries1; 198 extern int sysctl_tcp_retries2; 199 extern int sysctl_tcp_orphan_retries; 200 extern int sysctl_tcp_syncookies; 201 extern int sysctl_tcp_retrans_collapse; 202 extern int sysctl_tcp_stdurg; 203 extern int sysctl_tcp_rfc1337; 204 extern int sysctl_tcp_abort_on_overflow; 205 extern int sysctl_tcp_max_orphans; 206 extern int sysctl_tcp_fack; 207 extern int sysctl_tcp_reordering; 208 extern int sysctl_tcp_ecn; 209 extern int sysctl_tcp_dsack; 210 extern int sysctl_tcp_mem[3]; 211 extern int sysctl_tcp_wmem[3]; 212 extern int sysctl_tcp_rmem[3]; 213 extern int sysctl_tcp_app_win; 214 extern int sysctl_tcp_adv_win_scale; 215 extern int sysctl_tcp_tw_reuse; 216 extern int sysctl_tcp_frto; 217 extern int sysctl_tcp_low_latency; 218 extern int sysctl_tcp_nometrics_save; 219 extern int sysctl_tcp_moderate_rcvbuf; 220 extern int sysctl_tcp_tso_win_divisor; 221 extern int sysctl_tcp_abc; 222 223 extern atomic_t tcp_memory_allocated; 224 extern atomic_t tcp_sockets_allocated; 225 extern int tcp_memory_pressure; 226 227 /* 228 * The next routines deal with comparing 32 bit unsigned ints 229 * and worry about wraparound (automatic with unsigned arithmetic). 230 */ 231 232 static inline int before(__u32 seq1, __u32 seq2) 233 { 234 return (__s32)(seq1-seq2) < 0; 235 } 236 237 static inline int after(__u32 seq1, __u32 seq2) 238 { 239 return (__s32)(seq2-seq1) < 0; 240 } 241 242 243 /* is s2<=s1<=s3 ? */ 244 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3) 245 { 246 return seq3 - seq2 >= seq1 - seq2; 247 } 248 249 250 extern struct proto tcp_prot; 251 252 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics); 253 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field) 254 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field) 255 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field) 256 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field) 257 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val) 258 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val) 259 260 extern void tcp_v4_err(struct sk_buff *skb, u32); 261 262 extern void tcp_shutdown (struct sock *sk, int how); 263 264 extern int tcp_v4_rcv(struct sk_buff *skb); 265 266 extern int tcp_v4_remember_stamp(struct sock *sk); 267 268 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 269 270 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, 271 struct msghdr *msg, size_t size); 272 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags); 273 274 extern int tcp_ioctl(struct sock *sk, 275 int cmd, 276 unsigned long arg); 277 278 extern int tcp_rcv_state_process(struct sock *sk, 279 struct sk_buff *skb, 280 struct tcphdr *th, 281 unsigned len); 282 283 extern int tcp_rcv_established(struct sock *sk, 284 struct sk_buff *skb, 285 struct tcphdr *th, 286 unsigned len); 287 288 extern void tcp_rcv_space_adjust(struct sock *sk); 289 290 extern int tcp_twsk_unique(struct sock *sk, 291 struct sock *sktw, void *twp); 292 293 static inline void tcp_dec_quickack_mode(struct sock *sk, 294 const unsigned int pkts) 295 { 296 struct inet_connection_sock *icsk = inet_csk(sk); 297 298 if (icsk->icsk_ack.quick) { 299 if (pkts >= icsk->icsk_ack.quick) { 300 icsk->icsk_ack.quick = 0; 301 /* Leaving quickack mode we deflate ATO. */ 302 icsk->icsk_ack.ato = TCP_ATO_MIN; 303 } else 304 icsk->icsk_ack.quick -= pkts; 305 } 306 } 307 308 extern void tcp_enter_quickack_mode(struct sock *sk); 309 310 static inline void tcp_clear_options(struct tcp_options_received *rx_opt) 311 { 312 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0; 313 } 314 315 enum tcp_tw_status 316 { 317 TCP_TW_SUCCESS = 0, 318 TCP_TW_RST = 1, 319 TCP_TW_ACK = 2, 320 TCP_TW_SYN = 3 321 }; 322 323 324 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 325 struct sk_buff *skb, 326 const struct tcphdr *th); 327 328 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb, 329 struct request_sock *req, 330 struct request_sock **prev); 331 extern int tcp_child_process(struct sock *parent, 332 struct sock *child, 333 struct sk_buff *skb); 334 extern void tcp_enter_frto(struct sock *sk); 335 extern void tcp_enter_loss(struct sock *sk, int how); 336 extern void tcp_clear_retrans(struct tcp_sock *tp); 337 extern void tcp_update_metrics(struct sock *sk); 338 339 extern void tcp_close(struct sock *sk, 340 long timeout); 341 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait); 342 343 extern int tcp_getsockopt(struct sock *sk, int level, 344 int optname, 345 char __user *optval, 346 int __user *optlen); 347 extern int tcp_setsockopt(struct sock *sk, int level, 348 int optname, char __user *optval, 349 int optlen); 350 extern void tcp_set_keepalive(struct sock *sk, int val); 351 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, 352 struct msghdr *msg, 353 size_t len, int nonblock, 354 int flags, int *addr_len); 355 356 extern void tcp_parse_options(struct sk_buff *skb, 357 struct tcp_options_received *opt_rx, 358 int estab); 359 360 /* 361 * TCP v4 functions exported for the inet6 API 362 */ 363 364 extern void tcp_v4_send_check(struct sock *sk, int len, 365 struct sk_buff *skb); 366 367 extern int tcp_v4_conn_request(struct sock *sk, 368 struct sk_buff *skb); 369 370 extern struct sock * tcp_create_openreq_child(struct sock *sk, 371 struct request_sock *req, 372 struct sk_buff *skb); 373 374 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, 375 struct sk_buff *skb, 376 struct request_sock *req, 377 struct dst_entry *dst); 378 379 extern int tcp_v4_do_rcv(struct sock *sk, 380 struct sk_buff *skb); 381 382 extern int tcp_v4_connect(struct sock *sk, 383 struct sockaddr *uaddr, 384 int addr_len); 385 386 extern int tcp_connect(struct sock *sk); 387 388 extern struct sk_buff * tcp_make_synack(struct sock *sk, 389 struct dst_entry *dst, 390 struct request_sock *req); 391 392 extern int tcp_disconnect(struct sock *sk, int flags); 393 394 extern void tcp_unhash(struct sock *sk); 395 396 extern int tcp_v4_hash_connecting(struct sock *sk); 397 398 399 /* From syncookies.c */ 400 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 401 struct ip_options *opt); 402 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, 403 __u16 *mss); 404 405 /* tcp_output.c */ 406 407 extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp, 408 unsigned int cur_mss, int nonagle); 409 extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp); 410 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); 411 extern void tcp_xmit_retransmit_queue(struct sock *); 412 extern void tcp_simple_retransmit(struct sock *); 413 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32); 414 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 415 416 extern void tcp_send_probe0(struct sock *); 417 extern void tcp_send_partial(struct sock *); 418 extern int tcp_write_wakeup(struct sock *); 419 extern void tcp_send_fin(struct sock *sk); 420 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority); 421 extern int tcp_send_synack(struct sock *); 422 extern void tcp_push_one(struct sock *, unsigned int mss_now); 423 extern void tcp_send_ack(struct sock *sk); 424 extern void tcp_send_delayed_ack(struct sock *sk); 425 426 /* tcp_input.c */ 427 extern void tcp_cwnd_application_limited(struct sock *sk); 428 429 /* tcp_timer.c */ 430 extern void tcp_init_xmit_timers(struct sock *); 431 static inline void tcp_clear_xmit_timers(struct sock *sk) 432 { 433 inet_csk_clear_xmit_timers(sk); 434 } 435 436 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 437 extern unsigned int tcp_current_mss(struct sock *sk, int large); 438 439 /* tcp.c */ 440 extern void tcp_get_info(struct sock *, struct tcp_info *); 441 442 /* Read 'sendfile()'-style from a TCP socket */ 443 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 444 unsigned int, size_t); 445 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 446 sk_read_actor_t recv_actor); 447 448 extern void tcp_initialize_rcv_mss(struct sock *sk); 449 450 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 451 { 452 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 453 ntohl(TCP_FLAG_ACK) | 454 snd_wnd); 455 } 456 457 static inline void tcp_fast_path_on(struct tcp_sock *tp) 458 { 459 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 460 } 461 462 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp) 463 { 464 if (skb_queue_empty(&tp->out_of_order_queue) && 465 tp->rcv_wnd && 466 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 467 !tp->urg_data) 468 tcp_fast_path_on(tp); 469 } 470 471 /* Compute the actual receive window we are currently advertising. 472 * Rcv_nxt can be after the window if our peer push more data 473 * than the offered window. 474 */ 475 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 476 { 477 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 478 479 if (win < 0) 480 win = 0; 481 return (u32) win; 482 } 483 484 /* Choose a new window, without checks for shrinking, and without 485 * scaling applied to the result. The caller does these things 486 * if necessary. This is a "raw" window selection. 487 */ 488 extern u32 __tcp_select_window(struct sock *sk); 489 490 /* TCP timestamps are only 32-bits, this causes a slight 491 * complication on 64-bit systems since we store a snapshot 492 * of jiffies in the buffer control blocks below. We decided 493 * to use only the low 32-bits of jiffies and hide the ugly 494 * casts with the following macro. 495 */ 496 #define tcp_time_stamp ((__u32)(jiffies)) 497 498 /* This is what the send packet queuing engine uses to pass 499 * TCP per-packet control information to the transmission 500 * code. We also store the host-order sequence numbers in 501 * here too. This is 36 bytes on 32-bit architectures, 502 * 40 bytes on 64-bit machines, if this grows please adjust 503 * skbuff.h:skbuff->cb[xxx] size appropriately. 504 */ 505 struct tcp_skb_cb { 506 union { 507 struct inet_skb_parm h4; 508 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE) 509 struct inet6_skb_parm h6; 510 #endif 511 } header; /* For incoming frames */ 512 __u32 seq; /* Starting sequence number */ 513 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 514 __u32 when; /* used to compute rtt's */ 515 __u8 flags; /* TCP header flags. */ 516 517 /* NOTE: These must match up to the flags byte in a 518 * real TCP header. 519 */ 520 #define TCPCB_FLAG_FIN 0x01 521 #define TCPCB_FLAG_SYN 0x02 522 #define TCPCB_FLAG_RST 0x04 523 #define TCPCB_FLAG_PSH 0x08 524 #define TCPCB_FLAG_ACK 0x10 525 #define TCPCB_FLAG_URG 0x20 526 #define TCPCB_FLAG_ECE 0x40 527 #define TCPCB_FLAG_CWR 0x80 528 529 __u8 sacked; /* State flags for SACK/FACK. */ 530 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 531 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 532 #define TCPCB_LOST 0x04 /* SKB is lost */ 533 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 534 535 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 536 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 537 538 #define TCPCB_URG 0x20 /* Urgent pointer advanced here */ 539 540 #define TCPCB_AT_TAIL (TCPCB_URG) 541 542 __u16 urg_ptr; /* Valid w/URG flags is set. */ 543 __u32 ack_seq; /* Sequence number ACK'd */ 544 }; 545 546 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 547 548 #include <net/tcp_ecn.h> 549 550 /* Due to TSO, an SKB can be composed of multiple actual 551 * packets. To keep these tracked properly, we use this. 552 */ 553 static inline int tcp_skb_pcount(const struct sk_buff *skb) 554 { 555 return skb_shinfo(skb)->tso_segs; 556 } 557 558 /* This is valid iff tcp_skb_pcount() > 1. */ 559 static inline int tcp_skb_mss(const struct sk_buff *skb) 560 { 561 return skb_shinfo(skb)->tso_size; 562 } 563 564 static inline void tcp_dec_pcount_approx(__u32 *count, 565 const struct sk_buff *skb) 566 { 567 if (*count) { 568 *count -= tcp_skb_pcount(skb); 569 if ((int)*count < 0) 570 *count = 0; 571 } 572 } 573 574 static inline void tcp_packets_out_inc(struct sock *sk, 575 struct tcp_sock *tp, 576 const struct sk_buff *skb) 577 { 578 int orig = tp->packets_out; 579 580 tp->packets_out += tcp_skb_pcount(skb); 581 if (!orig) 582 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, 583 inet_csk(sk)->icsk_rto, TCP_RTO_MAX); 584 } 585 586 static inline void tcp_packets_out_dec(struct tcp_sock *tp, 587 const struct sk_buff *skb) 588 { 589 tp->packets_out -= tcp_skb_pcount(skb); 590 } 591 592 /* Events passed to congestion control interface */ 593 enum tcp_ca_event { 594 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 595 CA_EVENT_CWND_RESTART, /* congestion window restart */ 596 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 597 CA_EVENT_FRTO, /* fast recovery timeout */ 598 CA_EVENT_LOSS, /* loss timeout */ 599 CA_EVENT_FAST_ACK, /* in sequence ack */ 600 CA_EVENT_SLOW_ACK, /* other ack */ 601 }; 602 603 /* 604 * Interface for adding new TCP congestion control handlers 605 */ 606 #define TCP_CA_NAME_MAX 16 607 struct tcp_congestion_ops { 608 struct list_head list; 609 610 /* initialize private data (optional) */ 611 void (*init)(struct sock *sk); 612 /* cleanup private data (optional) */ 613 void (*release)(struct sock *sk); 614 615 /* return slow start threshold (required) */ 616 u32 (*ssthresh)(struct sock *sk); 617 /* lower bound for congestion window (optional) */ 618 u32 (*min_cwnd)(struct sock *sk); 619 /* do new cwnd calculation (required) */ 620 void (*cong_avoid)(struct sock *sk, u32 ack, 621 u32 rtt, u32 in_flight, int good_ack); 622 /* round trip time sample per acked packet (optional) */ 623 void (*rtt_sample)(struct sock *sk, u32 usrtt); 624 /* call before changing ca_state (optional) */ 625 void (*set_state)(struct sock *sk, u8 new_state); 626 /* call when cwnd event occurs (optional) */ 627 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 628 /* new value of cwnd after loss (optional) */ 629 u32 (*undo_cwnd)(struct sock *sk); 630 /* hook for packet ack accounting (optional) */ 631 void (*pkts_acked)(struct sock *sk, u32 num_acked); 632 /* get info for inet_diag (optional) */ 633 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 634 635 char name[TCP_CA_NAME_MAX]; 636 struct module *owner; 637 }; 638 639 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type); 640 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 641 642 extern void tcp_init_congestion_control(struct sock *sk); 643 extern void tcp_cleanup_congestion_control(struct sock *sk); 644 extern int tcp_set_default_congestion_control(const char *name); 645 extern void tcp_get_default_congestion_control(char *name); 646 extern int tcp_set_congestion_control(struct sock *sk, const char *name); 647 extern void tcp_slow_start(struct tcp_sock *tp); 648 649 extern struct tcp_congestion_ops tcp_init_congestion_ops; 650 extern u32 tcp_reno_ssthresh(struct sock *sk); 651 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, 652 u32 rtt, u32 in_flight, int flag); 653 extern u32 tcp_reno_min_cwnd(struct sock *sk); 654 extern struct tcp_congestion_ops tcp_reno; 655 656 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 657 { 658 struct inet_connection_sock *icsk = inet_csk(sk); 659 660 if (icsk->icsk_ca_ops->set_state) 661 icsk->icsk_ca_ops->set_state(sk, ca_state); 662 icsk->icsk_ca_state = ca_state; 663 } 664 665 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 666 { 667 const struct inet_connection_sock *icsk = inet_csk(sk); 668 669 if (icsk->icsk_ca_ops->cwnd_event) 670 icsk->icsk_ca_ops->cwnd_event(sk, event); 671 } 672 673 /* This determines how many packets are "in the network" to the best 674 * of our knowledge. In many cases it is conservative, but where 675 * detailed information is available from the receiver (via SACK 676 * blocks etc.) we can make more aggressive calculations. 677 * 678 * Use this for decisions involving congestion control, use just 679 * tp->packets_out to determine if the send queue is empty or not. 680 * 681 * Read this equation as: 682 * 683 * "Packets sent once on transmission queue" MINUS 684 * "Packets left network, but not honestly ACKed yet" PLUS 685 * "Packets fast retransmitted" 686 */ 687 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 688 { 689 return (tp->packets_out - tp->left_out + tp->retrans_out); 690 } 691 692 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 693 * The exception is rate halving phase, when cwnd is decreasing towards 694 * ssthresh. 695 */ 696 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 697 { 698 const struct tcp_sock *tp = tcp_sk(sk); 699 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery)) 700 return tp->snd_ssthresh; 701 else 702 return max(tp->snd_ssthresh, 703 ((tp->snd_cwnd >> 1) + 704 (tp->snd_cwnd >> 2))); 705 } 706 707 static inline void tcp_sync_left_out(struct tcp_sock *tp) 708 { 709 if (tp->rx_opt.sack_ok && 710 (tp->sacked_out >= tp->packets_out - tp->lost_out)) 711 tp->sacked_out = tp->packets_out - tp->lost_out; 712 tp->left_out = tp->sacked_out + tp->lost_out; 713 } 714 715 extern void tcp_enter_cwr(struct sock *sk); 716 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst); 717 718 /* Slow start with delack produces 3 packets of burst, so that 719 * it is safe "de facto". 720 */ 721 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 722 { 723 return 3; 724 } 725 726 /* RFC2861 Check whether we are limited by application or congestion window 727 * This is the inverse of cwnd check in tcp_tso_should_defer 728 */ 729 static inline int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight) 730 { 731 const struct tcp_sock *tp = tcp_sk(sk); 732 u32 left; 733 734 if (in_flight >= tp->snd_cwnd) 735 return 1; 736 737 if (!(sk->sk_route_caps & NETIF_F_TSO)) 738 return 0; 739 740 left = tp->snd_cwnd - in_flight; 741 if (sysctl_tcp_tso_win_divisor) 742 return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd; 743 else 744 return left <= tcp_max_burst(tp); 745 } 746 747 static inline void tcp_minshall_update(struct tcp_sock *tp, int mss, 748 const struct sk_buff *skb) 749 { 750 if (skb->len < mss) 751 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 752 } 753 754 static inline void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp) 755 { 756 const struct inet_connection_sock *icsk = inet_csk(sk); 757 if (!tp->packets_out && !icsk->icsk_pending) 758 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 759 icsk->icsk_rto, TCP_RTO_MAX); 760 } 761 762 static inline void tcp_push_pending_frames(struct sock *sk, 763 struct tcp_sock *tp) 764 { 765 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle); 766 } 767 768 static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq) 769 { 770 tp->snd_wl1 = seq; 771 } 772 773 static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq) 774 { 775 tp->snd_wl1 = seq; 776 } 777 778 /* 779 * Calculate(/check) TCP checksum 780 */ 781 static inline u16 tcp_v4_check(struct tcphdr *th, int len, 782 unsigned long saddr, unsigned long daddr, 783 unsigned long base) 784 { 785 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 786 } 787 788 static inline int __tcp_checksum_complete(struct sk_buff *skb) 789 { 790 return __skb_checksum_complete(skb); 791 } 792 793 static inline int tcp_checksum_complete(struct sk_buff *skb) 794 { 795 return skb->ip_summed != CHECKSUM_UNNECESSARY && 796 __tcp_checksum_complete(skb); 797 } 798 799 /* Prequeue for VJ style copy to user, combined with checksumming. */ 800 801 static inline void tcp_prequeue_init(struct tcp_sock *tp) 802 { 803 tp->ucopy.task = NULL; 804 tp->ucopy.len = 0; 805 tp->ucopy.memory = 0; 806 skb_queue_head_init(&tp->ucopy.prequeue); 807 } 808 809 /* Packet is added to VJ-style prequeue for processing in process 810 * context, if a reader task is waiting. Apparently, this exciting 811 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) 812 * failed somewhere. Latency? Burstiness? Well, at least now we will 813 * see, why it failed. 8)8) --ANK 814 * 815 * NOTE: is this not too big to inline? 816 */ 817 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb) 818 { 819 struct tcp_sock *tp = tcp_sk(sk); 820 821 if (!sysctl_tcp_low_latency && tp->ucopy.task) { 822 __skb_queue_tail(&tp->ucopy.prequeue, skb); 823 tp->ucopy.memory += skb->truesize; 824 if (tp->ucopy.memory > sk->sk_rcvbuf) { 825 struct sk_buff *skb1; 826 827 BUG_ON(sock_owned_by_user(sk)); 828 829 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { 830 sk->sk_backlog_rcv(sk, skb1); 831 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED); 832 } 833 834 tp->ucopy.memory = 0; 835 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { 836 wake_up_interruptible(sk->sk_sleep); 837 if (!inet_csk_ack_scheduled(sk)) 838 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 839 (3 * TCP_RTO_MIN) / 4, 840 TCP_RTO_MAX); 841 } 842 return 1; 843 } 844 return 0; 845 } 846 847 848 #undef STATE_TRACE 849 850 #ifdef STATE_TRACE 851 static const char *statename[]={ 852 "Unused","Established","Syn Sent","Syn Recv", 853 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 854 "Close Wait","Last ACK","Listen","Closing" 855 }; 856 #endif 857 858 static inline void tcp_set_state(struct sock *sk, int state) 859 { 860 int oldstate = sk->sk_state; 861 862 switch (state) { 863 case TCP_ESTABLISHED: 864 if (oldstate != TCP_ESTABLISHED) 865 TCP_INC_STATS(TCP_MIB_CURRESTAB); 866 break; 867 868 case TCP_CLOSE: 869 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED) 870 TCP_INC_STATS(TCP_MIB_ESTABRESETS); 871 872 sk->sk_prot->unhash(sk); 873 if (inet_csk(sk)->icsk_bind_hash && 874 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) 875 inet_put_port(&tcp_hashinfo, sk); 876 /* fall through */ 877 default: 878 if (oldstate==TCP_ESTABLISHED) 879 TCP_DEC_STATS(TCP_MIB_CURRESTAB); 880 } 881 882 /* Change state AFTER socket is unhashed to avoid closed 883 * socket sitting in hash tables. 884 */ 885 sk->sk_state = state; 886 887 #ifdef STATE_TRACE 888 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]); 889 #endif 890 } 891 892 static inline void tcp_done(struct sock *sk) 893 { 894 tcp_set_state(sk, TCP_CLOSE); 895 tcp_clear_xmit_timers(sk); 896 897 sk->sk_shutdown = SHUTDOWN_MASK; 898 899 if (!sock_flag(sk, SOCK_DEAD)) 900 sk->sk_state_change(sk); 901 else 902 inet_csk_destroy_sock(sk); 903 } 904 905 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 906 { 907 rx_opt->dsack = 0; 908 rx_opt->eff_sacks = 0; 909 rx_opt->num_sacks = 0; 910 } 911 912 /* Determine a window scaling and initial window to offer. */ 913 extern void tcp_select_initial_window(int __space, __u32 mss, 914 __u32 *rcv_wnd, __u32 *window_clamp, 915 int wscale_ok, __u8 *rcv_wscale); 916 917 static inline int tcp_win_from_space(int space) 918 { 919 return sysctl_tcp_adv_win_scale<=0 ? 920 (space>>(-sysctl_tcp_adv_win_scale)) : 921 space - (space>>sysctl_tcp_adv_win_scale); 922 } 923 924 /* Note: caller must be prepared to deal with negative returns */ 925 static inline int tcp_space(const struct sock *sk) 926 { 927 return tcp_win_from_space(sk->sk_rcvbuf - 928 atomic_read(&sk->sk_rmem_alloc)); 929 } 930 931 static inline int tcp_full_space(const struct sock *sk) 932 { 933 return tcp_win_from_space(sk->sk_rcvbuf); 934 } 935 936 static inline void tcp_openreq_init(struct request_sock *req, 937 struct tcp_options_received *rx_opt, 938 struct sk_buff *skb) 939 { 940 struct inet_request_sock *ireq = inet_rsk(req); 941 942 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 943 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 944 req->mss = rx_opt->mss_clamp; 945 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 946 ireq->tstamp_ok = rx_opt->tstamp_ok; 947 ireq->sack_ok = rx_opt->sack_ok; 948 ireq->snd_wscale = rx_opt->snd_wscale; 949 ireq->wscale_ok = rx_opt->wscale_ok; 950 ireq->acked = 0; 951 ireq->ecn_ok = 0; 952 ireq->rmt_port = skb->h.th->source; 953 } 954 955 extern void tcp_enter_memory_pressure(void); 956 957 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 958 { 959 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 960 } 961 962 static inline int keepalive_time_when(const struct tcp_sock *tp) 963 { 964 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 965 } 966 967 static inline int tcp_fin_time(const struct sock *sk) 968 { 969 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 970 const int rto = inet_csk(sk)->icsk_rto; 971 972 if (fin_timeout < (rto << 2) - (rto >> 1)) 973 fin_timeout = (rto << 2) - (rto >> 1); 974 975 return fin_timeout; 976 } 977 978 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst) 979 { 980 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0) 981 return 0; 982 if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS) 983 return 0; 984 985 /* RST segments are not recommended to carry timestamp, 986 and, if they do, it is recommended to ignore PAWS because 987 "their cleanup function should take precedence over timestamps." 988 Certainly, it is mistake. It is necessary to understand the reasons 989 of this constraint to relax it: if peer reboots, clock may go 990 out-of-sync and half-open connections will not be reset. 991 Actually, the problem would be not existing if all 992 the implementations followed draft about maintaining clock 993 via reboots. Linux-2.2 DOES NOT! 994 995 However, we can relax time bounds for RST segments to MSL. 996 */ 997 if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 998 return 0; 999 return 1; 1000 } 1001 1002 #define TCP_CHECK_TIMER(sk) do { } while (0) 1003 1004 static inline int tcp_use_frto(const struct sock *sk) 1005 { 1006 const struct tcp_sock *tp = tcp_sk(sk); 1007 1008 /* F-RTO must be activated in sysctl and there must be some 1009 * unsent new data, and the advertised window should allow 1010 * sending it. 1011 */ 1012 return (sysctl_tcp_frto && sk->sk_send_head && 1013 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq, 1014 tp->snd_una + tp->snd_wnd)); 1015 } 1016 1017 static inline void tcp_mib_init(void) 1018 { 1019 /* See RFC 2012 */ 1020 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1); 1021 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1022 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1023 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1); 1024 } 1025 1026 /*from STCP */ 1027 static inline void clear_all_retrans_hints(struct tcp_sock *tp){ 1028 tp->lost_skb_hint = NULL; 1029 tp->scoreboard_skb_hint = NULL; 1030 tp->retransmit_skb_hint = NULL; 1031 tp->forward_skb_hint = NULL; 1032 tp->fastpath_skb_hint = NULL; 1033 } 1034 1035 /* /proc */ 1036 enum tcp_seq_states { 1037 TCP_SEQ_STATE_LISTENING, 1038 TCP_SEQ_STATE_OPENREQ, 1039 TCP_SEQ_STATE_ESTABLISHED, 1040 TCP_SEQ_STATE_TIME_WAIT, 1041 }; 1042 1043 struct tcp_seq_afinfo { 1044 struct module *owner; 1045 char *name; 1046 sa_family_t family; 1047 int (*seq_show) (struct seq_file *m, void *v); 1048 struct file_operations *seq_fops; 1049 }; 1050 1051 struct tcp_iter_state { 1052 sa_family_t family; 1053 enum tcp_seq_states state; 1054 struct sock *syn_wait_sk; 1055 int bucket, sbucket, num, uid; 1056 struct seq_operations seq_ops; 1057 }; 1058 1059 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo); 1060 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo); 1061 1062 extern struct request_sock_ops tcp_request_sock_ops; 1063 1064 extern int tcp_v4_destroy_sock(struct sock *sk); 1065 1066 #ifdef CONFIG_PROC_FS 1067 extern int tcp4_proc_init(void); 1068 extern void tcp4_proc_exit(void); 1069 #endif 1070 1071 extern void tcp_v4_init(struct net_proto_family *ops); 1072 extern void tcp_init(void); 1073 1074 #endif /* _TCP_H */ 1075