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/list.h> 25#include <linux/tcp.h> 26#include <linux/slab.h> 27#include <linux/cache.h> 28#include <linux/percpu.h> 29#include <linux/skbuff.h> 30#include <linux/dmaengine.h> 31#include <linux/crypto.h> 32 33#include <net/inet_connection_sock.h> 34#include <net/inet_timewait_sock.h> 35#include <net/inet_hashtables.h> 36#include <net/checksum.h> 37#include <net/request_sock.h> 38#include <net/sock.h> 39#include <net/snmp.h> 40#include <net/ip.h> 41#include <net/tcp_states.h> 42 43#include <linux/seq_file.h> 44 45extern struct inet_hashinfo tcp_hashinfo; 46 47extern atomic_t tcp_orphan_count; 48extern void tcp_time_wait(struct sock *sk, int state, int timeo); 49 50#define MAX_TCP_HEADER (128 + MAX_HEADER) 51 52/* 53 * Never offer a window over 32767 without using window scaling. Some 54 * poor stacks do signed 16bit maths! 55 */ 56#define MAX_TCP_WINDOW 32767U 57 58/* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 59#define TCP_MIN_MSS 88U 60 61/* Minimal RCV_MSS. */ 62#define TCP_MIN_RCVMSS 536U 63 64/* The least MTU to use for probing */ 65#define TCP_BASE_MSS 512 66 67/* After receiving this amount of duplicate ACKs fast retransmit starts. */ 68#define TCP_FASTRETRANS_THRESH 3 69 70/* Maximal reordering. */ 71#define TCP_MAX_REORDERING 127 72 73/* Maximal number of ACKs sent quickly to accelerate slow-start. */ 74#define TCP_MAX_QUICKACKS 16U 75 76/* urg_data states */ 77#define TCP_URG_VALID 0x0100 78#define TCP_URG_NOTYET 0x0200 79#define TCP_URG_READ 0x0400 80 81#define TCP_RETR1 3 /* 82 * This is how many retries it does before it 83 * tries to figure out if the gateway is 84 * down. Minimal RFC value is 3; it corresponds 85 * to ~3sec-8min depending on RTO. 86 */ 87 88#define TCP_RETR2 15 /* 89 * This should take at least 90 * 90 minutes to time out. 91 * RFC1122 says that the limit is 100 sec. 92 * 15 is ~13-30min depending on RTO. 93 */ 94 95#define TCP_SYN_RETRIES 5 /* number of times to retry active opening a 96 * connection: ~180sec is RFC minimum */ 97 98#define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a 99 * connection: ~180sec is RFC minimum */ 100 101 102#define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned 103 * socket. 7 is ~50sec-16min. 104 */ 105 106 107#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 108 * state, about 60 seconds */ 109#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 110 /* BSD style FIN_WAIT2 deadlock breaker. 111 * It used to be 3min, new value is 60sec, 112 * to combine FIN-WAIT-2 timeout with 113 * TIME-WAIT timer. 114 */ 115 116#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 117#if HZ >= 100 118#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 119#define TCP_ATO_MIN ((unsigned)(HZ/25)) 120#else 121#define TCP_DELACK_MIN 4U 122#define TCP_ATO_MIN 4U 123#endif 124#define TCP_RTO_MAX ((unsigned)(120*HZ)) 125#define TCP_RTO_MIN ((unsigned)(HZ/5)) 126#define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */ 127 128#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 129 * for local resources. 130 */ 131 132#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 133#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 134#define TCP_KEEPALIVE_INTVL (75*HZ) 135 136#define MAX_TCP_KEEPIDLE 32767 137#define MAX_TCP_KEEPINTVL 32767 138#define MAX_TCP_KEEPCNT 127 139#define MAX_TCP_SYNCNT 127 140 141#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 142 143#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 144#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 145 * after this time. It should be equal 146 * (or greater than) TCP_TIMEWAIT_LEN 147 * to provide reliability equal to one 148 * provided by timewait state. 149 */ 150#define TCP_PAWS_WINDOW 1 /* Replay window for per-host 151 * timestamps. It must be less than 152 * minimal timewait lifetime. 153 */ 154/* 155 * TCP option 156 */ 157 158#define TCPOPT_NOP 1 /* Padding */ 159#define TCPOPT_EOL 0 /* End of options */ 160#define TCPOPT_MSS 2 /* Segment size negotiating */ 161#define TCPOPT_WINDOW 3 /* Window scaling */ 162#define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 163#define TCPOPT_SACK 5 /* SACK Block */ 164#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 165#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 166 167/* 168 * TCP option lengths 169 */ 170 171#define TCPOLEN_MSS 4 172#define TCPOLEN_WINDOW 3 173#define TCPOLEN_SACK_PERM 2 174#define TCPOLEN_TIMESTAMP 10 175#define TCPOLEN_MD5SIG 18 176 177/* But this is what stacks really send out. */ 178#define TCPOLEN_TSTAMP_ALIGNED 12 179#define TCPOLEN_WSCALE_ALIGNED 4 180#define TCPOLEN_SACKPERM_ALIGNED 4 181#define TCPOLEN_SACK_BASE 2 182#define TCPOLEN_SACK_BASE_ALIGNED 4 183#define TCPOLEN_SACK_PERBLOCK 8 184#define TCPOLEN_MD5SIG_ALIGNED 20 185 186/* Flags in tp->nonagle */ 187#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 188#define TCP_NAGLE_CORK 2 /* Socket is corked */ 189#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 190 191extern struct inet_timewait_death_row tcp_death_row; 192 193/* sysctl variables for tcp */ 194extern int sysctl_tcp_timestamps; 195extern int sysctl_tcp_window_scaling; 196extern int sysctl_tcp_sack; 197extern int sysctl_tcp_fin_timeout; 198extern int sysctl_tcp_keepalive_time; 199extern int sysctl_tcp_keepalive_probes; 200extern int sysctl_tcp_keepalive_intvl; 201extern int sysctl_tcp_syn_retries; 202extern int sysctl_tcp_synack_retries; 203extern int sysctl_tcp_retries1; 204extern int sysctl_tcp_retries2; 205extern int sysctl_tcp_orphan_retries; 206extern int sysctl_tcp_syncookies; 207extern int sysctl_tcp_retrans_collapse; 208extern int sysctl_tcp_stdurg; 209extern int sysctl_tcp_rfc1337; 210extern int sysctl_tcp_abort_on_overflow; 211extern int sysctl_tcp_max_orphans; 212extern int sysctl_tcp_fack; 213extern int sysctl_tcp_reordering; 214extern int sysctl_tcp_ecn; 215extern int sysctl_tcp_dsack; 216extern int sysctl_tcp_mem[3]; 217extern int sysctl_tcp_wmem[3]; 218extern int sysctl_tcp_rmem[3]; 219extern int sysctl_tcp_app_win; 220extern int sysctl_tcp_adv_win_scale; 221extern int sysctl_tcp_tw_reuse; 222extern int sysctl_tcp_frto; 223extern int sysctl_tcp_frto_response; 224extern int sysctl_tcp_low_latency; 225extern int sysctl_tcp_dma_copybreak; 226extern int sysctl_tcp_nometrics_save; 227extern int sysctl_tcp_moderate_rcvbuf; 228extern int sysctl_tcp_tso_win_divisor; 229extern int sysctl_tcp_abc; 230extern int sysctl_tcp_mtu_probing; 231extern int sysctl_tcp_base_mss; 232extern int sysctl_tcp_workaround_signed_windows; 233extern int sysctl_tcp_slow_start_after_idle; 234extern int sysctl_tcp_max_ssthresh; 235 236extern atomic_t tcp_memory_allocated; 237extern atomic_t tcp_sockets_allocated; 238extern int tcp_memory_pressure; 239 240/* 241 * The next routines deal with comparing 32 bit unsigned ints 242 * and worry about wraparound (automatic with unsigned arithmetic). 243 */ 244 245static inline int before(__u32 seq1, __u32 seq2) 246{ 247 return (__s32)(seq1-seq2) < 0; 248} 249#define after(seq2, seq1) before(seq1, seq2) 250 251/* is s2<=s1<=s3 ? */ 252static inline int between(__u32 seq1, __u32 seq2, __u32 seq3) 253{ 254 return seq3 - seq2 >= seq1 - seq2; 255} 256 257static inline int tcp_too_many_orphans(struct sock *sk, int num) 258{ 259 return (num > sysctl_tcp_max_orphans) || 260 (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 261 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]); 262} 263 264extern struct proto tcp_prot; 265 266DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics); 267#define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field) 268#define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field) 269#define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field) 270#define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field) 271#define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val) 272#define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val) 273 274extern void tcp_v4_err(struct sk_buff *skb, u32); 275 276extern void tcp_shutdown (struct sock *sk, int how); 277 278extern int tcp_v4_rcv(struct sk_buff *skb); 279 280extern int tcp_v4_remember_stamp(struct sock *sk); 281 282extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 283 284extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, 285 struct msghdr *msg, size_t size); 286extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags); 287 288extern int tcp_ioctl(struct sock *sk, 289 int cmd, 290 unsigned long arg); 291 292extern int tcp_rcv_state_process(struct sock *sk, 293 struct sk_buff *skb, 294 struct tcphdr *th, 295 unsigned len); 296 297extern int tcp_rcv_established(struct sock *sk, 298 struct sk_buff *skb, 299 struct tcphdr *th, 300 unsigned len); 301 302extern void tcp_rcv_space_adjust(struct sock *sk); 303 304extern void tcp_cleanup_rbuf(struct sock *sk, int copied); 305 306extern int tcp_twsk_unique(struct sock *sk, 307 struct sock *sktw, void *twp); 308 309extern void tcp_twsk_destructor(struct sock *sk); 310 311static inline void tcp_dec_quickack_mode(struct sock *sk, 312 const unsigned int pkts) 313{ 314 struct inet_connection_sock *icsk = inet_csk(sk); 315 316 if (icsk->icsk_ack.quick) { 317 if (pkts >= icsk->icsk_ack.quick) { 318 icsk->icsk_ack.quick = 0; 319 /* Leaving quickack mode we deflate ATO. */ 320 icsk->icsk_ack.ato = TCP_ATO_MIN; 321 } else 322 icsk->icsk_ack.quick -= pkts; 323 } 324} 325 326extern void tcp_enter_quickack_mode(struct sock *sk); 327 328static inline void tcp_clear_options(struct tcp_options_received *rx_opt) 329{ 330 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0; 331} 332 333enum tcp_tw_status 334{ 335 TCP_TW_SUCCESS = 0, 336 TCP_TW_RST = 1, 337 TCP_TW_ACK = 2, 338 TCP_TW_SYN = 3 339}; 340 341 342extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 343 struct sk_buff *skb, 344 const struct tcphdr *th); 345 346extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb, 347 struct request_sock *req, 348 struct request_sock **prev); 349extern int tcp_child_process(struct sock *parent, 350 struct sock *child, 351 struct sk_buff *skb); 352extern int tcp_use_frto(struct sock *sk); 353extern void tcp_enter_frto(struct sock *sk); 354extern void tcp_enter_loss(struct sock *sk, int how); 355extern void tcp_clear_retrans(struct tcp_sock *tp); 356extern void tcp_update_metrics(struct sock *sk); 357 358extern void tcp_close(struct sock *sk, 359 long timeout); 360extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait); 361 362extern int tcp_getsockopt(struct sock *sk, int level, 363 int optname, 364 char __user *optval, 365 int __user *optlen); 366extern int tcp_setsockopt(struct sock *sk, int level, 367 int optname, char __user *optval, 368 int optlen); 369extern int compat_tcp_getsockopt(struct sock *sk, 370 int level, int optname, 371 char __user *optval, int __user *optlen); 372extern int compat_tcp_setsockopt(struct sock *sk, 373 int level, int optname, 374 char __user *optval, int optlen); 375extern void tcp_set_keepalive(struct sock *sk, int val); 376extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, 377 struct msghdr *msg, 378 size_t len, int nonblock, 379 int flags, int *addr_len); 380 381extern void tcp_parse_options(struct sk_buff *skb, 382 struct tcp_options_received *opt_rx, 383 int estab); 384 385/* 386 * TCP v4 functions exported for the inet6 API 387 */ 388 389extern void tcp_v4_send_check(struct sock *sk, int len, 390 struct sk_buff *skb); 391 392extern int tcp_v4_conn_request(struct sock *sk, 393 struct sk_buff *skb); 394 395extern struct sock * tcp_create_openreq_child(struct sock *sk, 396 struct request_sock *req, 397 struct sk_buff *skb); 398 399extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, 400 struct sk_buff *skb, 401 struct request_sock *req, 402 struct dst_entry *dst); 403 404extern int tcp_v4_do_rcv(struct sock *sk, 405 struct sk_buff *skb); 406 407extern int tcp_v4_connect(struct sock *sk, 408 struct sockaddr *uaddr, 409 int addr_len); 410 411extern int tcp_connect(struct sock *sk); 412 413extern struct sk_buff * tcp_make_synack(struct sock *sk, 414 struct dst_entry *dst, 415 struct request_sock *req); 416 417extern int tcp_disconnect(struct sock *sk, int flags); 418 419extern void tcp_unhash(struct sock *sk); 420 421/* From syncookies.c */ 422extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 423 struct ip_options *opt); 424extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, 425 __u16 *mss); 426 427/* tcp_output.c */ 428 429extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 430 int nonagle); 431extern int tcp_may_send_now(struct sock *sk); 432extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); 433extern void tcp_xmit_retransmit_queue(struct sock *); 434extern void tcp_simple_retransmit(struct sock *); 435extern int tcp_trim_head(struct sock *, struct sk_buff *, u32); 436extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 437 438extern void tcp_send_probe0(struct sock *); 439extern void tcp_send_partial(struct sock *); 440extern int tcp_write_wakeup(struct sock *); 441extern void tcp_send_fin(struct sock *sk); 442extern void tcp_send_active_reset(struct sock *sk, gfp_t priority); 443extern int tcp_send_synack(struct sock *); 444extern void tcp_push_one(struct sock *, unsigned int mss_now); 445extern void tcp_send_ack(struct sock *sk); 446extern void tcp_send_delayed_ack(struct sock *sk); 447 448/* tcp_input.c */ 449extern void tcp_cwnd_application_limited(struct sock *sk); 450 451/* tcp_timer.c */ 452extern void tcp_init_xmit_timers(struct sock *); 453static inline void tcp_clear_xmit_timers(struct sock *sk) 454{ 455 inet_csk_clear_xmit_timers(sk); 456} 457 458extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 459extern unsigned int tcp_current_mss(struct sock *sk, int large); 460 461/* tcp.c */ 462extern void tcp_get_info(struct sock *, struct tcp_info *); 463 464/* Read 'sendfile()'-style from a TCP socket */ 465typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 466 unsigned int, size_t); 467extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 468 sk_read_actor_t recv_actor); 469 470extern void tcp_initialize_rcv_mss(struct sock *sk); 471 472extern int tcp_mtu_to_mss(struct sock *sk, int pmtu); 473extern int tcp_mss_to_mtu(struct sock *sk, int mss); 474extern void tcp_mtup_init(struct sock *sk); 475 476static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 477{ 478 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 479 ntohl(TCP_FLAG_ACK) | 480 snd_wnd); 481} 482 483static inline void tcp_fast_path_on(struct tcp_sock *tp) 484{ 485 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 486} 487 488static inline void tcp_fast_path_check(struct sock *sk) 489{ 490 struct tcp_sock *tp = tcp_sk(sk); 491 492 if (skb_queue_empty(&tp->out_of_order_queue) && 493 tp->rcv_wnd && 494 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 495 !tp->urg_data) 496 tcp_fast_path_on(tp); 497} 498 499/* Compute the actual receive window we are currently advertising. 500 * Rcv_nxt can be after the window if our peer push more data 501 * than the offered window. 502 */ 503static inline u32 tcp_receive_window(const struct tcp_sock *tp) 504{ 505 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 506 507 if (win < 0) 508 win = 0; 509 return (u32) win; 510} 511 512/* Choose a new window, without checks for shrinking, and without 513 * scaling applied to the result. The caller does these things 514 * if necessary. This is a "raw" window selection. 515 */ 516extern u32 __tcp_select_window(struct sock *sk); 517 518/* TCP timestamps are only 32-bits, this causes a slight 519 * complication on 64-bit systems since we store a snapshot 520 * of jiffies in the buffer control blocks below. We decided 521 * to use only the low 32-bits of jiffies and hide the ugly 522 * casts with the following macro. 523 */ 524#define tcp_time_stamp ((__u32)(jiffies)) 525 526struct tcp_skb_cb { 527 union { 528 struct inet_skb_parm h4; 529#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 530 struct inet6_skb_parm h6; 531#endif 532 } header; /* For incoming frames */ 533 __u32 seq; /* Starting sequence number */ 534 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 535 __u32 when; /* used to compute rtt's */ 536 __u8 flags; /* TCP header flags. */ 537 538 /* NOTE: These must match up to the flags byte in a 539 * real TCP header. 540 */ 541#define TCPCB_FLAG_FIN 0x01 542#define TCPCB_FLAG_SYN 0x02 543#define TCPCB_FLAG_RST 0x04 544#define TCPCB_FLAG_PSH 0x08 545#define TCPCB_FLAG_ACK 0x10 546#define TCPCB_FLAG_URG 0x20 547#define TCPCB_FLAG_ECE 0x40 548#define TCPCB_FLAG_CWR 0x80 549 550 __u8 sacked; /* State flags for SACK/FACK. */ 551#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 552#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 553#define TCPCB_LOST 0x04 /* SKB is lost */ 554#define TCPCB_TAGBITS 0x07 /* All tag bits */ 555 556#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 557#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 558 559#define TCPCB_URG 0x20 /* Urgent pointer advanced here */ 560 561#define TCPCB_AT_TAIL (TCPCB_URG) 562 563 __u16 urg_ptr; /* Valid w/URG flags is set. */ 564 __u32 ack_seq; /* Sequence number ACK'd */ 565}; 566 567#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 568 569#include <net/tcp_ecn.h> 570 571/* Due to TSO, an SKB can be composed of multiple actual 572 * packets. To keep these tracked properly, we use this. 573 */ 574static inline int tcp_skb_pcount(const struct sk_buff *skb) 575{ 576 return skb_shinfo(skb)->gso_segs; 577} 578 579/* This is valid iff tcp_skb_pcount() > 1. */ 580static inline int tcp_skb_mss(const struct sk_buff *skb) 581{ 582 return skb_shinfo(skb)->gso_size; 583} 584 585static inline void tcp_dec_pcount_approx(__u32 *count, 586 const struct sk_buff *skb) 587{ 588 if (*count) { 589 *count -= tcp_skb_pcount(skb); 590 if ((int)*count < 0) 591 *count = 0; 592 } 593} 594 595static inline void tcp_packets_out_inc(struct sock *sk, 596 const struct sk_buff *skb) 597{ 598 struct tcp_sock *tp = tcp_sk(sk); 599 int orig = tp->packets_out; 600 601 tp->packets_out += tcp_skb_pcount(skb); 602 if (!orig) 603 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, 604 inet_csk(sk)->icsk_rto, TCP_RTO_MAX); 605} 606 607static inline void tcp_packets_out_dec(struct tcp_sock *tp, 608 const struct sk_buff *skb) 609{ 610 tp->packets_out -= tcp_skb_pcount(skb); 611} 612 613/* Events passed to congestion control interface */ 614enum tcp_ca_event { 615 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 616 CA_EVENT_CWND_RESTART, /* congestion window restart */ 617 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 618 CA_EVENT_FRTO, /* fast recovery timeout */ 619 CA_EVENT_LOSS, /* loss timeout */ 620 CA_EVENT_FAST_ACK, /* in sequence ack */ 621 CA_EVENT_SLOW_ACK, /* other ack */ 622}; 623 624/* 625 * Interface for adding new TCP congestion control handlers 626 */ 627#define TCP_CA_NAME_MAX 16 628#define TCP_CA_MAX 128 629#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 630 631#define TCP_CONG_NON_RESTRICTED 0x1 632#define TCP_CONG_RTT_STAMP 0x2 633 634struct tcp_congestion_ops { 635 struct list_head list; 636 unsigned long flags; 637 638 /* initialize private data (optional) */ 639 void (*init)(struct sock *sk); 640 /* cleanup private data (optional) */ 641 void (*release)(struct sock *sk); 642 643 /* return slow start threshold (required) */ 644 u32 (*ssthresh)(struct sock *sk); 645 /* lower bound for congestion window (optional) */ 646 u32 (*min_cwnd)(const struct sock *sk); 647 /* do new cwnd calculation (required) */ 648 void (*cong_avoid)(struct sock *sk, u32 ack, 649 u32 rtt, u32 in_flight, int good_ack); 650 /* call before changing ca_state (optional) */ 651 void (*set_state)(struct sock *sk, u8 new_state); 652 /* call when cwnd event occurs (optional) */ 653 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 654 /* new value of cwnd after loss (optional) */ 655 u32 (*undo_cwnd)(struct sock *sk); 656 /* hook for packet ack accounting (optional) */ 657 void (*pkts_acked)(struct sock *sk, u32 num_acked, ktime_t last); 658 /* get info for inet_diag (optional) */ 659 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 660 661 char name[TCP_CA_NAME_MAX]; 662 struct module *owner; 663}; 664 665extern int tcp_register_congestion_control(struct tcp_congestion_ops *type); 666extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 667 668extern void tcp_init_congestion_control(struct sock *sk); 669extern void tcp_cleanup_congestion_control(struct sock *sk); 670extern int tcp_set_default_congestion_control(const char *name); 671extern void tcp_get_default_congestion_control(char *name); 672extern void tcp_get_available_congestion_control(char *buf, size_t len); 673extern void tcp_get_allowed_congestion_control(char *buf, size_t len); 674extern int tcp_set_allowed_congestion_control(char *allowed); 675extern int tcp_set_congestion_control(struct sock *sk, const char *name); 676extern void tcp_slow_start(struct tcp_sock *tp); 677 678extern struct tcp_congestion_ops tcp_init_congestion_ops; 679extern u32 tcp_reno_ssthresh(struct sock *sk); 680extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, 681 u32 rtt, u32 in_flight, int flag); 682extern u32 tcp_reno_min_cwnd(const struct sock *sk); 683extern struct tcp_congestion_ops tcp_reno; 684 685static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 686{ 687 struct inet_connection_sock *icsk = inet_csk(sk); 688 689 if (icsk->icsk_ca_ops->set_state) 690 icsk->icsk_ca_ops->set_state(sk, ca_state); 691 icsk->icsk_ca_state = ca_state; 692} 693 694static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 695{ 696 const struct inet_connection_sock *icsk = inet_csk(sk); 697 698 if (icsk->icsk_ca_ops->cwnd_event) 699 icsk->icsk_ca_ops->cwnd_event(sk, event); 700} 701 702/* This determines how many packets are "in the network" to the best 703 * of our knowledge. In many cases it is conservative, but where 704 * detailed information is available from the receiver (via SACK 705 * blocks etc.) we can make more aggressive calculations. 706 * 707 * Use this for decisions involving congestion control, use just 708 * tp->packets_out to determine if the send queue is empty or not. 709 * 710 * Read this equation as: 711 * 712 * "Packets sent once on transmission queue" MINUS 713 * "Packets left network, but not honestly ACKed yet" PLUS 714 * "Packets fast retransmitted" 715 */ 716static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 717{ 718 return (tp->packets_out - tp->left_out + tp->retrans_out); 719} 720 721/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 722 * The exception is rate halving phase, when cwnd is decreasing towards 723 * ssthresh. 724 */ 725static inline __u32 tcp_current_ssthresh(const struct sock *sk) 726{ 727 const struct tcp_sock *tp = tcp_sk(sk); 728 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery)) 729 return tp->snd_ssthresh; 730 else 731 return max(tp->snd_ssthresh, 732 ((tp->snd_cwnd >> 1) + 733 (tp->snd_cwnd >> 2))); 734} 735 736static inline void tcp_sync_left_out(struct tcp_sock *tp) 737{ 738 BUG_ON(tp->rx_opt.sack_ok && 739 (tp->sacked_out + tp->lost_out > tp->packets_out)); 740 tp->left_out = tp->sacked_out + tp->lost_out; 741} 742 743extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh); 744extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst); 745 746/* Slow start with delack produces 3 packets of burst, so that 747 * it is safe "de facto". 748 */ 749static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 750{ 751 return 3; 752} 753 754/* RFC2861 Check whether we are limited by application or congestion window 755 * This is the inverse of cwnd check in tcp_tso_should_defer 756 */ 757static inline int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight) 758{ 759 const struct tcp_sock *tp = tcp_sk(sk); 760 u32 left; 761 762 if (in_flight >= tp->snd_cwnd) 763 return 1; 764 765 if (!sk_can_gso(sk)) 766 return 0; 767 768 left = tp->snd_cwnd - in_flight; 769 if (sysctl_tcp_tso_win_divisor) 770 return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd; 771 else 772 return left <= tcp_max_burst(tp); 773} 774 775static inline void tcp_minshall_update(struct tcp_sock *tp, int mss, 776 const struct sk_buff *skb) 777{ 778 if (skb->len < mss) 779 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 780} 781 782static inline void tcp_check_probe_timer(struct sock *sk) 783{ 784 struct tcp_sock *tp = tcp_sk(sk); 785 const struct inet_connection_sock *icsk = inet_csk(sk); 786 787 if (!tp->packets_out && !icsk->icsk_pending) 788 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 789 icsk->icsk_rto, TCP_RTO_MAX); 790} 791 792static inline void tcp_push_pending_frames(struct sock *sk) 793{ 794 struct tcp_sock *tp = tcp_sk(sk); 795 796 __tcp_push_pending_frames(sk, tcp_current_mss(sk, 1), tp->nonagle); 797} 798 799static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq) 800{ 801 tp->snd_wl1 = seq; 802} 803 804static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq) 805{ 806 tp->snd_wl1 = seq; 807} 808 809/* 810 * Calculate(/check) TCP checksum 811 */ 812static inline __sum16 tcp_v4_check(int len, __be32 saddr, 813 __be32 daddr, __wsum base) 814{ 815 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 816} 817 818static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 819{ 820 return __skb_checksum_complete(skb); 821} 822 823static inline int tcp_checksum_complete(struct sk_buff *skb) 824{ 825 return !skb_csum_unnecessary(skb) && 826 __tcp_checksum_complete(skb); 827} 828 829/* Prequeue for VJ style copy to user, combined with checksumming. */ 830 831static inline void tcp_prequeue_init(struct tcp_sock *tp) 832{ 833 tp->ucopy.task = NULL; 834 tp->ucopy.len = 0; 835 tp->ucopy.memory = 0; 836 skb_queue_head_init(&tp->ucopy.prequeue); 837#ifdef CONFIG_NET_DMA 838 tp->ucopy.dma_chan = NULL; 839 tp->ucopy.wakeup = 0; 840 tp->ucopy.pinned_list = NULL; 841 tp->ucopy.dma_cookie = 0; 842#endif 843} 844 845/* Packet is added to VJ-style prequeue for processing in process 846 * context, if a reader task is waiting. Apparently, this exciting 847 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) 848 * failed somewhere. Latency? Burstiness? Well, at least now we will 849 * see, why it failed. 8)8) --ANK 850 * 851 * NOTE: is this not too big to inline? 852 */ 853static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb) 854{ 855 struct tcp_sock *tp = tcp_sk(sk); 856 857 if (!sysctl_tcp_low_latency && tp->ucopy.task) { 858 __skb_queue_tail(&tp->ucopy.prequeue, skb); 859 tp->ucopy.memory += skb->truesize; 860 if (tp->ucopy.memory > sk->sk_rcvbuf) { 861 struct sk_buff *skb1; 862 863 BUG_ON(sock_owned_by_user(sk)); 864 865 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { 866 sk->sk_backlog_rcv(sk, skb1); 867 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED); 868 } 869 870 tp->ucopy.memory = 0; 871 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { 872 wake_up_interruptible(sk->sk_sleep); 873 if (!inet_csk_ack_scheduled(sk)) 874 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 875 (3 * TCP_RTO_MIN) / 4, 876 TCP_RTO_MAX); 877 } 878 return 1; 879 } 880 return 0; 881} 882 883 884#undef STATE_TRACE 885 886#ifdef STATE_TRACE 887static const char *statename[]={ 888 "Unused","Established","Syn Sent","Syn Recv", 889 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 890 "Close Wait","Last ACK","Listen","Closing" 891}; 892#endif 893 894static inline void tcp_set_state(struct sock *sk, int state) 895{ 896 int oldstate = sk->sk_state; 897 898 switch (state) { 899 case TCP_ESTABLISHED: 900 if (oldstate != TCP_ESTABLISHED) 901 TCP_INC_STATS(TCP_MIB_CURRESTAB); 902 break; 903 904 case TCP_CLOSE: 905 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED) 906 TCP_INC_STATS(TCP_MIB_ESTABRESETS); 907 908 sk->sk_prot->unhash(sk); 909 if (inet_csk(sk)->icsk_bind_hash && 910 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) 911 inet_put_port(&tcp_hashinfo, sk); 912 /* fall through */ 913 default: 914 if (oldstate==TCP_ESTABLISHED) 915 TCP_DEC_STATS(TCP_MIB_CURRESTAB); 916 } 917 918 /* Change state AFTER socket is unhashed to avoid closed 919 * socket sitting in hash tables. 920 */ 921 sk->sk_state = state; 922 923#ifdef STATE_TRACE 924 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]); 925#endif 926} 927 928extern void tcp_done(struct sock *sk); 929 930static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 931{ 932 rx_opt->dsack = 0; 933 rx_opt->eff_sacks = 0; 934 rx_opt->num_sacks = 0; 935} 936 937/* Determine a window scaling and initial window to offer. */ 938extern void tcp_select_initial_window(int __space, __u32 mss, 939 __u32 *rcv_wnd, __u32 *window_clamp, 940 int wscale_ok, __u8 *rcv_wscale); 941 942static inline int tcp_win_from_space(int space) 943{ 944 return sysctl_tcp_adv_win_scale<=0 ? 945 (space>>(-sysctl_tcp_adv_win_scale)) : 946 space - (space>>sysctl_tcp_adv_win_scale); 947} 948 949/* Note: caller must be prepared to deal with negative returns */ 950static inline int tcp_space(const struct sock *sk) 951{ 952 return tcp_win_from_space(sk->sk_rcvbuf - 953 atomic_read(&sk->sk_rmem_alloc)); 954} 955 956static inline int tcp_full_space(const struct sock *sk) 957{ 958 return tcp_win_from_space(sk->sk_rcvbuf); 959} 960 961static inline void tcp_openreq_init(struct request_sock *req, 962 struct tcp_options_received *rx_opt, 963 struct sk_buff *skb) 964{ 965 struct inet_request_sock *ireq = inet_rsk(req); 966 967 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 968 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 969 req->mss = rx_opt->mss_clamp; 970 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 971 ireq->tstamp_ok = rx_opt->tstamp_ok; 972 ireq->sack_ok = rx_opt->sack_ok; 973 ireq->snd_wscale = rx_opt->snd_wscale; 974 ireq->wscale_ok = rx_opt->wscale_ok; 975 ireq->acked = 0; 976 ireq->ecn_ok = 0; 977 ireq->rmt_port = tcp_hdr(skb)->source; 978} 979 980extern void tcp_enter_memory_pressure(void); 981 982static inline int keepalive_intvl_when(const struct tcp_sock *tp) 983{ 984 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 985} 986 987static inline int keepalive_time_when(const struct tcp_sock *tp) 988{ 989 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 990} 991 992static inline int tcp_fin_time(const struct sock *sk) 993{ 994 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 995 const int rto = inet_csk(sk)->icsk_rto; 996 997 if (fin_timeout < (rto << 2) - (rto >> 1)) 998 fin_timeout = (rto << 2) - (rto >> 1); 999 1000 return fin_timeout; 1001} 1002 1003static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst) 1004{ 1005 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0) 1006 return 0; 1007 if (get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS) 1008 return 0; 1009 1010 /* RST segments are not recommended to carry timestamp, 1011 and, if they do, it is recommended to ignore PAWS because 1012 "their cleanup function should take precedence over timestamps." 1013 Certainly, it is mistake. It is necessary to understand the reasons 1014 of this constraint to relax it: if peer reboots, clock may go 1015 out-of-sync and half-open connections will not be reset. 1016 Actually, the problem would be not existing if all 1017 the implementations followed draft about maintaining clock 1018 via reboots. Linux-2.2 DOES NOT! 1019 1020 However, we can relax time bounds for RST segments to MSL. 1021 */ 1022 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1023 return 0; 1024 return 1; 1025} 1026 1027#define TCP_CHECK_TIMER(sk) do { } while (0) 1028 1029static inline void tcp_mib_init(void) 1030{ 1031 /* See RFC 2012 */ 1032 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1); 1033 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1034 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1035 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1); 1036} 1037 1038/*from STCP */ 1039static inline void clear_all_retrans_hints(struct tcp_sock *tp){ 1040 tp->lost_skb_hint = NULL; 1041 tp->scoreboard_skb_hint = NULL; 1042 tp->retransmit_skb_hint = NULL; 1043 tp->forward_skb_hint = NULL; 1044 tp->fastpath_skb_hint = NULL; 1045} 1046 1047/* MD5 Signature */ 1048struct crypto_hash; 1049 1050/* - key database */ 1051struct tcp_md5sig_key { 1052 u8 *key; 1053 u8 keylen; 1054}; 1055 1056struct tcp4_md5sig_key { 1057 u8 *key; 1058 u16 keylen; 1059 __be32 addr; 1060}; 1061 1062struct tcp6_md5sig_key { 1063 u8 *key; 1064 u16 keylen; 1065 struct in6_addr addr; 1066}; 1067 1068/* - sock block */ 1069struct tcp_md5sig_info { 1070 struct tcp4_md5sig_key *keys4; 1071#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1072 struct tcp6_md5sig_key *keys6; 1073 u32 entries6; 1074 u32 alloced6; 1075#endif 1076 u32 entries4; 1077 u32 alloced4; 1078}; 1079 1080/* - pseudo header */ 1081struct tcp4_pseudohdr { 1082 __be32 saddr; 1083 __be32 daddr; 1084 __u8 pad; 1085 __u8 protocol; 1086 __be16 len; 1087}; 1088 1089struct tcp6_pseudohdr { 1090 struct in6_addr saddr; 1091 struct in6_addr daddr; 1092 __be32 len; 1093 __be32 protocol; /* including padding */ 1094}; 1095 1096union tcp_md5sum_block { 1097 struct tcp4_pseudohdr ip4; 1098#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1099 struct tcp6_pseudohdr ip6; 1100#endif 1101}; 1102 1103/* - pool: digest algorithm, hash description and scratch buffer */ 1104struct tcp_md5sig_pool { 1105 struct hash_desc md5_desc; 1106 union tcp_md5sum_block md5_blk; 1107}; 1108 1109#define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */ 1110 1111/* - functions */ 1112extern int tcp_v4_calc_md5_hash(char *md5_hash, 1113 struct tcp_md5sig_key *key, 1114 struct sock *sk, 1115 struct dst_entry *dst, 1116 struct request_sock *req, 1117 struct tcphdr *th, 1118 int protocol, int tcplen); 1119extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1120 struct sock *addr_sk); 1121 1122extern int tcp_v4_md5_do_add(struct sock *sk, 1123 __be32 addr, 1124 u8 *newkey, 1125 u8 newkeylen); 1126 1127extern int tcp_v4_md5_do_del(struct sock *sk, 1128 __be32 addr); 1129 1130extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void); 1131extern void tcp_free_md5sig_pool(void); 1132 1133extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu); 1134extern void __tcp_put_md5sig_pool(void); 1135 1136static inline 1137struct tcp_md5sig_pool *tcp_get_md5sig_pool(void) 1138{ 1139 int cpu = get_cpu(); 1140 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu); 1141 if (!ret) 1142 put_cpu(); 1143 return ret; 1144} 1145 1146static inline void tcp_put_md5sig_pool(void) 1147{ 1148 __tcp_put_md5sig_pool(); 1149 put_cpu(); 1150} 1151 1152/* write queue abstraction */ 1153static inline void tcp_write_queue_purge(struct sock *sk) 1154{ 1155 struct sk_buff *skb; 1156 1157 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1158 sk_stream_free_skb(sk, skb); 1159 sk_stream_mem_reclaim(sk); 1160} 1161 1162static inline struct sk_buff *tcp_write_queue_head(struct sock *sk) 1163{ 1164 struct sk_buff *skb = sk->sk_write_queue.next; 1165 if (skb == (struct sk_buff *) &sk->sk_write_queue) 1166 return NULL; 1167 return skb; 1168} 1169 1170static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk) 1171{ 1172 struct sk_buff *skb = sk->sk_write_queue.prev; 1173 if (skb == (struct sk_buff *) &sk->sk_write_queue) 1174 return NULL; 1175 return skb; 1176} 1177 1178static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb) 1179{ 1180 return skb->next; 1181} 1182 1183#define tcp_for_write_queue(skb, sk) \ 1184 for (skb = (sk)->sk_write_queue.next; \ 1185 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \ 1186 skb = skb->next) 1187 1188#define tcp_for_write_queue_from(skb, sk) \ 1189 for (; (skb != (struct sk_buff *)&(sk)->sk_write_queue);\ 1190 skb = skb->next) 1191 1192static inline struct sk_buff *tcp_send_head(struct sock *sk) 1193{ 1194 return sk->sk_send_head; 1195} 1196 1197static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb) 1198{ 1199 struct tcp_sock *tp = tcp_sk(sk); 1200 1201 sk->sk_send_head = skb->next; 1202 if (sk->sk_send_head == (struct sk_buff *)&sk->sk_write_queue) 1203 sk->sk_send_head = NULL; 1204 /* Don't override Nagle indefinately with F-RTO */ 1205 if (tp->frto_counter == 2) 1206 tp->frto_counter = 3; 1207} 1208 1209static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1210{ 1211 if (sk->sk_send_head == skb_unlinked) 1212 sk->sk_send_head = NULL; 1213} 1214 1215static inline void tcp_init_send_head(struct sock *sk) 1216{ 1217 sk->sk_send_head = NULL; 1218} 1219 1220static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1221{ 1222 __skb_queue_tail(&sk->sk_write_queue, skb); 1223} 1224 1225static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1226{ 1227 __tcp_add_write_queue_tail(sk, skb); 1228 1229 /* Queue it, remembering where we must start sending. */ 1230 if (sk->sk_send_head == NULL) 1231 sk->sk_send_head = skb; 1232} 1233 1234static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1235{ 1236 __skb_queue_head(&sk->sk_write_queue, skb); 1237} 1238 1239/* Insert buff after skb on the write queue of sk. */ 1240static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1241 struct sk_buff *buff, 1242 struct sock *sk) 1243{ 1244 __skb_append(skb, buff, &sk->sk_write_queue); 1245} 1246 1247/* Insert skb between prev and next on the write queue of sk. */ 1248static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1249 struct sk_buff *skb, 1250 struct sock *sk) 1251{ 1252 __skb_insert(new, skb->prev, skb, &sk->sk_write_queue); 1253} 1254 1255static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1256{ 1257 __skb_unlink(skb, &sk->sk_write_queue); 1258} 1259 1260static inline int tcp_skb_is_last(const struct sock *sk, 1261 const struct sk_buff *skb) 1262{ 1263 return skb->next == (struct sk_buff *)&sk->sk_write_queue; 1264} 1265 1266static inline int tcp_write_queue_empty(struct sock *sk) 1267{ 1268 return skb_queue_empty(&sk->sk_write_queue); 1269} 1270 1271/* /proc */ 1272enum tcp_seq_states { 1273 TCP_SEQ_STATE_LISTENING, 1274 TCP_SEQ_STATE_OPENREQ, 1275 TCP_SEQ_STATE_ESTABLISHED, 1276 TCP_SEQ_STATE_TIME_WAIT, 1277}; 1278 1279struct tcp_seq_afinfo { 1280 struct module *owner; 1281 char *name; 1282 sa_family_t family; 1283 int (*seq_show) (struct seq_file *m, void *v); 1284 struct file_operations *seq_fops; 1285}; 1286 1287struct tcp_iter_state { 1288 sa_family_t family; 1289 enum tcp_seq_states state; 1290 struct sock *syn_wait_sk; 1291 int bucket, sbucket, num, uid; 1292 struct seq_operations seq_ops; 1293}; 1294 1295extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo); 1296extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo); 1297 1298extern struct request_sock_ops tcp_request_sock_ops; 1299 1300extern int tcp_v4_destroy_sock(struct sock *sk); 1301 1302extern int tcp_v4_gso_send_check(struct sk_buff *skb); 1303extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features); 1304#ifdef CONFIG_INET_GRO 1305extern struct sk_buff **tcp_gro_receive(struct sk_buff **head, 1306 struct sk_buff *skb); 1307extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head, 1308 struct sk_buff *skb); 1309extern int tcp_gro_complete(struct sk_buff *skb); 1310extern int tcp4_gro_complete(struct sk_buff *skb); 1311#endif /* CONFIG_INET_GRO */ 1312 1313#ifdef CONFIG_PROC_FS 1314extern int tcp4_proc_init(void); 1315extern void tcp4_proc_exit(void); 1316#endif 1317 1318/* TCP af-specific functions */ 1319struct tcp_sock_af_ops { 1320#ifdef CONFIG_TCP_MD5SIG 1321 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1322 struct sock *addr_sk); 1323 int (*calc_md5_hash) (char *location, 1324 struct tcp_md5sig_key *md5, 1325 struct sock *sk, 1326 struct dst_entry *dst, 1327 struct request_sock *req, 1328 struct tcphdr *th, 1329 int protocol, int len); 1330 int (*md5_add) (struct sock *sk, 1331 struct sock *addr_sk, 1332 u8 *newkey, 1333 u8 len); 1334 int (*md5_parse) (struct sock *sk, 1335 char __user *optval, 1336 int optlen); 1337#endif 1338}; 1339 1340struct tcp_request_sock_ops { 1341#ifdef CONFIG_TCP_MD5SIG 1342 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1343 struct request_sock *req); 1344#endif 1345}; 1346 1347extern void tcp_v4_init(struct net_proto_family *ops); 1348extern void tcp_init(void); 1349 1350#endif /* _TCP_H */ 1351