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 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 /* The least MTU to use for probing */ 64 #define TCP_BASE_MSS 512 65 66 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 67 #define TCP_FASTRETRANS_THRESH 3 68 69 /* Maximal reordering. */ 70 #define TCP_MAX_REORDERING 127 71 72 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 73 #define TCP_MAX_QUICKACKS 16U 74 75 /* urg_data states */ 76 #define TCP_URG_VALID 0x0100 77 #define TCP_URG_NOTYET 0x0200 78 #define TCP_URG_READ 0x0400 79 80 #define TCP_RETR1 3 /* 81 * This is how many retries it does before it 82 * tries to figure out if the gateway is 83 * down. Minimal RFC value is 3; it corresponds 84 * to ~3sec-8min depending on RTO. 85 */ 86 87 #define TCP_RETR2 15 /* 88 * This should take at least 89 * 90 minutes to time out. 90 * RFC1122 says that the limit is 100 sec. 91 * 15 is ~13-30min depending on RTO. 92 */ 93 94 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a 95 * connection: ~180sec is RFC minimum */ 96 97 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a 98 * connection: ~180sec is RFC minimum */ 99 100 101 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned 102 * socket. 7 is ~50sec-16min. 103 */ 104 105 106 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 107 * state, about 60 seconds */ 108 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 109 /* BSD style FIN_WAIT2 deadlock breaker. 110 * It used to be 3min, new value is 60sec, 111 * to combine FIN-WAIT-2 timeout with 112 * TIME-WAIT timer. 113 */ 114 115 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 116 #if HZ >= 100 117 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 118 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 119 #else 120 #define TCP_DELACK_MIN 4U 121 #define TCP_ATO_MIN 4U 122 #endif 123 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 124 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 125 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */ 126 127 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 128 * for local resources. 129 */ 130 131 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 132 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 133 #define TCP_KEEPALIVE_INTVL (75*HZ) 134 135 #define MAX_TCP_KEEPIDLE 32767 136 #define MAX_TCP_KEEPINTVL 32767 137 #define MAX_TCP_KEEPCNT 127 138 #define MAX_TCP_SYNCNT 127 139 140 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 141 #define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */ 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 */ 165 166 /* 167 * TCP option lengths 168 */ 169 170 #define TCPOLEN_MSS 4 171 #define TCPOLEN_WINDOW 3 172 #define TCPOLEN_SACK_PERM 2 173 #define TCPOLEN_TIMESTAMP 10 174 175 /* But this is what stacks really send out. */ 176 #define TCPOLEN_TSTAMP_ALIGNED 12 177 #define TCPOLEN_WSCALE_ALIGNED 4 178 #define TCPOLEN_SACKPERM_ALIGNED 4 179 #define TCPOLEN_SACK_BASE 2 180 #define TCPOLEN_SACK_BASE_ALIGNED 4 181 #define TCPOLEN_SACK_PERBLOCK 8 182 183 /* Flags in tp->nonagle */ 184 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 185 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 186 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 187 188 extern struct inet_timewait_death_row tcp_death_row; 189 190 /* sysctl variables for tcp */ 191 extern int sysctl_tcp_timestamps; 192 extern int sysctl_tcp_window_scaling; 193 extern int sysctl_tcp_sack; 194 extern int sysctl_tcp_fin_timeout; 195 extern int sysctl_tcp_keepalive_time; 196 extern int sysctl_tcp_keepalive_probes; 197 extern int sysctl_tcp_keepalive_intvl; 198 extern int sysctl_tcp_syn_retries; 199 extern int sysctl_tcp_synack_retries; 200 extern int sysctl_tcp_retries1; 201 extern int sysctl_tcp_retries2; 202 extern int sysctl_tcp_orphan_retries; 203 extern int sysctl_tcp_syncookies; 204 extern int sysctl_tcp_retrans_collapse; 205 extern int sysctl_tcp_stdurg; 206 extern int sysctl_tcp_rfc1337; 207 extern int sysctl_tcp_abort_on_overflow; 208 extern int sysctl_tcp_max_orphans; 209 extern int sysctl_tcp_fack; 210 extern int sysctl_tcp_reordering; 211 extern int sysctl_tcp_ecn; 212 extern int sysctl_tcp_dsack; 213 extern int sysctl_tcp_mem[3]; 214 extern int sysctl_tcp_wmem[3]; 215 extern int sysctl_tcp_rmem[3]; 216 extern int sysctl_tcp_app_win; 217 extern int sysctl_tcp_adv_win_scale; 218 extern int sysctl_tcp_tw_reuse; 219 extern int sysctl_tcp_frto; 220 extern int sysctl_tcp_low_latency; 221 extern int sysctl_tcp_dma_copybreak; 222 extern int sysctl_tcp_nometrics_save; 223 extern int sysctl_tcp_moderate_rcvbuf; 224 extern int sysctl_tcp_tso_win_divisor; 225 extern int sysctl_tcp_abc; 226 extern int sysctl_tcp_mtu_probing; 227 extern int sysctl_tcp_base_mss; 228 extern int sysctl_tcp_workaround_signed_windows; 229 extern int sysctl_tcp_slow_start_after_idle; 230 231 extern atomic_t tcp_memory_allocated; 232 extern atomic_t tcp_sockets_allocated; 233 extern int tcp_memory_pressure; 234 235 /* 236 * The next routines deal with comparing 32 bit unsigned ints 237 * and worry about wraparound (automatic with unsigned arithmetic). 238 */ 239 240 static inline int before(__u32 seq1, __u32 seq2) 241 { 242 return (__s32)(seq1-seq2) < 0; 243 } 244 245 static inline int after(__u32 seq1, __u32 seq2) 246 { 247 return (__s32)(seq2-seq1) < 0; 248 } 249 250 251 /* is s2<=s1<=s3 ? */ 252 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3) 253 { 254 return seq3 - seq2 >= seq1 - seq2; 255 } 256 257 258 extern struct proto tcp_prot; 259 260 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics); 261 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field) 262 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field) 263 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field) 264 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field) 265 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val) 266 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val) 267 268 extern void tcp_v4_err(struct sk_buff *skb, u32); 269 270 extern void tcp_shutdown (struct sock *sk, int how); 271 272 extern int tcp_v4_rcv(struct sk_buff *skb); 273 274 extern int tcp_v4_remember_stamp(struct sock *sk); 275 276 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 277 278 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, 279 struct msghdr *msg, size_t size); 280 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags); 281 282 extern int tcp_ioctl(struct sock *sk, 283 int cmd, 284 unsigned long arg); 285 286 extern int tcp_rcv_state_process(struct sock *sk, 287 struct sk_buff *skb, 288 struct tcphdr *th, 289 unsigned len); 290 291 extern int tcp_rcv_established(struct sock *sk, 292 struct sk_buff *skb, 293 struct tcphdr *th, 294 unsigned len); 295 296 extern void tcp_rcv_space_adjust(struct sock *sk); 297 298 extern void tcp_cleanup_rbuf(struct sock *sk, int copied); 299 300 extern int tcp_twsk_unique(struct sock *sk, 301 struct sock *sktw, void *twp); 302 303 static inline void tcp_dec_quickack_mode(struct sock *sk, 304 const unsigned int pkts) 305 { 306 struct inet_connection_sock *icsk = inet_csk(sk); 307 308 if (icsk->icsk_ack.quick) { 309 if (pkts >= icsk->icsk_ack.quick) { 310 icsk->icsk_ack.quick = 0; 311 /* Leaving quickack mode we deflate ATO. */ 312 icsk->icsk_ack.ato = TCP_ATO_MIN; 313 } else 314 icsk->icsk_ack.quick -= pkts; 315 } 316 } 317 318 extern void tcp_enter_quickack_mode(struct sock *sk); 319 320 static inline void tcp_clear_options(struct tcp_options_received *rx_opt) 321 { 322 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0; 323 } 324 325 enum tcp_tw_status 326 { 327 TCP_TW_SUCCESS = 0, 328 TCP_TW_RST = 1, 329 TCP_TW_ACK = 2, 330 TCP_TW_SYN = 3 331 }; 332 333 334 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 335 struct sk_buff *skb, 336 const struct tcphdr *th); 337 338 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb, 339 struct request_sock *req, 340 struct request_sock **prev); 341 extern int tcp_child_process(struct sock *parent, 342 struct sock *child, 343 struct sk_buff *skb); 344 extern void tcp_enter_frto(struct sock *sk); 345 extern void tcp_enter_loss(struct sock *sk, int how); 346 extern void tcp_clear_retrans(struct tcp_sock *tp); 347 extern void tcp_update_metrics(struct sock *sk); 348 349 extern void tcp_close(struct sock *sk, 350 long timeout); 351 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait); 352 353 extern int tcp_getsockopt(struct sock *sk, int level, 354 int optname, 355 char __user *optval, 356 int __user *optlen); 357 extern int tcp_setsockopt(struct sock *sk, int level, 358 int optname, char __user *optval, 359 int optlen); 360 extern int compat_tcp_getsockopt(struct sock *sk, 361 int level, int optname, 362 char __user *optval, int __user *optlen); 363 extern int compat_tcp_setsockopt(struct sock *sk, 364 int level, int optname, 365 char __user *optval, int optlen); 366 extern void tcp_set_keepalive(struct sock *sk, int val); 367 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, 368 struct msghdr *msg, 369 size_t len, int nonblock, 370 int flags, int *addr_len); 371 372 extern void tcp_parse_options(struct sk_buff *skb, 373 struct tcp_options_received *opt_rx, 374 int estab); 375 376 /* 377 * TCP v4 functions exported for the inet6 API 378 */ 379 380 extern void tcp_v4_send_check(struct sock *sk, int len, 381 struct sk_buff *skb); 382 383 extern int tcp_v4_conn_request(struct sock *sk, 384 struct sk_buff *skb); 385 386 extern struct sock * tcp_create_openreq_child(struct sock *sk, 387 struct request_sock *req, 388 struct sk_buff *skb); 389 390 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, 391 struct sk_buff *skb, 392 struct request_sock *req, 393 struct dst_entry *dst); 394 395 extern int tcp_v4_do_rcv(struct sock *sk, 396 struct sk_buff *skb); 397 398 extern int tcp_v4_connect(struct sock *sk, 399 struct sockaddr *uaddr, 400 int addr_len); 401 402 extern int tcp_connect(struct sock *sk); 403 404 extern struct sk_buff * tcp_make_synack(struct sock *sk, 405 struct dst_entry *dst, 406 struct request_sock *req); 407 408 extern int tcp_disconnect(struct sock *sk, int flags); 409 410 extern void tcp_unhash(struct sock *sk); 411 412 /* From syncookies.c */ 413 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 414 struct ip_options *opt); 415 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, 416 __u16 *mss); 417 418 /* tcp_output.c */ 419 420 extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp, 421 unsigned int cur_mss, int nonagle); 422 extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp); 423 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); 424 extern void tcp_xmit_retransmit_queue(struct sock *); 425 extern void tcp_simple_retransmit(struct sock *); 426 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32); 427 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 428 429 extern void tcp_send_probe0(struct sock *); 430 extern void tcp_send_partial(struct sock *); 431 extern int tcp_write_wakeup(struct sock *); 432 extern void tcp_send_fin(struct sock *sk); 433 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority); 434 extern int tcp_send_synack(struct sock *); 435 extern void tcp_push_one(struct sock *, unsigned int mss_now); 436 extern void tcp_send_ack(struct sock *sk); 437 extern void tcp_send_delayed_ack(struct sock *sk); 438 439 /* tcp_input.c */ 440 extern void tcp_cwnd_application_limited(struct sock *sk); 441 442 /* tcp_timer.c */ 443 extern void tcp_init_xmit_timers(struct sock *); 444 static inline void tcp_clear_xmit_timers(struct sock *sk) 445 { 446 inet_csk_clear_xmit_timers(sk); 447 } 448 449 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 450 extern unsigned int tcp_current_mss(struct sock *sk, int large); 451 452 /* tcp.c */ 453 extern void tcp_get_info(struct sock *, struct tcp_info *); 454 455 /* Read 'sendfile()'-style from a TCP socket */ 456 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 457 unsigned int, size_t); 458 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 459 sk_read_actor_t recv_actor); 460 461 extern void tcp_initialize_rcv_mss(struct sock *sk); 462 463 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu); 464 extern int tcp_mss_to_mtu(struct sock *sk, int mss); 465 extern void tcp_mtup_init(struct sock *sk); 466 467 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 468 { 469 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 470 ntohl(TCP_FLAG_ACK) | 471 snd_wnd); 472 } 473 474 static inline void tcp_fast_path_on(struct tcp_sock *tp) 475 { 476 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 477 } 478 479 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp) 480 { 481 if (skb_queue_empty(&tp->out_of_order_queue) && 482 tp->rcv_wnd && 483 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 484 !tp->urg_data) 485 tcp_fast_path_on(tp); 486 } 487 488 /* Compute the actual receive window we are currently advertising. 489 * Rcv_nxt can be after the window if our peer push more data 490 * than the offered window. 491 */ 492 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 493 { 494 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 495 496 if (win < 0) 497 win = 0; 498 return (u32) win; 499 } 500 501 /* Choose a new window, without checks for shrinking, and without 502 * scaling applied to the result. The caller does these things 503 * if necessary. This is a "raw" window selection. 504 */ 505 extern u32 __tcp_select_window(struct sock *sk); 506 507 /* TCP timestamps are only 32-bits, this causes a slight 508 * complication on 64-bit systems since we store a snapshot 509 * of jiffies in the buffer control blocks below. We decided 510 * to use only the low 32-bits of jiffies and hide the ugly 511 * casts with the following macro. 512 */ 513 #define tcp_time_stamp ((__u32)(jiffies)) 514 515 /* This is what the send packet queuing engine uses to pass 516 * TCP per-packet control information to the transmission 517 * code. We also store the host-order sequence numbers in 518 * here too. This is 36 bytes on 32-bit architectures, 519 * 40 bytes on 64-bit machines, if this grows please adjust 520 * skbuff.h:skbuff->cb[xxx] size appropriately. 521 */ 522 struct tcp_skb_cb { 523 union { 524 struct inet_skb_parm h4; 525 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE) 526 struct inet6_skb_parm h6; 527 #endif 528 } header; /* For incoming frames */ 529 __u32 seq; /* Starting sequence number */ 530 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 531 __u32 when; /* used to compute rtt's */ 532 __u8 flags; /* TCP header flags. */ 533 534 /* NOTE: These must match up to the flags byte in a 535 * real TCP header. 536 */ 537 #define TCPCB_FLAG_FIN 0x01 538 #define TCPCB_FLAG_SYN 0x02 539 #define TCPCB_FLAG_RST 0x04 540 #define TCPCB_FLAG_PSH 0x08 541 #define TCPCB_FLAG_ACK 0x10 542 #define TCPCB_FLAG_URG 0x20 543 #define TCPCB_FLAG_ECE 0x40 544 #define TCPCB_FLAG_CWR 0x80 545 546 __u8 sacked; /* State flags for SACK/FACK. */ 547 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 548 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 549 #define TCPCB_LOST 0x04 /* SKB is lost */ 550 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 551 552 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 553 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 554 555 #define TCPCB_URG 0x20 /* Urgent pointer advanced here */ 556 557 #define TCPCB_AT_TAIL (TCPCB_URG) 558 559 __u16 urg_ptr; /* Valid w/URG flags is set. */ 560 __u32 ack_seq; /* Sequence number ACK'd */ 561 }; 562 563 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 564 565 #include <net/tcp_ecn.h> 566 567 /* Due to TSO, an SKB can be composed of multiple actual 568 * packets. To keep these tracked properly, we use this. 569 */ 570 static inline int tcp_skb_pcount(const struct sk_buff *skb) 571 { 572 return skb_shinfo(skb)->tso_segs; 573 } 574 575 /* This is valid iff tcp_skb_pcount() > 1. */ 576 static inline int tcp_skb_mss(const struct sk_buff *skb) 577 { 578 return skb_shinfo(skb)->tso_size; 579 } 580 581 static inline void tcp_dec_pcount_approx(__u32 *count, 582 const struct sk_buff *skb) 583 { 584 if (*count) { 585 *count -= tcp_skb_pcount(skb); 586 if ((int)*count < 0) 587 *count = 0; 588 } 589 } 590 591 static inline void tcp_packets_out_inc(struct sock *sk, 592 struct tcp_sock *tp, 593 const struct sk_buff *skb) 594 { 595 int orig = tp->packets_out; 596 597 tp->packets_out += tcp_skb_pcount(skb); 598 if (!orig) 599 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, 600 inet_csk(sk)->icsk_rto, TCP_RTO_MAX); 601 } 602 603 static inline void tcp_packets_out_dec(struct tcp_sock *tp, 604 const struct sk_buff *skb) 605 { 606 tp->packets_out -= tcp_skb_pcount(skb); 607 } 608 609 /* Events passed to congestion control interface */ 610 enum tcp_ca_event { 611 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 612 CA_EVENT_CWND_RESTART, /* congestion window restart */ 613 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 614 CA_EVENT_FRTO, /* fast recovery timeout */ 615 CA_EVENT_LOSS, /* loss timeout */ 616 CA_EVENT_FAST_ACK, /* in sequence ack */ 617 CA_EVENT_SLOW_ACK, /* other ack */ 618 }; 619 620 /* 621 * Interface for adding new TCP congestion control handlers 622 */ 623 #define TCP_CA_NAME_MAX 16 624 struct tcp_congestion_ops { 625 struct list_head list; 626 627 /* initialize private data (optional) */ 628 void (*init)(struct sock *sk); 629 /* cleanup private data (optional) */ 630 void (*release)(struct sock *sk); 631 632 /* return slow start threshold (required) */ 633 u32 (*ssthresh)(struct sock *sk); 634 /* lower bound for congestion window (optional) */ 635 u32 (*min_cwnd)(const struct sock *sk); 636 /* do new cwnd calculation (required) */ 637 void (*cong_avoid)(struct sock *sk, u32 ack, 638 u32 rtt, u32 in_flight, int good_ack); 639 /* round trip time sample per acked packet (optional) */ 640 void (*rtt_sample)(struct sock *sk, u32 usrtt); 641 /* call before changing ca_state (optional) */ 642 void (*set_state)(struct sock *sk, u8 new_state); 643 /* call when cwnd event occurs (optional) */ 644 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 645 /* new value of cwnd after loss (optional) */ 646 u32 (*undo_cwnd)(struct sock *sk); 647 /* hook for packet ack accounting (optional) */ 648 void (*pkts_acked)(struct sock *sk, u32 num_acked); 649 /* get info for inet_diag (optional) */ 650 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 651 652 char name[TCP_CA_NAME_MAX]; 653 struct module *owner; 654 }; 655 656 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type); 657 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 658 659 extern void tcp_init_congestion_control(struct sock *sk); 660 extern void tcp_cleanup_congestion_control(struct sock *sk); 661 extern int tcp_set_default_congestion_control(const char *name); 662 extern void tcp_get_default_congestion_control(char *name); 663 extern int tcp_set_congestion_control(struct sock *sk, const char *name); 664 extern void tcp_slow_start(struct tcp_sock *tp); 665 666 extern struct tcp_congestion_ops tcp_init_congestion_ops; 667 extern u32 tcp_reno_ssthresh(struct sock *sk); 668 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, 669 u32 rtt, u32 in_flight, int flag); 670 extern u32 tcp_reno_min_cwnd(const struct sock *sk); 671 extern struct tcp_congestion_ops tcp_reno; 672 673 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 674 { 675 struct inet_connection_sock *icsk = inet_csk(sk); 676 677 if (icsk->icsk_ca_ops->set_state) 678 icsk->icsk_ca_ops->set_state(sk, ca_state); 679 icsk->icsk_ca_state = ca_state; 680 } 681 682 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 683 { 684 const struct inet_connection_sock *icsk = inet_csk(sk); 685 686 if (icsk->icsk_ca_ops->cwnd_event) 687 icsk->icsk_ca_ops->cwnd_event(sk, event); 688 } 689 690 /* This determines how many packets are "in the network" to the best 691 * of our knowledge. In many cases it is conservative, but where 692 * detailed information is available from the receiver (via SACK 693 * blocks etc.) we can make more aggressive calculations. 694 * 695 * Use this for decisions involving congestion control, use just 696 * tp->packets_out to determine if the send queue is empty or not. 697 * 698 * Read this equation as: 699 * 700 * "Packets sent once on transmission queue" MINUS 701 * "Packets left network, but not honestly ACKed yet" PLUS 702 * "Packets fast retransmitted" 703 */ 704 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 705 { 706 return (tp->packets_out - tp->left_out + tp->retrans_out); 707 } 708 709 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 710 * The exception is rate halving phase, when cwnd is decreasing towards 711 * ssthresh. 712 */ 713 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 714 { 715 const struct tcp_sock *tp = tcp_sk(sk); 716 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery)) 717 return tp->snd_ssthresh; 718 else 719 return max(tp->snd_ssthresh, 720 ((tp->snd_cwnd >> 1) + 721 (tp->snd_cwnd >> 2))); 722 } 723 724 static inline void tcp_sync_left_out(struct tcp_sock *tp) 725 { 726 if (tp->rx_opt.sack_ok && 727 (tp->sacked_out >= tp->packets_out - tp->lost_out)) 728 tp->sacked_out = tp->packets_out - tp->lost_out; 729 tp->left_out = tp->sacked_out + tp->lost_out; 730 } 731 732 extern void tcp_enter_cwr(struct sock *sk); 733 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst); 734 735 /* Slow start with delack produces 3 packets of burst, so that 736 * it is safe "de facto". 737 */ 738 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 739 { 740 return 3; 741 } 742 743 /* RFC2861 Check whether we are limited by application or congestion window 744 * This is the inverse of cwnd check in tcp_tso_should_defer 745 */ 746 static inline int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight) 747 { 748 const struct tcp_sock *tp = tcp_sk(sk); 749 u32 left; 750 751 if (in_flight >= tp->snd_cwnd) 752 return 1; 753 754 if (!(sk->sk_route_caps & NETIF_F_TSO)) 755 return 0; 756 757 left = tp->snd_cwnd - in_flight; 758 if (sysctl_tcp_tso_win_divisor) 759 return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd; 760 else 761 return left <= tcp_max_burst(tp); 762 } 763 764 static inline void tcp_minshall_update(struct tcp_sock *tp, int mss, 765 const struct sk_buff *skb) 766 { 767 if (skb->len < mss) 768 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 769 } 770 771 static inline void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp) 772 { 773 const struct inet_connection_sock *icsk = inet_csk(sk); 774 if (!tp->packets_out && !icsk->icsk_pending) 775 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 776 icsk->icsk_rto, TCP_RTO_MAX); 777 } 778 779 static inline void tcp_push_pending_frames(struct sock *sk, 780 struct tcp_sock *tp) 781 { 782 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle); 783 } 784 785 static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq) 786 { 787 tp->snd_wl1 = seq; 788 } 789 790 static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq) 791 { 792 tp->snd_wl1 = seq; 793 } 794 795 /* 796 * Calculate(/check) TCP checksum 797 */ 798 static inline u16 tcp_v4_check(struct tcphdr *th, int len, 799 unsigned long saddr, unsigned long daddr, 800 unsigned long base) 801 { 802 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 803 } 804 805 static inline int __tcp_checksum_complete(struct sk_buff *skb) 806 { 807 return __skb_checksum_complete(skb); 808 } 809 810 static inline int tcp_checksum_complete(struct sk_buff *skb) 811 { 812 return skb->ip_summed != CHECKSUM_UNNECESSARY && 813 __tcp_checksum_complete(skb); 814 } 815 816 /* Prequeue for VJ style copy to user, combined with checksumming. */ 817 818 static inline void tcp_prequeue_init(struct tcp_sock *tp) 819 { 820 tp->ucopy.task = NULL; 821 tp->ucopy.len = 0; 822 tp->ucopy.memory = 0; 823 skb_queue_head_init(&tp->ucopy.prequeue); 824 #ifdef CONFIG_NET_DMA 825 tp->ucopy.dma_chan = NULL; 826 tp->ucopy.wakeup = 0; 827 tp->ucopy.pinned_list = NULL; 828 tp->ucopy.dma_cookie = 0; 829 #endif 830 } 831 832 /* Packet is added to VJ-style prequeue for processing in process 833 * context, if a reader task is waiting. Apparently, this exciting 834 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) 835 * failed somewhere. Latency? Burstiness? Well, at least now we will 836 * see, why it failed. 8)8) --ANK 837 * 838 * NOTE: is this not too big to inline? 839 */ 840 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb) 841 { 842 struct tcp_sock *tp = tcp_sk(sk); 843 844 if (!sysctl_tcp_low_latency && tp->ucopy.task) { 845 __skb_queue_tail(&tp->ucopy.prequeue, skb); 846 tp->ucopy.memory += skb->truesize; 847 if (tp->ucopy.memory > sk->sk_rcvbuf) { 848 struct sk_buff *skb1; 849 850 BUG_ON(sock_owned_by_user(sk)); 851 852 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { 853 sk->sk_backlog_rcv(sk, skb1); 854 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED); 855 } 856 857 tp->ucopy.memory = 0; 858 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { 859 wake_up_interruptible(sk->sk_sleep); 860 if (!inet_csk_ack_scheduled(sk)) 861 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 862 (3 * TCP_RTO_MIN) / 4, 863 TCP_RTO_MAX); 864 } 865 return 1; 866 } 867 return 0; 868 } 869 870 871 #undef STATE_TRACE 872 873 #ifdef STATE_TRACE 874 static const char *statename[]={ 875 "Unused","Established","Syn Sent","Syn Recv", 876 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 877 "Close Wait","Last ACK","Listen","Closing" 878 }; 879 #endif 880 881 static inline void tcp_set_state(struct sock *sk, int state) 882 { 883 int oldstate = sk->sk_state; 884 885 switch (state) { 886 case TCP_ESTABLISHED: 887 if (oldstate != TCP_ESTABLISHED) 888 TCP_INC_STATS(TCP_MIB_CURRESTAB); 889 break; 890 891 case TCP_CLOSE: 892 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED) 893 TCP_INC_STATS(TCP_MIB_ESTABRESETS); 894 895 sk->sk_prot->unhash(sk); 896 if (inet_csk(sk)->icsk_bind_hash && 897 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) 898 inet_put_port(&tcp_hashinfo, sk); 899 /* fall through */ 900 default: 901 if (oldstate==TCP_ESTABLISHED) 902 TCP_DEC_STATS(TCP_MIB_CURRESTAB); 903 } 904 905 /* Change state AFTER socket is unhashed to avoid closed 906 * socket sitting in hash tables. 907 */ 908 sk->sk_state = state; 909 910 #ifdef STATE_TRACE 911 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]); 912 #endif 913 } 914 915 static inline void tcp_done(struct sock *sk) 916 { 917 tcp_set_state(sk, TCP_CLOSE); 918 tcp_clear_xmit_timers(sk); 919 920 sk->sk_shutdown = SHUTDOWN_MASK; 921 922 if (!sock_flag(sk, SOCK_DEAD)) 923 sk->sk_state_change(sk); 924 else 925 inet_csk_destroy_sock(sk); 926 } 927 928 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 929 { 930 rx_opt->dsack = 0; 931 rx_opt->eff_sacks = 0; 932 rx_opt->num_sacks = 0; 933 } 934 935 /* Determine a window scaling and initial window to offer. */ 936 extern void tcp_select_initial_window(int __space, __u32 mss, 937 __u32 *rcv_wnd, __u32 *window_clamp, 938 int wscale_ok, __u8 *rcv_wscale); 939 940 static inline int tcp_win_from_space(int space) 941 { 942 return sysctl_tcp_adv_win_scale<=0 ? 943 (space>>(-sysctl_tcp_adv_win_scale)) : 944 space - (space>>sysctl_tcp_adv_win_scale); 945 } 946 947 /* Note: caller must be prepared to deal with negative returns */ 948 static inline int tcp_space(const struct sock *sk) 949 { 950 return tcp_win_from_space(sk->sk_rcvbuf - 951 atomic_read(&sk->sk_rmem_alloc)); 952 } 953 954 static inline int tcp_full_space(const struct sock *sk) 955 { 956 return tcp_win_from_space(sk->sk_rcvbuf); 957 } 958 959 static inline void tcp_openreq_init(struct request_sock *req, 960 struct tcp_options_received *rx_opt, 961 struct sk_buff *skb) 962 { 963 struct inet_request_sock *ireq = inet_rsk(req); 964 965 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 966 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 967 req->mss = rx_opt->mss_clamp; 968 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 969 ireq->tstamp_ok = rx_opt->tstamp_ok; 970 ireq->sack_ok = rx_opt->sack_ok; 971 ireq->snd_wscale = rx_opt->snd_wscale; 972 ireq->wscale_ok = rx_opt->wscale_ok; 973 ireq->acked = 0; 974 ireq->ecn_ok = 0; 975 ireq->rmt_port = skb->h.th->source; 976 } 977 978 extern void tcp_enter_memory_pressure(void); 979 980 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 981 { 982 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 983 } 984 985 static inline int keepalive_time_when(const struct tcp_sock *tp) 986 { 987 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 988 } 989 990 static inline int tcp_fin_time(const struct sock *sk) 991 { 992 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 993 const int rto = inet_csk(sk)->icsk_rto; 994 995 if (fin_timeout < (rto << 2) - (rto >> 1)) 996 fin_timeout = (rto << 2) - (rto >> 1); 997 998 return fin_timeout; 999 } 1000 1001 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst) 1002 { 1003 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0) 1004 return 0; 1005 if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS) 1006 return 0; 1007 1008 /* RST segments are not recommended to carry timestamp, 1009 and, if they do, it is recommended to ignore PAWS because 1010 "their cleanup function should take precedence over timestamps." 1011 Certainly, it is mistake. It is necessary to understand the reasons 1012 of this constraint to relax it: if peer reboots, clock may go 1013 out-of-sync and half-open connections will not be reset. 1014 Actually, the problem would be not existing if all 1015 the implementations followed draft about maintaining clock 1016 via reboots. Linux-2.2 DOES NOT! 1017 1018 However, we can relax time bounds for RST segments to MSL. 1019 */ 1020 if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1021 return 0; 1022 return 1; 1023 } 1024 1025 #define TCP_CHECK_TIMER(sk) do { } while (0) 1026 1027 static inline int tcp_use_frto(const struct sock *sk) 1028 { 1029 const struct tcp_sock *tp = tcp_sk(sk); 1030 1031 /* F-RTO must be activated in sysctl and there must be some 1032 * unsent new data, and the advertised window should allow 1033 * sending it. 1034 */ 1035 return (sysctl_tcp_frto && sk->sk_send_head && 1036 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq, 1037 tp->snd_una + tp->snd_wnd)); 1038 } 1039 1040 static inline void tcp_mib_init(void) 1041 { 1042 /* See RFC 2012 */ 1043 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1); 1044 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1045 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1046 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1); 1047 } 1048 1049 /*from STCP */ 1050 static inline void clear_all_retrans_hints(struct tcp_sock *tp){ 1051 tp->lost_skb_hint = NULL; 1052 tp->scoreboard_skb_hint = NULL; 1053 tp->retransmit_skb_hint = NULL; 1054 tp->forward_skb_hint = NULL; 1055 tp->fastpath_skb_hint = NULL; 1056 } 1057 1058 /* /proc */ 1059 enum tcp_seq_states { 1060 TCP_SEQ_STATE_LISTENING, 1061 TCP_SEQ_STATE_OPENREQ, 1062 TCP_SEQ_STATE_ESTABLISHED, 1063 TCP_SEQ_STATE_TIME_WAIT, 1064 }; 1065 1066 struct tcp_seq_afinfo { 1067 struct module *owner; 1068 char *name; 1069 sa_family_t family; 1070 int (*seq_show) (struct seq_file *m, void *v); 1071 struct file_operations *seq_fops; 1072 }; 1073 1074 struct tcp_iter_state { 1075 sa_family_t family; 1076 enum tcp_seq_states state; 1077 struct sock *syn_wait_sk; 1078 int bucket, sbucket, num, uid; 1079 struct seq_operations seq_ops; 1080 }; 1081 1082 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo); 1083 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo); 1084 1085 extern struct request_sock_ops tcp_request_sock_ops; 1086 1087 extern int tcp_v4_destroy_sock(struct sock *sk); 1088 1089 #ifdef CONFIG_PROC_FS 1090 extern int tcp4_proc_init(void); 1091 extern void tcp4_proc_exit(void); 1092 #endif 1093 1094 extern void tcp_v4_init(struct net_proto_family *ops); 1095 extern void tcp_init(void); 1096 1097 #endif /* _TCP_H */ 1098