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 #include <linux/cryptohash.h> 33 34 #include <net/inet_connection_sock.h> 35 #include <net/inet_timewait_sock.h> 36 #include <net/inet_hashtables.h> 37 #include <net/checksum.h> 38 #include <net/request_sock.h> 39 #include <net/sock.h> 40 #include <net/snmp.h> 41 #include <net/ip.h> 42 #include <net/tcp_states.h> 43 #include <net/inet_ecn.h> 44 45 #include <linux/seq_file.h> 46 47 extern struct inet_hashinfo tcp_hashinfo; 48 49 extern atomic_t tcp_orphan_count; 50 extern void tcp_time_wait(struct sock *sk, int state, int timeo); 51 52 #define MAX_TCP_HEADER (128 + MAX_HEADER) 53 #define MAX_TCP_OPTION_SPACE 40 54 55 /* 56 * Never offer a window over 32767 without using window scaling. Some 57 * poor stacks do signed 16bit maths! 58 */ 59 #define MAX_TCP_WINDOW 32767U 60 61 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 62 #define TCP_MIN_MSS 88U 63 64 /* Minimal RCV_MSS. */ 65 #define TCP_MIN_RCVMSS 536U 66 67 /* The least MTU to use for probing */ 68 #define TCP_BASE_MSS 512 69 70 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 71 #define TCP_FASTRETRANS_THRESH 3 72 73 /* Maximal reordering. */ 74 #define TCP_MAX_REORDERING 127 75 76 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 77 #define TCP_MAX_QUICKACKS 16U 78 79 /* urg_data states */ 80 #define TCP_URG_VALID 0x0100 81 #define TCP_URG_NOTYET 0x0200 82 #define TCP_URG_READ 0x0400 83 84 #define TCP_RETR1 3 /* 85 * This is how many retries it does before it 86 * tries to figure out if the gateway is 87 * down. Minimal RFC value is 3; it corresponds 88 * to ~3sec-8min depending on RTO. 89 */ 90 91 #define TCP_RETR2 15 /* 92 * This should take at least 93 * 90 minutes to time out. 94 * RFC1122 says that the limit is 100 sec. 95 * 15 is ~13-30min depending on RTO. 96 */ 97 98 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a 99 * connection: ~180sec is RFC minimum */ 100 101 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a 102 * connection: ~180sec is RFC minimum */ 103 104 105 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned 106 * socket. 7 is ~50sec-16min. 107 */ 108 109 110 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 111 * state, about 60 seconds */ 112 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 113 /* BSD style FIN_WAIT2 deadlock breaker. 114 * It used to be 3min, new value is 60sec, 115 * to combine FIN-WAIT-2 timeout with 116 * TIME-WAIT timer. 117 */ 118 119 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 120 #if HZ >= 100 121 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 122 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 123 #else 124 #define TCP_DELACK_MIN 4U 125 #define TCP_ATO_MIN 4U 126 #endif 127 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 128 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 129 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */ 130 131 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 132 * for local resources. 133 */ 134 135 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 136 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 137 #define TCP_KEEPALIVE_INTVL (75*HZ) 138 139 #define MAX_TCP_KEEPIDLE 32767 140 #define MAX_TCP_KEEPINTVL 32767 141 #define MAX_TCP_KEEPCNT 127 142 #define MAX_TCP_SYNCNT 127 143 144 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 145 146 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 147 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 148 * after this time. It should be equal 149 * (or greater than) TCP_TIMEWAIT_LEN 150 * to provide reliability equal to one 151 * provided by timewait state. 152 */ 153 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 154 * timestamps. It must be less than 155 * minimal timewait lifetime. 156 */ 157 /* 158 * TCP option 159 */ 160 161 #define TCPOPT_NOP 1 /* Padding */ 162 #define TCPOPT_EOL 0 /* End of options */ 163 #define TCPOPT_MSS 2 /* Segment size negotiating */ 164 #define TCPOPT_WINDOW 3 /* Window scaling */ 165 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 166 #define TCPOPT_SACK 5 /* SACK Block */ 167 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 168 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 169 170 /* 171 * TCP option lengths 172 */ 173 174 #define TCPOLEN_MSS 4 175 #define TCPOLEN_WINDOW 3 176 #define TCPOLEN_SACK_PERM 2 177 #define TCPOLEN_TIMESTAMP 10 178 #define TCPOLEN_MD5SIG 18 179 180 /* But this is what stacks really send out. */ 181 #define TCPOLEN_TSTAMP_ALIGNED 12 182 #define TCPOLEN_WSCALE_ALIGNED 4 183 #define TCPOLEN_SACKPERM_ALIGNED 4 184 #define TCPOLEN_SACK_BASE 2 185 #define TCPOLEN_SACK_BASE_ALIGNED 4 186 #define TCPOLEN_SACK_PERBLOCK 8 187 #define TCPOLEN_MD5SIG_ALIGNED 20 188 #define TCPOLEN_MSS_ALIGNED 4 189 190 /* Flags in tp->nonagle */ 191 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 192 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 193 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 194 195 extern struct inet_timewait_death_row tcp_death_row; 196 197 /* sysctl variables for tcp */ 198 extern int sysctl_tcp_timestamps; 199 extern int sysctl_tcp_window_scaling; 200 extern int sysctl_tcp_sack; 201 extern int sysctl_tcp_fin_timeout; 202 extern int sysctl_tcp_keepalive_time; 203 extern int sysctl_tcp_keepalive_probes; 204 extern int sysctl_tcp_keepalive_intvl; 205 extern int sysctl_tcp_syn_retries; 206 extern int sysctl_tcp_synack_retries; 207 extern int sysctl_tcp_retries1; 208 extern int sysctl_tcp_retries2; 209 extern int sysctl_tcp_orphan_retries; 210 extern int sysctl_tcp_syncookies; 211 extern int sysctl_tcp_retrans_collapse; 212 extern int sysctl_tcp_stdurg; 213 extern int sysctl_tcp_rfc1337; 214 extern int sysctl_tcp_abort_on_overflow; 215 extern int sysctl_tcp_max_orphans; 216 extern int sysctl_tcp_fack; 217 extern int sysctl_tcp_reordering; 218 extern int sysctl_tcp_ecn; 219 extern int sysctl_tcp_dsack; 220 extern int sysctl_tcp_mem[3]; 221 extern int sysctl_tcp_wmem[3]; 222 extern int sysctl_tcp_rmem[3]; 223 extern int sysctl_tcp_app_win; 224 extern int sysctl_tcp_adv_win_scale; 225 extern int sysctl_tcp_tw_reuse; 226 extern int sysctl_tcp_frto; 227 extern int sysctl_tcp_frto_response; 228 extern int sysctl_tcp_low_latency; 229 extern int sysctl_tcp_dma_copybreak; 230 extern int sysctl_tcp_nometrics_save; 231 extern int sysctl_tcp_moderate_rcvbuf; 232 extern int sysctl_tcp_tso_win_divisor; 233 extern int sysctl_tcp_abc; 234 extern int sysctl_tcp_mtu_probing; 235 extern int sysctl_tcp_base_mss; 236 extern int sysctl_tcp_workaround_signed_windows; 237 extern int sysctl_tcp_slow_start_after_idle; 238 extern int sysctl_tcp_max_ssthresh; 239 240 extern atomic_t tcp_memory_allocated; 241 extern atomic_t tcp_sockets_allocated; 242 extern int tcp_memory_pressure; 243 244 /* 245 * The next routines deal with comparing 32 bit unsigned ints 246 * and worry about wraparound (automatic with unsigned arithmetic). 247 */ 248 249 static inline int before(__u32 seq1, __u32 seq2) 250 { 251 return (__s32)(seq1-seq2) < 0; 252 } 253 #define after(seq2, seq1) before(seq1, seq2) 254 255 /* is s2<=s1<=s3 ? */ 256 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3) 257 { 258 return seq3 - seq2 >= seq1 - seq2; 259 } 260 261 static inline int tcp_too_many_orphans(struct sock *sk, int num) 262 { 263 return (num > sysctl_tcp_max_orphans) || 264 (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 265 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]); 266 } 267 268 extern struct proto tcp_prot; 269 270 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 271 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 272 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 273 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 274 275 extern void tcp_v4_err(struct sk_buff *skb, u32); 276 277 extern void tcp_shutdown (struct sock *sk, int how); 278 279 extern int tcp_v4_rcv(struct sk_buff *skb); 280 281 extern int tcp_v4_remember_stamp(struct sock *sk); 282 283 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 284 285 extern int tcp_sendmsg(struct kiocb *iocb, struct socket *sock, 286 struct msghdr *msg, size_t size); 287 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags); 288 289 extern int tcp_ioctl(struct sock *sk, 290 int cmd, 291 unsigned long arg); 292 293 extern int tcp_rcv_state_process(struct sock *sk, 294 struct sk_buff *skb, 295 struct tcphdr *th, 296 unsigned len); 297 298 extern int tcp_rcv_established(struct sock *sk, 299 struct sk_buff *skb, 300 struct tcphdr *th, 301 unsigned len); 302 303 extern void tcp_rcv_space_adjust(struct sock *sk); 304 305 extern void tcp_cleanup_rbuf(struct sock *sk, int copied); 306 307 extern int tcp_twsk_unique(struct sock *sk, 308 struct sock *sktw, void *twp); 309 310 extern void tcp_twsk_destructor(struct sock *sk); 311 312 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 313 struct pipe_inode_info *pipe, size_t len, unsigned int flags); 314 315 static inline void tcp_dec_quickack_mode(struct sock *sk, 316 const unsigned int pkts) 317 { 318 struct inet_connection_sock *icsk = inet_csk(sk); 319 320 if (icsk->icsk_ack.quick) { 321 if (pkts >= icsk->icsk_ack.quick) { 322 icsk->icsk_ack.quick = 0; 323 /* Leaving quickack mode we deflate ATO. */ 324 icsk->icsk_ack.ato = TCP_ATO_MIN; 325 } else 326 icsk->icsk_ack.quick -= pkts; 327 } 328 } 329 330 extern void tcp_enter_quickack_mode(struct sock *sk); 331 332 static inline void tcp_clear_options(struct tcp_options_received *rx_opt) 333 { 334 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0; 335 } 336 337 #define TCP_ECN_OK 1 338 #define TCP_ECN_QUEUE_CWR 2 339 #define TCP_ECN_DEMAND_CWR 4 340 341 static __inline__ void 342 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th) 343 { 344 if (sysctl_tcp_ecn && th->ece && th->cwr) 345 inet_rsk(req)->ecn_ok = 1; 346 } 347 348 enum tcp_tw_status 349 { 350 TCP_TW_SUCCESS = 0, 351 TCP_TW_RST = 1, 352 TCP_TW_ACK = 2, 353 TCP_TW_SYN = 3 354 }; 355 356 357 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 358 struct sk_buff *skb, 359 const struct tcphdr *th); 360 361 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb, 362 struct request_sock *req, 363 struct request_sock **prev); 364 extern int tcp_child_process(struct sock *parent, 365 struct sock *child, 366 struct sk_buff *skb); 367 extern int tcp_use_frto(struct sock *sk); 368 extern void tcp_enter_frto(struct sock *sk); 369 extern void tcp_enter_loss(struct sock *sk, int how); 370 extern void tcp_clear_retrans(struct tcp_sock *tp); 371 extern void tcp_update_metrics(struct sock *sk); 372 373 extern void tcp_close(struct sock *sk, 374 long timeout); 375 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait); 376 377 extern int tcp_getsockopt(struct sock *sk, int level, 378 int optname, 379 char __user *optval, 380 int __user *optlen); 381 extern int tcp_setsockopt(struct sock *sk, int level, 382 int optname, char __user *optval, 383 int optlen); 384 extern int compat_tcp_getsockopt(struct sock *sk, 385 int level, int optname, 386 char __user *optval, int __user *optlen); 387 extern int compat_tcp_setsockopt(struct sock *sk, 388 int level, int optname, 389 char __user *optval, int optlen); 390 extern void tcp_set_keepalive(struct sock *sk, int val); 391 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, 392 struct msghdr *msg, 393 size_t len, int nonblock, 394 int flags, int *addr_len); 395 396 extern void tcp_parse_options(struct sk_buff *skb, 397 struct tcp_options_received *opt_rx, 398 int estab); 399 400 extern u8 *tcp_parse_md5sig_option(struct tcphdr *th); 401 402 /* 403 * TCP v4 functions exported for the inet6 API 404 */ 405 406 extern void tcp_v4_send_check(struct sock *sk, int len, 407 struct sk_buff *skb); 408 409 extern int tcp_v4_conn_request(struct sock *sk, 410 struct sk_buff *skb); 411 412 extern struct sock * tcp_create_openreq_child(struct sock *sk, 413 struct request_sock *req, 414 struct sk_buff *skb); 415 416 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, 417 struct sk_buff *skb, 418 struct request_sock *req, 419 struct dst_entry *dst); 420 421 extern int tcp_v4_do_rcv(struct sock *sk, 422 struct sk_buff *skb); 423 424 extern int tcp_v4_connect(struct sock *sk, 425 struct sockaddr *uaddr, 426 int addr_len); 427 428 extern int tcp_connect(struct sock *sk); 429 430 extern struct sk_buff * tcp_make_synack(struct sock *sk, 431 struct dst_entry *dst, 432 struct request_sock *req); 433 434 extern int tcp_disconnect(struct sock *sk, int flags); 435 436 437 /* From syncookies.c */ 438 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS]; 439 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 440 struct ip_options *opt); 441 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, 442 __u16 *mss); 443 444 extern __u32 cookie_init_timestamp(struct request_sock *req); 445 extern void cookie_check_timestamp(struct tcp_options_received *tcp_opt); 446 447 /* From net/ipv6/syncookies.c */ 448 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 449 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb, 450 __u16 *mss); 451 452 /* tcp_output.c */ 453 454 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 455 int nonagle); 456 extern int tcp_may_send_now(struct sock *sk); 457 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); 458 extern void tcp_xmit_retransmit_queue(struct sock *); 459 extern void tcp_simple_retransmit(struct sock *); 460 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32); 461 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 462 463 extern void tcp_send_probe0(struct sock *); 464 extern void tcp_send_partial(struct sock *); 465 extern int tcp_write_wakeup(struct sock *); 466 extern void tcp_send_fin(struct sock *sk); 467 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority); 468 extern int tcp_send_synack(struct sock *); 469 extern void tcp_push_one(struct sock *, unsigned int mss_now); 470 extern void tcp_send_ack(struct sock *sk); 471 extern void tcp_send_delayed_ack(struct sock *sk); 472 473 /* tcp_input.c */ 474 extern void tcp_cwnd_application_limited(struct sock *sk); 475 476 /* tcp_timer.c */ 477 extern void tcp_init_xmit_timers(struct sock *); 478 static inline void tcp_clear_xmit_timers(struct sock *sk) 479 { 480 inet_csk_clear_xmit_timers(sk); 481 } 482 483 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 484 extern unsigned int tcp_current_mss(struct sock *sk, int large); 485 486 /* tcp.c */ 487 extern void tcp_get_info(struct sock *, struct tcp_info *); 488 489 /* Read 'sendfile()'-style from a TCP socket */ 490 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 491 unsigned int, size_t); 492 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 493 sk_read_actor_t recv_actor); 494 495 extern void tcp_initialize_rcv_mss(struct sock *sk); 496 497 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu); 498 extern int tcp_mss_to_mtu(struct sock *sk, int mss); 499 extern void tcp_mtup_init(struct sock *sk); 500 501 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 502 { 503 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 504 ntohl(TCP_FLAG_ACK) | 505 snd_wnd); 506 } 507 508 static inline void tcp_fast_path_on(struct tcp_sock *tp) 509 { 510 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 511 } 512 513 static inline void tcp_fast_path_check(struct sock *sk) 514 { 515 struct tcp_sock *tp = tcp_sk(sk); 516 517 if (skb_queue_empty(&tp->out_of_order_queue) && 518 tp->rcv_wnd && 519 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 520 !tp->urg_data) 521 tcp_fast_path_on(tp); 522 } 523 524 /* Compute the actual receive window we are currently advertising. 525 * Rcv_nxt can be after the window if our peer push more data 526 * than the offered window. 527 */ 528 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 529 { 530 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 531 532 if (win < 0) 533 win = 0; 534 return (u32) win; 535 } 536 537 /* Choose a new window, without checks for shrinking, and without 538 * scaling applied to the result. The caller does these things 539 * if necessary. This is a "raw" window selection. 540 */ 541 extern u32 __tcp_select_window(struct sock *sk); 542 543 /* TCP timestamps are only 32-bits, this causes a slight 544 * complication on 64-bit systems since we store a snapshot 545 * of jiffies in the buffer control blocks below. We decided 546 * to use only the low 32-bits of jiffies and hide the ugly 547 * casts with the following macro. 548 */ 549 #define tcp_time_stamp ((__u32)(jiffies)) 550 551 /* This is what the send packet queuing engine uses to pass 552 * TCP per-packet control information to the transmission 553 * code. We also store the host-order sequence numbers in 554 * here too. This is 36 bytes on 32-bit architectures, 555 * 40 bytes on 64-bit machines, if this grows please adjust 556 * skbuff.h:skbuff->cb[xxx] size appropriately. 557 */ 558 struct tcp_skb_cb { 559 union { 560 struct inet_skb_parm h4; 561 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE) 562 struct inet6_skb_parm h6; 563 #endif 564 } header; /* For incoming frames */ 565 __u32 seq; /* Starting sequence number */ 566 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 567 __u32 when; /* used to compute rtt's */ 568 __u8 flags; /* TCP header flags. */ 569 570 /* NOTE: These must match up to the flags byte in a 571 * real TCP header. 572 */ 573 #define TCPCB_FLAG_FIN 0x01 574 #define TCPCB_FLAG_SYN 0x02 575 #define TCPCB_FLAG_RST 0x04 576 #define TCPCB_FLAG_PSH 0x08 577 #define TCPCB_FLAG_ACK 0x10 578 #define TCPCB_FLAG_URG 0x20 579 #define TCPCB_FLAG_ECE 0x40 580 #define TCPCB_FLAG_CWR 0x80 581 582 __u8 sacked; /* State flags for SACK/FACK. */ 583 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 584 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 585 #define TCPCB_LOST 0x04 /* SKB is lost */ 586 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 587 588 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 589 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 590 591 __u16 urg_ptr; /* Valid w/URG flags is set. */ 592 __u32 ack_seq; /* Sequence number ACK'd */ 593 }; 594 595 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 596 597 /* Due to TSO, an SKB can be composed of multiple actual 598 * packets. To keep these tracked properly, we use this. 599 */ 600 static inline int tcp_skb_pcount(const struct sk_buff *skb) 601 { 602 return skb_shinfo(skb)->gso_segs; 603 } 604 605 /* This is valid iff tcp_skb_pcount() > 1. */ 606 static inline int tcp_skb_mss(const struct sk_buff *skb) 607 { 608 return skb_shinfo(skb)->gso_size; 609 } 610 611 static inline void tcp_dec_pcount_approx_int(__u32 *count, const int decr) 612 { 613 if (*count) { 614 *count -= decr; 615 if ((int)*count < 0) 616 *count = 0; 617 } 618 } 619 620 static inline void tcp_dec_pcount_approx(__u32 *count, 621 const struct sk_buff *skb) 622 { 623 tcp_dec_pcount_approx_int(count, tcp_skb_pcount(skb)); 624 } 625 626 /* Events passed to congestion control interface */ 627 enum tcp_ca_event { 628 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 629 CA_EVENT_CWND_RESTART, /* congestion window restart */ 630 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 631 CA_EVENT_FRTO, /* fast recovery timeout */ 632 CA_EVENT_LOSS, /* loss timeout */ 633 CA_EVENT_FAST_ACK, /* in sequence ack */ 634 CA_EVENT_SLOW_ACK, /* other ack */ 635 }; 636 637 /* 638 * Interface for adding new TCP congestion control handlers 639 */ 640 #define TCP_CA_NAME_MAX 16 641 #define TCP_CA_MAX 128 642 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 643 644 #define TCP_CONG_NON_RESTRICTED 0x1 645 #define TCP_CONG_RTT_STAMP 0x2 646 647 struct tcp_congestion_ops { 648 struct list_head list; 649 unsigned long flags; 650 651 /* initialize private data (optional) */ 652 void (*init)(struct sock *sk); 653 /* cleanup private data (optional) */ 654 void (*release)(struct sock *sk); 655 656 /* return slow start threshold (required) */ 657 u32 (*ssthresh)(struct sock *sk); 658 /* lower bound for congestion window (optional) */ 659 u32 (*min_cwnd)(const struct sock *sk); 660 /* do new cwnd calculation (required) */ 661 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight); 662 /* call before changing ca_state (optional) */ 663 void (*set_state)(struct sock *sk, u8 new_state); 664 /* call when cwnd event occurs (optional) */ 665 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 666 /* new value of cwnd after loss (optional) */ 667 u32 (*undo_cwnd)(struct sock *sk); 668 /* hook for packet ack accounting (optional) */ 669 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 670 /* get info for inet_diag (optional) */ 671 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 672 673 char name[TCP_CA_NAME_MAX]; 674 struct module *owner; 675 }; 676 677 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type); 678 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 679 680 extern void tcp_init_congestion_control(struct sock *sk); 681 extern void tcp_cleanup_congestion_control(struct sock *sk); 682 extern int tcp_set_default_congestion_control(const char *name); 683 extern void tcp_get_default_congestion_control(char *name); 684 extern void tcp_get_available_congestion_control(char *buf, size_t len); 685 extern void tcp_get_allowed_congestion_control(char *buf, size_t len); 686 extern int tcp_set_allowed_congestion_control(char *allowed); 687 extern int tcp_set_congestion_control(struct sock *sk, const char *name); 688 extern void tcp_slow_start(struct tcp_sock *tp); 689 690 extern struct tcp_congestion_ops tcp_init_congestion_ops; 691 extern u32 tcp_reno_ssthresh(struct sock *sk); 692 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight); 693 extern u32 tcp_reno_min_cwnd(const struct sock *sk); 694 extern struct tcp_congestion_ops tcp_reno; 695 696 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 697 { 698 struct inet_connection_sock *icsk = inet_csk(sk); 699 700 if (icsk->icsk_ca_ops->set_state) 701 icsk->icsk_ca_ops->set_state(sk, ca_state); 702 icsk->icsk_ca_state = ca_state; 703 } 704 705 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 706 { 707 const struct inet_connection_sock *icsk = inet_csk(sk); 708 709 if (icsk->icsk_ca_ops->cwnd_event) 710 icsk->icsk_ca_ops->cwnd_event(sk, event); 711 } 712 713 /* These functions determine how the current flow behaves in respect of SACK 714 * handling. SACK is negotiated with the peer, and therefore it can vary 715 * between different flows. 716 * 717 * tcp_is_sack - SACK enabled 718 * tcp_is_reno - No SACK 719 * tcp_is_fack - FACK enabled, implies SACK enabled 720 */ 721 static inline int tcp_is_sack(const struct tcp_sock *tp) 722 { 723 return tp->rx_opt.sack_ok; 724 } 725 726 static inline int tcp_is_reno(const struct tcp_sock *tp) 727 { 728 return !tcp_is_sack(tp); 729 } 730 731 static inline int tcp_is_fack(const struct tcp_sock *tp) 732 { 733 return tp->rx_opt.sack_ok & 2; 734 } 735 736 static inline void tcp_enable_fack(struct tcp_sock *tp) 737 { 738 tp->rx_opt.sack_ok |= 2; 739 } 740 741 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 742 { 743 return tp->sacked_out + tp->lost_out; 744 } 745 746 /* This determines how many packets are "in the network" to the best 747 * of our knowledge. In many cases it is conservative, but where 748 * detailed information is available from the receiver (via SACK 749 * blocks etc.) we can make more aggressive calculations. 750 * 751 * Use this for decisions involving congestion control, use just 752 * tp->packets_out to determine if the send queue is empty or not. 753 * 754 * Read this equation as: 755 * 756 * "Packets sent once on transmission queue" MINUS 757 * "Packets left network, but not honestly ACKed yet" PLUS 758 * "Packets fast retransmitted" 759 */ 760 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 761 { 762 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 763 } 764 765 extern int tcp_limit_reno_sacked(struct tcp_sock *tp); 766 767 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 768 * The exception is rate halving phase, when cwnd is decreasing towards 769 * ssthresh. 770 */ 771 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 772 { 773 const struct tcp_sock *tp = tcp_sk(sk); 774 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery)) 775 return tp->snd_ssthresh; 776 else 777 return max(tp->snd_ssthresh, 778 ((tp->snd_cwnd >> 1) + 779 (tp->snd_cwnd >> 2))); 780 } 781 782 /* Use define here intentionally to get WARN_ON location shown at the caller */ 783 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 784 785 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh); 786 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst); 787 788 /* Slow start with delack produces 3 packets of burst, so that 789 * it is safe "de facto". This will be the default - same as 790 * the default reordering threshold - but if reordering increases, 791 * we must be able to allow cwnd to burst at least this much in order 792 * to not pull it back when holes are filled. 793 */ 794 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 795 { 796 return tp->reordering; 797 } 798 799 /* Returns end sequence number of the receiver's advertised window */ 800 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 801 { 802 return tp->snd_una + tp->snd_wnd; 803 } 804 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight); 805 806 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss, 807 const struct sk_buff *skb) 808 { 809 if (skb->len < mss) 810 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 811 } 812 813 static inline void tcp_check_probe_timer(struct sock *sk) 814 { 815 struct tcp_sock *tp = tcp_sk(sk); 816 const struct inet_connection_sock *icsk = inet_csk(sk); 817 818 if (!tp->packets_out && !icsk->icsk_pending) 819 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 820 icsk->icsk_rto, TCP_RTO_MAX); 821 } 822 823 static inline void tcp_push_pending_frames(struct sock *sk) 824 { 825 struct tcp_sock *tp = tcp_sk(sk); 826 827 __tcp_push_pending_frames(sk, tcp_current_mss(sk, 1), tp->nonagle); 828 } 829 830 static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq) 831 { 832 tp->snd_wl1 = seq; 833 } 834 835 static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq) 836 { 837 tp->snd_wl1 = seq; 838 } 839 840 /* 841 * Calculate(/check) TCP checksum 842 */ 843 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 844 __be32 daddr, __wsum base) 845 { 846 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 847 } 848 849 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 850 { 851 return __skb_checksum_complete(skb); 852 } 853 854 static inline int tcp_checksum_complete(struct sk_buff *skb) 855 { 856 return !skb_csum_unnecessary(skb) && 857 __tcp_checksum_complete(skb); 858 } 859 860 /* Prequeue for VJ style copy to user, combined with checksumming. */ 861 862 static inline void tcp_prequeue_init(struct tcp_sock *tp) 863 { 864 tp->ucopy.task = NULL; 865 tp->ucopy.len = 0; 866 tp->ucopy.memory = 0; 867 skb_queue_head_init(&tp->ucopy.prequeue); 868 #ifdef CONFIG_NET_DMA 869 tp->ucopy.dma_chan = NULL; 870 tp->ucopy.wakeup = 0; 871 tp->ucopy.pinned_list = NULL; 872 tp->ucopy.dma_cookie = 0; 873 #endif 874 } 875 876 /* Packet is added to VJ-style prequeue for processing in process 877 * context, if a reader task is waiting. Apparently, this exciting 878 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) 879 * failed somewhere. Latency? Burstiness? Well, at least now we will 880 * see, why it failed. 8)8) --ANK 881 * 882 * NOTE: is this not too big to inline? 883 */ 884 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb) 885 { 886 struct tcp_sock *tp = tcp_sk(sk); 887 888 if (!sysctl_tcp_low_latency && tp->ucopy.task) { 889 __skb_queue_tail(&tp->ucopy.prequeue, skb); 890 tp->ucopy.memory += skb->truesize; 891 if (tp->ucopy.memory > sk->sk_rcvbuf) { 892 struct sk_buff *skb1; 893 894 BUG_ON(sock_owned_by_user(sk)); 895 896 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { 897 sk->sk_backlog_rcv(sk, skb1); 898 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPPREQUEUEDROPPED); 899 } 900 901 tp->ucopy.memory = 0; 902 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { 903 wake_up_interruptible(sk->sk_sleep); 904 if (!inet_csk_ack_scheduled(sk)) 905 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 906 (3 * TCP_RTO_MIN) / 4, 907 TCP_RTO_MAX); 908 } 909 return 1; 910 } 911 return 0; 912 } 913 914 915 #undef STATE_TRACE 916 917 #ifdef STATE_TRACE 918 static const char *statename[]={ 919 "Unused","Established","Syn Sent","Syn Recv", 920 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 921 "Close Wait","Last ACK","Listen","Closing" 922 }; 923 #endif 924 extern void tcp_set_state(struct sock *sk, int state); 925 926 extern void tcp_done(struct sock *sk); 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 req->cookie_ts = 0; 967 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 968 req->mss = rx_opt->mss_clamp; 969 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 970 ireq->tstamp_ok = rx_opt->tstamp_ok; 971 ireq->sack_ok = rx_opt->sack_ok; 972 ireq->snd_wscale = rx_opt->snd_wscale; 973 ireq->wscale_ok = rx_opt->wscale_ok; 974 ireq->acked = 0; 975 ireq->ecn_ok = 0; 976 ireq->rmt_port = tcp_hdr(skb)->source; 977 } 978 979 extern void tcp_enter_memory_pressure(struct sock *sk); 980 981 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 982 { 983 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 984 } 985 986 static inline int keepalive_time_when(const struct tcp_sock *tp) 987 { 988 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 989 } 990 991 static inline int tcp_fin_time(const struct sock *sk) 992 { 993 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 994 const int rto = inet_csk(sk)->icsk_rto; 995 996 if (fin_timeout < (rto << 2) - (rto >> 1)) 997 fin_timeout = (rto << 2) - (rto >> 1); 998 999 return fin_timeout; 1000 } 1001 1002 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst) 1003 { 1004 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0) 1005 return 0; 1006 if (get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS) 1007 return 0; 1008 1009 /* RST segments are not recommended to carry timestamp, 1010 and, if they do, it is recommended to ignore PAWS because 1011 "their cleanup function should take precedence over timestamps." 1012 Certainly, it is mistake. It is necessary to understand the reasons 1013 of this constraint to relax it: if peer reboots, clock may go 1014 out-of-sync and half-open connections will not be reset. 1015 Actually, the problem would be not existing if all 1016 the implementations followed draft about maintaining clock 1017 via reboots. Linux-2.2 DOES NOT! 1018 1019 However, we can relax time bounds for RST segments to MSL. 1020 */ 1021 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1022 return 0; 1023 return 1; 1024 } 1025 1026 #define TCP_CHECK_TIMER(sk) do { } while (0) 1027 1028 static inline void tcp_mib_init(struct net *net) 1029 { 1030 /* See RFC 2012 */ 1031 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1032 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1033 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1034 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1035 } 1036 1037 /* from STCP */ 1038 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1039 { 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 } 1045 1046 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1047 { 1048 tcp_clear_retrans_hints_partial(tp); 1049 } 1050 1051 /* MD5 Signature */ 1052 struct crypto_hash; 1053 1054 /* - key database */ 1055 struct tcp_md5sig_key { 1056 u8 *key; 1057 u8 keylen; 1058 }; 1059 1060 struct tcp4_md5sig_key { 1061 struct tcp_md5sig_key base; 1062 __be32 addr; 1063 }; 1064 1065 struct tcp6_md5sig_key { 1066 struct tcp_md5sig_key base; 1067 #if 0 1068 u32 scope_id; /* XXX */ 1069 #endif 1070 struct in6_addr addr; 1071 }; 1072 1073 /* - sock block */ 1074 struct tcp_md5sig_info { 1075 struct tcp4_md5sig_key *keys4; 1076 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1077 struct tcp6_md5sig_key *keys6; 1078 u32 entries6; 1079 u32 alloced6; 1080 #endif 1081 u32 entries4; 1082 u32 alloced4; 1083 }; 1084 1085 /* - pseudo header */ 1086 struct tcp4_pseudohdr { 1087 __be32 saddr; 1088 __be32 daddr; 1089 __u8 pad; 1090 __u8 protocol; 1091 __be16 len; 1092 }; 1093 1094 struct tcp6_pseudohdr { 1095 struct in6_addr saddr; 1096 struct in6_addr daddr; 1097 __be32 len; 1098 __be32 protocol; /* including padding */ 1099 }; 1100 1101 union tcp_md5sum_block { 1102 struct tcp4_pseudohdr ip4; 1103 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1104 struct tcp6_pseudohdr ip6; 1105 #endif 1106 }; 1107 1108 /* - pool: digest algorithm, hash description and scratch buffer */ 1109 struct tcp_md5sig_pool { 1110 struct hash_desc md5_desc; 1111 union tcp_md5sum_block md5_blk; 1112 }; 1113 1114 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */ 1115 1116 /* - functions */ 1117 extern int tcp_v4_md5_hash_skb(char *md5_hash, 1118 struct tcp_md5sig_key *key, 1119 struct sock *sk, 1120 struct request_sock *req, 1121 struct sk_buff *skb); 1122 1123 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1124 struct sock *addr_sk); 1125 1126 extern int tcp_v4_md5_do_add(struct sock *sk, 1127 __be32 addr, 1128 u8 *newkey, 1129 u8 newkeylen); 1130 1131 extern int tcp_v4_md5_do_del(struct sock *sk, 1132 __be32 addr); 1133 1134 #ifdef CONFIG_TCP_MD5SIG 1135 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \ 1136 &(struct tcp_md5sig_key) { \ 1137 .key = (twsk)->tw_md5_key, \ 1138 .keylen = (twsk)->tw_md5_keylen, \ 1139 } : NULL) 1140 #else 1141 #define tcp_twsk_md5_key(twsk) NULL 1142 #endif 1143 1144 extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void); 1145 extern void tcp_free_md5sig_pool(void); 1146 1147 extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu); 1148 extern void __tcp_put_md5sig_pool(void); 1149 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *); 1150 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *, 1151 unsigned header_len); 1152 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1153 struct tcp_md5sig_key *key); 1154 1155 static inline 1156 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void) 1157 { 1158 int cpu = get_cpu(); 1159 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu); 1160 if (!ret) 1161 put_cpu(); 1162 return ret; 1163 } 1164 1165 static inline void tcp_put_md5sig_pool(void) 1166 { 1167 __tcp_put_md5sig_pool(); 1168 put_cpu(); 1169 } 1170 1171 /* write queue abstraction */ 1172 static inline void tcp_write_queue_purge(struct sock *sk) 1173 { 1174 struct sk_buff *skb; 1175 1176 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1177 sk_wmem_free_skb(sk, skb); 1178 sk_mem_reclaim(sk); 1179 } 1180 1181 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk) 1182 { 1183 struct sk_buff *skb = sk->sk_write_queue.next; 1184 if (skb == (struct sk_buff *) &sk->sk_write_queue) 1185 return NULL; 1186 return skb; 1187 } 1188 1189 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk) 1190 { 1191 struct sk_buff *skb = sk->sk_write_queue.prev; 1192 if (skb == (struct sk_buff *) &sk->sk_write_queue) 1193 return NULL; 1194 return skb; 1195 } 1196 1197 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb) 1198 { 1199 return skb->next; 1200 } 1201 1202 #define tcp_for_write_queue(skb, sk) \ 1203 for (skb = (sk)->sk_write_queue.next; \ 1204 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \ 1205 skb = skb->next) 1206 1207 #define tcp_for_write_queue_from(skb, sk) \ 1208 for (; (skb != (struct sk_buff *)&(sk)->sk_write_queue);\ 1209 skb = skb->next) 1210 1211 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1212 for (tmp = skb->next; \ 1213 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \ 1214 skb = tmp, tmp = skb->next) 1215 1216 static inline struct sk_buff *tcp_send_head(struct sock *sk) 1217 { 1218 return sk->sk_send_head; 1219 } 1220 1221 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb) 1222 { 1223 sk->sk_send_head = skb->next; 1224 if (sk->sk_send_head == (struct sk_buff *)&sk->sk_write_queue) 1225 sk->sk_send_head = NULL; 1226 } 1227 1228 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1229 { 1230 if (sk->sk_send_head == skb_unlinked) 1231 sk->sk_send_head = NULL; 1232 } 1233 1234 static inline void tcp_init_send_head(struct sock *sk) 1235 { 1236 sk->sk_send_head = NULL; 1237 } 1238 1239 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1240 { 1241 __skb_queue_tail(&sk->sk_write_queue, skb); 1242 } 1243 1244 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1245 { 1246 __tcp_add_write_queue_tail(sk, skb); 1247 1248 /* Queue it, remembering where we must start sending. */ 1249 if (sk->sk_send_head == NULL) { 1250 sk->sk_send_head = skb; 1251 1252 if (tcp_sk(sk)->highest_sack == NULL) 1253 tcp_sk(sk)->highest_sack = skb; 1254 } 1255 } 1256 1257 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1258 { 1259 __skb_queue_head(&sk->sk_write_queue, skb); 1260 } 1261 1262 /* Insert buff after skb on the write queue of sk. */ 1263 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1264 struct sk_buff *buff, 1265 struct sock *sk) 1266 { 1267 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1268 } 1269 1270 /* Insert skb between prev and next on the write queue of sk. */ 1271 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1272 struct sk_buff *skb, 1273 struct sock *sk) 1274 { 1275 __skb_insert(new, skb->prev, skb, &sk->sk_write_queue); 1276 1277 if (sk->sk_send_head == skb) 1278 sk->sk_send_head = new; 1279 } 1280 1281 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1282 { 1283 __skb_unlink(skb, &sk->sk_write_queue); 1284 } 1285 1286 static inline int tcp_skb_is_last(const struct sock *sk, 1287 const struct sk_buff *skb) 1288 { 1289 return skb->next == (struct sk_buff *)&sk->sk_write_queue; 1290 } 1291 1292 static inline int tcp_write_queue_empty(struct sock *sk) 1293 { 1294 return skb_queue_empty(&sk->sk_write_queue); 1295 } 1296 1297 /* Start sequence of the highest skb with SACKed bit, valid only if 1298 * sacked > 0 or when the caller has ensured validity by itself. 1299 */ 1300 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1301 { 1302 if (!tp->sacked_out) 1303 return tp->snd_una; 1304 1305 if (tp->highest_sack == NULL) 1306 return tp->snd_nxt; 1307 1308 return TCP_SKB_CB(tp->highest_sack)->seq; 1309 } 1310 1311 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1312 { 1313 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1314 tcp_write_queue_next(sk, skb); 1315 } 1316 1317 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1318 { 1319 return tcp_sk(sk)->highest_sack; 1320 } 1321 1322 static inline void tcp_highest_sack_reset(struct sock *sk) 1323 { 1324 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1325 } 1326 1327 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1328 static inline void tcp_highest_sack_combine(struct sock *sk, 1329 struct sk_buff *old, 1330 struct sk_buff *new) 1331 { 1332 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1333 tcp_sk(sk)->highest_sack = new; 1334 } 1335 1336 /* /proc */ 1337 enum tcp_seq_states { 1338 TCP_SEQ_STATE_LISTENING, 1339 TCP_SEQ_STATE_OPENREQ, 1340 TCP_SEQ_STATE_ESTABLISHED, 1341 TCP_SEQ_STATE_TIME_WAIT, 1342 }; 1343 1344 struct tcp_seq_afinfo { 1345 char *name; 1346 sa_family_t family; 1347 struct file_operations seq_fops; 1348 struct seq_operations seq_ops; 1349 }; 1350 1351 struct tcp_iter_state { 1352 struct seq_net_private p; 1353 sa_family_t family; 1354 enum tcp_seq_states state; 1355 struct sock *syn_wait_sk; 1356 int bucket, sbucket, num, uid; 1357 }; 1358 1359 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1360 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1361 1362 extern struct request_sock_ops tcp_request_sock_ops; 1363 extern struct request_sock_ops tcp6_request_sock_ops; 1364 1365 extern void tcp_v4_destroy_sock(struct sock *sk); 1366 1367 extern int tcp_v4_gso_send_check(struct sk_buff *skb); 1368 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features); 1369 1370 #ifdef CONFIG_PROC_FS 1371 extern int tcp4_proc_init(void); 1372 extern void tcp4_proc_exit(void); 1373 #endif 1374 1375 /* TCP af-specific functions */ 1376 struct tcp_sock_af_ops { 1377 #ifdef CONFIG_TCP_MD5SIG 1378 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1379 struct sock *addr_sk); 1380 int (*calc_md5_hash) (char *location, 1381 struct tcp_md5sig_key *md5, 1382 struct sock *sk, 1383 struct request_sock *req, 1384 struct sk_buff *skb); 1385 int (*md5_add) (struct sock *sk, 1386 struct sock *addr_sk, 1387 u8 *newkey, 1388 u8 len); 1389 int (*md5_parse) (struct sock *sk, 1390 char __user *optval, 1391 int optlen); 1392 #endif 1393 }; 1394 1395 struct tcp_request_sock_ops { 1396 #ifdef CONFIG_TCP_MD5SIG 1397 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1398 struct request_sock *req); 1399 #endif 1400 }; 1401 1402 extern void tcp_v4_init(void); 1403 extern void tcp_init(void); 1404 1405 #endif /* _TCP_H */ 1406