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 54 /* 55 * Never offer a window over 32767 without using window scaling. Some 56 * poor stacks do signed 16bit maths! 57 */ 58 #define MAX_TCP_WINDOW 32767U 59 60 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 61 #define TCP_MIN_MSS 88U 62 63 /* Minimal RCV_MSS. */ 64 #define TCP_MIN_RCVMSS 536U 65 66 /* The least MTU to use for probing */ 67 #define TCP_BASE_MSS 512 68 69 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 70 #define TCP_FASTRETRANS_THRESH 3 71 72 /* Maximal reordering. */ 73 #define TCP_MAX_REORDERING 127 74 75 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 76 #define TCP_MAX_QUICKACKS 16U 77 78 /* urg_data states */ 79 #define TCP_URG_VALID 0x0100 80 #define TCP_URG_NOTYET 0x0200 81 #define TCP_URG_READ 0x0400 82 83 #define TCP_RETR1 3 /* 84 * This is how many retries it does before it 85 * tries to figure out if the gateway is 86 * down. Minimal RFC value is 3; it corresponds 87 * to ~3sec-8min depending on RTO. 88 */ 89 90 #define TCP_RETR2 15 /* 91 * This should take at least 92 * 90 minutes to time out. 93 * RFC1122 says that the limit is 100 sec. 94 * 15 is ~13-30min depending on RTO. 95 */ 96 97 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a 98 * connection: ~180sec is RFC minimum */ 99 100 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a 101 * connection: ~180sec is RFC minimum */ 102 103 104 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned 105 * socket. 7 is ~50sec-16min. 106 */ 107 108 109 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 110 * state, about 60 seconds */ 111 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 112 /* BSD style FIN_WAIT2 deadlock breaker. 113 * It used to be 3min, new value is 60sec, 114 * to combine FIN-WAIT-2 timeout with 115 * TIME-WAIT timer. 116 */ 117 118 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 119 #if HZ >= 100 120 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 121 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 122 #else 123 #define TCP_DELACK_MIN 4U 124 #define TCP_ATO_MIN 4U 125 #endif 126 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 127 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 128 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */ 129 130 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 131 * for local resources. 132 */ 133 134 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 135 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 136 #define TCP_KEEPALIVE_INTVL (75*HZ) 137 138 #define MAX_TCP_KEEPIDLE 32767 139 #define MAX_TCP_KEEPINTVL 32767 140 #define MAX_TCP_KEEPCNT 127 141 #define MAX_TCP_SYNCNT 127 142 #define MAX_TCP_ACCEPT_DEFERRED 65535 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 20 188 189 /* Flags in tp->nonagle */ 190 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 191 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 192 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 193 194 extern struct inet_timewait_death_row tcp_death_row; 195 196 /* sysctl variables for tcp */ 197 extern int sysctl_tcp_timestamps; 198 extern int sysctl_tcp_window_scaling; 199 extern int sysctl_tcp_sack; 200 extern int sysctl_tcp_fin_timeout; 201 extern int sysctl_tcp_keepalive_time; 202 extern int sysctl_tcp_keepalive_probes; 203 extern int sysctl_tcp_keepalive_intvl; 204 extern int sysctl_tcp_syn_retries; 205 extern int sysctl_tcp_synack_retries; 206 extern int sysctl_tcp_retries1; 207 extern int sysctl_tcp_retries2; 208 extern int sysctl_tcp_orphan_retries; 209 extern int sysctl_tcp_syncookies; 210 extern int sysctl_tcp_retrans_collapse; 211 extern int sysctl_tcp_stdurg; 212 extern int sysctl_tcp_rfc1337; 213 extern int sysctl_tcp_abort_on_overflow; 214 extern int sysctl_tcp_max_orphans; 215 extern int sysctl_tcp_fack; 216 extern int sysctl_tcp_reordering; 217 extern int sysctl_tcp_ecn; 218 extern int sysctl_tcp_dsack; 219 extern int sysctl_tcp_mem[3]; 220 extern int sysctl_tcp_wmem[3]; 221 extern int sysctl_tcp_rmem[3]; 222 extern int sysctl_tcp_app_win; 223 extern int sysctl_tcp_adv_win_scale; 224 extern int sysctl_tcp_tw_reuse; 225 extern int sysctl_tcp_frto; 226 extern int sysctl_tcp_frto_response; 227 extern int sysctl_tcp_low_latency; 228 extern int sysctl_tcp_dma_copybreak; 229 extern int sysctl_tcp_nometrics_save; 230 extern int sysctl_tcp_moderate_rcvbuf; 231 extern int sysctl_tcp_tso_win_divisor; 232 extern int sysctl_tcp_abc; 233 extern int sysctl_tcp_mtu_probing; 234 extern int sysctl_tcp_base_mss; 235 extern int sysctl_tcp_workaround_signed_windows; 236 extern int sysctl_tcp_slow_start_after_idle; 237 extern int sysctl_tcp_max_ssthresh; 238 239 extern atomic_t tcp_memory_allocated; 240 extern atomic_t tcp_sockets_allocated; 241 extern int tcp_memory_pressure; 242 243 /* 244 * The next routines deal with comparing 32 bit unsigned ints 245 * and worry about wraparound (automatic with unsigned arithmetic). 246 */ 247 248 static inline int before(__u32 seq1, __u32 seq2) 249 { 250 return (__s32)(seq1-seq2) < 0; 251 } 252 #define after(seq2, seq1) before(seq1, seq2) 253 254 /* is s2<=s1<=s3 ? */ 255 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3) 256 { 257 return seq3 - seq2 >= seq1 - seq2; 258 } 259 260 static inline int tcp_too_many_orphans(struct sock *sk, int num) 261 { 262 return (num > sysctl_tcp_max_orphans) || 263 (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 264 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]); 265 } 266 267 extern struct proto tcp_prot; 268 269 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics); 270 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field) 271 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field) 272 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field) 273 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field) 274 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val) 275 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val) 276 277 extern void tcp_v4_err(struct sk_buff *skb, u32); 278 279 extern void tcp_shutdown (struct sock *sk, int how); 280 281 extern int tcp_v4_rcv(struct sk_buff *skb); 282 283 extern int tcp_v4_remember_stamp(struct sock *sk); 284 285 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 286 287 extern int tcp_sendmsg(struct kiocb *iocb, struct socket *sock, 288 struct msghdr *msg, size_t size); 289 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags); 290 291 extern int tcp_ioctl(struct sock *sk, 292 int cmd, 293 unsigned long arg); 294 295 extern int tcp_rcv_state_process(struct sock *sk, 296 struct sk_buff *skb, 297 struct tcphdr *th, 298 unsigned len); 299 300 extern int tcp_rcv_established(struct sock *sk, 301 struct sk_buff *skb, 302 struct tcphdr *th, 303 unsigned len); 304 305 extern void tcp_rcv_space_adjust(struct sock *sk); 306 307 extern void tcp_cleanup_rbuf(struct sock *sk, int copied); 308 309 extern int tcp_twsk_unique(struct sock *sk, 310 struct sock *sktw, void *twp); 311 312 extern void tcp_twsk_destructor(struct sock *sk); 313 314 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 315 struct pipe_inode_info *pipe, size_t len, unsigned int flags); 316 317 static inline void tcp_dec_quickack_mode(struct sock *sk, 318 const unsigned int pkts) 319 { 320 struct inet_connection_sock *icsk = inet_csk(sk); 321 322 if (icsk->icsk_ack.quick) { 323 if (pkts >= icsk->icsk_ack.quick) { 324 icsk->icsk_ack.quick = 0; 325 /* Leaving quickack mode we deflate ATO. */ 326 icsk->icsk_ack.ato = TCP_ATO_MIN; 327 } else 328 icsk->icsk_ack.quick -= pkts; 329 } 330 } 331 332 extern void tcp_enter_quickack_mode(struct sock *sk); 333 334 static inline void tcp_clear_options(struct tcp_options_received *rx_opt) 335 { 336 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0; 337 } 338 339 #define TCP_ECN_OK 1 340 #define TCP_ECN_QUEUE_CWR 2 341 #define TCP_ECN_DEMAND_CWR 4 342 343 static __inline__ void 344 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th) 345 { 346 if (sysctl_tcp_ecn && th->ece && th->cwr) 347 inet_rsk(req)->ecn_ok = 1; 348 } 349 350 enum tcp_tw_status 351 { 352 TCP_TW_SUCCESS = 0, 353 TCP_TW_RST = 1, 354 TCP_TW_ACK = 2, 355 TCP_TW_SYN = 3 356 }; 357 358 359 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 360 struct sk_buff *skb, 361 const struct tcphdr *th); 362 363 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb, 364 struct request_sock *req, 365 struct request_sock **prev); 366 extern int tcp_child_process(struct sock *parent, 367 struct sock *child, 368 struct sk_buff *skb); 369 extern int tcp_use_frto(struct sock *sk); 370 extern void tcp_enter_frto(struct sock *sk); 371 extern void tcp_enter_loss(struct sock *sk, int how); 372 extern void tcp_clear_retrans(struct tcp_sock *tp); 373 extern void tcp_update_metrics(struct sock *sk); 374 375 extern void tcp_close(struct sock *sk, 376 long timeout); 377 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait); 378 379 extern int tcp_getsockopt(struct sock *sk, int level, 380 int optname, 381 char __user *optval, 382 int __user *optlen); 383 extern int tcp_setsockopt(struct sock *sk, int level, 384 int optname, char __user *optval, 385 int optlen); 386 extern int compat_tcp_getsockopt(struct sock *sk, 387 int level, int optname, 388 char __user *optval, int __user *optlen); 389 extern int compat_tcp_setsockopt(struct sock *sk, 390 int level, int optname, 391 char __user *optval, int optlen); 392 extern void tcp_set_keepalive(struct sock *sk, int val); 393 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, 394 struct msghdr *msg, 395 size_t len, int nonblock, 396 int flags, int *addr_len); 397 398 extern void tcp_parse_options(struct sk_buff *skb, 399 struct tcp_options_received *opt_rx, 400 int estab); 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 extern void tcp_unhash(struct sock *sk); 437 438 /* From syncookies.c */ 439 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS]; 440 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 441 struct ip_options *opt); 442 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, 443 __u16 *mss); 444 445 extern __u32 cookie_init_timestamp(struct request_sock *req); 446 extern void cookie_check_timestamp(struct tcp_options_received *tcp_opt); 447 448 /* From net/ipv6/syncookies.c */ 449 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 450 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb, 451 __u16 *mss); 452 453 /* tcp_output.c */ 454 455 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 456 int nonagle); 457 extern int tcp_may_send_now(struct sock *sk); 458 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); 459 extern void tcp_xmit_retransmit_queue(struct sock *); 460 extern void tcp_simple_retransmit(struct sock *); 461 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32); 462 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 463 464 extern void tcp_send_probe0(struct sock *); 465 extern void tcp_send_partial(struct sock *); 466 extern int tcp_write_wakeup(struct sock *); 467 extern void tcp_send_fin(struct sock *sk); 468 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority); 469 extern int tcp_send_synack(struct sock *); 470 extern void tcp_push_one(struct sock *, unsigned int mss_now); 471 extern void tcp_send_ack(struct sock *sk); 472 extern void tcp_send_delayed_ack(struct sock *sk); 473 474 /* tcp_input.c */ 475 extern void tcp_cwnd_application_limited(struct sock *sk); 476 477 /* tcp_timer.c */ 478 extern void tcp_init_xmit_timers(struct sock *); 479 static inline void tcp_clear_xmit_timers(struct sock *sk) 480 { 481 inet_csk_clear_xmit_timers(sk); 482 } 483 484 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 485 extern unsigned int tcp_current_mss(struct sock *sk, int large); 486 487 /* tcp.c */ 488 extern void tcp_get_info(struct sock *, struct tcp_info *); 489 490 /* Read 'sendfile()'-style from a TCP socket */ 491 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 492 unsigned int, size_t); 493 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 494 sk_read_actor_t recv_actor); 495 496 extern void tcp_initialize_rcv_mss(struct sock *sk); 497 498 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu); 499 extern int tcp_mss_to_mtu(struct sock *sk, int mss); 500 extern void tcp_mtup_init(struct sock *sk); 501 502 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 503 { 504 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 505 ntohl(TCP_FLAG_ACK) | 506 snd_wnd); 507 } 508 509 static inline void tcp_fast_path_on(struct tcp_sock *tp) 510 { 511 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 512 } 513 514 static inline void tcp_fast_path_check(struct sock *sk) 515 { 516 struct tcp_sock *tp = tcp_sk(sk); 517 518 if (skb_queue_empty(&tp->out_of_order_queue) && 519 tp->rcv_wnd && 520 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 521 !tp->urg_data) 522 tcp_fast_path_on(tp); 523 } 524 525 /* Compute the actual receive window we are currently advertising. 526 * Rcv_nxt can be after the window if our peer push more data 527 * than the offered window. 528 */ 529 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 530 { 531 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 532 533 if (win < 0) 534 win = 0; 535 return (u32) win; 536 } 537 538 /* Choose a new window, without checks for shrinking, and without 539 * scaling applied to the result. The caller does these things 540 * if necessary. This is a "raw" window selection. 541 */ 542 extern u32 __tcp_select_window(struct sock *sk); 543 544 /* TCP timestamps are only 32-bits, this causes a slight 545 * complication on 64-bit systems since we store a snapshot 546 * of jiffies in the buffer control blocks below. We decided 547 * to use only the low 32-bits of jiffies and hide the ugly 548 * casts with the following macro. 549 */ 550 #define tcp_time_stamp ((__u32)(jiffies)) 551 552 /* This is what the send packet queuing engine uses to pass 553 * TCP per-packet control information to the transmission 554 * code. We also store the host-order sequence numbers in 555 * here too. This is 36 bytes on 32-bit architectures, 556 * 40 bytes on 64-bit machines, if this grows please adjust 557 * skbuff.h:skbuff->cb[xxx] size appropriately. 558 */ 559 struct tcp_skb_cb { 560 union { 561 struct inet_skb_parm h4; 562 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE) 563 struct inet6_skb_parm h6; 564 #endif 565 } header; /* For incoming frames */ 566 __u32 seq; /* Starting sequence number */ 567 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 568 __u32 when; /* used to compute rtt's */ 569 __u8 flags; /* TCP header flags. */ 570 571 /* NOTE: These must match up to the flags byte in a 572 * real TCP header. 573 */ 574 #define TCPCB_FLAG_FIN 0x01 575 #define TCPCB_FLAG_SYN 0x02 576 #define TCPCB_FLAG_RST 0x04 577 #define TCPCB_FLAG_PSH 0x08 578 #define TCPCB_FLAG_ACK 0x10 579 #define TCPCB_FLAG_URG 0x20 580 #define TCPCB_FLAG_ECE 0x40 581 #define TCPCB_FLAG_CWR 0x80 582 583 __u8 sacked; /* State flags for SACK/FACK. */ 584 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 585 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 586 #define TCPCB_LOST 0x04 /* SKB is lost */ 587 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 588 589 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 590 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 591 592 __u16 urg_ptr; /* Valid w/URG flags is set. */ 593 __u32 ack_seq; /* Sequence number ACK'd */ 594 }; 595 596 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 597 598 /* Due to TSO, an SKB can be composed of multiple actual 599 * packets. To keep these tracked properly, we use this. 600 */ 601 static inline int tcp_skb_pcount(const struct sk_buff *skb) 602 { 603 return skb_shinfo(skb)->gso_segs; 604 } 605 606 /* This is valid iff tcp_skb_pcount() > 1. */ 607 static inline int tcp_skb_mss(const struct sk_buff *skb) 608 { 609 return skb_shinfo(skb)->gso_size; 610 } 611 612 static inline void tcp_dec_pcount_approx_int(__u32 *count, const int decr) 613 { 614 if (*count) { 615 *count -= decr; 616 if ((int)*count < 0) 617 *count = 0; 618 } 619 } 620 621 static inline void tcp_dec_pcount_approx(__u32 *count, 622 const struct sk_buff *skb) 623 { 624 tcp_dec_pcount_approx_int(count, tcp_skb_pcount(skb)); 625 } 626 627 /* Events passed to congestion control interface */ 628 enum tcp_ca_event { 629 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 630 CA_EVENT_CWND_RESTART, /* congestion window restart */ 631 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 632 CA_EVENT_FRTO, /* fast recovery timeout */ 633 CA_EVENT_LOSS, /* loss timeout */ 634 CA_EVENT_FAST_ACK, /* in sequence ack */ 635 CA_EVENT_SLOW_ACK, /* other ack */ 636 }; 637 638 /* 639 * Interface for adding new TCP congestion control handlers 640 */ 641 #define TCP_CA_NAME_MAX 16 642 #define TCP_CA_MAX 128 643 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 644 645 #define TCP_CONG_NON_RESTRICTED 0x1 646 #define TCP_CONG_RTT_STAMP 0x2 647 648 struct tcp_congestion_ops { 649 struct list_head list; 650 unsigned long flags; 651 652 /* initialize private data (optional) */ 653 void (*init)(struct sock *sk); 654 /* cleanup private data (optional) */ 655 void (*release)(struct sock *sk); 656 657 /* return slow start threshold (required) */ 658 u32 (*ssthresh)(struct sock *sk); 659 /* lower bound for congestion window (optional) */ 660 u32 (*min_cwnd)(const struct sock *sk); 661 /* do new cwnd calculation (required) */ 662 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight); 663 /* call before changing ca_state (optional) */ 664 void (*set_state)(struct sock *sk, u8 new_state); 665 /* call when cwnd event occurs (optional) */ 666 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 667 /* new value of cwnd after loss (optional) */ 668 u32 (*undo_cwnd)(struct sock *sk); 669 /* hook for packet ack accounting (optional) */ 670 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 671 /* get info for inet_diag (optional) */ 672 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 673 674 char name[TCP_CA_NAME_MAX]; 675 struct module *owner; 676 }; 677 678 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type); 679 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 680 681 extern void tcp_init_congestion_control(struct sock *sk); 682 extern void tcp_cleanup_congestion_control(struct sock *sk); 683 extern int tcp_set_default_congestion_control(const char *name); 684 extern void tcp_get_default_congestion_control(char *name); 685 extern void tcp_get_available_congestion_control(char *buf, size_t len); 686 extern void tcp_get_allowed_congestion_control(char *buf, size_t len); 687 extern int tcp_set_allowed_congestion_control(char *allowed); 688 extern int tcp_set_congestion_control(struct sock *sk, const char *name); 689 extern void tcp_slow_start(struct tcp_sock *tp); 690 691 extern struct tcp_congestion_ops tcp_init_congestion_ops; 692 extern u32 tcp_reno_ssthresh(struct sock *sk); 693 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight); 694 extern u32 tcp_reno_min_cwnd(const struct sock *sk); 695 extern struct tcp_congestion_ops tcp_reno; 696 697 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 698 { 699 struct inet_connection_sock *icsk = inet_csk(sk); 700 701 if (icsk->icsk_ca_ops->set_state) 702 icsk->icsk_ca_ops->set_state(sk, ca_state); 703 icsk->icsk_ca_state = ca_state; 704 } 705 706 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 707 { 708 const struct inet_connection_sock *icsk = inet_csk(sk); 709 710 if (icsk->icsk_ca_ops->cwnd_event) 711 icsk->icsk_ca_ops->cwnd_event(sk, event); 712 } 713 714 /* These functions determine how the current flow behaves in respect of SACK 715 * handling. SACK is negotiated with the peer, and therefore it can vary 716 * between different flows. 717 * 718 * tcp_is_sack - SACK enabled 719 * tcp_is_reno - No SACK 720 * tcp_is_fack - FACK enabled, implies SACK enabled 721 */ 722 static inline int tcp_is_sack(const struct tcp_sock *tp) 723 { 724 return tp->rx_opt.sack_ok; 725 } 726 727 static inline int tcp_is_reno(const struct tcp_sock *tp) 728 { 729 return !tcp_is_sack(tp); 730 } 731 732 static inline int tcp_is_fack(const struct tcp_sock *tp) 733 { 734 return tp->rx_opt.sack_ok & 2; 735 } 736 737 static inline void tcp_enable_fack(struct tcp_sock *tp) 738 { 739 tp->rx_opt.sack_ok |= 2; 740 } 741 742 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 743 { 744 return tp->sacked_out + tp->lost_out; 745 } 746 747 /* This determines how many packets are "in the network" to the best 748 * of our knowledge. In many cases it is conservative, but where 749 * detailed information is available from the receiver (via SACK 750 * blocks etc.) we can make more aggressive calculations. 751 * 752 * Use this for decisions involving congestion control, use just 753 * tp->packets_out to determine if the send queue is empty or not. 754 * 755 * Read this equation as: 756 * 757 * "Packets sent once on transmission queue" MINUS 758 * "Packets left network, but not honestly ACKed yet" PLUS 759 * "Packets fast retransmitted" 760 */ 761 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 762 { 763 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 764 } 765 766 extern int tcp_limit_reno_sacked(struct tcp_sock *tp); 767 768 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 769 * The exception is rate halving phase, when cwnd is decreasing towards 770 * ssthresh. 771 */ 772 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 773 { 774 const struct tcp_sock *tp = tcp_sk(sk); 775 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery)) 776 return tp->snd_ssthresh; 777 else 778 return max(tp->snd_ssthresh, 779 ((tp->snd_cwnd >> 1) + 780 (tp->snd_cwnd >> 2))); 781 } 782 783 /* Use define here intentionally to get WARN_ON location shown at the caller */ 784 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 785 786 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh); 787 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst); 788 789 /* Slow start with delack produces 3 packets of burst, so that 790 * it is safe "de facto". This will be the default - same as 791 * the default reordering threshold - but if reordering increases, 792 * we must be able to allow cwnd to burst at least this much in order 793 * to not pull it back when holes are filled. 794 */ 795 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 796 { 797 return tp->reordering; 798 } 799 800 /* Returns end sequence number of the receiver's advertised window */ 801 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 802 { 803 return tp->snd_una + tp->snd_wnd; 804 } 805 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight); 806 807 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss, 808 const struct sk_buff *skb) 809 { 810 if (skb->len < mss) 811 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 812 } 813 814 static inline void tcp_check_probe_timer(struct sock *sk) 815 { 816 struct tcp_sock *tp = tcp_sk(sk); 817 const struct inet_connection_sock *icsk = inet_csk(sk); 818 819 if (!tp->packets_out && !icsk->icsk_pending) 820 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 821 icsk->icsk_rto, TCP_RTO_MAX); 822 } 823 824 static inline void tcp_push_pending_frames(struct sock *sk) 825 { 826 struct tcp_sock *tp = tcp_sk(sk); 827 828 __tcp_push_pending_frames(sk, tcp_current_mss(sk, 1), tp->nonagle); 829 } 830 831 static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq) 832 { 833 tp->snd_wl1 = seq; 834 } 835 836 static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq) 837 { 838 tp->snd_wl1 = seq; 839 } 840 841 /* 842 * Calculate(/check) TCP checksum 843 */ 844 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 845 __be32 daddr, __wsum base) 846 { 847 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 848 } 849 850 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 851 { 852 return __skb_checksum_complete(skb); 853 } 854 855 static inline int tcp_checksum_complete(struct sk_buff *skb) 856 { 857 return !skb_csum_unnecessary(skb) && 858 __tcp_checksum_complete(skb); 859 } 860 861 /* Prequeue for VJ style copy to user, combined with checksumming. */ 862 863 static inline void tcp_prequeue_init(struct tcp_sock *tp) 864 { 865 tp->ucopy.task = NULL; 866 tp->ucopy.len = 0; 867 tp->ucopy.memory = 0; 868 skb_queue_head_init(&tp->ucopy.prequeue); 869 #ifdef CONFIG_NET_DMA 870 tp->ucopy.dma_chan = NULL; 871 tp->ucopy.wakeup = 0; 872 tp->ucopy.pinned_list = NULL; 873 tp->ucopy.dma_cookie = 0; 874 #endif 875 } 876 877 /* Packet is added to VJ-style prequeue for processing in process 878 * context, if a reader task is waiting. Apparently, this exciting 879 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) 880 * failed somewhere. Latency? Burstiness? Well, at least now we will 881 * see, why it failed. 8)8) --ANK 882 * 883 * NOTE: is this not too big to inline? 884 */ 885 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb) 886 { 887 struct tcp_sock *tp = tcp_sk(sk); 888 889 if (!sysctl_tcp_low_latency && tp->ucopy.task) { 890 __skb_queue_tail(&tp->ucopy.prequeue, skb); 891 tp->ucopy.memory += skb->truesize; 892 if (tp->ucopy.memory > sk->sk_rcvbuf) { 893 struct sk_buff *skb1; 894 895 BUG_ON(sock_owned_by_user(sk)); 896 897 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { 898 sk->sk_backlog_rcv(sk, skb1); 899 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED); 900 } 901 902 tp->ucopy.memory = 0; 903 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { 904 wake_up_interruptible(sk->sk_sleep); 905 if (!inet_csk_ack_scheduled(sk)) 906 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 907 (3 * TCP_RTO_MIN) / 4, 908 TCP_RTO_MAX); 909 } 910 return 1; 911 } 912 return 0; 913 } 914 915 916 #undef STATE_TRACE 917 918 #ifdef STATE_TRACE 919 static const char *statename[]={ 920 "Unused","Established","Syn Sent","Syn Recv", 921 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 922 "Close Wait","Last ACK","Listen","Closing" 923 }; 924 #endif 925 extern void tcp_set_state(struct sock *sk, int state); 926 927 extern void tcp_done(struct sock *sk); 928 929 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 930 { 931 rx_opt->dsack = 0; 932 rx_opt->eff_sacks = 0; 933 rx_opt->num_sacks = 0; 934 } 935 936 /* Determine a window scaling and initial window to offer. */ 937 extern void tcp_select_initial_window(int __space, __u32 mss, 938 __u32 *rcv_wnd, __u32 *window_clamp, 939 int wscale_ok, __u8 *rcv_wscale); 940 941 static inline int tcp_win_from_space(int space) 942 { 943 return sysctl_tcp_adv_win_scale<=0 ? 944 (space>>(-sysctl_tcp_adv_win_scale)) : 945 space - (space>>sysctl_tcp_adv_win_scale); 946 } 947 948 /* Note: caller must be prepared to deal with negative returns */ 949 static inline int tcp_space(const struct sock *sk) 950 { 951 return tcp_win_from_space(sk->sk_rcvbuf - 952 atomic_read(&sk->sk_rmem_alloc)); 953 } 954 955 static inline int tcp_full_space(const struct sock *sk) 956 { 957 return tcp_win_from_space(sk->sk_rcvbuf); 958 } 959 960 static inline void tcp_openreq_init(struct request_sock *req, 961 struct tcp_options_received *rx_opt, 962 struct sk_buff *skb) 963 { 964 struct inet_request_sock *ireq = inet_rsk(req); 965 966 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 967 req->cookie_ts = 0; 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 } 979 980 extern void tcp_enter_memory_pressure(void); 981 982 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 983 { 984 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 985 } 986 987 static inline int keepalive_time_when(const struct tcp_sock *tp) 988 { 989 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 990 } 991 992 static 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 1003 static 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 1029 static 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 */ 1039 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1040 { 1041 tp->lost_skb_hint = NULL; 1042 tp->scoreboard_skb_hint = NULL; 1043 tp->retransmit_skb_hint = NULL; 1044 tp->forward_skb_hint = NULL; 1045 } 1046 1047 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1048 { 1049 tcp_clear_retrans_hints_partial(tp); 1050 } 1051 1052 /* MD5 Signature */ 1053 struct crypto_hash; 1054 1055 /* - key database */ 1056 struct tcp_md5sig_key { 1057 u8 *key; 1058 u8 keylen; 1059 }; 1060 1061 struct tcp4_md5sig_key { 1062 struct tcp_md5sig_key base; 1063 __be32 addr; 1064 }; 1065 1066 struct tcp6_md5sig_key { 1067 struct tcp_md5sig_key base; 1068 #if 0 1069 u32 scope_id; /* XXX */ 1070 #endif 1071 struct in6_addr addr; 1072 }; 1073 1074 /* - sock block */ 1075 struct tcp_md5sig_info { 1076 struct tcp4_md5sig_key *keys4; 1077 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1078 struct tcp6_md5sig_key *keys6; 1079 u32 entries6; 1080 u32 alloced6; 1081 #endif 1082 u32 entries4; 1083 u32 alloced4; 1084 }; 1085 1086 /* - pseudo header */ 1087 struct tcp4_pseudohdr { 1088 __be32 saddr; 1089 __be32 daddr; 1090 __u8 pad; 1091 __u8 protocol; 1092 __be16 len; 1093 }; 1094 1095 struct tcp6_pseudohdr { 1096 struct in6_addr saddr; 1097 struct in6_addr daddr; 1098 __be32 len; 1099 __be32 protocol; /* including padding */ 1100 }; 1101 1102 union tcp_md5sum_block { 1103 struct tcp4_pseudohdr ip4; 1104 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1105 struct tcp6_pseudohdr ip6; 1106 #endif 1107 }; 1108 1109 /* - pool: digest algorithm, hash description and scratch buffer */ 1110 struct tcp_md5sig_pool { 1111 struct hash_desc md5_desc; 1112 union tcp_md5sum_block md5_blk; 1113 }; 1114 1115 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */ 1116 1117 /* - functions */ 1118 extern int tcp_v4_calc_md5_hash(char *md5_hash, 1119 struct tcp_md5sig_key *key, 1120 struct sock *sk, 1121 struct dst_entry *dst, 1122 struct request_sock *req, 1123 struct tcphdr *th, 1124 int protocol, 1125 unsigned int tcplen); 1126 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1127 struct sock *addr_sk); 1128 1129 extern int tcp_v4_md5_do_add(struct sock *sk, 1130 __be32 addr, 1131 u8 *newkey, 1132 u8 newkeylen); 1133 1134 extern int tcp_v4_md5_do_del(struct sock *sk, 1135 __be32 addr); 1136 1137 extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void); 1138 extern void tcp_free_md5sig_pool(void); 1139 1140 extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu); 1141 extern void __tcp_put_md5sig_pool(void); 1142 1143 static inline 1144 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void) 1145 { 1146 int cpu = get_cpu(); 1147 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu); 1148 if (!ret) 1149 put_cpu(); 1150 return ret; 1151 } 1152 1153 static inline void tcp_put_md5sig_pool(void) 1154 { 1155 __tcp_put_md5sig_pool(); 1156 put_cpu(); 1157 } 1158 1159 /* write queue abstraction */ 1160 static inline void tcp_write_queue_purge(struct sock *sk) 1161 { 1162 struct sk_buff *skb; 1163 1164 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1165 sk_wmem_free_skb(sk, skb); 1166 sk_mem_reclaim(sk); 1167 } 1168 1169 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk) 1170 { 1171 struct sk_buff *skb = sk->sk_write_queue.next; 1172 if (skb == (struct sk_buff *) &sk->sk_write_queue) 1173 return NULL; 1174 return skb; 1175 } 1176 1177 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk) 1178 { 1179 struct sk_buff *skb = sk->sk_write_queue.prev; 1180 if (skb == (struct sk_buff *) &sk->sk_write_queue) 1181 return NULL; 1182 return skb; 1183 } 1184 1185 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb) 1186 { 1187 return skb->next; 1188 } 1189 1190 #define tcp_for_write_queue(skb, sk) \ 1191 for (skb = (sk)->sk_write_queue.next; \ 1192 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \ 1193 skb = skb->next) 1194 1195 #define tcp_for_write_queue_from(skb, sk) \ 1196 for (; (skb != (struct sk_buff *)&(sk)->sk_write_queue);\ 1197 skb = skb->next) 1198 1199 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1200 for (tmp = skb->next; \ 1201 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \ 1202 skb = tmp, tmp = skb->next) 1203 1204 static inline struct sk_buff *tcp_send_head(struct sock *sk) 1205 { 1206 return sk->sk_send_head; 1207 } 1208 1209 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb) 1210 { 1211 sk->sk_send_head = skb->next; 1212 if (sk->sk_send_head == (struct sk_buff *)&sk->sk_write_queue) 1213 sk->sk_send_head = NULL; 1214 } 1215 1216 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1217 { 1218 if (sk->sk_send_head == skb_unlinked) 1219 sk->sk_send_head = NULL; 1220 } 1221 1222 static inline void tcp_init_send_head(struct sock *sk) 1223 { 1224 sk->sk_send_head = NULL; 1225 } 1226 1227 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1228 { 1229 __skb_queue_tail(&sk->sk_write_queue, skb); 1230 } 1231 1232 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1233 { 1234 __tcp_add_write_queue_tail(sk, skb); 1235 1236 /* Queue it, remembering where we must start sending. */ 1237 if (sk->sk_send_head == NULL) { 1238 sk->sk_send_head = skb; 1239 1240 if (tcp_sk(sk)->highest_sack == NULL) 1241 tcp_sk(sk)->highest_sack = skb; 1242 } 1243 } 1244 1245 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1246 { 1247 __skb_queue_head(&sk->sk_write_queue, skb); 1248 } 1249 1250 /* Insert buff after skb on the write queue of sk. */ 1251 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1252 struct sk_buff *buff, 1253 struct sock *sk) 1254 { 1255 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1256 } 1257 1258 /* Insert skb between prev and next on the write queue of sk. */ 1259 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1260 struct sk_buff *skb, 1261 struct sock *sk) 1262 { 1263 __skb_insert(new, skb->prev, skb, &sk->sk_write_queue); 1264 1265 if (sk->sk_send_head == skb) 1266 sk->sk_send_head = new; 1267 } 1268 1269 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1270 { 1271 __skb_unlink(skb, &sk->sk_write_queue); 1272 } 1273 1274 static inline int tcp_skb_is_last(const struct sock *sk, 1275 const struct sk_buff *skb) 1276 { 1277 return skb->next == (struct sk_buff *)&sk->sk_write_queue; 1278 } 1279 1280 static inline int tcp_write_queue_empty(struct sock *sk) 1281 { 1282 return skb_queue_empty(&sk->sk_write_queue); 1283 } 1284 1285 /* Start sequence of the highest skb with SACKed bit, valid only if 1286 * sacked > 0 or when the caller has ensured validity by itself. 1287 */ 1288 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1289 { 1290 if (!tp->sacked_out) 1291 return tp->snd_una; 1292 1293 if (tp->highest_sack == NULL) 1294 return tp->snd_nxt; 1295 1296 return TCP_SKB_CB(tp->highest_sack)->seq; 1297 } 1298 1299 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1300 { 1301 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1302 tcp_write_queue_next(sk, skb); 1303 } 1304 1305 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1306 { 1307 return tcp_sk(sk)->highest_sack; 1308 } 1309 1310 static inline void tcp_highest_sack_reset(struct sock *sk) 1311 { 1312 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1313 } 1314 1315 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1316 static inline void tcp_highest_sack_combine(struct sock *sk, 1317 struct sk_buff *old, 1318 struct sk_buff *new) 1319 { 1320 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1321 tcp_sk(sk)->highest_sack = new; 1322 } 1323 1324 /* /proc */ 1325 enum tcp_seq_states { 1326 TCP_SEQ_STATE_LISTENING, 1327 TCP_SEQ_STATE_OPENREQ, 1328 TCP_SEQ_STATE_ESTABLISHED, 1329 TCP_SEQ_STATE_TIME_WAIT, 1330 }; 1331 1332 struct tcp_seq_afinfo { 1333 char *name; 1334 sa_family_t family; 1335 struct file_operations seq_fops; 1336 struct seq_operations seq_ops; 1337 }; 1338 1339 struct tcp_iter_state { 1340 struct seq_net_private p; 1341 sa_family_t family; 1342 enum tcp_seq_states state; 1343 struct sock *syn_wait_sk; 1344 int bucket, sbucket, num, uid; 1345 }; 1346 1347 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1348 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1349 1350 extern struct request_sock_ops tcp_request_sock_ops; 1351 extern struct request_sock_ops tcp6_request_sock_ops; 1352 1353 extern int tcp_v4_destroy_sock(struct sock *sk); 1354 1355 extern int tcp_v4_gso_send_check(struct sk_buff *skb); 1356 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features); 1357 1358 #ifdef CONFIG_PROC_FS 1359 extern int tcp4_proc_init(void); 1360 extern void tcp4_proc_exit(void); 1361 #endif 1362 1363 /* TCP af-specific functions */ 1364 struct tcp_sock_af_ops { 1365 #ifdef CONFIG_TCP_MD5SIG 1366 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1367 struct sock *addr_sk); 1368 int (*calc_md5_hash) (char *location, 1369 struct tcp_md5sig_key *md5, 1370 struct sock *sk, 1371 struct dst_entry *dst, 1372 struct request_sock *req, 1373 struct tcphdr *th, 1374 int protocol, 1375 unsigned int len); 1376 int (*md5_add) (struct sock *sk, 1377 struct sock *addr_sk, 1378 u8 *newkey, 1379 u8 len); 1380 int (*md5_parse) (struct sock *sk, 1381 char __user *optval, 1382 int optlen); 1383 #endif 1384 }; 1385 1386 struct tcp_request_sock_ops { 1387 #ifdef CONFIG_TCP_MD5SIG 1388 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1389 struct request_sock *req); 1390 #endif 1391 }; 1392 1393 extern void tcp_v4_init(void); 1394 extern void tcp_init(void); 1395 1396 #endif /* _TCP_H */ 1397