1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * Definitions for the TCP module. 7 * 8 * Version: @(#)tcp.h 1.0.5 05/23/93 9 * 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * 13 * This program is free software; you can redistribute it and/or 14 * modify it under the terms of the GNU General Public License 15 * as published by the Free Software Foundation; either version 16 * 2 of the License, or (at your option) any later version. 17 */ 18 #ifndef _TCP_H 19 #define _TCP_H 20 21 #define TCP_DEBUG 1 22 #define FASTRETRANS_DEBUG 1 23 24 #include <linux/list.h> 25 #include <linux/tcp.h> 26 #include <linux/slab.h> 27 #include <linux/cache.h> 28 #include <linux/percpu.h> 29 #include <linux/skbuff.h> 30 #include <linux/dmaengine.h> 31 #include <linux/crypto.h> 32 33 #include <net/inet_connection_sock.h> 34 #include <net/inet_timewait_sock.h> 35 #include <net/inet_hashtables.h> 36 #include <net/checksum.h> 37 #include <net/request_sock.h> 38 #include <net/sock.h> 39 #include <net/snmp.h> 40 #include <net/ip.h> 41 #include <net/tcp_states.h> 42 #include <net/inet_ecn.h> 43 44 #include <linux/seq_file.h> 45 46 extern struct inet_hashinfo tcp_hashinfo; 47 48 extern atomic_t tcp_orphan_count; 49 extern void tcp_time_wait(struct sock *sk, int state, int timeo); 50 51 #define MAX_TCP_HEADER (128 + MAX_HEADER) 52 53 /* 54 * Never offer a window over 32767 without using window scaling. Some 55 * poor stacks do signed 16bit maths! 56 */ 57 #define MAX_TCP_WINDOW 32767U 58 59 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 60 #define TCP_MIN_MSS 88U 61 62 /* Minimal RCV_MSS. */ 63 #define TCP_MIN_RCVMSS 536U 64 65 /* The least MTU to use for probing */ 66 #define TCP_BASE_MSS 512 67 68 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 69 #define TCP_FASTRETRANS_THRESH 3 70 71 /* Maximal reordering. */ 72 #define TCP_MAX_REORDERING 127 73 74 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 75 #define TCP_MAX_QUICKACKS 16U 76 77 /* urg_data states */ 78 #define TCP_URG_VALID 0x0100 79 #define TCP_URG_NOTYET 0x0200 80 #define TCP_URG_READ 0x0400 81 82 #define TCP_RETR1 3 /* 83 * This is how many retries it does before it 84 * tries to figure out if the gateway is 85 * down. Minimal RFC value is 3; it corresponds 86 * to ~3sec-8min depending on RTO. 87 */ 88 89 #define TCP_RETR2 15 /* 90 * This should take at least 91 * 90 minutes to time out. 92 * RFC1122 says that the limit is 100 sec. 93 * 15 is ~13-30min depending on RTO. 94 */ 95 96 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a 97 * connection: ~180sec is RFC minimum */ 98 99 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a 100 * connection: ~180sec is RFC minimum */ 101 102 103 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned 104 * socket. 7 is ~50sec-16min. 105 */ 106 107 108 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 109 * state, about 60 seconds */ 110 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 111 /* BSD style FIN_WAIT2 deadlock breaker. 112 * It used to be 3min, new value is 60sec, 113 * to combine FIN-WAIT-2 timeout with 114 * TIME-WAIT timer. 115 */ 116 117 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 118 #if HZ >= 100 119 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 120 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 121 #else 122 #define TCP_DELACK_MIN 4U 123 #define TCP_ATO_MIN 4U 124 #endif 125 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 126 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 127 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */ 128 129 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 130 * for local resources. 131 */ 132 133 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 134 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 135 #define TCP_KEEPALIVE_INTVL (75*HZ) 136 137 #define MAX_TCP_KEEPIDLE 32767 138 #define MAX_TCP_KEEPINTVL 32767 139 #define MAX_TCP_KEEPCNT 127 140 #define MAX_TCP_SYNCNT 127 141 142 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 143 144 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 145 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 146 * after this time. It should be equal 147 * (or greater than) TCP_TIMEWAIT_LEN 148 * to provide reliability equal to one 149 * provided by timewait state. 150 */ 151 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 152 * timestamps. It must be less than 153 * minimal timewait lifetime. 154 */ 155 /* 156 * TCP option 157 */ 158 159 #define TCPOPT_NOP 1 /* Padding */ 160 #define TCPOPT_EOL 0 /* End of options */ 161 #define TCPOPT_MSS 2 /* Segment size negotiating */ 162 #define TCPOPT_WINDOW 3 /* Window scaling */ 163 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 164 #define TCPOPT_SACK 5 /* SACK Block */ 165 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 166 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 167 168 /* 169 * TCP option lengths 170 */ 171 172 #define TCPOLEN_MSS 4 173 #define TCPOLEN_WINDOW 3 174 #define TCPOLEN_SACK_PERM 2 175 #define TCPOLEN_TIMESTAMP 10 176 #define TCPOLEN_MD5SIG 18 177 178 /* But this is what stacks really send out. */ 179 #define TCPOLEN_TSTAMP_ALIGNED 12 180 #define TCPOLEN_WSCALE_ALIGNED 4 181 #define TCPOLEN_SACKPERM_ALIGNED 4 182 #define TCPOLEN_SACK_BASE 2 183 #define TCPOLEN_SACK_BASE_ALIGNED 4 184 #define TCPOLEN_SACK_PERBLOCK 8 185 #define TCPOLEN_MD5SIG_ALIGNED 20 186 187 /* Flags in tp->nonagle */ 188 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 189 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 190 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 191 192 extern struct inet_timewait_death_row tcp_death_row; 193 194 /* sysctl variables for tcp */ 195 extern int sysctl_tcp_timestamps; 196 extern int sysctl_tcp_window_scaling; 197 extern int sysctl_tcp_sack; 198 extern int sysctl_tcp_fin_timeout; 199 extern int sysctl_tcp_keepalive_time; 200 extern int sysctl_tcp_keepalive_probes; 201 extern int sysctl_tcp_keepalive_intvl; 202 extern int sysctl_tcp_syn_retries; 203 extern int sysctl_tcp_synack_retries; 204 extern int sysctl_tcp_retries1; 205 extern int sysctl_tcp_retries2; 206 extern int sysctl_tcp_orphan_retries; 207 extern int sysctl_tcp_syncookies; 208 extern int sysctl_tcp_retrans_collapse; 209 extern int sysctl_tcp_stdurg; 210 extern int sysctl_tcp_rfc1337; 211 extern int sysctl_tcp_abort_on_overflow; 212 extern int sysctl_tcp_max_orphans; 213 extern int sysctl_tcp_fack; 214 extern int sysctl_tcp_reordering; 215 extern int sysctl_tcp_ecn; 216 extern int sysctl_tcp_dsack; 217 extern int sysctl_tcp_mem[3]; 218 extern int sysctl_tcp_wmem[3]; 219 extern int sysctl_tcp_rmem[3]; 220 extern int sysctl_tcp_app_win; 221 extern int sysctl_tcp_adv_win_scale; 222 extern int sysctl_tcp_tw_reuse; 223 extern int sysctl_tcp_frto; 224 extern int sysctl_tcp_frto_response; 225 extern int sysctl_tcp_low_latency; 226 extern int sysctl_tcp_dma_copybreak; 227 extern int sysctl_tcp_nometrics_save; 228 extern int sysctl_tcp_moderate_rcvbuf; 229 extern int sysctl_tcp_tso_win_divisor; 230 extern int sysctl_tcp_abc; 231 extern int sysctl_tcp_mtu_probing; 232 extern int sysctl_tcp_base_mss; 233 extern int sysctl_tcp_workaround_signed_windows; 234 extern int sysctl_tcp_slow_start_after_idle; 235 extern int sysctl_tcp_max_ssthresh; 236 237 extern atomic_t tcp_memory_allocated; 238 extern atomic_t tcp_sockets_allocated; 239 extern int tcp_memory_pressure; 240 241 /* 242 * The next routines deal with comparing 32 bit unsigned ints 243 * and worry about wraparound (automatic with unsigned arithmetic). 244 */ 245 246 static inline int before(__u32 seq1, __u32 seq2) 247 { 248 return (__s32)(seq1-seq2) < 0; 249 } 250 #define after(seq2, seq1) before(seq1, seq2) 251 252 /* is s2<=s1<=s3 ? */ 253 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3) 254 { 255 return seq3 - seq2 >= seq1 - seq2; 256 } 257 258 static inline int tcp_too_many_orphans(struct sock *sk, int num) 259 { 260 return (num > sysctl_tcp_max_orphans) || 261 (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 262 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]); 263 } 264 265 extern struct proto tcp_prot; 266 267 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics); 268 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field) 269 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field) 270 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field) 271 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field) 272 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val) 273 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(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 /* 401 * TCP v4 functions exported for the inet6 API 402 */ 403 404 extern void tcp_v4_send_check(struct sock *sk, int len, 405 struct sk_buff *skb); 406 407 extern int tcp_v4_conn_request(struct sock *sk, 408 struct sk_buff *skb); 409 410 extern struct sock * tcp_create_openreq_child(struct sock *sk, 411 struct request_sock *req, 412 struct sk_buff *skb); 413 414 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, 415 struct sk_buff *skb, 416 struct request_sock *req, 417 struct dst_entry *dst); 418 419 extern int tcp_v4_do_rcv(struct sock *sk, 420 struct sk_buff *skb); 421 422 extern int tcp_v4_connect(struct sock *sk, 423 struct sockaddr *uaddr, 424 int addr_len); 425 426 extern int tcp_connect(struct sock *sk); 427 428 extern struct sk_buff * tcp_make_synack(struct sock *sk, 429 struct dst_entry *dst, 430 struct request_sock *req); 431 432 extern int tcp_disconnect(struct sock *sk, int flags); 433 434 extern void tcp_unhash(struct sock *sk); 435 436 /* From syncookies.c */ 437 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 438 struct ip_options *opt); 439 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, 440 __u16 *mss); 441 442 /* tcp_output.c */ 443 444 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 445 int nonagle); 446 extern int tcp_may_send_now(struct sock *sk); 447 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); 448 extern void tcp_xmit_retransmit_queue(struct sock *); 449 extern void tcp_simple_retransmit(struct sock *); 450 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32); 451 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 452 453 extern void tcp_send_probe0(struct sock *); 454 extern void tcp_send_partial(struct sock *); 455 extern int tcp_write_wakeup(struct sock *); 456 extern void tcp_send_fin(struct sock *sk); 457 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority); 458 extern int tcp_send_synack(struct sock *); 459 extern void tcp_push_one(struct sock *, unsigned int mss_now); 460 extern void tcp_send_ack(struct sock *sk); 461 extern void tcp_send_delayed_ack(struct sock *sk); 462 463 /* tcp_input.c */ 464 extern void tcp_cwnd_application_limited(struct sock *sk); 465 466 /* tcp_timer.c */ 467 extern void tcp_init_xmit_timers(struct sock *); 468 static inline void tcp_clear_xmit_timers(struct sock *sk) 469 { 470 inet_csk_clear_xmit_timers(sk); 471 } 472 473 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 474 extern unsigned int tcp_current_mss(struct sock *sk, int large); 475 476 /* tcp.c */ 477 extern void tcp_get_info(struct sock *, struct tcp_info *); 478 479 /* Read 'sendfile()'-style from a TCP socket */ 480 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 481 unsigned int, size_t); 482 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 483 sk_read_actor_t recv_actor); 484 485 extern void tcp_initialize_rcv_mss(struct sock *sk); 486 487 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu); 488 extern int tcp_mss_to_mtu(struct sock *sk, int mss); 489 extern void tcp_mtup_init(struct sock *sk); 490 491 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 492 { 493 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 494 ntohl(TCP_FLAG_ACK) | 495 snd_wnd); 496 } 497 498 static inline void tcp_fast_path_on(struct tcp_sock *tp) 499 { 500 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 501 } 502 503 static inline void tcp_fast_path_check(struct sock *sk) 504 { 505 struct tcp_sock *tp = tcp_sk(sk); 506 507 if (skb_queue_empty(&tp->out_of_order_queue) && 508 tp->rcv_wnd && 509 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 510 !tp->urg_data) 511 tcp_fast_path_on(tp); 512 } 513 514 /* Compute the actual receive window we are currently advertising. 515 * Rcv_nxt can be after the window if our peer push more data 516 * than the offered window. 517 */ 518 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 519 { 520 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 521 522 if (win < 0) 523 win = 0; 524 return (u32) win; 525 } 526 527 /* Choose a new window, without checks for shrinking, and without 528 * scaling applied to the result. The caller does these things 529 * if necessary. This is a "raw" window selection. 530 */ 531 extern u32 __tcp_select_window(struct sock *sk); 532 533 /* TCP timestamps are only 32-bits, this causes a slight 534 * complication on 64-bit systems since we store a snapshot 535 * of jiffies in the buffer control blocks below. We decided 536 * to use only the low 32-bits of jiffies and hide the ugly 537 * casts with the following macro. 538 */ 539 #define tcp_time_stamp ((__u32)(jiffies)) 540 541 /* This is what the send packet queuing engine uses to pass 542 * TCP per-packet control information to the transmission 543 * code. We also store the host-order sequence numbers in 544 * here too. This is 36 bytes on 32-bit architectures, 545 * 40 bytes on 64-bit machines, if this grows please adjust 546 * skbuff.h:skbuff->cb[xxx] size appropriately. 547 */ 548 struct tcp_skb_cb { 549 union { 550 struct inet_skb_parm h4; 551 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE) 552 struct inet6_skb_parm h6; 553 #endif 554 } header; /* For incoming frames */ 555 __u32 seq; /* Starting sequence number */ 556 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 557 __u32 when; /* used to compute rtt's */ 558 __u8 flags; /* TCP header flags. */ 559 560 /* NOTE: These must match up to the flags byte in a 561 * real TCP header. 562 */ 563 #define TCPCB_FLAG_FIN 0x01 564 #define TCPCB_FLAG_SYN 0x02 565 #define TCPCB_FLAG_RST 0x04 566 #define TCPCB_FLAG_PSH 0x08 567 #define TCPCB_FLAG_ACK 0x10 568 #define TCPCB_FLAG_URG 0x20 569 #define TCPCB_FLAG_ECE 0x40 570 #define TCPCB_FLAG_CWR 0x80 571 572 __u8 sacked; /* State flags for SACK/FACK. */ 573 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 574 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 575 #define TCPCB_LOST 0x04 /* SKB is lost */ 576 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 577 578 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 579 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 580 581 #define TCPCB_URG 0x20 /* Urgent pointer advanced here */ 582 583 #define TCPCB_AT_TAIL (TCPCB_URG) 584 585 __u16 urg_ptr; /* Valid w/URG flags is set. */ 586 __u32 ack_seq; /* Sequence number ACK'd */ 587 }; 588 589 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 590 591 /* Due to TSO, an SKB can be composed of multiple actual 592 * packets. To keep these tracked properly, we use this. 593 */ 594 static inline int tcp_skb_pcount(const struct sk_buff *skb) 595 { 596 return skb_shinfo(skb)->gso_segs; 597 } 598 599 /* This is valid iff tcp_skb_pcount() > 1. */ 600 static inline int tcp_skb_mss(const struct sk_buff *skb) 601 { 602 return skb_shinfo(skb)->gso_size; 603 } 604 605 static inline void tcp_dec_pcount_approx_int(__u32 *count, const int decr) 606 { 607 if (*count) { 608 *count -= decr; 609 if ((int)*count < 0) 610 *count = 0; 611 } 612 } 613 614 static inline void tcp_dec_pcount_approx(__u32 *count, 615 const struct sk_buff *skb) 616 { 617 tcp_dec_pcount_approx_int(count, tcp_skb_pcount(skb)); 618 } 619 620 /* Events passed to congestion control interface */ 621 enum tcp_ca_event { 622 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 623 CA_EVENT_CWND_RESTART, /* congestion window restart */ 624 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 625 CA_EVENT_FRTO, /* fast recovery timeout */ 626 CA_EVENT_LOSS, /* loss timeout */ 627 CA_EVENT_FAST_ACK, /* in sequence ack */ 628 CA_EVENT_SLOW_ACK, /* other ack */ 629 }; 630 631 /* 632 * Interface for adding new TCP congestion control handlers 633 */ 634 #define TCP_CA_NAME_MAX 16 635 #define TCP_CA_MAX 128 636 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 637 638 #define TCP_CONG_NON_RESTRICTED 0x1 639 #define TCP_CONG_RTT_STAMP 0x2 640 641 struct tcp_congestion_ops { 642 struct list_head list; 643 unsigned long flags; 644 645 /* initialize private data (optional) */ 646 void (*init)(struct sock *sk); 647 /* cleanup private data (optional) */ 648 void (*release)(struct sock *sk); 649 650 /* return slow start threshold (required) */ 651 u32 (*ssthresh)(struct sock *sk); 652 /* lower bound for congestion window (optional) */ 653 u32 (*min_cwnd)(const struct sock *sk); 654 /* do new cwnd calculation (required) */ 655 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight); 656 /* call before changing ca_state (optional) */ 657 void (*set_state)(struct sock *sk, u8 new_state); 658 /* call when cwnd event occurs (optional) */ 659 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 660 /* new value of cwnd after loss (optional) */ 661 u32 (*undo_cwnd)(struct sock *sk); 662 /* hook for packet ack accounting (optional) */ 663 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 664 /* get info for inet_diag (optional) */ 665 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 666 667 char name[TCP_CA_NAME_MAX]; 668 struct module *owner; 669 }; 670 671 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type); 672 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 673 674 extern void tcp_init_congestion_control(struct sock *sk); 675 extern void tcp_cleanup_congestion_control(struct sock *sk); 676 extern int tcp_set_default_congestion_control(const char *name); 677 extern void tcp_get_default_congestion_control(char *name); 678 extern void tcp_get_available_congestion_control(char *buf, size_t len); 679 extern void tcp_get_allowed_congestion_control(char *buf, size_t len); 680 extern int tcp_set_allowed_congestion_control(char *allowed); 681 extern int tcp_set_congestion_control(struct sock *sk, const char *name); 682 extern void tcp_slow_start(struct tcp_sock *tp); 683 684 extern struct tcp_congestion_ops tcp_init_congestion_ops; 685 extern u32 tcp_reno_ssthresh(struct sock *sk); 686 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight); 687 extern u32 tcp_reno_min_cwnd(const struct sock *sk); 688 extern struct tcp_congestion_ops tcp_reno; 689 690 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 691 { 692 struct inet_connection_sock *icsk = inet_csk(sk); 693 694 if (icsk->icsk_ca_ops->set_state) 695 icsk->icsk_ca_ops->set_state(sk, ca_state); 696 icsk->icsk_ca_state = ca_state; 697 } 698 699 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 700 { 701 const struct inet_connection_sock *icsk = inet_csk(sk); 702 703 if (icsk->icsk_ca_ops->cwnd_event) 704 icsk->icsk_ca_ops->cwnd_event(sk, event); 705 } 706 707 /* These functions determine how the current flow behaves in respect of SACK 708 * handling. SACK is negotiated with the peer, and therefore it can vary 709 * between different flows. 710 * 711 * tcp_is_sack - SACK enabled 712 * tcp_is_reno - No SACK 713 * tcp_is_fack - FACK enabled, implies SACK enabled 714 */ 715 static inline int tcp_is_sack(const struct tcp_sock *tp) 716 { 717 return tp->rx_opt.sack_ok; 718 } 719 720 static inline int tcp_is_reno(const struct tcp_sock *tp) 721 { 722 return !tcp_is_sack(tp); 723 } 724 725 static inline int tcp_is_fack(const struct tcp_sock *tp) 726 { 727 return tp->rx_opt.sack_ok & 2; 728 } 729 730 static inline void tcp_enable_fack(struct tcp_sock *tp) 731 { 732 tp->rx_opt.sack_ok |= 2; 733 } 734 735 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 736 { 737 return tp->sacked_out + tp->lost_out; 738 } 739 740 /* This determines how many packets are "in the network" to the best 741 * of our knowledge. In many cases it is conservative, but where 742 * detailed information is available from the receiver (via SACK 743 * blocks etc.) we can make more aggressive calculations. 744 * 745 * Use this for decisions involving congestion control, use just 746 * tp->packets_out to determine if the send queue is empty or not. 747 * 748 * Read this equation as: 749 * 750 * "Packets sent once on transmission queue" MINUS 751 * "Packets left network, but not honestly ACKed yet" PLUS 752 * "Packets fast retransmitted" 753 */ 754 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 755 { 756 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 757 } 758 759 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 760 * The exception is rate halving phase, when cwnd is decreasing towards 761 * ssthresh. 762 */ 763 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 764 { 765 const struct tcp_sock *tp = tcp_sk(sk); 766 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery)) 767 return tp->snd_ssthresh; 768 else 769 return max(tp->snd_ssthresh, 770 ((tp->snd_cwnd >> 1) + 771 (tp->snd_cwnd >> 2))); 772 } 773 774 /* Use define here intentionally to get WARN_ON location shown at the caller */ 775 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 776 777 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh); 778 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst); 779 780 /* Slow start with delack produces 3 packets of burst, so that 781 * it is safe "de facto". 782 */ 783 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 784 { 785 return 3; 786 } 787 788 /* RFC2861 Check whether we are limited by application or congestion window 789 * This is the inverse of cwnd check in tcp_tso_should_defer 790 */ 791 static inline int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight) 792 { 793 const struct tcp_sock *tp = tcp_sk(sk); 794 u32 left; 795 796 if (in_flight >= tp->snd_cwnd) 797 return 1; 798 799 if (!sk_can_gso(sk)) 800 return 0; 801 802 left = tp->snd_cwnd - in_flight; 803 if (sysctl_tcp_tso_win_divisor) 804 return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd; 805 else 806 return left <= tcp_max_burst(tp); 807 } 808 809 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss, 810 const struct sk_buff *skb) 811 { 812 if (skb->len < mss) 813 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 814 } 815 816 static inline void tcp_check_probe_timer(struct sock *sk) 817 { 818 struct tcp_sock *tp = tcp_sk(sk); 819 const struct inet_connection_sock *icsk = inet_csk(sk); 820 821 if (!tp->packets_out && !icsk->icsk_pending) 822 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 823 icsk->icsk_rto, TCP_RTO_MAX); 824 } 825 826 static inline void tcp_push_pending_frames(struct sock *sk) 827 { 828 struct tcp_sock *tp = tcp_sk(sk); 829 830 __tcp_push_pending_frames(sk, tcp_current_mss(sk, 1), tp->nonagle); 831 } 832 833 static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq) 834 { 835 tp->snd_wl1 = seq; 836 } 837 838 static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq) 839 { 840 tp->snd_wl1 = seq; 841 } 842 843 /* 844 * Calculate(/check) TCP checksum 845 */ 846 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 847 __be32 daddr, __wsum base) 848 { 849 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 850 } 851 852 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 853 { 854 return __skb_checksum_complete(skb); 855 } 856 857 static inline int tcp_checksum_complete(struct sk_buff *skb) 858 { 859 return !skb_csum_unnecessary(skb) && 860 __tcp_checksum_complete(skb); 861 } 862 863 /* Prequeue for VJ style copy to user, combined with checksumming. */ 864 865 static inline void tcp_prequeue_init(struct tcp_sock *tp) 866 { 867 tp->ucopy.task = NULL; 868 tp->ucopy.len = 0; 869 tp->ucopy.memory = 0; 870 skb_queue_head_init(&tp->ucopy.prequeue); 871 #ifdef CONFIG_NET_DMA 872 tp->ucopy.dma_chan = NULL; 873 tp->ucopy.wakeup = 0; 874 tp->ucopy.pinned_list = NULL; 875 tp->ucopy.dma_cookie = 0; 876 #endif 877 } 878 879 /* Packet is added to VJ-style prequeue for processing in process 880 * context, if a reader task is waiting. Apparently, this exciting 881 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) 882 * failed somewhere. Latency? Burstiness? Well, at least now we will 883 * see, why it failed. 8)8) --ANK 884 * 885 * NOTE: is this not too big to inline? 886 */ 887 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb) 888 { 889 struct tcp_sock *tp = tcp_sk(sk); 890 891 if (!sysctl_tcp_low_latency && tp->ucopy.task) { 892 __skb_queue_tail(&tp->ucopy.prequeue, skb); 893 tp->ucopy.memory += skb->truesize; 894 if (tp->ucopy.memory > sk->sk_rcvbuf) { 895 struct sk_buff *skb1; 896 897 BUG_ON(sock_owned_by_user(sk)); 898 899 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { 900 sk->sk_backlog_rcv(sk, skb1); 901 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED); 902 } 903 904 tp->ucopy.memory = 0; 905 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { 906 wake_up_interruptible(sk->sk_sleep); 907 if (!inet_csk_ack_scheduled(sk)) 908 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 909 (3 * TCP_RTO_MIN) / 4, 910 TCP_RTO_MAX); 911 } 912 return 1; 913 } 914 return 0; 915 } 916 917 918 #undef STATE_TRACE 919 920 #ifdef STATE_TRACE 921 static const char *statename[]={ 922 "Unused","Established","Syn Sent","Syn Recv", 923 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 924 "Close Wait","Last ACK","Listen","Closing" 925 }; 926 #endif 927 928 static inline void tcp_set_state(struct sock *sk, int state) 929 { 930 int oldstate = sk->sk_state; 931 932 switch (state) { 933 case TCP_ESTABLISHED: 934 if (oldstate != TCP_ESTABLISHED) 935 TCP_INC_STATS(TCP_MIB_CURRESTAB); 936 break; 937 938 case TCP_CLOSE: 939 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED) 940 TCP_INC_STATS(TCP_MIB_ESTABRESETS); 941 942 sk->sk_prot->unhash(sk); 943 if (inet_csk(sk)->icsk_bind_hash && 944 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) 945 inet_put_port(&tcp_hashinfo, sk); 946 /* fall through */ 947 default: 948 if (oldstate==TCP_ESTABLISHED) 949 TCP_DEC_STATS(TCP_MIB_CURRESTAB); 950 } 951 952 /* Change state AFTER socket is unhashed to avoid closed 953 * socket sitting in hash tables. 954 */ 955 sk->sk_state = state; 956 957 #ifdef STATE_TRACE 958 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]); 959 #endif 960 } 961 962 extern void tcp_done(struct sock *sk); 963 964 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 965 { 966 rx_opt->dsack = 0; 967 rx_opt->eff_sacks = 0; 968 rx_opt->num_sacks = 0; 969 } 970 971 /* Determine a window scaling and initial window to offer. */ 972 extern void tcp_select_initial_window(int __space, __u32 mss, 973 __u32 *rcv_wnd, __u32 *window_clamp, 974 int wscale_ok, __u8 *rcv_wscale); 975 976 static inline int tcp_win_from_space(int space) 977 { 978 return sysctl_tcp_adv_win_scale<=0 ? 979 (space>>(-sysctl_tcp_adv_win_scale)) : 980 space - (space>>sysctl_tcp_adv_win_scale); 981 } 982 983 /* Note: caller must be prepared to deal with negative returns */ 984 static inline int tcp_space(const struct sock *sk) 985 { 986 return tcp_win_from_space(sk->sk_rcvbuf - 987 atomic_read(&sk->sk_rmem_alloc)); 988 } 989 990 static inline int tcp_full_space(const struct sock *sk) 991 { 992 return tcp_win_from_space(sk->sk_rcvbuf); 993 } 994 995 static inline void tcp_openreq_init(struct request_sock *req, 996 struct tcp_options_received *rx_opt, 997 struct sk_buff *skb) 998 { 999 struct inet_request_sock *ireq = inet_rsk(req); 1000 1001 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 1002 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 1003 req->mss = rx_opt->mss_clamp; 1004 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 1005 ireq->tstamp_ok = rx_opt->tstamp_ok; 1006 ireq->sack_ok = rx_opt->sack_ok; 1007 ireq->snd_wscale = rx_opt->snd_wscale; 1008 ireq->wscale_ok = rx_opt->wscale_ok; 1009 ireq->acked = 0; 1010 ireq->ecn_ok = 0; 1011 ireq->rmt_port = tcp_hdr(skb)->source; 1012 } 1013 1014 extern void tcp_enter_memory_pressure(void); 1015 1016 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1017 { 1018 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1019 } 1020 1021 static inline int keepalive_time_when(const struct tcp_sock *tp) 1022 { 1023 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1024 } 1025 1026 static inline int tcp_fin_time(const struct sock *sk) 1027 { 1028 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1029 const int rto = inet_csk(sk)->icsk_rto; 1030 1031 if (fin_timeout < (rto << 2) - (rto >> 1)) 1032 fin_timeout = (rto << 2) - (rto >> 1); 1033 1034 return fin_timeout; 1035 } 1036 1037 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst) 1038 { 1039 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0) 1040 return 0; 1041 if (get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS) 1042 return 0; 1043 1044 /* RST segments are not recommended to carry timestamp, 1045 and, if they do, it is recommended to ignore PAWS because 1046 "their cleanup function should take precedence over timestamps." 1047 Certainly, it is mistake. It is necessary to understand the reasons 1048 of this constraint to relax it: if peer reboots, clock may go 1049 out-of-sync and half-open connections will not be reset. 1050 Actually, the problem would be not existing if all 1051 the implementations followed draft about maintaining clock 1052 via reboots. Linux-2.2 DOES NOT! 1053 1054 However, we can relax time bounds for RST segments to MSL. 1055 */ 1056 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1057 return 0; 1058 return 1; 1059 } 1060 1061 #define TCP_CHECK_TIMER(sk) do { } while (0) 1062 1063 static inline void tcp_mib_init(void) 1064 { 1065 /* See RFC 2012 */ 1066 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1); 1067 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1068 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1069 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1); 1070 } 1071 1072 /* from STCP */ 1073 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1074 { 1075 tp->lost_skb_hint = NULL; 1076 tp->scoreboard_skb_hint = NULL; 1077 tp->retransmit_skb_hint = NULL; 1078 tp->forward_skb_hint = NULL; 1079 } 1080 1081 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1082 { 1083 tcp_clear_retrans_hints_partial(tp); 1084 } 1085 1086 /* MD5 Signature */ 1087 struct crypto_hash; 1088 1089 /* - key database */ 1090 struct tcp_md5sig_key { 1091 u8 *key; 1092 u8 keylen; 1093 }; 1094 1095 struct tcp4_md5sig_key { 1096 struct tcp_md5sig_key base; 1097 __be32 addr; 1098 }; 1099 1100 struct tcp6_md5sig_key { 1101 struct tcp_md5sig_key base; 1102 #if 0 1103 u32 scope_id; /* XXX */ 1104 #endif 1105 struct in6_addr addr; 1106 }; 1107 1108 /* - sock block */ 1109 struct tcp_md5sig_info { 1110 struct tcp4_md5sig_key *keys4; 1111 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1112 struct tcp6_md5sig_key *keys6; 1113 u32 entries6; 1114 u32 alloced6; 1115 #endif 1116 u32 entries4; 1117 u32 alloced4; 1118 }; 1119 1120 /* - pseudo header */ 1121 struct tcp4_pseudohdr { 1122 __be32 saddr; 1123 __be32 daddr; 1124 __u8 pad; 1125 __u8 protocol; 1126 __be16 len; 1127 }; 1128 1129 struct tcp6_pseudohdr { 1130 struct in6_addr saddr; 1131 struct in6_addr daddr; 1132 __be32 len; 1133 __be32 protocol; /* including padding */ 1134 }; 1135 1136 union tcp_md5sum_block { 1137 struct tcp4_pseudohdr ip4; 1138 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1139 struct tcp6_pseudohdr ip6; 1140 #endif 1141 }; 1142 1143 /* - pool: digest algorithm, hash description and scratch buffer */ 1144 struct tcp_md5sig_pool { 1145 struct hash_desc md5_desc; 1146 union tcp_md5sum_block md5_blk; 1147 }; 1148 1149 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */ 1150 1151 /* - functions */ 1152 extern int tcp_v4_calc_md5_hash(char *md5_hash, 1153 struct tcp_md5sig_key *key, 1154 struct sock *sk, 1155 struct dst_entry *dst, 1156 struct request_sock *req, 1157 struct tcphdr *th, 1158 int protocol, 1159 unsigned int tcplen); 1160 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1161 struct sock *addr_sk); 1162 1163 extern int tcp_v4_md5_do_add(struct sock *sk, 1164 __be32 addr, 1165 u8 *newkey, 1166 u8 newkeylen); 1167 1168 extern int tcp_v4_md5_do_del(struct sock *sk, 1169 __be32 addr); 1170 1171 extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void); 1172 extern void tcp_free_md5sig_pool(void); 1173 1174 extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu); 1175 extern void __tcp_put_md5sig_pool(void); 1176 1177 static inline 1178 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void) 1179 { 1180 int cpu = get_cpu(); 1181 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu); 1182 if (!ret) 1183 put_cpu(); 1184 return ret; 1185 } 1186 1187 static inline void tcp_put_md5sig_pool(void) 1188 { 1189 __tcp_put_md5sig_pool(); 1190 put_cpu(); 1191 } 1192 1193 /* write queue abstraction */ 1194 static inline void tcp_write_queue_purge(struct sock *sk) 1195 { 1196 struct sk_buff *skb; 1197 1198 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1199 sk_wmem_free_skb(sk, skb); 1200 sk_mem_reclaim(sk); 1201 } 1202 1203 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk) 1204 { 1205 struct sk_buff *skb = sk->sk_write_queue.next; 1206 if (skb == (struct sk_buff *) &sk->sk_write_queue) 1207 return NULL; 1208 return skb; 1209 } 1210 1211 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk) 1212 { 1213 struct sk_buff *skb = sk->sk_write_queue.prev; 1214 if (skb == (struct sk_buff *) &sk->sk_write_queue) 1215 return NULL; 1216 return skb; 1217 } 1218 1219 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb) 1220 { 1221 return skb->next; 1222 } 1223 1224 #define tcp_for_write_queue(skb, sk) \ 1225 for (skb = (sk)->sk_write_queue.next; \ 1226 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \ 1227 skb = skb->next) 1228 1229 #define tcp_for_write_queue_from(skb, sk) \ 1230 for (; (skb != (struct sk_buff *)&(sk)->sk_write_queue);\ 1231 skb = skb->next) 1232 1233 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1234 for (tmp = skb->next; \ 1235 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \ 1236 skb = tmp, tmp = skb->next) 1237 1238 static inline struct sk_buff *tcp_send_head(struct sock *sk) 1239 { 1240 return sk->sk_send_head; 1241 } 1242 1243 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb) 1244 { 1245 sk->sk_send_head = skb->next; 1246 if (sk->sk_send_head == (struct sk_buff *)&sk->sk_write_queue) 1247 sk->sk_send_head = NULL; 1248 } 1249 1250 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1251 { 1252 if (sk->sk_send_head == skb_unlinked) 1253 sk->sk_send_head = NULL; 1254 } 1255 1256 static inline void tcp_init_send_head(struct sock *sk) 1257 { 1258 sk->sk_send_head = NULL; 1259 } 1260 1261 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1262 { 1263 __skb_queue_tail(&sk->sk_write_queue, skb); 1264 } 1265 1266 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1267 { 1268 __tcp_add_write_queue_tail(sk, skb); 1269 1270 /* Queue it, remembering where we must start sending. */ 1271 if (sk->sk_send_head == NULL) { 1272 sk->sk_send_head = skb; 1273 1274 if (tcp_sk(sk)->highest_sack == NULL) 1275 tcp_sk(sk)->highest_sack = skb; 1276 } 1277 } 1278 1279 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1280 { 1281 __skb_queue_head(&sk->sk_write_queue, skb); 1282 } 1283 1284 /* Insert buff after skb on the write queue of sk. */ 1285 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1286 struct sk_buff *buff, 1287 struct sock *sk) 1288 { 1289 __skb_append(skb, buff, &sk->sk_write_queue); 1290 } 1291 1292 /* Insert skb between prev and next on the write queue of sk. */ 1293 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1294 struct sk_buff *skb, 1295 struct sock *sk) 1296 { 1297 __skb_insert(new, skb->prev, skb, &sk->sk_write_queue); 1298 1299 if (sk->sk_send_head == skb) 1300 sk->sk_send_head = new; 1301 } 1302 1303 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1304 { 1305 __skb_unlink(skb, &sk->sk_write_queue); 1306 } 1307 1308 static inline int tcp_skb_is_last(const struct sock *sk, 1309 const struct sk_buff *skb) 1310 { 1311 return skb->next == (struct sk_buff *)&sk->sk_write_queue; 1312 } 1313 1314 static inline int tcp_write_queue_empty(struct sock *sk) 1315 { 1316 return skb_queue_empty(&sk->sk_write_queue); 1317 } 1318 1319 /* Start sequence of the highest skb with SACKed bit, valid only if 1320 * sacked > 0 or when the caller has ensured validity by itself. 1321 */ 1322 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1323 { 1324 if (!tp->sacked_out) 1325 return tp->snd_una; 1326 1327 if (tp->highest_sack == NULL) 1328 return tp->snd_nxt; 1329 1330 return TCP_SKB_CB(tp->highest_sack)->seq; 1331 } 1332 1333 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1334 { 1335 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1336 tcp_write_queue_next(sk, skb); 1337 } 1338 1339 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1340 { 1341 return tcp_sk(sk)->highest_sack; 1342 } 1343 1344 static inline void tcp_highest_sack_reset(struct sock *sk) 1345 { 1346 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1347 } 1348 1349 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1350 static inline void tcp_highest_sack_combine(struct sock *sk, 1351 struct sk_buff *old, 1352 struct sk_buff *new) 1353 { 1354 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1355 tcp_sk(sk)->highest_sack = new; 1356 } 1357 1358 /* /proc */ 1359 enum tcp_seq_states { 1360 TCP_SEQ_STATE_LISTENING, 1361 TCP_SEQ_STATE_OPENREQ, 1362 TCP_SEQ_STATE_ESTABLISHED, 1363 TCP_SEQ_STATE_TIME_WAIT, 1364 }; 1365 1366 struct tcp_seq_afinfo { 1367 struct module *owner; 1368 char *name; 1369 sa_family_t family; 1370 int (*seq_show) (struct seq_file *m, void *v); 1371 struct file_operations *seq_fops; 1372 }; 1373 1374 struct tcp_iter_state { 1375 sa_family_t family; 1376 enum tcp_seq_states state; 1377 struct sock *syn_wait_sk; 1378 int bucket, sbucket, num, uid; 1379 struct seq_operations seq_ops; 1380 }; 1381 1382 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo); 1383 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo); 1384 1385 extern struct request_sock_ops tcp_request_sock_ops; 1386 1387 extern int tcp_v4_destroy_sock(struct sock *sk); 1388 1389 extern int tcp_v4_gso_send_check(struct sk_buff *skb); 1390 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features); 1391 1392 #ifdef CONFIG_PROC_FS 1393 extern int tcp4_proc_init(void); 1394 extern void tcp4_proc_exit(void); 1395 #endif 1396 1397 /* TCP af-specific functions */ 1398 struct tcp_sock_af_ops { 1399 #ifdef CONFIG_TCP_MD5SIG 1400 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1401 struct sock *addr_sk); 1402 int (*calc_md5_hash) (char *location, 1403 struct tcp_md5sig_key *md5, 1404 struct sock *sk, 1405 struct dst_entry *dst, 1406 struct request_sock *req, 1407 struct tcphdr *th, 1408 int protocol, 1409 unsigned int len); 1410 int (*md5_add) (struct sock *sk, 1411 struct sock *addr_sk, 1412 u8 *newkey, 1413 u8 len); 1414 int (*md5_parse) (struct sock *sk, 1415 char __user *optval, 1416 int optlen); 1417 #endif 1418 }; 1419 1420 struct tcp_request_sock_ops { 1421 #ifdef CONFIG_TCP_MD5SIG 1422 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1423 struct request_sock *req); 1424 #endif 1425 }; 1426 1427 extern void tcp_v4_init(struct net_proto_family *ops); 1428 extern void tcp_init(void); 1429 1430 #endif /* _TCP_H */ 1431