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 struct percpu_counter 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 struct percpu_counter 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); 485 486 /* Bound MSS / TSO packet size with the half of the window */ 487 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 488 { 489 if (tp->max_window && pktsize > (tp->max_window >> 1)) 490 return max(tp->max_window >> 1, 68U - tp->tcp_header_len); 491 else 492 return pktsize; 493 } 494 495 /* tcp.c */ 496 extern void tcp_get_info(struct sock *, struct tcp_info *); 497 498 /* Read 'sendfile()'-style from a TCP socket */ 499 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 500 unsigned int, size_t); 501 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 502 sk_read_actor_t recv_actor); 503 504 extern void tcp_initialize_rcv_mss(struct sock *sk); 505 506 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu); 507 extern int tcp_mss_to_mtu(struct sock *sk, int mss); 508 extern void tcp_mtup_init(struct sock *sk); 509 510 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 511 { 512 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 513 ntohl(TCP_FLAG_ACK) | 514 snd_wnd); 515 } 516 517 static inline void tcp_fast_path_on(struct tcp_sock *tp) 518 { 519 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 520 } 521 522 static inline void tcp_fast_path_check(struct sock *sk) 523 { 524 struct tcp_sock *tp = tcp_sk(sk); 525 526 if (skb_queue_empty(&tp->out_of_order_queue) && 527 tp->rcv_wnd && 528 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 529 !tp->urg_data) 530 tcp_fast_path_on(tp); 531 } 532 533 /* Compute the actual receive window we are currently advertising. 534 * Rcv_nxt can be after the window if our peer push more data 535 * than the offered window. 536 */ 537 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 538 { 539 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 540 541 if (win < 0) 542 win = 0; 543 return (u32) win; 544 } 545 546 /* Choose a new window, without checks for shrinking, and without 547 * scaling applied to the result. The caller does these things 548 * if necessary. This is a "raw" window selection. 549 */ 550 extern u32 __tcp_select_window(struct sock *sk); 551 552 /* TCP timestamps are only 32-bits, this causes a slight 553 * complication on 64-bit systems since we store a snapshot 554 * of jiffies in the buffer control blocks below. We decided 555 * to use only the low 32-bits of jiffies and hide the ugly 556 * casts with the following macro. 557 */ 558 #define tcp_time_stamp ((__u32)(jiffies)) 559 560 /* This is what the send packet queuing engine uses to pass 561 * TCP per-packet control information to the transmission 562 * code. We also store the host-order sequence numbers in 563 * here too. This is 36 bytes on 32-bit architectures, 564 * 40 bytes on 64-bit machines, if this grows please adjust 565 * skbuff.h:skbuff->cb[xxx] size appropriately. 566 */ 567 struct tcp_skb_cb { 568 union { 569 struct inet_skb_parm h4; 570 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE) 571 struct inet6_skb_parm h6; 572 #endif 573 } header; /* For incoming frames */ 574 __u32 seq; /* Starting sequence number */ 575 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 576 __u32 when; /* used to compute rtt's */ 577 __u8 flags; /* TCP header flags. */ 578 579 /* NOTE: These must match up to the flags byte in a 580 * real TCP header. 581 */ 582 #define TCPCB_FLAG_FIN 0x01 583 #define TCPCB_FLAG_SYN 0x02 584 #define TCPCB_FLAG_RST 0x04 585 #define TCPCB_FLAG_PSH 0x08 586 #define TCPCB_FLAG_ACK 0x10 587 #define TCPCB_FLAG_URG 0x20 588 #define TCPCB_FLAG_ECE 0x40 589 #define TCPCB_FLAG_CWR 0x80 590 591 __u8 sacked; /* State flags for SACK/FACK. */ 592 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 593 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 594 #define TCPCB_LOST 0x04 /* SKB is lost */ 595 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 596 597 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 598 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 599 600 __u32 ack_seq; /* Sequence number ACK'd */ 601 }; 602 603 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 604 605 /* Due to TSO, an SKB can be composed of multiple actual 606 * packets. To keep these tracked properly, we use this. 607 */ 608 static inline int tcp_skb_pcount(const struct sk_buff *skb) 609 { 610 return skb_shinfo(skb)->gso_segs; 611 } 612 613 /* This is valid iff tcp_skb_pcount() > 1. */ 614 static inline int tcp_skb_mss(const struct sk_buff *skb) 615 { 616 return skb_shinfo(skb)->gso_size; 617 } 618 619 /* Events passed to congestion control interface */ 620 enum tcp_ca_event { 621 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 622 CA_EVENT_CWND_RESTART, /* congestion window restart */ 623 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 624 CA_EVENT_FRTO, /* fast recovery timeout */ 625 CA_EVENT_LOSS, /* loss timeout */ 626 CA_EVENT_FAST_ACK, /* in sequence ack */ 627 CA_EVENT_SLOW_ACK, /* other ack */ 628 }; 629 630 /* 631 * Interface for adding new TCP congestion control handlers 632 */ 633 #define TCP_CA_NAME_MAX 16 634 #define TCP_CA_MAX 128 635 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 636 637 #define TCP_CONG_NON_RESTRICTED 0x1 638 #define TCP_CONG_RTT_STAMP 0x2 639 640 struct tcp_congestion_ops { 641 struct list_head list; 642 unsigned long flags; 643 644 /* initialize private data (optional) */ 645 void (*init)(struct sock *sk); 646 /* cleanup private data (optional) */ 647 void (*release)(struct sock *sk); 648 649 /* return slow start threshold (required) */ 650 u32 (*ssthresh)(struct sock *sk); 651 /* lower bound for congestion window (optional) */ 652 u32 (*min_cwnd)(const struct sock *sk); 653 /* do new cwnd calculation (required) */ 654 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight); 655 /* call before changing ca_state (optional) */ 656 void (*set_state)(struct sock *sk, u8 new_state); 657 /* call when cwnd event occurs (optional) */ 658 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 659 /* new value of cwnd after loss (optional) */ 660 u32 (*undo_cwnd)(struct sock *sk); 661 /* hook for packet ack accounting (optional) */ 662 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 663 /* get info for inet_diag (optional) */ 664 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 665 666 char name[TCP_CA_NAME_MAX]; 667 struct module *owner; 668 }; 669 670 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type); 671 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 672 673 extern void tcp_init_congestion_control(struct sock *sk); 674 extern void tcp_cleanup_congestion_control(struct sock *sk); 675 extern int tcp_set_default_congestion_control(const char *name); 676 extern void tcp_get_default_congestion_control(char *name); 677 extern void tcp_get_available_congestion_control(char *buf, size_t len); 678 extern void tcp_get_allowed_congestion_control(char *buf, size_t len); 679 extern int tcp_set_allowed_congestion_control(char *allowed); 680 extern int tcp_set_congestion_control(struct sock *sk, const char *name); 681 extern void tcp_slow_start(struct tcp_sock *tp); 682 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 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". This will be the default - same as 782 * the default reordering threshold - but if reordering increases, 783 * we must be able to allow cwnd to burst at least this much in order 784 * to not pull it back when holes are filled. 785 */ 786 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 787 { 788 return tp->reordering; 789 } 790 791 /* Returns end sequence number of the receiver's advertised window */ 792 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 793 { 794 return tp->snd_una + tp->snd_wnd; 795 } 796 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight); 797 798 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss, 799 const struct sk_buff *skb) 800 { 801 if (skb->len < mss) 802 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 803 } 804 805 static inline void tcp_check_probe_timer(struct sock *sk) 806 { 807 struct tcp_sock *tp = tcp_sk(sk); 808 const struct inet_connection_sock *icsk = inet_csk(sk); 809 810 if (!tp->packets_out && !icsk->icsk_pending) 811 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 812 icsk->icsk_rto, TCP_RTO_MAX); 813 } 814 815 static inline void tcp_push_pending_frames(struct sock *sk) 816 { 817 struct tcp_sock *tp = tcp_sk(sk); 818 819 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 820 } 821 822 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 823 { 824 tp->snd_wl1 = seq; 825 } 826 827 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 828 { 829 tp->snd_wl1 = seq; 830 } 831 832 /* 833 * Calculate(/check) TCP checksum 834 */ 835 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 836 __be32 daddr, __wsum base) 837 { 838 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 839 } 840 841 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 842 { 843 return __skb_checksum_complete(skb); 844 } 845 846 static inline int tcp_checksum_complete(struct sk_buff *skb) 847 { 848 return !skb_csum_unnecessary(skb) && 849 __tcp_checksum_complete(skb); 850 } 851 852 /* Prequeue for VJ style copy to user, combined with checksumming. */ 853 854 static inline void tcp_prequeue_init(struct tcp_sock *tp) 855 { 856 tp->ucopy.task = NULL; 857 tp->ucopy.len = 0; 858 tp->ucopy.memory = 0; 859 skb_queue_head_init(&tp->ucopy.prequeue); 860 #ifdef CONFIG_NET_DMA 861 tp->ucopy.dma_chan = NULL; 862 tp->ucopy.wakeup = 0; 863 tp->ucopy.pinned_list = NULL; 864 tp->ucopy.dma_cookie = 0; 865 #endif 866 } 867 868 /* Packet is added to VJ-style prequeue for processing in process 869 * context, if a reader task is waiting. Apparently, this exciting 870 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) 871 * failed somewhere. Latency? Burstiness? Well, at least now we will 872 * see, why it failed. 8)8) --ANK 873 * 874 * NOTE: is this not too big to inline? 875 */ 876 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb) 877 { 878 struct tcp_sock *tp = tcp_sk(sk); 879 880 if (!sysctl_tcp_low_latency && tp->ucopy.task) { 881 __skb_queue_tail(&tp->ucopy.prequeue, skb); 882 tp->ucopy.memory += skb->truesize; 883 if (tp->ucopy.memory > sk->sk_rcvbuf) { 884 struct sk_buff *skb1; 885 886 BUG_ON(sock_owned_by_user(sk)); 887 888 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { 889 sk_backlog_rcv(sk, skb1); 890 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPPREQUEUEDROPPED); 891 } 892 893 tp->ucopy.memory = 0; 894 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { 895 wake_up_interruptible(sk->sk_sleep); 896 if (!inet_csk_ack_scheduled(sk)) 897 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 898 (3 * TCP_RTO_MIN) / 4, 899 TCP_RTO_MAX); 900 } 901 return 1; 902 } 903 return 0; 904 } 905 906 907 #undef STATE_TRACE 908 909 #ifdef STATE_TRACE 910 static const char *statename[]={ 911 "Unused","Established","Syn Sent","Syn Recv", 912 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 913 "Close Wait","Last ACK","Listen","Closing" 914 }; 915 #endif 916 extern void tcp_set_state(struct sock *sk, int state); 917 918 extern void tcp_done(struct sock *sk); 919 920 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 921 { 922 rx_opt->dsack = 0; 923 rx_opt->num_sacks = 0; 924 } 925 926 /* Determine a window scaling and initial window to offer. */ 927 extern void tcp_select_initial_window(int __space, __u32 mss, 928 __u32 *rcv_wnd, __u32 *window_clamp, 929 int wscale_ok, __u8 *rcv_wscale); 930 931 static inline int tcp_win_from_space(int space) 932 { 933 return sysctl_tcp_adv_win_scale<=0 ? 934 (space>>(-sysctl_tcp_adv_win_scale)) : 935 space - (space>>sysctl_tcp_adv_win_scale); 936 } 937 938 /* Note: caller must be prepared to deal with negative returns */ 939 static inline int tcp_space(const struct sock *sk) 940 { 941 return tcp_win_from_space(sk->sk_rcvbuf - 942 atomic_read(&sk->sk_rmem_alloc)); 943 } 944 945 static inline int tcp_full_space(const struct sock *sk) 946 { 947 return tcp_win_from_space(sk->sk_rcvbuf); 948 } 949 950 static inline void tcp_openreq_init(struct request_sock *req, 951 struct tcp_options_received *rx_opt, 952 struct sk_buff *skb) 953 { 954 struct inet_request_sock *ireq = inet_rsk(req); 955 956 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 957 req->cookie_ts = 0; 958 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 959 req->mss = rx_opt->mss_clamp; 960 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 961 ireq->tstamp_ok = rx_opt->tstamp_ok; 962 ireq->sack_ok = rx_opt->sack_ok; 963 ireq->snd_wscale = rx_opt->snd_wscale; 964 ireq->wscale_ok = rx_opt->wscale_ok; 965 ireq->acked = 0; 966 ireq->ecn_ok = 0; 967 ireq->rmt_port = tcp_hdr(skb)->source; 968 ireq->loc_port = tcp_hdr(skb)->dest; 969 } 970 971 extern void tcp_enter_memory_pressure(struct sock *sk); 972 973 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 974 { 975 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 976 } 977 978 static inline int keepalive_time_when(const struct tcp_sock *tp) 979 { 980 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 981 } 982 983 static inline int tcp_fin_time(const struct sock *sk) 984 { 985 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 986 const int rto = inet_csk(sk)->icsk_rto; 987 988 if (fin_timeout < (rto << 2) - (rto >> 1)) 989 fin_timeout = (rto << 2) - (rto >> 1); 990 991 return fin_timeout; 992 } 993 994 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, 995 int paws_win) 996 { 997 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 998 return 1; 999 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1000 return 1; 1001 1002 return 0; 1003 } 1004 1005 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt, 1006 int rst) 1007 { 1008 if (tcp_paws_check(rx_opt, 0)) 1009 return 0; 1010 1011 /* RST segments are not recommended to carry timestamp, 1012 and, if they do, it is recommended to ignore PAWS because 1013 "their cleanup function should take precedence over timestamps." 1014 Certainly, it is mistake. It is necessary to understand the reasons 1015 of this constraint to relax it: if peer reboots, clock may go 1016 out-of-sync and half-open connections will not be reset. 1017 Actually, the problem would be not existing if all 1018 the implementations followed draft about maintaining clock 1019 via reboots. Linux-2.2 DOES NOT! 1020 1021 However, we can relax time bounds for RST segments to MSL. 1022 */ 1023 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1024 return 0; 1025 return 1; 1026 } 1027 1028 #define TCP_CHECK_TIMER(sk) do { } while (0) 1029 1030 static inline void tcp_mib_init(struct net *net) 1031 { 1032 /* See RFC 2012 */ 1033 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1034 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1035 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1036 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1037 } 1038 1039 /* from STCP */ 1040 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1041 { 1042 tp->lost_skb_hint = NULL; 1043 tp->scoreboard_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 tp->retransmit_skb_hint = NULL; 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_md5_hash_skb(char *md5_hash, 1119 struct tcp_md5sig_key *key, 1120 struct sock *sk, 1121 struct request_sock *req, 1122 struct sk_buff *skb); 1123 1124 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1125 struct sock *addr_sk); 1126 1127 extern int tcp_v4_md5_do_add(struct sock *sk, 1128 __be32 addr, 1129 u8 *newkey, 1130 u8 newkeylen); 1131 1132 extern int tcp_v4_md5_do_del(struct sock *sk, 1133 __be32 addr); 1134 1135 #ifdef CONFIG_TCP_MD5SIG 1136 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \ 1137 &(struct tcp_md5sig_key) { \ 1138 .key = (twsk)->tw_md5_key, \ 1139 .keylen = (twsk)->tw_md5_keylen, \ 1140 } : NULL) 1141 #else 1142 #define tcp_twsk_md5_key(twsk) NULL 1143 #endif 1144 1145 extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void); 1146 extern void tcp_free_md5sig_pool(void); 1147 1148 extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu); 1149 extern void __tcp_put_md5sig_pool(void); 1150 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *); 1151 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *, 1152 unsigned header_len); 1153 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1154 struct tcp_md5sig_key *key); 1155 1156 static inline 1157 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void) 1158 { 1159 int cpu = get_cpu(); 1160 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu); 1161 if (!ret) 1162 put_cpu(); 1163 return ret; 1164 } 1165 1166 static inline void tcp_put_md5sig_pool(void) 1167 { 1168 __tcp_put_md5sig_pool(); 1169 put_cpu(); 1170 } 1171 1172 /* write queue abstraction */ 1173 static inline void tcp_write_queue_purge(struct sock *sk) 1174 { 1175 struct sk_buff *skb; 1176 1177 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1178 sk_wmem_free_skb(sk, skb); 1179 sk_mem_reclaim(sk); 1180 } 1181 1182 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk) 1183 { 1184 return skb_peek(&sk->sk_write_queue); 1185 } 1186 1187 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk) 1188 { 1189 return skb_peek_tail(&sk->sk_write_queue); 1190 } 1191 1192 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb) 1193 { 1194 return skb_queue_next(&sk->sk_write_queue, skb); 1195 } 1196 1197 static inline struct sk_buff *tcp_write_queue_prev(struct sock *sk, struct sk_buff *skb) 1198 { 1199 return skb_queue_prev(&sk->sk_write_queue, skb); 1200 } 1201 1202 #define tcp_for_write_queue(skb, sk) \ 1203 skb_queue_walk(&(sk)->sk_write_queue, skb) 1204 1205 #define tcp_for_write_queue_from(skb, sk) \ 1206 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1207 1208 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1209 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1210 1211 static inline struct sk_buff *tcp_send_head(struct sock *sk) 1212 { 1213 return sk->sk_send_head; 1214 } 1215 1216 static inline bool tcp_skb_is_last(const struct sock *sk, 1217 const struct sk_buff *skb) 1218 { 1219 return skb_queue_is_last(&sk->sk_write_queue, skb); 1220 } 1221 1222 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb) 1223 { 1224 if (tcp_skb_is_last(sk, skb)) 1225 sk->sk_send_head = NULL; 1226 else 1227 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1228 } 1229 1230 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1231 { 1232 if (sk->sk_send_head == skb_unlinked) 1233 sk->sk_send_head = NULL; 1234 } 1235 1236 static inline void tcp_init_send_head(struct sock *sk) 1237 { 1238 sk->sk_send_head = NULL; 1239 } 1240 1241 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1242 { 1243 __skb_queue_tail(&sk->sk_write_queue, skb); 1244 } 1245 1246 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1247 { 1248 __tcp_add_write_queue_tail(sk, skb); 1249 1250 /* Queue it, remembering where we must start sending. */ 1251 if (sk->sk_send_head == NULL) { 1252 sk->sk_send_head = skb; 1253 1254 if (tcp_sk(sk)->highest_sack == NULL) 1255 tcp_sk(sk)->highest_sack = skb; 1256 } 1257 } 1258 1259 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1260 { 1261 __skb_queue_head(&sk->sk_write_queue, skb); 1262 } 1263 1264 /* Insert buff after skb on the write queue of sk. */ 1265 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1266 struct sk_buff *buff, 1267 struct sock *sk) 1268 { 1269 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1270 } 1271 1272 /* Insert new before skb on the write queue of sk. */ 1273 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1274 struct sk_buff *skb, 1275 struct sock *sk) 1276 { 1277 __skb_queue_before(&sk->sk_write_queue, skb, new); 1278 1279 if (sk->sk_send_head == skb) 1280 sk->sk_send_head = new; 1281 } 1282 1283 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1284 { 1285 __skb_unlink(skb, &sk->sk_write_queue); 1286 } 1287 1288 static inline int tcp_write_queue_empty(struct sock *sk) 1289 { 1290 return skb_queue_empty(&sk->sk_write_queue); 1291 } 1292 1293 /* Start sequence of the highest skb with SACKed bit, valid only if 1294 * sacked > 0 or when the caller has ensured validity by itself. 1295 */ 1296 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1297 { 1298 if (!tp->sacked_out) 1299 return tp->snd_una; 1300 1301 if (tp->highest_sack == NULL) 1302 return tp->snd_nxt; 1303 1304 return TCP_SKB_CB(tp->highest_sack)->seq; 1305 } 1306 1307 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1308 { 1309 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1310 tcp_write_queue_next(sk, skb); 1311 } 1312 1313 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1314 { 1315 return tcp_sk(sk)->highest_sack; 1316 } 1317 1318 static inline void tcp_highest_sack_reset(struct sock *sk) 1319 { 1320 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1321 } 1322 1323 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1324 static inline void tcp_highest_sack_combine(struct sock *sk, 1325 struct sk_buff *old, 1326 struct sk_buff *new) 1327 { 1328 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1329 tcp_sk(sk)->highest_sack = new; 1330 } 1331 1332 /* /proc */ 1333 enum tcp_seq_states { 1334 TCP_SEQ_STATE_LISTENING, 1335 TCP_SEQ_STATE_OPENREQ, 1336 TCP_SEQ_STATE_ESTABLISHED, 1337 TCP_SEQ_STATE_TIME_WAIT, 1338 }; 1339 1340 struct tcp_seq_afinfo { 1341 char *name; 1342 sa_family_t family; 1343 struct file_operations seq_fops; 1344 struct seq_operations seq_ops; 1345 }; 1346 1347 struct tcp_iter_state { 1348 struct seq_net_private p; 1349 sa_family_t family; 1350 enum tcp_seq_states state; 1351 struct sock *syn_wait_sk; 1352 int bucket, sbucket, num, uid; 1353 }; 1354 1355 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1356 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1357 1358 extern struct request_sock_ops tcp_request_sock_ops; 1359 extern struct request_sock_ops tcp6_request_sock_ops; 1360 1361 extern void tcp_v4_destroy_sock(struct sock *sk); 1362 1363 extern int tcp_v4_gso_send_check(struct sk_buff *skb); 1364 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features); 1365 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head, 1366 struct sk_buff *skb); 1367 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head, 1368 struct sk_buff *skb); 1369 extern int tcp_gro_complete(struct sk_buff *skb); 1370 extern int tcp4_gro_complete(struct sk_buff *skb); 1371 1372 #ifdef CONFIG_PROC_FS 1373 extern int tcp4_proc_init(void); 1374 extern void tcp4_proc_exit(void); 1375 #endif 1376 1377 /* TCP af-specific functions */ 1378 struct tcp_sock_af_ops { 1379 #ifdef CONFIG_TCP_MD5SIG 1380 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1381 struct sock *addr_sk); 1382 int (*calc_md5_hash) (char *location, 1383 struct tcp_md5sig_key *md5, 1384 struct sock *sk, 1385 struct request_sock *req, 1386 struct sk_buff *skb); 1387 int (*md5_add) (struct sock *sk, 1388 struct sock *addr_sk, 1389 u8 *newkey, 1390 u8 len); 1391 int (*md5_parse) (struct sock *sk, 1392 char __user *optval, 1393 int optlen); 1394 #endif 1395 }; 1396 1397 struct tcp_request_sock_ops { 1398 #ifdef CONFIG_TCP_MD5SIG 1399 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1400 struct request_sock *req); 1401 #endif 1402 }; 1403 1404 extern void tcp_v4_init(void); 1405 extern void tcp_init(void); 1406 1407 #endif /* _TCP_H */ 1408