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#include <linux/kref.h> 34 35#include <net/inet_connection_sock.h> 36#include <net/inet_timewait_sock.h> 37#include <net/inet_hashtables.h> 38#include <net/checksum.h> 39#include <net/request_sock.h> 40#include <net/sock.h> 41#include <net/snmp.h> 42#include <net/ip.h> 43#include <net/tcp_states.h> 44#include <net/inet_ecn.h> 45#include <net/dst.h> 46 47#include <linux/seq_file.h> 48 49extern struct inet_hashinfo tcp_hashinfo; 50 51extern struct percpu_counter tcp_orphan_count; 52extern void tcp_time_wait(struct sock *sk, int state, int timeo); 53 54#define MAX_TCP_HEADER (128 + MAX_HEADER) 55#define MAX_TCP_OPTION_SPACE 40 56 57/* 58 * Never offer a window over 32767 without using window scaling. Some 59 * poor stacks do signed 16bit maths! 60 */ 61#define MAX_TCP_WINDOW 32767U 62 63/* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 64#define TCP_MIN_MSS 88U 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 127 142 143#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 144 145#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 146#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 147 * after this time. It should be equal 148 * (or greater than) TCP_TIMEWAIT_LEN 149 * to provide reliability equal to one 150 * provided by timewait state. 151 */ 152#define TCP_PAWS_WINDOW 1 /* Replay window for per-host 153 * timestamps. It must be less than 154 * minimal timewait lifetime. 155 */ 156/* 157 * TCP option 158 */ 159 160#define TCPOPT_NOP 1 /* Padding */ 161#define TCPOPT_EOL 0 /* End of options */ 162#define TCPOPT_MSS 2 /* Segment size negotiating */ 163#define TCPOPT_WINDOW 3 /* Window scaling */ 164#define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 165#define TCPOPT_SACK 5 /* SACK Block */ 166#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 167#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 168#define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */ 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#define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */ 180#define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */ 181#define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN) 182#define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX) 183 184/* But this is what stacks really send out. */ 185#define TCPOLEN_TSTAMP_ALIGNED 12 186#define TCPOLEN_WSCALE_ALIGNED 4 187#define TCPOLEN_SACKPERM_ALIGNED 4 188#define TCPOLEN_SACK_BASE 2 189#define TCPOLEN_SACK_BASE_ALIGNED 4 190#define TCPOLEN_SACK_PERBLOCK 8 191#define TCPOLEN_MD5SIG_ALIGNED 20 192#define TCPOLEN_MSS_ALIGNED 4 193 194/* Flags in tp->nonagle */ 195#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 196#define TCP_NAGLE_CORK 2 /* Socket is corked */ 197#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 198 199/* TCP thin-stream limits */ 200#define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 201 202extern struct inet_timewait_death_row tcp_death_row; 203 204/* sysctl variables for tcp */ 205extern int sysctl_tcp_timestamps; 206extern int sysctl_tcp_window_scaling; 207extern int sysctl_tcp_sack; 208extern int sysctl_tcp_fin_timeout; 209extern int sysctl_tcp_keepalive_time; 210extern int sysctl_tcp_keepalive_probes; 211extern int sysctl_tcp_keepalive_intvl; 212extern int sysctl_tcp_syn_retries; 213extern int sysctl_tcp_synack_retries; 214extern int sysctl_tcp_retries1; 215extern int sysctl_tcp_retries2; 216extern int sysctl_tcp_orphan_retries; 217extern int sysctl_tcp_syncookies; 218extern int sysctl_tcp_retrans_collapse; 219extern int sysctl_tcp_stdurg; 220extern int sysctl_tcp_rfc1337; 221extern int sysctl_tcp_abort_on_overflow; 222extern int sysctl_tcp_max_orphans; 223extern int sysctl_tcp_fack; 224extern int sysctl_tcp_reordering; 225extern int sysctl_tcp_ecn; 226extern int sysctl_tcp_dsack; 227extern int sysctl_tcp_mem[3]; 228extern int sysctl_tcp_wmem[3]; 229extern int sysctl_tcp_rmem[3]; 230extern int sysctl_tcp_app_win; 231extern int sysctl_tcp_adv_win_scale; 232extern int sysctl_tcp_tw_reuse; 233extern int sysctl_tcp_frto; 234extern int sysctl_tcp_frto_response; 235extern int sysctl_tcp_low_latency; 236extern int sysctl_tcp_dma_copybreak; 237extern int sysctl_tcp_nometrics_save; 238extern int sysctl_tcp_moderate_rcvbuf; 239extern int sysctl_tcp_tso_win_divisor; 240extern int sysctl_tcp_abc; 241extern int sysctl_tcp_mtu_probing; 242extern int sysctl_tcp_base_mss; 243extern int sysctl_tcp_workaround_signed_windows; 244extern int sysctl_tcp_slow_start_after_idle; 245extern int sysctl_tcp_max_ssthresh; 246extern int sysctl_tcp_cookie_size; 247extern int sysctl_tcp_thin_linear_timeouts; 248extern int sysctl_tcp_thin_dupack; 249 250extern atomic_t tcp_memory_allocated; 251extern struct percpu_counter tcp_sockets_allocated; 252extern int tcp_memory_pressure; 253 254/* 255 * The next routines deal with comparing 32 bit unsigned ints 256 * and worry about wraparound (automatic with unsigned arithmetic). 257 */ 258 259static inline int before(__u32 seq1, __u32 seq2) 260{ 261 return (__s32)(seq1-seq2) < 0; 262} 263#define after(seq2, seq1) before(seq1, seq2) 264 265/* is s2<=s1<=s3 ? */ 266static inline int between(__u32 seq1, __u32 seq2, __u32 seq3) 267{ 268 return seq3 - seq2 >= seq1 - seq2; 269} 270 271static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 272{ 273 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 274 int orphans = percpu_counter_read_positive(ocp); 275 276 if (orphans << shift > sysctl_tcp_max_orphans) { 277 orphans = percpu_counter_sum_positive(ocp); 278 if (orphans << shift > sysctl_tcp_max_orphans) 279 return true; 280 } 281 282 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 283 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]) 284 return true; 285 return false; 286} 287 288/* syncookies: remember time of last synqueue overflow */ 289static inline void tcp_synq_overflow(struct sock *sk) 290{ 291 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies; 292} 293 294/* syncookies: no recent synqueue overflow on this listening socket? */ 295static inline int tcp_synq_no_recent_overflow(const struct sock *sk) 296{ 297 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 298 return time_after(jiffies, last_overflow + TCP_TIMEOUT_INIT); 299} 300 301extern struct proto tcp_prot; 302 303#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 304#define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 305#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 306#define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 307#define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 308 309extern void tcp_v4_err(struct sk_buff *skb, u32); 310 311extern void tcp_shutdown (struct sock *sk, int how); 312 313extern int tcp_v4_rcv(struct sk_buff *skb); 314 315extern int tcp_v4_remember_stamp(struct sock *sk); 316extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 317extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 318 size_t size); 319extern int tcp_sendpage(struct sock *sk, struct page *page, int offset, 320 size_t size, int flags); 321extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 322extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 323 struct tcphdr *th, unsigned len); 324extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 325 struct tcphdr *th, unsigned len); 326extern void tcp_rcv_space_adjust(struct sock *sk); 327extern void tcp_cleanup_rbuf(struct sock *sk, int copied); 328extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 329extern void tcp_twsk_destructor(struct sock *sk); 330extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 331 struct pipe_inode_info *pipe, size_t len, 332 unsigned int flags); 333 334static inline void tcp_dec_quickack_mode(struct sock *sk, 335 const unsigned int pkts) 336{ 337 struct inet_connection_sock *icsk = inet_csk(sk); 338 339 if (icsk->icsk_ack.quick) { 340 if (pkts >= icsk->icsk_ack.quick) { 341 icsk->icsk_ack.quick = 0; 342 /* Leaving quickack mode we deflate ATO. */ 343 icsk->icsk_ack.ato = TCP_ATO_MIN; 344 } else 345 icsk->icsk_ack.quick -= pkts; 346 } 347} 348 349extern void tcp_enter_quickack_mode(struct sock *sk); 350 351#define TCP_ECN_OK 1 352#define TCP_ECN_QUEUE_CWR 2 353#define TCP_ECN_DEMAND_CWR 4 354 355static __inline__ void 356TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th) 357{ 358 if (sysctl_tcp_ecn && th->ece && th->cwr) 359 inet_rsk(req)->ecn_ok = 1; 360} 361 362enum tcp_tw_status { 363 TCP_TW_SUCCESS = 0, 364 TCP_TW_RST = 1, 365 TCP_TW_ACK = 2, 366 TCP_TW_SYN = 3 367}; 368 369 370extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 371 struct sk_buff *skb, 372 const struct tcphdr *th); 373extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb, 374 struct request_sock *req, 375 struct request_sock **prev); 376extern int tcp_child_process(struct sock *parent, struct sock *child, 377 struct sk_buff *skb); 378extern int tcp_use_frto(struct sock *sk); 379extern void tcp_enter_frto(struct sock *sk); 380extern void tcp_enter_loss(struct sock *sk, int how); 381extern void tcp_clear_retrans(struct tcp_sock *tp); 382extern void tcp_update_metrics(struct sock *sk); 383extern void tcp_close(struct sock *sk, long timeout); 384extern unsigned int tcp_poll(struct file * file, struct socket *sock, 385 struct poll_table_struct *wait); 386extern int tcp_getsockopt(struct sock *sk, int level, int optname, 387 char __user *optval, int __user *optlen); 388extern int tcp_setsockopt(struct sock *sk, int level, int optname, 389 char __user *optval, unsigned int optlen); 390extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 391 char __user *optval, int __user *optlen); 392extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 393 char __user *optval, unsigned int optlen); 394extern void tcp_set_keepalive(struct sock *sk, int val); 395extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req); 396extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 397 size_t len, int nonblock, int flags, int *addr_len); 398extern void tcp_parse_options(struct sk_buff *skb, 399 struct tcp_options_received *opt_rx, u8 **hvpp, 400 int estab); 401extern u8 *tcp_parse_md5sig_option(struct tcphdr *th); 402 403/* 404 * TCP v4 functions exported for the inet6 API 405 */ 406 407extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 408extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 409extern struct sock * tcp_create_openreq_child(struct sock *sk, 410 struct request_sock *req, 411 struct sk_buff *skb); 412extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 413 struct request_sock *req, 414 struct dst_entry *dst); 415extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 416extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, 417 int addr_len); 418extern int tcp_connect(struct sock *sk); 419extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst, 420 struct request_sock *req, 421 struct request_values *rvp); 422extern int tcp_disconnect(struct sock *sk, int flags); 423 424 425/* From syncookies.c */ 426extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS]; 427extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 428 struct ip_options *opt); 429extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, 430 __u16 *mss); 431 432extern __u32 cookie_init_timestamp(struct request_sock *req); 433extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *); 434 435/* From net/ipv6/syncookies.c */ 436extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 437extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb, 438 __u16 *mss); 439 440/* tcp_output.c */ 441 442extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 443 int nonagle); 444extern int tcp_may_send_now(struct sock *sk); 445extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); 446extern void tcp_retransmit_timer(struct sock *sk); 447extern void tcp_xmit_retransmit_queue(struct sock *); 448extern void tcp_simple_retransmit(struct sock *); 449extern int tcp_trim_head(struct sock *, struct sk_buff *, u32); 450extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 451 452extern void tcp_send_probe0(struct sock *); 453extern void tcp_send_partial(struct sock *); 454extern int tcp_write_wakeup(struct sock *); 455extern void tcp_send_fin(struct sock *sk); 456extern void tcp_send_active_reset(struct sock *sk, gfp_t priority); 457extern int tcp_send_synack(struct sock *); 458extern void tcp_push_one(struct sock *, unsigned int mss_now); 459extern void tcp_send_ack(struct sock *sk); 460extern void tcp_send_delayed_ack(struct sock *sk); 461 462/* tcp_input.c */ 463extern void tcp_cwnd_application_limited(struct sock *sk); 464 465/* tcp_timer.c */ 466extern void tcp_init_xmit_timers(struct sock *); 467static inline void tcp_clear_xmit_timers(struct sock *sk) 468{ 469 inet_csk_clear_xmit_timers(sk); 470} 471 472extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 473extern unsigned int tcp_current_mss(struct sock *sk); 474 475/* Bound MSS / TSO packet size with the half of the window */ 476static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 477{ 478 int cutoff; 479 480 /* When peer uses tiny windows, there is no use in packetizing 481 * to sub-MSS pieces for the sake of SWS or making sure there 482 * are enough packets in the pipe for fast recovery. 483 * 484 * On the other hand, for extremely large MSS devices, handling 485 * smaller than MSS windows in this way does make sense. 486 */ 487 if (tp->max_window >= 512) 488 cutoff = (tp->max_window >> 1); 489 else 490 cutoff = tp->max_window; 491 492 if (cutoff && pktsize > cutoff) 493 return max_t(int, cutoff, 68U - tp->tcp_header_len); 494 else 495 return pktsize; 496} 497 498/* tcp.c */ 499extern void tcp_get_info(struct sock *, struct tcp_info *); 500 501/* Read 'sendfile()'-style from a TCP socket */ 502typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 503 unsigned int, size_t); 504extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 505 sk_read_actor_t recv_actor); 506 507extern void tcp_initialize_rcv_mss(struct sock *sk); 508 509extern int tcp_mtu_to_mss(struct sock *sk, int pmtu); 510extern int tcp_mss_to_mtu(struct sock *sk, int mss); 511extern void tcp_mtup_init(struct sock *sk); 512 513static inline void tcp_bound_rto(const struct sock *sk) 514{ 515 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 516 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 517} 518 519static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 520{ 521 return (tp->srtt >> 3) + tp->rttvar; 522} 523 524static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 525{ 526 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 527 ntohl(TCP_FLAG_ACK) | 528 snd_wnd); 529} 530 531static inline void tcp_fast_path_on(struct tcp_sock *tp) 532{ 533 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 534} 535 536static inline void tcp_fast_path_check(struct sock *sk) 537{ 538 struct tcp_sock *tp = tcp_sk(sk); 539 540 if (skb_queue_empty(&tp->out_of_order_queue) && 541 tp->rcv_wnd && 542 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 543 !tp->urg_data) 544 tcp_fast_path_on(tp); 545} 546 547/* Compute the actual rto_min value */ 548static inline u32 tcp_rto_min(struct sock *sk) 549{ 550 struct dst_entry *dst = __sk_dst_get(sk); 551 u32 rto_min = TCP_RTO_MIN; 552 553 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 554 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 555 return rto_min; 556} 557 558/* Compute the actual receive window we are currently advertising. 559 * Rcv_nxt can be after the window if our peer push more data 560 * than the offered window. 561 */ 562static inline u32 tcp_receive_window(const struct tcp_sock *tp) 563{ 564 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 565 566 if (win < 0) 567 win = 0; 568 return (u32) win; 569} 570 571/* Choose a new window, without checks for shrinking, and without 572 * scaling applied to the result. The caller does these things 573 * if necessary. This is a "raw" window selection. 574 */ 575extern u32 __tcp_select_window(struct sock *sk); 576 577/* TCP timestamps are only 32-bits, this causes a slight 578 * complication on 64-bit systems since we store a snapshot 579 * of jiffies in the buffer control blocks below. We decided 580 * to use only the low 32-bits of jiffies and hide the ugly 581 * casts with the following macro. 582 */ 583#define tcp_time_stamp ((__u32)(jiffies)) 584 585#define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 586 587#define TCPHDR_FIN 0x01 588#define TCPHDR_SYN 0x02 589#define TCPHDR_RST 0x04 590#define TCPHDR_PSH 0x08 591#define TCPHDR_ACK 0x10 592#define TCPHDR_URG 0x20 593#define TCPHDR_ECE 0x40 594#define TCPHDR_CWR 0x80 595 596struct tcp_skb_cb { 597 union { 598 struct inet_skb_parm h4; 599#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 600 struct inet6_skb_parm h6; 601#endif 602 } header; /* For incoming frames */ 603 __u32 seq; /* Starting sequence number */ 604 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 605 __u32 when; /* used to compute rtt's */ 606 __u8 flags; /* TCP header flags. */ 607 __u8 sacked; /* State flags for SACK/FACK. */ 608#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 609#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 610#define TCPCB_LOST 0x04 /* SKB is lost */ 611#define TCPCB_TAGBITS 0x07 /* All tag bits */ 612 613#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 614#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 615 616 __u32 ack_seq; /* Sequence number ACK'd */ 617}; 618 619#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 620 621/* Due to TSO, an SKB can be composed of multiple actual 622 * packets. To keep these tracked properly, we use this. 623 */ 624static inline int tcp_skb_pcount(const struct sk_buff *skb) 625{ 626 return skb_shinfo(skb)->gso_segs; 627} 628 629/* This is valid iff tcp_skb_pcount() > 1. */ 630static inline int tcp_skb_mss(const struct sk_buff *skb) 631{ 632 return skb_shinfo(skb)->gso_size; 633} 634 635/* Events passed to congestion control interface */ 636enum tcp_ca_event { 637 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 638 CA_EVENT_CWND_RESTART, /* congestion window restart */ 639 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 640 CA_EVENT_FRTO, /* fast recovery timeout */ 641 CA_EVENT_LOSS, /* loss timeout */ 642 CA_EVENT_FAST_ACK, /* in sequence ack */ 643 CA_EVENT_SLOW_ACK, /* other ack */ 644}; 645 646/* 647 * Interface for adding new TCP congestion control handlers 648 */ 649#define TCP_CA_NAME_MAX 16 650#define TCP_CA_MAX 128 651#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 652 653#define TCP_CONG_NON_RESTRICTED 0x1 654#define TCP_CONG_RTT_STAMP 0x2 655 656struct tcp_congestion_ops { 657 struct list_head list; 658 unsigned long flags; 659 660 /* initialize private data (optional) */ 661 void (*init)(struct sock *sk); 662 /* cleanup private data (optional) */ 663 void (*release)(struct sock *sk); 664 665 /* return slow start threshold (required) */ 666 u32 (*ssthresh)(struct sock *sk); 667 /* lower bound for congestion window (optional) */ 668 u32 (*min_cwnd)(const struct sock *sk); 669 /* do new cwnd calculation (required) */ 670 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight); 671 /* call before changing ca_state (optional) */ 672 void (*set_state)(struct sock *sk, u8 new_state); 673 /* call when cwnd event occurs (optional) */ 674 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 675 /* new value of cwnd after loss (optional) */ 676 u32 (*undo_cwnd)(struct sock *sk); 677 /* hook for packet ack accounting (optional) */ 678 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 679 /* get info for inet_diag (optional) */ 680 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 681 682 char name[TCP_CA_NAME_MAX]; 683 struct module *owner; 684}; 685 686extern int tcp_register_congestion_control(struct tcp_congestion_ops *type); 687extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 688 689extern void tcp_init_congestion_control(struct sock *sk); 690extern void tcp_cleanup_congestion_control(struct sock *sk); 691extern int tcp_set_default_congestion_control(const char *name); 692extern void tcp_get_default_congestion_control(char *name); 693extern void tcp_get_available_congestion_control(char *buf, size_t len); 694extern void tcp_get_allowed_congestion_control(char *buf, size_t len); 695extern int tcp_set_allowed_congestion_control(char *allowed); 696extern int tcp_set_congestion_control(struct sock *sk, const char *name); 697extern void tcp_slow_start(struct tcp_sock *tp); 698extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 699 700extern struct tcp_congestion_ops tcp_init_congestion_ops; 701extern u32 tcp_reno_ssthresh(struct sock *sk); 702extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight); 703extern u32 tcp_reno_min_cwnd(const struct sock *sk); 704extern struct tcp_congestion_ops tcp_reno; 705 706static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 707{ 708 struct inet_connection_sock *icsk = inet_csk(sk); 709 710 if (icsk->icsk_ca_ops->set_state) 711 icsk->icsk_ca_ops->set_state(sk, ca_state); 712 icsk->icsk_ca_state = ca_state; 713} 714 715static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 716{ 717 const struct inet_connection_sock *icsk = inet_csk(sk); 718 719 if (icsk->icsk_ca_ops->cwnd_event) 720 icsk->icsk_ca_ops->cwnd_event(sk, event); 721} 722 723/* These functions determine how the current flow behaves in respect of SACK 724 * handling. SACK is negotiated with the peer, and therefore it can vary 725 * between different flows. 726 * 727 * tcp_is_sack - SACK enabled 728 * tcp_is_reno - No SACK 729 * tcp_is_fack - FACK enabled, implies SACK enabled 730 */ 731static inline int tcp_is_sack(const struct tcp_sock *tp) 732{ 733 return tp->rx_opt.sack_ok; 734} 735 736static inline int tcp_is_reno(const struct tcp_sock *tp) 737{ 738 return !tcp_is_sack(tp); 739} 740 741static inline int tcp_is_fack(const struct tcp_sock *tp) 742{ 743 return tp->rx_opt.sack_ok & 2; 744} 745 746static inline void tcp_enable_fack(struct tcp_sock *tp) 747{ 748 tp->rx_opt.sack_ok |= 2; 749} 750 751static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 752{ 753 return tp->sacked_out + tp->lost_out; 754} 755 756/* This determines how many packets are "in the network" to the best 757 * of our knowledge. In many cases it is conservative, but where 758 * detailed information is available from the receiver (via SACK 759 * blocks etc.) we can make more aggressive calculations. 760 * 761 * Use this for decisions involving congestion control, use just 762 * tp->packets_out to determine if the send queue is empty or not. 763 * 764 * Read this equation as: 765 * 766 * "Packets sent once on transmission queue" MINUS 767 * "Packets left network, but not honestly ACKed yet" PLUS 768 * "Packets fast retransmitted" 769 */ 770static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 771{ 772 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 773} 774 775#define TCP_INFINITE_SSTHRESH 0x7fffffff 776 777static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 778{ 779 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 780} 781 782/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 783 * The exception is rate halving phase, when cwnd is decreasing towards 784 * ssthresh. 785 */ 786static inline __u32 tcp_current_ssthresh(const struct sock *sk) 787{ 788 const struct tcp_sock *tp = tcp_sk(sk); 789 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery)) 790 return tp->snd_ssthresh; 791 else 792 return max(tp->snd_ssthresh, 793 ((tp->snd_cwnd >> 1) + 794 (tp->snd_cwnd >> 2))); 795} 796 797/* Use define here intentionally to get WARN_ON location shown at the caller */ 798#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 799 800extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh); 801extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst); 802 803/* Slow start with delack produces 3 packets of burst, so that 804 * it is safe "de facto". This will be the default - same as 805 * the default reordering threshold - but if reordering increases, 806 * we must be able to allow cwnd to burst at least this much in order 807 * to not pull it back when holes are filled. 808 */ 809static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 810{ 811 return tp->reordering; 812} 813 814/* Returns end sequence number of the receiver's advertised window */ 815static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 816{ 817 return tp->snd_una + tp->snd_wnd; 818} 819extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight); 820 821static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss, 822 const struct sk_buff *skb) 823{ 824 if (skb->len < mss) 825 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 826} 827 828static inline void tcp_check_probe_timer(struct sock *sk) 829{ 830 struct tcp_sock *tp = tcp_sk(sk); 831 const struct inet_connection_sock *icsk = inet_csk(sk); 832 833 if (!tp->packets_out && !icsk->icsk_pending) 834 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 835 icsk->icsk_rto, TCP_RTO_MAX); 836} 837 838static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 839{ 840 tp->snd_wl1 = seq; 841} 842 843static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 844{ 845 tp->snd_wl1 = seq; 846} 847 848/* 849 * Calculate(/check) TCP checksum 850 */ 851static inline __sum16 tcp_v4_check(int len, __be32 saddr, 852 __be32 daddr, __wsum base) 853{ 854 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 855} 856 857static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 858{ 859 return __skb_checksum_complete(skb); 860} 861 862static inline int tcp_checksum_complete(struct sk_buff *skb) 863{ 864 return !skb_csum_unnecessary(skb) && 865 __tcp_checksum_complete(skb); 866} 867 868/* Prequeue for VJ style copy to user, combined with checksumming. */ 869 870static inline void tcp_prequeue_init(struct tcp_sock *tp) 871{ 872 tp->ucopy.task = NULL; 873 tp->ucopy.len = 0; 874 tp->ucopy.memory = 0; 875 skb_queue_head_init(&tp->ucopy.prequeue); 876#ifdef CONFIG_NET_DMA 877 tp->ucopy.dma_chan = NULL; 878 tp->ucopy.wakeup = 0; 879 tp->ucopy.pinned_list = NULL; 880 tp->ucopy.dma_cookie = 0; 881#endif 882} 883 884/* Packet is added to VJ-style prequeue for processing in process 885 * context, if a reader task is waiting. Apparently, this exciting 886 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) 887 * failed somewhere. Latency? Burstiness? Well, at least now we will 888 * see, why it failed. 8)8) --ANK 889 * 890 * NOTE: is this not too big to inline? 891 */ 892static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb) 893{ 894 struct tcp_sock *tp = tcp_sk(sk); 895 896 if (sysctl_tcp_low_latency || !tp->ucopy.task) 897 return 0; 898 899 __skb_queue_tail(&tp->ucopy.prequeue, skb); 900 tp->ucopy.memory += skb->truesize; 901 if (tp->ucopy.memory > sk->sk_rcvbuf) { 902 struct sk_buff *skb1; 903 904 BUG_ON(sock_owned_by_user(sk)); 905 906 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { 907 sk_backlog_rcv(sk, skb1); 908 NET_INC_STATS_BH(sock_net(sk), 909 LINUX_MIB_TCPPREQUEUEDROPPED); 910 } 911 912 tp->ucopy.memory = 0; 913 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { 914 wake_up_interruptible_sync_poll(sk_sleep(sk), 915 POLLIN | POLLRDNORM | POLLRDBAND); 916 if (!inet_csk_ack_scheduled(sk)) 917 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 918 (3 * tcp_rto_min(sk)) / 4, 919 TCP_RTO_MAX); 920 } 921 return 1; 922} 923 924 925#undef STATE_TRACE 926 927#ifdef STATE_TRACE 928static const char *statename[]={ 929 "Unused","Established","Syn Sent","Syn Recv", 930 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 931 "Close Wait","Last ACK","Listen","Closing" 932}; 933#endif 934extern void tcp_set_state(struct sock *sk, int state); 935 936extern void tcp_done(struct sock *sk); 937 938static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 939{ 940 rx_opt->dsack = 0; 941 rx_opt->num_sacks = 0; 942} 943 944/* Determine a window scaling and initial window to offer. */ 945extern void tcp_select_initial_window(int __space, __u32 mss, 946 __u32 *rcv_wnd, __u32 *window_clamp, 947 int wscale_ok, __u8 *rcv_wscale, 948 __u32 init_rcv_wnd); 949 950static inline int tcp_win_from_space(int space) 951{ 952 return sysctl_tcp_adv_win_scale<=0 ? 953 (space>>(-sysctl_tcp_adv_win_scale)) : 954 space - (space>>sysctl_tcp_adv_win_scale); 955} 956 957/* Note: caller must be prepared to deal with negative returns */ 958static inline int tcp_space(const struct sock *sk) 959{ 960 return tcp_win_from_space(sk->sk_rcvbuf - 961 atomic_read(&sk->sk_rmem_alloc)); 962} 963 964static inline int tcp_full_space(const struct sock *sk) 965{ 966 return tcp_win_from_space(sk->sk_rcvbuf); 967} 968 969static inline void tcp_openreq_init(struct request_sock *req, 970 struct tcp_options_received *rx_opt, 971 struct sk_buff *skb) 972{ 973 struct inet_request_sock *ireq = inet_rsk(req); 974 975 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 976 req->cookie_ts = 0; 977 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 978 req->mss = rx_opt->mss_clamp; 979 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 980 ireq->tstamp_ok = rx_opt->tstamp_ok; 981 ireq->sack_ok = rx_opt->sack_ok; 982 ireq->snd_wscale = rx_opt->snd_wscale; 983 ireq->wscale_ok = rx_opt->wscale_ok; 984 ireq->acked = 0; 985 ireq->ecn_ok = 0; 986 ireq->rmt_port = tcp_hdr(skb)->source; 987 ireq->loc_port = tcp_hdr(skb)->dest; 988} 989 990extern void tcp_enter_memory_pressure(struct sock *sk); 991 992static inline int keepalive_intvl_when(const struct tcp_sock *tp) 993{ 994 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 995} 996 997static inline int keepalive_time_when(const struct tcp_sock *tp) 998{ 999 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1000} 1001 1002static inline int keepalive_probes(const struct tcp_sock *tp) 1003{ 1004 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1005} 1006 1007static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1008{ 1009 const struct inet_connection_sock *icsk = &tp->inet_conn; 1010 1011 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1012 tcp_time_stamp - tp->rcv_tstamp); 1013} 1014 1015static inline int tcp_fin_time(const struct sock *sk) 1016{ 1017 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1018 const int rto = inet_csk(sk)->icsk_rto; 1019 1020 if (fin_timeout < (rto << 2) - (rto >> 1)) 1021 fin_timeout = (rto << 2) - (rto >> 1); 1022 1023 return fin_timeout; 1024} 1025 1026static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, 1027 int paws_win) 1028{ 1029 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1030 return 1; 1031 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1032 return 1; 1033 1034 return 0; 1035} 1036 1037static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt, 1038 int rst) 1039{ 1040 if (tcp_paws_check(rx_opt, 0)) 1041 return 0; 1042 1043 /* RST segments are not recommended to carry timestamp, 1044 and, if they do, it is recommended to ignore PAWS because 1045 "their cleanup function should take precedence over timestamps." 1046 Certainly, it is mistake. It is necessary to understand the reasons 1047 of this constraint to relax it: if peer reboots, clock may go 1048 out-of-sync and half-open connections will not be reset. 1049 Actually, the problem would be not existing if all 1050 the implementations followed draft about maintaining clock 1051 via reboots. Linux-2.2 DOES NOT! 1052 1053 However, we can relax time bounds for RST segments to MSL. 1054 */ 1055 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1056 return 0; 1057 return 1; 1058} 1059 1060#define TCP_CHECK_TIMER(sk) do { } while (0) 1061 1062static inline void tcp_mib_init(struct net *net) 1063{ 1064 /* See RFC 2012 */ 1065 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1066 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1067 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1068 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1069} 1070 1071/* from STCP */ 1072static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1073{ 1074 tp->lost_skb_hint = NULL; 1075 tp->scoreboard_skb_hint = NULL; 1076} 1077 1078static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1079{ 1080 tcp_clear_retrans_hints_partial(tp); 1081 tp->retransmit_skb_hint = NULL; 1082} 1083 1084/* MD5 Signature */ 1085struct crypto_hash; 1086 1087/* - key database */ 1088struct tcp_md5sig_key { 1089 u8 *key; 1090 u8 keylen; 1091}; 1092 1093struct tcp4_md5sig_key { 1094 struct tcp_md5sig_key base; 1095 __be32 addr; 1096}; 1097 1098struct tcp6_md5sig_key { 1099 struct tcp_md5sig_key base; 1100 struct in6_addr addr; 1101}; 1102 1103/* - sock block */ 1104struct tcp_md5sig_info { 1105 struct tcp4_md5sig_key *keys4; 1106#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1107 struct tcp6_md5sig_key *keys6; 1108 u32 entries6; 1109 u32 alloced6; 1110#endif 1111 u32 entries4; 1112 u32 alloced4; 1113}; 1114 1115/* - pseudo header */ 1116struct tcp4_pseudohdr { 1117 __be32 saddr; 1118 __be32 daddr; 1119 __u8 pad; 1120 __u8 protocol; 1121 __be16 len; 1122}; 1123 1124struct tcp6_pseudohdr { 1125 struct in6_addr saddr; 1126 struct in6_addr daddr; 1127 __be32 len; 1128 __be32 protocol; /* including padding */ 1129}; 1130 1131union tcp_md5sum_block { 1132 struct tcp4_pseudohdr ip4; 1133#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 1134 struct tcp6_pseudohdr ip6; 1135#endif 1136}; 1137 1138/* - pool: digest algorithm, hash description and scratch buffer */ 1139struct tcp_md5sig_pool { 1140 struct hash_desc md5_desc; 1141 union tcp_md5sum_block md5_blk; 1142}; 1143 1144#define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */ 1145 1146/* - functions */ 1147extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, 1148 struct sock *sk, struct request_sock *req, 1149 struct sk_buff *skb); 1150extern struct tcp_md5sig_key * tcp_v4_md5_lookup(struct sock *sk, 1151 struct sock *addr_sk); 1152extern int tcp_v4_md5_do_add(struct sock *sk, __be32 addr, u8 *newkey, 1153 u8 newkeylen); 1154extern int tcp_v4_md5_do_del(struct sock *sk, __be32 addr); 1155 1156#ifdef CONFIG_TCP_MD5SIG 1157#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \ 1158 &(struct tcp_md5sig_key) { \ 1159 .key = (twsk)->tw_md5_key, \ 1160 .keylen = (twsk)->tw_md5_keylen, \ 1161 } : NULL) 1162#else 1163#define tcp_twsk_md5_key(twsk) NULL 1164#endif 1165 1166extern struct tcp_md5sig_pool * __percpu *tcp_alloc_md5sig_pool(struct sock *); 1167extern void tcp_free_md5sig_pool(void); 1168 1169extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1170extern void tcp_put_md5sig_pool(void); 1171 1172extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *); 1173extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *, 1174 unsigned header_len); 1175extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1176 struct tcp_md5sig_key *key); 1177 1178/* write queue abstraction */ 1179static inline void tcp_write_queue_purge(struct sock *sk) 1180{ 1181 struct sk_buff *skb; 1182 1183 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1184 sk_wmem_free_skb(sk, skb); 1185 sk_mem_reclaim(sk); 1186 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1187} 1188 1189static inline struct sk_buff *tcp_write_queue_head(struct sock *sk) 1190{ 1191 return skb_peek(&sk->sk_write_queue); 1192} 1193 1194static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk) 1195{ 1196 return skb_peek_tail(&sk->sk_write_queue); 1197} 1198 1199static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb) 1200{ 1201 return skb_queue_next(&sk->sk_write_queue, skb); 1202} 1203 1204static inline struct sk_buff *tcp_write_queue_prev(struct sock *sk, struct sk_buff *skb) 1205{ 1206 return skb_queue_prev(&sk->sk_write_queue, skb); 1207} 1208 1209#define tcp_for_write_queue(skb, sk) \ 1210 skb_queue_walk(&(sk)->sk_write_queue, skb) 1211 1212#define tcp_for_write_queue_from(skb, sk) \ 1213 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1214 1215#define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1216 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1217 1218static inline struct sk_buff *tcp_send_head(struct sock *sk) 1219{ 1220 return sk->sk_send_head; 1221} 1222 1223static inline bool tcp_skb_is_last(const struct sock *sk, 1224 const struct sk_buff *skb) 1225{ 1226 return skb_queue_is_last(&sk->sk_write_queue, skb); 1227} 1228 1229static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb) 1230{ 1231 if (tcp_skb_is_last(sk, skb)) 1232 sk->sk_send_head = NULL; 1233 else 1234 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1235} 1236 1237static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1238{ 1239 if (sk->sk_send_head == skb_unlinked) 1240 sk->sk_send_head = NULL; 1241} 1242 1243static inline void tcp_init_send_head(struct sock *sk) 1244{ 1245 sk->sk_send_head = NULL; 1246} 1247 1248static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1249{ 1250 __skb_queue_tail(&sk->sk_write_queue, skb); 1251} 1252 1253static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1254{ 1255 __tcp_add_write_queue_tail(sk, skb); 1256 1257 /* Queue it, remembering where we must start sending. */ 1258 if (sk->sk_send_head == NULL) { 1259 sk->sk_send_head = skb; 1260 1261 if (tcp_sk(sk)->highest_sack == NULL) 1262 tcp_sk(sk)->highest_sack = skb; 1263 } 1264} 1265 1266static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1267{ 1268 __skb_queue_head(&sk->sk_write_queue, skb); 1269} 1270 1271/* Insert buff after skb on the write queue of sk. */ 1272static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1273 struct sk_buff *buff, 1274 struct sock *sk) 1275{ 1276 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1277} 1278 1279/* Insert new before skb on the write queue of sk. */ 1280static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1281 struct sk_buff *skb, 1282 struct sock *sk) 1283{ 1284 __skb_queue_before(&sk->sk_write_queue, skb, new); 1285 1286 if (sk->sk_send_head == skb) 1287 sk->sk_send_head = new; 1288} 1289 1290static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1291{ 1292 __skb_unlink(skb, &sk->sk_write_queue); 1293} 1294 1295static inline int tcp_write_queue_empty(struct sock *sk) 1296{ 1297 return skb_queue_empty(&sk->sk_write_queue); 1298} 1299 1300static inline void tcp_push_pending_frames(struct sock *sk) 1301{ 1302 if (tcp_send_head(sk)) { 1303 struct tcp_sock *tp = tcp_sk(sk); 1304 1305 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1306 } 1307} 1308 1309/* Start sequence of the highest skb with SACKed bit, valid only if 1310 * sacked > 0 or when the caller has ensured validity by itself. 1311 */ 1312static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1313{ 1314 if (!tp->sacked_out) 1315 return tp->snd_una; 1316 1317 if (tp->highest_sack == NULL) 1318 return tp->snd_nxt; 1319 1320 return TCP_SKB_CB(tp->highest_sack)->seq; 1321} 1322 1323static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1324{ 1325 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1326 tcp_write_queue_next(sk, skb); 1327} 1328 1329static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1330{ 1331 return tcp_sk(sk)->highest_sack; 1332} 1333 1334static inline void tcp_highest_sack_reset(struct sock *sk) 1335{ 1336 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1337} 1338 1339/* Called when old skb is about to be deleted (to be combined with new skb) */ 1340static inline void tcp_highest_sack_combine(struct sock *sk, 1341 struct sk_buff *old, 1342 struct sk_buff *new) 1343{ 1344 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1345 tcp_sk(sk)->highest_sack = new; 1346} 1347 1348/* Determines whether this is a thin stream (which may suffer from 1349 * increased latency). Used to trigger latency-reducing mechanisms. 1350 */ 1351static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp) 1352{ 1353 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1354} 1355 1356/* /proc */ 1357enum tcp_seq_states { 1358 TCP_SEQ_STATE_LISTENING, 1359 TCP_SEQ_STATE_OPENREQ, 1360 TCP_SEQ_STATE_ESTABLISHED, 1361 TCP_SEQ_STATE_TIME_WAIT, 1362}; 1363 1364struct tcp_seq_afinfo { 1365 char *name; 1366 sa_family_t family; 1367 struct file_operations seq_fops; 1368 struct seq_operations seq_ops; 1369}; 1370 1371struct tcp_iter_state { 1372 struct seq_net_private p; 1373 sa_family_t family; 1374 enum tcp_seq_states state; 1375 struct sock *syn_wait_sk; 1376 int bucket, offset, sbucket, num, uid; 1377 loff_t last_pos; 1378}; 1379 1380extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1381extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1382 1383extern struct request_sock_ops tcp_request_sock_ops; 1384extern struct request_sock_ops tcp6_request_sock_ops; 1385 1386extern void tcp_v4_destroy_sock(struct sock *sk); 1387 1388extern int tcp_v4_gso_send_check(struct sk_buff *skb); 1389extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features); 1390extern struct sk_buff **tcp_gro_receive(struct sk_buff **head, 1391 struct sk_buff *skb); 1392extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head, 1393 struct sk_buff *skb); 1394extern int tcp_gro_complete(struct sk_buff *skb); 1395extern int tcp4_gro_complete(struct sk_buff *skb); 1396 1397#ifdef CONFIG_PROC_FS 1398extern int tcp4_proc_init(void); 1399extern void tcp4_proc_exit(void); 1400#endif 1401 1402/* TCP af-specific functions */ 1403struct tcp_sock_af_ops { 1404#ifdef CONFIG_TCP_MD5SIG 1405 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1406 struct sock *addr_sk); 1407 int (*calc_md5_hash) (char *location, 1408 struct tcp_md5sig_key *md5, 1409 struct sock *sk, 1410 struct request_sock *req, 1411 struct sk_buff *skb); 1412 int (*md5_add) (struct sock *sk, 1413 struct sock *addr_sk, 1414 u8 *newkey, 1415 u8 len); 1416 int (*md5_parse) (struct sock *sk, 1417 char __user *optval, 1418 int optlen); 1419#endif 1420}; 1421 1422struct tcp_request_sock_ops { 1423#ifdef CONFIG_TCP_MD5SIG 1424 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1425 struct request_sock *req); 1426 int (*calc_md5_hash) (char *location, 1427 struct tcp_md5sig_key *md5, 1428 struct sock *sk, 1429 struct request_sock *req, 1430 struct sk_buff *skb); 1431#endif 1432}; 1433 1434/* Using SHA1 for now, define some constants. 1435 */ 1436#define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS) 1437#define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4) 1438#define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS) 1439 1440extern int tcp_cookie_generator(u32 *bakery); 1441 1442/** 1443 * struct tcp_cookie_values - each socket needs extra space for the 1444 * cookies, together with (optional) space for any SYN data. 1445 * 1446 * A tcp_sock contains a pointer to the current value, and this is 1447 * cloned to the tcp_timewait_sock. 1448 * 1449 * @cookie_pair: variable data from the option exchange. 1450 * 1451 * @cookie_desired: user specified tcpct_cookie_desired. Zero 1452 * indicates default (sysctl_tcp_cookie_size). 1453 * After cookie sent, remembers size of cookie. 1454 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX. 1455 * 1456 * @s_data_desired: user specified tcpct_s_data_desired. When the 1457 * constant payload is specified (@s_data_constant), 1458 * holds its length instead. 1459 * Range 0 to TCP_MSS_DESIRED. 1460 * 1461 * @s_data_payload: constant data that is to be included in the 1462 * payload of SYN or SYNACK segments when the 1463 * cookie option is present. 1464 */ 1465struct tcp_cookie_values { 1466 struct kref kref; 1467 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE]; 1468 u8 cookie_pair_size; 1469 u8 cookie_desired; 1470 u16 s_data_desired:11, 1471 s_data_constant:1, 1472 s_data_in:1, 1473 s_data_out:1, 1474 s_data_unused:2; 1475 u8 s_data_payload[0]; 1476}; 1477 1478static inline void tcp_cookie_values_release(struct kref *kref) 1479{ 1480 kfree(container_of(kref, struct tcp_cookie_values, kref)); 1481} 1482 1483/* The length of constant payload data. Note that s_data_desired is 1484 * overloaded, depending on s_data_constant: either the length of constant 1485 * data (returned here) or the limit on variable data. 1486 */ 1487static inline int tcp_s_data_size(const struct tcp_sock *tp) 1488{ 1489 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant) 1490 ? tp->cookie_values->s_data_desired 1491 : 0; 1492} 1493 1494/** 1495 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace. 1496 * 1497 * As tcp_request_sock has already been extended in other places, the 1498 * only remaining method is to pass stack values along as function 1499 * parameters. These parameters are not needed after sending SYNACK. 1500 * 1501 * @cookie_bakery: cryptographic secret and message workspace. 1502 * 1503 * @cookie_plus: bytes in authenticator/cookie option, copied from 1504 * struct tcp_options_received (above). 1505 */ 1506struct tcp_extend_values { 1507 struct request_values rv; 1508 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS]; 1509 u8 cookie_plus:6, 1510 cookie_out_never:1, 1511 cookie_in_always:1; 1512}; 1513 1514static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp) 1515{ 1516 return (struct tcp_extend_values *)rvp; 1517} 1518 1519extern void tcp_v4_init(void); 1520extern void tcp_init(void); 1521 1522#endif /* _TCP_H */ 1523