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 FASTRETRANS_DEBUG 1 22 23#include <linux/list.h> 24#include <linux/tcp.h> 25#include <linux/bug.h> 26#include <linux/slab.h> 27#include <linux/cache.h> 28#include <linux/percpu.h> 29#include <linux/skbuff.h> 30#include <linux/crypto.h> 31#include <linux/cryptohash.h> 32#include <linux/kref.h> 33#include <linux/ktime.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#include <linux/memcontrol.h> 49 50extern struct inet_hashinfo tcp_hashinfo; 51 52extern struct percpu_counter tcp_orphan_count; 53void tcp_time_wait(struct sock *sk, int state, int timeo); 54 55#define MAX_TCP_HEADER (128 + MAX_HEADER) 56#define MAX_TCP_OPTION_SPACE 40 57 58/* 59 * Never offer a window over 32767 without using window scaling. Some 60 * poor stacks do signed 16bit maths! 61 */ 62#define MAX_TCP_WINDOW 32767U 63 64/* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 65#define TCP_MIN_MSS 88U 66 67/* The least MTU to use for probing */ 68#define TCP_BASE_MSS 1024 69 70/* probing interval, default to 10 minutes as per RFC4821 */ 71#define TCP_PROBE_INTERVAL 600 72 73/* Specify interval when tcp mtu probing will stop */ 74#define TCP_PROBE_THRESHOLD 8 75 76/* After receiving this amount of duplicate ACKs fast retransmit starts. */ 77#define TCP_FASTRETRANS_THRESH 3 78 79/* Maximal number of ACKs sent quickly to accelerate slow-start. */ 80#define TCP_MAX_QUICKACKS 16U 81 82/* urg_data states */ 83#define TCP_URG_VALID 0x0100 84#define TCP_URG_NOTYET 0x0200 85#define TCP_URG_READ 0x0400 86 87#define TCP_RETR1 3 /* 88 * This is how many retries it does before it 89 * tries to figure out if the gateway is 90 * down. Minimal RFC value is 3; it corresponds 91 * to ~3sec-8min depending on RTO. 92 */ 93 94#define TCP_RETR2 15 /* 95 * This should take at least 96 * 90 minutes to time out. 97 * RFC1122 says that the limit is 100 sec. 98 * 15 is ~13-30min depending on RTO. 99 */ 100 101#define TCP_SYN_RETRIES 6 /* This is how many retries are done 102 * when active opening a connection. 103 * RFC1122 says the minimum retry MUST 104 * be at least 180secs. Nevertheless 105 * this value is corresponding to 106 * 63secs of retransmission with the 107 * current initial RTO. 108 */ 109 110#define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 111 * when passive opening a connection. 112 * This is corresponding to 31secs of 113 * retransmission with the current 114 * initial RTO. 115 */ 116 117#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 118 * state, about 60 seconds */ 119#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 120 /* BSD style FIN_WAIT2 deadlock breaker. 121 * It used to be 3min, new value is 60sec, 122 * to combine FIN-WAIT-2 timeout with 123 * TIME-WAIT timer. 124 */ 125 126#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 127#if HZ >= 100 128#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 129#define TCP_ATO_MIN ((unsigned)(HZ/25)) 130#else 131#define TCP_DELACK_MIN 4U 132#define TCP_ATO_MIN 4U 133#endif 134#define TCP_RTO_MAX ((unsigned)(120*HZ)) 135#define TCP_RTO_MIN ((unsigned)(HZ/5)) 136#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 137#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 138 * used as a fallback RTO for the 139 * initial data transmission if no 140 * valid RTT sample has been acquired, 141 * most likely due to retrans in 3WHS. 142 */ 143 144#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 145 * for local resources. 146 */ 147 148#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 149#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 150#define TCP_KEEPALIVE_INTVL (75*HZ) 151 152#define MAX_TCP_KEEPIDLE 32767 153#define MAX_TCP_KEEPINTVL 32767 154#define MAX_TCP_KEEPCNT 127 155#define MAX_TCP_SYNCNT 127 156 157#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 158 159#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 160#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 161 * after this time. It should be equal 162 * (or greater than) TCP_TIMEWAIT_LEN 163 * to provide reliability equal to one 164 * provided by timewait state. 165 */ 166#define TCP_PAWS_WINDOW 1 /* Replay window for per-host 167 * timestamps. It must be less than 168 * minimal timewait lifetime. 169 */ 170/* 171 * TCP option 172 */ 173 174#define TCPOPT_NOP 1 /* Padding */ 175#define TCPOPT_EOL 0 /* End of options */ 176#define TCPOPT_MSS 2 /* Segment size negotiating */ 177#define TCPOPT_WINDOW 3 /* Window scaling */ 178#define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 179#define TCPOPT_SACK 5 /* SACK Block */ 180#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 181#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 182#define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 183#define TCPOPT_EXP 254 /* Experimental */ 184/* Magic number to be after the option value for sharing TCP 185 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 186 */ 187#define TCPOPT_FASTOPEN_MAGIC 0xF989 188 189/* 190 * TCP option lengths 191 */ 192 193#define TCPOLEN_MSS 4 194#define TCPOLEN_WINDOW 3 195#define TCPOLEN_SACK_PERM 2 196#define TCPOLEN_TIMESTAMP 10 197#define TCPOLEN_MD5SIG 18 198#define TCPOLEN_FASTOPEN_BASE 2 199#define TCPOLEN_EXP_FASTOPEN_BASE 4 200 201/* But this is what stacks really send out. */ 202#define TCPOLEN_TSTAMP_ALIGNED 12 203#define TCPOLEN_WSCALE_ALIGNED 4 204#define TCPOLEN_SACKPERM_ALIGNED 4 205#define TCPOLEN_SACK_BASE 2 206#define TCPOLEN_SACK_BASE_ALIGNED 4 207#define TCPOLEN_SACK_PERBLOCK 8 208#define TCPOLEN_MD5SIG_ALIGNED 20 209#define TCPOLEN_MSS_ALIGNED 4 210 211/* Flags in tp->nonagle */ 212#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 213#define TCP_NAGLE_CORK 2 /* Socket is corked */ 214#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 215 216/* TCP thin-stream limits */ 217#define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 218 219/* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */ 220#define TCP_INIT_CWND 10 221 222/* Bit Flags for sysctl_tcp_fastopen */ 223#define TFO_CLIENT_ENABLE 1 224#define TFO_SERVER_ENABLE 2 225#define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 226 227/* Accept SYN data w/o any cookie option */ 228#define TFO_SERVER_COOKIE_NOT_REQD 0x200 229 230/* Force enable TFO on all listeners, i.e., not requiring the 231 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen. 232 */ 233#define TFO_SERVER_WO_SOCKOPT1 0x400 234#define TFO_SERVER_WO_SOCKOPT2 0x800 235 236extern struct inet_timewait_death_row tcp_death_row; 237 238/* sysctl variables for tcp */ 239extern int sysctl_tcp_timestamps; 240extern int sysctl_tcp_window_scaling; 241extern int sysctl_tcp_sack; 242extern int sysctl_tcp_fin_timeout; 243extern int sysctl_tcp_keepalive_time; 244extern int sysctl_tcp_keepalive_probes; 245extern int sysctl_tcp_keepalive_intvl; 246extern int sysctl_tcp_syn_retries; 247extern int sysctl_tcp_synack_retries; 248extern int sysctl_tcp_retries1; 249extern int sysctl_tcp_retries2; 250extern int sysctl_tcp_orphan_retries; 251extern int sysctl_tcp_syncookies; 252extern int sysctl_tcp_fastopen; 253extern int sysctl_tcp_retrans_collapse; 254extern int sysctl_tcp_stdurg; 255extern int sysctl_tcp_rfc1337; 256extern int sysctl_tcp_abort_on_overflow; 257extern int sysctl_tcp_max_orphans; 258extern int sysctl_tcp_fack; 259extern int sysctl_tcp_reordering; 260extern int sysctl_tcp_max_reordering; 261extern int sysctl_tcp_dsack; 262extern long sysctl_tcp_mem[3]; 263extern int sysctl_tcp_wmem[3]; 264extern int sysctl_tcp_rmem[3]; 265extern int sysctl_tcp_app_win; 266extern int sysctl_tcp_adv_win_scale; 267extern int sysctl_tcp_tw_reuse; 268extern int sysctl_tcp_frto; 269extern int sysctl_tcp_low_latency; 270extern int sysctl_tcp_nometrics_save; 271extern int sysctl_tcp_moderate_rcvbuf; 272extern int sysctl_tcp_tso_win_divisor; 273extern int sysctl_tcp_workaround_signed_windows; 274extern int sysctl_tcp_slow_start_after_idle; 275extern int sysctl_tcp_thin_linear_timeouts; 276extern int sysctl_tcp_thin_dupack; 277extern int sysctl_tcp_early_retrans; 278extern int sysctl_tcp_limit_output_bytes; 279extern int sysctl_tcp_challenge_ack_limit; 280extern unsigned int sysctl_tcp_notsent_lowat; 281extern int sysctl_tcp_min_tso_segs; 282extern int sysctl_tcp_autocorking; 283extern int sysctl_tcp_invalid_ratelimit; 284 285extern atomic_long_t tcp_memory_allocated; 286extern struct percpu_counter tcp_sockets_allocated; 287extern int tcp_memory_pressure; 288 289/* 290 * The next routines deal with comparing 32 bit unsigned ints 291 * and worry about wraparound (automatic with unsigned arithmetic). 292 */ 293 294static inline bool before(__u32 seq1, __u32 seq2) 295{ 296 return (__s32)(seq1-seq2) < 0; 297} 298#define after(seq2, seq1) before(seq1, seq2) 299 300/* is s2<=s1<=s3 ? */ 301static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 302{ 303 return seq3 - seq2 >= seq1 - seq2; 304} 305 306static inline bool tcp_out_of_memory(struct sock *sk) 307{ 308 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 309 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 310 return true; 311 return false; 312} 313 314static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 315{ 316 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 317 int orphans = percpu_counter_read_positive(ocp); 318 319 if (orphans << shift > sysctl_tcp_max_orphans) { 320 orphans = percpu_counter_sum_positive(ocp); 321 if (orphans << shift > sysctl_tcp_max_orphans) 322 return true; 323 } 324 return false; 325} 326 327bool tcp_check_oom(struct sock *sk, int shift); 328 329/* syncookies: remember time of last synqueue overflow */ 330static inline void tcp_synq_overflow(struct sock *sk) 331{ 332 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies; 333} 334 335/* syncookies: no recent synqueue overflow on this listening socket? */ 336static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 337{ 338 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 339 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK); 340} 341 342extern struct proto tcp_prot; 343 344#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 345#define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 346#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 347#define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 348#define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 349 350void tcp_tasklet_init(void); 351 352void tcp_v4_err(struct sk_buff *skb, u32); 353 354void tcp_shutdown(struct sock *sk, int how); 355 356void tcp_v4_early_demux(struct sk_buff *skb); 357int tcp_v4_rcv(struct sk_buff *skb); 358 359int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 360int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 361int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 362 int flags); 363void tcp_release_cb(struct sock *sk); 364void tcp_wfree(struct sk_buff *skb); 365void tcp_write_timer_handler(struct sock *sk); 366void tcp_delack_timer_handler(struct sock *sk); 367int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 368int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 369 const struct tcphdr *th, unsigned int len); 370void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 371 const struct tcphdr *th, unsigned int len); 372void tcp_rcv_space_adjust(struct sock *sk); 373int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 374void tcp_twsk_destructor(struct sock *sk); 375ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 376 struct pipe_inode_info *pipe, size_t len, 377 unsigned int flags); 378 379static inline void tcp_dec_quickack_mode(struct sock *sk, 380 const unsigned int pkts) 381{ 382 struct inet_connection_sock *icsk = inet_csk(sk); 383 384 if (icsk->icsk_ack.quick) { 385 if (pkts >= icsk->icsk_ack.quick) { 386 icsk->icsk_ack.quick = 0; 387 /* Leaving quickack mode we deflate ATO. */ 388 icsk->icsk_ack.ato = TCP_ATO_MIN; 389 } else 390 icsk->icsk_ack.quick -= pkts; 391 } 392} 393 394#define TCP_ECN_OK 1 395#define TCP_ECN_QUEUE_CWR 2 396#define TCP_ECN_DEMAND_CWR 4 397#define TCP_ECN_SEEN 8 398 399enum tcp_tw_status { 400 TCP_TW_SUCCESS = 0, 401 TCP_TW_RST = 1, 402 TCP_TW_ACK = 2, 403 TCP_TW_SYN = 3 404}; 405 406 407enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 408 struct sk_buff *skb, 409 const struct tcphdr *th); 410struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 411 struct request_sock *req, bool fastopen); 412int tcp_child_process(struct sock *parent, struct sock *child, 413 struct sk_buff *skb); 414void tcp_enter_loss(struct sock *sk); 415void tcp_clear_retrans(struct tcp_sock *tp); 416void tcp_update_metrics(struct sock *sk); 417void tcp_init_metrics(struct sock *sk); 418void tcp_metrics_init(void); 419bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, 420 bool paws_check, bool timestamps); 421bool tcp_remember_stamp(struct sock *sk); 422bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw); 423void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst); 424void tcp_disable_fack(struct tcp_sock *tp); 425void tcp_close(struct sock *sk, long timeout); 426void tcp_init_sock(struct sock *sk); 427unsigned int tcp_poll(struct file *file, struct socket *sock, 428 struct poll_table_struct *wait); 429int tcp_getsockopt(struct sock *sk, int level, int optname, 430 char __user *optval, int __user *optlen); 431int tcp_setsockopt(struct sock *sk, int level, int optname, 432 char __user *optval, unsigned int optlen); 433int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 434 char __user *optval, int __user *optlen); 435int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 436 char __user *optval, unsigned int optlen); 437void tcp_set_keepalive(struct sock *sk, int val); 438void tcp_syn_ack_timeout(const struct request_sock *req); 439int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 440 int flags, int *addr_len); 441void tcp_parse_options(const struct sk_buff *skb, 442 struct tcp_options_received *opt_rx, 443 int estab, struct tcp_fastopen_cookie *foc); 444const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 445 446/* 447 * TCP v4 functions exported for the inet6 API 448 */ 449 450void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 451void tcp_v4_mtu_reduced(struct sock *sk); 452void tcp_req_err(struct sock *sk, u32 seq); 453int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 454struct sock *tcp_create_openreq_child(struct sock *sk, 455 struct request_sock *req, 456 struct sk_buff *skb); 457void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 458struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 459 struct request_sock *req, 460 struct dst_entry *dst); 461int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 462int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 463int tcp_connect(struct sock *sk); 464struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst, 465 struct request_sock *req, 466 struct tcp_fastopen_cookie *foc); 467int tcp_disconnect(struct sock *sk, int flags); 468 469void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 470int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 471void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 472 473/* From syncookies.c */ 474int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 475 u32 cookie); 476struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 477#ifdef CONFIG_SYN_COOKIES 478 479/* Syncookies use a monotonic timer which increments every 60 seconds. 480 * This counter is used both as a hash input and partially encoded into 481 * the cookie value. A cookie is only validated further if the delta 482 * between the current counter value and the encoded one is less than this, 483 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 484 * the counter advances immediately after a cookie is generated). 485 */ 486#define MAX_SYNCOOKIE_AGE 2 487 488static inline u32 tcp_cookie_time(void) 489{ 490 u64 val = get_jiffies_64(); 491 492 do_div(val, 60 * HZ); 493 return val; 494} 495 496u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 497 u16 *mssp); 498__u32 cookie_v4_init_sequence(struct sock *sk, const struct sk_buff *skb, 499 __u16 *mss); 500__u32 cookie_init_timestamp(struct request_sock *req); 501bool cookie_timestamp_decode(struct tcp_options_received *opt); 502bool cookie_ecn_ok(const struct tcp_options_received *opt, 503 const struct net *net, const struct dst_entry *dst); 504 505/* From net/ipv6/syncookies.c */ 506int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 507 u32 cookie); 508struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 509 510u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 511 const struct tcphdr *th, u16 *mssp); 512__u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb, 513 __u16 *mss); 514#endif 515/* tcp_output.c */ 516 517void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 518 int nonagle); 519bool tcp_may_send_now(struct sock *sk); 520int __tcp_retransmit_skb(struct sock *, struct sk_buff *); 521int tcp_retransmit_skb(struct sock *, struct sk_buff *); 522void tcp_retransmit_timer(struct sock *sk); 523void tcp_xmit_retransmit_queue(struct sock *); 524void tcp_simple_retransmit(struct sock *); 525int tcp_trim_head(struct sock *, struct sk_buff *, u32); 526int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t); 527 528void tcp_send_probe0(struct sock *); 529void tcp_send_partial(struct sock *); 530int tcp_write_wakeup(struct sock *); 531void tcp_send_fin(struct sock *sk); 532void tcp_send_active_reset(struct sock *sk, gfp_t priority); 533int tcp_send_synack(struct sock *); 534void tcp_push_one(struct sock *, unsigned int mss_now); 535void tcp_send_ack(struct sock *sk); 536void tcp_send_delayed_ack(struct sock *sk); 537void tcp_send_loss_probe(struct sock *sk); 538bool tcp_schedule_loss_probe(struct sock *sk); 539 540/* tcp_input.c */ 541void tcp_resume_early_retransmit(struct sock *sk); 542void tcp_rearm_rto(struct sock *sk); 543void tcp_reset(struct sock *sk); 544 545/* tcp_timer.c */ 546void tcp_init_xmit_timers(struct sock *); 547static inline void tcp_clear_xmit_timers(struct sock *sk) 548{ 549 inet_csk_clear_xmit_timers(sk); 550} 551 552unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 553unsigned int tcp_current_mss(struct sock *sk); 554 555/* Bound MSS / TSO packet size with the half of the window */ 556static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 557{ 558 int cutoff; 559 560 /* When peer uses tiny windows, there is no use in packetizing 561 * to sub-MSS pieces for the sake of SWS or making sure there 562 * are enough packets in the pipe for fast recovery. 563 * 564 * On the other hand, for extremely large MSS devices, handling 565 * smaller than MSS windows in this way does make sense. 566 */ 567 if (tp->max_window >= 512) 568 cutoff = (tp->max_window >> 1); 569 else 570 cutoff = tp->max_window; 571 572 if (cutoff && pktsize > cutoff) 573 return max_t(int, cutoff, 68U - tp->tcp_header_len); 574 else 575 return pktsize; 576} 577 578/* tcp.c */ 579void tcp_get_info(struct sock *, struct tcp_info *); 580 581/* Read 'sendfile()'-style from a TCP socket */ 582typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 583 unsigned int, size_t); 584int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 585 sk_read_actor_t recv_actor); 586 587void tcp_initialize_rcv_mss(struct sock *sk); 588 589int tcp_mtu_to_mss(struct sock *sk, int pmtu); 590int tcp_mss_to_mtu(struct sock *sk, int mss); 591void tcp_mtup_init(struct sock *sk); 592void tcp_init_buffer_space(struct sock *sk); 593 594static inline void tcp_bound_rto(const struct sock *sk) 595{ 596 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 597 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 598} 599 600static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 601{ 602 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 603} 604 605static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 606{ 607 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 608 ntohl(TCP_FLAG_ACK) | 609 snd_wnd); 610} 611 612static inline void tcp_fast_path_on(struct tcp_sock *tp) 613{ 614 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 615} 616 617static inline void tcp_fast_path_check(struct sock *sk) 618{ 619 struct tcp_sock *tp = tcp_sk(sk); 620 621 if (skb_queue_empty(&tp->out_of_order_queue) && 622 tp->rcv_wnd && 623 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 624 !tp->urg_data) 625 tcp_fast_path_on(tp); 626} 627 628/* Compute the actual rto_min value */ 629static inline u32 tcp_rto_min(struct sock *sk) 630{ 631 const struct dst_entry *dst = __sk_dst_get(sk); 632 u32 rto_min = TCP_RTO_MIN; 633 634 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 635 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 636 return rto_min; 637} 638 639static inline u32 tcp_rto_min_us(struct sock *sk) 640{ 641 return jiffies_to_usecs(tcp_rto_min(sk)); 642} 643 644static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 645{ 646 return dst_metric_locked(dst, RTAX_CC_ALGO); 647} 648 649/* Compute the actual receive window we are currently advertising. 650 * Rcv_nxt can be after the window if our peer push more data 651 * than the offered window. 652 */ 653static inline u32 tcp_receive_window(const struct tcp_sock *tp) 654{ 655 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 656 657 if (win < 0) 658 win = 0; 659 return (u32) win; 660} 661 662/* Choose a new window, without checks for shrinking, and without 663 * scaling applied to the result. The caller does these things 664 * if necessary. This is a "raw" window selection. 665 */ 666u32 __tcp_select_window(struct sock *sk); 667 668void tcp_send_window_probe(struct sock *sk); 669 670/* TCP timestamps are only 32-bits, this causes a slight 671 * complication on 64-bit systems since we store a snapshot 672 * of jiffies in the buffer control blocks below. We decided 673 * to use only the low 32-bits of jiffies and hide the ugly 674 * casts with the following macro. 675 */ 676#define tcp_time_stamp ((__u32)(jiffies)) 677 678static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 679{ 680 return skb->skb_mstamp.stamp_jiffies; 681} 682 683 684#define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 685 686#define TCPHDR_FIN 0x01 687#define TCPHDR_SYN 0x02 688#define TCPHDR_RST 0x04 689#define TCPHDR_PSH 0x08 690#define TCPHDR_ACK 0x10 691#define TCPHDR_URG 0x20 692#define TCPHDR_ECE 0x40 693#define TCPHDR_CWR 0x80 694 695/* This is what the send packet queuing engine uses to pass 696 * TCP per-packet control information to the transmission code. 697 * We also store the host-order sequence numbers in here too. 698 * This is 44 bytes if IPV6 is enabled. 699 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 700 */ 701struct tcp_skb_cb { 702 __u32 seq; /* Starting sequence number */ 703 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 704 union { 705 /* Note : tcp_tw_isn is used in input path only 706 * (isn chosen by tcp_timewait_state_process()) 707 * 708 * tcp_gso_segs is used in write queue only, 709 * cf tcp_skb_pcount() 710 */ 711 __u32 tcp_tw_isn; 712 __u32 tcp_gso_segs; 713 }; 714 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 715 716 __u8 sacked; /* State flags for SACK/FACK. */ 717#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 718#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 719#define TCPCB_LOST 0x04 /* SKB is lost */ 720#define TCPCB_TAGBITS 0x07 /* All tag bits */ 721#define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 722#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 723#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 724 TCPCB_REPAIRED) 725 726 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 727 /* 1 byte hole */ 728 __u32 ack_seq; /* Sequence number ACK'd */ 729 union { 730 struct inet_skb_parm h4; 731#if IS_ENABLED(CONFIG_IPV6) 732 struct inet6_skb_parm h6; 733#endif 734 } header; /* For incoming frames */ 735}; 736 737#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 738 739 740#if IS_ENABLED(CONFIG_IPV6) 741/* This is the variant of inet6_iif() that must be used by TCP, 742 * as TCP moves IP6CB into a different location in skb->cb[] 743 */ 744static inline int tcp_v6_iif(const struct sk_buff *skb) 745{ 746 return TCP_SKB_CB(skb)->header.h6.iif; 747} 748#endif 749 750/* Due to TSO, an SKB can be composed of multiple actual 751 * packets. To keep these tracked properly, we use this. 752 */ 753static inline int tcp_skb_pcount(const struct sk_buff *skb) 754{ 755 return TCP_SKB_CB(skb)->tcp_gso_segs; 756} 757 758static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 759{ 760 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 761} 762 763static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 764{ 765 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 766} 767 768/* This is valid iff tcp_skb_pcount() > 1. */ 769static inline int tcp_skb_mss(const struct sk_buff *skb) 770{ 771 return skb_shinfo(skb)->gso_size; 772} 773 774/* Events passed to congestion control interface */ 775enum tcp_ca_event { 776 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 777 CA_EVENT_CWND_RESTART, /* congestion window restart */ 778 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 779 CA_EVENT_LOSS, /* loss timeout */ 780 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 781 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 782 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 783 CA_EVENT_NON_DELAYED_ACK, 784}; 785 786/* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 787enum tcp_ca_ack_event_flags { 788 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 789 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 790 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 791}; 792 793/* 794 * Interface for adding new TCP congestion control handlers 795 */ 796#define TCP_CA_NAME_MAX 16 797#define TCP_CA_MAX 128 798#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 799 800#define TCP_CA_UNSPEC 0 801 802/* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 803#define TCP_CONG_NON_RESTRICTED 0x1 804/* Requires ECN/ECT set on all packets */ 805#define TCP_CONG_NEEDS_ECN 0x2 806 807union tcp_cc_info; 808 809struct tcp_congestion_ops { 810 struct list_head list; 811 u32 key; 812 u32 flags; 813 814 /* initialize private data (optional) */ 815 void (*init)(struct sock *sk); 816 /* cleanup private data (optional) */ 817 void (*release)(struct sock *sk); 818 819 /* return slow start threshold (required) */ 820 u32 (*ssthresh)(struct sock *sk); 821 /* do new cwnd calculation (required) */ 822 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 823 /* call before changing ca_state (optional) */ 824 void (*set_state)(struct sock *sk, u8 new_state); 825 /* call when cwnd event occurs (optional) */ 826 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 827 /* call when ack arrives (optional) */ 828 void (*in_ack_event)(struct sock *sk, u32 flags); 829 /* new value of cwnd after loss (optional) */ 830 u32 (*undo_cwnd)(struct sock *sk); 831 /* hook for packet ack accounting (optional) */ 832 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 833 /* get info for inet_diag (optional) */ 834 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 835 union tcp_cc_info *info); 836 837 char name[TCP_CA_NAME_MAX]; 838 struct module *owner; 839}; 840 841int tcp_register_congestion_control(struct tcp_congestion_ops *type); 842void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 843 844void tcp_assign_congestion_control(struct sock *sk); 845void tcp_init_congestion_control(struct sock *sk); 846void tcp_cleanup_congestion_control(struct sock *sk); 847int tcp_set_default_congestion_control(const char *name); 848void tcp_get_default_congestion_control(char *name); 849void tcp_get_available_congestion_control(char *buf, size_t len); 850void tcp_get_allowed_congestion_control(char *buf, size_t len); 851int tcp_set_allowed_congestion_control(char *allowed); 852int tcp_set_congestion_control(struct sock *sk, const char *name); 853u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 854void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 855 856u32 tcp_reno_ssthresh(struct sock *sk); 857void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 858extern struct tcp_congestion_ops tcp_reno; 859 860struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 861u32 tcp_ca_get_key_by_name(const char *name); 862#ifdef CONFIG_INET 863char *tcp_ca_get_name_by_key(u32 key, char *buffer); 864#else 865static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 866{ 867 return NULL; 868} 869#endif 870 871static inline bool tcp_ca_needs_ecn(const struct sock *sk) 872{ 873 const struct inet_connection_sock *icsk = inet_csk(sk); 874 875 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 876} 877 878static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 879{ 880 struct inet_connection_sock *icsk = inet_csk(sk); 881 882 if (icsk->icsk_ca_ops->set_state) 883 icsk->icsk_ca_ops->set_state(sk, ca_state); 884 icsk->icsk_ca_state = ca_state; 885} 886 887static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 888{ 889 const struct inet_connection_sock *icsk = inet_csk(sk); 890 891 if (icsk->icsk_ca_ops->cwnd_event) 892 icsk->icsk_ca_ops->cwnd_event(sk, event); 893} 894 895/* These functions determine how the current flow behaves in respect of SACK 896 * handling. SACK is negotiated with the peer, and therefore it can vary 897 * between different flows. 898 * 899 * tcp_is_sack - SACK enabled 900 * tcp_is_reno - No SACK 901 * tcp_is_fack - FACK enabled, implies SACK enabled 902 */ 903static inline int tcp_is_sack(const struct tcp_sock *tp) 904{ 905 return tp->rx_opt.sack_ok; 906} 907 908static inline bool tcp_is_reno(const struct tcp_sock *tp) 909{ 910 return !tcp_is_sack(tp); 911} 912 913static inline bool tcp_is_fack(const struct tcp_sock *tp) 914{ 915 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 916} 917 918static inline void tcp_enable_fack(struct tcp_sock *tp) 919{ 920 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 921} 922 923/* TCP early-retransmit (ER) is similar to but more conservative than 924 * the thin-dupack feature. Enable ER only if thin-dupack is disabled. 925 */ 926static inline void tcp_enable_early_retrans(struct tcp_sock *tp) 927{ 928 tp->do_early_retrans = sysctl_tcp_early_retrans && 929 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack && 930 sysctl_tcp_reordering == 3; 931} 932 933static inline void tcp_disable_early_retrans(struct tcp_sock *tp) 934{ 935 tp->do_early_retrans = 0; 936} 937 938static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 939{ 940 return tp->sacked_out + tp->lost_out; 941} 942 943/* This determines how many packets are "in the network" to the best 944 * of our knowledge. In many cases it is conservative, but where 945 * detailed information is available from the receiver (via SACK 946 * blocks etc.) we can make more aggressive calculations. 947 * 948 * Use this for decisions involving congestion control, use just 949 * tp->packets_out to determine if the send queue is empty or not. 950 * 951 * Read this equation as: 952 * 953 * "Packets sent once on transmission queue" MINUS 954 * "Packets left network, but not honestly ACKed yet" PLUS 955 * "Packets fast retransmitted" 956 */ 957static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 958{ 959 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 960} 961 962#define TCP_INFINITE_SSTHRESH 0x7fffffff 963 964static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 965{ 966 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 967} 968 969static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 970{ 971 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 972 (1 << inet_csk(sk)->icsk_ca_state); 973} 974 975/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 976 * The exception is cwnd reduction phase, when cwnd is decreasing towards 977 * ssthresh. 978 */ 979static inline __u32 tcp_current_ssthresh(const struct sock *sk) 980{ 981 const struct tcp_sock *tp = tcp_sk(sk); 982 983 if (tcp_in_cwnd_reduction(sk)) 984 return tp->snd_ssthresh; 985 else 986 return max(tp->snd_ssthresh, 987 ((tp->snd_cwnd >> 1) + 988 (tp->snd_cwnd >> 2))); 989} 990 991/* Use define here intentionally to get WARN_ON location shown at the caller */ 992#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 993 994void tcp_enter_cwr(struct sock *sk); 995__u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 996 997/* The maximum number of MSS of available cwnd for which TSO defers 998 * sending if not using sysctl_tcp_tso_win_divisor. 999 */ 1000static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1001{ 1002 return 3; 1003} 1004 1005/* Slow start with delack produces 3 packets of burst, so that 1006 * it is safe "de facto". This will be the default - same as 1007 * the default reordering threshold - but if reordering increases, 1008 * we must be able to allow cwnd to burst at least this much in order 1009 * to not pull it back when holes are filled. 1010 */ 1011static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 1012{ 1013 return tp->reordering; 1014} 1015 1016/* Returns end sequence number of the receiver's advertised window */ 1017static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1018{ 1019 return tp->snd_una + tp->snd_wnd; 1020} 1021 1022/* We follow the spirit of RFC2861 to validate cwnd but implement a more 1023 * flexible approach. The RFC suggests cwnd should not be raised unless 1024 * it was fully used previously. And that's exactly what we do in 1025 * congestion avoidance mode. But in slow start we allow cwnd to grow 1026 * as long as the application has used half the cwnd. 1027 * Example : 1028 * cwnd is 10 (IW10), but application sends 9 frames. 1029 * We allow cwnd to reach 18 when all frames are ACKed. 1030 * This check is safe because it's as aggressive as slow start which already 1031 * risks 100% overshoot. The advantage is that we discourage application to 1032 * either send more filler packets or data to artificially blow up the cwnd 1033 * usage, and allow application-limited process to probe bw more aggressively. 1034 */ 1035static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1036{ 1037 const struct tcp_sock *tp = tcp_sk(sk); 1038 1039 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1040 if (tp->snd_cwnd <= tp->snd_ssthresh) 1041 return tp->snd_cwnd < 2 * tp->max_packets_out; 1042 1043 return tp->is_cwnd_limited; 1044} 1045 1046static inline void tcp_check_probe_timer(struct sock *sk) 1047{ 1048 const struct tcp_sock *tp = tcp_sk(sk); 1049 const struct inet_connection_sock *icsk = inet_csk(sk); 1050 1051 if (!tp->packets_out && !icsk->icsk_pending) 1052 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1053 icsk->icsk_rto, TCP_RTO_MAX); 1054} 1055 1056static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1057{ 1058 tp->snd_wl1 = seq; 1059} 1060 1061static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1062{ 1063 tp->snd_wl1 = seq; 1064} 1065 1066/* 1067 * Calculate(/check) TCP checksum 1068 */ 1069static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1070 __be32 daddr, __wsum base) 1071{ 1072 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1073} 1074 1075static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1076{ 1077 return __skb_checksum_complete(skb); 1078} 1079 1080static inline bool tcp_checksum_complete(struct sk_buff *skb) 1081{ 1082 return !skb_csum_unnecessary(skb) && 1083 __tcp_checksum_complete(skb); 1084} 1085 1086/* Prequeue for VJ style copy to user, combined with checksumming. */ 1087 1088static inline void tcp_prequeue_init(struct tcp_sock *tp) 1089{ 1090 tp->ucopy.task = NULL; 1091 tp->ucopy.len = 0; 1092 tp->ucopy.memory = 0; 1093 skb_queue_head_init(&tp->ucopy.prequeue); 1094} 1095 1096bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); 1097 1098#undef STATE_TRACE 1099 1100#ifdef STATE_TRACE 1101static const char *statename[]={ 1102 "Unused","Established","Syn Sent","Syn Recv", 1103 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1104 "Close Wait","Last ACK","Listen","Closing" 1105}; 1106#endif 1107void tcp_set_state(struct sock *sk, int state); 1108 1109void tcp_done(struct sock *sk); 1110 1111static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1112{ 1113 rx_opt->dsack = 0; 1114 rx_opt->num_sacks = 0; 1115} 1116 1117u32 tcp_default_init_rwnd(u32 mss); 1118 1119/* Determine a window scaling and initial window to offer. */ 1120void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1121 __u32 *window_clamp, int wscale_ok, 1122 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1123 1124static inline int tcp_win_from_space(int space) 1125{ 1126 return sysctl_tcp_adv_win_scale<=0 ? 1127 (space>>(-sysctl_tcp_adv_win_scale)) : 1128 space - (space>>sysctl_tcp_adv_win_scale); 1129} 1130 1131/* Note: caller must be prepared to deal with negative returns */ 1132static inline int tcp_space(const struct sock *sk) 1133{ 1134 return tcp_win_from_space(sk->sk_rcvbuf - 1135 atomic_read(&sk->sk_rmem_alloc)); 1136} 1137 1138static inline int tcp_full_space(const struct sock *sk) 1139{ 1140 return tcp_win_from_space(sk->sk_rcvbuf); 1141} 1142 1143extern void tcp_openreq_init_rwin(struct request_sock *req, 1144 struct sock *sk, struct dst_entry *dst); 1145 1146void tcp_enter_memory_pressure(struct sock *sk); 1147 1148static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1149{ 1150 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1151} 1152 1153static inline int keepalive_time_when(const struct tcp_sock *tp) 1154{ 1155 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1156} 1157 1158static inline int keepalive_probes(const struct tcp_sock *tp) 1159{ 1160 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1161} 1162 1163static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1164{ 1165 const struct inet_connection_sock *icsk = &tp->inet_conn; 1166 1167 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1168 tcp_time_stamp - tp->rcv_tstamp); 1169} 1170 1171static inline int tcp_fin_time(const struct sock *sk) 1172{ 1173 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1174 const int rto = inet_csk(sk)->icsk_rto; 1175 1176 if (fin_timeout < (rto << 2) - (rto >> 1)) 1177 fin_timeout = (rto << 2) - (rto >> 1); 1178 1179 return fin_timeout; 1180} 1181 1182static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1183 int paws_win) 1184{ 1185 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1186 return true; 1187 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1188 return true; 1189 /* 1190 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1191 * then following tcp messages have valid values. Ignore 0 value, 1192 * or else 'negative' tsval might forbid us to accept their packets. 1193 */ 1194 if (!rx_opt->ts_recent) 1195 return true; 1196 return false; 1197} 1198 1199static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1200 int rst) 1201{ 1202 if (tcp_paws_check(rx_opt, 0)) 1203 return false; 1204 1205 /* RST segments are not recommended to carry timestamp, 1206 and, if they do, it is recommended to ignore PAWS because 1207 "their cleanup function should take precedence over timestamps." 1208 Certainly, it is mistake. It is necessary to understand the reasons 1209 of this constraint to relax it: if peer reboots, clock may go 1210 out-of-sync and half-open connections will not be reset. 1211 Actually, the problem would be not existing if all 1212 the implementations followed draft about maintaining clock 1213 via reboots. Linux-2.2 DOES NOT! 1214 1215 However, we can relax time bounds for RST segments to MSL. 1216 */ 1217 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1218 return false; 1219 return true; 1220} 1221 1222bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1223 int mib_idx, u32 *last_oow_ack_time); 1224 1225static inline void tcp_mib_init(struct net *net) 1226{ 1227 /* See RFC 2012 */ 1228 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1229 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1230 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1231 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1232} 1233 1234/* from STCP */ 1235static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1236{ 1237 tp->lost_skb_hint = NULL; 1238} 1239 1240static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1241{ 1242 tcp_clear_retrans_hints_partial(tp); 1243 tp->retransmit_skb_hint = NULL; 1244} 1245 1246/* MD5 Signature */ 1247struct crypto_hash; 1248 1249union tcp_md5_addr { 1250 struct in_addr a4; 1251#if IS_ENABLED(CONFIG_IPV6) 1252 struct in6_addr a6; 1253#endif 1254}; 1255 1256/* - key database */ 1257struct tcp_md5sig_key { 1258 struct hlist_node node; 1259 u8 keylen; 1260 u8 family; /* AF_INET or AF_INET6 */ 1261 union tcp_md5_addr addr; 1262 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1263 struct rcu_head rcu; 1264}; 1265 1266/* - sock block */ 1267struct tcp_md5sig_info { 1268 struct hlist_head head; 1269 struct rcu_head rcu; 1270}; 1271 1272/* - pseudo header */ 1273struct tcp4_pseudohdr { 1274 __be32 saddr; 1275 __be32 daddr; 1276 __u8 pad; 1277 __u8 protocol; 1278 __be16 len; 1279}; 1280 1281struct tcp6_pseudohdr { 1282 struct in6_addr saddr; 1283 struct in6_addr daddr; 1284 __be32 len; 1285 __be32 protocol; /* including padding */ 1286}; 1287 1288union tcp_md5sum_block { 1289 struct tcp4_pseudohdr ip4; 1290#if IS_ENABLED(CONFIG_IPV6) 1291 struct tcp6_pseudohdr ip6; 1292#endif 1293}; 1294 1295/* - pool: digest algorithm, hash description and scratch buffer */ 1296struct tcp_md5sig_pool { 1297 struct hash_desc md5_desc; 1298 union tcp_md5sum_block md5_blk; 1299}; 1300 1301/* - functions */ 1302int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1303 const struct sock *sk, const struct sk_buff *skb); 1304int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1305 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp); 1306int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1307 int family); 1308struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1309 const struct sock *addr_sk); 1310 1311#ifdef CONFIG_TCP_MD5SIG 1312struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1313 const union tcp_md5_addr *addr, 1314 int family); 1315#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1316#else 1317static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1318 const union tcp_md5_addr *addr, 1319 int family) 1320{ 1321 return NULL; 1322} 1323#define tcp_twsk_md5_key(twsk) NULL 1324#endif 1325 1326bool tcp_alloc_md5sig_pool(void); 1327 1328struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1329static inline void tcp_put_md5sig_pool(void) 1330{ 1331 local_bh_enable(); 1332} 1333 1334int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 1335int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1336 unsigned int header_len); 1337int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1338 const struct tcp_md5sig_key *key); 1339 1340/* From tcp_fastopen.c */ 1341void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1342 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1343 unsigned long *last_syn_loss); 1344void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1345 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1346 u16 try_exp); 1347struct tcp_fastopen_request { 1348 /* Fast Open cookie. Size 0 means a cookie request */ 1349 struct tcp_fastopen_cookie cookie; 1350 struct msghdr *data; /* data in MSG_FASTOPEN */ 1351 size_t size; 1352 int copied; /* queued in tcp_connect() */ 1353}; 1354void tcp_free_fastopen_req(struct tcp_sock *tp); 1355 1356extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1357int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1358bool tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1359 struct request_sock *req, 1360 struct tcp_fastopen_cookie *foc, 1361 struct dst_entry *dst); 1362void tcp_fastopen_init_key_once(bool publish); 1363#define TCP_FASTOPEN_KEY_LENGTH 16 1364 1365/* Fastopen key context */ 1366struct tcp_fastopen_context { 1367 struct crypto_cipher *tfm; 1368 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1369 struct rcu_head rcu; 1370}; 1371 1372/* write queue abstraction */ 1373static inline void tcp_write_queue_purge(struct sock *sk) 1374{ 1375 struct sk_buff *skb; 1376 1377 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1378 sk_wmem_free_skb(sk, skb); 1379 sk_mem_reclaim(sk); 1380 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1381} 1382 1383static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1384{ 1385 return skb_peek(&sk->sk_write_queue); 1386} 1387 1388static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1389{ 1390 return skb_peek_tail(&sk->sk_write_queue); 1391} 1392 1393static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1394 const struct sk_buff *skb) 1395{ 1396 return skb_queue_next(&sk->sk_write_queue, skb); 1397} 1398 1399static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1400 const struct sk_buff *skb) 1401{ 1402 return skb_queue_prev(&sk->sk_write_queue, skb); 1403} 1404 1405#define tcp_for_write_queue(skb, sk) \ 1406 skb_queue_walk(&(sk)->sk_write_queue, skb) 1407 1408#define tcp_for_write_queue_from(skb, sk) \ 1409 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1410 1411#define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1412 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1413 1414static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1415{ 1416 return sk->sk_send_head; 1417} 1418 1419static inline bool tcp_skb_is_last(const struct sock *sk, 1420 const struct sk_buff *skb) 1421{ 1422 return skb_queue_is_last(&sk->sk_write_queue, skb); 1423} 1424 1425static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1426{ 1427 if (tcp_skb_is_last(sk, skb)) 1428 sk->sk_send_head = NULL; 1429 else 1430 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1431} 1432 1433static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1434{ 1435 if (sk->sk_send_head == skb_unlinked) 1436 sk->sk_send_head = NULL; 1437} 1438 1439static inline void tcp_init_send_head(struct sock *sk) 1440{ 1441 sk->sk_send_head = NULL; 1442} 1443 1444static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1445{ 1446 __skb_queue_tail(&sk->sk_write_queue, skb); 1447} 1448 1449static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1450{ 1451 __tcp_add_write_queue_tail(sk, skb); 1452 1453 /* Queue it, remembering where we must start sending. */ 1454 if (sk->sk_send_head == NULL) { 1455 sk->sk_send_head = skb; 1456 1457 if (tcp_sk(sk)->highest_sack == NULL) 1458 tcp_sk(sk)->highest_sack = skb; 1459 } 1460} 1461 1462static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1463{ 1464 __skb_queue_head(&sk->sk_write_queue, skb); 1465} 1466 1467/* Insert buff after skb on the write queue of sk. */ 1468static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1469 struct sk_buff *buff, 1470 struct sock *sk) 1471{ 1472 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1473} 1474 1475/* Insert new before skb on the write queue of sk. */ 1476static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1477 struct sk_buff *skb, 1478 struct sock *sk) 1479{ 1480 __skb_queue_before(&sk->sk_write_queue, skb, new); 1481 1482 if (sk->sk_send_head == skb) 1483 sk->sk_send_head = new; 1484} 1485 1486static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1487{ 1488 __skb_unlink(skb, &sk->sk_write_queue); 1489} 1490 1491static inline bool tcp_write_queue_empty(struct sock *sk) 1492{ 1493 return skb_queue_empty(&sk->sk_write_queue); 1494} 1495 1496static inline void tcp_push_pending_frames(struct sock *sk) 1497{ 1498 if (tcp_send_head(sk)) { 1499 struct tcp_sock *tp = tcp_sk(sk); 1500 1501 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1502 } 1503} 1504 1505/* Start sequence of the skb just after the highest skb with SACKed 1506 * bit, valid only if sacked_out > 0 or when the caller has ensured 1507 * validity by itself. 1508 */ 1509static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1510{ 1511 if (!tp->sacked_out) 1512 return tp->snd_una; 1513 1514 if (tp->highest_sack == NULL) 1515 return tp->snd_nxt; 1516 1517 return TCP_SKB_CB(tp->highest_sack)->seq; 1518} 1519 1520static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1521{ 1522 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1523 tcp_write_queue_next(sk, skb); 1524} 1525 1526static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1527{ 1528 return tcp_sk(sk)->highest_sack; 1529} 1530 1531static inline void tcp_highest_sack_reset(struct sock *sk) 1532{ 1533 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1534} 1535 1536/* Called when old skb is about to be deleted (to be combined with new skb) */ 1537static inline void tcp_highest_sack_combine(struct sock *sk, 1538 struct sk_buff *old, 1539 struct sk_buff *new) 1540{ 1541 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1542 tcp_sk(sk)->highest_sack = new; 1543} 1544 1545/* Determines whether this is a thin stream (which may suffer from 1546 * increased latency). Used to trigger latency-reducing mechanisms. 1547 */ 1548static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1549{ 1550 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1551} 1552 1553/* /proc */ 1554enum tcp_seq_states { 1555 TCP_SEQ_STATE_LISTENING, 1556 TCP_SEQ_STATE_OPENREQ, 1557 TCP_SEQ_STATE_ESTABLISHED, 1558}; 1559 1560int tcp_seq_open(struct inode *inode, struct file *file); 1561 1562struct tcp_seq_afinfo { 1563 char *name; 1564 sa_family_t family; 1565 const struct file_operations *seq_fops; 1566 struct seq_operations seq_ops; 1567}; 1568 1569struct tcp_iter_state { 1570 struct seq_net_private p; 1571 sa_family_t family; 1572 enum tcp_seq_states state; 1573 struct sock *syn_wait_sk; 1574 int bucket, offset, sbucket, num; 1575 kuid_t uid; 1576 loff_t last_pos; 1577}; 1578 1579int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1580void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1581 1582extern struct request_sock_ops tcp_request_sock_ops; 1583extern struct request_sock_ops tcp6_request_sock_ops; 1584 1585void tcp_v4_destroy_sock(struct sock *sk); 1586 1587struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1588 netdev_features_t features); 1589struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1590int tcp_gro_complete(struct sk_buff *skb); 1591 1592void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1593 1594static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1595{ 1596 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat; 1597} 1598 1599static inline bool tcp_stream_memory_free(const struct sock *sk) 1600{ 1601 const struct tcp_sock *tp = tcp_sk(sk); 1602 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1603 1604 return notsent_bytes < tcp_notsent_lowat(tp); 1605} 1606 1607#ifdef CONFIG_PROC_FS 1608int tcp4_proc_init(void); 1609void tcp4_proc_exit(void); 1610#endif 1611 1612int tcp_rtx_synack(struct sock *sk, struct request_sock *req); 1613int tcp_conn_request(struct request_sock_ops *rsk_ops, 1614 const struct tcp_request_sock_ops *af_ops, 1615 struct sock *sk, struct sk_buff *skb); 1616 1617/* TCP af-specific functions */ 1618struct tcp_sock_af_ops { 1619#ifdef CONFIG_TCP_MD5SIG 1620 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1621 const struct sock *addr_sk); 1622 int (*calc_md5_hash)(char *location, 1623 const struct tcp_md5sig_key *md5, 1624 const struct sock *sk, 1625 const struct sk_buff *skb); 1626 int (*md5_parse)(struct sock *sk, 1627 char __user *optval, 1628 int optlen); 1629#endif 1630}; 1631 1632struct tcp_request_sock_ops { 1633 u16 mss_clamp; 1634#ifdef CONFIG_TCP_MD5SIG 1635 struct tcp_md5sig_key *(*req_md5_lookup)(struct sock *sk, 1636 const struct sock *addr_sk); 1637 int (*calc_md5_hash) (char *location, 1638 const struct tcp_md5sig_key *md5, 1639 const struct sock *sk, 1640 const struct sk_buff *skb); 1641#endif 1642 void (*init_req)(struct request_sock *req, struct sock *sk, 1643 struct sk_buff *skb); 1644#ifdef CONFIG_SYN_COOKIES 1645 __u32 (*cookie_init_seq)(struct sock *sk, const struct sk_buff *skb, 1646 __u16 *mss); 1647#endif 1648 struct dst_entry *(*route_req)(struct sock *sk, struct flowi *fl, 1649 const struct request_sock *req, 1650 bool *strict); 1651 __u32 (*init_seq)(const struct sk_buff *skb); 1652 int (*send_synack)(struct sock *sk, struct dst_entry *dst, 1653 struct flowi *fl, struct request_sock *req, 1654 u16 queue_mapping, struct tcp_fastopen_cookie *foc); 1655 void (*queue_hash_add)(struct sock *sk, struct request_sock *req, 1656 const unsigned long timeout); 1657}; 1658 1659#ifdef CONFIG_SYN_COOKIES 1660static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1661 struct sock *sk, struct sk_buff *skb, 1662 __u16 *mss) 1663{ 1664 return ops->cookie_init_seq(sk, skb, mss); 1665} 1666#else 1667static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1668 struct sock *sk, struct sk_buff *skb, 1669 __u16 *mss) 1670{ 1671 return 0; 1672} 1673#endif 1674 1675int tcpv4_offload_init(void); 1676 1677void tcp_v4_init(void); 1678void tcp_init(void); 1679 1680/* 1681 * Save and compile IPv4 options, return a pointer to it 1682 */ 1683static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb) 1684{ 1685 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1686 struct ip_options_rcu *dopt = NULL; 1687 1688 if (opt->optlen) { 1689 int opt_size = sizeof(*dopt) + opt->optlen; 1690 1691 dopt = kmalloc(opt_size, GFP_ATOMIC); 1692 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) { 1693 kfree(dopt); 1694 dopt = NULL; 1695 } 1696 } 1697 return dopt; 1698} 1699 1700/* locally generated TCP pure ACKs have skb->truesize == 2 1701 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1702 * This is much faster than dissecting the packet to find out. 1703 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1704 */ 1705static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1706{ 1707 return skb->truesize == 2; 1708} 1709 1710static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1711{ 1712 skb->truesize = 2; 1713} 1714 1715#endif /* _TCP_H */ 1716