linux/include/net/tcp.h
<<
>>
Prefs
   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/cryptohash.h>
  31#include <linux/kref.h>
  32#include <linux/ktime.h>
  33
  34#include <net/inet_connection_sock.h>
  35#include <net/inet_timewait_sock.h>
  36#include <net/inet_hashtables.h>
  37#include <net/checksum.h>
  38#include <net/request_sock.h>
  39#include <net/sock.h>
  40#include <net/snmp.h>
  41#include <net/ip.h>
  42#include <net/tcp_states.h>
  43#include <net/inet_ecn.h>
  44#include <net/dst.h>
  45
  46#include <linux/seq_file.h>
  47#include <linux/memcontrol.h>
  48
  49extern struct inet_hashinfo tcp_hashinfo;
  50
  51extern struct percpu_counter tcp_orphan_count;
  52void 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            1024
  68
  69/* probing interval, default to 10 minutes as per RFC4821 */
  70#define TCP_PROBE_INTERVAL      600
  71
  72/* Specify interval when tcp mtu probing will stop */
  73#define TCP_PROBE_THRESHOLD     8
  74
  75/* After receiving this amount of duplicate ACKs fast retransmit starts. */
  76#define TCP_FASTRETRANS_THRESH 3
  77
  78/* Maximal number of ACKs sent quickly to accelerate slow-start. */
  79#define TCP_MAX_QUICKACKS       16U
  80
  81/* urg_data states */
  82#define TCP_URG_VALID   0x0100
  83#define TCP_URG_NOTYET  0x0200
  84#define TCP_URG_READ    0x0400
  85
  86#define TCP_RETR1       3       /*
  87                                 * This is how many retries it does before it
  88                                 * tries to figure out if the gateway is
  89                                 * down. Minimal RFC value is 3; it corresponds
  90                                 * to ~3sec-8min depending on RTO.
  91                                 */
  92
  93#define TCP_RETR2       15      /*
  94                                 * This should take at least
  95                                 * 90 minutes to time out.
  96                                 * RFC1122 says that the limit is 100 sec.
  97                                 * 15 is ~13-30min depending on RTO.
  98                                 */
  99
 100#define TCP_SYN_RETRIES  6      /* This is how many retries are done
 101                                 * when active opening a connection.
 102                                 * RFC1122 says the minimum retry MUST
 103                                 * be at least 180secs.  Nevertheless
 104                                 * this value is corresponding to
 105                                 * 63secs of retransmission with the
 106                                 * current initial RTO.
 107                                 */
 108
 109#define TCP_SYNACK_RETRIES 5    /* This is how may retries are done
 110                                 * when passive opening a connection.
 111                                 * This is corresponding to 31secs of
 112                                 * retransmission with the current
 113                                 * initial RTO.
 114                                 */
 115
 116#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
 117                                  * state, about 60 seconds     */
 118#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
 119                                 /* BSD style FIN_WAIT2 deadlock breaker.
 120                                  * It used to be 3min, new value is 60sec,
 121                                  * to combine FIN-WAIT-2 timeout with
 122                                  * TIME-WAIT timer.
 123                                  */
 124
 125#define TCP_DELACK_MAX  ((unsigned)(HZ/5))      /* maximal time to delay before sending an ACK */
 126#if HZ >= 100
 127#define TCP_DELACK_MIN  ((unsigned)(HZ/25))     /* minimal time to delay before sending an ACK */
 128#define TCP_ATO_MIN     ((unsigned)(HZ/25))
 129#else
 130#define TCP_DELACK_MIN  4U
 131#define TCP_ATO_MIN     4U
 132#endif
 133#define TCP_RTO_MAX     ((unsigned)(120*HZ))
 134#define TCP_RTO_MIN     ((unsigned)(HZ/5))
 135#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ))     /* RFC6298 2.1 initial RTO value        */
 136#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
 137                                                 * used as a fallback RTO for the
 138                                                 * initial data transmission if no
 139                                                 * valid RTT sample has been acquired,
 140                                                 * most likely due to retrans in 3WHS.
 141                                                 */
 142
 143#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
 144                                                         * for local resources.
 145                                                         */
 146#define TCP_REO_TIMEOUT_MIN     (2000) /* Min RACK reordering timeout in usec */
 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 rfc6928 */
 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.
 232 */
 233#define TFO_SERVER_WO_SOCKOPT1  0x400
 234
 235
 236/* sysctl variables for tcp */
 237extern int sysctl_tcp_timestamps;
 238extern int sysctl_tcp_window_scaling;
 239extern int sysctl_tcp_sack;
 240extern int sysctl_tcp_fastopen;
 241extern int sysctl_tcp_retrans_collapse;
 242extern int sysctl_tcp_stdurg;
 243extern int sysctl_tcp_rfc1337;
 244extern int sysctl_tcp_abort_on_overflow;
 245extern int sysctl_tcp_max_orphans;
 246extern int sysctl_tcp_fack;
 247extern int sysctl_tcp_reordering;
 248extern int sysctl_tcp_max_reordering;
 249extern int sysctl_tcp_dsack;
 250extern long sysctl_tcp_mem[3];
 251extern int sysctl_tcp_wmem[3];
 252extern int sysctl_tcp_rmem[3];
 253extern int sysctl_tcp_app_win;
 254extern int sysctl_tcp_adv_win_scale;
 255extern int sysctl_tcp_frto;
 256extern int sysctl_tcp_low_latency;
 257extern int sysctl_tcp_nometrics_save;
 258extern int sysctl_tcp_moderate_rcvbuf;
 259extern int sysctl_tcp_tso_win_divisor;
 260extern int sysctl_tcp_workaround_signed_windows;
 261extern int sysctl_tcp_slow_start_after_idle;
 262extern int sysctl_tcp_thin_linear_timeouts;
 263extern int sysctl_tcp_thin_dupack;
 264extern int sysctl_tcp_early_retrans;
 265extern int sysctl_tcp_recovery;
 266#define TCP_RACK_LOSS_DETECTION  0x1 /* Use RACK to detect losses */
 267
 268extern int sysctl_tcp_limit_output_bytes;
 269extern int sysctl_tcp_challenge_ack_limit;
 270extern int sysctl_tcp_min_tso_segs;
 271extern int sysctl_tcp_min_rtt_wlen;
 272extern int sysctl_tcp_autocorking;
 273extern int sysctl_tcp_invalid_ratelimit;
 274extern int sysctl_tcp_pacing_ss_ratio;
 275extern int sysctl_tcp_pacing_ca_ratio;
 276
 277extern atomic_long_t tcp_memory_allocated;
 278extern struct percpu_counter tcp_sockets_allocated;
 279extern int tcp_memory_pressure;
 280
 281/* optimized version of sk_under_memory_pressure() for TCP sockets */
 282static inline bool tcp_under_memory_pressure(const struct sock *sk)
 283{
 284        if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
 285            mem_cgroup_under_socket_pressure(sk->sk_memcg))
 286                return true;
 287
 288        return tcp_memory_pressure;
 289}
 290/*
 291 * The next routines deal with comparing 32 bit unsigned ints
 292 * and worry about wraparound (automatic with unsigned arithmetic).
 293 */
 294
 295static inline bool before(__u32 seq1, __u32 seq2)
 296{
 297        return (__s32)(seq1-seq2) < 0;
 298}
 299#define after(seq2, seq1)       before(seq1, seq2)
 300
 301/* is s2<=s1<=s3 ? */
 302static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
 303{
 304        return seq3 - seq2 >= seq1 - seq2;
 305}
 306
 307static inline bool tcp_out_of_memory(struct sock *sk)
 308{
 309        if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
 310            sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
 311                return true;
 312        return false;
 313}
 314
 315void sk_forced_mem_schedule(struct sock *sk, int size);
 316
 317static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
 318{
 319        struct percpu_counter *ocp = sk->sk_prot->orphan_count;
 320        int orphans = percpu_counter_read_positive(ocp);
 321
 322        if (orphans << shift > sysctl_tcp_max_orphans) {
 323                orphans = percpu_counter_sum_positive(ocp);
 324                if (orphans << shift > sysctl_tcp_max_orphans)
 325                        return true;
 326        }
 327        return false;
 328}
 329
 330bool tcp_check_oom(struct sock *sk, int shift);
 331
 332
 333extern struct proto tcp_prot;
 334
 335#define TCP_INC_STATS(net, field)       SNMP_INC_STATS((net)->mib.tcp_statistics, field)
 336#define __TCP_INC_STATS(net, field)     __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
 337#define TCP_DEC_STATS(net, field)       SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
 338#define TCP_ADD_STATS(net, field, val)  SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
 339
 340void tcp_tasklet_init(void);
 341
 342void tcp_v4_err(struct sk_buff *skb, u32);
 343
 344void tcp_shutdown(struct sock *sk, int how);
 345
 346void tcp_v4_early_demux(struct sk_buff *skb);
 347int tcp_v4_rcv(struct sk_buff *skb);
 348
 349int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
 350int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
 351int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
 352                 int flags);
 353void tcp_release_cb(struct sock *sk);
 354void tcp_wfree(struct sk_buff *skb);
 355void tcp_write_timer_handler(struct sock *sk);
 356void tcp_delack_timer_handler(struct sock *sk);
 357int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
 358int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
 359void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
 360                         const struct tcphdr *th, unsigned int len);
 361void tcp_rcv_space_adjust(struct sock *sk);
 362int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
 363void tcp_twsk_destructor(struct sock *sk);
 364ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
 365                        struct pipe_inode_info *pipe, size_t len,
 366                        unsigned int flags);
 367
 368static inline void tcp_dec_quickack_mode(struct sock *sk,
 369                                         const unsigned int pkts)
 370{
 371        struct inet_connection_sock *icsk = inet_csk(sk);
 372
 373        if (icsk->icsk_ack.quick) {
 374                if (pkts >= icsk->icsk_ack.quick) {
 375                        icsk->icsk_ack.quick = 0;
 376                        /* Leaving quickack mode we deflate ATO. */
 377                        icsk->icsk_ack.ato   = TCP_ATO_MIN;
 378                } else
 379                        icsk->icsk_ack.quick -= pkts;
 380        }
 381}
 382
 383#define TCP_ECN_OK              1
 384#define TCP_ECN_QUEUE_CWR       2
 385#define TCP_ECN_DEMAND_CWR      4
 386#define TCP_ECN_SEEN            8
 387
 388enum tcp_tw_status {
 389        TCP_TW_SUCCESS = 0,
 390        TCP_TW_RST = 1,
 391        TCP_TW_ACK = 2,
 392        TCP_TW_SYN = 3
 393};
 394
 395
 396enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
 397                                              struct sk_buff *skb,
 398                                              const struct tcphdr *th);
 399struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
 400                           struct request_sock *req, bool fastopen);
 401int tcp_child_process(struct sock *parent, struct sock *child,
 402                      struct sk_buff *skb);
 403void tcp_enter_loss(struct sock *sk);
 404void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag);
 405void tcp_clear_retrans(struct tcp_sock *tp);
 406void tcp_update_metrics(struct sock *sk);
 407void tcp_init_metrics(struct sock *sk);
 408void tcp_metrics_init(void);
 409bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
 410                        bool paws_check, bool timestamps);
 411bool tcp_remember_stamp(struct sock *sk);
 412bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
 413void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
 414void tcp_disable_fack(struct tcp_sock *tp);
 415void tcp_close(struct sock *sk, long timeout);
 416void tcp_init_sock(struct sock *sk);
 417unsigned int tcp_poll(struct file *file, struct socket *sock,
 418                      struct poll_table_struct *wait);
 419int tcp_getsockopt(struct sock *sk, int level, int optname,
 420                   char __user *optval, int __user *optlen);
 421int tcp_setsockopt(struct sock *sk, int level, int optname,
 422                   char __user *optval, unsigned int optlen);
 423int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
 424                          char __user *optval, int __user *optlen);
 425int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
 426                          char __user *optval, unsigned int optlen);
 427void tcp_set_keepalive(struct sock *sk, int val);
 428void tcp_syn_ack_timeout(const struct request_sock *req);
 429int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
 430                int flags, int *addr_len);
 431void tcp_parse_options(const struct sk_buff *skb,
 432                       struct tcp_options_received *opt_rx,
 433                       int estab, struct tcp_fastopen_cookie *foc);
 434const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
 435
 436/*
 437 *      TCP v4 functions exported for the inet6 API
 438 */
 439
 440void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
 441void tcp_v4_mtu_reduced(struct sock *sk);
 442void tcp_req_err(struct sock *sk, u32 seq, bool abort);
 443int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
 444struct sock *tcp_create_openreq_child(const struct sock *sk,
 445                                      struct request_sock *req,
 446                                      struct sk_buff *skb);
 447void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
 448struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
 449                                  struct request_sock *req,
 450                                  struct dst_entry *dst,
 451                                  struct request_sock *req_unhash,
 452                                  bool *own_req);
 453int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
 454int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
 455int tcp_connect(struct sock *sk);
 456enum tcp_synack_type {
 457        TCP_SYNACK_NORMAL,
 458        TCP_SYNACK_FASTOPEN,
 459        TCP_SYNACK_COOKIE,
 460};
 461struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
 462                                struct request_sock *req,
 463                                struct tcp_fastopen_cookie *foc,
 464                                enum tcp_synack_type synack_type);
 465int tcp_disconnect(struct sock *sk, int flags);
 466
 467void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
 468int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
 469void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
 470
 471/* From syncookies.c */
 472struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
 473                                 struct request_sock *req,
 474                                 struct dst_entry *dst);
 475int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
 476                      u32 cookie);
 477struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
 478#ifdef CONFIG_SYN_COOKIES
 479
 480/* Syncookies use a monotonic timer which increments every 60 seconds.
 481 * This counter is used both as a hash input and partially encoded into
 482 * the cookie value.  A cookie is only validated further if the delta
 483 * between the current counter value and the encoded one is less than this,
 484 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
 485 * the counter advances immediately after a cookie is generated).
 486 */
 487#define MAX_SYNCOOKIE_AGE       2
 488#define TCP_SYNCOOKIE_PERIOD    (60 * HZ)
 489#define TCP_SYNCOOKIE_VALID     (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
 490
 491/* syncookies: remember time of last synqueue overflow
 492 * But do not dirty this field too often (once per second is enough)
 493 * It is racy as we do not hold a lock, but race is very minor.
 494 */
 495static inline void tcp_synq_overflow(const struct sock *sk)
 496{
 497        unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
 498        unsigned long now = jiffies;
 499
 500        if (time_after(now, last_overflow + HZ))
 501                tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
 502}
 503
 504/* syncookies: no recent synqueue overflow on this listening socket? */
 505static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
 506{
 507        unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
 508
 509        return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
 510}
 511
 512static inline u32 tcp_cookie_time(void)
 513{
 514        u64 val = get_jiffies_64();
 515
 516        do_div(val, TCP_SYNCOOKIE_PERIOD);
 517        return val;
 518}
 519
 520u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
 521                              u16 *mssp);
 522__u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
 523__u32 cookie_init_timestamp(struct request_sock *req);
 524bool cookie_timestamp_decode(struct tcp_options_received *opt);
 525bool cookie_ecn_ok(const struct tcp_options_received *opt,
 526                   const struct net *net, const struct dst_entry *dst);
 527
 528/* From net/ipv6/syncookies.c */
 529int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
 530                      u32 cookie);
 531struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
 532
 533u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
 534                              const struct tcphdr *th, u16 *mssp);
 535__u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
 536#endif
 537/* tcp_output.c */
 538
 539u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
 540                     int min_tso_segs);
 541void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
 542                               int nonagle);
 543bool tcp_may_send_now(struct sock *sk);
 544int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
 545int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
 546void tcp_retransmit_timer(struct sock *sk);
 547void tcp_xmit_retransmit_queue(struct sock *);
 548void tcp_simple_retransmit(struct sock *);
 549void tcp_enter_recovery(struct sock *sk, bool ece_ack);
 550int tcp_trim_head(struct sock *, struct sk_buff *, u32);
 551int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
 552
 553void tcp_send_probe0(struct sock *);
 554void tcp_send_partial(struct sock *);
 555int tcp_write_wakeup(struct sock *, int mib);
 556void tcp_send_fin(struct sock *sk);
 557void tcp_send_active_reset(struct sock *sk, gfp_t priority);
 558int tcp_send_synack(struct sock *);
 559void tcp_push_one(struct sock *, unsigned int mss_now);
 560void tcp_send_ack(struct sock *sk);
 561void tcp_send_delayed_ack(struct sock *sk);
 562void tcp_send_loss_probe(struct sock *sk);
 563bool tcp_schedule_loss_probe(struct sock *sk);
 564void tcp_skb_collapse_tstamp(struct sk_buff *skb,
 565                             const struct sk_buff *next_skb);
 566
 567/* tcp_input.c */
 568void tcp_rearm_rto(struct sock *sk);
 569void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
 570void tcp_reset(struct sock *sk);
 571void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
 572void tcp_fin(struct sock *sk);
 573
 574/* tcp_timer.c */
 575void tcp_init_xmit_timers(struct sock *);
 576static inline void tcp_clear_xmit_timers(struct sock *sk)
 577{
 578        inet_csk_clear_xmit_timers(sk);
 579}
 580
 581unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
 582unsigned int tcp_current_mss(struct sock *sk);
 583
 584/* Bound MSS / TSO packet size with the half of the window */
 585static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
 586{
 587        int cutoff;
 588
 589        /* When peer uses tiny windows, there is no use in packetizing
 590         * to sub-MSS pieces for the sake of SWS or making sure there
 591         * are enough packets in the pipe for fast recovery.
 592         *
 593         * On the other hand, for extremely large MSS devices, handling
 594         * smaller than MSS windows in this way does make sense.
 595         */
 596        if (tp->max_window > TCP_MSS_DEFAULT)
 597                cutoff = (tp->max_window >> 1);
 598        else
 599                cutoff = tp->max_window;
 600
 601        if (cutoff && pktsize > cutoff)
 602                return max_t(int, cutoff, 68U - tp->tcp_header_len);
 603        else
 604                return pktsize;
 605}
 606
 607/* tcp.c */
 608void tcp_get_info(struct sock *, struct tcp_info *);
 609
 610/* Read 'sendfile()'-style from a TCP socket */
 611int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
 612                  sk_read_actor_t recv_actor);
 613
 614void tcp_initialize_rcv_mss(struct sock *sk);
 615
 616int tcp_mtu_to_mss(struct sock *sk, int pmtu);
 617int tcp_mss_to_mtu(struct sock *sk, int mss);
 618void tcp_mtup_init(struct sock *sk);
 619void tcp_init_buffer_space(struct sock *sk);
 620
 621static inline void tcp_bound_rto(const struct sock *sk)
 622{
 623        if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
 624                inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
 625}
 626
 627static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
 628{
 629        return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
 630}
 631
 632static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
 633{
 634        tp->pred_flags = htonl((tp->tcp_header_len << 26) |
 635                               ntohl(TCP_FLAG_ACK) |
 636                               snd_wnd);
 637}
 638
 639static inline void tcp_fast_path_on(struct tcp_sock *tp)
 640{
 641        __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
 642}
 643
 644static inline void tcp_fast_path_check(struct sock *sk)
 645{
 646        struct tcp_sock *tp = tcp_sk(sk);
 647
 648        if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
 649            tp->rcv_wnd &&
 650            atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
 651            !tp->urg_data)
 652                tcp_fast_path_on(tp);
 653}
 654
 655/* Compute the actual rto_min value */
 656static inline u32 tcp_rto_min(struct sock *sk)
 657{
 658        const struct dst_entry *dst = __sk_dst_get(sk);
 659        u32 rto_min = TCP_RTO_MIN;
 660
 661        if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
 662                rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
 663        return rto_min;
 664}
 665
 666static inline u32 tcp_rto_min_us(struct sock *sk)
 667{
 668        return jiffies_to_usecs(tcp_rto_min(sk));
 669}
 670
 671static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
 672{
 673        return dst_metric_locked(dst, RTAX_CC_ALGO);
 674}
 675
 676/* Minimum RTT in usec. ~0 means not available. */
 677static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
 678{
 679        return minmax_get(&tp->rtt_min);
 680}
 681
 682/* Compute the actual receive window we are currently advertising.
 683 * Rcv_nxt can be after the window if our peer push more data
 684 * than the offered window.
 685 */
 686static inline u32 tcp_receive_window(const struct tcp_sock *tp)
 687{
 688        s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
 689
 690        if (win < 0)
 691                win = 0;
 692        return (u32) win;
 693}
 694
 695/* Choose a new window, without checks for shrinking, and without
 696 * scaling applied to the result.  The caller does these things
 697 * if necessary.  This is a "raw" window selection.
 698 */
 699u32 __tcp_select_window(struct sock *sk);
 700
 701void tcp_send_window_probe(struct sock *sk);
 702
 703/* TCP timestamps are only 32-bits, this causes a slight
 704 * complication on 64-bit systems since we store a snapshot
 705 * of jiffies in the buffer control blocks below.  We decided
 706 * to use only the low 32-bits of jiffies and hide the ugly
 707 * casts with the following macro.
 708 */
 709#define tcp_time_stamp          ((__u32)(jiffies))
 710
 711static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
 712{
 713        return skb->skb_mstamp.stamp_jiffies;
 714}
 715
 716
 717#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
 718
 719#define TCPHDR_FIN 0x01
 720#define TCPHDR_SYN 0x02
 721#define TCPHDR_RST 0x04
 722#define TCPHDR_PSH 0x08
 723#define TCPHDR_ACK 0x10
 724#define TCPHDR_URG 0x20
 725#define TCPHDR_ECE 0x40
 726#define TCPHDR_CWR 0x80
 727
 728#define TCPHDR_SYN_ECN  (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
 729
 730/* This is what the send packet queuing engine uses to pass
 731 * TCP per-packet control information to the transmission code.
 732 * We also store the host-order sequence numbers in here too.
 733 * This is 44 bytes if IPV6 is enabled.
 734 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
 735 */
 736struct tcp_skb_cb {
 737        __u32           seq;            /* Starting sequence number     */
 738        __u32           end_seq;        /* SEQ + FIN + SYN + datalen    */
 739        union {
 740                /* Note : tcp_tw_isn is used in input path only
 741                 *        (isn chosen by tcp_timewait_state_process())
 742                 *
 743                 *        tcp_gso_segs/size are used in write queue only,
 744                 *        cf tcp_skb_pcount()/tcp_skb_mss()
 745                 */
 746                __u32           tcp_tw_isn;
 747                struct {
 748                        u16     tcp_gso_segs;
 749                        u16     tcp_gso_size;
 750                };
 751        };
 752        __u8            tcp_flags;      /* TCP header flags. (tcp[13])  */
 753
 754        __u8            sacked;         /* State flags for SACK/FACK.   */
 755#define TCPCB_SACKED_ACKED      0x01    /* SKB ACK'd by a SACK block    */
 756#define TCPCB_SACKED_RETRANS    0x02    /* SKB retransmitted            */
 757#define TCPCB_LOST              0x04    /* SKB is lost                  */
 758#define TCPCB_TAGBITS           0x07    /* All tag bits                 */
 759#define TCPCB_REPAIRED          0x10    /* SKB repaired (no skb_mstamp) */
 760#define TCPCB_EVER_RETRANS      0x80    /* Ever retransmitted frame     */
 761#define TCPCB_RETRANS           (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
 762                                TCPCB_REPAIRED)
 763
 764        __u8            ip_dsfield;     /* IPv4 tos or IPv6 dsfield     */
 765        __u8            txstamp_ack:1,  /* Record TX timestamp for ack? */
 766                        eor:1,          /* Is skb MSG_EOR marked? */
 767                        unused:6;
 768        __u32           ack_seq;        /* Sequence number ACK'd        */
 769        union {
 770                struct {
 771                        /* There is space for up to 24 bytes */
 772                        __u32 in_flight:30,/* Bytes in flight at transmit */
 773                              is_app_limited:1, /* cwnd not fully used? */
 774                              unused:1;
 775                        /* pkts S/ACKed so far upon tx of skb, incl retrans: */
 776                        __u32 delivered;
 777                        /* start of send pipeline phase */
 778                        struct skb_mstamp first_tx_mstamp;
 779                        /* when we reached the "delivered" count */
 780                        struct skb_mstamp delivered_mstamp;
 781                } tx;   /* only used for outgoing skbs */
 782                union {
 783                        struct inet_skb_parm    h4;
 784#if IS_ENABLED(CONFIG_IPV6)
 785                        struct inet6_skb_parm   h6;
 786#endif
 787                } header;       /* For incoming skbs */
 788        };
 789};
 790
 791#define TCP_SKB_CB(__skb)       ((struct tcp_skb_cb *)&((__skb)->cb[0]))
 792
 793
 794#if IS_ENABLED(CONFIG_IPV6)
 795/* This is the variant of inet6_iif() that must be used by TCP,
 796 * as TCP moves IP6CB into a different location in skb->cb[]
 797 */
 798static inline int tcp_v6_iif(const struct sk_buff *skb)
 799{
 800        bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
 801
 802        return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
 803}
 804#endif
 805
 806/* TCP_SKB_CB reference means this can not be used from early demux */
 807static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
 808{
 809#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
 810        if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
 811            skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
 812                return true;
 813#endif
 814        return false;
 815}
 816
 817/* Due to TSO, an SKB can be composed of multiple actual
 818 * packets.  To keep these tracked properly, we use this.
 819 */
 820static inline int tcp_skb_pcount(const struct sk_buff *skb)
 821{
 822        return TCP_SKB_CB(skb)->tcp_gso_segs;
 823}
 824
 825static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
 826{
 827        TCP_SKB_CB(skb)->tcp_gso_segs = segs;
 828}
 829
 830static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
 831{
 832        TCP_SKB_CB(skb)->tcp_gso_segs += segs;
 833}
 834
 835/* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
 836static inline int tcp_skb_mss(const struct sk_buff *skb)
 837{
 838        return TCP_SKB_CB(skb)->tcp_gso_size;
 839}
 840
 841static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
 842{
 843        return likely(!TCP_SKB_CB(skb)->eor);
 844}
 845
 846/* Events passed to congestion control interface */
 847enum tcp_ca_event {
 848        CA_EVENT_TX_START,      /* first transmit when no packets in flight */
 849        CA_EVENT_CWND_RESTART,  /* congestion window restart */
 850        CA_EVENT_COMPLETE_CWR,  /* end of congestion recovery */
 851        CA_EVENT_LOSS,          /* loss timeout */
 852        CA_EVENT_ECN_NO_CE,     /* ECT set, but not CE marked */
 853        CA_EVENT_ECN_IS_CE,     /* received CE marked IP packet */
 854        CA_EVENT_DELAYED_ACK,   /* Delayed ack is sent */
 855        CA_EVENT_NON_DELAYED_ACK,
 856};
 857
 858/* Information about inbound ACK, passed to cong_ops->in_ack_event() */
 859enum tcp_ca_ack_event_flags {
 860        CA_ACK_SLOWPATH         = (1 << 0),     /* In slow path processing */
 861        CA_ACK_WIN_UPDATE       = (1 << 1),     /* ACK updated window */
 862        CA_ACK_ECE              = (1 << 2),     /* ECE bit is set on ack */
 863};
 864
 865/*
 866 * Interface for adding new TCP congestion control handlers
 867 */
 868#define TCP_CA_NAME_MAX 16
 869#define TCP_CA_MAX      128
 870#define TCP_CA_BUF_MAX  (TCP_CA_NAME_MAX*TCP_CA_MAX)
 871
 872#define TCP_CA_UNSPEC   0
 873
 874/* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
 875#define TCP_CONG_NON_RESTRICTED 0x1
 876/* Requires ECN/ECT set on all packets */
 877#define TCP_CONG_NEEDS_ECN      0x2
 878
 879union tcp_cc_info;
 880
 881struct ack_sample {
 882        u32 pkts_acked;
 883        s32 rtt_us;
 884        u32 in_flight;
 885};
 886
 887/* A rate sample measures the number of (original/retransmitted) data
 888 * packets delivered "delivered" over an interval of time "interval_us".
 889 * The tcp_rate.c code fills in the rate sample, and congestion
 890 * control modules that define a cong_control function to run at the end
 891 * of ACK processing can optionally chose to consult this sample when
 892 * setting cwnd and pacing rate.
 893 * A sample is invalid if "delivered" or "interval_us" is negative.
 894 */
 895struct rate_sample {
 896        struct  skb_mstamp prior_mstamp; /* starting timestamp for interval */
 897        u32  prior_delivered;   /* tp->delivered at "prior_mstamp" */
 898        s32  delivered;         /* number of packets delivered over interval */
 899        long interval_us;       /* time for tp->delivered to incr "delivered" */
 900        long rtt_us;            /* RTT of last (S)ACKed packet (or -1) */
 901        int  losses;            /* number of packets marked lost upon ACK */
 902        u32  acked_sacked;      /* number of packets newly (S)ACKed upon ACK */
 903        u32  prior_in_flight;   /* in flight before this ACK */
 904        bool is_app_limited;    /* is sample from packet with bubble in pipe? */
 905        bool is_retrans;        /* is sample from retransmission? */
 906};
 907
 908struct tcp_congestion_ops {
 909        struct list_head        list;
 910        u32 key;
 911        u32 flags;
 912
 913        /* initialize private data (optional) */
 914        void (*init)(struct sock *sk);
 915        /* cleanup private data  (optional) */
 916        void (*release)(struct sock *sk);
 917
 918        /* return slow start threshold (required) */
 919        u32 (*ssthresh)(struct sock *sk);
 920        /* do new cwnd calculation (required) */
 921        void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
 922        /* call before changing ca_state (optional) */
 923        void (*set_state)(struct sock *sk, u8 new_state);
 924        /* call when cwnd event occurs (optional) */
 925        void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
 926        /* call when ack arrives (optional) */
 927        void (*in_ack_event)(struct sock *sk, u32 flags);
 928        /* new value of cwnd after loss (optional) */
 929        u32  (*undo_cwnd)(struct sock *sk);
 930        /* hook for packet ack accounting (optional) */
 931        void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
 932        /* suggest number of segments for each skb to transmit (optional) */
 933        u32 (*tso_segs_goal)(struct sock *sk);
 934        /* returns the multiplier used in tcp_sndbuf_expand (optional) */
 935        u32 (*sndbuf_expand)(struct sock *sk);
 936        /* call when packets are delivered to update cwnd and pacing rate,
 937         * after all the ca_state processing. (optional)
 938         */
 939        void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
 940        /* get info for inet_diag (optional) */
 941        size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
 942                           union tcp_cc_info *info);
 943
 944        char            name[TCP_CA_NAME_MAX];
 945        struct module   *owner;
 946};
 947
 948int tcp_register_congestion_control(struct tcp_congestion_ops *type);
 949void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
 950
 951void tcp_assign_congestion_control(struct sock *sk);
 952void tcp_init_congestion_control(struct sock *sk);
 953void tcp_cleanup_congestion_control(struct sock *sk);
 954int tcp_set_default_congestion_control(const char *name);
 955void tcp_get_default_congestion_control(char *name);
 956void tcp_get_available_congestion_control(char *buf, size_t len);
 957void tcp_get_allowed_congestion_control(char *buf, size_t len);
 958int tcp_set_allowed_congestion_control(char *allowed);
 959int tcp_set_congestion_control(struct sock *sk, const char *name);
 960u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
 961void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
 962
 963u32 tcp_reno_ssthresh(struct sock *sk);
 964u32 tcp_reno_undo_cwnd(struct sock *sk);
 965void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
 966extern struct tcp_congestion_ops tcp_reno;
 967
 968struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
 969u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
 970#ifdef CONFIG_INET
 971char *tcp_ca_get_name_by_key(u32 key, char *buffer);
 972#else
 973static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
 974{
 975        return NULL;
 976}
 977#endif
 978
 979static inline bool tcp_ca_needs_ecn(const struct sock *sk)
 980{
 981        const struct inet_connection_sock *icsk = inet_csk(sk);
 982
 983        return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
 984}
 985
 986static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
 987{
 988        struct inet_connection_sock *icsk = inet_csk(sk);
 989
 990        if (icsk->icsk_ca_ops->set_state)
 991                icsk->icsk_ca_ops->set_state(sk, ca_state);
 992        icsk->icsk_ca_state = ca_state;
 993}
 994
 995static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
 996{
 997        const struct inet_connection_sock *icsk = inet_csk(sk);
 998
 999        if (icsk->icsk_ca_ops->cwnd_event)
1000                icsk->icsk_ca_ops->cwnd_event(sk, event);
1001}
1002
1003/* From tcp_rate.c */
1004void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1005void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1006                            struct rate_sample *rs);
1007void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1008                  struct skb_mstamp *now, struct rate_sample *rs);
1009void tcp_rate_check_app_limited(struct sock *sk);
1010
1011/* These functions determine how the current flow behaves in respect of SACK
1012 * handling. SACK is negotiated with the peer, and therefore it can vary
1013 * between different flows.
1014 *
1015 * tcp_is_sack - SACK enabled
1016 * tcp_is_reno - No SACK
1017 * tcp_is_fack - FACK enabled, implies SACK enabled
1018 */
1019static inline int tcp_is_sack(const struct tcp_sock *tp)
1020{
1021        return tp->rx_opt.sack_ok;
1022}
1023
1024static inline bool tcp_is_reno(const struct tcp_sock *tp)
1025{
1026        return !tcp_is_sack(tp);
1027}
1028
1029static inline bool tcp_is_fack(const struct tcp_sock *tp)
1030{
1031        return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
1032}
1033
1034static inline void tcp_enable_fack(struct tcp_sock *tp)
1035{
1036        tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
1037}
1038
1039static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1040{
1041        return tp->sacked_out + tp->lost_out;
1042}
1043
1044/* This determines how many packets are "in the network" to the best
1045 * of our knowledge.  In many cases it is conservative, but where
1046 * detailed information is available from the receiver (via SACK
1047 * blocks etc.) we can make more aggressive calculations.
1048 *
1049 * Use this for decisions involving congestion control, use just
1050 * tp->packets_out to determine if the send queue is empty or not.
1051 *
1052 * Read this equation as:
1053 *
1054 *      "Packets sent once on transmission queue" MINUS
1055 *      "Packets left network, but not honestly ACKed yet" PLUS
1056 *      "Packets fast retransmitted"
1057 */
1058static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1059{
1060        return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1061}
1062
1063#define TCP_INFINITE_SSTHRESH   0x7fffffff
1064
1065static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1066{
1067        return tp->snd_cwnd < tp->snd_ssthresh;
1068}
1069
1070static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1071{
1072        return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1073}
1074
1075static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1076{
1077        return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1078               (1 << inet_csk(sk)->icsk_ca_state);
1079}
1080
1081/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1082 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1083 * ssthresh.
1084 */
1085static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1086{
1087        const struct tcp_sock *tp = tcp_sk(sk);
1088
1089        if (tcp_in_cwnd_reduction(sk))
1090                return tp->snd_ssthresh;
1091        else
1092                return max(tp->snd_ssthresh,
1093                           ((tp->snd_cwnd >> 1) +
1094                            (tp->snd_cwnd >> 2)));
1095}
1096
1097/* Use define here intentionally to get WARN_ON location shown at the caller */
1098#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1099
1100void tcp_enter_cwr(struct sock *sk);
1101__u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1102
1103/* The maximum number of MSS of available cwnd for which TSO defers
1104 * sending if not using sysctl_tcp_tso_win_divisor.
1105 */
1106static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1107{
1108        return 3;
1109}
1110
1111/* Returns end sequence number of the receiver's advertised window */
1112static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1113{
1114        return tp->snd_una + tp->snd_wnd;
1115}
1116
1117/* We follow the spirit of RFC2861 to validate cwnd but implement a more
1118 * flexible approach. The RFC suggests cwnd should not be raised unless
1119 * it was fully used previously. And that's exactly what we do in
1120 * congestion avoidance mode. But in slow start we allow cwnd to grow
1121 * as long as the application has used half the cwnd.
1122 * Example :
1123 *    cwnd is 10 (IW10), but application sends 9 frames.
1124 *    We allow cwnd to reach 18 when all frames are ACKed.
1125 * This check is safe because it's as aggressive as slow start which already
1126 * risks 100% overshoot. The advantage is that we discourage application to
1127 * either send more filler packets or data to artificially blow up the cwnd
1128 * usage, and allow application-limited process to probe bw more aggressively.
1129 */
1130static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1131{
1132        const struct tcp_sock *tp = tcp_sk(sk);
1133
1134        /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1135        if (tcp_in_slow_start(tp))
1136                return tp->snd_cwnd < 2 * tp->max_packets_out;
1137
1138        return tp->is_cwnd_limited;
1139}
1140
1141/* Something is really bad, we could not queue an additional packet,
1142 * because qdisc is full or receiver sent a 0 window.
1143 * We do not want to add fuel to the fire, or abort too early,
1144 * so make sure the timer we arm now is at least 200ms in the future,
1145 * regardless of current icsk_rto value (as it could be ~2ms)
1146 */
1147static inline unsigned long tcp_probe0_base(const struct sock *sk)
1148{
1149        return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1150}
1151
1152/* Variant of inet_csk_rto_backoff() used for zero window probes */
1153static inline unsigned long tcp_probe0_when(const struct sock *sk,
1154                                            unsigned long max_when)
1155{
1156        u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1157
1158        return (unsigned long)min_t(u64, when, max_when);
1159}
1160
1161static inline void tcp_check_probe_timer(struct sock *sk)
1162{
1163        if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1164                inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1165                                          tcp_probe0_base(sk), TCP_RTO_MAX);
1166}
1167
1168static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1169{
1170        tp->snd_wl1 = seq;
1171}
1172
1173static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1174{
1175        tp->snd_wl1 = seq;
1176}
1177
1178/*
1179 * Calculate(/check) TCP checksum
1180 */
1181static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1182                                   __be32 daddr, __wsum base)
1183{
1184        return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1185}
1186
1187static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1188{
1189        return __skb_checksum_complete(skb);
1190}
1191
1192static inline bool tcp_checksum_complete(struct sk_buff *skb)
1193{
1194        return !skb_csum_unnecessary(skb) &&
1195                __tcp_checksum_complete(skb);
1196}
1197
1198/* Prequeue for VJ style copy to user, combined with checksumming. */
1199
1200static inline void tcp_prequeue_init(struct tcp_sock *tp)
1201{
1202        tp->ucopy.task = NULL;
1203        tp->ucopy.len = 0;
1204        tp->ucopy.memory = 0;
1205        skb_queue_head_init(&tp->ucopy.prequeue);
1206}
1207
1208bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1209bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1210int tcp_filter(struct sock *sk, struct sk_buff *skb);
1211
1212#undef STATE_TRACE
1213
1214#ifdef STATE_TRACE
1215static const char *statename[]={
1216        "Unused","Established","Syn Sent","Syn Recv",
1217        "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1218        "Close Wait","Last ACK","Listen","Closing"
1219};
1220#endif
1221void tcp_set_state(struct sock *sk, int state);
1222
1223void tcp_done(struct sock *sk);
1224
1225int tcp_abort(struct sock *sk, int err);
1226
1227static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1228{
1229        rx_opt->dsack = 0;
1230        rx_opt->num_sacks = 0;
1231}
1232
1233u32 tcp_default_init_rwnd(u32 mss);
1234void tcp_cwnd_restart(struct sock *sk, s32 delta);
1235
1236static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1237{
1238        struct tcp_sock *tp = tcp_sk(sk);
1239        s32 delta;
1240
1241        if (!sysctl_tcp_slow_start_after_idle || tp->packets_out)
1242                return;
1243        delta = tcp_time_stamp - tp->lsndtime;
1244        if (delta > inet_csk(sk)->icsk_rto)
1245                tcp_cwnd_restart(sk, delta);
1246}
1247
1248/* Determine a window scaling and initial window to offer. */
1249void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1250                               __u32 *window_clamp, int wscale_ok,
1251                               __u8 *rcv_wscale, __u32 init_rcv_wnd);
1252
1253static inline int tcp_win_from_space(int space)
1254{
1255        return sysctl_tcp_adv_win_scale<=0 ?
1256                (space>>(-sysctl_tcp_adv_win_scale)) :
1257                space - (space>>sysctl_tcp_adv_win_scale);
1258}
1259
1260/* Note: caller must be prepared to deal with negative returns */
1261static inline int tcp_space(const struct sock *sk)
1262{
1263        return tcp_win_from_space(sk->sk_rcvbuf -
1264                                  atomic_read(&sk->sk_rmem_alloc));
1265}
1266
1267static inline int tcp_full_space(const struct sock *sk)
1268{
1269        return tcp_win_from_space(sk->sk_rcvbuf);
1270}
1271
1272extern void tcp_openreq_init_rwin(struct request_sock *req,
1273                                  const struct sock *sk_listener,
1274                                  const struct dst_entry *dst);
1275
1276void tcp_enter_memory_pressure(struct sock *sk);
1277
1278static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1279{
1280        struct net *net = sock_net((struct sock *)tp);
1281
1282        return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1283}
1284
1285static inline int keepalive_time_when(const struct tcp_sock *tp)
1286{
1287        struct net *net = sock_net((struct sock *)tp);
1288
1289        return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1290}
1291
1292static inline int keepalive_probes(const struct tcp_sock *tp)
1293{
1294        struct net *net = sock_net((struct sock *)tp);
1295
1296        return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1297}
1298
1299static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1300{
1301        const struct inet_connection_sock *icsk = &tp->inet_conn;
1302
1303        return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1304                          tcp_time_stamp - tp->rcv_tstamp);
1305}
1306
1307static inline int tcp_fin_time(const struct sock *sk)
1308{
1309        int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1310        const int rto = inet_csk(sk)->icsk_rto;
1311
1312        if (fin_timeout < (rto << 2) - (rto >> 1))
1313                fin_timeout = (rto << 2) - (rto >> 1);
1314
1315        return fin_timeout;
1316}
1317
1318static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1319                                  int paws_win)
1320{
1321        if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1322                return true;
1323        if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1324                return true;
1325        /*
1326         * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1327         * then following tcp messages have valid values. Ignore 0 value,
1328         * or else 'negative' tsval might forbid us to accept their packets.
1329         */
1330        if (!rx_opt->ts_recent)
1331                return true;
1332        return false;
1333}
1334
1335static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1336                                   int rst)
1337{
1338        if (tcp_paws_check(rx_opt, 0))
1339                return false;
1340
1341        /* RST segments are not recommended to carry timestamp,
1342           and, if they do, it is recommended to ignore PAWS because
1343           "their cleanup function should take precedence over timestamps."
1344           Certainly, it is mistake. It is necessary to understand the reasons
1345           of this constraint to relax it: if peer reboots, clock may go
1346           out-of-sync and half-open connections will not be reset.
1347           Actually, the problem would be not existing if all
1348           the implementations followed draft about maintaining clock
1349           via reboots. Linux-2.2 DOES NOT!
1350
1351           However, we can relax time bounds for RST segments to MSL.
1352         */
1353        if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1354                return false;
1355        return true;
1356}
1357
1358bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1359                          int mib_idx, u32 *last_oow_ack_time);
1360
1361static inline void tcp_mib_init(struct net *net)
1362{
1363        /* See RFC 2012 */
1364        TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1365        TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1366        TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1367        TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1368}
1369
1370/* from STCP */
1371static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1372{
1373        tp->lost_skb_hint = NULL;
1374}
1375
1376static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1377{
1378        tcp_clear_retrans_hints_partial(tp);
1379        tp->retransmit_skb_hint = NULL;
1380}
1381
1382union tcp_md5_addr {
1383        struct in_addr  a4;
1384#if IS_ENABLED(CONFIG_IPV6)
1385        struct in6_addr a6;
1386#endif
1387};
1388
1389/* - key database */
1390struct tcp_md5sig_key {
1391        struct hlist_node       node;
1392        u8                      keylen;
1393        u8                      family; /* AF_INET or AF_INET6 */
1394        union tcp_md5_addr      addr;
1395        u8                      key[TCP_MD5SIG_MAXKEYLEN];
1396        struct rcu_head         rcu;
1397};
1398
1399/* - sock block */
1400struct tcp_md5sig_info {
1401        struct hlist_head       head;
1402        struct rcu_head         rcu;
1403};
1404
1405/* - pseudo header */
1406struct tcp4_pseudohdr {
1407        __be32          saddr;
1408        __be32          daddr;
1409        __u8            pad;
1410        __u8            protocol;
1411        __be16          len;
1412};
1413
1414struct tcp6_pseudohdr {
1415        struct in6_addr saddr;
1416        struct in6_addr daddr;
1417        __be32          len;
1418        __be32          protocol;       /* including padding */
1419};
1420
1421union tcp_md5sum_block {
1422        struct tcp4_pseudohdr ip4;
1423#if IS_ENABLED(CONFIG_IPV6)
1424        struct tcp6_pseudohdr ip6;
1425#endif
1426};
1427
1428/* - pool: digest algorithm, hash description and scratch buffer */
1429struct tcp_md5sig_pool {
1430        struct ahash_request    *md5_req;
1431        void                    *scratch;
1432};
1433
1434/* - functions */
1435int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1436                        const struct sock *sk, const struct sk_buff *skb);
1437int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1438                   int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1439int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1440                   int family);
1441struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1442                                         const struct sock *addr_sk);
1443
1444#ifdef CONFIG_TCP_MD5SIG
1445struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1446                                         const union tcp_md5_addr *addr,
1447                                         int family);
1448#define tcp_twsk_md5_key(twsk)  ((twsk)->tw_md5_key)
1449#else
1450static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1451                                         const union tcp_md5_addr *addr,
1452                                         int family)
1453{
1454        return NULL;
1455}
1456#define tcp_twsk_md5_key(twsk)  NULL
1457#endif
1458
1459bool tcp_alloc_md5sig_pool(void);
1460
1461struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1462static inline void tcp_put_md5sig_pool(void)
1463{
1464        local_bh_enable();
1465}
1466
1467int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1468                          unsigned int header_len);
1469int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1470                     const struct tcp_md5sig_key *key);
1471
1472/* From tcp_fastopen.c */
1473void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1474                            struct tcp_fastopen_cookie *cookie, int *syn_loss,
1475                            unsigned long *last_syn_loss);
1476void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1477                            struct tcp_fastopen_cookie *cookie, bool syn_lost,
1478                            u16 try_exp);
1479struct tcp_fastopen_request {
1480        /* Fast Open cookie. Size 0 means a cookie request */
1481        struct tcp_fastopen_cookie      cookie;
1482        struct msghdr                   *data;  /* data in MSG_FASTOPEN */
1483        size_t                          size;
1484        int                             copied; /* queued in tcp_connect() */
1485};
1486void tcp_free_fastopen_req(struct tcp_sock *tp);
1487
1488extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1489int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1490void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1491struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1492                              struct request_sock *req,
1493                              struct tcp_fastopen_cookie *foc,
1494                              struct dst_entry *dst);
1495void tcp_fastopen_init_key_once(bool publish);
1496bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1497                             struct tcp_fastopen_cookie *cookie);
1498bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1499#define TCP_FASTOPEN_KEY_LENGTH 16
1500
1501/* Fastopen key context */
1502struct tcp_fastopen_context {
1503        struct crypto_cipher    *tfm;
1504        __u8                    key[TCP_FASTOPEN_KEY_LENGTH];
1505        struct rcu_head         rcu;
1506};
1507
1508/* Latencies incurred by various limits for a sender. They are
1509 * chronograph-like stats that are mutually exclusive.
1510 */
1511enum tcp_chrono {
1512        TCP_CHRONO_UNSPEC,
1513        TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1514        TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1515        TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1516        __TCP_CHRONO_MAX,
1517};
1518
1519void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1520void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1521
1522/* write queue abstraction */
1523static inline void tcp_write_queue_purge(struct sock *sk)
1524{
1525        struct sk_buff *skb;
1526
1527        tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
1528        while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1529                sk_wmem_free_skb(sk, skb);
1530        sk_mem_reclaim(sk);
1531        tcp_clear_all_retrans_hints(tcp_sk(sk));
1532}
1533
1534static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1535{
1536        return skb_peek(&sk->sk_write_queue);
1537}
1538
1539static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1540{
1541        return skb_peek_tail(&sk->sk_write_queue);
1542}
1543
1544static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1545                                                   const struct sk_buff *skb)
1546{
1547        return skb_queue_next(&sk->sk_write_queue, skb);
1548}
1549
1550static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1551                                                   const struct sk_buff *skb)
1552{
1553        return skb_queue_prev(&sk->sk_write_queue, skb);
1554}
1555
1556#define tcp_for_write_queue(skb, sk)                                    \
1557        skb_queue_walk(&(sk)->sk_write_queue, skb)
1558
1559#define tcp_for_write_queue_from(skb, sk)                               \
1560        skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1561
1562#define tcp_for_write_queue_from_safe(skb, tmp, sk)                     \
1563        skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1564
1565static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1566{
1567        return sk->sk_send_head;
1568}
1569
1570static inline bool tcp_skb_is_last(const struct sock *sk,
1571                                   const struct sk_buff *skb)
1572{
1573        return skb_queue_is_last(&sk->sk_write_queue, skb);
1574}
1575
1576static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1577{
1578        if (tcp_skb_is_last(sk, skb))
1579                sk->sk_send_head = NULL;
1580        else
1581                sk->sk_send_head = tcp_write_queue_next(sk, skb);
1582}
1583
1584static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1585{
1586        if (sk->sk_send_head == skb_unlinked) {
1587                sk->sk_send_head = NULL;
1588                tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
1589        }
1590        if (tcp_sk(sk)->highest_sack == skb_unlinked)
1591                tcp_sk(sk)->highest_sack = NULL;
1592}
1593
1594static inline void tcp_init_send_head(struct sock *sk)
1595{
1596        sk->sk_send_head = NULL;
1597}
1598
1599static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1600{
1601        __skb_queue_tail(&sk->sk_write_queue, skb);
1602}
1603
1604static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1605{
1606        __tcp_add_write_queue_tail(sk, skb);
1607
1608        /* Queue it, remembering where we must start sending. */
1609        if (sk->sk_send_head == NULL) {
1610                sk->sk_send_head = skb;
1611                tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1612
1613                if (tcp_sk(sk)->highest_sack == NULL)
1614                        tcp_sk(sk)->highest_sack = skb;
1615        }
1616}
1617
1618static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1619{
1620        __skb_queue_head(&sk->sk_write_queue, skb);
1621}
1622
1623/* Insert buff after skb on the write queue of sk.  */
1624static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1625                                                struct sk_buff *buff,
1626                                                struct sock *sk)
1627{
1628        __skb_queue_after(&sk->sk_write_queue, skb, buff);
1629}
1630
1631/* Insert new before skb on the write queue of sk.  */
1632static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1633                                                  struct sk_buff *skb,
1634                                                  struct sock *sk)
1635{
1636        __skb_queue_before(&sk->sk_write_queue, skb, new);
1637
1638        if (sk->sk_send_head == skb)
1639                sk->sk_send_head = new;
1640}
1641
1642static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1643{
1644        __skb_unlink(skb, &sk->sk_write_queue);
1645}
1646
1647static inline bool tcp_write_queue_empty(struct sock *sk)
1648{
1649        return skb_queue_empty(&sk->sk_write_queue);
1650}
1651
1652static inline void tcp_push_pending_frames(struct sock *sk)
1653{
1654        if (tcp_send_head(sk)) {
1655                struct tcp_sock *tp = tcp_sk(sk);
1656
1657                __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1658        }
1659}
1660
1661/* Start sequence of the skb just after the highest skb with SACKed
1662 * bit, valid only if sacked_out > 0 or when the caller has ensured
1663 * validity by itself.
1664 */
1665static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1666{
1667        if (!tp->sacked_out)
1668                return tp->snd_una;
1669
1670        if (tp->highest_sack == NULL)
1671                return tp->snd_nxt;
1672
1673        return TCP_SKB_CB(tp->highest_sack)->seq;
1674}
1675
1676static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1677{
1678        tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1679                                                tcp_write_queue_next(sk, skb);
1680}
1681
1682static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1683{
1684        return tcp_sk(sk)->highest_sack;
1685}
1686
1687static inline void tcp_highest_sack_reset(struct sock *sk)
1688{
1689        tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1690}
1691
1692/* Called when old skb is about to be deleted (to be combined with new skb) */
1693static inline void tcp_highest_sack_combine(struct sock *sk,
1694                                            struct sk_buff *old,
1695                                            struct sk_buff *new)
1696{
1697        if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1698                tcp_sk(sk)->highest_sack = new;
1699}
1700
1701/* This helper checks if socket has IP_TRANSPARENT set */
1702static inline bool inet_sk_transparent(const struct sock *sk)
1703{
1704        switch (sk->sk_state) {
1705        case TCP_TIME_WAIT:
1706                return inet_twsk(sk)->tw_transparent;
1707        case TCP_NEW_SYN_RECV:
1708                return inet_rsk(inet_reqsk(sk))->no_srccheck;
1709        }
1710        return inet_sk(sk)->transparent;
1711}
1712
1713/* Determines whether this is a thin stream (which may suffer from
1714 * increased latency). Used to trigger latency-reducing mechanisms.
1715 */
1716static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1717{
1718        return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1719}
1720
1721/* /proc */
1722enum tcp_seq_states {
1723        TCP_SEQ_STATE_LISTENING,
1724        TCP_SEQ_STATE_ESTABLISHED,
1725};
1726
1727int tcp_seq_open(struct inode *inode, struct file *file);
1728
1729struct tcp_seq_afinfo {
1730        char                            *name;
1731        sa_family_t                     family;
1732        const struct file_operations    *seq_fops;
1733        struct seq_operations           seq_ops;
1734};
1735
1736struct tcp_iter_state {
1737        struct seq_net_private  p;
1738        sa_family_t             family;
1739        enum tcp_seq_states     state;
1740        struct sock             *syn_wait_sk;
1741        int                     bucket, offset, sbucket, num;
1742        loff_t                  last_pos;
1743};
1744
1745int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1746void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1747
1748extern struct request_sock_ops tcp_request_sock_ops;
1749extern struct request_sock_ops tcp6_request_sock_ops;
1750
1751void tcp_v4_destroy_sock(struct sock *sk);
1752
1753struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1754                                netdev_features_t features);
1755struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1756int tcp_gro_complete(struct sk_buff *skb);
1757
1758void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1759
1760static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1761{
1762        struct net *net = sock_net((struct sock *)tp);
1763        return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1764}
1765
1766static inline bool tcp_stream_memory_free(const struct sock *sk)
1767{
1768        const struct tcp_sock *tp = tcp_sk(sk);
1769        u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1770
1771        return notsent_bytes < tcp_notsent_lowat(tp);
1772}
1773
1774#ifdef CONFIG_PROC_FS
1775int tcp4_proc_init(void);
1776void tcp4_proc_exit(void);
1777#endif
1778
1779int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1780int tcp_conn_request(struct request_sock_ops *rsk_ops,
1781                     const struct tcp_request_sock_ops *af_ops,
1782                     struct sock *sk, struct sk_buff *skb);
1783
1784/* TCP af-specific functions */
1785struct tcp_sock_af_ops {
1786#ifdef CONFIG_TCP_MD5SIG
1787        struct tcp_md5sig_key   *(*md5_lookup) (const struct sock *sk,
1788                                                const struct sock *addr_sk);
1789        int             (*calc_md5_hash)(char *location,
1790                                         const struct tcp_md5sig_key *md5,
1791                                         const struct sock *sk,
1792                                         const struct sk_buff *skb);
1793        int             (*md5_parse)(struct sock *sk,
1794                                     char __user *optval,
1795                                     int optlen);
1796#endif
1797};
1798
1799struct tcp_request_sock_ops {
1800        u16 mss_clamp;
1801#ifdef CONFIG_TCP_MD5SIG
1802        struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1803                                                 const struct sock *addr_sk);
1804        int             (*calc_md5_hash) (char *location,
1805                                          const struct tcp_md5sig_key *md5,
1806                                          const struct sock *sk,
1807                                          const struct sk_buff *skb);
1808#endif
1809        void (*init_req)(struct request_sock *req,
1810                         const struct sock *sk_listener,
1811                         struct sk_buff *skb);
1812#ifdef CONFIG_SYN_COOKIES
1813        __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1814                                 __u16 *mss);
1815#endif
1816        struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1817                                       const struct request_sock *req,
1818                                       bool *strict);
1819        __u32 (*init_seq)(const struct sk_buff *skb, u32 *tsoff);
1820        int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1821                           struct flowi *fl, struct request_sock *req,
1822                           struct tcp_fastopen_cookie *foc,
1823                           enum tcp_synack_type synack_type);
1824};
1825
1826#ifdef CONFIG_SYN_COOKIES
1827static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1828                                         const struct sock *sk, struct sk_buff *skb,
1829                                         __u16 *mss)
1830{
1831        tcp_synq_overflow(sk);
1832        __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1833        return ops->cookie_init_seq(skb, mss);
1834}
1835#else
1836static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1837                                         const struct sock *sk, struct sk_buff *skb,
1838                                         __u16 *mss)
1839{
1840        return 0;
1841}
1842#endif
1843
1844int tcpv4_offload_init(void);
1845
1846void tcp_v4_init(void);
1847void tcp_init(void);
1848
1849/* tcp_recovery.c */
1850extern void tcp_rack_mark_lost(struct sock *sk, const struct skb_mstamp *now);
1851extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
1852                             const struct skb_mstamp *xmit_time,
1853                             const struct skb_mstamp *ack_time);
1854extern void tcp_rack_reo_timeout(struct sock *sk);
1855
1856/*
1857 * Save and compile IPv4 options, return a pointer to it
1858 */
1859static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1860{
1861        const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1862        struct ip_options_rcu *dopt = NULL;
1863
1864        if (opt->optlen) {
1865                int opt_size = sizeof(*dopt) + opt->optlen;
1866
1867                dopt = kmalloc(opt_size, GFP_ATOMIC);
1868                if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1869                        kfree(dopt);
1870                        dopt = NULL;
1871                }
1872        }
1873        return dopt;
1874}
1875
1876/* locally generated TCP pure ACKs have skb->truesize == 2
1877 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1878 * This is much faster than dissecting the packet to find out.
1879 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1880 */
1881static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1882{
1883        return skb->truesize == 2;
1884}
1885
1886static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
1887{
1888        skb->truesize = 2;
1889}
1890
1891static inline int tcp_inq(struct sock *sk)
1892{
1893        struct tcp_sock *tp = tcp_sk(sk);
1894        int answ;
1895
1896        if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1897                answ = 0;
1898        } else if (sock_flag(sk, SOCK_URGINLINE) ||
1899                   !tp->urg_data ||
1900                   before(tp->urg_seq, tp->copied_seq) ||
1901                   !before(tp->urg_seq, tp->rcv_nxt)) {
1902
1903                answ = tp->rcv_nxt - tp->copied_seq;
1904
1905                /* Subtract 1, if FIN was received */
1906                if (answ && sock_flag(sk, SOCK_DONE))
1907                        answ--;
1908        } else {
1909                answ = tp->urg_seq - tp->copied_seq;
1910        }
1911
1912        return answ;
1913}
1914
1915int tcp_peek_len(struct socket *sock);
1916
1917static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
1918{
1919        u16 segs_in;
1920
1921        segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1922        tp->segs_in += segs_in;
1923        if (skb->len > tcp_hdrlen(skb))
1924                tp->data_segs_in += segs_in;
1925}
1926
1927/*
1928 * TCP listen path runs lockless.
1929 * We forced "struct sock" to be const qualified to make sure
1930 * we don't modify one of its field by mistake.
1931 * Here, we increment sk_drops which is an atomic_t, so we can safely
1932 * make sock writable again.
1933 */
1934static inline void tcp_listendrop(const struct sock *sk)
1935{
1936        atomic_inc(&((struct sock *)sk)->sk_drops);
1937        __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
1938}
1939
1940#endif  /* _TCP_H */
1941