Linux Audio

Check our new training course

Loading...
   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/dmaengine.h>
  31#include <linux/crypto.h>
  32#include <linux/cryptohash.h>
  33#include <linux/kref.h>
  34
  35#include <net/inet_connection_sock.h>
  36#include <net/inet_timewait_sock.h>
  37#include <net/inet_hashtables.h>
  38#include <net/checksum.h>
  39#include <net/request_sock.h>
  40#include <net/sock.h>
  41#include <net/snmp.h>
  42#include <net/ip.h>
  43#include <net/tcp_states.h>
  44#include <net/inet_ecn.h>
  45#include <net/dst.h>
  46
  47#include <linux/seq_file.h>
  48#include <linux/memcontrol.h>
  49
  50extern struct inet_hashinfo tcp_hashinfo;
  51
  52extern struct percpu_counter tcp_orphan_count;
  53extern void 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/* Offer an initial receive window of 10 mss. */
  65#define TCP_DEFAULT_INIT_RCVWND	10
  66
  67/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
  68#define TCP_MIN_MSS		88U
  69
  70/* The least MTU to use for probing */
  71#define TCP_BASE_MSS		512
  72
  73/* After receiving this amount of duplicate ACKs fast retransmit starts. */
  74#define TCP_FASTRETRANS_THRESH 3
  75
  76/* Maximal reordering. */
  77#define TCP_MAX_REORDERING	127
  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	 5	/* number of times to retry active opening a
 102				 * connection: ~180sec is RFC minimum	*/
 103
 104#define TCP_SYNACK_RETRIES 5	/* number of times to retry passive opening a
 105				 * connection: ~180sec is RFC minimum	*/
 106
 107#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
 108				  * state, about 60 seconds	*/
 109#define TCP_FIN_TIMEOUT	TCP_TIMEWAIT_LEN
 110                                 /* BSD style FIN_WAIT2 deadlock breaker.
 111				  * It used to be 3min, new value is 60sec,
 112				  * to combine FIN-WAIT-2 timeout with
 113				  * TIME-WAIT timer.
 114				  */
 115
 116#define TCP_DELACK_MAX	((unsigned)(HZ/5))	/* maximal time to delay before sending an ACK */
 117#if HZ >= 100
 118#define TCP_DELACK_MIN	((unsigned)(HZ/25))	/* minimal time to delay before sending an ACK */
 119#define TCP_ATO_MIN	((unsigned)(HZ/25))
 120#else
 121#define TCP_DELACK_MIN	4U
 122#define TCP_ATO_MIN	4U
 123#endif
 124#define TCP_RTO_MAX	((unsigned)(120*HZ))
 125#define TCP_RTO_MIN	((unsigned)(HZ/5))
 126#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ))	/* RFC6298 2.1 initial RTO value	*/
 127#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ))	/* RFC 1122 initial RTO value, now
 128						 * used as a fallback RTO for the
 129						 * initial data transmission if no
 130						 * valid RTT sample has been acquired,
 131						 * most likely due to retrans in 3WHS.
 132						 */
 133
 134#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
 135					                 * for local resources.
 136					                 */
 137
 138#define TCP_KEEPALIVE_TIME	(120*60*HZ)	/* two hours */
 139#define TCP_KEEPALIVE_PROBES	9		/* Max of 9 keepalive probes	*/
 140#define TCP_KEEPALIVE_INTVL	(75*HZ)
 141
 142#define MAX_TCP_KEEPIDLE	32767
 143#define MAX_TCP_KEEPINTVL	32767
 144#define MAX_TCP_KEEPCNT		127
 145#define MAX_TCP_SYNCNT		127
 146
 147#define TCP_SYNQ_INTERVAL	(HZ/5)	/* Period of SYNACK timer */
 148
 149#define TCP_PAWS_24DAYS	(60 * 60 * 24 * 24)
 150#define TCP_PAWS_MSL	60		/* Per-host timestamps are invalidated
 151					 * after this time. It should be equal
 152					 * (or greater than) TCP_TIMEWAIT_LEN
 153					 * to provide reliability equal to one
 154					 * provided by timewait state.
 155					 */
 156#define TCP_PAWS_WINDOW	1		/* Replay window for per-host
 157					 * timestamps. It must be less than
 158					 * minimal timewait lifetime.
 159					 */
 160/*
 161 *	TCP option
 162 */
 163 
 164#define TCPOPT_NOP		1	/* Padding */
 165#define TCPOPT_EOL		0	/* End of options */
 166#define TCPOPT_MSS		2	/* Segment size negotiating */
 167#define TCPOPT_WINDOW		3	/* Window scaling */
 168#define TCPOPT_SACK_PERM        4       /* SACK Permitted */
 169#define TCPOPT_SACK             5       /* SACK Block */
 170#define TCPOPT_TIMESTAMP	8	/* Better RTT estimations/PAWS */
 171#define TCPOPT_MD5SIG		19	/* MD5 Signature (RFC2385) */
 172#define TCPOPT_COOKIE		253	/* Cookie extension (experimental) */
 173
 174/*
 175 *     TCP option lengths
 176 */
 177
 178#define TCPOLEN_MSS            4
 179#define TCPOLEN_WINDOW         3
 180#define TCPOLEN_SACK_PERM      2
 181#define TCPOLEN_TIMESTAMP      10
 182#define TCPOLEN_MD5SIG         18
 183#define TCPOLEN_COOKIE_BASE    2	/* Cookie-less header extension */
 184#define TCPOLEN_COOKIE_PAIR    3	/* Cookie pair header extension */
 185#define TCPOLEN_COOKIE_MIN     (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN)
 186#define TCPOLEN_COOKIE_MAX     (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX)
 187
 188/* But this is what stacks really send out. */
 189#define TCPOLEN_TSTAMP_ALIGNED		12
 190#define TCPOLEN_WSCALE_ALIGNED		4
 191#define TCPOLEN_SACKPERM_ALIGNED	4
 192#define TCPOLEN_SACK_BASE		2
 193#define TCPOLEN_SACK_BASE_ALIGNED	4
 194#define TCPOLEN_SACK_PERBLOCK		8
 195#define TCPOLEN_MD5SIG_ALIGNED		20
 196#define TCPOLEN_MSS_ALIGNED		4
 197
 198/* Flags in tp->nonagle */
 199#define TCP_NAGLE_OFF		1	/* Nagle's algo is disabled */
 200#define TCP_NAGLE_CORK		2	/* Socket is corked	    */
 201#define TCP_NAGLE_PUSH		4	/* Cork is overridden for already queued data */
 202
 203/* TCP thin-stream limits */
 204#define TCP_THIN_LINEAR_RETRIES 6       /* After 6 linear retries, do exp. backoff */
 205
 206/* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
 207#define TCP_INIT_CWND		10
 208
 209extern struct inet_timewait_death_row tcp_death_row;
 210
 211/* sysctl variables for tcp */
 212extern int sysctl_tcp_timestamps;
 213extern int sysctl_tcp_window_scaling;
 214extern int sysctl_tcp_sack;
 215extern int sysctl_tcp_fin_timeout;
 216extern int sysctl_tcp_keepalive_time;
 217extern int sysctl_tcp_keepalive_probes;
 218extern int sysctl_tcp_keepalive_intvl;
 219extern int sysctl_tcp_syn_retries;
 220extern int sysctl_tcp_synack_retries;
 221extern int sysctl_tcp_retries1;
 222extern int sysctl_tcp_retries2;
 223extern int sysctl_tcp_orphan_retries;
 224extern int sysctl_tcp_syncookies;
 225extern int sysctl_tcp_retrans_collapse;
 226extern int sysctl_tcp_stdurg;
 227extern int sysctl_tcp_rfc1337;
 228extern int sysctl_tcp_abort_on_overflow;
 229extern int sysctl_tcp_max_orphans;
 230extern int sysctl_tcp_fack;
 231extern int sysctl_tcp_reordering;
 232extern int sysctl_tcp_ecn;
 233extern int sysctl_tcp_dsack;
 234extern int sysctl_tcp_wmem[3];
 235extern int sysctl_tcp_rmem[3];
 236extern int sysctl_tcp_app_win;
 237extern int sysctl_tcp_adv_win_scale;
 238extern int sysctl_tcp_tw_reuse;
 239extern int sysctl_tcp_frto;
 240extern int sysctl_tcp_frto_response;
 241extern int sysctl_tcp_low_latency;
 242extern int sysctl_tcp_dma_copybreak;
 243extern int sysctl_tcp_nometrics_save;
 244extern int sysctl_tcp_moderate_rcvbuf;
 245extern int sysctl_tcp_tso_win_divisor;
 246extern int sysctl_tcp_abc;
 247extern int sysctl_tcp_mtu_probing;
 248extern int sysctl_tcp_base_mss;
 249extern int sysctl_tcp_workaround_signed_windows;
 250extern int sysctl_tcp_slow_start_after_idle;
 251extern int sysctl_tcp_max_ssthresh;
 252extern int sysctl_tcp_cookie_size;
 253extern int sysctl_tcp_thin_linear_timeouts;
 254extern int sysctl_tcp_thin_dupack;
 255extern int sysctl_tcp_early_retrans;
 256
 257extern atomic_long_t tcp_memory_allocated;
 258extern struct percpu_counter tcp_sockets_allocated;
 259extern int tcp_memory_pressure;
 260
 261/*
 262 * The next routines deal with comparing 32 bit unsigned ints
 263 * and worry about wraparound (automatic with unsigned arithmetic).
 264 */
 265
 266static inline bool before(__u32 seq1, __u32 seq2)
 267{
 268        return (__s32)(seq1-seq2) < 0;
 269}
 270#define after(seq2, seq1) 	before(seq1, seq2)
 271
 272/* is s2<=s1<=s3 ? */
 273static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
 274{
 275	return seq3 - seq2 >= seq1 - seq2;
 276}
 277
 278static inline bool tcp_out_of_memory(struct sock *sk)
 279{
 280	if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
 281	    sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
 282		return true;
 283	return false;
 284}
 285
 286static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
 287{
 288	struct percpu_counter *ocp = sk->sk_prot->orphan_count;
 289	int orphans = percpu_counter_read_positive(ocp);
 290
 291	if (orphans << shift > sysctl_tcp_max_orphans) {
 292		orphans = percpu_counter_sum_positive(ocp);
 293		if (orphans << shift > sysctl_tcp_max_orphans)
 294			return true;
 295	}
 296	return false;
 297}
 298
 299extern bool tcp_check_oom(struct sock *sk, int shift);
 300
 301/* syncookies: remember time of last synqueue overflow */
 302static inline void tcp_synq_overflow(struct sock *sk)
 303{
 304	tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
 305}
 306
 307/* syncookies: no recent synqueue overflow on this listening socket? */
 308static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
 309{
 310	unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
 311	return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
 312}
 313
 314extern struct proto tcp_prot;
 315
 316#define TCP_INC_STATS(net, field)	SNMP_INC_STATS((net)->mib.tcp_statistics, field)
 317#define TCP_INC_STATS_BH(net, field)	SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
 318#define TCP_DEC_STATS(net, field)	SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
 319#define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
 320#define TCP_ADD_STATS(net, field, val)	SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
 321
 322extern void tcp_init_mem(struct net *net);
 323
 324extern void tcp_v4_err(struct sk_buff *skb, u32);
 325
 326extern void tcp_shutdown (struct sock *sk, int how);
 327
 328extern int tcp_v4_rcv(struct sk_buff *skb);
 329
 330extern struct inet_peer *tcp_v4_get_peer(struct sock *sk, bool *release_it);
 331extern void *tcp_v4_tw_get_peer(struct sock *sk);
 332extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
 333extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
 334		       size_t size);
 335extern int tcp_sendpage(struct sock *sk, struct page *page, int offset,
 336			size_t size, int flags);
 337extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
 338extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
 339				 const struct tcphdr *th, unsigned int len);
 340extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
 341			       const struct tcphdr *th, unsigned int len);
 342extern void tcp_rcv_space_adjust(struct sock *sk);
 343extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
 344extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
 345extern void tcp_twsk_destructor(struct sock *sk);
 346extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
 347			       struct pipe_inode_info *pipe, size_t len,
 348			       unsigned int flags);
 349
 350static inline void tcp_dec_quickack_mode(struct sock *sk,
 351					 const unsigned int pkts)
 352{
 353	struct inet_connection_sock *icsk = inet_csk(sk);
 354
 355	if (icsk->icsk_ack.quick) {
 356		if (pkts >= icsk->icsk_ack.quick) {
 357			icsk->icsk_ack.quick = 0;
 358			/* Leaving quickack mode we deflate ATO. */
 359			icsk->icsk_ack.ato   = TCP_ATO_MIN;
 360		} else
 361			icsk->icsk_ack.quick -= pkts;
 362	}
 363}
 364
 365#define	TCP_ECN_OK		1
 366#define	TCP_ECN_QUEUE_CWR	2
 367#define	TCP_ECN_DEMAND_CWR	4
 368#define	TCP_ECN_SEEN		8
 369
 370enum tcp_tw_status {
 371	TCP_TW_SUCCESS = 0,
 372	TCP_TW_RST = 1,
 373	TCP_TW_ACK = 2,
 374	TCP_TW_SYN = 3
 375};
 376
 377
 378extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
 379						     struct sk_buff *skb,
 380						     const struct tcphdr *th);
 381extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
 382				   struct request_sock *req,
 383				   struct request_sock **prev);
 384extern int tcp_child_process(struct sock *parent, struct sock *child,
 385			     struct sk_buff *skb);
 386extern bool tcp_use_frto(struct sock *sk);
 387extern void tcp_enter_frto(struct sock *sk);
 388extern void tcp_enter_loss(struct sock *sk, int how);
 389extern void tcp_clear_retrans(struct tcp_sock *tp);
 390extern void tcp_update_metrics(struct sock *sk);
 391extern void tcp_close(struct sock *sk, long timeout);
 392extern void tcp_init_sock(struct sock *sk);
 393extern unsigned int tcp_poll(struct file * file, struct socket *sock,
 394			     struct poll_table_struct *wait);
 395extern int tcp_getsockopt(struct sock *sk, int level, int optname,
 396			  char __user *optval, int __user *optlen);
 397extern int tcp_setsockopt(struct sock *sk, int level, int optname,
 398			  char __user *optval, unsigned int optlen);
 399extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
 400				 char __user *optval, int __user *optlen);
 401extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
 402				 char __user *optval, unsigned int optlen);
 403extern void tcp_set_keepalive(struct sock *sk, int val);
 404extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
 405extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
 406		       size_t len, int nonblock, int flags, int *addr_len);
 407extern void tcp_parse_options(const struct sk_buff *skb,
 408			      struct tcp_options_received *opt_rx, const u8 **hvpp,
 409			      int estab);
 410extern const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
 411
 412/*
 413 *	TCP v4 functions exported for the inet6 API
 414 */
 415
 416extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
 417extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
 418extern struct sock * tcp_create_openreq_child(struct sock *sk,
 419					      struct request_sock *req,
 420					      struct sk_buff *skb);
 421extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
 422					  struct request_sock *req,
 423					  struct dst_entry *dst);
 424extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
 425extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr,
 426			  int addr_len);
 427extern int tcp_connect(struct sock *sk);
 428extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst,
 429					struct request_sock *req,
 430					struct request_values *rvp);
 431extern int tcp_disconnect(struct sock *sk, int flags);
 432
 433void tcp_connect_init(struct sock *sk);
 434void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
 435int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
 436
 437/* From syncookies.c */
 438extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
 439extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 
 440				    struct ip_options *opt);
 441#ifdef CONFIG_SYN_COOKIES
 442extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, 
 443				     __u16 *mss);
 444#else
 445static inline __u32 cookie_v4_init_sequence(struct sock *sk,
 446					    struct sk_buff *skb,
 447					    __u16 *mss)
 448{
 449	return 0;
 450}
 451#endif
 452
 453extern __u32 cookie_init_timestamp(struct request_sock *req);
 454extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *);
 455
 456/* From net/ipv6/syncookies.c */
 457extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
 458#ifdef CONFIG_SYN_COOKIES
 459extern __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb,
 460				     __u16 *mss);
 461#else
 462static inline __u32 cookie_v6_init_sequence(struct sock *sk,
 463					    struct sk_buff *skb,
 464					    __u16 *mss)
 465{
 466	return 0;
 467}
 468#endif
 469/* tcp_output.c */
 470
 471extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
 472				      int nonagle);
 473extern bool tcp_may_send_now(struct sock *sk);
 474extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
 475extern void tcp_retransmit_timer(struct sock *sk);
 476extern void tcp_xmit_retransmit_queue(struct sock *);
 477extern void tcp_simple_retransmit(struct sock *);
 478extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
 479extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
 480
 481extern void tcp_send_probe0(struct sock *);
 482extern void tcp_send_partial(struct sock *);
 483extern int tcp_write_wakeup(struct sock *);
 484extern void tcp_send_fin(struct sock *sk);
 485extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
 486extern int tcp_send_synack(struct sock *);
 487extern bool tcp_syn_flood_action(struct sock *sk,
 488				 const struct sk_buff *skb,
 489				 const char *proto);
 490extern void tcp_push_one(struct sock *, unsigned int mss_now);
 491extern void tcp_send_ack(struct sock *sk);
 492extern void tcp_send_delayed_ack(struct sock *sk);
 493
 494/* tcp_input.c */
 495extern void tcp_cwnd_application_limited(struct sock *sk);
 496extern void tcp_resume_early_retransmit(struct sock *sk);
 497extern void tcp_rearm_rto(struct sock *sk);
 498
 499/* tcp_timer.c */
 500extern void tcp_init_xmit_timers(struct sock *);
 501static inline void tcp_clear_xmit_timers(struct sock *sk)
 502{
 503	inet_csk_clear_xmit_timers(sk);
 504}
 505
 506extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
 507extern unsigned int tcp_current_mss(struct sock *sk);
 508
 509/* Bound MSS / TSO packet size with the half of the window */
 510static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
 511{
 512	int cutoff;
 513
 514	/* When peer uses tiny windows, there is no use in packetizing
 515	 * to sub-MSS pieces for the sake of SWS or making sure there
 516	 * are enough packets in the pipe for fast recovery.
 517	 *
 518	 * On the other hand, for extremely large MSS devices, handling
 519	 * smaller than MSS windows in this way does make sense.
 520	 */
 521	if (tp->max_window >= 512)
 522		cutoff = (tp->max_window >> 1);
 523	else
 524		cutoff = tp->max_window;
 525
 526	if (cutoff && pktsize > cutoff)
 527		return max_t(int, cutoff, 68U - tp->tcp_header_len);
 528	else
 529		return pktsize;
 530}
 531
 532/* tcp.c */
 533extern void tcp_get_info(const struct sock *, struct tcp_info *);
 534
 535/* Read 'sendfile()'-style from a TCP socket */
 536typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
 537				unsigned int, size_t);
 538extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
 539			 sk_read_actor_t recv_actor);
 540
 541extern void tcp_initialize_rcv_mss(struct sock *sk);
 542
 543extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
 544extern int tcp_mss_to_mtu(struct sock *sk, int mss);
 545extern void tcp_mtup_init(struct sock *sk);
 546extern void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt);
 547
 548static inline void tcp_bound_rto(const struct sock *sk)
 549{
 550	if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
 551		inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
 552}
 553
 554static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
 555{
 556	return (tp->srtt >> 3) + tp->rttvar;
 557}
 558
 559static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
 560{
 561	tp->pred_flags = htonl((tp->tcp_header_len << 26) |
 562			       ntohl(TCP_FLAG_ACK) |
 563			       snd_wnd);
 564}
 565
 566static inline void tcp_fast_path_on(struct tcp_sock *tp)
 567{
 568	__tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
 569}
 570
 571static inline void tcp_fast_path_check(struct sock *sk)
 572{
 573	struct tcp_sock *tp = tcp_sk(sk);
 574
 575	if (skb_queue_empty(&tp->out_of_order_queue) &&
 576	    tp->rcv_wnd &&
 577	    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
 578	    !tp->urg_data)
 579		tcp_fast_path_on(tp);
 580}
 581
 582/* Compute the actual rto_min value */
 583static inline u32 tcp_rto_min(struct sock *sk)
 584{
 585	const struct dst_entry *dst = __sk_dst_get(sk);
 586	u32 rto_min = TCP_RTO_MIN;
 587
 588	if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
 589		rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
 590	return rto_min;
 591}
 592
 593/* Compute the actual receive window we are currently advertising.
 594 * Rcv_nxt can be after the window if our peer push more data
 595 * than the offered window.
 596 */
 597static inline u32 tcp_receive_window(const struct tcp_sock *tp)
 598{
 599	s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
 600
 601	if (win < 0)
 602		win = 0;
 603	return (u32) win;
 604}
 605
 606/* Choose a new window, without checks for shrinking, and without
 607 * scaling applied to the result.  The caller does these things
 608 * if necessary.  This is a "raw" window selection.
 609 */
 610extern u32 __tcp_select_window(struct sock *sk);
 611
 612void tcp_send_window_probe(struct sock *sk);
 613
 614/* TCP timestamps are only 32-bits, this causes a slight
 615 * complication on 64-bit systems since we store a snapshot
 616 * of jiffies in the buffer control blocks below.  We decided
 617 * to use only the low 32-bits of jiffies and hide the ugly
 618 * casts with the following macro.
 619 */
 620#define tcp_time_stamp		((__u32)(jiffies))
 621
 622#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
 623
 624#define TCPHDR_FIN 0x01
 625#define TCPHDR_SYN 0x02
 626#define TCPHDR_RST 0x04
 627#define TCPHDR_PSH 0x08
 628#define TCPHDR_ACK 0x10
 629#define TCPHDR_URG 0x20
 630#define TCPHDR_ECE 0x40
 631#define TCPHDR_CWR 0x80
 632
 633/* This is what the send packet queuing engine uses to pass
 634 * TCP per-packet control information to the transmission code.
 635 * We also store the host-order sequence numbers in here too.
 636 * This is 44 bytes if IPV6 is enabled.
 637 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
 638 */
 639struct tcp_skb_cb {
 640	union {
 641		struct inet_skb_parm	h4;
 642#if IS_ENABLED(CONFIG_IPV6)
 643		struct inet6_skb_parm	h6;
 644#endif
 645	} header;	/* For incoming frames		*/
 646	__u32		seq;		/* Starting sequence number	*/
 647	__u32		end_seq;	/* SEQ + FIN + SYN + datalen	*/
 648	__u32		when;		/* used to compute rtt's	*/
 649	__u8		tcp_flags;	/* TCP header flags. (tcp[13])	*/
 650
 651	__u8		sacked;		/* State flags for SACK/FACK.	*/
 652#define TCPCB_SACKED_ACKED	0x01	/* SKB ACK'd by a SACK block	*/
 653#define TCPCB_SACKED_RETRANS	0x02	/* SKB retransmitted		*/
 654#define TCPCB_LOST		0x04	/* SKB is lost			*/
 655#define TCPCB_TAGBITS		0x07	/* All tag bits			*/
 656#define TCPCB_EVER_RETRANS	0x80	/* Ever retransmitted frame	*/
 657#define TCPCB_RETRANS		(TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
 658
 659	__u8		ip_dsfield;	/* IPv4 tos or IPv6 dsfield	*/
 660	/* 1 byte hole */
 661	__u32		ack_seq;	/* Sequence number ACK'd	*/
 662};
 663
 664#define TCP_SKB_CB(__skb)	((struct tcp_skb_cb *)&((__skb)->cb[0]))
 665
 666/* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set
 667 *
 668 * If we receive a SYN packet with these bits set, it means a network is
 669 * playing bad games with TOS bits. In order to avoid possible false congestion
 670 * notifications, we disable TCP ECN negociation.
 671 */
 672static inline void
 673TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb)
 674{
 675	const struct tcphdr *th = tcp_hdr(skb);
 676
 677	if (sysctl_tcp_ecn && th->ece && th->cwr &&
 678	    INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield))
 679		inet_rsk(req)->ecn_ok = 1;
 680}
 681
 682/* Due to TSO, an SKB can be composed of multiple actual
 683 * packets.  To keep these tracked properly, we use this.
 684 */
 685static inline int tcp_skb_pcount(const struct sk_buff *skb)
 686{
 687	return skb_shinfo(skb)->gso_segs;
 688}
 689
 690/* This is valid iff tcp_skb_pcount() > 1. */
 691static inline int tcp_skb_mss(const struct sk_buff *skb)
 692{
 693	return skb_shinfo(skb)->gso_size;
 694}
 695
 696/* Events passed to congestion control interface */
 697enum tcp_ca_event {
 698	CA_EVENT_TX_START,	/* first transmit when no packets in flight */
 699	CA_EVENT_CWND_RESTART,	/* congestion window restart */
 700	CA_EVENT_COMPLETE_CWR,	/* end of congestion recovery */
 701	CA_EVENT_FRTO,		/* fast recovery timeout */
 702	CA_EVENT_LOSS,		/* loss timeout */
 703	CA_EVENT_FAST_ACK,	/* in sequence ack */
 704	CA_EVENT_SLOW_ACK,	/* other ack */
 705};
 706
 707/*
 708 * Interface for adding new TCP congestion control handlers
 709 */
 710#define TCP_CA_NAME_MAX	16
 711#define TCP_CA_MAX	128
 712#define TCP_CA_BUF_MAX	(TCP_CA_NAME_MAX*TCP_CA_MAX)
 713
 714#define TCP_CONG_NON_RESTRICTED 0x1
 715#define TCP_CONG_RTT_STAMP	0x2
 716
 717struct tcp_congestion_ops {
 718	struct list_head	list;
 719	unsigned long flags;
 720
 721	/* initialize private data (optional) */
 722	void (*init)(struct sock *sk);
 723	/* cleanup private data  (optional) */
 724	void (*release)(struct sock *sk);
 725
 726	/* return slow start threshold (required) */
 727	u32 (*ssthresh)(struct sock *sk);
 728	/* lower bound for congestion window (optional) */
 729	u32 (*min_cwnd)(const struct sock *sk);
 730	/* do new cwnd calculation (required) */
 731	void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
 732	/* call before changing ca_state (optional) */
 733	void (*set_state)(struct sock *sk, u8 new_state);
 734	/* call when cwnd event occurs (optional) */
 735	void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
 736	/* new value of cwnd after loss (optional) */
 737	u32  (*undo_cwnd)(struct sock *sk);
 738	/* hook for packet ack accounting (optional) */
 739	void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
 740	/* get info for inet_diag (optional) */
 741	void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
 742
 743	char 		name[TCP_CA_NAME_MAX];
 744	struct module 	*owner;
 745};
 746
 747extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
 748extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
 749
 750extern void tcp_init_congestion_control(struct sock *sk);
 751extern void tcp_cleanup_congestion_control(struct sock *sk);
 752extern int tcp_set_default_congestion_control(const char *name);
 753extern void tcp_get_default_congestion_control(char *name);
 754extern void tcp_get_available_congestion_control(char *buf, size_t len);
 755extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
 756extern int tcp_set_allowed_congestion_control(char *allowed);
 757extern int tcp_set_congestion_control(struct sock *sk, const char *name);
 758extern void tcp_slow_start(struct tcp_sock *tp);
 759extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
 760
 761extern struct tcp_congestion_ops tcp_init_congestion_ops;
 762extern u32 tcp_reno_ssthresh(struct sock *sk);
 763extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
 764extern u32 tcp_reno_min_cwnd(const struct sock *sk);
 765extern struct tcp_congestion_ops tcp_reno;
 766
 767static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
 768{
 769	struct inet_connection_sock *icsk = inet_csk(sk);
 770
 771	if (icsk->icsk_ca_ops->set_state)
 772		icsk->icsk_ca_ops->set_state(sk, ca_state);
 773	icsk->icsk_ca_state = ca_state;
 774}
 775
 776static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
 777{
 778	const struct inet_connection_sock *icsk = inet_csk(sk);
 779
 780	if (icsk->icsk_ca_ops->cwnd_event)
 781		icsk->icsk_ca_ops->cwnd_event(sk, event);
 782}
 783
 784/* These functions determine how the current flow behaves in respect of SACK
 785 * handling. SACK is negotiated with the peer, and therefore it can vary
 786 * between different flows.
 787 *
 788 * tcp_is_sack - SACK enabled
 789 * tcp_is_reno - No SACK
 790 * tcp_is_fack - FACK enabled, implies SACK enabled
 791 */
 792static inline int tcp_is_sack(const struct tcp_sock *tp)
 793{
 794	return tp->rx_opt.sack_ok;
 795}
 796
 797static inline bool tcp_is_reno(const struct tcp_sock *tp)
 798{
 799	return !tcp_is_sack(tp);
 800}
 801
 802static inline bool tcp_is_fack(const struct tcp_sock *tp)
 803{
 804	return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
 805}
 806
 807static inline void tcp_enable_fack(struct tcp_sock *tp)
 808{
 809	tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
 810}
 811
 812/* TCP early-retransmit (ER) is similar to but more conservative than
 813 * the thin-dupack feature.  Enable ER only if thin-dupack is disabled.
 814 */
 815static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
 816{
 817	tp->do_early_retrans = sysctl_tcp_early_retrans &&
 818		!sysctl_tcp_thin_dupack && sysctl_tcp_reordering == 3;
 819	tp->early_retrans_delayed = 0;
 820}
 821
 822static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
 823{
 824	tp->do_early_retrans = 0;
 825}
 826
 827static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
 828{
 829	return tp->sacked_out + tp->lost_out;
 830}
 831
 832/* This determines how many packets are "in the network" to the best
 833 * of our knowledge.  In many cases it is conservative, but where
 834 * detailed information is available from the receiver (via SACK
 835 * blocks etc.) we can make more aggressive calculations.
 836 *
 837 * Use this for decisions involving congestion control, use just
 838 * tp->packets_out to determine if the send queue is empty or not.
 839 *
 840 * Read this equation as:
 841 *
 842 *	"Packets sent once on transmission queue" MINUS
 843 *	"Packets left network, but not honestly ACKed yet" PLUS
 844 *	"Packets fast retransmitted"
 845 */
 846static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
 847{
 848	return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
 849}
 850
 851#define TCP_INFINITE_SSTHRESH	0x7fffffff
 852
 853static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
 854{
 855	return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
 856}
 857
 858/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
 859 * The exception is rate halving phase, when cwnd is decreasing towards
 860 * ssthresh.
 861 */
 862static inline __u32 tcp_current_ssthresh(const struct sock *sk)
 863{
 864	const struct tcp_sock *tp = tcp_sk(sk);
 865
 866	if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
 867		return tp->snd_ssthresh;
 868	else
 869		return max(tp->snd_ssthresh,
 870			   ((tp->snd_cwnd >> 1) +
 871			    (tp->snd_cwnd >> 2)));
 872}
 873
 874/* Use define here intentionally to get WARN_ON location shown at the caller */
 875#define tcp_verify_left_out(tp)	WARN_ON(tcp_left_out(tp) > tp->packets_out)
 876
 877extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
 878extern __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
 879
 880/* The maximum number of MSS of available cwnd for which TSO defers
 881 * sending if not using sysctl_tcp_tso_win_divisor.
 882 */
 883static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
 884{
 885	return 3;
 886}
 887
 888/* Slow start with delack produces 3 packets of burst, so that
 889 * it is safe "de facto".  This will be the default - same as
 890 * the default reordering threshold - but if reordering increases,
 891 * we must be able to allow cwnd to burst at least this much in order
 892 * to not pull it back when holes are filled.
 893 */
 894static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
 895{
 896	return tp->reordering;
 897}
 898
 899/* Returns end sequence number of the receiver's advertised window */
 900static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
 901{
 902	return tp->snd_una + tp->snd_wnd;
 903}
 904extern bool tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
 905
 906static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
 907				       const struct sk_buff *skb)
 908{
 909	if (skb->len < mss)
 910		tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
 911}
 912
 913static inline void tcp_check_probe_timer(struct sock *sk)
 914{
 915	const struct tcp_sock *tp = tcp_sk(sk);
 916	const struct inet_connection_sock *icsk = inet_csk(sk);
 917
 918	if (!tp->packets_out && !icsk->icsk_pending)
 919		inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
 920					  icsk->icsk_rto, TCP_RTO_MAX);
 921}
 922
 923static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
 924{
 925	tp->snd_wl1 = seq;
 926}
 927
 928static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
 929{
 930	tp->snd_wl1 = seq;
 931}
 932
 933/*
 934 * Calculate(/check) TCP checksum
 935 */
 936static inline __sum16 tcp_v4_check(int len, __be32 saddr,
 937				   __be32 daddr, __wsum base)
 938{
 939	return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
 940}
 941
 942static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
 943{
 944	return __skb_checksum_complete(skb);
 945}
 946
 947static inline bool tcp_checksum_complete(struct sk_buff *skb)
 948{
 949	return !skb_csum_unnecessary(skb) &&
 950		__tcp_checksum_complete(skb);
 951}
 952
 953/* Prequeue for VJ style copy to user, combined with checksumming. */
 954
 955static inline void tcp_prequeue_init(struct tcp_sock *tp)
 956{
 957	tp->ucopy.task = NULL;
 958	tp->ucopy.len = 0;
 959	tp->ucopy.memory = 0;
 960	skb_queue_head_init(&tp->ucopy.prequeue);
 961#ifdef CONFIG_NET_DMA
 962	tp->ucopy.dma_chan = NULL;
 963	tp->ucopy.wakeup = 0;
 964	tp->ucopy.pinned_list = NULL;
 965	tp->ucopy.dma_cookie = 0;
 966#endif
 967}
 968
 969/* Packet is added to VJ-style prequeue for processing in process
 970 * context, if a reader task is waiting. Apparently, this exciting
 971 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
 972 * failed somewhere. Latency? Burstiness? Well, at least now we will
 973 * see, why it failed. 8)8)				  --ANK
 974 *
 975 * NOTE: is this not too big to inline?
 976 */
 977static inline bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
 978{
 979	struct tcp_sock *tp = tcp_sk(sk);
 980
 981	if (sysctl_tcp_low_latency || !tp->ucopy.task)
 982		return false;
 983
 984	__skb_queue_tail(&tp->ucopy.prequeue, skb);
 985	tp->ucopy.memory += skb->truesize;
 986	if (tp->ucopy.memory > sk->sk_rcvbuf) {
 987		struct sk_buff *skb1;
 988
 989		BUG_ON(sock_owned_by_user(sk));
 990
 991		while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
 992			sk_backlog_rcv(sk, skb1);
 993			NET_INC_STATS_BH(sock_net(sk),
 994					 LINUX_MIB_TCPPREQUEUEDROPPED);
 995		}
 996
 997		tp->ucopy.memory = 0;
 998	} else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
 999		wake_up_interruptible_sync_poll(sk_sleep(sk),
1000					   POLLIN | POLLRDNORM | POLLRDBAND);
1001		if (!inet_csk_ack_scheduled(sk))
1002			inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1003						  (3 * tcp_rto_min(sk)) / 4,
1004						  TCP_RTO_MAX);
1005	}
1006	return true;
1007}
1008
1009
1010#undef STATE_TRACE
1011
1012#ifdef STATE_TRACE
1013static const char *statename[]={
1014	"Unused","Established","Syn Sent","Syn Recv",
1015	"Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1016	"Close Wait","Last ACK","Listen","Closing"
1017};
1018#endif
1019extern void tcp_set_state(struct sock *sk, int state);
1020
1021extern void tcp_done(struct sock *sk);
1022
1023static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1024{
1025	rx_opt->dsack = 0;
1026	rx_opt->num_sacks = 0;
1027}
1028
1029/* Determine a window scaling and initial window to offer. */
1030extern void tcp_select_initial_window(int __space, __u32 mss,
1031				      __u32 *rcv_wnd, __u32 *window_clamp,
1032				      int wscale_ok, __u8 *rcv_wscale,
1033				      __u32 init_rcv_wnd);
1034
1035static inline int tcp_win_from_space(int space)
1036{
1037	return sysctl_tcp_adv_win_scale<=0 ?
1038		(space>>(-sysctl_tcp_adv_win_scale)) :
1039		space - (space>>sysctl_tcp_adv_win_scale);
1040}
1041
1042/* Note: caller must be prepared to deal with negative returns */ 
1043static inline int tcp_space(const struct sock *sk)
1044{
1045	return tcp_win_from_space(sk->sk_rcvbuf -
1046				  atomic_read(&sk->sk_rmem_alloc));
1047} 
1048
1049static inline int tcp_full_space(const struct sock *sk)
1050{
1051	return tcp_win_from_space(sk->sk_rcvbuf); 
1052}
1053
1054static inline void tcp_openreq_init(struct request_sock *req,
1055				    struct tcp_options_received *rx_opt,
1056				    struct sk_buff *skb)
1057{
1058	struct inet_request_sock *ireq = inet_rsk(req);
1059
1060	req->rcv_wnd = 0;		/* So that tcp_send_synack() knows! */
1061	req->cookie_ts = 0;
1062	tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1063	req->mss = rx_opt->mss_clamp;
1064	req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1065	ireq->tstamp_ok = rx_opt->tstamp_ok;
1066	ireq->sack_ok = rx_opt->sack_ok;
1067	ireq->snd_wscale = rx_opt->snd_wscale;
1068	ireq->wscale_ok = rx_opt->wscale_ok;
1069	ireq->acked = 0;
1070	ireq->ecn_ok = 0;
1071	ireq->rmt_port = tcp_hdr(skb)->source;
1072	ireq->loc_port = tcp_hdr(skb)->dest;
1073}
1074
1075extern void tcp_enter_memory_pressure(struct sock *sk);
1076
1077static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1078{
1079	return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1080}
1081
1082static inline int keepalive_time_when(const struct tcp_sock *tp)
1083{
1084	return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1085}
1086
1087static inline int keepalive_probes(const struct tcp_sock *tp)
1088{
1089	return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1090}
1091
1092static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1093{
1094	const struct inet_connection_sock *icsk = &tp->inet_conn;
1095
1096	return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1097			  tcp_time_stamp - tp->rcv_tstamp);
1098}
1099
1100static inline int tcp_fin_time(const struct sock *sk)
1101{
1102	int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1103	const int rto = inet_csk(sk)->icsk_rto;
1104
1105	if (fin_timeout < (rto << 2) - (rto >> 1))
1106		fin_timeout = (rto << 2) - (rto >> 1);
1107
1108	return fin_timeout;
1109}
1110
1111static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1112				  int paws_win)
1113{
1114	if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1115		return true;
1116	if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1117		return true;
1118	/*
1119	 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1120	 * then following tcp messages have valid values. Ignore 0 value,
1121	 * or else 'negative' tsval might forbid us to accept their packets.
1122	 */
1123	if (!rx_opt->ts_recent)
1124		return true;
1125	return false;
1126}
1127
1128static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1129				   int rst)
1130{
1131	if (tcp_paws_check(rx_opt, 0))
1132		return false;
1133
1134	/* RST segments are not recommended to carry timestamp,
1135	   and, if they do, it is recommended to ignore PAWS because
1136	   "their cleanup function should take precedence over timestamps."
1137	   Certainly, it is mistake. It is necessary to understand the reasons
1138	   of this constraint to relax it: if peer reboots, clock may go
1139	   out-of-sync and half-open connections will not be reset.
1140	   Actually, the problem would be not existing if all
1141	   the implementations followed draft about maintaining clock
1142	   via reboots. Linux-2.2 DOES NOT!
1143
1144	   However, we can relax time bounds for RST segments to MSL.
1145	 */
1146	if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1147		return false;
1148	return true;
1149}
1150
1151static inline void tcp_mib_init(struct net *net)
1152{
1153	/* See RFC 2012 */
1154	TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1155	TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1156	TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1157	TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1158}
1159
1160/* from STCP */
1161static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1162{
1163	tp->lost_skb_hint = NULL;
1164	tp->scoreboard_skb_hint = NULL;
1165}
1166
1167static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1168{
1169	tcp_clear_retrans_hints_partial(tp);
1170	tp->retransmit_skb_hint = NULL;
1171}
1172
1173/* MD5 Signature */
1174struct crypto_hash;
1175
1176union tcp_md5_addr {
1177	struct in_addr  a4;
1178#if IS_ENABLED(CONFIG_IPV6)
1179	struct in6_addr	a6;
1180#endif
1181};
1182
1183/* - key database */
1184struct tcp_md5sig_key {
1185	struct hlist_node	node;
1186	u8			keylen;
1187	u8			family; /* AF_INET or AF_INET6 */
1188	union tcp_md5_addr	addr;
1189	u8			key[TCP_MD5SIG_MAXKEYLEN];
1190	struct rcu_head		rcu;
1191};
1192
1193/* - sock block */
1194struct tcp_md5sig_info {
1195	struct hlist_head	head;
1196	struct rcu_head		rcu;
1197};
1198
1199/* - pseudo header */
1200struct tcp4_pseudohdr {
1201	__be32		saddr;
1202	__be32		daddr;
1203	__u8		pad;
1204	__u8		protocol;
1205	__be16		len;
1206};
1207
1208struct tcp6_pseudohdr {
1209	struct in6_addr	saddr;
1210	struct in6_addr daddr;
1211	__be32		len;
1212	__be32		protocol;	/* including padding */
1213};
1214
1215union tcp_md5sum_block {
1216	struct tcp4_pseudohdr ip4;
1217#if IS_ENABLED(CONFIG_IPV6)
1218	struct tcp6_pseudohdr ip6;
1219#endif
1220};
1221
1222/* - pool: digest algorithm, hash description and scratch buffer */
1223struct tcp_md5sig_pool {
1224	struct hash_desc	md5_desc;
1225	union tcp_md5sum_block	md5_blk;
1226};
1227
1228/* - functions */
1229extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1230			       const struct sock *sk,
1231			       const struct request_sock *req,
1232			       const struct sk_buff *skb);
1233extern int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1234			  int family, const u8 *newkey,
1235			  u8 newkeylen, gfp_t gfp);
1236extern int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1237			  int family);
1238extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1239					 struct sock *addr_sk);
1240
1241#ifdef CONFIG_TCP_MD5SIG
1242extern struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1243			const union tcp_md5_addr *addr, int family);
1244#define tcp_twsk_md5_key(twsk)	((twsk)->tw_md5_key)
1245#else
1246static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1247					 const union tcp_md5_addr *addr,
1248					 int family)
1249{
1250	return NULL;
1251}
1252#define tcp_twsk_md5_key(twsk)	NULL
1253#endif
1254
1255extern struct tcp_md5sig_pool __percpu *tcp_alloc_md5sig_pool(struct sock *);
1256extern void tcp_free_md5sig_pool(void);
1257
1258extern struct tcp_md5sig_pool	*tcp_get_md5sig_pool(void);
1259extern void tcp_put_md5sig_pool(void);
1260
1261extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1262extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1263				 unsigned int header_len);
1264extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1265			    const struct tcp_md5sig_key *key);
1266
1267/* write queue abstraction */
1268static inline void tcp_write_queue_purge(struct sock *sk)
1269{
1270	struct sk_buff *skb;
1271
1272	while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1273		sk_wmem_free_skb(sk, skb);
1274	sk_mem_reclaim(sk);
1275	tcp_clear_all_retrans_hints(tcp_sk(sk));
1276}
1277
1278static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1279{
1280	return skb_peek(&sk->sk_write_queue);
1281}
1282
1283static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1284{
1285	return skb_peek_tail(&sk->sk_write_queue);
1286}
1287
1288static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1289						   const struct sk_buff *skb)
1290{
1291	return skb_queue_next(&sk->sk_write_queue, skb);
1292}
1293
1294static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1295						   const struct sk_buff *skb)
1296{
1297	return skb_queue_prev(&sk->sk_write_queue, skb);
1298}
1299
1300#define tcp_for_write_queue(skb, sk)					\
1301	skb_queue_walk(&(sk)->sk_write_queue, skb)
1302
1303#define tcp_for_write_queue_from(skb, sk)				\
1304	skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1305
1306#define tcp_for_write_queue_from_safe(skb, tmp, sk)			\
1307	skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1308
1309static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1310{
1311	return sk->sk_send_head;
1312}
1313
1314static inline bool tcp_skb_is_last(const struct sock *sk,
1315				   const struct sk_buff *skb)
1316{
1317	return skb_queue_is_last(&sk->sk_write_queue, skb);
1318}
1319
1320static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1321{
1322	if (tcp_skb_is_last(sk, skb))
1323		sk->sk_send_head = NULL;
1324	else
1325		sk->sk_send_head = tcp_write_queue_next(sk, skb);
1326}
1327
1328static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1329{
1330	if (sk->sk_send_head == skb_unlinked)
1331		sk->sk_send_head = NULL;
1332}
1333
1334static inline void tcp_init_send_head(struct sock *sk)
1335{
1336	sk->sk_send_head = NULL;
1337}
1338
1339static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1340{
1341	__skb_queue_tail(&sk->sk_write_queue, skb);
1342}
1343
1344static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1345{
1346	__tcp_add_write_queue_tail(sk, skb);
1347
1348	/* Queue it, remembering where we must start sending. */
1349	if (sk->sk_send_head == NULL) {
1350		sk->sk_send_head = skb;
1351
1352		if (tcp_sk(sk)->highest_sack == NULL)
1353			tcp_sk(sk)->highest_sack = skb;
1354	}
1355}
1356
1357static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1358{
1359	__skb_queue_head(&sk->sk_write_queue, skb);
1360}
1361
1362/* Insert buff after skb on the write queue of sk.  */
1363static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1364						struct sk_buff *buff,
1365						struct sock *sk)
1366{
1367	__skb_queue_after(&sk->sk_write_queue, skb, buff);
1368}
1369
1370/* Insert new before skb on the write queue of sk.  */
1371static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1372						  struct sk_buff *skb,
1373						  struct sock *sk)
1374{
1375	__skb_queue_before(&sk->sk_write_queue, skb, new);
1376
1377	if (sk->sk_send_head == skb)
1378		sk->sk_send_head = new;
1379}
1380
1381static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1382{
1383	__skb_unlink(skb, &sk->sk_write_queue);
1384}
1385
1386static inline bool tcp_write_queue_empty(struct sock *sk)
1387{
1388	return skb_queue_empty(&sk->sk_write_queue);
1389}
1390
1391static inline void tcp_push_pending_frames(struct sock *sk)
1392{
1393	if (tcp_send_head(sk)) {
1394		struct tcp_sock *tp = tcp_sk(sk);
1395
1396		__tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1397	}
1398}
1399
1400/* Start sequence of the skb just after the highest skb with SACKed
1401 * bit, valid only if sacked_out > 0 or when the caller has ensured
1402 * validity by itself.
1403 */
1404static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1405{
1406	if (!tp->sacked_out)
1407		return tp->snd_una;
1408
1409	if (tp->highest_sack == NULL)
1410		return tp->snd_nxt;
1411
1412	return TCP_SKB_CB(tp->highest_sack)->seq;
1413}
1414
1415static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1416{
1417	tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1418						tcp_write_queue_next(sk, skb);
1419}
1420
1421static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1422{
1423	return tcp_sk(sk)->highest_sack;
1424}
1425
1426static inline void tcp_highest_sack_reset(struct sock *sk)
1427{
1428	tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1429}
1430
1431/* Called when old skb is about to be deleted (to be combined with new skb) */
1432static inline void tcp_highest_sack_combine(struct sock *sk,
1433					    struct sk_buff *old,
1434					    struct sk_buff *new)
1435{
1436	if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1437		tcp_sk(sk)->highest_sack = new;
1438}
1439
1440/* Determines whether this is a thin stream (which may suffer from
1441 * increased latency). Used to trigger latency-reducing mechanisms.
1442 */
1443static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1444{
1445	return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1446}
1447
1448/* /proc */
1449enum tcp_seq_states {
1450	TCP_SEQ_STATE_LISTENING,
1451	TCP_SEQ_STATE_OPENREQ,
1452	TCP_SEQ_STATE_ESTABLISHED,
1453	TCP_SEQ_STATE_TIME_WAIT,
1454};
1455
1456int tcp_seq_open(struct inode *inode, struct file *file);
1457
1458struct tcp_seq_afinfo {
1459	char				*name;
1460	sa_family_t			family;
1461	const struct file_operations	*seq_fops;
1462	struct seq_operations		seq_ops;
1463};
1464
1465struct tcp_iter_state {
1466	struct seq_net_private	p;
1467	sa_family_t		family;
1468	enum tcp_seq_states	state;
1469	struct sock		*syn_wait_sk;
1470	int			bucket, offset, sbucket, num, uid;
1471	loff_t			last_pos;
1472};
1473
1474extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1475extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1476
1477extern struct request_sock_ops tcp_request_sock_ops;
1478extern struct request_sock_ops tcp6_request_sock_ops;
1479
1480extern void tcp_v4_destroy_sock(struct sock *sk);
1481
1482extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1483extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb,
1484				       netdev_features_t features);
1485extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1486					struct sk_buff *skb);
1487extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1488					 struct sk_buff *skb);
1489extern int tcp_gro_complete(struct sk_buff *skb);
1490extern int tcp4_gro_complete(struct sk_buff *skb);
1491
1492#ifdef CONFIG_PROC_FS
1493extern int tcp4_proc_init(void);
1494extern void tcp4_proc_exit(void);
1495#endif
1496
1497/* TCP af-specific functions */
1498struct tcp_sock_af_ops {
1499#ifdef CONFIG_TCP_MD5SIG
1500	struct tcp_md5sig_key	*(*md5_lookup) (struct sock *sk,
1501						struct sock *addr_sk);
1502	int			(*calc_md5_hash) (char *location,
1503						  struct tcp_md5sig_key *md5,
1504						  const struct sock *sk,
1505						  const struct request_sock *req,
1506						  const struct sk_buff *skb);
1507	int			(*md5_parse) (struct sock *sk,
1508					      char __user *optval,
1509					      int optlen);
1510#endif
1511};
1512
1513struct tcp_request_sock_ops {
1514#ifdef CONFIG_TCP_MD5SIG
1515	struct tcp_md5sig_key	*(*md5_lookup) (struct sock *sk,
1516						struct request_sock *req);
1517	int			(*calc_md5_hash) (char *location,
1518						  struct tcp_md5sig_key *md5,
1519						  const struct sock *sk,
1520						  const struct request_sock *req,
1521						  const struct sk_buff *skb);
1522#endif
1523};
1524
1525/* Using SHA1 for now, define some constants.
1526 */
1527#define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS)
1528#define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4)
1529#define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS)
1530
1531extern int tcp_cookie_generator(u32 *bakery);
1532
1533/**
1534 *	struct tcp_cookie_values - each socket needs extra space for the
1535 *	cookies, together with (optional) space for any SYN data.
1536 *
1537 *	A tcp_sock contains a pointer to the current value, and this is
1538 *	cloned to the tcp_timewait_sock.
1539 *
1540 * @cookie_pair:	variable data from the option exchange.
1541 *
1542 * @cookie_desired:	user specified tcpct_cookie_desired.  Zero
1543 *			indicates default (sysctl_tcp_cookie_size).
1544 *			After cookie sent, remembers size of cookie.
1545 *			Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX.
1546 *
1547 * @s_data_desired:	user specified tcpct_s_data_desired.  When the
1548 *			constant payload is specified (@s_data_constant),
1549 *			holds its length instead.
1550 *			Range 0 to TCP_MSS_DESIRED.
1551 *
1552 * @s_data_payload:	constant data that is to be included in the
1553 *			payload of SYN or SYNACK segments when the
1554 *			cookie option is present.
1555 */
1556struct tcp_cookie_values {
1557	struct kref	kref;
1558	u8		cookie_pair[TCP_COOKIE_PAIR_SIZE];
1559	u8		cookie_pair_size;
1560	u8		cookie_desired;
1561	u16		s_data_desired:11,
1562			s_data_constant:1,
1563			s_data_in:1,
1564			s_data_out:1,
1565			s_data_unused:2;
1566	u8		s_data_payload[0];
1567};
1568
1569static inline void tcp_cookie_values_release(struct kref *kref)
1570{
1571	kfree(container_of(kref, struct tcp_cookie_values, kref));
1572}
1573
1574/* The length of constant payload data.  Note that s_data_desired is
1575 * overloaded, depending on s_data_constant: either the length of constant
1576 * data (returned here) or the limit on variable data.
1577 */
1578static inline int tcp_s_data_size(const struct tcp_sock *tp)
1579{
1580	return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant)
1581		? tp->cookie_values->s_data_desired
1582		: 0;
1583}
1584
1585/**
1586 *	struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace.
1587 *
1588 *	As tcp_request_sock has already been extended in other places, the
1589 *	only remaining method is to pass stack values along as function
1590 *	parameters.  These parameters are not needed after sending SYNACK.
1591 *
1592 * @cookie_bakery:	cryptographic secret and message workspace.
1593 *
1594 * @cookie_plus:	bytes in authenticator/cookie option, copied from
1595 *			struct tcp_options_received (above).
1596 */
1597struct tcp_extend_values {
1598	struct request_values		rv;
1599	u32				cookie_bakery[COOKIE_WORKSPACE_WORDS];
1600	u8				cookie_plus:6,
1601					cookie_out_never:1,
1602					cookie_in_always:1;
1603};
1604
1605static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp)
1606{
1607	return (struct tcp_extend_values *)rvp;
1608}
1609
1610extern void tcp_v4_init(void);
1611extern void tcp_init(void);
1612
1613#endif	/* _TCP_H */