Linux Audio

Check our new training course

Loading...
v6.9.4
   1/* SPDX-License-Identifier: GPL-2.0-or-later */
   2/*
   3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   4 *		operating system.  INET is implemented using the  BSD Socket
   5 *		interface as the means of communication with the user level.
   6 *
   7 *		Definitions for the TCP module.
   8 *
   9 * Version:	@(#)tcp.h	1.0.5	05/23/93
  10 *
  11 * Authors:	Ross Biro
  12 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
 
 
 
 
 
  13 */
  14#ifndef _TCP_H
  15#define _TCP_H
  16
  17#define FASTRETRANS_DEBUG 1
  18
  19#include <linux/list.h>
  20#include <linux/tcp.h>
  21#include <linux/bug.h>
  22#include <linux/slab.h>
  23#include <linux/cache.h>
  24#include <linux/percpu.h>
  25#include <linux/skbuff.h>
 
  26#include <linux/kref.h>
  27#include <linux/ktime.h>
  28#include <linux/indirect_call_wrapper.h>
  29
  30#include <net/inet_connection_sock.h>
  31#include <net/inet_timewait_sock.h>
  32#include <net/inet_hashtables.h>
  33#include <net/checksum.h>
  34#include <net/request_sock.h>
  35#include <net/sock_reuseport.h>
  36#include <net/sock.h>
  37#include <net/snmp.h>
  38#include <net/ip.h>
  39#include <net/tcp_states.h>
  40#include <net/tcp_ao.h>
  41#include <net/inet_ecn.h>
  42#include <net/dst.h>
  43#include <net/mptcp.h>
  44
  45#include <linux/seq_file.h>
  46#include <linux/memcontrol.h>
  47#include <linux/bpf-cgroup.h>
  48#include <linux/siphash.h>
  49
  50extern struct inet_hashinfo tcp_hashinfo;
  51
  52DECLARE_PER_CPU(unsigned int, tcp_orphan_count);
  53int tcp_orphan_count_sum(void);
  54
  55void tcp_time_wait(struct sock *sk, int state, int timeo);
  56
  57#define MAX_TCP_HEADER	L1_CACHE_ALIGN(128 + MAX_HEADER)
  58#define MAX_TCP_OPTION_SPACE 40
  59#define TCP_MIN_SND_MSS		48
  60#define TCP_MIN_GSO_SIZE	(TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
  61
  62/*
  63 * Never offer a window over 32767 without using window scaling. Some
  64 * poor stacks do signed 16bit maths!
  65 */
  66#define MAX_TCP_WINDOW		32767U
  67
  68/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
  69#define TCP_MIN_MSS		88U
  70
  71/* The initial MTU to use for probing */
  72#define TCP_BASE_MSS		1024
  73
  74/* probing interval, default to 10 minutes as per RFC4821 */
  75#define TCP_PROBE_INTERVAL	600
  76
  77/* Specify interval when tcp mtu probing will stop */
  78#define TCP_PROBE_THRESHOLD	8
  79
  80/* After receiving this amount of duplicate ACKs fast retransmit starts. */
  81#define TCP_FASTRETRANS_THRESH 3
  82
  83/* Maximal number of ACKs sent quickly to accelerate slow-start. */
  84#define TCP_MAX_QUICKACKS	16U
  85
  86/* Maximal number of window scale according to RFC1323 */
  87#define TCP_MAX_WSCALE		14U
  88
  89/* urg_data states */
  90#define TCP_URG_VALID	0x0100
  91#define TCP_URG_NOTYET	0x0200
  92#define TCP_URG_READ	0x0400
  93
  94#define TCP_RETR1	3	/*
  95				 * This is how many retries it does before it
  96				 * tries to figure out if the gateway is
  97				 * down. Minimal RFC value is 3; it corresponds
  98				 * to ~3sec-8min depending on RTO.
  99				 */
 100
 101#define TCP_RETR2	15	/*
 102				 * This should take at least
 103				 * 90 minutes to time out.
 104				 * RFC1122 says that the limit is 100 sec.
 105				 * 15 is ~13-30min depending on RTO.
 106				 */
 107
 108#define TCP_SYN_RETRIES	 6	/* This is how many retries are done
 109				 * when active opening a connection.
 110				 * RFC1122 says the minimum retry MUST
 111				 * be at least 180secs.  Nevertheless
 112				 * this value is corresponding to
 113				 * 63secs of retransmission with the
 114				 * current initial RTO.
 115				 */
 116
 117#define TCP_SYNACK_RETRIES 5	/* This is how may retries are done
 118				 * when passive opening a connection.
 119				 * This is corresponding to 31secs of
 120				 * retransmission with the current
 121				 * initial RTO.
 122				 */
 123
 124#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
 125				  * state, about 60 seconds	*/
 126#define TCP_FIN_TIMEOUT	TCP_TIMEWAIT_LEN
 127                                 /* BSD style FIN_WAIT2 deadlock breaker.
 128				  * It used to be 3min, new value is 60sec,
 129				  * to combine FIN-WAIT-2 timeout with
 130				  * TIME-WAIT timer.
 131				  */
 132#define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */
 133
 134#define TCP_DELACK_MAX	((unsigned)(HZ/5))	/* maximal time to delay before sending an ACK */
 135static_assert((1 << ATO_BITS) > TCP_DELACK_MAX);
 136
 137#if HZ >= 100
 138#define TCP_DELACK_MIN	((unsigned)(HZ/25))	/* minimal time to delay before sending an ACK */
 139#define TCP_ATO_MIN	((unsigned)(HZ/25))
 140#else
 141#define TCP_DELACK_MIN	4U
 142#define TCP_ATO_MIN	4U
 143#endif
 144#define TCP_RTO_MAX	((unsigned)(120*HZ))
 145#define TCP_RTO_MIN	((unsigned)(HZ/5))
 146#define TCP_TIMEOUT_MIN	(2U) /* Min timeout for TCP timers in jiffies */
 147
 148#define TCP_TIMEOUT_MIN_US (2*USEC_PER_MSEC) /* Min TCP timeout in microsecs */
 149
 150#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ))	/* RFC6298 2.1 initial RTO value	*/
 151#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ))	/* RFC 1122 initial RTO value, now
 152						 * used as a fallback RTO for the
 153						 * initial data transmission if no
 154						 * valid RTT sample has been acquired,
 155						 * most likely due to retrans in 3WHS.
 156						 */
 157
 158#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
 159					                 * for local resources.
 160					                 */
 
 161#define TCP_KEEPALIVE_TIME	(120*60*HZ)	/* two hours */
 162#define TCP_KEEPALIVE_PROBES	9		/* Max of 9 keepalive probes	*/
 163#define TCP_KEEPALIVE_INTVL	(75*HZ)
 164
 165#define MAX_TCP_KEEPIDLE	32767
 166#define MAX_TCP_KEEPINTVL	32767
 167#define MAX_TCP_KEEPCNT		127
 168#define MAX_TCP_SYNCNT		127
 169
 170/* Ensure that TCP PAWS checks are relaxed after ~2147 seconds
 171 * to avoid overflows. This assumes a clock smaller than 1 Mhz.
 172 * Default clock is 1 Khz, tcp_usec_ts uses 1 Mhz.
 173 */
 174#define TCP_PAWS_WRAP (INT_MAX / USEC_PER_SEC)
 175
 
 176#define TCP_PAWS_MSL	60		/* Per-host timestamps are invalidated
 177					 * after this time. It should be equal
 178					 * (or greater than) TCP_TIMEWAIT_LEN
 179					 * to provide reliability equal to one
 180					 * provided by timewait state.
 181					 */
 182#define TCP_PAWS_WINDOW	1		/* Replay window for per-host
 183					 * timestamps. It must be less than
 184					 * minimal timewait lifetime.
 185					 */
 186/*
 187 *	TCP option
 188 */
 189
 190#define TCPOPT_NOP		1	/* Padding */
 191#define TCPOPT_EOL		0	/* End of options */
 192#define TCPOPT_MSS		2	/* Segment size negotiating */
 193#define TCPOPT_WINDOW		3	/* Window scaling */
 194#define TCPOPT_SACK_PERM        4       /* SACK Permitted */
 195#define TCPOPT_SACK             5       /* SACK Block */
 196#define TCPOPT_TIMESTAMP	8	/* Better RTT estimations/PAWS */
 197#define TCPOPT_MD5SIG		19	/* MD5 Signature (RFC2385) */
 198#define TCPOPT_AO		29	/* Authentication Option (RFC5925) */
 199#define TCPOPT_MPTCP		30	/* Multipath TCP (RFC6824) */
 200#define TCPOPT_FASTOPEN		34	/* Fast open (RFC7413) */
 201#define TCPOPT_EXP		254	/* Experimental */
 202/* Magic number to be after the option value for sharing TCP
 203 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
 204 */
 205#define TCPOPT_FASTOPEN_MAGIC	0xF989
 206#define TCPOPT_SMC_MAGIC	0xE2D4C3D9
 207
 208/*
 209 *     TCP option lengths
 210 */
 211
 212#define TCPOLEN_MSS            4
 213#define TCPOLEN_WINDOW         3
 214#define TCPOLEN_SACK_PERM      2
 215#define TCPOLEN_TIMESTAMP      10
 216#define TCPOLEN_MD5SIG         18
 217#define TCPOLEN_FASTOPEN_BASE  2
 218#define TCPOLEN_EXP_FASTOPEN_BASE  4
 219#define TCPOLEN_EXP_SMC_BASE   6
 220
 221/* But this is what stacks really send out. */
 222#define TCPOLEN_TSTAMP_ALIGNED		12
 223#define TCPOLEN_WSCALE_ALIGNED		4
 224#define TCPOLEN_SACKPERM_ALIGNED	4
 225#define TCPOLEN_SACK_BASE		2
 226#define TCPOLEN_SACK_BASE_ALIGNED	4
 227#define TCPOLEN_SACK_PERBLOCK		8
 228#define TCPOLEN_MD5SIG_ALIGNED		20
 229#define TCPOLEN_MSS_ALIGNED		4
 230#define TCPOLEN_EXP_SMC_BASE_ALIGNED	8
 231
 232/* Flags in tp->nonagle */
 233#define TCP_NAGLE_OFF		1	/* Nagle's algo is disabled */
 234#define TCP_NAGLE_CORK		2	/* Socket is corked	    */
 235#define TCP_NAGLE_PUSH		4	/* Cork is overridden for already queued data */
 236
 237/* TCP thin-stream limits */
 238#define TCP_THIN_LINEAR_RETRIES 6       /* After 6 linear retries, do exp. backoff */
 239
 240/* TCP initial congestion window as per rfc6928 */
 241#define TCP_INIT_CWND		10
 242
 243/* Bit Flags for sysctl_tcp_fastopen */
 244#define	TFO_CLIENT_ENABLE	1
 245#define	TFO_SERVER_ENABLE	2
 246#define	TFO_CLIENT_NO_COOKIE	4	/* Data in SYN w/o cookie option */
 247
 248/* Accept SYN data w/o any cookie option */
 249#define	TFO_SERVER_COOKIE_NOT_REQD	0x200
 250
 251/* Force enable TFO on all listeners, i.e., not requiring the
 252 * TCP_FASTOPEN socket option.
 253 */
 254#define	TFO_SERVER_WO_SOCKOPT1	0x400
 255
 
 256
 257/* sysctl variables for tcp */
 
 
 
 
 
 
 
 
 258extern int sysctl_tcp_max_orphans;
 
 
 
 
 259extern long sysctl_tcp_mem[3];
 260
 261#define TCP_RACK_LOSS_DETECTION  0x1 /* Use RACK to detect losses */
 262#define TCP_RACK_STATIC_REO_WND  0x2 /* Use static RACK reo wnd */
 263#define TCP_RACK_NO_DUPTHRESH    0x4 /* Do not use DUPACK threshold in RACK */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 264
 265extern atomic_long_t tcp_memory_allocated;
 266DECLARE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc);
 267
 268extern struct percpu_counter tcp_sockets_allocated;
 269extern unsigned long tcp_memory_pressure;
 270
 271/* optimized version of sk_under_memory_pressure() for TCP sockets */
 272static inline bool tcp_under_memory_pressure(const struct sock *sk)
 273{
 274	if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
 275	    mem_cgroup_under_socket_pressure(sk->sk_memcg))
 276		return true;
 277
 278	return READ_ONCE(tcp_memory_pressure);
 279}
 280/*
 281 * The next routines deal with comparing 32 bit unsigned ints
 282 * and worry about wraparound (automatic with unsigned arithmetic).
 283 */
 284
 285static inline bool before(__u32 seq1, __u32 seq2)
 286{
 287        return (__s32)(seq1-seq2) < 0;
 288}
 289#define after(seq2, seq1) 	before(seq1, seq2)
 290
 291/* is s2<=s1<=s3 ? */
 292static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
 293{
 294	return seq3 - seq2 >= seq1 - seq2;
 295}
 296
 297static inline bool tcp_out_of_memory(struct sock *sk)
 298{
 299	if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
 300	    sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
 301		return true;
 302	return false;
 303}
 304
 305static inline void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
 
 
 306{
 307	sk_wmem_queued_add(sk, -skb->truesize);
 308	if (!skb_zcopy_pure(skb))
 309		sk_mem_uncharge(sk, skb->truesize);
 310	else
 311		sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb)));
 312	__kfree_skb(skb);
 313}
 314
 315void sk_forced_mem_schedule(struct sock *sk, int size);
 
 
 
 
 
 
 316
 317bool tcp_check_oom(struct sock *sk, int shift);
 318
 319
 320extern struct proto tcp_prot;
 321
 322#define TCP_INC_STATS(net, field)	SNMP_INC_STATS((net)->mib.tcp_statistics, field)
 323#define __TCP_INC_STATS(net, field)	__SNMP_INC_STATS((net)->mib.tcp_statistics, field)
 324#define TCP_DEC_STATS(net, field)	SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
 325#define TCP_ADD_STATS(net, field, val)	SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
 326
 327void tcp_tasklet_init(void);
 328
 329int tcp_v4_err(struct sk_buff *skb, u32);
 330
 331void tcp_shutdown(struct sock *sk, int how);
 332
 333int tcp_v4_early_demux(struct sk_buff *skb);
 334int tcp_v4_rcv(struct sk_buff *skb);
 335
 336void tcp_remove_empty_skb(struct sock *sk);
 337int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
 338int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
 339int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *copied,
 340			 size_t size, struct ubuf_info *uarg);
 341void tcp_splice_eof(struct socket *sock);
 342int tcp_send_mss(struct sock *sk, int *size_goal, int flags);
 343int tcp_wmem_schedule(struct sock *sk, int copy);
 344void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle,
 345	      int size_goal);
 346void tcp_release_cb(struct sock *sk);
 347void tcp_wfree(struct sk_buff *skb);
 348void tcp_write_timer_handler(struct sock *sk);
 349void tcp_delack_timer_handler(struct sock *sk);
 350int tcp_ioctl(struct sock *sk, int cmd, int *karg);
 351enum skb_drop_reason tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
 352void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
 
 353void tcp_rcv_space_adjust(struct sock *sk);
 354int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
 355void tcp_twsk_destructor(struct sock *sk);
 356void tcp_twsk_purge(struct list_head *net_exit_list, int family);
 357ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
 358			struct pipe_inode_info *pipe, size_t len,
 359			unsigned int flags);
 360struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, gfp_t gfp,
 361				     bool force_schedule);
 362
 363static inline void tcp_dec_quickack_mode(struct sock *sk)
 
 364{
 365	struct inet_connection_sock *icsk = inet_csk(sk);
 366
 367	if (icsk->icsk_ack.quick) {
 368		/* How many ACKs S/ACKing new data have we sent? */
 369		const unsigned int pkts = inet_csk_ack_scheduled(sk) ? 1 : 0;
 370
 371		if (pkts >= icsk->icsk_ack.quick) {
 372			icsk->icsk_ack.quick = 0;
 373			/* Leaving quickack mode we deflate ATO. */
 374			icsk->icsk_ack.ato   = TCP_ATO_MIN;
 375		} else
 376			icsk->icsk_ack.quick -= pkts;
 377	}
 378}
 379
 380#define	TCP_ECN_OK		1
 381#define	TCP_ECN_QUEUE_CWR	2
 382#define	TCP_ECN_DEMAND_CWR	4
 383#define	TCP_ECN_SEEN		8
 384
 385enum tcp_tw_status {
 386	TCP_TW_SUCCESS = 0,
 387	TCP_TW_RST = 1,
 388	TCP_TW_ACK = 2,
 389	TCP_TW_SYN = 3
 390};
 391
 392
 393enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
 394					      struct sk_buff *skb,
 395					      const struct tcphdr *th);
 396struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
 397			   struct request_sock *req, bool fastopen,
 398			   bool *lost_race);
 399enum skb_drop_reason tcp_child_process(struct sock *parent, struct sock *child,
 400				       struct sk_buff *skb);
 401void tcp_enter_loss(struct sock *sk);
 402void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag);
 403void tcp_clear_retrans(struct tcp_sock *tp);
 404void tcp_update_metrics(struct sock *sk);
 405void tcp_init_metrics(struct sock *sk);
 406void tcp_metrics_init(void);
 407bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
 408void __tcp_close(struct sock *sk, long timeout);
 
 
 
 
 409void tcp_close(struct sock *sk, long timeout);
 410void tcp_init_sock(struct sock *sk);
 411void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb);
 412__poll_t tcp_poll(struct file *file, struct socket *sock,
 413		      struct poll_table_struct *wait);
 414int do_tcp_getsockopt(struct sock *sk, int level,
 415		      int optname, sockptr_t optval, sockptr_t optlen);
 416int tcp_getsockopt(struct sock *sk, int level, int optname,
 417		   char __user *optval, int __user *optlen);
 418bool tcp_bpf_bypass_getsockopt(int level, int optname);
 419int do_tcp_setsockopt(struct sock *sk, int level, int optname,
 420		      sockptr_t optval, unsigned int optlen);
 421int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
 422		   unsigned int optlen);
 
 423void tcp_set_keepalive(struct sock *sk, int val);
 424void tcp_syn_ack_timeout(const struct request_sock *req);
 425int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
 426		int flags, int *addr_len);
 427int tcp_set_rcvlowat(struct sock *sk, int val);
 428int tcp_set_window_clamp(struct sock *sk, int val);
 429void tcp_update_recv_tstamps(struct sk_buff *skb,
 430			     struct scm_timestamping_internal *tss);
 431void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk,
 432			struct scm_timestamping_internal *tss);
 433void tcp_data_ready(struct sock *sk);
 434#ifdef CONFIG_MMU
 435int tcp_mmap(struct file *file, struct socket *sock,
 436	     struct vm_area_struct *vma);
 437#endif
 438void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
 439		       struct tcp_options_received *opt_rx,
 440		       int estab, struct tcp_fastopen_cookie *foc);
 
 441
 442/*
 443 *	BPF SKB-less helpers
 444 */
 445u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
 446			 struct tcphdr *th, u32 *cookie);
 447u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
 448			 struct tcphdr *th, u32 *cookie);
 449u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss);
 450u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops,
 451			  const struct tcp_request_sock_ops *af_ops,
 452			  struct sock *sk, struct tcphdr *th);
 453/*
 454 *	TCP v4 functions exported for the inet6 API
 455 */
 456
 457void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
 458void tcp_v4_mtu_reduced(struct sock *sk);
 459void tcp_req_err(struct sock *sk, u32 seq, bool abort);
 460void tcp_ld_RTO_revert(struct sock *sk, u32 seq);
 461int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
 462struct sock *tcp_create_openreq_child(const struct sock *sk,
 463				      struct request_sock *req,
 464				      struct sk_buff *skb);
 465void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
 466struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
 467				  struct request_sock *req,
 468				  struct dst_entry *dst,
 469				  struct request_sock *req_unhash,
 470				  bool *own_req);
 471int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
 472int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
 473int tcp_connect(struct sock *sk);
 474enum tcp_synack_type {
 475	TCP_SYNACK_NORMAL,
 476	TCP_SYNACK_FASTOPEN,
 477	TCP_SYNACK_COOKIE,
 478};
 479struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
 480				struct request_sock *req,
 481				struct tcp_fastopen_cookie *foc,
 482				enum tcp_synack_type synack_type,
 483				struct sk_buff *syn_skb);
 484int tcp_disconnect(struct sock *sk, int flags);
 485
 486void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
 487int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
 488void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
 489
 490/* From syncookies.c */
 491struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
 492				 struct request_sock *req,
 493				 struct dst_entry *dst);
 494int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th);
 
 495struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
 496struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
 497					    struct sock *sk, struct sk_buff *skb,
 498					    struct tcp_options_received *tcp_opt,
 499					    int mss, u32 tsoff);
 500
 501#if IS_ENABLED(CONFIG_BPF)
 502struct bpf_tcp_req_attrs {
 503	u32 rcv_tsval;
 504	u32 rcv_tsecr;
 505	u16 mss;
 506	u8 rcv_wscale;
 507	u8 snd_wscale;
 508	u8 ecn_ok;
 509	u8 wscale_ok;
 510	u8 sack_ok;
 511	u8 tstamp_ok;
 512	u8 usec_ts_ok;
 513	u8 reserved[3];
 514};
 515#endif
 516
 517#ifdef CONFIG_SYN_COOKIES
 518
 519/* Syncookies use a monotonic timer which increments every 60 seconds.
 520 * This counter is used both as a hash input and partially encoded into
 521 * the cookie value.  A cookie is only validated further if the delta
 522 * between the current counter value and the encoded one is less than this,
 523 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
 524 * the counter advances immediately after a cookie is generated).
 525 */
 526#define MAX_SYNCOOKIE_AGE	2
 527#define TCP_SYNCOOKIE_PERIOD	(60 * HZ)
 528#define TCP_SYNCOOKIE_VALID	(MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
 529
 530/* syncookies: remember time of last synqueue overflow
 531 * But do not dirty this field too often (once per second is enough)
 532 * It is racy as we do not hold a lock, but race is very minor.
 533 */
 534static inline void tcp_synq_overflow(const struct sock *sk)
 535{
 536	unsigned int last_overflow;
 537	unsigned int now = jiffies;
 538
 539	if (sk->sk_reuseport) {
 540		struct sock_reuseport *reuse;
 541
 542		reuse = rcu_dereference(sk->sk_reuseport_cb);
 543		if (likely(reuse)) {
 544			last_overflow = READ_ONCE(reuse->synq_overflow_ts);
 545			if (!time_between32(now, last_overflow,
 546					    last_overflow + HZ))
 547				WRITE_ONCE(reuse->synq_overflow_ts, now);
 548			return;
 549		}
 550	}
 551
 552	last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
 553	if (!time_between32(now, last_overflow, last_overflow + HZ))
 554		WRITE_ONCE(tcp_sk_rw(sk)->rx_opt.ts_recent_stamp, now);
 555}
 556
 557/* syncookies: no recent synqueue overflow on this listening socket? */
 558static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
 559{
 560	unsigned int last_overflow;
 561	unsigned int now = jiffies;
 562
 563	if (sk->sk_reuseport) {
 564		struct sock_reuseport *reuse;
 565
 566		reuse = rcu_dereference(sk->sk_reuseport_cb);
 567		if (likely(reuse)) {
 568			last_overflow = READ_ONCE(reuse->synq_overflow_ts);
 569			return !time_between32(now, last_overflow - HZ,
 570					       last_overflow +
 571					       TCP_SYNCOOKIE_VALID);
 572		}
 573	}
 574
 575	last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
 576
 577	/* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
 578	 * then we're under synflood. However, we have to use
 579	 * 'last_overflow - HZ' as lower bound. That's because a concurrent
 580	 * tcp_synq_overflow() could update .ts_recent_stamp after we read
 581	 * jiffies but before we store .ts_recent_stamp into last_overflow,
 582	 * which could lead to rejecting a valid syncookie.
 583	 */
 584	return !time_between32(now, last_overflow - HZ,
 585			       last_overflow + TCP_SYNCOOKIE_VALID);
 586}
 587
 588static inline u32 tcp_cookie_time(void)
 589{
 590	u64 val = get_jiffies_64();
 591
 592	do_div(val, TCP_SYNCOOKIE_PERIOD);
 593	return val;
 594}
 595
 596/* Convert one nsec 64bit timestamp to ts (ms or usec resolution) */
 597static inline u64 tcp_ns_to_ts(bool usec_ts, u64 val)
 598{
 599	if (usec_ts)
 600		return div_u64(val, NSEC_PER_USEC);
 601
 602	return div_u64(val, NSEC_PER_MSEC);
 603}
 604
 605u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
 606			      u16 *mssp);
 607__u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
 608u64 cookie_init_timestamp(struct request_sock *req, u64 now);
 609bool cookie_timestamp_decode(const struct net *net,
 610			     struct tcp_options_received *opt);
 611
 612static inline bool cookie_ecn_ok(const struct net *net, const struct dst_entry *dst)
 613{
 614	return READ_ONCE(net->ipv4.sysctl_tcp_ecn) ||
 615		dst_feature(dst, RTAX_FEATURE_ECN);
 616}
 617
 618#if IS_ENABLED(CONFIG_BPF)
 619static inline bool cookie_bpf_ok(struct sk_buff *skb)
 620{
 621	return skb->sk;
 622}
 623
 624struct request_sock *cookie_bpf_check(struct sock *sk, struct sk_buff *skb);
 625#else
 626static inline bool cookie_bpf_ok(struct sk_buff *skb)
 627{
 628	return false;
 629}
 630
 631static inline struct request_sock *cookie_bpf_check(struct net *net, struct sock *sk,
 632						    struct sk_buff *skb)
 633{
 634	return NULL;
 635}
 636#endif
 637
 638/* From net/ipv6/syncookies.c */
 639int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th);
 
 640struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
 641
 642u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
 643			      const struct tcphdr *th, u16 *mssp);
 644__u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
 645#endif
 646/* tcp_output.c */
 647
 648void tcp_skb_entail(struct sock *sk, struct sk_buff *skb);
 649void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb);
 650void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
 651			       int nonagle);
 
 652int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
 653int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
 654void tcp_retransmit_timer(struct sock *sk);
 655void tcp_xmit_retransmit_queue(struct sock *);
 656void tcp_simple_retransmit(struct sock *);
 657void tcp_enter_recovery(struct sock *sk, bool ece_ack);
 658int tcp_trim_head(struct sock *, struct sk_buff *, u32);
 659enum tcp_queue {
 660	TCP_FRAG_IN_WRITE_QUEUE,
 661	TCP_FRAG_IN_RTX_QUEUE,
 662};
 663int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
 664		 struct sk_buff *skb, u32 len,
 665		 unsigned int mss_now, gfp_t gfp);
 666
 667void tcp_send_probe0(struct sock *);
 
 668int tcp_write_wakeup(struct sock *, int mib);
 669void tcp_send_fin(struct sock *sk);
 670void tcp_send_active_reset(struct sock *sk, gfp_t priority);
 671int tcp_send_synack(struct sock *);
 672void tcp_push_one(struct sock *, unsigned int mss_now);
 673void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
 674void tcp_send_ack(struct sock *sk);
 675void tcp_send_delayed_ack(struct sock *sk);
 676void tcp_send_loss_probe(struct sock *sk);
 677bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
 678void tcp_skb_collapse_tstamp(struct sk_buff *skb,
 679			     const struct sk_buff *next_skb);
 680
 681/* tcp_input.c */
 
 682void tcp_rearm_rto(struct sock *sk);
 683void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
 684void tcp_reset(struct sock *sk, struct sk_buff *skb);
 
 685void tcp_fin(struct sock *sk);
 686void tcp_check_space(struct sock *sk);
 687void tcp_sack_compress_send_ack(struct sock *sk);
 688
 689/* tcp_timer.c */
 690void tcp_init_xmit_timers(struct sock *);
 691static inline void tcp_clear_xmit_timers(struct sock *sk)
 692{
 693	if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
 694		__sock_put(sk);
 695
 696	if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
 697		__sock_put(sk);
 698
 699	inet_csk_clear_xmit_timers(sk);
 700}
 701
 702unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
 703unsigned int tcp_current_mss(struct sock *sk);
 704u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when);
 705
 706/* Bound MSS / TSO packet size with the half of the window */
 707static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
 708{
 709	int cutoff;
 710
 711	/* When peer uses tiny windows, there is no use in packetizing
 712	 * to sub-MSS pieces for the sake of SWS or making sure there
 713	 * are enough packets in the pipe for fast recovery.
 714	 *
 715	 * On the other hand, for extremely large MSS devices, handling
 716	 * smaller than MSS windows in this way does make sense.
 717	 */
 718	if (tp->max_window > TCP_MSS_DEFAULT)
 719		cutoff = (tp->max_window >> 1);
 720	else
 721		cutoff = tp->max_window;
 722
 723	if (cutoff && pktsize > cutoff)
 724		return max_t(int, cutoff, 68U - tp->tcp_header_len);
 725	else
 726		return pktsize;
 727}
 728
 729/* tcp.c */
 730void tcp_get_info(struct sock *, struct tcp_info *);
 731
 732/* Read 'sendfile()'-style from a TCP socket */
 733int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
 734		  sk_read_actor_t recv_actor);
 735int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor);
 736struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off);
 737void tcp_read_done(struct sock *sk, size_t len);
 738
 739void tcp_initialize_rcv_mss(struct sock *sk);
 740
 741int tcp_mtu_to_mss(struct sock *sk, int pmtu);
 742int tcp_mss_to_mtu(struct sock *sk, int mss);
 743void tcp_mtup_init(struct sock *sk);
 
 744
 745static inline void tcp_bound_rto(const struct sock *sk)
 746{
 747	if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
 748		inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
 749}
 750
 751static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
 752{
 753	return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
 754}
 755
 756static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
 757{
 758	/* mptcp hooks are only on the slow path */
 759	if (sk_is_mptcp((struct sock *)tp))
 760		return;
 761
 762	tp->pred_flags = htonl((tp->tcp_header_len << 26) |
 763			       ntohl(TCP_FLAG_ACK) |
 764			       snd_wnd);
 765}
 766
 767static inline void tcp_fast_path_on(struct tcp_sock *tp)
 768{
 769	__tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
 770}
 771
 772static inline void tcp_fast_path_check(struct sock *sk)
 773{
 774	struct tcp_sock *tp = tcp_sk(sk);
 775
 776	if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
 777	    tp->rcv_wnd &&
 778	    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
 779	    !tp->urg_data)
 780		tcp_fast_path_on(tp);
 781}
 782
 783u32 tcp_delack_max(const struct sock *sk);
 784
 785/* Compute the actual rto_min value */
 786static inline u32 tcp_rto_min(const struct sock *sk)
 787{
 788	const struct dst_entry *dst = __sk_dst_get(sk);
 789	u32 rto_min = inet_csk(sk)->icsk_rto_min;
 790
 791	if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
 792		rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
 793	return rto_min;
 794}
 795
 796static inline u32 tcp_rto_min_us(const struct sock *sk)
 797{
 798	return jiffies_to_usecs(tcp_rto_min(sk));
 799}
 800
 801static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
 802{
 803	return dst_metric_locked(dst, RTAX_CC_ALGO);
 804}
 805
 806/* Minimum RTT in usec. ~0 means not available. */
 807static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
 808{
 809	return minmax_get(&tp->rtt_min);
 810}
 811
 812/* Compute the actual receive window we are currently advertising.
 813 * Rcv_nxt can be after the window if our peer push more data
 814 * than the offered window.
 815 */
 816static inline u32 tcp_receive_window(const struct tcp_sock *tp)
 817{
 818	s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
 819
 820	if (win < 0)
 821		win = 0;
 822	return (u32) win;
 823}
 824
 825/* Choose a new window, without checks for shrinking, and without
 826 * scaling applied to the result.  The caller does these things
 827 * if necessary.  This is a "raw" window selection.
 828 */
 829u32 __tcp_select_window(struct sock *sk);
 830
 831void tcp_send_window_probe(struct sock *sk);
 832
 833/* TCP uses 32bit jiffies to save some space.
 834 * Note that this is different from tcp_time_stamp, which
 835 * historically has been the same until linux-4.13.
 
 
 836 */
 837#define tcp_jiffies32 ((u32)jiffies)
 838
 839/*
 840 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
 841 * It is no longer tied to jiffies, but to 1 ms clock.
 842 * Note: double check if you want to use tcp_jiffies32 instead of this.
 843 */
 844#define TCP_TS_HZ	1000
 845
 846static inline u64 tcp_clock_ns(void)
 847{
 848	return ktime_get_ns();
 849}
 850
 851static inline u64 tcp_clock_us(void)
 852{
 853	return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
 854}
 855
 856static inline u64 tcp_clock_ms(void)
 857{
 858	return div_u64(tcp_clock_ns(), NSEC_PER_MSEC);
 859}
 860
 861/* TCP Timestamp included in TS option (RFC 1323) can either use ms
 862 * or usec resolution. Each socket carries a flag to select one or other
 863 * resolution, as the route attribute could change anytime.
 864 * Each flow must stick to initial resolution.
 865 */
 866static inline u32 tcp_clock_ts(bool usec_ts)
 867{
 868	return usec_ts ? tcp_clock_us() : tcp_clock_ms();
 869}
 870
 871static inline u32 tcp_time_stamp_ms(const struct tcp_sock *tp)
 872{
 873	return div_u64(tp->tcp_mstamp, USEC_PER_MSEC);
 874}
 875
 876static inline u32 tcp_time_stamp_ts(const struct tcp_sock *tp)
 877{
 878	if (tp->tcp_usec_ts)
 879		return tp->tcp_mstamp;
 880	return tcp_time_stamp_ms(tp);
 881}
 882
 883void tcp_mstamp_refresh(struct tcp_sock *tp);
 884
 885static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
 886{
 887	return max_t(s64, t1 - t0, 0);
 888}
 889
 890/* provide the departure time in us unit */
 891static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
 892{
 893	return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
 894}
 895
 896/* Provide skb TSval in usec or ms unit */
 897static inline u32 tcp_skb_timestamp_ts(bool usec_ts, const struct sk_buff *skb)
 898{
 899	if (usec_ts)
 900		return tcp_skb_timestamp_us(skb);
 901
 902	return div_u64(skb->skb_mstamp_ns, NSEC_PER_MSEC);
 903}
 904
 905static inline u32 tcp_tw_tsval(const struct tcp_timewait_sock *tcptw)
 906{
 907	return tcp_clock_ts(tcptw->tw_sk.tw_usec_ts) + tcptw->tw_ts_offset;
 908}
 909
 910static inline u32 tcp_rsk_tsval(const struct tcp_request_sock *treq)
 911{
 912	return tcp_clock_ts(treq->req_usec_ts) + treq->ts_off;
 913}
 914
 915#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
 916
 917#define TCPHDR_FIN 0x01
 918#define TCPHDR_SYN 0x02
 919#define TCPHDR_RST 0x04
 920#define TCPHDR_PSH 0x08
 921#define TCPHDR_ACK 0x10
 922#define TCPHDR_URG 0x20
 923#define TCPHDR_ECE 0x40
 924#define TCPHDR_CWR 0x80
 925
 926#define TCPHDR_SYN_ECN	(TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
 927
 928/* This is what the send packet queuing engine uses to pass
 929 * TCP per-packet control information to the transmission code.
 930 * We also store the host-order sequence numbers in here too.
 931 * This is 44 bytes if IPV6 is enabled.
 932 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
 933 */
 934struct tcp_skb_cb {
 935	__u32		seq;		/* Starting sequence number	*/
 936	__u32		end_seq;	/* SEQ + FIN + SYN + datalen	*/
 937	union {
 938		/* Note : tcp_tw_isn is used in input path only
 939		 *	  (isn chosen by tcp_timewait_state_process())
 940		 *
 941		 * 	  tcp_gso_segs/size are used in write queue only,
 942		 *	  cf tcp_skb_pcount()/tcp_skb_mss()
 943		 */
 944		__u32		tcp_tw_isn;
 945		struct {
 946			u16	tcp_gso_segs;
 947			u16	tcp_gso_size;
 948		};
 949	};
 950	__u8		tcp_flags;	/* TCP header flags. (tcp[13])	*/
 951
 952	__u8		sacked;		/* State flags for SACK.	*/
 953#define TCPCB_SACKED_ACKED	0x01	/* SKB ACK'd by a SACK block	*/
 954#define TCPCB_SACKED_RETRANS	0x02	/* SKB retransmitted		*/
 955#define TCPCB_LOST		0x04	/* SKB is lost			*/
 956#define TCPCB_TAGBITS		0x07	/* All tag bits			*/
 957#define TCPCB_REPAIRED		0x10	/* SKB repaired (no skb_mstamp_ns)	*/
 958#define TCPCB_EVER_RETRANS	0x80	/* Ever retransmitted frame	*/
 959#define TCPCB_RETRANS		(TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
 960				TCPCB_REPAIRED)
 961
 962	__u8		ip_dsfield;	/* IPv4 tos or IPv6 dsfield	*/
 963	__u8		txstamp_ack:1,	/* Record TX timestamp for ack? */
 964			eor:1,		/* Is skb MSG_EOR marked? */
 965			has_rxtstamp:1,	/* SKB has a RX timestamp	*/
 966			unused:5;
 967	__u32		ack_seq;	/* Sequence number ACK'd	*/
 968	union {
 969		struct {
 970#define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1)
 971			/* There is space for up to 24 bytes */
 972			__u32 is_app_limited:1, /* cwnd not fully used? */
 973			      delivered_ce:20,
 974			      unused:11;
 975			/* pkts S/ACKed so far upon tx of skb, incl retrans: */
 976			__u32 delivered;
 977			/* start of send pipeline phase */
 978			u64 first_tx_mstamp;
 979			/* when we reached the "delivered" count */
 980			u64 delivered_mstamp;
 981		} tx;   /* only used for outgoing skbs */
 982		union {
 983			struct inet_skb_parm	h4;
 984#if IS_ENABLED(CONFIG_IPV6)
 985			struct inet6_skb_parm	h6;
 986#endif
 987		} header;	/* For incoming skbs */
 988	};
 989};
 990
 991#define TCP_SKB_CB(__skb)	((struct tcp_skb_cb *)&((__skb)->cb[0]))
 992
 993extern const struct inet_connection_sock_af_ops ipv4_specific;
 994
 995#if IS_ENABLED(CONFIG_IPV6)
 996/* This is the variant of inet6_iif() that must be used by TCP,
 997 * as TCP moves IP6CB into a different location in skb->cb[]
 998 */
 999static inline int tcp_v6_iif(const struct sk_buff *skb)
1000{
1001	return TCP_SKB_CB(skb)->header.h6.iif;
1002}
1003
1004static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
1005{
1006	bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
1007
1008	return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
1009}
1010
1011/* TCP_SKB_CB reference means this can not be used from early demux */
1012static inline int tcp_v6_sdif(const struct sk_buff *skb)
1013{
1014#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
1015	if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
1016		return TCP_SKB_CB(skb)->header.h6.iif;
1017#endif
1018	return 0;
1019}
1020
1021extern const struct inet_connection_sock_af_ops ipv6_specific;
1022
1023INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb));
1024INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb));
1025void tcp_v6_early_demux(struct sk_buff *skb);
1026
1027#endif
1028
1029/* TCP_SKB_CB reference means this can not be used from early demux */
1030static inline int tcp_v4_sdif(struct sk_buff *skb)
1031{
1032#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
1033	if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
1034		return TCP_SKB_CB(skb)->header.h4.iif;
 
1035#endif
1036	return 0;
1037}
1038
1039/* Due to TSO, an SKB can be composed of multiple actual
1040 * packets.  To keep these tracked properly, we use this.
1041 */
1042static inline int tcp_skb_pcount(const struct sk_buff *skb)
1043{
1044	return TCP_SKB_CB(skb)->tcp_gso_segs;
1045}
1046
1047static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
1048{
1049	TCP_SKB_CB(skb)->tcp_gso_segs = segs;
1050}
1051
1052static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
1053{
1054	TCP_SKB_CB(skb)->tcp_gso_segs += segs;
1055}
1056
1057/* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
1058static inline int tcp_skb_mss(const struct sk_buff *skb)
1059{
1060	return TCP_SKB_CB(skb)->tcp_gso_size;
1061}
1062
1063static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
1064{
1065	return likely(!TCP_SKB_CB(skb)->eor);
1066}
1067
1068static inline bool tcp_skb_can_collapse(const struct sk_buff *to,
1069					const struct sk_buff *from)
1070{
1071	return likely(tcp_skb_can_collapse_to(to) &&
1072		      mptcp_skb_can_collapse(to, from) &&
1073		      skb_pure_zcopy_same(to, from));
1074}
1075
1076/* Events passed to congestion control interface */
1077enum tcp_ca_event {
1078	CA_EVENT_TX_START,	/* first transmit when no packets in flight */
1079	CA_EVENT_CWND_RESTART,	/* congestion window restart */
1080	CA_EVENT_COMPLETE_CWR,	/* end of congestion recovery */
1081	CA_EVENT_LOSS,		/* loss timeout */
1082	CA_EVENT_ECN_NO_CE,	/* ECT set, but not CE marked */
1083	CA_EVENT_ECN_IS_CE,	/* received CE marked IP packet */
 
 
1084};
1085
1086/* Information about inbound ACK, passed to cong_ops->in_ack_event() */
1087enum tcp_ca_ack_event_flags {
1088	CA_ACK_SLOWPATH		= (1 << 0),	/* In slow path processing */
1089	CA_ACK_WIN_UPDATE	= (1 << 1),	/* ACK updated window */
1090	CA_ACK_ECE		= (1 << 2),	/* ECE bit is set on ack */
1091};
1092
1093/*
1094 * Interface for adding new TCP congestion control handlers
1095 */
1096#define TCP_CA_NAME_MAX	16
1097#define TCP_CA_MAX	128
1098#define TCP_CA_BUF_MAX	(TCP_CA_NAME_MAX*TCP_CA_MAX)
1099
1100#define TCP_CA_UNSPEC	0
1101
1102/* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
1103#define TCP_CONG_NON_RESTRICTED 0x1
1104/* Requires ECN/ECT set on all packets */
1105#define TCP_CONG_NEEDS_ECN	0x2
1106#define TCP_CONG_MASK	(TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN)
1107
1108union tcp_cc_info;
1109
1110struct ack_sample {
1111	u32 pkts_acked;
1112	s32 rtt_us;
1113	u32 in_flight;
1114};
1115
1116/* A rate sample measures the number of (original/retransmitted) data
1117 * packets delivered "delivered" over an interval of time "interval_us".
1118 * The tcp_rate.c code fills in the rate sample, and congestion
1119 * control modules that define a cong_control function to run at the end
1120 * of ACK processing can optionally chose to consult this sample when
1121 * setting cwnd and pacing rate.
1122 * A sample is invalid if "delivered" or "interval_us" is negative.
1123 */
1124struct rate_sample {
1125	u64  prior_mstamp; /* starting timestamp for interval */
1126	u32  prior_delivered;	/* tp->delivered at "prior_mstamp" */
1127	u32  prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */
1128	s32  delivered;		/* number of packets delivered over interval */
1129	s32  delivered_ce;	/* number of packets delivered w/ CE marks*/
1130	long interval_us;	/* time for tp->delivered to incr "delivered" */
1131	u32 snd_interval_us;	/* snd interval for delivered packets */
1132	u32 rcv_interval_us;	/* rcv interval for delivered packets */
1133	long rtt_us;		/* RTT of last (S)ACKed packet (or -1) */
1134	int  losses;		/* number of packets marked lost upon ACK */
1135	u32  acked_sacked;	/* number of packets newly (S)ACKed upon ACK */
1136	u32  prior_in_flight;	/* in flight before this ACK */
1137	u32  last_end_seq;	/* end_seq of most recently ACKed packet */
1138	bool is_app_limited;	/* is sample from packet with bubble in pipe? */
1139	bool is_retrans;	/* is sample from retransmission? */
1140	bool is_ack_delayed;	/* is this (likely) a delayed ACK? */
1141};
1142
1143struct tcp_congestion_ops {
1144/* fast path fields are put first to fill one cache line */
 
 
 
 
 
 
 
1145
1146	/* return slow start threshold (required) */
1147	u32 (*ssthresh)(struct sock *sk);
1148
1149	/* do new cwnd calculation (required) */
1150	void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1151
1152	/* call before changing ca_state (optional) */
1153	void (*set_state)(struct sock *sk, u8 new_state);
1154
1155	/* call when cwnd event occurs (optional) */
1156	void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1157
1158	/* call when ack arrives (optional) */
1159	void (*in_ack_event)(struct sock *sk, u32 flags);
1160
 
1161	/* hook for packet ack accounting (optional) */
1162	void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1163
1164	/* override sysctl_tcp_min_tso_segs */
1165	u32 (*min_tso_segs)(struct sock *sk);
1166
1167	/* call when packets are delivered to update cwnd and pacing rate,
1168	 * after all the ca_state processing. (optional)
1169	 */
1170	void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1171
1172
1173	/* new value of cwnd after loss (required) */
1174	u32  (*undo_cwnd)(struct sock *sk);
1175	/* returns the multiplier used in tcp_sndbuf_expand (optional) */
1176	u32 (*sndbuf_expand)(struct sock *sk);
1177
1178/* control/slow paths put last */
1179	/* get info for inet_diag (optional) */
1180	size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1181			   union tcp_cc_info *info);
1182
1183	char 			name[TCP_CA_NAME_MAX];
1184	struct module		*owner;
1185	struct list_head	list;
1186	u32			key;
1187	u32			flags;
1188
1189	/* initialize private data (optional) */
1190	void (*init)(struct sock *sk);
1191	/* cleanup private data  (optional) */
1192	void (*release)(struct sock *sk);
1193} ____cacheline_aligned_in_smp;
1194
1195int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1196void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1197int tcp_update_congestion_control(struct tcp_congestion_ops *type,
1198				  struct tcp_congestion_ops *old_type);
1199int tcp_validate_congestion_control(struct tcp_congestion_ops *ca);
1200
1201void tcp_assign_congestion_control(struct sock *sk);
1202void tcp_init_congestion_control(struct sock *sk);
1203void tcp_cleanup_congestion_control(struct sock *sk);
1204int tcp_set_default_congestion_control(struct net *net, const char *name);
1205void tcp_get_default_congestion_control(struct net *net, char *name);
1206void tcp_get_available_congestion_control(char *buf, size_t len);
1207void tcp_get_allowed_congestion_control(char *buf, size_t len);
1208int tcp_set_allowed_congestion_control(char *allowed);
1209int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
1210			       bool cap_net_admin);
1211u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1212void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1213
1214u32 tcp_reno_ssthresh(struct sock *sk);
1215u32 tcp_reno_undo_cwnd(struct sock *sk);
1216void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1217extern struct tcp_congestion_ops tcp_reno;
1218
1219struct tcp_congestion_ops *tcp_ca_find(const char *name);
1220struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1221u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1222#ifdef CONFIG_INET
1223char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1224#else
1225static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1226{
1227	return NULL;
1228}
1229#endif
1230
1231static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1232{
1233	const struct inet_connection_sock *icsk = inet_csk(sk);
1234
1235	return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1236}
1237
 
 
 
 
 
 
 
 
 
1238static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1239{
1240	const struct inet_connection_sock *icsk = inet_csk(sk);
1241
1242	if (icsk->icsk_ca_ops->cwnd_event)
1243		icsk->icsk_ca_ops->cwnd_event(sk, event);
1244}
1245
1246/* From tcp_cong.c */
1247void tcp_set_ca_state(struct sock *sk, const u8 ca_state);
1248
1249/* From tcp_rate.c */
1250void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1251void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1252			    struct rate_sample *rs);
1253void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1254		  bool is_sack_reneg, struct rate_sample *rs);
1255void tcp_rate_check_app_limited(struct sock *sk);
1256
1257static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2)
1258{
1259	return t1 > t2 || (t1 == t2 && after(seq1, seq2));
1260}
1261
1262/* These functions determine how the current flow behaves in respect of SACK
1263 * handling. SACK is negotiated with the peer, and therefore it can vary
1264 * between different flows.
1265 *
1266 * tcp_is_sack - SACK enabled
1267 * tcp_is_reno - No SACK
 
1268 */
1269static inline int tcp_is_sack(const struct tcp_sock *tp)
1270{
1271	return likely(tp->rx_opt.sack_ok);
1272}
1273
1274static inline bool tcp_is_reno(const struct tcp_sock *tp)
1275{
1276	return !tcp_is_sack(tp);
1277}
1278
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1279static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1280{
1281	return tp->sacked_out + tp->lost_out;
1282}
1283
1284/* This determines how many packets are "in the network" to the best
1285 * of our knowledge.  In many cases it is conservative, but where
1286 * detailed information is available from the receiver (via SACK
1287 * blocks etc.) we can make more aggressive calculations.
1288 *
1289 * Use this for decisions involving congestion control, use just
1290 * tp->packets_out to determine if the send queue is empty or not.
1291 *
1292 * Read this equation as:
1293 *
1294 *	"Packets sent once on transmission queue" MINUS
1295 *	"Packets left network, but not honestly ACKed yet" PLUS
1296 *	"Packets fast retransmitted"
1297 */
1298static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1299{
1300	return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1301}
1302
1303#define TCP_INFINITE_SSTHRESH	0x7fffffff
1304
1305static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp)
1306{
1307	return tp->snd_cwnd;
1308}
1309
1310static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val)
1311{
1312	WARN_ON_ONCE((int)val <= 0);
1313	tp->snd_cwnd = val;
1314}
1315
1316static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1317{
1318	return tcp_snd_cwnd(tp) < tp->snd_ssthresh;
1319}
1320
1321static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1322{
1323	return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1324}
1325
1326static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1327{
1328	return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1329	       (1 << inet_csk(sk)->icsk_ca_state);
1330}
1331
1332/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1333 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1334 * ssthresh.
1335 */
1336static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1337{
1338	const struct tcp_sock *tp = tcp_sk(sk);
1339
1340	if (tcp_in_cwnd_reduction(sk))
1341		return tp->snd_ssthresh;
1342	else
1343		return max(tp->snd_ssthresh,
1344			   ((tcp_snd_cwnd(tp) >> 1) +
1345			    (tcp_snd_cwnd(tp) >> 2)));
1346}
1347
1348/* Use define here intentionally to get WARN_ON location shown at the caller */
1349#define tcp_verify_left_out(tp)	WARN_ON(tcp_left_out(tp) > tp->packets_out)
1350
1351void tcp_enter_cwr(struct sock *sk);
1352__u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1353
1354/* The maximum number of MSS of available cwnd for which TSO defers
1355 * sending if not using sysctl_tcp_tso_win_divisor.
1356 */
1357static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1358{
1359	return 3;
1360}
1361
1362/* Returns end sequence number of the receiver's advertised window */
1363static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1364{
1365	return tp->snd_una + tp->snd_wnd;
1366}
1367
1368/* We follow the spirit of RFC2861 to validate cwnd but implement a more
1369 * flexible approach. The RFC suggests cwnd should not be raised unless
1370 * it was fully used previously. And that's exactly what we do in
1371 * congestion avoidance mode. But in slow start we allow cwnd to grow
1372 * as long as the application has used half the cwnd.
1373 * Example :
1374 *    cwnd is 10 (IW10), but application sends 9 frames.
1375 *    We allow cwnd to reach 18 when all frames are ACKed.
1376 * This check is safe because it's as aggressive as slow start which already
1377 * risks 100% overshoot. The advantage is that we discourage application to
1378 * either send more filler packets or data to artificially blow up the cwnd
1379 * usage, and allow application-limited process to probe bw more aggressively.
1380 */
1381static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1382{
1383	const struct tcp_sock *tp = tcp_sk(sk);
1384
1385	if (tp->is_cwnd_limited)
1386		return true;
1387
1388	/* If in slow start, ensure cwnd grows to twice what was ACKed. */
1389	if (tcp_in_slow_start(tp))
1390		return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out;
1391
1392	return false;
1393}
1394
1395/* BBR congestion control needs pacing.
1396 * Same remark for SO_MAX_PACING_RATE.
1397 * sch_fq packet scheduler is efficiently handling pacing,
1398 * but is not always installed/used.
1399 * Return true if TCP stack should pace packets itself.
1400 */
1401static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1402{
1403	return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1404}
1405
1406/* Estimates in how many jiffies next packet for this flow can be sent.
1407 * Scheduling a retransmit timer too early would be silly.
1408 */
1409static inline unsigned long tcp_pacing_delay(const struct sock *sk)
1410{
1411	s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache;
1412
1413	return delay > 0 ? nsecs_to_jiffies(delay) : 0;
1414}
1415
1416static inline void tcp_reset_xmit_timer(struct sock *sk,
1417					const int what,
1418					unsigned long when,
1419					const unsigned long max_when)
1420{
1421	inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk),
1422				  max_when);
1423}
1424
1425/* Something is really bad, we could not queue an additional packet,
1426 * because qdisc is full or receiver sent a 0 window, or we are paced.
1427 * We do not want to add fuel to the fire, or abort too early,
1428 * so make sure the timer we arm now is at least 200ms in the future,
1429 * regardless of current icsk_rto value (as it could be ~2ms)
1430 */
1431static inline unsigned long tcp_probe0_base(const struct sock *sk)
1432{
1433	return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1434}
1435
1436/* Variant of inet_csk_rto_backoff() used for zero window probes */
1437static inline unsigned long tcp_probe0_when(const struct sock *sk,
1438					    unsigned long max_when)
1439{
1440	u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1,
1441			   inet_csk(sk)->icsk_backoff);
1442	u64 when = (u64)tcp_probe0_base(sk) << backoff;
1443
1444	return (unsigned long)min_t(u64, when, max_when);
1445}
1446
1447static inline void tcp_check_probe_timer(struct sock *sk)
1448{
1449	if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1450		tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1451				     tcp_probe0_base(sk), TCP_RTO_MAX);
1452}
1453
1454static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1455{
1456	tp->snd_wl1 = seq;
1457}
1458
1459static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1460{
1461	tp->snd_wl1 = seq;
1462}
1463
1464/*
1465 * Calculate(/check) TCP checksum
1466 */
1467static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1468				   __be32 daddr, __wsum base)
1469{
1470	return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base);
 
 
 
 
 
1471}
1472
1473static inline bool tcp_checksum_complete(struct sk_buff *skb)
1474{
1475	return !skb_csum_unnecessary(skb) &&
1476		__skb_checksum_complete(skb);
1477}
1478
1479bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb,
1480		     enum skb_drop_reason *reason);
1481
 
 
 
 
 
 
 
1482
 
 
1483int tcp_filter(struct sock *sk, struct sk_buff *skb);
 
 
 
 
 
 
 
 
 
 
1484void tcp_set_state(struct sock *sk, int state);
 
1485void tcp_done(struct sock *sk);
 
1486int tcp_abort(struct sock *sk, int err);
1487
1488static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1489{
1490	rx_opt->dsack = 0;
1491	rx_opt->num_sacks = 0;
1492}
1493
 
1494void tcp_cwnd_restart(struct sock *sk, s32 delta);
1495
1496static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1497{
1498	const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1499	struct tcp_sock *tp = tcp_sk(sk);
1500	s32 delta;
1501
1502	if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) ||
1503	    tp->packets_out || ca_ops->cong_control)
1504		return;
1505	delta = tcp_jiffies32 - tp->lsndtime;
1506	if (delta > inet_csk(sk)->icsk_rto)
1507		tcp_cwnd_restart(sk, delta);
1508}
1509
1510/* Determine a window scaling and initial window to offer. */
1511void tcp_select_initial_window(const struct sock *sk, int __space,
1512			       __u32 mss, __u32 *rcv_wnd,
1513			       __u32 *window_clamp, int wscale_ok,
1514			       __u8 *rcv_wscale, __u32 init_rcv_wnd);
1515
1516static inline int __tcp_win_from_space(u8 scaling_ratio, int space)
1517{
1518	s64 scaled_space = (s64)space * scaling_ratio;
1519
1520	return scaled_space >> TCP_RMEM_TO_WIN_SCALE;
1521}
1522
1523static inline int tcp_win_from_space(const struct sock *sk, int space)
1524{
1525	return __tcp_win_from_space(tcp_sk(sk)->scaling_ratio, space);
1526}
1527
1528/* inverse of __tcp_win_from_space() */
1529static inline int __tcp_space_from_win(u8 scaling_ratio, int win)
1530{
1531	u64 val = (u64)win << TCP_RMEM_TO_WIN_SCALE;
1532
1533	do_div(val, scaling_ratio);
1534	return val;
1535}
1536
1537static inline int tcp_space_from_win(const struct sock *sk, int win)
1538{
1539	return __tcp_space_from_win(tcp_sk(sk)->scaling_ratio, win);
1540}
1541
1542/* Assume a 50% default for skb->len/skb->truesize ratio.
1543 * This may be adjusted later in tcp_measure_rcv_mss().
1544 */
1545#define TCP_DEFAULT_SCALING_RATIO (1 << (TCP_RMEM_TO_WIN_SCALE - 1))
1546
1547static inline void tcp_scaling_ratio_init(struct sock *sk)
1548{
1549	tcp_sk(sk)->scaling_ratio = TCP_DEFAULT_SCALING_RATIO;
1550}
1551
1552/* Note: caller must be prepared to deal with negative returns */
1553static inline int tcp_space(const struct sock *sk)
1554{
1555	return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) -
1556				  READ_ONCE(sk->sk_backlog.len) -
1557				  atomic_read(&sk->sk_rmem_alloc));
1558}
1559
1560static inline int tcp_full_space(const struct sock *sk)
1561{
1562	return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1563}
1564
1565static inline void __tcp_adjust_rcv_ssthresh(struct sock *sk, u32 new_ssthresh)
1566{
1567	int unused_mem = sk_unused_reserved_mem(sk);
1568	struct tcp_sock *tp = tcp_sk(sk);
1569
1570	tp->rcv_ssthresh = min(tp->rcv_ssthresh, new_ssthresh);
1571	if (unused_mem)
1572		tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh,
1573					 tcp_win_from_space(sk, unused_mem));
1574}
1575
1576static inline void tcp_adjust_rcv_ssthresh(struct sock *sk)
1577{
1578	__tcp_adjust_rcv_ssthresh(sk, 4U * tcp_sk(sk)->advmss);
1579}
1580
1581void tcp_cleanup_rbuf(struct sock *sk, int copied);
1582void __tcp_cleanup_rbuf(struct sock *sk, int copied);
1583
1584
1585/* We provision sk_rcvbuf around 200% of sk_rcvlowat.
1586 * If 87.5 % (7/8) of the space has been consumed, we want to override
1587 * SO_RCVLOWAT constraint, since we are receiving skbs with too small
1588 * len/truesize ratio.
1589 */
1590static inline bool tcp_rmem_pressure(const struct sock *sk)
1591{
1592	int rcvbuf, threshold;
1593
1594	if (tcp_under_memory_pressure(sk))
1595		return true;
1596
1597	rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1598	threshold = rcvbuf - (rcvbuf >> 3);
1599
1600	return atomic_read(&sk->sk_rmem_alloc) > threshold;
1601}
1602
1603static inline bool tcp_epollin_ready(const struct sock *sk, int target)
1604{
1605	const struct tcp_sock *tp = tcp_sk(sk);
1606	int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq);
1607
1608	if (avail <= 0)
1609		return false;
1610
1611	return (avail >= target) || tcp_rmem_pressure(sk) ||
1612	       (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss);
1613}
1614
1615extern void tcp_openreq_init_rwin(struct request_sock *req,
1616				  const struct sock *sk_listener,
1617				  const struct dst_entry *dst);
1618
1619void tcp_enter_memory_pressure(struct sock *sk);
1620void tcp_leave_memory_pressure(struct sock *sk);
1621
1622static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1623{
1624	struct net *net = sock_net((struct sock *)tp);
1625	int val;
1626
1627	/* Paired with WRITE_ONCE() in tcp_sock_set_keepintvl()
1628	 * and do_tcp_setsockopt().
1629	 */
1630	val = READ_ONCE(tp->keepalive_intvl);
1631
1632	return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl);
1633}
1634
1635static inline int keepalive_time_when(const struct tcp_sock *tp)
1636{
1637	struct net *net = sock_net((struct sock *)tp);
1638	int val;
1639
1640	/* Paired with WRITE_ONCE() in tcp_sock_set_keepidle_locked() */
1641	val = READ_ONCE(tp->keepalive_time);
1642
1643	return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time);
1644}
1645
1646static inline int keepalive_probes(const struct tcp_sock *tp)
1647{
1648	struct net *net = sock_net((struct sock *)tp);
1649	int val;
1650
1651	/* Paired with WRITE_ONCE() in tcp_sock_set_keepcnt()
1652	 * and do_tcp_setsockopt().
1653	 */
1654	val = READ_ONCE(tp->keepalive_probes);
1655
1656	return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes);
1657}
1658
1659static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1660{
1661	const struct inet_connection_sock *icsk = &tp->inet_conn;
1662
1663	return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1664			  tcp_jiffies32 - tp->rcv_tstamp);
1665}
1666
1667static inline int tcp_fin_time(const struct sock *sk)
1668{
1669	int fin_timeout = tcp_sk(sk)->linger2 ? :
1670		READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout);
1671	const int rto = inet_csk(sk)->icsk_rto;
1672
1673	if (fin_timeout < (rto << 2) - (rto >> 1))
1674		fin_timeout = (rto << 2) - (rto >> 1);
1675
1676	return fin_timeout;
1677}
1678
1679static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1680				  int paws_win)
1681{
1682	if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1683		return true;
1684	if (unlikely(!time_before32(ktime_get_seconds(),
1685				    rx_opt->ts_recent_stamp + TCP_PAWS_WRAP)))
1686		return true;
1687	/*
1688	 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1689	 * then following tcp messages have valid values. Ignore 0 value,
1690	 * or else 'negative' tsval might forbid us to accept their packets.
1691	 */
1692	if (!rx_opt->ts_recent)
1693		return true;
1694	return false;
1695}
1696
1697static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1698				   int rst)
1699{
1700	if (tcp_paws_check(rx_opt, 0))
1701		return false;
1702
1703	/* RST segments are not recommended to carry timestamp,
1704	   and, if they do, it is recommended to ignore PAWS because
1705	   "their cleanup function should take precedence over timestamps."
1706	   Certainly, it is mistake. It is necessary to understand the reasons
1707	   of this constraint to relax it: if peer reboots, clock may go
1708	   out-of-sync and half-open connections will not be reset.
1709	   Actually, the problem would be not existing if all
1710	   the implementations followed draft about maintaining clock
1711	   via reboots. Linux-2.2 DOES NOT!
1712
1713	   However, we can relax time bounds for RST segments to MSL.
1714	 */
1715	if (rst && !time_before32(ktime_get_seconds(),
1716				  rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1717		return false;
1718	return true;
1719}
1720
1721bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1722			  int mib_idx, u32 *last_oow_ack_time);
1723
1724static inline void tcp_mib_init(struct net *net)
1725{
1726	/* See RFC 2012 */
1727	TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1728	TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1729	TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1730	TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1731}
1732
1733/* from STCP */
1734static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1735{
1736	tp->lost_skb_hint = NULL;
1737}
1738
1739static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1740{
1741	tcp_clear_retrans_hints_partial(tp);
1742	tp->retransmit_skb_hint = NULL;
1743}
1744
1745#define tcp_md5_addr tcp_ao_addr
 
 
 
 
 
1746
1747/* - key database */
1748struct tcp_md5sig_key {
1749	struct hlist_node	node;
1750	u8			keylen;
1751	u8			family; /* AF_INET or AF_INET6 */
1752	u8			prefixlen;
1753	u8			flags;
1754	union tcp_md5_addr	addr;
1755	int			l3index; /* set if key added with L3 scope */
1756	u8			key[TCP_MD5SIG_MAXKEYLEN];
1757	struct rcu_head		rcu;
1758};
1759
1760/* - sock block */
1761struct tcp_md5sig_info {
1762	struct hlist_head	head;
1763	struct rcu_head		rcu;
1764};
1765
1766/* - pseudo header */
1767struct tcp4_pseudohdr {
1768	__be32		saddr;
1769	__be32		daddr;
1770	__u8		pad;
1771	__u8		protocol;
1772	__be16		len;
1773};
1774
1775struct tcp6_pseudohdr {
1776	struct in6_addr	saddr;
1777	struct in6_addr daddr;
1778	__be32		len;
1779	__be32		protocol;	/* including padding */
1780};
1781
1782union tcp_md5sum_block {
1783	struct tcp4_pseudohdr ip4;
1784#if IS_ENABLED(CONFIG_IPV6)
1785	struct tcp6_pseudohdr ip6;
1786#endif
1787};
1788
1789/*
1790 * struct tcp_sigpool - per-CPU pool of ahash_requests
1791 * @scratch: per-CPU temporary area, that can be used between
1792 *	     tcp_sigpool_start() and tcp_sigpool_end() to perform
1793 *	     crypto request
1794 * @req: pre-allocated ahash request
1795 */
1796struct tcp_sigpool {
1797	void *scratch;
1798	struct ahash_request *req;
1799};
1800
1801int tcp_sigpool_alloc_ahash(const char *alg, size_t scratch_size);
1802void tcp_sigpool_get(unsigned int id);
1803void tcp_sigpool_release(unsigned int id);
1804int tcp_sigpool_hash_skb_data(struct tcp_sigpool *hp,
1805			      const struct sk_buff *skb,
1806			      unsigned int header_len);
1807
1808/**
1809 * tcp_sigpool_start - disable bh and start using tcp_sigpool_ahash
1810 * @id: tcp_sigpool that was previously allocated by tcp_sigpool_alloc_ahash()
1811 * @c: returned tcp_sigpool for usage (uninitialized on failure)
1812 *
1813 * Returns 0 on success, error otherwise.
1814 */
1815int tcp_sigpool_start(unsigned int id, struct tcp_sigpool *c);
1816/**
1817 * tcp_sigpool_end - enable bh and stop using tcp_sigpool
1818 * @c: tcp_sigpool context that was returned by tcp_sigpool_start()
1819 */
1820void tcp_sigpool_end(struct tcp_sigpool *c);
1821size_t tcp_sigpool_algo(unsigned int id, char *buf, size_t buf_len);
1822/* - functions */
1823int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1824			const struct sock *sk, const struct sk_buff *skb);
1825int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1826		   int family, u8 prefixlen, int l3index, u8 flags,
1827		   const u8 *newkey, u8 newkeylen);
1828int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr,
1829		     int family, u8 prefixlen, int l3index,
1830		     struct tcp_md5sig_key *key);
1831
1832int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1833		   int family, u8 prefixlen, int l3index, u8 flags);
1834void tcp_clear_md5_list(struct sock *sk);
1835struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1836					 const struct sock *addr_sk);
1837
1838#ifdef CONFIG_TCP_MD5SIG
1839struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1840					   const union tcp_md5_addr *addr,
1841					   int family, bool any_l3index);
1842static inline struct tcp_md5sig_key *
1843tcp_md5_do_lookup(const struct sock *sk, int l3index,
1844		  const union tcp_md5_addr *addr, int family)
1845{
1846	if (!static_branch_unlikely(&tcp_md5_needed.key))
1847		return NULL;
1848	return __tcp_md5_do_lookup(sk, l3index, addr, family, false);
1849}
1850
1851static inline struct tcp_md5sig_key *
1852tcp_md5_do_lookup_any_l3index(const struct sock *sk,
1853			      const union tcp_md5_addr *addr, int family)
1854{
1855	if (!static_branch_unlikely(&tcp_md5_needed.key))
1856		return NULL;
1857	return __tcp_md5_do_lookup(sk, 0, addr, family, true);
1858}
1859
1860enum skb_drop_reason
1861tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb,
1862		     const void *saddr, const void *daddr,
1863		     int family, int l3index, const __u8 *hash_location);
1864
1865
1866#define tcp_twsk_md5_key(twsk)	((twsk)->tw_md5_key)
1867#else
1868static inline struct tcp_md5sig_key *
1869tcp_md5_do_lookup(const struct sock *sk, int l3index,
1870		  const union tcp_md5_addr *addr, int family)
1871{
1872	return NULL;
1873}
 
 
1874
1875static inline struct tcp_md5sig_key *
1876tcp_md5_do_lookup_any_l3index(const struct sock *sk,
1877			      const union tcp_md5_addr *addr, int family)
1878{
1879	return NULL;
1880}
1881
1882static inline enum skb_drop_reason
1883tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb,
1884		     const void *saddr, const void *daddr,
1885		     int family, int l3index, const __u8 *hash_location)
1886{
1887	return SKB_NOT_DROPPED_YET;
1888}
1889#define tcp_twsk_md5_key(twsk)	NULL
1890#endif
1891
1892int tcp_md5_alloc_sigpool(void);
1893void tcp_md5_release_sigpool(void);
1894void tcp_md5_add_sigpool(void);
1895extern int tcp_md5_sigpool_id;
1896
1897int tcp_md5_hash_key(struct tcp_sigpool *hp,
1898		     const struct tcp_md5sig_key *key);
1899
1900/* From tcp_fastopen.c */
1901void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1902			    struct tcp_fastopen_cookie *cookie);
 
1903void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1904			    struct tcp_fastopen_cookie *cookie, bool syn_lost,
1905			    u16 try_exp);
1906struct tcp_fastopen_request {
1907	/* Fast Open cookie. Size 0 means a cookie request */
1908	struct tcp_fastopen_cookie	cookie;
1909	struct msghdr			*data;  /* data in MSG_FASTOPEN */
1910	size_t				size;
1911	int				copied;	/* queued in tcp_connect() */
1912	struct ubuf_info		*uarg;
1913};
1914void tcp_free_fastopen_req(struct tcp_sock *tp);
1915void tcp_fastopen_destroy_cipher(struct sock *sk);
1916void tcp_fastopen_ctx_destroy(struct net *net);
1917int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1918			      void *primary_key, void *backup_key);
1919int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk,
1920			    u64 *key);
1921void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1922struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1923			      struct request_sock *req,
1924			      struct tcp_fastopen_cookie *foc,
1925			      const struct dst_entry *dst);
1926void tcp_fastopen_init_key_once(struct net *net);
1927bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1928			     struct tcp_fastopen_cookie *cookie);
1929bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1930#define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t)
1931#define TCP_FASTOPEN_KEY_MAX 2
1932#define TCP_FASTOPEN_KEY_BUF_LENGTH \
1933	(TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX)
1934
1935/* Fastopen key context */
1936struct tcp_fastopen_context {
1937	siphash_key_t	key[TCP_FASTOPEN_KEY_MAX];
1938	int		num;
1939	struct rcu_head	rcu;
1940};
1941
1942void tcp_fastopen_active_disable(struct sock *sk);
1943bool tcp_fastopen_active_should_disable(struct sock *sk);
1944void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1945void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1946
1947/* Caller needs to wrap with rcu_read_(un)lock() */
1948static inline
1949struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk)
1950{
1951	struct tcp_fastopen_context *ctx;
1952
1953	ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx);
1954	if (!ctx)
1955		ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx);
1956	return ctx;
1957}
1958
1959static inline
1960bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc,
1961			       const struct tcp_fastopen_cookie *orig)
1962{
1963	if (orig->len == TCP_FASTOPEN_COOKIE_SIZE &&
1964	    orig->len == foc->len &&
1965	    !memcmp(orig->val, foc->val, foc->len))
1966		return true;
1967	return false;
1968}
1969
1970static inline
1971int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx)
1972{
1973	return ctx->num;
1974}
1975
1976/* Latencies incurred by various limits for a sender. They are
1977 * chronograph-like stats that are mutually exclusive.
1978 */
1979enum tcp_chrono {
1980	TCP_CHRONO_UNSPEC,
1981	TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1982	TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1983	TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1984	__TCP_CHRONO_MAX,
1985};
1986
1987void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1988void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1989
1990/* This helper is needed, because skb->tcp_tsorted_anchor uses
1991 * the same memory storage than skb->destructor/_skb_refdst
1992 */
1993static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1994{
1995	skb->destructor = NULL;
1996	skb->_skb_refdst = 0UL;
1997}
1998
1999#define tcp_skb_tsorted_save(skb) {		\
2000	unsigned long _save = skb->_skb_refdst;	\
2001	skb->_skb_refdst = 0UL;
2002
2003#define tcp_skb_tsorted_restore(skb)		\
2004	skb->_skb_refdst = _save;		\
 
 
 
2005}
2006
2007void tcp_write_queue_purge(struct sock *sk);
 
 
 
2008
2009static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
2010{
2011	return skb_rb_first(&sk->tcp_rtx_queue);
2012}
2013
2014static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
 
2015{
2016	return skb_rb_last(&sk->tcp_rtx_queue);
2017}
2018
2019static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
 
2020{
2021	return skb_peek_tail(&sk->sk_write_queue);
2022}
2023
 
 
 
 
 
 
2024#define tcp_for_write_queue_from_safe(skb, tmp, sk)			\
2025	skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
2026
2027static inline struct sk_buff *tcp_send_head(const struct sock *sk)
2028{
2029	return skb_peek(&sk->sk_write_queue);
2030}
2031
2032static inline bool tcp_skb_is_last(const struct sock *sk,
2033				   const struct sk_buff *skb)
2034{
2035	return skb_queue_is_last(&sk->sk_write_queue, skb);
2036}
2037
2038/**
2039 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue
2040 * @sk: socket
2041 *
2042 * Since the write queue can have a temporary empty skb in it,
2043 * we must not use "return skb_queue_empty(&sk->sk_write_queue)"
2044 */
2045static inline bool tcp_write_queue_empty(const struct sock *sk)
2046{
2047	const struct tcp_sock *tp = tcp_sk(sk);
 
 
 
 
2048
2049	return tp->write_seq == tp->snd_nxt;
 
 
 
 
 
 
 
2050}
2051
2052static inline bool tcp_rtx_queue_empty(const struct sock *sk)
2053{
2054	return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
2055}
2056
2057static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
2058{
2059	return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
2060}
2061
2062static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
2063{
2064	__skb_queue_tail(&sk->sk_write_queue, skb);
2065
2066	/* Queue it, remembering where we must start sending. */
2067	if (sk->sk_write_queue.next == skb)
 
2068		tcp_chrono_start(sk, TCP_CHRONO_BUSY);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2069}
2070
2071/* Insert new before skb on the write queue of sk.  */
2072static inline void tcp_insert_write_queue_before(struct sk_buff *new,
2073						  struct sk_buff *skb,
2074						  struct sock *sk)
2075{
2076	__skb_queue_before(&sk->sk_write_queue, skb, new);
 
 
 
2077}
2078
2079static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
2080{
2081	tcp_skb_tsorted_anchor_cleanup(skb);
2082	__skb_unlink(skb, &sk->sk_write_queue);
2083}
2084
2085void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
2086
2087static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
2088{
2089	tcp_skb_tsorted_anchor_cleanup(skb);
2090	rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
2091}
2092
2093static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
2094{
2095	list_del(&skb->tcp_tsorted_anchor);
2096	tcp_rtx_queue_unlink(skb, sk);
2097	tcp_wmem_free_skb(sk, skb);
2098}
2099
2100static inline void tcp_push_pending_frames(struct sock *sk)
2101{
2102	if (tcp_send_head(sk)) {
2103		struct tcp_sock *tp = tcp_sk(sk);
2104
2105		__tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
2106	}
2107}
2108
2109/* Start sequence of the skb just after the highest skb with SACKed
2110 * bit, valid only if sacked_out > 0 or when the caller has ensured
2111 * validity by itself.
2112 */
2113static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
2114{
2115	if (!tp->sacked_out)
2116		return tp->snd_una;
2117
2118	if (tp->highest_sack == NULL)
2119		return tp->snd_nxt;
2120
2121	return TCP_SKB_CB(tp->highest_sack)->seq;
2122}
2123
2124static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
2125{
2126	tcp_sk(sk)->highest_sack = skb_rb_next(skb);
 
2127}
2128
2129static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
2130{
2131	return tcp_sk(sk)->highest_sack;
2132}
2133
2134static inline void tcp_highest_sack_reset(struct sock *sk)
2135{
2136	tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
2137}
2138
2139/* Called when old skb is about to be deleted and replaced by new skb */
2140static inline void tcp_highest_sack_replace(struct sock *sk,
2141					    struct sk_buff *old,
2142					    struct sk_buff *new)
2143{
2144	if (old == tcp_highest_sack(sk))
2145		tcp_sk(sk)->highest_sack = new;
2146}
2147
2148/* This helper checks if socket has IP_TRANSPARENT set */
2149static inline bool inet_sk_transparent(const struct sock *sk)
2150{
2151	switch (sk->sk_state) {
2152	case TCP_TIME_WAIT:
2153		return inet_twsk(sk)->tw_transparent;
2154	case TCP_NEW_SYN_RECV:
2155		return inet_rsk(inet_reqsk(sk))->no_srccheck;
2156	}
2157	return inet_test_bit(TRANSPARENT, sk);
2158}
2159
2160/* Determines whether this is a thin stream (which may suffer from
2161 * increased latency). Used to trigger latency-reducing mechanisms.
2162 */
2163static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
2164{
2165	return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
2166}
2167
2168/* /proc */
2169enum tcp_seq_states {
2170	TCP_SEQ_STATE_LISTENING,
2171	TCP_SEQ_STATE_ESTABLISHED,
2172};
2173
2174void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
2175void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
2176void tcp_seq_stop(struct seq_file *seq, void *v);
2177
2178struct tcp_seq_afinfo {
 
2179	sa_family_t			family;
 
 
2180};
2181
2182struct tcp_iter_state {
2183	struct seq_net_private	p;
 
2184	enum tcp_seq_states	state;
2185	struct sock		*syn_wait_sk;
2186	int			bucket, offset, sbucket, num;
2187	loff_t			last_pos;
2188};
2189
 
 
 
2190extern struct request_sock_ops tcp_request_sock_ops;
2191extern struct request_sock_ops tcp6_request_sock_ops;
2192
2193void tcp_v4_destroy_sock(struct sock *sk);
2194
2195struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
2196				netdev_features_t features);
2197struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb);
2198INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff));
2199INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb));
2200INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff));
2201INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb));
2202#ifdef CONFIG_INET
2203void tcp_gro_complete(struct sk_buff *skb);
2204#else
2205static inline void tcp_gro_complete(struct sk_buff *skb) { }
2206#endif
2207
2208void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
2209
2210static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
2211{
2212	struct net *net = sock_net((struct sock *)tp);
2213	u32 val;
 
2214
2215	val = READ_ONCE(tp->notsent_lowat);
 
 
 
2216
2217	return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat);
2218}
2219
2220bool tcp_stream_memory_free(const struct sock *sk, int wake);
2221
2222#ifdef CONFIG_PROC_FS
2223int tcp4_proc_init(void);
2224void tcp4_proc_exit(void);
2225#endif
2226
2227int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
2228int tcp_conn_request(struct request_sock_ops *rsk_ops,
2229		     const struct tcp_request_sock_ops *af_ops,
2230		     struct sock *sk, struct sk_buff *skb);
2231
2232/* TCP af-specific functions */
2233struct tcp_sock_af_ops {
2234#ifdef CONFIG_TCP_MD5SIG
2235	struct tcp_md5sig_key	*(*md5_lookup) (const struct sock *sk,
2236						const struct sock *addr_sk);
2237	int		(*calc_md5_hash)(char *location,
2238					 const struct tcp_md5sig_key *md5,
2239					 const struct sock *sk,
2240					 const struct sk_buff *skb);
2241	int		(*md5_parse)(struct sock *sk,
2242				     int optname,
2243				     sockptr_t optval,
2244				     int optlen);
2245#endif
2246#ifdef CONFIG_TCP_AO
2247	int (*ao_parse)(struct sock *sk, int optname, sockptr_t optval, int optlen);
2248	struct tcp_ao_key *(*ao_lookup)(const struct sock *sk,
2249					struct sock *addr_sk,
2250					int sndid, int rcvid);
2251	int (*ao_calc_key_sk)(struct tcp_ao_key *mkt, u8 *key,
2252			      const struct sock *sk,
2253			      __be32 sisn, __be32 disn, bool send);
2254	int (*calc_ao_hash)(char *location, struct tcp_ao_key *ao,
2255			    const struct sock *sk, const struct sk_buff *skb,
2256			    const u8 *tkey, int hash_offset, u32 sne);
2257#endif
2258};
2259
2260struct tcp_request_sock_ops {
2261	u16 mss_clamp;
2262#ifdef CONFIG_TCP_MD5SIG
2263	struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
2264						 const struct sock *addr_sk);
2265	int		(*calc_md5_hash) (char *location,
2266					  const struct tcp_md5sig_key *md5,
2267					  const struct sock *sk,
2268					  const struct sk_buff *skb);
2269#endif
2270#ifdef CONFIG_TCP_AO
2271	struct tcp_ao_key *(*ao_lookup)(const struct sock *sk,
2272					struct request_sock *req,
2273					int sndid, int rcvid);
2274	int (*ao_calc_key)(struct tcp_ao_key *mkt, u8 *key, struct request_sock *sk);
2275	int (*ao_synack_hash)(char *ao_hash, struct tcp_ao_key *mkt,
2276			      struct request_sock *req, const struct sk_buff *skb,
2277			      int hash_offset, u32 sne);
2278#endif
2279#ifdef CONFIG_SYN_COOKIES
2280	__u32 (*cookie_init_seq)(const struct sk_buff *skb,
2281				 __u16 *mss);
2282#endif
2283	struct dst_entry *(*route_req)(const struct sock *sk,
2284				       struct sk_buff *skb,
2285				       struct flowi *fl,
2286				       struct request_sock *req);
2287	u32 (*init_seq)(const struct sk_buff *skb);
2288	u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
2289	int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
2290			   struct flowi *fl, struct request_sock *req,
2291			   struct tcp_fastopen_cookie *foc,
2292			   enum tcp_synack_type synack_type,
2293			   struct sk_buff *syn_skb);
2294};
2295
2296extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops;
2297#if IS_ENABLED(CONFIG_IPV6)
2298extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops;
2299#endif
2300
2301#ifdef CONFIG_SYN_COOKIES
2302static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2303					 const struct sock *sk, struct sk_buff *skb,
2304					 __u16 *mss)
2305{
2306	tcp_synq_overflow(sk);
2307	__NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
2308	return ops->cookie_init_seq(skb, mss);
2309}
2310#else
2311static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2312					 const struct sock *sk, struct sk_buff *skb,
2313					 __u16 *mss)
2314{
2315	return 0;
2316}
2317#endif
2318
2319struct tcp_key {
2320	union {
2321		struct {
2322			struct tcp_ao_key *ao_key;
2323			char *traffic_key;
2324			u32 sne;
2325			u8 rcv_next;
2326		};
2327		struct tcp_md5sig_key *md5_key;
2328	};
2329	enum {
2330		TCP_KEY_NONE = 0,
2331		TCP_KEY_MD5,
2332		TCP_KEY_AO,
2333	} type;
2334};
2335
2336static inline void tcp_get_current_key(const struct sock *sk,
2337				       struct tcp_key *out)
2338{
2339#if defined(CONFIG_TCP_AO) || defined(CONFIG_TCP_MD5SIG)
2340	const struct tcp_sock *tp = tcp_sk(sk);
2341#endif
2342
2343#ifdef CONFIG_TCP_AO
2344	if (static_branch_unlikely(&tcp_ao_needed.key)) {
2345		struct tcp_ao_info *ao;
2346
2347		ao = rcu_dereference_protected(tp->ao_info,
2348					       lockdep_sock_is_held(sk));
2349		if (ao) {
2350			out->ao_key = READ_ONCE(ao->current_key);
2351			out->type = TCP_KEY_AO;
2352			return;
2353		}
2354	}
2355#endif
2356#ifdef CONFIG_TCP_MD5SIG
2357	if (static_branch_unlikely(&tcp_md5_needed.key) &&
2358	    rcu_access_pointer(tp->md5sig_info)) {
2359		out->md5_key = tp->af_specific->md5_lookup(sk, sk);
2360		if (out->md5_key) {
2361			out->type = TCP_KEY_MD5;
2362			return;
2363		}
2364	}
2365#endif
2366	out->type = TCP_KEY_NONE;
2367}
2368
2369static inline bool tcp_key_is_md5(const struct tcp_key *key)
2370{
2371#ifdef CONFIG_TCP_MD5SIG
2372	if (static_branch_unlikely(&tcp_md5_needed.key) &&
2373	    key->type == TCP_KEY_MD5)
2374		return true;
2375#endif
2376	return false;
2377}
2378
2379static inline bool tcp_key_is_ao(const struct tcp_key *key)
2380{
2381#ifdef CONFIG_TCP_AO
2382	if (static_branch_unlikely(&tcp_ao_needed.key) &&
2383	    key->type == TCP_KEY_AO)
2384		return true;
2385#endif
2386	return false;
2387}
2388
2389int tcpv4_offload_init(void);
2390
2391void tcp_v4_init(void);
2392void tcp_init(void);
2393
2394/* tcp_recovery.c */
2395void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
2396void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
2397extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
2398				u32 reo_wnd);
2399extern bool tcp_rack_mark_lost(struct sock *sk);
2400extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
2401			     u64 xmit_time);
2402extern void tcp_rack_reo_timeout(struct sock *sk);
2403extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
2404
2405/* tcp_plb.c */
2406
2407/*
2408 * Scaling factor for fractions in PLB. For example, tcp_plb_update_state
2409 * expects cong_ratio which represents fraction of traffic that experienced
2410 * congestion over a single RTT. In order to avoid floating point operations,
2411 * this fraction should be mapped to (1 << TCP_PLB_SCALE) and passed in.
2412 */
2413#define TCP_PLB_SCALE 8
2414
2415/* State for PLB (Protective Load Balancing) for a single TCP connection. */
2416struct tcp_plb_state {
2417	u8	consec_cong_rounds:5, /* consecutive congested rounds */
2418		unused:3;
2419	u32	pause_until; /* jiffies32 when PLB can resume rerouting */
2420};
2421
2422static inline void tcp_plb_init(const struct sock *sk,
2423				struct tcp_plb_state *plb)
2424{
2425	plb->consec_cong_rounds = 0;
2426	plb->pause_until = 0;
2427}
2428void tcp_plb_update_state(const struct sock *sk, struct tcp_plb_state *plb,
2429			  const int cong_ratio);
2430void tcp_plb_check_rehash(struct sock *sk, struct tcp_plb_state *plb);
2431void tcp_plb_update_state_upon_rto(struct sock *sk, struct tcp_plb_state *plb);
2432
2433/* At how many usecs into the future should the RTO fire? */
2434static inline s64 tcp_rto_delta_us(const struct sock *sk)
2435{
2436	const struct sk_buff *skb = tcp_rtx_queue_head(sk);
2437	u32 rto = inet_csk(sk)->icsk_rto;
2438	u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto);
2439
2440	return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
2441}
2442
2443/*
2444 * Save and compile IPv4 options, return a pointer to it
2445 */
2446static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
2447							 struct sk_buff *skb)
2448{
2449	const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2450	struct ip_options_rcu *dopt = NULL;
2451
2452	if (opt->optlen) {
2453		int opt_size = sizeof(*dopt) + opt->optlen;
2454
2455		dopt = kmalloc(opt_size, GFP_ATOMIC);
2456		if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
2457			kfree(dopt);
2458			dopt = NULL;
2459		}
2460	}
2461	return dopt;
2462}
2463
2464/* locally generated TCP pure ACKs have skb->truesize == 2
2465 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2466 * This is much faster than dissecting the packet to find out.
2467 * (Think of GRE encapsulations, IPv4, IPv6, ...)
2468 */
2469static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2470{
2471	return skb->truesize == 2;
2472}
2473
2474static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2475{
2476	skb->truesize = 2;
2477}
2478
2479static inline int tcp_inq(struct sock *sk)
2480{
2481	struct tcp_sock *tp = tcp_sk(sk);
2482	int answ;
2483
2484	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2485		answ = 0;
2486	} else if (sock_flag(sk, SOCK_URGINLINE) ||
2487		   !tp->urg_data ||
2488		   before(tp->urg_seq, tp->copied_seq) ||
2489		   !before(tp->urg_seq, tp->rcv_nxt)) {
2490
2491		answ = tp->rcv_nxt - tp->copied_seq;
2492
2493		/* Subtract 1, if FIN was received */
2494		if (answ && sock_flag(sk, SOCK_DONE))
2495			answ--;
2496	} else {
2497		answ = tp->urg_seq - tp->copied_seq;
2498	}
2499
2500	return answ;
2501}
2502
2503int tcp_peek_len(struct socket *sock);
2504
2505static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2506{
2507	u16 segs_in;
2508
2509	segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2510
2511	/* We update these fields while other threads might
2512	 * read them from tcp_get_info()
2513	 */
2514	WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in);
2515	if (skb->len > tcp_hdrlen(skb))
2516		WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in);
2517}
2518
2519/*
2520 * TCP listen path runs lockless.
2521 * We forced "struct sock" to be const qualified to make sure
2522 * we don't modify one of its field by mistake.
2523 * Here, we increment sk_drops which is an atomic_t, so we can safely
2524 * make sock writable again.
2525 */
2526static inline void tcp_listendrop(const struct sock *sk)
2527{
2528	atomic_inc(&((struct sock *)sk)->sk_drops);
2529	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2530}
2531
2532enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2533
2534/*
2535 * Interface for adding Upper Level Protocols over TCP
2536 */
2537
2538#define TCP_ULP_NAME_MAX	16
2539#define TCP_ULP_MAX		128
2540#define TCP_ULP_BUF_MAX		(TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2541
2542struct tcp_ulp_ops {
2543	struct list_head	list;
2544
2545	/* initialize ulp */
2546	int (*init)(struct sock *sk);
2547	/* update ulp */
2548	void (*update)(struct sock *sk, struct proto *p,
2549		       void (*write_space)(struct sock *sk));
2550	/* cleanup ulp */
2551	void (*release)(struct sock *sk);
2552	/* diagnostic */
2553	int (*get_info)(struct sock *sk, struct sk_buff *skb);
2554	size_t (*get_info_size)(const struct sock *sk);
2555	/* clone ulp */
2556	void (*clone)(const struct request_sock *req, struct sock *newsk,
2557		      const gfp_t priority);
2558
2559	char		name[TCP_ULP_NAME_MAX];
2560	struct module	*owner;
2561};
2562int tcp_register_ulp(struct tcp_ulp_ops *type);
2563void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2564int tcp_set_ulp(struct sock *sk, const char *name);
2565void tcp_get_available_ulp(char *buf, size_t len);
2566void tcp_cleanup_ulp(struct sock *sk);
2567void tcp_update_ulp(struct sock *sk, struct proto *p,
2568		    void (*write_space)(struct sock *sk));
2569
2570#define MODULE_ALIAS_TCP_ULP(name)				\
2571	__MODULE_INFO(alias, alias_userspace, name);		\
2572	__MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2573
2574#ifdef CONFIG_NET_SOCK_MSG
2575struct sk_msg;
2576struct sk_psock;
2577
2578#ifdef CONFIG_BPF_SYSCALL
2579int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore);
2580void tcp_bpf_clone(const struct sock *sk, struct sock *newsk);
2581#endif /* CONFIG_BPF_SYSCALL */
2582
2583#ifdef CONFIG_INET
2584void tcp_eat_skb(struct sock *sk, struct sk_buff *skb);
2585#else
2586static inline void tcp_eat_skb(struct sock *sk, struct sk_buff *skb)
2587{
2588}
2589#endif
2590
2591int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress,
2592			  struct sk_msg *msg, u32 bytes, int flags);
2593#endif /* CONFIG_NET_SOCK_MSG */
2594
2595#if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG)
2596static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk)
2597{
2598}
2599#endif
2600
2601#ifdef CONFIG_CGROUP_BPF
2602static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2603				      struct sk_buff *skb,
2604				      unsigned int end_offset)
2605{
2606	skops->skb = skb;
2607	skops->skb_data_end = skb->data + end_offset;
2608}
2609#else
2610static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2611				      struct sk_buff *skb,
2612				      unsigned int end_offset)
2613{
2614}
2615#endif
2616
2617/* Call BPF_SOCK_OPS program that returns an int. If the return value
2618 * is < 0, then the BPF op failed (for example if the loaded BPF
2619 * program does not support the chosen operation or there is no BPF
2620 * program loaded).
2621 */
2622#ifdef CONFIG_BPF
2623static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2624{
2625	struct bpf_sock_ops_kern sock_ops;
2626	int ret;
2627
2628	memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2629	if (sk_fullsock(sk)) {
2630		sock_ops.is_fullsock = 1;
2631		sock_owned_by_me(sk);
2632	}
2633
2634	sock_ops.sk = sk;
2635	sock_ops.op = op;
2636	if (nargs > 0)
2637		memcpy(sock_ops.args, args, nargs * sizeof(*args));
2638
2639	ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2640	if (ret == 0)
2641		ret = sock_ops.reply;
2642	else
2643		ret = -1;
2644	return ret;
2645}
2646
2647static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2648{
2649	u32 args[2] = {arg1, arg2};
2650
2651	return tcp_call_bpf(sk, op, 2, args);
2652}
2653
2654static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2655				    u32 arg3)
2656{
2657	u32 args[3] = {arg1, arg2, arg3};
2658
2659	return tcp_call_bpf(sk, op, 3, args);
2660}
2661
2662#else
2663static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2664{
2665	return -EPERM;
2666}
2667
2668static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2669{
2670	return -EPERM;
2671}
2672
2673static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2674				    u32 arg3)
2675{
2676	return -EPERM;
2677}
2678
2679#endif
2680
2681static inline u32 tcp_timeout_init(struct sock *sk)
2682{
2683	int timeout;
2684
2685	timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2686
2687	if (timeout <= 0)
2688		timeout = TCP_TIMEOUT_INIT;
2689	return min_t(int, timeout, TCP_RTO_MAX);
2690}
2691
2692static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2693{
2694	int rwnd;
2695
2696	rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2697
2698	if (rwnd < 0)
2699		rwnd = 0;
2700	return rwnd;
2701}
2702
2703static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2704{
2705	return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2706}
2707
2708static inline void tcp_bpf_rtt(struct sock *sk)
2709{
2710	if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG))
2711		tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL);
2712}
2713
2714#if IS_ENABLED(CONFIG_SMC)
2715extern struct static_key_false tcp_have_smc;
2716#endif
2717
2718#if IS_ENABLED(CONFIG_TLS_DEVICE)
2719void clean_acked_data_enable(struct inet_connection_sock *icsk,
2720			     void (*cad)(struct sock *sk, u32 ack_seq));
2721void clean_acked_data_disable(struct inet_connection_sock *icsk);
2722void clean_acked_data_flush(void);
2723#endif
2724
2725DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
2726static inline void tcp_add_tx_delay(struct sk_buff *skb,
2727				    const struct tcp_sock *tp)
2728{
2729	if (static_branch_unlikely(&tcp_tx_delay_enabled))
2730		skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC;
2731}
2732
2733/* Compute Earliest Departure Time for some control packets
2734 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
2735 */
2736static inline u64 tcp_transmit_time(const struct sock *sk)
2737{
2738	if (static_branch_unlikely(&tcp_tx_delay_enabled)) {
2739		u32 delay = (sk->sk_state == TCP_TIME_WAIT) ?
2740			tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay;
2741
2742		return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;
2743	}
2744	return 0;
2745}
2746
2747static inline int tcp_parse_auth_options(const struct tcphdr *th,
2748		const u8 **md5_hash, const struct tcp_ao_hdr **aoh)
2749{
2750	const u8 *md5_tmp, *ao_tmp;
2751	int ret;
2752
2753	ret = tcp_do_parse_auth_options(th, &md5_tmp, &ao_tmp);
2754	if (ret)
2755		return ret;
2756
2757	if (md5_hash)
2758		*md5_hash = md5_tmp;
2759
2760	if (aoh) {
2761		if (!ao_tmp)
2762			*aoh = NULL;
2763		else
2764			*aoh = (struct tcp_ao_hdr *)(ao_tmp - 2);
2765	}
2766
2767	return 0;
2768}
2769
2770static inline bool tcp_ao_required(struct sock *sk, const void *saddr,
2771				   int family, int l3index, bool stat_inc)
2772{
2773#ifdef CONFIG_TCP_AO
2774	struct tcp_ao_info *ao_info;
2775	struct tcp_ao_key *ao_key;
2776
2777	if (!static_branch_unlikely(&tcp_ao_needed.key))
2778		return false;
2779
2780	ao_info = rcu_dereference_check(tcp_sk(sk)->ao_info,
2781					lockdep_sock_is_held(sk));
2782	if (!ao_info)
2783		return false;
2784
2785	ao_key = tcp_ao_do_lookup(sk, l3index, saddr, family, -1, -1);
2786	if (ao_info->ao_required || ao_key) {
2787		if (stat_inc) {
2788			NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOREQUIRED);
2789			atomic64_inc(&ao_info->counters.ao_required);
2790		}
2791		return true;
2792	}
2793#endif
2794	return false;
2795}
2796
2797/* Called with rcu_read_lock() */
2798static inline enum skb_drop_reason
2799tcp_inbound_hash(struct sock *sk, const struct request_sock *req,
2800		 const struct sk_buff *skb,
2801		 const void *saddr, const void *daddr,
2802		 int family, int dif, int sdif)
2803{
2804	const struct tcphdr *th = tcp_hdr(skb);
2805	const struct tcp_ao_hdr *aoh;
2806	const __u8 *md5_location;
2807	int l3index;
2808
2809	/* Invalid option or two times meet any of auth options */
2810	if (tcp_parse_auth_options(th, &md5_location, &aoh)) {
2811		tcp_hash_fail("TCP segment has incorrect auth options set",
2812			      family, skb, "");
2813		return SKB_DROP_REASON_TCP_AUTH_HDR;
2814	}
2815
2816	if (req) {
2817		if (tcp_rsk_used_ao(req) != !!aoh) {
2818			NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOBAD);
2819			tcp_hash_fail("TCP connection can't start/end using TCP-AO",
2820				      family, skb, "%s",
2821				      !aoh ? "missing AO" : "AO signed");
2822			return SKB_DROP_REASON_TCP_AOFAILURE;
2823		}
2824	}
2825
2826	/* sdif set, means packet ingressed via a device
2827	 * in an L3 domain and dif is set to the l3mdev
2828	 */
2829	l3index = sdif ? dif : 0;
2830
2831	/* Fast path: unsigned segments */
2832	if (likely(!md5_location && !aoh)) {
2833		/* Drop if there's TCP-MD5 or TCP-AO key with any rcvid/sndid
2834		 * for the remote peer. On TCP-AO established connection
2835		 * the last key is impossible to remove, so there's
2836		 * always at least one current_key.
2837		 */
2838		if (tcp_ao_required(sk, saddr, family, l3index, true)) {
2839			tcp_hash_fail("AO hash is required, but not found",
2840					family, skb, "L3 index %d", l3index);
2841			return SKB_DROP_REASON_TCP_AONOTFOUND;
2842		}
2843		if (unlikely(tcp_md5_do_lookup(sk, l3index, saddr, family))) {
2844			NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
2845			tcp_hash_fail("MD5 Hash not found",
2846				      family, skb, "L3 index %d", l3index);
2847			return SKB_DROP_REASON_TCP_MD5NOTFOUND;
2848		}
2849		return SKB_NOT_DROPPED_YET;
2850	}
2851
2852	if (aoh)
2853		return tcp_inbound_ao_hash(sk, skb, family, req, l3index, aoh);
2854
2855	return tcp_inbound_md5_hash(sk, skb, saddr, daddr, family,
2856				    l3index, md5_location);
2857}
2858
2859#endif	/* _TCP_H */
v4.10.11
 
   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
 147#define TCP_KEEPALIVE_TIME	(120*60*HZ)	/* two hours */
 148#define TCP_KEEPALIVE_PROBES	9		/* Max of 9 keepalive probes	*/
 149#define TCP_KEEPALIVE_INTVL	(75*HZ)
 150
 151#define MAX_TCP_KEEPIDLE	32767
 152#define MAX_TCP_KEEPINTVL	32767
 153#define MAX_TCP_KEEPCNT		127
 154#define MAX_TCP_SYNCNT		127
 155
 156#define TCP_SYNQ_INTERVAL	(HZ/5)	/* Period of SYNACK timer */
 
 
 
 
 157
 158#define TCP_PAWS_24DAYS	(60 * 60 * 24 * 24)
 159#define TCP_PAWS_MSL	60		/* Per-host timestamps are invalidated
 160					 * after this time. It should be equal
 161					 * (or greater than) TCP_TIMEWAIT_LEN
 162					 * to provide reliability equal to one
 163					 * provided by timewait state.
 164					 */
 165#define TCP_PAWS_WINDOW	1		/* Replay window for per-host
 166					 * timestamps. It must be less than
 167					 * minimal timewait lifetime.
 168					 */
 169/*
 170 *	TCP option
 171 */
 172
 173#define TCPOPT_NOP		1	/* Padding */
 174#define TCPOPT_EOL		0	/* End of options */
 175#define TCPOPT_MSS		2	/* Segment size negotiating */
 176#define TCPOPT_WINDOW		3	/* Window scaling */
 177#define TCPOPT_SACK_PERM        4       /* SACK Permitted */
 178#define TCPOPT_SACK             5       /* SACK Block */
 179#define TCPOPT_TIMESTAMP	8	/* Better RTT estimations/PAWS */
 180#define TCPOPT_MD5SIG		19	/* MD5 Signature (RFC2385) */
 
 
 181#define TCPOPT_FASTOPEN		34	/* Fast open (RFC7413) */
 182#define TCPOPT_EXP		254	/* Experimental */
 183/* Magic number to be after the option value for sharing TCP
 184 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
 185 */
 186#define TCPOPT_FASTOPEN_MAGIC	0xF989
 
 187
 188/*
 189 *     TCP option lengths
 190 */
 191
 192#define TCPOLEN_MSS            4
 193#define TCPOLEN_WINDOW         3
 194#define TCPOLEN_SACK_PERM      2
 195#define TCPOLEN_TIMESTAMP      10
 196#define TCPOLEN_MD5SIG         18
 197#define TCPOLEN_FASTOPEN_BASE  2
 198#define TCPOLEN_EXP_FASTOPEN_BASE  4
 
 199
 200/* But this is what stacks really send out. */
 201#define TCPOLEN_TSTAMP_ALIGNED		12
 202#define TCPOLEN_WSCALE_ALIGNED		4
 203#define TCPOLEN_SACKPERM_ALIGNED	4
 204#define TCPOLEN_SACK_BASE		2
 205#define TCPOLEN_SACK_BASE_ALIGNED	4
 206#define TCPOLEN_SACK_PERBLOCK		8
 207#define TCPOLEN_MD5SIG_ALIGNED		20
 208#define TCPOLEN_MSS_ALIGNED		4
 
 209
 210/* Flags in tp->nonagle */
 211#define TCP_NAGLE_OFF		1	/* Nagle's algo is disabled */
 212#define TCP_NAGLE_CORK		2	/* Socket is corked	    */
 213#define TCP_NAGLE_PUSH		4	/* Cork is overridden for already queued data */
 214
 215/* TCP thin-stream limits */
 216#define TCP_THIN_LINEAR_RETRIES 6       /* After 6 linear retries, do exp. backoff */
 217
 218/* TCP initial congestion window as per rfc6928 */
 219#define TCP_INIT_CWND		10
 220
 221/* Bit Flags for sysctl_tcp_fastopen */
 222#define	TFO_CLIENT_ENABLE	1
 223#define	TFO_SERVER_ENABLE	2
 224#define	TFO_CLIENT_NO_COOKIE	4	/* Data in SYN w/o cookie option */
 225
 226/* Accept SYN data w/o any cookie option */
 227#define	TFO_SERVER_COOKIE_NOT_REQD	0x200
 228
 229/* Force enable TFO on all listeners, i.e., not requiring the
 230 * TCP_FASTOPEN socket option.
 231 */
 232#define	TFO_SERVER_WO_SOCKOPT1	0x400
 233
 234extern struct inet_timewait_death_row tcp_death_row;
 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_limit_output_bytes;
 266extern int sysctl_tcp_challenge_ack_limit;
 267extern int sysctl_tcp_min_tso_segs;
 268extern int sysctl_tcp_min_rtt_wlen;
 269extern int sysctl_tcp_autocorking;
 270extern int sysctl_tcp_invalid_ratelimit;
 271extern int sysctl_tcp_pacing_ss_ratio;
 272extern int sysctl_tcp_pacing_ca_ratio;
 273
 274extern atomic_long_t tcp_memory_allocated;
 
 
 275extern struct percpu_counter tcp_sockets_allocated;
 276extern int tcp_memory_pressure;
 277
 278/* optimized version of sk_under_memory_pressure() for TCP sockets */
 279static inline bool tcp_under_memory_pressure(const struct sock *sk)
 280{
 281	if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
 282	    mem_cgroup_under_socket_pressure(sk->sk_memcg))
 283		return true;
 284
 285	return tcp_memory_pressure;
 286}
 287/*
 288 * The next routines deal with comparing 32 bit unsigned ints
 289 * and worry about wraparound (automatic with unsigned arithmetic).
 290 */
 291
 292static inline bool before(__u32 seq1, __u32 seq2)
 293{
 294        return (__s32)(seq1-seq2) < 0;
 295}
 296#define after(seq2, seq1) 	before(seq1, seq2)
 297
 298/* is s2<=s1<=s3 ? */
 299static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
 300{
 301	return seq3 - seq2 >= seq1 - seq2;
 302}
 303
 304static inline bool tcp_out_of_memory(struct sock *sk)
 305{
 306	if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
 307	    sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
 308		return true;
 309	return false;
 310}
 311
 312void sk_forced_mem_schedule(struct sock *sk, int size);
 313
 314static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
 315{
 316	struct percpu_counter *ocp = sk->sk_prot->orphan_count;
 317	int orphans = percpu_counter_read_positive(ocp);
 
 
 
 
 
 318
 319	if (orphans << shift > sysctl_tcp_max_orphans) {
 320		orphans = percpu_counter_sum_positive(ocp);
 321		if (orphans << shift > sysctl_tcp_max_orphans)
 322			return true;
 323	}
 324	return false;
 325}
 326
 327bool tcp_check_oom(struct sock *sk, int shift);
 328
 329
 330extern struct proto tcp_prot;
 331
 332#define TCP_INC_STATS(net, field)	SNMP_INC_STATS((net)->mib.tcp_statistics, field)
 333#define __TCP_INC_STATS(net, field)	__SNMP_INC_STATS((net)->mib.tcp_statistics, field)
 334#define TCP_DEC_STATS(net, field)	SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
 335#define TCP_ADD_STATS(net, field, val)	SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
 336
 337void tcp_tasklet_init(void);
 338
 339void tcp_v4_err(struct sk_buff *skb, u32);
 340
 341void tcp_shutdown(struct sock *sk, int how);
 342
 343void tcp_v4_early_demux(struct sk_buff *skb);
 344int tcp_v4_rcv(struct sk_buff *skb);
 345
 346int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
 347int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
 348int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
 349		 int flags);
 
 
 
 
 
 
 350void tcp_release_cb(struct sock *sk);
 351void tcp_wfree(struct sk_buff *skb);
 352void tcp_write_timer_handler(struct sock *sk);
 353void tcp_delack_timer_handler(struct sock *sk);
 354int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
 355int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
 356void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
 357			 const struct tcphdr *th, unsigned int len);
 358void tcp_rcv_space_adjust(struct sock *sk);
 359int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
 360void tcp_twsk_destructor(struct sock *sk);
 
 361ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
 362			struct pipe_inode_info *pipe, size_t len,
 363			unsigned int flags);
 
 
 364
 365static inline void tcp_dec_quickack_mode(struct sock *sk,
 366					 const unsigned int pkts)
 367{
 368	struct inet_connection_sock *icsk = inet_csk(sk);
 369
 370	if (icsk->icsk_ack.quick) {
 
 
 
 371		if (pkts >= icsk->icsk_ack.quick) {
 372			icsk->icsk_ack.quick = 0;
 373			/* Leaving quickack mode we deflate ATO. */
 374			icsk->icsk_ack.ato   = TCP_ATO_MIN;
 375		} else
 376			icsk->icsk_ack.quick -= pkts;
 377	}
 378}
 379
 380#define	TCP_ECN_OK		1
 381#define	TCP_ECN_QUEUE_CWR	2
 382#define	TCP_ECN_DEMAND_CWR	4
 383#define	TCP_ECN_SEEN		8
 384
 385enum tcp_tw_status {
 386	TCP_TW_SUCCESS = 0,
 387	TCP_TW_RST = 1,
 388	TCP_TW_ACK = 2,
 389	TCP_TW_SYN = 3
 390};
 391
 392
 393enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
 394					      struct sk_buff *skb,
 395					      const struct tcphdr *th);
 396struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
 397			   struct request_sock *req, bool fastopen);
 398int tcp_child_process(struct sock *parent, struct sock *child,
 399		      struct sk_buff *skb);
 
 400void tcp_enter_loss(struct sock *sk);
 
 401void tcp_clear_retrans(struct tcp_sock *tp);
 402void tcp_update_metrics(struct sock *sk);
 403void tcp_init_metrics(struct sock *sk);
 404void tcp_metrics_init(void);
 405bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
 406			bool paws_check, bool timestamps);
 407bool tcp_remember_stamp(struct sock *sk);
 408bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
 409void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
 410void tcp_disable_fack(struct tcp_sock *tp);
 411void tcp_close(struct sock *sk, long timeout);
 412void tcp_init_sock(struct sock *sk);
 413unsigned int tcp_poll(struct file *file, struct socket *sock,
 
 414		      struct poll_table_struct *wait);
 
 
 415int tcp_getsockopt(struct sock *sk, int level, int optname,
 416		   char __user *optval, int __user *optlen);
 417int tcp_setsockopt(struct sock *sk, int level, int optname,
 418		   char __user *optval, unsigned int optlen);
 419int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
 420			  char __user *optval, int __user *optlen);
 421int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
 422			  char __user *optval, unsigned int optlen);
 423void tcp_set_keepalive(struct sock *sk, int val);
 424void tcp_syn_ack_timeout(const struct request_sock *req);
 425int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
 426		int flags, int *addr_len);
 427void tcp_parse_options(const struct sk_buff *skb,
 
 
 
 
 
 
 
 
 
 
 
 428		       struct tcp_options_received *opt_rx,
 429		       int estab, struct tcp_fastopen_cookie *foc);
 430const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
 431
 432/*
 
 
 
 
 
 
 
 
 
 
 
 433 *	TCP v4 functions exported for the inet6 API
 434 */
 435
 436void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
 437void tcp_v4_mtu_reduced(struct sock *sk);
 438void tcp_req_err(struct sock *sk, u32 seq, bool abort);
 
 439int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
 440struct sock *tcp_create_openreq_child(const struct sock *sk,
 441				      struct request_sock *req,
 442				      struct sk_buff *skb);
 443void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
 444struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
 445				  struct request_sock *req,
 446				  struct dst_entry *dst,
 447				  struct request_sock *req_unhash,
 448				  bool *own_req);
 449int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
 450int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
 451int tcp_connect(struct sock *sk);
 452enum tcp_synack_type {
 453	TCP_SYNACK_NORMAL,
 454	TCP_SYNACK_FASTOPEN,
 455	TCP_SYNACK_COOKIE,
 456};
 457struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
 458				struct request_sock *req,
 459				struct tcp_fastopen_cookie *foc,
 460				enum tcp_synack_type synack_type);
 
 461int tcp_disconnect(struct sock *sk, int flags);
 462
 463void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
 464int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
 465void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
 466
 467/* From syncookies.c */
 468struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
 469				 struct request_sock *req,
 470				 struct dst_entry *dst);
 471int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
 472		      u32 cookie);
 473struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 474#ifdef CONFIG_SYN_COOKIES
 475
 476/* Syncookies use a monotonic timer which increments every 60 seconds.
 477 * This counter is used both as a hash input and partially encoded into
 478 * the cookie value.  A cookie is only validated further if the delta
 479 * between the current counter value and the encoded one is less than this,
 480 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
 481 * the counter advances immediately after a cookie is generated).
 482 */
 483#define MAX_SYNCOOKIE_AGE	2
 484#define TCP_SYNCOOKIE_PERIOD	(60 * HZ)
 485#define TCP_SYNCOOKIE_VALID	(MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
 486
 487/* syncookies: remember time of last synqueue overflow
 488 * But do not dirty this field too often (once per second is enough)
 489 * It is racy as we do not hold a lock, but race is very minor.
 490 */
 491static inline void tcp_synq_overflow(const struct sock *sk)
 492{
 493	unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
 494	unsigned long now = jiffies;
 
 
 
 495
 496	if (time_after(now, last_overflow + HZ))
 497		tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
 
 
 
 
 
 
 
 
 
 
 
 498}
 499
 500/* syncookies: no recent synqueue overflow on this listening socket? */
 501static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
 502{
 503	unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
 
 
 
 
 
 
 
 
 
 
 
 
 
 504
 505	return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
 
 
 
 
 
 
 
 
 
 
 506}
 507
 508static inline u32 tcp_cookie_time(void)
 509{
 510	u64 val = get_jiffies_64();
 511
 512	do_div(val, TCP_SYNCOOKIE_PERIOD);
 513	return val;
 514}
 515
 
 
 
 
 
 
 
 
 
 516u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
 517			      u16 *mssp);
 518__u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
 519__u32 cookie_init_timestamp(struct request_sock *req);
 520bool cookie_timestamp_decode(struct tcp_options_received *opt);
 521bool cookie_ecn_ok(const struct tcp_options_received *opt,
 522		   const struct net *net, const struct dst_entry *dst);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 523
 524/* From net/ipv6/syncookies.c */
 525int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
 526		      u32 cookie);
 527struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
 528
 529u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
 530			      const struct tcphdr *th, u16 *mssp);
 531__u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
 532#endif
 533/* tcp_output.c */
 534
 535u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
 536		     int min_tso_segs);
 537void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
 538			       int nonagle);
 539bool tcp_may_send_now(struct sock *sk);
 540int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
 541int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
 542void tcp_retransmit_timer(struct sock *sk);
 543void tcp_xmit_retransmit_queue(struct sock *);
 544void tcp_simple_retransmit(struct sock *);
 
 545int tcp_trim_head(struct sock *, struct sk_buff *, u32);
 546int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
 
 
 
 
 
 
 547
 548void tcp_send_probe0(struct sock *);
 549void tcp_send_partial(struct sock *);
 550int tcp_write_wakeup(struct sock *, int mib);
 551void tcp_send_fin(struct sock *sk);
 552void tcp_send_active_reset(struct sock *sk, gfp_t priority);
 553int tcp_send_synack(struct sock *);
 554void tcp_push_one(struct sock *, unsigned int mss_now);
 
 555void tcp_send_ack(struct sock *sk);
 556void tcp_send_delayed_ack(struct sock *sk);
 557void tcp_send_loss_probe(struct sock *sk);
 558bool tcp_schedule_loss_probe(struct sock *sk);
 559void tcp_skb_collapse_tstamp(struct sk_buff *skb,
 560			     const struct sk_buff *next_skb);
 561
 562/* tcp_input.c */
 563void tcp_resume_early_retransmit(struct sock *sk);
 564void tcp_rearm_rto(struct sock *sk);
 565void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
 566void tcp_reset(struct sock *sk);
 567void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
 568void tcp_fin(struct sock *sk);
 
 
 569
 570/* tcp_timer.c */
 571void tcp_init_xmit_timers(struct sock *);
 572static inline void tcp_clear_xmit_timers(struct sock *sk)
 573{
 
 
 
 
 
 
 574	inet_csk_clear_xmit_timers(sk);
 575}
 576
 577unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
 578unsigned int tcp_current_mss(struct sock *sk);
 
 579
 580/* Bound MSS / TSO packet size with the half of the window */
 581static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
 582{
 583	int cutoff;
 584
 585	/* When peer uses tiny windows, there is no use in packetizing
 586	 * to sub-MSS pieces for the sake of SWS or making sure there
 587	 * are enough packets in the pipe for fast recovery.
 588	 *
 589	 * On the other hand, for extremely large MSS devices, handling
 590	 * smaller than MSS windows in this way does make sense.
 591	 */
 592	if (tp->max_window > TCP_MSS_DEFAULT)
 593		cutoff = (tp->max_window >> 1);
 594	else
 595		cutoff = tp->max_window;
 596
 597	if (cutoff && pktsize > cutoff)
 598		return max_t(int, cutoff, 68U - tp->tcp_header_len);
 599	else
 600		return pktsize;
 601}
 602
 603/* tcp.c */
 604void tcp_get_info(struct sock *, struct tcp_info *);
 605
 606/* Read 'sendfile()'-style from a TCP socket */
 607int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
 608		  sk_read_actor_t recv_actor);
 
 
 
 609
 610void tcp_initialize_rcv_mss(struct sock *sk);
 611
 612int tcp_mtu_to_mss(struct sock *sk, int pmtu);
 613int tcp_mss_to_mtu(struct sock *sk, int mss);
 614void tcp_mtup_init(struct sock *sk);
 615void tcp_init_buffer_space(struct sock *sk);
 616
 617static inline void tcp_bound_rto(const struct sock *sk)
 618{
 619	if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
 620		inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
 621}
 622
 623static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
 624{
 625	return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
 626}
 627
 628static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
 629{
 
 
 
 
 630	tp->pred_flags = htonl((tp->tcp_header_len << 26) |
 631			       ntohl(TCP_FLAG_ACK) |
 632			       snd_wnd);
 633}
 634
 635static inline void tcp_fast_path_on(struct tcp_sock *tp)
 636{
 637	__tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
 638}
 639
 640static inline void tcp_fast_path_check(struct sock *sk)
 641{
 642	struct tcp_sock *tp = tcp_sk(sk);
 643
 644	if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
 645	    tp->rcv_wnd &&
 646	    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
 647	    !tp->urg_data)
 648		tcp_fast_path_on(tp);
 649}
 650
 
 
 651/* Compute the actual rto_min value */
 652static inline u32 tcp_rto_min(struct sock *sk)
 653{
 654	const struct dst_entry *dst = __sk_dst_get(sk);
 655	u32 rto_min = TCP_RTO_MIN;
 656
 657	if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
 658		rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
 659	return rto_min;
 660}
 661
 662static inline u32 tcp_rto_min_us(struct sock *sk)
 663{
 664	return jiffies_to_usecs(tcp_rto_min(sk));
 665}
 666
 667static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
 668{
 669	return dst_metric_locked(dst, RTAX_CC_ALGO);
 670}
 671
 672/* Minimum RTT in usec. ~0 means not available. */
 673static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
 674{
 675	return minmax_get(&tp->rtt_min);
 676}
 677
 678/* Compute the actual receive window we are currently advertising.
 679 * Rcv_nxt can be after the window if our peer push more data
 680 * than the offered window.
 681 */
 682static inline u32 tcp_receive_window(const struct tcp_sock *tp)
 683{
 684	s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
 685
 686	if (win < 0)
 687		win = 0;
 688	return (u32) win;
 689}
 690
 691/* Choose a new window, without checks for shrinking, and without
 692 * scaling applied to the result.  The caller does these things
 693 * if necessary.  This is a "raw" window selection.
 694 */
 695u32 __tcp_select_window(struct sock *sk);
 696
 697void tcp_send_window_probe(struct sock *sk);
 698
 699/* TCP timestamps are only 32-bits, this causes a slight
 700 * complication on 64-bit systems since we store a snapshot
 701 * of jiffies in the buffer control blocks below.  We decided
 702 * to use only the low 32-bits of jiffies and hide the ugly
 703 * casts with the following macro.
 704 */
 705#define tcp_time_stamp		((__u32)(jiffies))
 706
 707static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 708{
 709	return skb->skb_mstamp.stamp_jiffies;
 710}
 711
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 712
 713#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
 714
 715#define TCPHDR_FIN 0x01
 716#define TCPHDR_SYN 0x02
 717#define TCPHDR_RST 0x04
 718#define TCPHDR_PSH 0x08
 719#define TCPHDR_ACK 0x10
 720#define TCPHDR_URG 0x20
 721#define TCPHDR_ECE 0x40
 722#define TCPHDR_CWR 0x80
 723
 724#define TCPHDR_SYN_ECN	(TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
 725
 726/* This is what the send packet queuing engine uses to pass
 727 * TCP per-packet control information to the transmission code.
 728 * We also store the host-order sequence numbers in here too.
 729 * This is 44 bytes if IPV6 is enabled.
 730 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
 731 */
 732struct tcp_skb_cb {
 733	__u32		seq;		/* Starting sequence number	*/
 734	__u32		end_seq;	/* SEQ + FIN + SYN + datalen	*/
 735	union {
 736		/* Note : tcp_tw_isn is used in input path only
 737		 *	  (isn chosen by tcp_timewait_state_process())
 738		 *
 739		 * 	  tcp_gso_segs/size are used in write queue only,
 740		 *	  cf tcp_skb_pcount()/tcp_skb_mss()
 741		 */
 742		__u32		tcp_tw_isn;
 743		struct {
 744			u16	tcp_gso_segs;
 745			u16	tcp_gso_size;
 746		};
 747	};
 748	__u8		tcp_flags;	/* TCP header flags. (tcp[13])	*/
 749
 750	__u8		sacked;		/* State flags for SACK/FACK.	*/
 751#define TCPCB_SACKED_ACKED	0x01	/* SKB ACK'd by a SACK block	*/
 752#define TCPCB_SACKED_RETRANS	0x02	/* SKB retransmitted		*/
 753#define TCPCB_LOST		0x04	/* SKB is lost			*/
 754#define TCPCB_TAGBITS		0x07	/* All tag bits			*/
 755#define TCPCB_REPAIRED		0x10	/* SKB repaired (no skb_mstamp)	*/
 756#define TCPCB_EVER_RETRANS	0x80	/* Ever retransmitted frame	*/
 757#define TCPCB_RETRANS		(TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
 758				TCPCB_REPAIRED)
 759
 760	__u8		ip_dsfield;	/* IPv4 tos or IPv6 dsfield	*/
 761	__u8		txstamp_ack:1,	/* Record TX timestamp for ack? */
 762			eor:1,		/* Is skb MSG_EOR marked? */
 763			unused:6;
 
 764	__u32		ack_seq;	/* Sequence number ACK'd	*/
 765	union {
 766		struct {
 
 767			/* There is space for up to 24 bytes */
 768			__u32 in_flight:30,/* Bytes in flight at transmit */
 769			      is_app_limited:1, /* cwnd not fully used? */
 770			      unused:1;
 771			/* pkts S/ACKed so far upon tx of skb, incl retrans: */
 772			__u32 delivered;
 773			/* start of send pipeline phase */
 774			struct skb_mstamp first_tx_mstamp;
 775			/* when we reached the "delivered" count */
 776			struct skb_mstamp delivered_mstamp;
 777		} tx;   /* only used for outgoing skbs */
 778		union {
 779			struct inet_skb_parm	h4;
 780#if IS_ENABLED(CONFIG_IPV6)
 781			struct inet6_skb_parm	h6;
 782#endif
 783		} header;	/* For incoming skbs */
 784	};
 785};
 786
 787#define TCP_SKB_CB(__skb)	((struct tcp_skb_cb *)&((__skb)->cb[0]))
 788
 
 789
 790#if IS_ENABLED(CONFIG_IPV6)
 791/* This is the variant of inet6_iif() that must be used by TCP,
 792 * as TCP moves IP6CB into a different location in skb->cb[]
 793 */
 794static inline int tcp_v6_iif(const struct sk_buff *skb)
 795{
 
 
 
 
 
 796	bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
 797
 798	return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
 799}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 800#endif
 801
 802/* TCP_SKB_CB reference means this can not be used from early demux */
 803static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
 804{
 805#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
 806	if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
 807	    skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
 808		return true;
 809#endif
 810	return false;
 811}
 812
 813/* Due to TSO, an SKB can be composed of multiple actual
 814 * packets.  To keep these tracked properly, we use this.
 815 */
 816static inline int tcp_skb_pcount(const struct sk_buff *skb)
 817{
 818	return TCP_SKB_CB(skb)->tcp_gso_segs;
 819}
 820
 821static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
 822{
 823	TCP_SKB_CB(skb)->tcp_gso_segs = segs;
 824}
 825
 826static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
 827{
 828	TCP_SKB_CB(skb)->tcp_gso_segs += segs;
 829}
 830
 831/* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
 832static inline int tcp_skb_mss(const struct sk_buff *skb)
 833{
 834	return TCP_SKB_CB(skb)->tcp_gso_size;
 835}
 836
 837static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
 838{
 839	return likely(!TCP_SKB_CB(skb)->eor);
 840}
 841
 
 
 
 
 
 
 
 
 842/* Events passed to congestion control interface */
 843enum tcp_ca_event {
 844	CA_EVENT_TX_START,	/* first transmit when no packets in flight */
 845	CA_EVENT_CWND_RESTART,	/* congestion window restart */
 846	CA_EVENT_COMPLETE_CWR,	/* end of congestion recovery */
 847	CA_EVENT_LOSS,		/* loss timeout */
 848	CA_EVENT_ECN_NO_CE,	/* ECT set, but not CE marked */
 849	CA_EVENT_ECN_IS_CE,	/* received CE marked IP packet */
 850	CA_EVENT_DELAYED_ACK,	/* Delayed ack is sent */
 851	CA_EVENT_NON_DELAYED_ACK,
 852};
 853
 854/* Information about inbound ACK, passed to cong_ops->in_ack_event() */
 855enum tcp_ca_ack_event_flags {
 856	CA_ACK_SLOWPATH		= (1 << 0),	/* In slow path processing */
 857	CA_ACK_WIN_UPDATE	= (1 << 1),	/* ACK updated window */
 858	CA_ACK_ECE		= (1 << 2),	/* ECE bit is set on ack */
 859};
 860
 861/*
 862 * Interface for adding new TCP congestion control handlers
 863 */
 864#define TCP_CA_NAME_MAX	16
 865#define TCP_CA_MAX	128
 866#define TCP_CA_BUF_MAX	(TCP_CA_NAME_MAX*TCP_CA_MAX)
 867
 868#define TCP_CA_UNSPEC	0
 869
 870/* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
 871#define TCP_CONG_NON_RESTRICTED 0x1
 872/* Requires ECN/ECT set on all packets */
 873#define TCP_CONG_NEEDS_ECN	0x2
 
 874
 875union tcp_cc_info;
 876
 877struct ack_sample {
 878	u32 pkts_acked;
 879	s32 rtt_us;
 880	u32 in_flight;
 881};
 882
 883/* A rate sample measures the number of (original/retransmitted) data
 884 * packets delivered "delivered" over an interval of time "interval_us".
 885 * The tcp_rate.c code fills in the rate sample, and congestion
 886 * control modules that define a cong_control function to run at the end
 887 * of ACK processing can optionally chose to consult this sample when
 888 * setting cwnd and pacing rate.
 889 * A sample is invalid if "delivered" or "interval_us" is negative.
 890 */
 891struct rate_sample {
 892	struct	skb_mstamp prior_mstamp; /* starting timestamp for interval */
 893	u32  prior_delivered;	/* tp->delivered at "prior_mstamp" */
 
 894	s32  delivered;		/* number of packets delivered over interval */
 
 895	long interval_us;	/* time for tp->delivered to incr "delivered" */
 
 
 896	long rtt_us;		/* RTT of last (S)ACKed packet (or -1) */
 897	int  losses;		/* number of packets marked lost upon ACK */
 898	u32  acked_sacked;	/* number of packets newly (S)ACKed upon ACK */
 899	u32  prior_in_flight;	/* in flight before this ACK */
 
 900	bool is_app_limited;	/* is sample from packet with bubble in pipe? */
 901	bool is_retrans;	/* is sample from retransmission? */
 
 902};
 903
 904struct tcp_congestion_ops {
 905	struct list_head	list;
 906	u32 key;
 907	u32 flags;
 908
 909	/* initialize private data (optional) */
 910	void (*init)(struct sock *sk);
 911	/* cleanup private data  (optional) */
 912	void (*release)(struct sock *sk);
 913
 914	/* return slow start threshold (required) */
 915	u32 (*ssthresh)(struct sock *sk);
 
 916	/* do new cwnd calculation (required) */
 917	void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
 
 918	/* call before changing ca_state (optional) */
 919	void (*set_state)(struct sock *sk, u8 new_state);
 
 920	/* call when cwnd event occurs (optional) */
 921	void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
 
 922	/* call when ack arrives (optional) */
 923	void (*in_ack_event)(struct sock *sk, u32 flags);
 924	/* new value of cwnd after loss (optional) */
 925	u32  (*undo_cwnd)(struct sock *sk);
 926	/* hook for packet ack accounting (optional) */
 927	void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
 928	/* suggest number of segments for each skb to transmit (optional) */
 929	u32 (*tso_segs_goal)(struct sock *sk);
 930	/* returns the multiplier used in tcp_sndbuf_expand (optional) */
 931	u32 (*sndbuf_expand)(struct sock *sk);
 932	/* call when packets are delivered to update cwnd and pacing rate,
 933	 * after all the ca_state processing. (optional)
 934	 */
 935	void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
 
 
 
 
 
 
 
 
 936	/* get info for inet_diag (optional) */
 937	size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
 938			   union tcp_cc_info *info);
 939
 940	char 		name[TCP_CA_NAME_MAX];
 941	struct module 	*owner;
 942};
 
 
 
 
 
 
 
 
 943
 944int tcp_register_congestion_control(struct tcp_congestion_ops *type);
 945void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
 
 
 
 946
 947void tcp_assign_congestion_control(struct sock *sk);
 948void tcp_init_congestion_control(struct sock *sk);
 949void tcp_cleanup_congestion_control(struct sock *sk);
 950int tcp_set_default_congestion_control(const char *name);
 951void tcp_get_default_congestion_control(char *name);
 952void tcp_get_available_congestion_control(char *buf, size_t len);
 953void tcp_get_allowed_congestion_control(char *buf, size_t len);
 954int tcp_set_allowed_congestion_control(char *allowed);
 955int tcp_set_congestion_control(struct sock *sk, const char *name);
 
 956u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
 957void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
 958
 959u32 tcp_reno_ssthresh(struct sock *sk);
 960u32 tcp_reno_undo_cwnd(struct sock *sk);
 961void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
 962extern struct tcp_congestion_ops tcp_reno;
 963
 
 964struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
 965u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
 966#ifdef CONFIG_INET
 967char *tcp_ca_get_name_by_key(u32 key, char *buffer);
 968#else
 969static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
 970{
 971	return NULL;
 972}
 973#endif
 974
 975static inline bool tcp_ca_needs_ecn(const struct sock *sk)
 976{
 977	const struct inet_connection_sock *icsk = inet_csk(sk);
 978
 979	return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
 980}
 981
 982static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
 983{
 984	struct inet_connection_sock *icsk = inet_csk(sk);
 985
 986	if (icsk->icsk_ca_ops->set_state)
 987		icsk->icsk_ca_ops->set_state(sk, ca_state);
 988	icsk->icsk_ca_state = ca_state;
 989}
 990
 991static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
 992{
 993	const struct inet_connection_sock *icsk = inet_csk(sk);
 994
 995	if (icsk->icsk_ca_ops->cwnd_event)
 996		icsk->icsk_ca_ops->cwnd_event(sk, event);
 997}
 998
 
 
 
 999/* From tcp_rate.c */
1000void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1001void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1002			    struct rate_sample *rs);
1003void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1004		  struct skb_mstamp *now, struct rate_sample *rs);
1005void tcp_rate_check_app_limited(struct sock *sk);
1006
 
 
 
 
 
1007/* These functions determine how the current flow behaves in respect of SACK
1008 * handling. SACK is negotiated with the peer, and therefore it can vary
1009 * between different flows.
1010 *
1011 * tcp_is_sack - SACK enabled
1012 * tcp_is_reno - No SACK
1013 * tcp_is_fack - FACK enabled, implies SACK enabled
1014 */
1015static inline int tcp_is_sack(const struct tcp_sock *tp)
1016{
1017	return tp->rx_opt.sack_ok;
1018}
1019
1020static inline bool tcp_is_reno(const struct tcp_sock *tp)
1021{
1022	return !tcp_is_sack(tp);
1023}
1024
1025static inline bool tcp_is_fack(const struct tcp_sock *tp)
1026{
1027	return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
1028}
1029
1030static inline void tcp_enable_fack(struct tcp_sock *tp)
1031{
1032	tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
1033}
1034
1035/* TCP early-retransmit (ER) is similar to but more conservative than
1036 * the thin-dupack feature.  Enable ER only if thin-dupack is disabled.
1037 */
1038static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
1039{
1040	struct net *net = sock_net((struct sock *)tp);
1041
1042	tp->do_early_retrans = sysctl_tcp_early_retrans &&
1043		sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
1044		net->ipv4.sysctl_tcp_reordering == 3;
1045}
1046
1047static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
1048{
1049	tp->do_early_retrans = 0;
1050}
1051
1052static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1053{
1054	return tp->sacked_out + tp->lost_out;
1055}
1056
1057/* This determines how many packets are "in the network" to the best
1058 * of our knowledge.  In many cases it is conservative, but where
1059 * detailed information is available from the receiver (via SACK
1060 * blocks etc.) we can make more aggressive calculations.
1061 *
1062 * Use this for decisions involving congestion control, use just
1063 * tp->packets_out to determine if the send queue is empty or not.
1064 *
1065 * Read this equation as:
1066 *
1067 *	"Packets sent once on transmission queue" MINUS
1068 *	"Packets left network, but not honestly ACKed yet" PLUS
1069 *	"Packets fast retransmitted"
1070 */
1071static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1072{
1073	return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1074}
1075
1076#define TCP_INFINITE_SSTHRESH	0x7fffffff
1077
 
 
 
 
 
 
 
 
 
 
 
1078static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1079{
1080	return tp->snd_cwnd < tp->snd_ssthresh;
1081}
1082
1083static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1084{
1085	return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1086}
1087
1088static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1089{
1090	return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1091	       (1 << inet_csk(sk)->icsk_ca_state);
1092}
1093
1094/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1095 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1096 * ssthresh.
1097 */
1098static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1099{
1100	const struct tcp_sock *tp = tcp_sk(sk);
1101
1102	if (tcp_in_cwnd_reduction(sk))
1103		return tp->snd_ssthresh;
1104	else
1105		return max(tp->snd_ssthresh,
1106			   ((tp->snd_cwnd >> 1) +
1107			    (tp->snd_cwnd >> 2)));
1108}
1109
1110/* Use define here intentionally to get WARN_ON location shown at the caller */
1111#define tcp_verify_left_out(tp)	WARN_ON(tcp_left_out(tp) > tp->packets_out)
1112
1113void tcp_enter_cwr(struct sock *sk);
1114__u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1115
1116/* The maximum number of MSS of available cwnd for which TSO defers
1117 * sending if not using sysctl_tcp_tso_win_divisor.
1118 */
1119static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1120{
1121	return 3;
1122}
1123
1124/* Returns end sequence number of the receiver's advertised window */
1125static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1126{
1127	return tp->snd_una + tp->snd_wnd;
1128}
1129
1130/* We follow the spirit of RFC2861 to validate cwnd but implement a more
1131 * flexible approach. The RFC suggests cwnd should not be raised unless
1132 * it was fully used previously. And that's exactly what we do in
1133 * congestion avoidance mode. But in slow start we allow cwnd to grow
1134 * as long as the application has used half the cwnd.
1135 * Example :
1136 *    cwnd is 10 (IW10), but application sends 9 frames.
1137 *    We allow cwnd to reach 18 when all frames are ACKed.
1138 * This check is safe because it's as aggressive as slow start which already
1139 * risks 100% overshoot. The advantage is that we discourage application to
1140 * either send more filler packets or data to artificially blow up the cwnd
1141 * usage, and allow application-limited process to probe bw more aggressively.
1142 */
1143static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1144{
1145	const struct tcp_sock *tp = tcp_sk(sk);
1146
 
 
 
1147	/* If in slow start, ensure cwnd grows to twice what was ACKed. */
1148	if (tcp_in_slow_start(tp))
1149		return tp->snd_cwnd < 2 * tp->max_packets_out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1150
1151	return tp->is_cwnd_limited;
 
 
 
 
 
 
1152}
1153
1154/* Something is really bad, we could not queue an additional packet,
1155 * because qdisc is full or receiver sent a 0 window.
1156 * We do not want to add fuel to the fire, or abort too early,
1157 * so make sure the timer we arm now is at least 200ms in the future,
1158 * regardless of current icsk_rto value (as it could be ~2ms)
1159 */
1160static inline unsigned long tcp_probe0_base(const struct sock *sk)
1161{
1162	return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1163}
1164
1165/* Variant of inet_csk_rto_backoff() used for zero window probes */
1166static inline unsigned long tcp_probe0_when(const struct sock *sk,
1167					    unsigned long max_when)
1168{
1169	u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
 
 
1170
1171	return (unsigned long)min_t(u64, when, max_when);
1172}
1173
1174static inline void tcp_check_probe_timer(struct sock *sk)
1175{
1176	if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1177		inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1178					  tcp_probe0_base(sk), TCP_RTO_MAX);
1179}
1180
1181static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1182{
1183	tp->snd_wl1 = seq;
1184}
1185
1186static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1187{
1188	tp->snd_wl1 = seq;
1189}
1190
1191/*
1192 * Calculate(/check) TCP checksum
1193 */
1194static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1195				   __be32 daddr, __wsum base)
1196{
1197	return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1198}
1199
1200static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1201{
1202	return __skb_checksum_complete(skb);
1203}
1204
1205static inline bool tcp_checksum_complete(struct sk_buff *skb)
1206{
1207	return !skb_csum_unnecessary(skb) &&
1208		__tcp_checksum_complete(skb);
1209}
1210
1211/* Prequeue for VJ style copy to user, combined with checksumming. */
 
1212
1213static inline void tcp_prequeue_init(struct tcp_sock *tp)
1214{
1215	tp->ucopy.task = NULL;
1216	tp->ucopy.len = 0;
1217	tp->ucopy.memory = 0;
1218	skb_queue_head_init(&tp->ucopy.prequeue);
1219}
1220
1221bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1222bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1223int tcp_filter(struct sock *sk, struct sk_buff *skb);
1224
1225#undef STATE_TRACE
1226
1227#ifdef STATE_TRACE
1228static const char *statename[]={
1229	"Unused","Established","Syn Sent","Syn Recv",
1230	"Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1231	"Close Wait","Last ACK","Listen","Closing"
1232};
1233#endif
1234void tcp_set_state(struct sock *sk, int state);
1235
1236void tcp_done(struct sock *sk);
1237
1238int tcp_abort(struct sock *sk, int err);
1239
1240static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1241{
1242	rx_opt->dsack = 0;
1243	rx_opt->num_sacks = 0;
1244}
1245
1246u32 tcp_default_init_rwnd(u32 mss);
1247void tcp_cwnd_restart(struct sock *sk, s32 delta);
1248
1249static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1250{
 
1251	struct tcp_sock *tp = tcp_sk(sk);
1252	s32 delta;
1253
1254	if (!sysctl_tcp_slow_start_after_idle || tp->packets_out)
 
1255		return;
1256	delta = tcp_time_stamp - tp->lsndtime;
1257	if (delta > inet_csk(sk)->icsk_rto)
1258		tcp_cwnd_restart(sk, delta);
1259}
1260
1261/* Determine a window scaling and initial window to offer. */
1262void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
 
1263			       __u32 *window_clamp, int wscale_ok,
1264			       __u8 *rcv_wscale, __u32 init_rcv_wnd);
1265
1266static inline int tcp_win_from_space(int space)
1267{
1268	return sysctl_tcp_adv_win_scale<=0 ?
1269		(space>>(-sysctl_tcp_adv_win_scale)) :
1270		space - (space>>sysctl_tcp_adv_win_scale);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1271}
1272
1273/* Note: caller must be prepared to deal with negative returns */
1274static inline int tcp_space(const struct sock *sk)
1275{
1276	return tcp_win_from_space(sk->sk_rcvbuf -
 
1277				  atomic_read(&sk->sk_rmem_alloc));
1278}
1279
1280static inline int tcp_full_space(const struct sock *sk)
1281{
1282	return tcp_win_from_space(sk->sk_rcvbuf);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1283}
1284
1285extern void tcp_openreq_init_rwin(struct request_sock *req,
1286				  const struct sock *sk_listener,
1287				  const struct dst_entry *dst);
1288
1289void tcp_enter_memory_pressure(struct sock *sk);
 
1290
1291static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1292{
1293	struct net *net = sock_net((struct sock *)tp);
 
1294
1295	return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
 
 
 
 
 
1296}
1297
1298static inline int keepalive_time_when(const struct tcp_sock *tp)
1299{
1300	struct net *net = sock_net((struct sock *)tp);
 
 
 
 
1301
1302	return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1303}
1304
1305static inline int keepalive_probes(const struct tcp_sock *tp)
1306{
1307	struct net *net = sock_net((struct sock *)tp);
 
1308
1309	return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
 
 
 
 
 
1310}
1311
1312static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1313{
1314	const struct inet_connection_sock *icsk = &tp->inet_conn;
1315
1316	return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1317			  tcp_time_stamp - tp->rcv_tstamp);
1318}
1319
1320static inline int tcp_fin_time(const struct sock *sk)
1321{
1322	int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
 
1323	const int rto = inet_csk(sk)->icsk_rto;
1324
1325	if (fin_timeout < (rto << 2) - (rto >> 1))
1326		fin_timeout = (rto << 2) - (rto >> 1);
1327
1328	return fin_timeout;
1329}
1330
1331static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1332				  int paws_win)
1333{
1334	if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1335		return true;
1336	if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
 
1337		return true;
1338	/*
1339	 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1340	 * then following tcp messages have valid values. Ignore 0 value,
1341	 * or else 'negative' tsval might forbid us to accept their packets.
1342	 */
1343	if (!rx_opt->ts_recent)
1344		return true;
1345	return false;
1346}
1347
1348static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1349				   int rst)
1350{
1351	if (tcp_paws_check(rx_opt, 0))
1352		return false;
1353
1354	/* RST segments are not recommended to carry timestamp,
1355	   and, if they do, it is recommended to ignore PAWS because
1356	   "their cleanup function should take precedence over timestamps."
1357	   Certainly, it is mistake. It is necessary to understand the reasons
1358	   of this constraint to relax it: if peer reboots, clock may go
1359	   out-of-sync and half-open connections will not be reset.
1360	   Actually, the problem would be not existing if all
1361	   the implementations followed draft about maintaining clock
1362	   via reboots. Linux-2.2 DOES NOT!
1363
1364	   However, we can relax time bounds for RST segments to MSL.
1365	 */
1366	if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
 
1367		return false;
1368	return true;
1369}
1370
1371bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1372			  int mib_idx, u32 *last_oow_ack_time);
1373
1374static inline void tcp_mib_init(struct net *net)
1375{
1376	/* See RFC 2012 */
1377	TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1378	TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1379	TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1380	TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1381}
1382
1383/* from STCP */
1384static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1385{
1386	tp->lost_skb_hint = NULL;
1387}
1388
1389static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1390{
1391	tcp_clear_retrans_hints_partial(tp);
1392	tp->retransmit_skb_hint = NULL;
1393}
1394
1395union tcp_md5_addr {
1396	struct in_addr  a4;
1397#if IS_ENABLED(CONFIG_IPV6)
1398	struct in6_addr	a6;
1399#endif
1400};
1401
1402/* - key database */
1403struct tcp_md5sig_key {
1404	struct hlist_node	node;
1405	u8			keylen;
1406	u8			family; /* AF_INET or AF_INET6 */
 
 
1407	union tcp_md5_addr	addr;
 
1408	u8			key[TCP_MD5SIG_MAXKEYLEN];
1409	struct rcu_head		rcu;
1410};
1411
1412/* - sock block */
1413struct tcp_md5sig_info {
1414	struct hlist_head	head;
1415	struct rcu_head		rcu;
1416};
1417
1418/* - pseudo header */
1419struct tcp4_pseudohdr {
1420	__be32		saddr;
1421	__be32		daddr;
1422	__u8		pad;
1423	__u8		protocol;
1424	__be16		len;
1425};
1426
1427struct tcp6_pseudohdr {
1428	struct in6_addr	saddr;
1429	struct in6_addr daddr;
1430	__be32		len;
1431	__be32		protocol;	/* including padding */
1432};
1433
1434union tcp_md5sum_block {
1435	struct tcp4_pseudohdr ip4;
1436#if IS_ENABLED(CONFIG_IPV6)
1437	struct tcp6_pseudohdr ip6;
1438#endif
1439};
1440
1441/* - pool: digest algorithm, hash description and scratch buffer */
1442struct tcp_md5sig_pool {
1443	struct ahash_request	*md5_req;
1444	void			*scratch;
1445};
1446
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1447/* - functions */
1448int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1449			const struct sock *sk, const struct sk_buff *skb);
1450int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1451		   int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
 
 
 
 
 
1452int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1453		   int family);
 
1454struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1455					 const struct sock *addr_sk);
1456
1457#ifdef CONFIG_TCP_MD5SIG
1458struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1459					 const union tcp_md5_addr *addr,
1460					 int family);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1461#define tcp_twsk_md5_key(twsk)	((twsk)->tw_md5_key)
1462#else
1463static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1464					 const union tcp_md5_addr *addr,
1465					 int family)
1466{
1467	return NULL;
1468}
1469#define tcp_twsk_md5_key(twsk)	NULL
1470#endif
1471
1472bool tcp_alloc_md5sig_pool(void);
 
 
 
 
 
1473
1474struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1475static inline void tcp_put_md5sig_pool(void)
 
 
1476{
1477	local_bh_enable();
1478}
 
 
1479
1480int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1481			  unsigned int header_len);
1482int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
 
 
 
1483		     const struct tcp_md5sig_key *key);
1484
1485/* From tcp_fastopen.c */
1486void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1487			    struct tcp_fastopen_cookie *cookie, int *syn_loss,
1488			    unsigned long *last_syn_loss);
1489void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1490			    struct tcp_fastopen_cookie *cookie, bool syn_lost,
1491			    u16 try_exp);
1492struct tcp_fastopen_request {
1493	/* Fast Open cookie. Size 0 means a cookie request */
1494	struct tcp_fastopen_cookie	cookie;
1495	struct msghdr			*data;  /* data in MSG_FASTOPEN */
1496	size_t				size;
1497	int				copied;	/* queued in tcp_connect() */
 
1498};
1499void tcp_free_fastopen_req(struct tcp_sock *tp);
1500
1501extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1502int tcp_fastopen_reset_cipher(void *key, unsigned int len);
 
 
 
1503void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1504struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1505			      struct request_sock *req,
1506			      struct tcp_fastopen_cookie *foc,
1507			      struct dst_entry *dst);
1508void tcp_fastopen_init_key_once(bool publish);
1509#define TCP_FASTOPEN_KEY_LENGTH 16
 
 
 
 
 
 
1510
1511/* Fastopen key context */
1512struct tcp_fastopen_context {
1513	struct crypto_cipher	*tfm;
1514	__u8			key[TCP_FASTOPEN_KEY_LENGTH];
1515	struct rcu_head		rcu;
1516};
1517
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1518/* Latencies incurred by various limits for a sender. They are
1519 * chronograph-like stats that are mutually exclusive.
1520 */
1521enum tcp_chrono {
1522	TCP_CHRONO_UNSPEC,
1523	TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1524	TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1525	TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1526	__TCP_CHRONO_MAX,
1527};
1528
1529void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1530void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1531
1532/* write queue abstraction */
1533static inline void tcp_write_queue_purge(struct sock *sk)
 
 
1534{
1535	struct sk_buff *skb;
 
 
 
 
 
 
1536
1537	tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
1538	while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1539		sk_wmem_free_skb(sk, skb);
1540	sk_mem_reclaim(sk);
1541	tcp_clear_all_retrans_hints(tcp_sk(sk));
1542}
1543
1544static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1545{
1546	return skb_peek(&sk->sk_write_queue);
1547}
1548
1549static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1550{
1551	return skb_peek_tail(&sk->sk_write_queue);
1552}
1553
1554static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1555						   const struct sk_buff *skb)
1556{
1557	return skb_queue_next(&sk->sk_write_queue, skb);
1558}
1559
1560static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1561						   const struct sk_buff *skb)
1562{
1563	return skb_queue_prev(&sk->sk_write_queue, skb);
1564}
1565
1566#define tcp_for_write_queue(skb, sk)					\
1567	skb_queue_walk(&(sk)->sk_write_queue, skb)
1568
1569#define tcp_for_write_queue_from(skb, sk)				\
1570	skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1571
1572#define tcp_for_write_queue_from_safe(skb, tmp, sk)			\
1573	skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1574
1575static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1576{
1577	return sk->sk_send_head;
1578}
1579
1580static inline bool tcp_skb_is_last(const struct sock *sk,
1581				   const struct sk_buff *skb)
1582{
1583	return skb_queue_is_last(&sk->sk_write_queue, skb);
1584}
1585
1586static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
 
 
 
 
 
 
 
1587{
1588	if (tcp_skb_is_last(sk, skb))
1589		sk->sk_send_head = NULL;
1590	else
1591		sk->sk_send_head = tcp_write_queue_next(sk, skb);
1592}
1593
1594static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1595{
1596	if (sk->sk_send_head == skb_unlinked) {
1597		sk->sk_send_head = NULL;
1598		tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
1599	}
1600	if (tcp_sk(sk)->highest_sack == skb_unlinked)
1601		tcp_sk(sk)->highest_sack = NULL;
1602}
1603
1604static inline void tcp_init_send_head(struct sock *sk)
1605{
1606	sk->sk_send_head = NULL;
1607}
1608
1609static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1610{
1611	__skb_queue_tail(&sk->sk_write_queue, skb);
1612}
1613
1614static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1615{
1616	__tcp_add_write_queue_tail(sk, skb);
1617
1618	/* Queue it, remembering where we must start sending. */
1619	if (sk->sk_send_head == NULL) {
1620		sk->sk_send_head = skb;
1621		tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1622
1623		if (tcp_sk(sk)->highest_sack == NULL)
1624			tcp_sk(sk)->highest_sack = skb;
1625	}
1626}
1627
1628static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1629{
1630	__skb_queue_head(&sk->sk_write_queue, skb);
1631}
1632
1633/* Insert buff after skb on the write queue of sk.  */
1634static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1635						struct sk_buff *buff,
1636						struct sock *sk)
1637{
1638	__skb_queue_after(&sk->sk_write_queue, skb, buff);
1639}
1640
1641/* Insert new before skb on the write queue of sk.  */
1642static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1643						  struct sk_buff *skb,
1644						  struct sock *sk)
1645{
1646	__skb_queue_before(&sk->sk_write_queue, skb, new);
1647
1648	if (sk->sk_send_head == skb)
1649		sk->sk_send_head = new;
1650}
1651
1652static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1653{
 
1654	__skb_unlink(skb, &sk->sk_write_queue);
1655}
1656
1657static inline bool tcp_write_queue_empty(struct sock *sk)
 
 
 
 
 
 
 
 
1658{
1659	return skb_queue_empty(&sk->sk_write_queue);
 
 
1660}
1661
1662static inline void tcp_push_pending_frames(struct sock *sk)
1663{
1664	if (tcp_send_head(sk)) {
1665		struct tcp_sock *tp = tcp_sk(sk);
1666
1667		__tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1668	}
1669}
1670
1671/* Start sequence of the skb just after the highest skb with SACKed
1672 * bit, valid only if sacked_out > 0 or when the caller has ensured
1673 * validity by itself.
1674 */
1675static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1676{
1677	if (!tp->sacked_out)
1678		return tp->snd_una;
1679
1680	if (tp->highest_sack == NULL)
1681		return tp->snd_nxt;
1682
1683	return TCP_SKB_CB(tp->highest_sack)->seq;
1684}
1685
1686static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1687{
1688	tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1689						tcp_write_queue_next(sk, skb);
1690}
1691
1692static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1693{
1694	return tcp_sk(sk)->highest_sack;
1695}
1696
1697static inline void tcp_highest_sack_reset(struct sock *sk)
1698{
1699	tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1700}
1701
1702/* Called when old skb is about to be deleted (to be combined with new skb) */
1703static inline void tcp_highest_sack_combine(struct sock *sk,
1704					    struct sk_buff *old,
1705					    struct sk_buff *new)
1706{
1707	if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1708		tcp_sk(sk)->highest_sack = new;
1709}
1710
1711/* This helper checks if socket has IP_TRANSPARENT set */
1712static inline bool inet_sk_transparent(const struct sock *sk)
1713{
1714	switch (sk->sk_state) {
1715	case TCP_TIME_WAIT:
1716		return inet_twsk(sk)->tw_transparent;
1717	case TCP_NEW_SYN_RECV:
1718		return inet_rsk(inet_reqsk(sk))->no_srccheck;
1719	}
1720	return inet_sk(sk)->transparent;
1721}
1722
1723/* Determines whether this is a thin stream (which may suffer from
1724 * increased latency). Used to trigger latency-reducing mechanisms.
1725 */
1726static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1727{
1728	return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1729}
1730
1731/* /proc */
1732enum tcp_seq_states {
1733	TCP_SEQ_STATE_LISTENING,
1734	TCP_SEQ_STATE_ESTABLISHED,
1735};
1736
1737int tcp_seq_open(struct inode *inode, struct file *file);
 
 
1738
1739struct tcp_seq_afinfo {
1740	char				*name;
1741	sa_family_t			family;
1742	const struct file_operations	*seq_fops;
1743	struct seq_operations		seq_ops;
1744};
1745
1746struct tcp_iter_state {
1747	struct seq_net_private	p;
1748	sa_family_t		family;
1749	enum tcp_seq_states	state;
1750	struct sock		*syn_wait_sk;
1751	int			bucket, offset, sbucket, num;
1752	loff_t			last_pos;
1753};
1754
1755int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1756void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1757
1758extern struct request_sock_ops tcp_request_sock_ops;
1759extern struct request_sock_ops tcp6_request_sock_ops;
1760
1761void tcp_v4_destroy_sock(struct sock *sk);
1762
1763struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1764				netdev_features_t features);
1765struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1766int tcp_gro_complete(struct sk_buff *skb);
 
 
 
 
 
 
 
 
1767
1768void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1769
1770static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1771{
1772	struct net *net = sock_net((struct sock *)tp);
1773	return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1774}
1775
1776static inline bool tcp_stream_memory_free(const struct sock *sk)
1777{
1778	const struct tcp_sock *tp = tcp_sk(sk);
1779	u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1780
1781	return notsent_bytes < tcp_notsent_lowat(tp);
1782}
1783
 
 
1784#ifdef CONFIG_PROC_FS
1785int tcp4_proc_init(void);
1786void tcp4_proc_exit(void);
1787#endif
1788
1789int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1790int tcp_conn_request(struct request_sock_ops *rsk_ops,
1791		     const struct tcp_request_sock_ops *af_ops,
1792		     struct sock *sk, struct sk_buff *skb);
1793
1794/* TCP af-specific functions */
1795struct tcp_sock_af_ops {
1796#ifdef CONFIG_TCP_MD5SIG
1797	struct tcp_md5sig_key	*(*md5_lookup) (const struct sock *sk,
1798						const struct sock *addr_sk);
1799	int		(*calc_md5_hash)(char *location,
1800					 const struct tcp_md5sig_key *md5,
1801					 const struct sock *sk,
1802					 const struct sk_buff *skb);
1803	int		(*md5_parse)(struct sock *sk,
1804				     char __user *optval,
 
1805				     int optlen);
1806#endif
 
 
 
 
 
 
 
 
 
 
 
 
1807};
1808
1809struct tcp_request_sock_ops {
1810	u16 mss_clamp;
1811#ifdef CONFIG_TCP_MD5SIG
1812	struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1813						 const struct sock *addr_sk);
1814	int		(*calc_md5_hash) (char *location,
1815					  const struct tcp_md5sig_key *md5,
1816					  const struct sock *sk,
1817					  const struct sk_buff *skb);
1818#endif
1819	void (*init_req)(struct request_sock *req,
1820			 const struct sock *sk_listener,
1821			 struct sk_buff *skb);
 
 
 
 
 
 
1822#ifdef CONFIG_SYN_COOKIES
1823	__u32 (*cookie_init_seq)(const struct sk_buff *skb,
1824				 __u16 *mss);
1825#endif
1826	struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1827				       const struct request_sock *req,
1828				       bool *strict);
1829	__u32 (*init_seq)(const struct sk_buff *skb, u32 *tsoff);
 
 
1830	int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1831			   struct flowi *fl, struct request_sock *req,
1832			   struct tcp_fastopen_cookie *foc,
1833			   enum tcp_synack_type synack_type);
 
1834};
1835
 
 
 
 
 
1836#ifdef CONFIG_SYN_COOKIES
1837static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1838					 const struct sock *sk, struct sk_buff *skb,
1839					 __u16 *mss)
1840{
1841	tcp_synq_overflow(sk);
1842	__NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1843	return ops->cookie_init_seq(skb, mss);
1844}
1845#else
1846static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1847					 const struct sock *sk, struct sk_buff *skb,
1848					 __u16 *mss)
1849{
1850	return 0;
1851}
1852#endif
1853
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1854int tcpv4_offload_init(void);
1855
1856void tcp_v4_init(void);
1857void tcp_init(void);
1858
1859/* tcp_recovery.c */
 
 
 
 
 
 
 
 
 
 
 
1860
1861/* Flags to enable various loss recovery features. See below */
1862extern int sysctl_tcp_recovery;
 
 
 
 
 
1863
1864/* Use TCP RACK to detect (some) tail and retransmit losses */
1865#define TCP_RACK_LOST_RETRANS  0x1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1866
1867extern int tcp_rack_mark_lost(struct sock *sk);
 
 
 
 
 
1868
1869extern void tcp_rack_advance(struct tcp_sock *tp,
1870			     const struct skb_mstamp *xmit_time, u8 sacked);
1871
1872/*
1873 * Save and compile IPv4 options, return a pointer to it
1874 */
1875static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
 
1876{
1877	const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1878	struct ip_options_rcu *dopt = NULL;
1879
1880	if (opt->optlen) {
1881		int opt_size = sizeof(*dopt) + opt->optlen;
1882
1883		dopt = kmalloc(opt_size, GFP_ATOMIC);
1884		if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1885			kfree(dopt);
1886			dopt = NULL;
1887		}
1888	}
1889	return dopt;
1890}
1891
1892/* locally generated TCP pure ACKs have skb->truesize == 2
1893 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1894 * This is much faster than dissecting the packet to find out.
1895 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1896 */
1897static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1898{
1899	return skb->truesize == 2;
1900}
1901
1902static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
1903{
1904	skb->truesize = 2;
1905}
1906
1907static inline int tcp_inq(struct sock *sk)
1908{
1909	struct tcp_sock *tp = tcp_sk(sk);
1910	int answ;
1911
1912	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1913		answ = 0;
1914	} else if (sock_flag(sk, SOCK_URGINLINE) ||
1915		   !tp->urg_data ||
1916		   before(tp->urg_seq, tp->copied_seq) ||
1917		   !before(tp->urg_seq, tp->rcv_nxt)) {
1918
1919		answ = tp->rcv_nxt - tp->copied_seq;
1920
1921		/* Subtract 1, if FIN was received */
1922		if (answ && sock_flag(sk, SOCK_DONE))
1923			answ--;
1924	} else {
1925		answ = tp->urg_seq - tp->copied_seq;
1926	}
1927
1928	return answ;
1929}
1930
1931int tcp_peek_len(struct socket *sock);
1932
1933static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
1934{
1935	u16 segs_in;
1936
1937	segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1938	tp->segs_in += segs_in;
 
 
 
 
1939	if (skb->len > tcp_hdrlen(skb))
1940		tp->data_segs_in += segs_in;
1941}
1942
1943/*
1944 * TCP listen path runs lockless.
1945 * We forced "struct sock" to be const qualified to make sure
1946 * we don't modify one of its field by mistake.
1947 * Here, we increment sk_drops which is an atomic_t, so we can safely
1948 * make sock writable again.
1949 */
1950static inline void tcp_listendrop(const struct sock *sk)
1951{
1952	atomic_inc(&((struct sock *)sk)->sk_drops);
1953	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1954}
1955
1956#endif	/* _TCP_H */