tcp.h

  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 protocol.
  8 *
  9 * Version:	@(#)tcp.h	1.0.2	04/28/93
 10 *
 11 * Author:	Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
 12 */
 13#ifndef _LINUX_TCP_H
 14#define _LINUX_TCP_H
 15
 16
 17#include <linux/skbuff.h>
 18#include <linux/win_minmax.h>
 19#include <net/sock.h>
 20#include <net/inet_connection_sock.h>
 21#include <net/inet_timewait_sock.h>
 22#include <uapi/linux/tcp.h>
 23
 24static inline struct tcphdr *tcp_hdr(const struct sk_buff *skb)
 25{
 26	return (struct tcphdr *)skb_transport_header(skb);
 27}
 28
 29static inline unsigned int __tcp_hdrlen(const struct tcphdr *th)
 30{
 31	return th->doff * 4;
 32}
 33
 34static inline unsigned int tcp_hdrlen(const struct sk_buff *skb)
 35{
 36	return __tcp_hdrlen(tcp_hdr(skb));
 37}
 38
 39static inline struct tcphdr *inner_tcp_hdr(const struct sk_buff *skb)
 40{
 41	return (struct tcphdr *)skb_inner_transport_header(skb);
 42}
 43
 44static inline unsigned int inner_tcp_hdrlen(const struct sk_buff *skb)
 45{
 46	return inner_tcp_hdr(skb)->doff * 4;
 47}
 48
 49/**
 50 * skb_tcp_all_headers - Returns size of all headers for a TCP packet
 51 * @skb: buffer
 52 *
 53 * Used in TX path, for a packet known to be a TCP one.
 54 *
 55 * if (skb_is_gso(skb)) {
 56 *         int hlen = skb_tcp_all_headers(skb);
 57 *         ...
 58 */
 59static inline int skb_tcp_all_headers(const struct sk_buff *skb)
 60{
 61	return skb_transport_offset(skb) + tcp_hdrlen(skb);
 62}
 63
 64/**
 65 * skb_inner_tcp_all_headers - Returns size of all headers for an encap TCP packet
 66 * @skb: buffer
 67 *
 68 * Used in TX path, for a packet known to be a TCP one.
 69 *
 70 * if (skb_is_gso(skb) && skb->encapsulation) {
 71 *         int hlen = skb_inner_tcp_all_headers(skb);
 72 *         ...
 73 */
 74static inline int skb_inner_tcp_all_headers(const struct sk_buff *skb)
 75{
 76	return skb_inner_transport_offset(skb) + inner_tcp_hdrlen(skb);
 77}
 78
 79static inline unsigned int tcp_optlen(const struct sk_buff *skb)
 80{
 81	return (tcp_hdr(skb)->doff - 5) * 4;
 82}
 83
 84/* TCP Fast Open */
 85#define TCP_FASTOPEN_COOKIE_MIN	4	/* Min Fast Open Cookie size in bytes */
 86#define TCP_FASTOPEN_COOKIE_MAX	16	/* Max Fast Open Cookie size in bytes */
 87#define TCP_FASTOPEN_COOKIE_SIZE 8	/* the size employed by this impl. */
 88
 89/* TCP Fast Open Cookie as stored in memory */
 90struct tcp_fastopen_cookie {
 91	__le64	val[DIV_ROUND_UP(TCP_FASTOPEN_COOKIE_MAX, sizeof(u64))];
 92	s8	len;
 93	bool	exp;	/* In RFC6994 experimental option format */
 94};
 95
 96/* This defines a selective acknowledgement block. */
 97struct tcp_sack_block_wire {
 98	__be32	start_seq;
 99	__be32	end_seq;
100};
101
102struct tcp_sack_block {
103	u32	start_seq;
104	u32	end_seq;
105};
106
107/*These are used to set the sack_ok field in struct tcp_options_received */
108#define TCP_SACK_SEEN     (1 << 0)   /*1 = peer is SACK capable, */
109#define TCP_DSACK_SEEN    (1 << 2)   /*1 = DSACK was received from peer*/
110
111struct tcp_options_received {
112/*	PAWS/RTTM data	*/
113	int	ts_recent_stamp;/* Time we stored ts_recent (for aging) */
114	u32	ts_recent;	/* Time stamp to echo next		*/
115	u32	rcv_tsval;	/* Time stamp value             	*/
116	u32	rcv_tsecr;	/* Time stamp echo reply        	*/
117	u16 	saw_tstamp : 1,	/* Saw TIMESTAMP on last packet		*/
118		tstamp_ok : 1,	/* TIMESTAMP seen on SYN packet		*/
119		dsack : 1,	/* D-SACK is scheduled			*/
120		wscale_ok : 1,	/* Wscale seen on SYN packet		*/
121		sack_ok : 3,	/* SACK seen on SYN packet		*/
122		smc_ok : 1,	/* SMC seen on SYN packet		*/
123		snd_wscale : 4,	/* Window scaling received from sender	*/
124		rcv_wscale : 4;	/* Window scaling to send to receiver	*/
125	u8	accecn:6,	/* AccECN index in header, 0=no options	*/
126		saw_unknown:1,	/* Received unknown option		*/
127		unused:1;
128	u8	num_sacks;	/* Number of SACK blocks		*/
129	u16	user_mss;	/* mss requested by user in ioctl	*/
130	u16	mss_clamp;	/* Maximal mss, negotiated at connection setup */
131};
132
133static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
134{
135	rx_opt->tstamp_ok = rx_opt->sack_ok = 0;
136	rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
137#if IS_ENABLED(CONFIG_SMC)
138	rx_opt->smc_ok = 0;
139#endif
140}
141
142/* This is the max number of SACKS that we'll generate and process. It's safe
143 * to increase this, although since:
144 *   size = TCPOLEN_SACK_BASE_ALIGNED (4) + n * TCPOLEN_SACK_PERBLOCK (8)
145 * only four options will fit in a standard TCP header */
146#define TCP_NUM_SACKS 4
147
148struct tcp_request_sock_ops;
149
150struct tcp_request_sock {
151	struct inet_request_sock 	req;
152	const struct tcp_request_sock_ops *af_specific;
153	u64				snt_synack; /* first SYNACK sent time */
154	bool				tfo_listener;
155	bool				is_mptcp;
156	bool				req_usec_ts;
157#if IS_ENABLED(CONFIG_MPTCP)
158	bool				drop_req;
159#endif
160	u32				txhash;
161	u32				rcv_isn;
162	u32				snt_isn;
163	u32				ts_off;
164	u32				snt_tsval_first;
165	u32				snt_tsval_last;
166	u32				last_oow_ack_time; /* last SYNACK */
167	u32				rcv_nxt; /* the ack # by SYNACK. For
168						  * FastOpen it's the seq#
169						  * after data-in-SYN.
170						  */
171	u8				syn_tos;
172	bool				accecn_ok;
173	u8				syn_ect_snt: 2,
174					syn_ect_rcv: 2,
175					accecn_fail_mode:4;
176	u8				saw_accecn_opt  :2;
177#ifdef CONFIG_TCP_AO
178	u8				ao_keyid;
179	u8				ao_rcv_next;
180	bool				used_tcp_ao;
181#endif
182};
183
184static inline struct tcp_request_sock *tcp_rsk(const struct request_sock *req)
185{
186	return (struct tcp_request_sock *)req;
187}
188
189static inline bool tcp_rsk_used_ao(const struct request_sock *req)
190{
191#ifndef CONFIG_TCP_AO
192	return false;
193#else
194	return tcp_rsk(req)->used_tcp_ao;
195#endif
196}
197
198#define TCP_RMEM_TO_WIN_SCALE 8
199
200struct tcp_sock {
201	/* Cacheline organization can be found documented in
202	 * Documentation/networking/net_cachelines/tcp_sock.rst.
203	 * Please update the document when adding new fields.
204	 */
205
206	/* inet_connection_sock has to be the first member of tcp_sock */
207	struct inet_connection_sock	inet_conn;
208
209	/* TX read-mostly hotpath cache lines */
210	__cacheline_group_begin(tcp_sock_read_tx);
211	u32	max_window;	/* Maximal window ever seen from peer	*/
212	u32	rcv_ssthresh;	/* Current window clamp			*/
213	u32	reordering;	/* Packet reordering metric.		*/
214	u32	notsent_lowat;	/* TCP_NOTSENT_LOWAT */
215	u16	gso_segs;	/* Max number of segs per GSO packet	*/
216	/* from STCP, retrans queue hinting */
217	struct sk_buff *retransmit_skb_hint;
218#if defined(CONFIG_TLS_DEVICE)
219	void (*tcp_clean_acked)(struct sock *sk, u32 acked_seq);
220#endif
221	__cacheline_group_end(tcp_sock_read_tx);
222
223	/* TXRX read-mostly hotpath cache lines */
224	__cacheline_group_begin(tcp_sock_read_txrx);
225	u32	tsoffset;	/* timestamp offset */
226	u32	snd_wnd;	/* The window we expect to receive	*/
227	u32	mss_cache;	/* Cached effective mss, not including SACKS */
228	u32	snd_cwnd;	/* Sending congestion window		*/
229	u32	prr_out;	/* Total number of pkts sent during Recovery. */
230	u32	lost_out;	/* Lost packets			*/
231	u32	sacked_out;	/* SACK'd packets			*/
232	u16	tcp_header_len;	/* Bytes of tcp header to send		*/
233	u8	scaling_ratio;	/* see tcp_win_from_space() */
234	u8	chrono_type : 2,	/* current chronograph type */
235		repair      : 1,
236		tcp_usec_ts : 1, /* TSval values in usec */
237		is_sack_reneg:1,    /* in recovery from loss with SACK reneg? */
238		is_cwnd_limited:1,/* forward progress limited by snd_cwnd? */
239		recvmsg_inq : 1;/* Indicate # of bytes in queue upon recvmsg */
240	__cacheline_group_end(tcp_sock_read_txrx);
241
242	/* RX read-mostly hotpath cache lines */
243	__cacheline_group_begin(tcp_sock_read_rx);
244	u32	copied_seq;	/* Head of yet unread data */
245	u32	snd_wl1;	/* Sequence for window update		*/
246	u32	tlp_high_seq;	/* snd_nxt at the time of TLP */
247	u32	rttvar_us;	/* smoothed mdev_max			*/
248	u32	retrans_out;	/* Retransmitted packets out		*/
249	u16	advmss;		/* Advertised MSS			*/
250	u16	urg_data;	/* Saved octet of OOB data and control flags */
251	u32	lost;		/* Total data packets lost incl. rexmits */
252	u32	snd_ssthresh;	/* Slow start size threshold		*/
253	struct  minmax rtt_min;
254	/* OOO segments go in this rbtree. Socket lock must be held. */
255	struct rb_root	out_of_order_queue;
256	__cacheline_group_end(tcp_sock_read_rx);
257
258	/* TX read-write hotpath cache lines */
259	__cacheline_group_begin(tcp_sock_write_tx) ____cacheline_aligned;
260	u32	segs_out;	/* RFC4898 tcpEStatsPerfSegsOut
261				 * The total number of segments sent.
262				 */
263	u32	data_segs_out;	/* RFC4898 tcpEStatsPerfDataSegsOut
264				 * total number of data segments sent.
265				 */
266	u64	bytes_sent;	/* RFC4898 tcpEStatsPerfHCDataOctetsOut
267				 * total number of data bytes sent.
268				 */
269	u32	snd_sml;	/* Last byte of the most recently transmitted small packet */
270	u32	chrono_start;	/* Start time in jiffies of a TCP chrono */
271	u32	chrono_stat[3];	/* Time in jiffies for chrono_stat stats */
272	u32	write_seq;	/* Tail(+1) of data held in tcp send buffer */
273	u32	pushed_seq;	/* Last pushed seq, required to talk to windows */
274	u32	lsndtime;	/* timestamp of last sent data packet (for restart window) */
275	u32	mdev_us;	/* medium deviation			*/
276	u32	rtt_seq;	/* sequence number to update rttvar	*/
277	u64	tcp_wstamp_ns;	/* departure time for next sent data packet */
278	u64	accecn_opt_tstamp;	/* Last AccECN option sent timestamp */
279	struct list_head tsorted_sent_queue; /* time-sorted sent but un-SACKed skbs */
280	struct sk_buff *highest_sack;   /* skb just after the highest
281					 * skb with SACKed bit set
282					 * (validity guaranteed only if
283					 * sacked_out > 0)
284					 */
285	u8	ecn_flags;	/* ECN status bits.			*/
286	__cacheline_group_end(tcp_sock_write_tx);
287
288	/* TXRX read-write hotpath cache lines */
289	__cacheline_group_begin(tcp_sock_write_txrx);
290/*
291 *	Header prediction flags
292 *	0x5?10 << 16 + snd_wnd in net byte order
293 */
294	u8	nonagle     : 4,/* Disable Nagle algorithm?             */
295		rate_app_limited:1;  /* rate_{delivered,interval_us} limited? */
296	u8	received_ce_pending:4, /* Not yet transmit cnt of received_ce */
297		unused2:4;
298	u8	accecn_minlen:2,/* Minimum length of AccECN option sent */
299		est_ecnfield:2,/* ECN field for AccECN delivered estimates */
300		accecn_opt_demand:2,/* Demand AccECN option for n next ACKs */
301		prev_ecnfield:2; /* ECN bits from the previous segment */
302	__be32	pred_flags;
303	u64	tcp_clock_cache; /* cache last tcp_clock_ns() (see tcp_mstamp_refresh()) */
304	u64	tcp_mstamp;	/* most recent packet received/sent */
305	u32	rcv_nxt;	/* What we want to receive next		*/
306	u32	snd_nxt;	/* Next sequence we send		*/
307	u32	snd_una;	/* First byte we want an ack for	*/
308	u32	window_clamp;	/* Maximal window to advertise		*/
309	u32	srtt_us;	/* smoothed round trip time << 3 in usecs */
310	u32	packets_out;	/* Packets which are "in flight"	*/
311	u32	snd_up;		/* Urgent pointer		*/
312	u32	delivered;	/* Total data packets delivered incl. rexmits */
313	u32	delivered_ce;	/* Like the above but only ECE marked packets */
314	u32	received_ce;	/* Like the above but for rcvd CE marked pkts */
315	u32	received_ecn_bytes[3]; /* received byte counters for three ECN
316					* types: INET_ECN_ECT_1, INET_ECN_ECT_0,
317					* and INET_ECN_CE
318					*/
319	u32	app_limited;	/* limited until "delivered" reaches this val */
320	u32	rcv_wnd;	/* Current receiver window		*/
321	u32	rcv_tstamp;	/* timestamp of last received ACK (for keepalives) */
322/*
323 *      Options received (usually on last packet, some only on SYN packets).
324 */
325	struct tcp_options_received rx_opt;
326	__cacheline_group_end(tcp_sock_write_txrx);
327
328	/* RX read-write hotpath cache lines */
329	__cacheline_group_begin(tcp_sock_write_rx) __aligned(8);
330	u64	bytes_received;
331				/* RFC4898 tcpEStatsAppHCThruOctetsReceived
332				 * sum(delta(rcv_nxt)), or how many bytes
333				 * were acked.
334				 */
335	u32	segs_in;	/* RFC4898 tcpEStatsPerfSegsIn
336				 * total number of segments in.
337				 */
338	u32	data_segs_in;	/* RFC4898 tcpEStatsPerfDataSegsIn
339				 * total number of data segments in.
340				 */
341	u32	rcv_wup;	/* rcv_nxt on last window update sent	*/
342	u32	max_packets_out;  /* max packets_out in last window */
343	u32	cwnd_usage_seq;  /* right edge of cwnd usage tracking flight */
344	u32	rate_delivered;    /* saved rate sample: packets delivered */
345	u32	rate_interval_us;  /* saved rate sample: time elapsed */
346	u32	rcv_rtt_last_tsecr;
347	u32	delivered_ecn_bytes[3];
348	u64	first_tx_mstamp;  /* start of window send phase */
349	u64	delivered_mstamp; /* time we reached "delivered" */
350	u64	bytes_acked;	/* RFC4898 tcpEStatsAppHCThruOctetsAcked
351				 * sum(delta(snd_una)), or how many bytes
352				 * were acked.
353				 */
354	struct {
355		u32	rtt_us;
356		u32	seq;
357		u64	time;
358	} rcv_rtt_est;
359/* Receiver queue space */
360	struct {
361		int	space;
362		u32	seq;
363		u64	time;
364	} rcvq_space;
365	__cacheline_group_end(tcp_sock_write_rx);
366	/* End of Hot Path */
367
368/*
369 *	RFC793 variables by their proper names. This means you can
370 *	read the code and the spec side by side (and laugh ...)
371 *	See RFC793 and RFC1122. The RFC writes these in capitals.
372 */
373	u32	dsack_dups;	/* RFC4898 tcpEStatsStackDSACKDups
374				 * total number of DSACK blocks received
375				 */
376	u32	compressed_ack_rcv_nxt;
377	struct list_head tsq_node; /* anchor in tsq_tasklet.head list */
378
379	/* Information of the most recently (s)acked skb */
380	struct tcp_rack {
381		u64 mstamp; /* (Re)sent time of the skb */
382		u32 rtt_us;  /* Associated RTT */
383		u32 end_seq; /* Ending TCP sequence of the skb */
384		u32 last_delivered; /* tp->delivered at last reo_wnd adj */
385		u8 reo_wnd_steps;   /* Allowed reordering window */
386#define TCP_RACK_RECOVERY_THRESH 16
387		u8 reo_wnd_persist:5, /* No. of recovery since last adj */
388		   dsack_seen:1, /* Whether DSACK seen after last adj */
389		   advanced:1;	 /* mstamp advanced since last lost marking */
390	} rack;
391	u8	compressed_ack;
392	u8	dup_ack_counter:2,
393		tlp_retrans:1,	/* TLP is a retransmission */
394		syn_ect_snt:2,	/* AccECN ECT memory, only */
395		syn_ect_rcv:2;	/* ... needed during 3WHS + first seqno */
396	u8	thin_lto    : 1,/* Use linear timeouts for thin streams */
397		fastopen_connect:1, /* FASTOPEN_CONNECT sockopt */
398		fastopen_no_cookie:1, /* Allow send/recv SYN+data without a cookie */
399		fastopen_client_fail:2, /* reason why fastopen failed */
400		frto        : 1;/* F-RTO (RFC5682) activated in CA_Loss */
401	u8	repair_queue;
402	u8	save_syn:2,	/* Save headers of SYN packet */
403		syn_data:1,	/* SYN includes data */
404		syn_fastopen:1,	/* SYN includes Fast Open option */
405		syn_fastopen_exp:1,/* SYN includes Fast Open exp. option */
406		syn_fastopen_ch:1, /* Active TFO re-enabling probe */
407		syn_data_acked:1,/* data in SYN is acked by SYN-ACK */
408		syn_fastopen_child:1; /* created TFO passive child socket */
409
410	u8	keepalive_probes; /* num of allowed keep alive probes	*/
411	u8	accecn_fail_mode:4,	/* AccECN failure handling */
412		saw_accecn_opt:2;	/* An AccECN option was seen */
413	u32	tcp_tx_delay;	/* delay (in usec) added to TX packets */
414
415/* RTT measurement */
416	u32	mdev_max_us;	/* maximal mdev for the last rtt period	*/
417
418	u32	reord_seen;	/* number of data packet reordering events */
419
420/*
421 *	Slow start and congestion control (see also Nagle, and Karn & Partridge)
422 */
423	u32	snd_cwnd_cnt;	/* Linear increase counter		*/
424	u32	snd_cwnd_clamp; /* Do not allow snd_cwnd to grow above this */
425	u32	snd_cwnd_used;
426	u32	snd_cwnd_stamp;
427	u32	prior_cwnd;	/* cwnd right before starting loss recovery */
428	u32	prr_delivered;	/* Number of newly delivered packets to
429				 * receiver in Recovery. */
430	u32	last_oow_ack_time;  /* timestamp of last out-of-window ACK */
431
432	struct hrtimer	pacing_timer;
433	struct hrtimer	compressed_ack_timer;
434
435	struct sk_buff	*ooo_last_skb; /* cache rb_last(out_of_order_queue) */
436
437	/* SACKs data, these 2 need to be together (see tcp_options_write) */
438	struct tcp_sack_block duplicate_sack[1]; /* D-SACK block */
439	struct tcp_sack_block selective_acks[4]; /* The SACKS themselves*/
440
441	struct tcp_sack_block recv_sack_cache[4];
442
443	u32	prior_ssthresh; /* ssthresh saved at recovery start	*/
444	u32	high_seq;	/* snd_nxt at onset of congestion	*/
445
446	u32	retrans_stamp;	/* Timestamp of the last retransmit,
447				 * also used in SYN-SENT to remember stamp of
448				 * the first SYN. */
449	u32	undo_marker;	/* snd_una upon a new recovery episode. */
450	int	undo_retrans;	/* number of undoable retransmissions. */
451	u32	mtu_info; /* We received an ICMP_FRAG_NEEDED / ICMPV6_PKT_TOOBIG
452			   * while socket was owned by user.
453			   */
454	u64	bytes_retrans;	/* RFC4898 tcpEStatsPerfOctetsRetrans
455				 * Total data bytes retransmitted
456				 */
457	u32	total_retrans;	/* Total retransmits for entire connection */
458	u32	rto_stamp;	/* Start time (ms) of last CA_Loss recovery */
459	u16	total_rto;	/* Total number of RTO timeouts, including
460				 * SYN/SYN-ACK and recurring timeouts.
461				 */
462	u16	total_rto_recoveries;	/* Total number of RTO recoveries,
463					 * including any unfinished recovery.
464					 */
465	u32	total_rto_time;	/* ms spent in (completed) RTO recoveries. */
466
467	u32	urg_seq;	/* Seq of received urgent pointer */
468	unsigned int		keepalive_time;	  /* time before keep alive takes place */
469	unsigned int		keepalive_intvl;  /* time interval between keep alive probes */
470
471	int			linger2;
472
473
474/* Sock_ops bpf program related variables */
475#ifdef CONFIG_BPF
476	u8	bpf_sock_ops_cb_flags;  /* Control calling BPF programs
477					 * values defined in uapi/linux/tcp.h
478					 */
479	u8	bpf_chg_cc_inprogress:1; /* In the middle of
480					  * bpf_setsockopt(TCP_CONGESTION),
481					  * it is to avoid the bpf_tcp_cc->init()
482					  * to recur itself by calling
483					  * bpf_setsockopt(TCP_CONGESTION, "itself").
484					  */
485#define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) (TP->bpf_sock_ops_cb_flags & ARG)
486#else
487#define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) 0
488#endif
489
490	u16 timeout_rehash;	/* Timeout-triggered rehash attempts */
491
492	u32 rcv_ooopack; /* Received out-of-order packets, for tcpinfo */
493
494/* TCP-specific MTU probe information. */
495	struct {
496		u32		  probe_seq_start;
497		u32		  probe_seq_end;
498	} mtu_probe;
499	u32     plb_rehash;     /* PLB-triggered rehash attempts */
500#if IS_ENABLED(CONFIG_MPTCP)
501	bool	is_mptcp;
502#endif
503#if IS_ENABLED(CONFIG_SMC)
504	bool	syn_smc;	/* SYN includes SMC */
505	bool	(*smc_hs_congested)(const struct sock *sk);
506#endif
507
508#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
509/* TCP AF-Specific parts; only used by TCP-AO/MD5 Signature support so far */
510	const struct tcp_sock_af_ops	*af_specific;
511
512#ifdef CONFIG_TCP_MD5SIG
513/* TCP MD5 Signature Option information */
514	struct tcp_md5sig_info	__rcu *md5sig_info;
515#endif
516#ifdef CONFIG_TCP_AO
517	struct tcp_ao_info	__rcu *ao_info;
518#endif
519#endif
520
521/* TCP fastopen related information */
522	struct tcp_fastopen_request *fastopen_req;
523	/* fastopen_rsk points to request_sock that resulted in this big
524	 * socket. Used to retransmit SYNACKs etc.
525	 */
526	struct request_sock __rcu *fastopen_rsk;
527	struct saved_syn *saved_syn;
528};
529
530enum tsq_enum {
531	TSQ_THROTTLED,
532	TSQ_QUEUED,
533	TCP_TSQ_DEFERRED,	   /* tcp_tasklet_func() found socket was owned */
534	TCP_WRITE_TIMER_DEFERRED,  /* tcp_write_timer() found socket was owned */
535	TCP_DELACK_TIMER_DEFERRED, /* tcp_delack_timer() found socket was owned */
536	TCP_MTU_REDUCED_DEFERRED,  /* tcp_v{4|6}_err() could not call
537				    * tcp_v{4|6}_mtu_reduced()
538				    */
539	TCP_ACK_DEFERRED,	   /* TX pure ack is deferred */
540};
541
542enum tsq_flags {
543	TSQF_THROTTLED			= BIT(TSQ_THROTTLED),
544	TSQF_QUEUED			= BIT(TSQ_QUEUED),
545	TCPF_TSQ_DEFERRED		= BIT(TCP_TSQ_DEFERRED),
546	TCPF_WRITE_TIMER_DEFERRED	= BIT(TCP_WRITE_TIMER_DEFERRED),
547	TCPF_DELACK_TIMER_DEFERRED	= BIT(TCP_DELACK_TIMER_DEFERRED),
548	TCPF_MTU_REDUCED_DEFERRED	= BIT(TCP_MTU_REDUCED_DEFERRED),
549	TCPF_ACK_DEFERRED		= BIT(TCP_ACK_DEFERRED),
550};
551
552#define tcp_sk(ptr) container_of_const(ptr, struct tcp_sock, inet_conn.icsk_inet.sk)
553
554/* Variant of tcp_sk() upgrading a const sock to a read/write tcp socket.
555 * Used in context of (lockless) tcp listeners.
556 */
557#define tcp_sk_rw(ptr) container_of(ptr, struct tcp_sock, inet_conn.icsk_inet.sk)
558
559struct tcp_timewait_sock {
560	struct inet_timewait_sock tw_sk;
561#define tw_rcv_nxt tw_sk.__tw_common.skc_tw_rcv_nxt
562#define tw_snd_nxt tw_sk.__tw_common.skc_tw_snd_nxt
563	u32			  tw_rcv_wnd;
564	u32			  tw_ts_offset;
565	u32			  tw_ts_recent;
566
567	/* The time we sent the last out-of-window ACK: */
568	u32			  tw_last_oow_ack_time;
569
570	int			  tw_ts_recent_stamp;
571	u32			  tw_tx_delay;
572#ifdef CONFIG_TCP_MD5SIG
573	struct tcp_md5sig_key	  *tw_md5_key;
574#endif
575#ifdef CONFIG_TCP_AO
576	struct tcp_ao_info	__rcu *ao_info;
577#endif
578};
579
580static inline struct tcp_timewait_sock *tcp_twsk(const struct sock *sk)
581{
582	return (struct tcp_timewait_sock *)sk;
583}
584
585static inline bool tcp_passive_fastopen(const struct sock *sk)
586{
587	return sk->sk_state == TCP_SYN_RECV &&
588	       rcu_access_pointer(tcp_sk(sk)->fastopen_rsk) != NULL;
589}
590
591static inline void fastopen_queue_tune(struct sock *sk, int backlog)
592{
593	struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
594	int somaxconn = READ_ONCE(sock_net(sk)->core.sysctl_somaxconn);
595
596	WRITE_ONCE(queue->fastopenq.max_qlen, min_t(unsigned int, backlog, somaxconn));
597}
598
599static inline void tcp_move_syn(struct tcp_sock *tp,
600				struct request_sock *req)
601{
602	tp->saved_syn = req->saved_syn;
603	req->saved_syn = NULL;
604}
605
606static inline void tcp_saved_syn_free(struct tcp_sock *tp)
607{
608	kfree(tp->saved_syn);
609	tp->saved_syn = NULL;
610}
611
612static inline u32 tcp_saved_syn_len(const struct saved_syn *saved_syn)
613{
614	return saved_syn->mac_hdrlen + saved_syn->network_hdrlen +
615		saved_syn->tcp_hdrlen;
616}
617
618struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk,
619					       const struct sk_buff *orig_skb,
620					       const struct sk_buff *ack_skb);
621
622static inline u16 tcp_mss_clamp(const struct tcp_sock *tp, u16 mss)
623{
624	/* We use READ_ONCE() here because socket might not be locked.
625	 * This happens for listeners.
626	 */
627	u16 user_mss = READ_ONCE(tp->rx_opt.user_mss);
628
629	return (user_mss && user_mss < mss) ? user_mss : mss;
630}
631
632int tcp_skb_shift(struct sk_buff *to, struct sk_buff *from, int pcount,
633		  int shiftlen);
634
635void __tcp_sock_set_cork(struct sock *sk, bool on);
636void tcp_sock_set_cork(struct sock *sk, bool on);
637int tcp_sock_set_keepcnt(struct sock *sk, int val);
638int tcp_sock_set_keepidle_locked(struct sock *sk, int val);
639int tcp_sock_set_keepidle(struct sock *sk, int val);
640int tcp_sock_set_keepintvl(struct sock *sk, int val);
641void __tcp_sock_set_nodelay(struct sock *sk, bool on);
642void tcp_sock_set_nodelay(struct sock *sk);
643void tcp_sock_set_quickack(struct sock *sk, int val);
644int tcp_sock_set_syncnt(struct sock *sk, int val);
645int tcp_sock_set_user_timeout(struct sock *sk, int val);
646int tcp_sock_set_maxseg(struct sock *sk, int val);
647
648static inline bool dst_tcp_usec_ts(const struct dst_entry *dst)
649{
650	return dst_feature(dst, RTAX_FEATURE_TCP_USEC_TS);
651}
652
653#endif	/* _LINUX_TCP_H */