xref: /linux/include/linux/tcp.h (revision d2c3a7eb)
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 
tcp_hdr(const struct sk_buff * skb)24 static inline struct tcphdr *tcp_hdr(const struct sk_buff *skb)
25 {
26 	return (struct tcphdr *)skb_transport_header(skb);
27 }
28 
__tcp_hdrlen(const struct tcphdr * th)29 static inline unsigned int __tcp_hdrlen(const struct tcphdr *th)
30 {
31 	return th->doff * 4;
32 }
33 
tcp_hdrlen(const struct sk_buff * skb)34 static inline unsigned int tcp_hdrlen(const struct sk_buff *skb)
35 {
36 	return __tcp_hdrlen(tcp_hdr(skb));
37 }
38 
inner_tcp_hdr(const struct sk_buff * skb)39 static inline struct tcphdr *inner_tcp_hdr(const struct sk_buff *skb)
40 {
41 	return (struct tcphdr *)skb_inner_transport_header(skb);
42 }
43 
inner_tcp_hdrlen(const struct sk_buff * skb)44 static 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  */
skb_tcp_all_headers(const struct sk_buff * skb)59 static 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  */
skb_inner_tcp_all_headers(const struct sk_buff * skb)74 static 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 
tcp_optlen(const struct sk_buff * skb)79 static 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 */
90 struct 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. */
97 struct tcp_sack_block_wire {
98 	__be32	start_seq;
99 	__be32	end_seq;
100 };
101 
102 struct 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 
111 struct 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	saw_unknown:1,	/* Received unknown option		*/
126 		unused:7;
127 	u8	num_sacks;	/* Number of SACK blocks		*/
128 	u16	user_mss;	/* mss requested by user in ioctl	*/
129 	u16	mss_clamp;	/* Maximal mss, negotiated at connection setup */
130 };
131 
tcp_clear_options(struct tcp_options_received * rx_opt)132 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
133 {
134 	rx_opt->tstamp_ok = rx_opt->sack_ok = 0;
135 	rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
136 #if IS_ENABLED(CONFIG_SMC)
137 	rx_opt->smc_ok = 0;
138 #endif
139 }
140 
141 /* This is the max number of SACKS that we'll generate and process. It's safe
142  * to increase this, although since:
143  *   size = TCPOLEN_SACK_BASE_ALIGNED (4) + n * TCPOLEN_SACK_PERBLOCK (8)
144  * only four options will fit in a standard TCP header */
145 #define TCP_NUM_SACKS 4
146 
147 struct tcp_request_sock_ops;
148 
149 struct tcp_request_sock {
150 	struct inet_request_sock 	req;
151 	const struct tcp_request_sock_ops *af_specific;
152 	u64				snt_synack; /* first SYNACK sent time */
153 	bool				tfo_listener;
154 	bool				is_mptcp;
155 	bool				req_usec_ts;
156 #if IS_ENABLED(CONFIG_MPTCP)
157 	bool				drop_req;
158 #endif
159 	u32				txhash;
160 	u32				rcv_isn;
161 	u32				snt_isn;
162 	u32				ts_off;
163 	u32				last_oow_ack_time; /* last SYNACK */
164 	u32				rcv_nxt; /* the ack # by SYNACK. For
165 						  * FastOpen it's the seq#
166 						  * after data-in-SYN.
167 						  */
168 	u8				syn_tos;
169 #ifdef CONFIG_TCP_AO
170 	u8				ao_keyid;
171 	u8				ao_rcv_next;
172 	bool				used_tcp_ao;
173 #endif
174 };
175 
tcp_rsk(const struct request_sock * req)176 static inline struct tcp_request_sock *tcp_rsk(const struct request_sock *req)
177 {
178 	return (struct tcp_request_sock *)req;
179 }
180 
tcp_rsk_used_ao(const struct request_sock * req)181 static inline bool tcp_rsk_used_ao(const struct request_sock *req)
182 {
183 #ifndef CONFIG_TCP_AO
184 	return false;
185 #else
186 	return tcp_rsk(req)->used_tcp_ao;
187 #endif
188 }
189 
190 #define TCP_RMEM_TO_WIN_SCALE 8
191 
192 struct tcp_sock {
193 	/* Cacheline organization can be found documented in
194 	 * Documentation/networking/net_cachelines/tcp_sock.rst.
195 	 * Please update the document when adding new fields.
196 	 */
197 
198 	/* inet_connection_sock has to be the first member of tcp_sock */
199 	struct inet_connection_sock	inet_conn;
200 
201 	/* TX read-mostly hotpath cache lines */
202 	__cacheline_group_begin(tcp_sock_read_tx);
203 	/* timestamp of last sent data packet (for restart window) */
204 	u32	max_window;	/* Maximal window ever seen from peer	*/
205 	u32	rcv_ssthresh;	/* Current window clamp			*/
206 	u32	reordering;	/* Packet reordering metric.		*/
207 	u32	notsent_lowat;	/* TCP_NOTSENT_LOWAT */
208 	u16	gso_segs;	/* Max number of segs per GSO packet	*/
209 	/* from STCP, retrans queue hinting */
210 	struct sk_buff *lost_skb_hint;
211 	struct sk_buff *retransmit_skb_hint;
212 	__cacheline_group_end(tcp_sock_read_tx);
213 
214 	/* TXRX read-mostly hotpath cache lines */
215 	__cacheline_group_begin(tcp_sock_read_txrx);
216 	u32	tsoffset;	/* timestamp offset */
217 	u32	snd_wnd;	/* The window we expect to receive	*/
218 	u32	mss_cache;	/* Cached effective mss, not including SACKS */
219 	u32	snd_cwnd;	/* Sending congestion window		*/
220 	u32	prr_out;	/* Total number of pkts sent during Recovery. */
221 	u32	lost_out;	/* Lost packets			*/
222 	u32	sacked_out;	/* SACK'd packets			*/
223 	u16	tcp_header_len;	/* Bytes of tcp header to send		*/
224 	u8	scaling_ratio;	/* see tcp_win_from_space() */
225 	u8	chrono_type : 2,	/* current chronograph type */
226 		repair      : 1,
227 		tcp_usec_ts : 1, /* TSval values in usec */
228 		is_sack_reneg:1,    /* in recovery from loss with SACK reneg? */
229 		is_cwnd_limited:1;/* forward progress limited by snd_cwnd? */
230 	__cacheline_group_end(tcp_sock_read_txrx);
231 
232 	/* RX read-mostly hotpath cache lines */
233 	__cacheline_group_begin(tcp_sock_read_rx);
234 	u32	copied_seq;	/* Head of yet unread data */
235 	u32	rcv_tstamp;	/* timestamp of last received ACK (for keepalives) */
236 	u32	snd_wl1;	/* Sequence for window update		*/
237 	u32	tlp_high_seq;	/* snd_nxt at the time of TLP */
238 	u32	rttvar_us;	/* smoothed mdev_max			*/
239 	u32	retrans_out;	/* Retransmitted packets out		*/
240 	u16	advmss;		/* Advertised MSS			*/
241 	u16	urg_data;	/* Saved octet of OOB data and control flags */
242 	u32	lost;		/* Total data packets lost incl. rexmits */
243 	struct  minmax rtt_min;
244 	/* OOO segments go in this rbtree. Socket lock must be held. */
245 	struct rb_root	out_of_order_queue;
246 	u32	snd_ssthresh;	/* Slow start size threshold		*/
247 	u8	recvmsg_inq : 1;/* Indicate # of bytes in queue upon recvmsg */
248 	__cacheline_group_end(tcp_sock_read_rx);
249 
250 	/* TX read-write hotpath cache lines */
251 	__cacheline_group_begin(tcp_sock_write_tx) ____cacheline_aligned;
252 	u32	segs_out;	/* RFC4898 tcpEStatsPerfSegsOut
253 				 * The total number of segments sent.
254 				 */
255 	u32	data_segs_out;	/* RFC4898 tcpEStatsPerfDataSegsOut
256 				 * total number of data segments sent.
257 				 */
258 	u64	bytes_sent;	/* RFC4898 tcpEStatsPerfHCDataOctetsOut
259 				 * total number of data bytes sent.
260 				 */
261 	u32	snd_sml;	/* Last byte of the most recently transmitted small packet */
262 	u32	chrono_start;	/* Start time in jiffies of a TCP chrono */
263 	u32	chrono_stat[3];	/* Time in jiffies for chrono_stat stats */
264 	u32	write_seq;	/* Tail(+1) of data held in tcp send buffer */
265 	u32	pushed_seq;	/* Last pushed seq, required to talk to windows */
266 	u32	lsndtime;
267 	u32	mdev_us;	/* medium deviation			*/
268 	u32	rtt_seq;	/* sequence number to update rttvar	*/
269 	u64	tcp_wstamp_ns;	/* departure time for next sent data packet */
270 	struct list_head tsorted_sent_queue; /* time-sorted sent but un-SACKed skbs */
271 	struct sk_buff *highest_sack;   /* skb just after the highest
272 					 * skb with SACKed bit set
273 					 * (validity guaranteed only if
274 					 * sacked_out > 0)
275 					 */
276 	u8	ecn_flags;	/* ECN status bits.			*/
277 	__cacheline_group_end(tcp_sock_write_tx);
278 
279 	/* TXRX read-write hotpath cache lines */
280 	__cacheline_group_begin(tcp_sock_write_txrx);
281 /*
282  *	Header prediction flags
283  *	0x5?10 << 16 + snd_wnd in net byte order
284  */
285 	__be32	pred_flags;
286 	u64	tcp_clock_cache; /* cache last tcp_clock_ns() (see tcp_mstamp_refresh()) */
287 	u64	tcp_mstamp;	/* most recent packet received/sent */
288 	u32	rcv_nxt;	/* What we want to receive next		*/
289 	u32	snd_nxt;	/* Next sequence we send		*/
290 	u32	snd_una;	/* First byte we want an ack for	*/
291 	u32	window_clamp;	/* Maximal window to advertise		*/
292 	u32	srtt_us;	/* smoothed round trip time << 3 in usecs */
293 	u32	packets_out;	/* Packets which are "in flight"	*/
294 	u32	snd_up;		/* Urgent pointer		*/
295 	u32	delivered;	/* Total data packets delivered incl. rexmits */
296 	u32	delivered_ce;	/* Like the above but only ECE marked packets */
297 	u32	app_limited;	/* limited until "delivered" reaches this val */
298 	u32	rcv_wnd;	/* Current receiver window		*/
299 /*
300  *      Options received (usually on last packet, some only on SYN packets).
301  */
302 	struct tcp_options_received rx_opt;
303 	u8	nonagle     : 4,/* Disable Nagle algorithm?             */
304 		rate_app_limited:1;  /* rate_{delivered,interval_us} limited? */
305 	__cacheline_group_end(tcp_sock_write_txrx);
306 
307 	/* RX read-write hotpath cache lines */
308 	__cacheline_group_begin(tcp_sock_write_rx) __aligned(8);
309 	u64	bytes_received;
310 				/* RFC4898 tcpEStatsAppHCThruOctetsReceived
311 				 * sum(delta(rcv_nxt)), or how many bytes
312 				 * were acked.
313 				 */
314 	u32	segs_in;	/* RFC4898 tcpEStatsPerfSegsIn
315 				 * total number of segments in.
316 				 */
317 	u32	data_segs_in;	/* RFC4898 tcpEStatsPerfDataSegsIn
318 				 * total number of data segments in.
319 				 */
320 	u32	rcv_wup;	/* rcv_nxt on last window update sent	*/
321 	u32	max_packets_out;  /* max packets_out in last window */
322 	u32	cwnd_usage_seq;  /* right edge of cwnd usage tracking flight */
323 	u32	rate_delivered;    /* saved rate sample: packets delivered */
324 	u32	rate_interval_us;  /* saved rate sample: time elapsed */
325 	u32	rcv_rtt_last_tsecr;
326 	u64	first_tx_mstamp;  /* start of window send phase */
327 	u64	delivered_mstamp; /* time we reached "delivered" */
328 	u64	bytes_acked;	/* RFC4898 tcpEStatsAppHCThruOctetsAcked
329 				 * sum(delta(snd_una)), or how many bytes
330 				 * were acked.
331 				 */
332 	struct {
333 		u32	rtt_us;
334 		u32	seq;
335 		u64	time;
336 	} rcv_rtt_est;
337 /* Receiver queue space */
338 	struct {
339 		u32	space;
340 		u32	seq;
341 		u64	time;
342 	} rcvq_space;
343 	__cacheline_group_end(tcp_sock_write_rx);
344 	/* End of Hot Path */
345 
346 /*
347  *	RFC793 variables by their proper names. This means you can
348  *	read the code and the spec side by side (and laugh ...)
349  *	See RFC793 and RFC1122. The RFC writes these in capitals.
350  */
351 	u32	dsack_dups;	/* RFC4898 tcpEStatsStackDSACKDups
352 				 * total number of DSACK blocks received
353 				 */
354 	u32	compressed_ack_rcv_nxt;
355 	struct list_head tsq_node; /* anchor in tsq_tasklet.head list */
356 
357 	/* Information of the most recently (s)acked skb */
358 	struct tcp_rack {
359 		u64 mstamp; /* (Re)sent time of the skb */
360 		u32 rtt_us;  /* Associated RTT */
361 		u32 end_seq; /* Ending TCP sequence of the skb */
362 		u32 last_delivered; /* tp->delivered at last reo_wnd adj */
363 		u8 reo_wnd_steps;   /* Allowed reordering window */
364 #define TCP_RACK_RECOVERY_THRESH 16
365 		u8 reo_wnd_persist:5, /* No. of recovery since last adj */
366 		   dsack_seen:1, /* Whether DSACK seen after last adj */
367 		   advanced:1;	 /* mstamp advanced since last lost marking */
368 	} rack;
369 	u8	compressed_ack;
370 	u8	dup_ack_counter:2,
371 		tlp_retrans:1,	/* TLP is a retransmission */
372 		unused:5;
373 	u8	thin_lto    : 1,/* Use linear timeouts for thin streams */
374 		fastopen_connect:1, /* FASTOPEN_CONNECT sockopt */
375 		fastopen_no_cookie:1, /* Allow send/recv SYN+data without a cookie */
376 		fastopen_client_fail:2, /* reason why fastopen failed */
377 		frto        : 1;/* F-RTO (RFC5682) activated in CA_Loss */
378 	u8	repair_queue;
379 	u8	save_syn:2,	/* Save headers of SYN packet */
380 		syn_data:1,	/* SYN includes data */
381 		syn_fastopen:1,	/* SYN includes Fast Open option */
382 		syn_fastopen_exp:1,/* SYN includes Fast Open exp. option */
383 		syn_fastopen_ch:1, /* Active TFO re-enabling probe */
384 		syn_data_acked:1;/* data in SYN is acked by SYN-ACK */
385 
386 	u8	keepalive_probes; /* num of allowed keep alive probes	*/
387 	u32	tcp_tx_delay;	/* delay (in usec) added to TX packets */
388 
389 /* RTT measurement */
390 	u32	mdev_max_us;	/* maximal mdev for the last rtt period	*/
391 
392 	u32	reord_seen;	/* number of data packet reordering events */
393 
394 /*
395  *	Slow start and congestion control (see also Nagle, and Karn & Partridge)
396  */
397 	u32	snd_cwnd_cnt;	/* Linear increase counter		*/
398 	u32	snd_cwnd_clamp; /* Do not allow snd_cwnd to grow above this */
399 	u32	snd_cwnd_used;
400 	u32	snd_cwnd_stamp;
401 	u32	prior_cwnd;	/* cwnd right before starting loss recovery */
402 	u32	prr_delivered;	/* Number of newly delivered packets to
403 				 * receiver in Recovery. */
404 	u32	last_oow_ack_time;  /* timestamp of last out-of-window ACK */
405 
406 	struct hrtimer	pacing_timer;
407 	struct hrtimer	compressed_ack_timer;
408 
409 	struct sk_buff	*ooo_last_skb; /* cache rb_last(out_of_order_queue) */
410 
411 	/* SACKs data, these 2 need to be together (see tcp_options_write) */
412 	struct tcp_sack_block duplicate_sack[1]; /* D-SACK block */
413 	struct tcp_sack_block selective_acks[4]; /* The SACKS themselves*/
414 
415 	struct tcp_sack_block recv_sack_cache[4];
416 
417 	int     lost_cnt_hint;
418 
419 	u32	prior_ssthresh; /* ssthresh saved at recovery start	*/
420 	u32	high_seq;	/* snd_nxt at onset of congestion	*/
421 
422 	u32	retrans_stamp;	/* Timestamp of the last retransmit,
423 				 * also used in SYN-SENT to remember stamp of
424 				 * the first SYN. */
425 	u32	undo_marker;	/* snd_una upon a new recovery episode. */
426 	int	undo_retrans;	/* number of undoable retransmissions. */
427 	u64	bytes_retrans;	/* RFC4898 tcpEStatsPerfOctetsRetrans
428 				 * Total data bytes retransmitted
429 				 */
430 	u32	total_retrans;	/* Total retransmits for entire connection */
431 	u32	rto_stamp;	/* Start time (ms) of last CA_Loss recovery */
432 	u16	total_rto;	/* Total number of RTO timeouts, including
433 				 * SYN/SYN-ACK and recurring timeouts.
434 				 */
435 	u16	total_rto_recoveries;	/* Total number of RTO recoveries,
436 					 * including any unfinished recovery.
437 					 */
438 	u32	total_rto_time;	/* ms spent in (completed) RTO recoveries. */
439 
440 	u32	urg_seq;	/* Seq of received urgent pointer */
441 	unsigned int		keepalive_time;	  /* time before keep alive takes place */
442 	unsigned int		keepalive_intvl;  /* time interval between keep alive probes */
443 
444 	int			linger2;
445 
446 
447 /* Sock_ops bpf program related variables */
448 #ifdef CONFIG_BPF
449 	u8	bpf_sock_ops_cb_flags;  /* Control calling BPF programs
450 					 * values defined in uapi/linux/tcp.h
451 					 */
452 	u8	bpf_chg_cc_inprogress:1; /* In the middle of
453 					  * bpf_setsockopt(TCP_CONGESTION),
454 					  * it is to avoid the bpf_tcp_cc->init()
455 					  * to recur itself by calling
456 					  * bpf_setsockopt(TCP_CONGESTION, "itself").
457 					  */
458 #define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) (TP->bpf_sock_ops_cb_flags & ARG)
459 #else
460 #define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) 0
461 #endif
462 
463 	u16 timeout_rehash;	/* Timeout-triggered rehash attempts */
464 
465 	u32 rcv_ooopack; /* Received out-of-order packets, for tcpinfo */
466 
467 /* TCP-specific MTU probe information. */
468 	struct {
469 		u32		  probe_seq_start;
470 		u32		  probe_seq_end;
471 	} mtu_probe;
472 	u32     plb_rehash;     /* PLB-triggered rehash attempts */
473 	u32	mtu_info; /* We received an ICMP_FRAG_NEEDED / ICMPV6_PKT_TOOBIG
474 			   * while socket was owned by user.
475 			   */
476 #if IS_ENABLED(CONFIG_MPTCP)
477 	bool	is_mptcp;
478 #endif
479 #if IS_ENABLED(CONFIG_SMC)
480 	bool	syn_smc;	/* SYN includes SMC */
481 	bool	(*smc_hs_congested)(const struct sock *sk);
482 #endif
483 
484 #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
485 /* TCP AF-Specific parts; only used by TCP-AO/MD5 Signature support so far */
486 	const struct tcp_sock_af_ops	*af_specific;
487 
488 #ifdef CONFIG_TCP_MD5SIG
489 /* TCP MD5 Signature Option information */
490 	struct tcp_md5sig_info	__rcu *md5sig_info;
491 #endif
492 #ifdef CONFIG_TCP_AO
493 	struct tcp_ao_info	__rcu *ao_info;
494 #endif
495 #endif
496 
497 /* TCP fastopen related information */
498 	struct tcp_fastopen_request *fastopen_req;
499 	/* fastopen_rsk points to request_sock that resulted in this big
500 	 * socket. Used to retransmit SYNACKs etc.
501 	 */
502 	struct request_sock __rcu *fastopen_rsk;
503 	struct saved_syn *saved_syn;
504 };
505 
506 enum tsq_enum {
507 	TSQ_THROTTLED,
508 	TSQ_QUEUED,
509 	TCP_TSQ_DEFERRED,	   /* tcp_tasklet_func() found socket was owned */
510 	TCP_WRITE_TIMER_DEFERRED,  /* tcp_write_timer() found socket was owned */
511 	TCP_DELACK_TIMER_DEFERRED, /* tcp_delack_timer() found socket was owned */
512 	TCP_MTU_REDUCED_DEFERRED,  /* tcp_v{4|6}_err() could not call
513 				    * tcp_v{4|6}_mtu_reduced()
514 				    */
515 	TCP_ACK_DEFERRED,	   /* TX pure ack is deferred */
516 };
517 
518 enum tsq_flags {
519 	TSQF_THROTTLED			= BIT(TSQ_THROTTLED),
520 	TSQF_QUEUED			= BIT(TSQ_QUEUED),
521 	TCPF_TSQ_DEFERRED		= BIT(TCP_TSQ_DEFERRED),
522 	TCPF_WRITE_TIMER_DEFERRED	= BIT(TCP_WRITE_TIMER_DEFERRED),
523 	TCPF_DELACK_TIMER_DEFERRED	= BIT(TCP_DELACK_TIMER_DEFERRED),
524 	TCPF_MTU_REDUCED_DEFERRED	= BIT(TCP_MTU_REDUCED_DEFERRED),
525 	TCPF_ACK_DEFERRED		= BIT(TCP_ACK_DEFERRED),
526 };
527 
528 #define tcp_sk(ptr) container_of_const(ptr, struct tcp_sock, inet_conn.icsk_inet.sk)
529 
530 /* Variant of tcp_sk() upgrading a const sock to a read/write tcp socket.
531  * Used in context of (lockless) tcp listeners.
532  */
533 #define tcp_sk_rw(ptr) container_of(ptr, struct tcp_sock, inet_conn.icsk_inet.sk)
534 
535 struct tcp_timewait_sock {
536 	struct inet_timewait_sock tw_sk;
537 #define tw_rcv_nxt tw_sk.__tw_common.skc_tw_rcv_nxt
538 #define tw_snd_nxt tw_sk.__tw_common.skc_tw_snd_nxt
539 	u32			  tw_rcv_wnd;
540 	u32			  tw_ts_offset;
541 	u32			  tw_ts_recent;
542 
543 	/* The time we sent the last out-of-window ACK: */
544 	u32			  tw_last_oow_ack_time;
545 
546 	int			  tw_ts_recent_stamp;
547 	u32			  tw_tx_delay;
548 #ifdef CONFIG_TCP_MD5SIG
549 	struct tcp_md5sig_key	  *tw_md5_key;
550 #endif
551 #ifdef CONFIG_TCP_AO
552 	struct tcp_ao_info	__rcu *ao_info;
553 #endif
554 };
555 
tcp_twsk(const struct sock * sk)556 static inline struct tcp_timewait_sock *tcp_twsk(const struct sock *sk)
557 {
558 	return (struct tcp_timewait_sock *)sk;
559 }
560 
tcp_passive_fastopen(const struct sock * sk)561 static inline bool tcp_passive_fastopen(const struct sock *sk)
562 {
563 	return sk->sk_state == TCP_SYN_RECV &&
564 	       rcu_access_pointer(tcp_sk(sk)->fastopen_rsk) != NULL;
565 }
566 
fastopen_queue_tune(struct sock * sk,int backlog)567 static inline void fastopen_queue_tune(struct sock *sk, int backlog)
568 {
569 	struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
570 	int somaxconn = READ_ONCE(sock_net(sk)->core.sysctl_somaxconn);
571 
572 	WRITE_ONCE(queue->fastopenq.max_qlen, min_t(unsigned int, backlog, somaxconn));
573 }
574 
tcp_move_syn(struct tcp_sock * tp,struct request_sock * req)575 static inline void tcp_move_syn(struct tcp_sock *tp,
576 				struct request_sock *req)
577 {
578 	tp->saved_syn = req->saved_syn;
579 	req->saved_syn = NULL;
580 }
581 
tcp_saved_syn_free(struct tcp_sock * tp)582 static inline void tcp_saved_syn_free(struct tcp_sock *tp)
583 {
584 	kfree(tp->saved_syn);
585 	tp->saved_syn = NULL;
586 }
587 
tcp_saved_syn_len(const struct saved_syn * saved_syn)588 static inline u32 tcp_saved_syn_len(const struct saved_syn *saved_syn)
589 {
590 	return saved_syn->mac_hdrlen + saved_syn->network_hdrlen +
591 		saved_syn->tcp_hdrlen;
592 }
593 
594 struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk,
595 					       const struct sk_buff *orig_skb,
596 					       const struct sk_buff *ack_skb);
597 
tcp_mss_clamp(const struct tcp_sock * tp,u16 mss)598 static inline u16 tcp_mss_clamp(const struct tcp_sock *tp, u16 mss)
599 {
600 	/* We use READ_ONCE() here because socket might not be locked.
601 	 * This happens for listeners.
602 	 */
603 	u16 user_mss = READ_ONCE(tp->rx_opt.user_mss);
604 
605 	return (user_mss && user_mss < mss) ? user_mss : mss;
606 }
607 
608 int tcp_skb_shift(struct sk_buff *to, struct sk_buff *from, int pcount,
609 		  int shiftlen);
610 
611 void __tcp_sock_set_cork(struct sock *sk, bool on);
612 void tcp_sock_set_cork(struct sock *sk, bool on);
613 int tcp_sock_set_keepcnt(struct sock *sk, int val);
614 int tcp_sock_set_keepidle_locked(struct sock *sk, int val);
615 int tcp_sock_set_keepidle(struct sock *sk, int val);
616 int tcp_sock_set_keepintvl(struct sock *sk, int val);
617 void __tcp_sock_set_nodelay(struct sock *sk, bool on);
618 void tcp_sock_set_nodelay(struct sock *sk);
619 void tcp_sock_set_quickack(struct sock *sk, int val);
620 int tcp_sock_set_syncnt(struct sock *sk, int val);
621 int tcp_sock_set_user_timeout(struct sock *sk, int val);
622 
dst_tcp_usec_ts(const struct dst_entry * dst)623 static inline bool dst_tcp_usec_ts(const struct dst_entry *dst)
624 {
625 	return dst_feature(dst, RTAX_FEATURE_TCP_USEC_TS);
626 }
627 
628 #endif	/* _LINUX_TCP_H */
629