xref: /linux-6.15/include/linux/tcp.h (revision c81ea720)
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, <[email protected]>
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 
24 static inline struct tcphdr *tcp_hdr(const struct sk_buff *skb)
25 {
26 	return (struct tcphdr *)skb_transport_header(skb);
27 }
28 
29 static inline unsigned int __tcp_hdrlen(const struct tcphdr *th)
30 {
31 	return th->doff * 4;
32 }
33 
34 static inline unsigned int tcp_hdrlen(const struct sk_buff *skb)
35 {
36 	return __tcp_hdrlen(tcp_hdr(skb));
37 }
38 
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 
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 static inline unsigned int tcp_optlen(const struct sk_buff *skb)
50 {
51 	return (tcp_hdr(skb)->doff - 5) * 4;
52 }
53 
54 /* TCP Fast Open */
55 #define TCP_FASTOPEN_COOKIE_MIN	4	/* Min Fast Open Cookie size in bytes */
56 #define TCP_FASTOPEN_COOKIE_MAX	16	/* Max Fast Open Cookie size in bytes */
57 #define TCP_FASTOPEN_COOKIE_SIZE 8	/* the size employed by this impl. */
58 
59 /* TCP Fast Open Cookie as stored in memory */
60 struct tcp_fastopen_cookie {
61 	__le64	val[DIV_ROUND_UP(TCP_FASTOPEN_COOKIE_MAX, sizeof(u64))];
62 	s8	len;
63 	bool	exp;	/* In RFC6994 experimental option format */
64 };
65 
66 /* This defines a selective acknowledgement block. */
67 struct tcp_sack_block_wire {
68 	__be32	start_seq;
69 	__be32	end_seq;
70 };
71 
72 struct tcp_sack_block {
73 	u32	start_seq;
74 	u32	end_seq;
75 };
76 
77 /*These are used to set the sack_ok field in struct tcp_options_received */
78 #define TCP_SACK_SEEN     (1 << 0)   /*1 = peer is SACK capable, */
79 #define TCP_DSACK_SEEN    (1 << 2)   /*1 = DSACK was received from peer*/
80 
81 #if IS_ENABLED(CONFIG_MPTCP)
82 struct mptcp_options_received {
83 	u64	sndr_key;
84 	u64	rcvr_key;
85 	u64	data_ack;
86 	u64	data_seq;
87 	u32	subflow_seq;
88 	u16	data_len;
89 	u8	mp_capable : 1,
90 		mp_join : 1,
91 		dss : 1;
92 	u8	use_map:1,
93 		dsn64:1,
94 		data_fin:1,
95 		use_ack:1,
96 		ack64:1,
97 		mpc_map:1,
98 		__unused:2;
99 };
100 #endif
101 
102 struct tcp_options_received {
103 /*	PAWS/RTTM data	*/
104 	int	ts_recent_stamp;/* Time we stored ts_recent (for aging) */
105 	u32	ts_recent;	/* Time stamp to echo next		*/
106 	u32	rcv_tsval;	/* Time stamp value             	*/
107 	u32	rcv_tsecr;	/* Time stamp echo reply        	*/
108 	u16 	saw_tstamp : 1,	/* Saw TIMESTAMP on last packet		*/
109 		tstamp_ok : 1,	/* TIMESTAMP seen on SYN packet		*/
110 		dsack : 1,	/* D-SACK is scheduled			*/
111 		wscale_ok : 1,	/* Wscale seen on SYN packet		*/
112 		sack_ok : 3,	/* SACK seen on SYN packet		*/
113 		smc_ok : 1,	/* SMC seen on SYN packet		*/
114 		snd_wscale : 4,	/* Window scaling received from sender	*/
115 		rcv_wscale : 4;	/* Window scaling to send to receiver	*/
116 	u8	num_sacks;	/* Number of SACK blocks		*/
117 	u16	user_mss;	/* mss requested by user in ioctl	*/
118 	u16	mss_clamp;	/* Maximal mss, negotiated at connection setup */
119 #if IS_ENABLED(CONFIG_MPTCP)
120 	struct mptcp_options_received	mptcp;
121 #endif
122 };
123 
124 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
125 {
126 	rx_opt->tstamp_ok = rx_opt->sack_ok = 0;
127 	rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
128 #if IS_ENABLED(CONFIG_SMC)
129 	rx_opt->smc_ok = 0;
130 #endif
131 #if IS_ENABLED(CONFIG_MPTCP)
132 	rx_opt->mptcp.mp_capable = 0;
133 	rx_opt->mptcp.mp_join = 0;
134 	rx_opt->mptcp.dss = 0;
135 #endif
136 }
137 
138 /* This is the max number of SACKS that we'll generate and process. It's safe
139  * to increase this, although since:
140  *   size = TCPOLEN_SACK_BASE_ALIGNED (4) + n * TCPOLEN_SACK_PERBLOCK (8)
141  * only four options will fit in a standard TCP header */
142 #define TCP_NUM_SACKS 4
143 
144 struct tcp_request_sock_ops;
145 
146 struct tcp_request_sock {
147 	struct inet_request_sock 	req;
148 	const struct tcp_request_sock_ops *af_specific;
149 	u64				snt_synack; /* first SYNACK sent time */
150 	bool				tfo_listener;
151 #if IS_ENABLED(CONFIG_MPTCP)
152 	bool				is_mptcp;
153 #endif
154 	u32				txhash;
155 	u32				rcv_isn;
156 	u32				snt_isn;
157 	u32				ts_off;
158 	u32				last_oow_ack_time; /* last SYNACK */
159 	u32				rcv_nxt; /* the ack # by SYNACK. For
160 						  * FastOpen it's the seq#
161 						  * after data-in-SYN.
162 						  */
163 };
164 
165 static inline struct tcp_request_sock *tcp_rsk(const struct request_sock *req)
166 {
167 	return (struct tcp_request_sock *)req;
168 }
169 
170 struct tcp_sock {
171 	/* inet_connection_sock has to be the first member of tcp_sock */
172 	struct inet_connection_sock	inet_conn;
173 	u16	tcp_header_len;	/* Bytes of tcp header to send		*/
174 	u16	gso_segs;	/* Max number of segs per GSO packet	*/
175 
176 /*
177  *	Header prediction flags
178  *	0x5?10 << 16 + snd_wnd in net byte order
179  */
180 	__be32	pred_flags;
181 
182 /*
183  *	RFC793 variables by their proper names. This means you can
184  *	read the code and the spec side by side (and laugh ...)
185  *	See RFC793 and RFC1122. The RFC writes these in capitals.
186  */
187 	u64	bytes_received;	/* RFC4898 tcpEStatsAppHCThruOctetsReceived
188 				 * sum(delta(rcv_nxt)), or how many bytes
189 				 * were acked.
190 				 */
191 	u32	segs_in;	/* RFC4898 tcpEStatsPerfSegsIn
192 				 * total number of segments in.
193 				 */
194 	u32	data_segs_in;	/* RFC4898 tcpEStatsPerfDataSegsIn
195 				 * total number of data segments in.
196 				 */
197  	u32	rcv_nxt;	/* What we want to receive next 	*/
198 	u32	copied_seq;	/* Head of yet unread data		*/
199 	u32	rcv_wup;	/* rcv_nxt on last window update sent	*/
200  	u32	snd_nxt;	/* Next sequence we send		*/
201 	u32	segs_out;	/* RFC4898 tcpEStatsPerfSegsOut
202 				 * The total number of segments sent.
203 				 */
204 	u32	data_segs_out;	/* RFC4898 tcpEStatsPerfDataSegsOut
205 				 * total number of data segments sent.
206 				 */
207 	u64	bytes_sent;	/* RFC4898 tcpEStatsPerfHCDataOctetsOut
208 				 * total number of data bytes sent.
209 				 */
210 	u64	bytes_acked;	/* RFC4898 tcpEStatsAppHCThruOctetsAcked
211 				 * sum(delta(snd_una)), or how many bytes
212 				 * were acked.
213 				 */
214 	u32	dsack_dups;	/* RFC4898 tcpEStatsStackDSACKDups
215 				 * total number of DSACK blocks received
216 				 */
217  	u32	snd_una;	/* First byte we want an ack for	*/
218  	u32	snd_sml;	/* Last byte of the most recently transmitted small packet */
219 	u32	rcv_tstamp;	/* timestamp of last received ACK (for keepalives) */
220 	u32	lsndtime;	/* timestamp of last sent data packet (for restart window) */
221 	u32	last_oow_ack_time;  /* timestamp of last out-of-window ACK */
222 	u32	compressed_ack_rcv_nxt;
223 
224 	u32	tsoffset;	/* timestamp offset */
225 
226 	struct list_head tsq_node; /* anchor in tsq_tasklet.head list */
227 	struct list_head tsorted_sent_queue; /* time-sorted sent but un-SACKed skbs */
228 
229 	u32	snd_wl1;	/* Sequence for window update		*/
230 	u32	snd_wnd;	/* The window we expect to receive	*/
231 	u32	max_window;	/* Maximal window ever seen from peer	*/
232 	u32	mss_cache;	/* Cached effective mss, not including SACKS */
233 
234 	u32	window_clamp;	/* Maximal window to advertise		*/
235 	u32	rcv_ssthresh;	/* Current window clamp			*/
236 
237 	/* Information of the most recently (s)acked skb */
238 	struct tcp_rack {
239 		u64 mstamp; /* (Re)sent time of the skb */
240 		u32 rtt_us;  /* Associated RTT */
241 		u32 end_seq; /* Ending TCP sequence of the skb */
242 		u32 last_delivered; /* tp->delivered at last reo_wnd adj */
243 		u8 reo_wnd_steps;   /* Allowed reordering window */
244 #define TCP_RACK_RECOVERY_THRESH 16
245 		u8 reo_wnd_persist:5, /* No. of recovery since last adj */
246 		   dsack_seen:1, /* Whether DSACK seen after last adj */
247 		   advanced:1;	 /* mstamp advanced since last lost marking */
248 	} rack;
249 	u16	advmss;		/* Advertised MSS			*/
250 	u8	compressed_ack;
251 	u32	chrono_start;	/* Start time in jiffies of a TCP chrono */
252 	u32	chrono_stat[3];	/* Time in jiffies for chrono_stat stats */
253 	u8	chrono_type:2,	/* current chronograph type */
254 		rate_app_limited:1,  /* rate_{delivered,interval_us} limited? */
255 		fastopen_connect:1, /* FASTOPEN_CONNECT sockopt */
256 		fastopen_no_cookie:1, /* Allow send/recv SYN+data without a cookie */
257 		is_sack_reneg:1,    /* in recovery from loss with SACK reneg? */
258 		fastopen_client_fail:2; /* reason why fastopen failed */
259 	u8	nonagle     : 4,/* Disable Nagle algorithm?             */
260 		thin_lto    : 1,/* Use linear timeouts for thin streams */
261 		recvmsg_inq : 1,/* Indicate # of bytes in queue upon recvmsg */
262 		repair      : 1,
263 		frto        : 1;/* F-RTO (RFC5682) activated in CA_Loss */
264 	u8	repair_queue;
265 	u8	syn_data:1,	/* SYN includes data */
266 		syn_fastopen:1,	/* SYN includes Fast Open option */
267 		syn_fastopen_exp:1,/* SYN includes Fast Open exp. option */
268 		syn_fastopen_ch:1, /* Active TFO re-enabling probe */
269 		syn_data_acked:1,/* data in SYN is acked by SYN-ACK */
270 		save_syn:1,	/* Save headers of SYN packet */
271 		is_cwnd_limited:1,/* forward progress limited by snd_cwnd? */
272 		syn_smc:1;	/* SYN includes SMC */
273 	u32	tlp_high_seq;	/* snd_nxt at the time of TLP retransmit. */
274 
275 	u32	tcp_tx_delay;	/* delay (in usec) added to TX packets */
276 	u64	tcp_wstamp_ns;	/* departure time for next sent data packet */
277 	u64	tcp_clock_cache; /* cache last tcp_clock_ns() (see tcp_mstamp_refresh()) */
278 
279 /* RTT measurement */
280 	u64	tcp_mstamp;	/* most recent packet received/sent */
281 	u32	srtt_us;	/* smoothed round trip time << 3 in usecs */
282 	u32	mdev_us;	/* medium deviation			*/
283 	u32	mdev_max_us;	/* maximal mdev for the last rtt period	*/
284 	u32	rttvar_us;	/* smoothed mdev_max			*/
285 	u32	rtt_seq;	/* sequence number to update rttvar	*/
286 	struct  minmax rtt_min;
287 
288 	u32	packets_out;	/* Packets which are "in flight"	*/
289 	u32	retrans_out;	/* Retransmitted packets out		*/
290 	u32	max_packets_out;  /* max packets_out in last window */
291 	u32	max_packets_seq;  /* right edge of max_packets_out flight */
292 
293 	u16	urg_data;	/* Saved octet of OOB data and control flags */
294 	u8	ecn_flags;	/* ECN status bits.			*/
295 	u8	keepalive_probes; /* num of allowed keep alive probes	*/
296 	u32	reordering;	/* Packet reordering metric.		*/
297 	u32	reord_seen;	/* number of data packet reordering events */
298 	u32	snd_up;		/* Urgent pointer		*/
299 
300 /*
301  *      Options received (usually on last packet, some only on SYN packets).
302  */
303 	struct tcp_options_received rx_opt;
304 
305 /*
306  *	Slow start and congestion control (see also Nagle, and Karn & Partridge)
307  */
308  	u32	snd_ssthresh;	/* Slow start size threshold		*/
309  	u32	snd_cwnd;	/* Sending congestion window		*/
310 	u32	snd_cwnd_cnt;	/* Linear increase counter		*/
311 	u32	snd_cwnd_clamp; /* Do not allow snd_cwnd to grow above this */
312 	u32	snd_cwnd_used;
313 	u32	snd_cwnd_stamp;
314 	u32	prior_cwnd;	/* cwnd right before starting loss recovery */
315 	u32	prr_delivered;	/* Number of newly delivered packets to
316 				 * receiver in Recovery. */
317 	u32	prr_out;	/* Total number of pkts sent during Recovery. */
318 	u32	delivered;	/* Total data packets delivered incl. rexmits */
319 	u32	delivered_ce;	/* Like the above but only ECE marked packets */
320 	u32	lost;		/* Total data packets lost incl. rexmits */
321 	u32	app_limited;	/* limited until "delivered" reaches this val */
322 	u64	first_tx_mstamp;  /* start of window send phase */
323 	u64	delivered_mstamp; /* time we reached "delivered" */
324 	u32	rate_delivered;    /* saved rate sample: packets delivered */
325 	u32	rate_interval_us;  /* saved rate sample: time elapsed */
326 
327  	u32	rcv_wnd;	/* Current receiver window		*/
328 	u32	write_seq;	/* Tail(+1) of data held in tcp send buffer */
329 	u32	notsent_lowat;	/* TCP_NOTSENT_LOWAT */
330 	u32	pushed_seq;	/* Last pushed seq, required to talk to windows */
331 	u32	lost_out;	/* Lost packets			*/
332 	u32	sacked_out;	/* SACK'd packets			*/
333 
334 	struct hrtimer	pacing_timer;
335 	struct hrtimer	compressed_ack_timer;
336 
337 	/* from STCP, retrans queue hinting */
338 	struct sk_buff* lost_skb_hint;
339 	struct sk_buff *retransmit_skb_hint;
340 
341 	/* OOO segments go in this rbtree. Socket lock must be held. */
342 	struct rb_root	out_of_order_queue;
343 	struct sk_buff	*ooo_last_skb; /* cache rb_last(out_of_order_queue) */
344 
345 	/* SACKs data, these 2 need to be together (see tcp_options_write) */
346 	struct tcp_sack_block duplicate_sack[1]; /* D-SACK block */
347 	struct tcp_sack_block selective_acks[4]; /* The SACKS themselves*/
348 
349 	struct tcp_sack_block recv_sack_cache[4];
350 
351 	struct sk_buff *highest_sack;   /* skb just after the highest
352 					 * skb with SACKed bit set
353 					 * (validity guaranteed only if
354 					 * sacked_out > 0)
355 					 */
356 
357 	int     lost_cnt_hint;
358 
359 	u32	prior_ssthresh; /* ssthresh saved at recovery start	*/
360 	u32	high_seq;	/* snd_nxt at onset of congestion	*/
361 
362 	u32	retrans_stamp;	/* Timestamp of the last retransmit,
363 				 * also used in SYN-SENT to remember stamp of
364 				 * the first SYN. */
365 	u32	undo_marker;	/* snd_una upon a new recovery episode. */
366 	int	undo_retrans;	/* number of undoable retransmissions. */
367 	u64	bytes_retrans;	/* RFC4898 tcpEStatsPerfOctetsRetrans
368 				 * Total data bytes retransmitted
369 				 */
370 	u32	total_retrans;	/* Total retransmits for entire connection */
371 
372 	u32	urg_seq;	/* Seq of received urgent pointer */
373 	unsigned int		keepalive_time;	  /* time before keep alive takes place */
374 	unsigned int		keepalive_intvl;  /* time interval between keep alive probes */
375 
376 	int			linger2;
377 
378 
379 /* Sock_ops bpf program related variables */
380 #ifdef CONFIG_BPF
381 	u8	bpf_sock_ops_cb_flags;  /* Control calling BPF programs
382 					 * values defined in uapi/linux/tcp.h
383 					 */
384 #define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) (TP->bpf_sock_ops_cb_flags & ARG)
385 #else
386 #define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) 0
387 #endif
388 
389 	u16 timeout_rehash;	/* Timeout-triggered rehash attempts */
390 
391 	u32 rcv_ooopack; /* Received out-of-order packets, for tcpinfo */
392 
393 /* Receiver side RTT estimation */
394 	u32 rcv_rtt_last_tsecr;
395 	struct {
396 		u32	rtt_us;
397 		u32	seq;
398 		u64	time;
399 	} rcv_rtt_est;
400 
401 /* Receiver queue space */
402 	struct {
403 		u32	space;
404 		u32	seq;
405 		u64	time;
406 	} rcvq_space;
407 
408 /* TCP-specific MTU probe information. */
409 	struct {
410 		u32		  probe_seq_start;
411 		u32		  probe_seq_end;
412 	} mtu_probe;
413 	u32	mtu_info; /* We received an ICMP_FRAG_NEEDED / ICMPV6_PKT_TOOBIG
414 			   * while socket was owned by user.
415 			   */
416 #if IS_ENABLED(CONFIG_MPTCP)
417 	bool	is_mptcp;
418 #endif
419 
420 #ifdef CONFIG_TCP_MD5SIG
421 /* TCP AF-Specific parts; only used by MD5 Signature support so far */
422 	const struct tcp_sock_af_ops	*af_specific;
423 
424 /* TCP MD5 Signature Option information */
425 	struct tcp_md5sig_info	__rcu *md5sig_info;
426 #endif
427 
428 /* TCP fastopen related information */
429 	struct tcp_fastopen_request *fastopen_req;
430 	/* fastopen_rsk points to request_sock that resulted in this big
431 	 * socket. Used to retransmit SYNACKs etc.
432 	 */
433 	struct request_sock __rcu *fastopen_rsk;
434 	u32	*saved_syn;
435 };
436 
437 enum tsq_enum {
438 	TSQ_THROTTLED,
439 	TSQ_QUEUED,
440 	TCP_TSQ_DEFERRED,	   /* tcp_tasklet_func() found socket was owned */
441 	TCP_WRITE_TIMER_DEFERRED,  /* tcp_write_timer() found socket was owned */
442 	TCP_DELACK_TIMER_DEFERRED, /* tcp_delack_timer() found socket was owned */
443 	TCP_MTU_REDUCED_DEFERRED,  /* tcp_v{4|6}_err() could not call
444 				    * tcp_v{4|6}_mtu_reduced()
445 				    */
446 };
447 
448 enum tsq_flags {
449 	TSQF_THROTTLED			= (1UL << TSQ_THROTTLED),
450 	TSQF_QUEUED			= (1UL << TSQ_QUEUED),
451 	TCPF_TSQ_DEFERRED		= (1UL << TCP_TSQ_DEFERRED),
452 	TCPF_WRITE_TIMER_DEFERRED	= (1UL << TCP_WRITE_TIMER_DEFERRED),
453 	TCPF_DELACK_TIMER_DEFERRED	= (1UL << TCP_DELACK_TIMER_DEFERRED),
454 	TCPF_MTU_REDUCED_DEFERRED	= (1UL << TCP_MTU_REDUCED_DEFERRED),
455 };
456 
457 static inline struct tcp_sock *tcp_sk(const struct sock *sk)
458 {
459 	return (struct tcp_sock *)sk;
460 }
461 
462 struct tcp_timewait_sock {
463 	struct inet_timewait_sock tw_sk;
464 #define tw_rcv_nxt tw_sk.__tw_common.skc_tw_rcv_nxt
465 #define tw_snd_nxt tw_sk.__tw_common.skc_tw_snd_nxt
466 	u32			  tw_rcv_wnd;
467 	u32			  tw_ts_offset;
468 	u32			  tw_ts_recent;
469 
470 	/* The time we sent the last out-of-window ACK: */
471 	u32			  tw_last_oow_ack_time;
472 
473 	int			  tw_ts_recent_stamp;
474 	u32			  tw_tx_delay;
475 #ifdef CONFIG_TCP_MD5SIG
476 	struct tcp_md5sig_key	  *tw_md5_key;
477 #endif
478 };
479 
480 static inline struct tcp_timewait_sock *tcp_twsk(const struct sock *sk)
481 {
482 	return (struct tcp_timewait_sock *)sk;
483 }
484 
485 static inline bool tcp_passive_fastopen(const struct sock *sk)
486 {
487 	return sk->sk_state == TCP_SYN_RECV &&
488 	       rcu_access_pointer(tcp_sk(sk)->fastopen_rsk) != NULL;
489 }
490 
491 static inline void fastopen_queue_tune(struct sock *sk, int backlog)
492 {
493 	struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
494 	int somaxconn = READ_ONCE(sock_net(sk)->core.sysctl_somaxconn);
495 
496 	queue->fastopenq.max_qlen = min_t(unsigned int, backlog, somaxconn);
497 }
498 
499 static inline void tcp_move_syn(struct tcp_sock *tp,
500 				struct request_sock *req)
501 {
502 	tp->saved_syn = req->saved_syn;
503 	req->saved_syn = NULL;
504 }
505 
506 static inline void tcp_saved_syn_free(struct tcp_sock *tp)
507 {
508 	kfree(tp->saved_syn);
509 	tp->saved_syn = NULL;
510 }
511 
512 struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk);
513 
514 static inline u16 tcp_mss_clamp(const struct tcp_sock *tp, u16 mss)
515 {
516 	/* We use READ_ONCE() here because socket might not be locked.
517 	 * This happens for listeners.
518 	 */
519 	u16 user_mss = READ_ONCE(tp->rx_opt.user_mss);
520 
521 	return (user_mss && user_mss < mss) ? user_mss : mss;
522 }
523 
524 int tcp_skb_shift(struct sk_buff *to, struct sk_buff *from, int pcount,
525 		  int shiftlen);
526 
527 #endif	/* _LINUX_TCP_H */
528