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 protocol.
7  *
8  * Version:	@(#)tcp.h	1.0.2	04/28/93
9  *
10  * Author:	Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11  *
12  *		This program is free software; you can redistribute it and/or
13  *		modify it under the terms of the GNU General Public License
14  *		as published by the Free Software Foundation; either version
15  *		2 of the License, or (at your option) any later version.
16  */
17 #ifndef _LINUX_TCP_H
18 #define _LINUX_TCP_H
19 
20 
21 #include <linux/skbuff.h>
22 #include <net/sock.h>
23 #include <net/inet_connection_sock.h>
24 #include <net/inet_timewait_sock.h>
25 #include <uapi/linux/tcp.h>
26 
tcp_hdr(const struct sk_buff * skb)27 static inline struct tcphdr *tcp_hdr(const struct sk_buff *skb)
28 {
29 	return (struct tcphdr *)skb_transport_header(skb);
30 }
31 
tcp_hdrlen(const struct sk_buff * skb)32 static inline unsigned int tcp_hdrlen(const struct sk_buff *skb)
33 {
34 	return tcp_hdr(skb)->doff * 4;
35 }
36 
inner_tcp_hdr(const struct sk_buff * skb)37 static inline struct tcphdr *inner_tcp_hdr(const struct sk_buff *skb)
38 {
39 	return (struct tcphdr *)skb_inner_transport_header(skb);
40 }
41 
inner_tcp_hdrlen(const struct sk_buff * skb)42 static inline unsigned int inner_tcp_hdrlen(const struct sk_buff *skb)
43 {
44 	return inner_tcp_hdr(skb)->doff * 4;
45 }
46 
tcp_optlen(const struct sk_buff * skb)47 static inline unsigned int tcp_optlen(const struct sk_buff *skb)
48 {
49 	return (tcp_hdr(skb)->doff - 5) * 4;
50 }
51 
52 /* TCP Fast Open */
53 #define TCP_FASTOPEN_COOKIE_MIN	4	/* Min Fast Open Cookie size in bytes */
54 #define TCP_FASTOPEN_COOKIE_MAX	16	/* Max Fast Open Cookie size in bytes */
55 #define TCP_FASTOPEN_COOKIE_SIZE 8	/* the size employed by this impl. */
56 
57 /* TCP Fast Open Cookie as stored in memory */
58 struct tcp_fastopen_cookie {
59 	s8	len;
60 	u8	val[TCP_FASTOPEN_COOKIE_MAX];
61 	bool	exp;	/* In RFC6994 experimental option format */
62 };
63 
64 /* This defines a selective acknowledgement block. */
65 struct tcp_sack_block_wire {
66 	__be32	start_seq;
67 	__be32	end_seq;
68 };
69 
70 struct tcp_sack_block {
71 	u32	start_seq;
72 	u32	end_seq;
73 };
74 
75 /*These are used to set the sack_ok field in struct tcp_options_received */
76 #define TCP_SACK_SEEN     (1 << 0)   /*1 = peer is SACK capable, */
77 #define TCP_FACK_ENABLED  (1 << 1)   /*1 = FACK is enabled locally*/
78 #define TCP_DSACK_SEEN    (1 << 2)   /*1 = DSACK was received from peer*/
79 
80 struct tcp_options_received {
81 /*	PAWS/RTTM data	*/
82 	long	ts_recent_stamp;/* Time we stored ts_recent (for aging) */
83 	u32	ts_recent;	/* Time stamp to echo next		*/
84 	u32	rcv_tsval;	/* Time stamp value             	*/
85 	u32	rcv_tsecr;	/* Time stamp echo reply        	*/
86 	u16 	saw_tstamp : 1,	/* Saw TIMESTAMP on last packet		*/
87 		tstamp_ok : 1,	/* TIMESTAMP seen on SYN packet		*/
88 		dsack : 1,	/* D-SACK is scheduled			*/
89 		wscale_ok : 1,	/* Wscale seen on SYN packet		*/
90 		sack_ok : 4,	/* SACK seen on SYN packet		*/
91 		snd_wscale : 4,	/* Window scaling received from sender	*/
92 		rcv_wscale : 4;	/* Window scaling to send to receiver	*/
93 	u8	num_sacks;	/* Number of SACK blocks		*/
94 	u16	user_mss;	/* mss requested by user in ioctl	*/
95 	u16	mss_clamp;	/* Maximal mss, negotiated at connection setup */
96 };
97 
tcp_clear_options(struct tcp_options_received * rx_opt)98 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
99 {
100 	rx_opt->tstamp_ok = rx_opt->sack_ok = 0;
101 	rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
102 }
103 
104 /* This is the max number of SACKS that we'll generate and process. It's safe
105  * to increase this, although since:
106  *   size = TCPOLEN_SACK_BASE_ALIGNED (4) + n * TCPOLEN_SACK_PERBLOCK (8)
107  * only four options will fit in a standard TCP header */
108 #define TCP_NUM_SACKS 4
109 
110 struct tcp_request_sock_ops;
111 
112 struct tcp_request_sock {
113 	struct inet_request_sock 	req;
114 	const struct tcp_request_sock_ops *af_specific;
115 	struct skb_mstamp		snt_synack; /* first SYNACK sent time */
116 	bool				tfo_listener;
117 	u32				txhash;
118 	u32				rcv_isn;
119 	u32				snt_isn;
120 	u32				last_oow_ack_time; /* last SYNACK */
121 	u32				rcv_nxt; /* the ack # by SYNACK. For
122 						  * FastOpen it's the seq#
123 						  * after data-in-SYN.
124 						  */
125 };
126 
tcp_rsk(const struct request_sock * req)127 static inline struct tcp_request_sock *tcp_rsk(const struct request_sock *req)
128 {
129 	return (struct tcp_request_sock *)req;
130 }
131 
132 struct tcp_sock {
133 	/* inet_connection_sock has to be the first member of tcp_sock */
134 	struct inet_connection_sock	inet_conn;
135 	u16	tcp_header_len;	/* Bytes of tcp header to send		*/
136 	u16	gso_segs;	/* Max number of segs per GSO packet	*/
137 
138 /*
139  *	Header prediction flags
140  *	0x5?10 << 16 + snd_wnd in net byte order
141  */
142 	__be32	pred_flags;
143 
144 /*
145  *	RFC793 variables by their proper names. This means you can
146  *	read the code and the spec side by side (and laugh ...)
147  *	See RFC793 and RFC1122. The RFC writes these in capitals.
148  */
149 	u64	bytes_received;	/* RFC4898 tcpEStatsAppHCThruOctetsReceived
150 				 * sum(delta(rcv_nxt)), or how many bytes
151 				 * were acked.
152 				 */
153 	u32	segs_in;	/* RFC4898 tcpEStatsPerfSegsIn
154 				 * total number of segments in.
155 				 */
156  	u32	rcv_nxt;	/* What we want to receive next 	*/
157 	u32	copied_seq;	/* Head of yet unread data		*/
158 	u32	rcv_wup;	/* rcv_nxt on last window update sent	*/
159  	u32	snd_nxt;	/* Next sequence we send		*/
160 	u32	segs_out;	/* RFC4898 tcpEStatsPerfSegsOut
161 				 * The total number of segments sent.
162 				 */
163 	u64	bytes_acked;	/* RFC4898 tcpEStatsAppHCThruOctetsAcked
164 				 * sum(delta(snd_una)), or how many bytes
165 				 * were acked.
166 				 */
167 	struct u64_stats_sync syncp; /* protects 64bit vars (cf tcp_get_info()) */
168 
169  	u32	snd_una;	/* First byte we want an ack for	*/
170  	u32	snd_sml;	/* Last byte of the most recently transmitted small packet */
171 	u32	rcv_tstamp;	/* timestamp of last received ACK (for keepalives) */
172 	u32	lsndtime;	/* timestamp of last sent data packet (for restart window) */
173 	u32	last_oow_ack_time;  /* timestamp of last out-of-window ACK */
174 
175 	u32	tsoffset;	/* timestamp offset */
176 
177 	struct list_head tsq_node; /* anchor in tsq_tasklet.head list */
178 	unsigned long	tsq_flags;
179 
180 	/* Data for direct copy to user */
181 	struct {
182 		struct sk_buff_head	prequeue;
183 		struct task_struct	*task;
184 		struct msghdr		*msg;
185 		int			memory;
186 		int			len;
187 	} ucopy;
188 
189 	u32	snd_wl1;	/* Sequence for window update		*/
190 	u32	snd_wnd;	/* The window we expect to receive	*/
191 	u32	max_window;	/* Maximal window ever seen from peer	*/
192 	u32	mss_cache;	/* Cached effective mss, not including SACKS */
193 
194 	u32	window_clamp;	/* Maximal window to advertise		*/
195 	u32	rcv_ssthresh;	/* Current window clamp			*/
196 
197 	/* Information of the most recently (s)acked skb */
198 	struct tcp_rack {
199 		struct skb_mstamp mstamp; /* (Re)sent time of the skb */
200 		u8 advanced; /* mstamp advanced since last lost marking */
201 		u8 reord;    /* reordering detected */
202 	} rack;
203 	u16	advmss;		/* Advertised MSS			*/
204 	u8	unused;
205 	u8	nonagle     : 4,/* Disable Nagle algorithm?             */
206 		thin_lto    : 1,/* Use linear timeouts for thin streams */
207 		thin_dupack : 1,/* Fast retransmit on first dupack      */
208 		repair      : 1,
209 		frto        : 1;/* F-RTO (RFC5682) activated in CA_Loss */
210 	u8	repair_queue;
211 	u8	do_early_retrans:1,/* Enable RFC5827 early-retransmit  */
212 		syn_data:1,	/* SYN includes data */
213 		syn_fastopen:1,	/* SYN includes Fast Open option */
214 		syn_fastopen_exp:1,/* SYN includes Fast Open exp. option */
215 		syn_data_acked:1,/* data in SYN is acked by SYN-ACK */
216 		save_syn:1,	/* Save headers of SYN packet */
217 		is_cwnd_limited:1;/* forward progress limited by snd_cwnd? */
218 	u32	tlp_high_seq;	/* snd_nxt at the time of TLP retransmit. */
219 
220 /* RTT measurement */
221 	u32	srtt_us;	/* smoothed round trip time << 3 in usecs */
222 	u32	mdev_us;	/* medium deviation			*/
223 	u32	mdev_max_us;	/* maximal mdev for the last rtt period	*/
224 	u32	rttvar_us;	/* smoothed mdev_max			*/
225 	u32	rtt_seq;	/* sequence number to update rttvar	*/
226 	struct rtt_meas {
227 		u32 rtt, ts;	/* RTT in usec and sampling time in jiffies. */
228 	} rtt_min[3];
229 
230 	u32	packets_out;	/* Packets which are "in flight"	*/
231 	u32	retrans_out;	/* Retransmitted packets out		*/
232 	u32	max_packets_out;  /* max packets_out in last window */
233 	u32	max_packets_seq;  /* right edge of max_packets_out flight */
234 
235 	u16	urg_data;	/* Saved octet of OOB data and control flags */
236 	u8	ecn_flags;	/* ECN status bits.			*/
237 	u8	keepalive_probes; /* num of allowed keep alive probes	*/
238 	u32	reordering;	/* Packet reordering metric.		*/
239 	u32	snd_up;		/* Urgent pointer		*/
240 
241 /*
242  *      Options received (usually on last packet, some only on SYN packets).
243  */
244 	struct tcp_options_received rx_opt;
245 
246 /*
247  *	Slow start and congestion control (see also Nagle, and Karn & Partridge)
248  */
249  	u32	snd_ssthresh;	/* Slow start size threshold		*/
250  	u32	snd_cwnd;	/* Sending congestion window		*/
251 	u32	snd_cwnd_cnt;	/* Linear increase counter		*/
252 	u32	snd_cwnd_clamp; /* Do not allow snd_cwnd to grow above this */
253 	u32	snd_cwnd_used;
254 	u32	snd_cwnd_stamp;
255 	u32	prior_cwnd;	/* Congestion window at start of Recovery. */
256 	u32	prr_delivered;	/* Number of newly delivered packets to
257 				 * receiver in Recovery. */
258 	u32	prr_out;	/* Total number of pkts sent during Recovery. */
259 
260  	u32	rcv_wnd;	/* Current receiver window		*/
261 	u32	write_seq;	/* Tail(+1) of data held in tcp send buffer */
262 	u32	notsent_lowat;	/* TCP_NOTSENT_LOWAT */
263 	u32	pushed_seq;	/* Last pushed seq, required to talk to windows */
264 	u32	lost_out;	/* Lost packets			*/
265 	u32	sacked_out;	/* SACK'd packets			*/
266 	u32	fackets_out;	/* FACK'd packets			*/
267 
268 	/* from STCP, retrans queue hinting */
269 	struct sk_buff* lost_skb_hint;
270 	struct sk_buff *retransmit_skb_hint;
271 
272 	/* OOO segments go in this list. Note that socket lock must be held,
273 	 * as we do not use sk_buff_head lock.
274 	 */
275 	struct sk_buff_head	out_of_order_queue;
276 
277 	/* SACKs data, these 2 need to be together (see tcp_options_write) */
278 	struct tcp_sack_block duplicate_sack[1]; /* D-SACK block */
279 	struct tcp_sack_block selective_acks[4]; /* The SACKS themselves*/
280 
281 	struct tcp_sack_block recv_sack_cache[4];
282 
283 	struct sk_buff *highest_sack;   /* skb just after the highest
284 					 * skb with SACKed bit set
285 					 * (validity guaranteed only if
286 					 * sacked_out > 0)
287 					 */
288 
289 	int     lost_cnt_hint;
290 	u32     retransmit_high;	/* L-bits may be on up to this seqno */
291 
292 	u32	prior_ssthresh; /* ssthresh saved at recovery start	*/
293 	u32	high_seq;	/* snd_nxt at onset of congestion	*/
294 
295 	u32	retrans_stamp;	/* Timestamp of the last retransmit,
296 				 * also used in SYN-SENT to remember stamp of
297 				 * the first SYN. */
298 	u32	undo_marker;	/* snd_una upon a new recovery episode. */
299 	int	undo_retrans;	/* number of undoable retransmissions. */
300 	u32	total_retrans;	/* Total retransmits for entire connection */
301 
302 	u32	urg_seq;	/* Seq of received urgent pointer */
303 	unsigned int		keepalive_time;	  /* time before keep alive takes place */
304 	unsigned int		keepalive_intvl;  /* time interval between keep alive probes */
305 
306 	int			linger2;
307 
308 /* Receiver side RTT estimation */
309 	struct {
310 		u32	rtt;
311 		u32	seq;
312 		u32	time;
313 	} rcv_rtt_est;
314 
315 /* Receiver queue space */
316 	struct {
317 		int	space;
318 		u32	seq;
319 		u32	time;
320 	} rcvq_space;
321 
322 /* TCP-specific MTU probe information. */
323 	struct {
324 		u32		  probe_seq_start;
325 		u32		  probe_seq_end;
326 	} mtu_probe;
327 	u32	mtu_info; /* We received an ICMP_FRAG_NEEDED / ICMPV6_PKT_TOOBIG
328 			   * while socket was owned by user.
329 			   */
330 
331 #ifdef CONFIG_TCP_MD5SIG
332 /* TCP AF-Specific parts; only used by MD5 Signature support so far */
333 	const struct tcp_sock_af_ops	*af_specific;
334 
335 /* TCP MD5 Signature Option information */
336 	struct tcp_md5sig_info	__rcu *md5sig_info;
337 #endif
338 
339 /* TCP fastopen related information */
340 	struct tcp_fastopen_request *fastopen_req;
341 	/* fastopen_rsk points to request_sock that resulted in this big
342 	 * socket. Used to retransmit SYNACKs etc.
343 	 */
344 	struct request_sock *fastopen_rsk;
345 	u32	*saved_syn;
346 };
347 
348 enum tsq_flags {
349 	TSQ_THROTTLED,
350 	TSQ_QUEUED,
351 	TCP_TSQ_DEFERRED,	   /* tcp_tasklet_func() found socket was owned */
352 	TCP_WRITE_TIMER_DEFERRED,  /* tcp_write_timer() found socket was owned */
353 	TCP_DELACK_TIMER_DEFERRED, /* tcp_delack_timer() found socket was owned */
354 	TCP_MTU_REDUCED_DEFERRED,  /* tcp_v{4|6}_err() could not call
355 				    * tcp_v{4|6}_mtu_reduced()
356 				    */
357 };
358 
tcp_sk(const struct sock * sk)359 static inline struct tcp_sock *tcp_sk(const struct sock *sk)
360 {
361 	return (struct tcp_sock *)sk;
362 }
363 
364 struct tcp_timewait_sock {
365 	struct inet_timewait_sock tw_sk;
366 #define tw_rcv_nxt tw_sk.__tw_common.skc_tw_rcv_nxt
367 #define tw_snd_nxt tw_sk.__tw_common.skc_tw_snd_nxt
368 	u32			  tw_rcv_wnd;
369 	u32			  tw_ts_offset;
370 	u32			  tw_ts_recent;
371 
372 	/* The time we sent the last out-of-window ACK: */
373 	u32			  tw_last_oow_ack_time;
374 
375 	long			  tw_ts_recent_stamp;
376 #ifdef CONFIG_TCP_MD5SIG
377 	struct tcp_md5sig_key	  *tw_md5_key;
378 #endif
379 };
380 
tcp_twsk(const struct sock * sk)381 static inline struct tcp_timewait_sock *tcp_twsk(const struct sock *sk)
382 {
383 	return (struct tcp_timewait_sock *)sk;
384 }
385 
tcp_passive_fastopen(const struct sock * sk)386 static inline bool tcp_passive_fastopen(const struct sock *sk)
387 {
388 	return (sk->sk_state == TCP_SYN_RECV &&
389 		tcp_sk(sk)->fastopen_rsk != NULL);
390 }
391 
fastopen_queue_tune(struct sock * sk,int backlog)392 static inline void fastopen_queue_tune(struct sock *sk, int backlog)
393 {
394 	struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
395 	int somaxconn = READ_ONCE(sock_net(sk)->core.sysctl_somaxconn);
396 
397 	queue->fastopenq.max_qlen = min_t(unsigned int, backlog, somaxconn);
398 }
399 
tcp_move_syn(struct tcp_sock * tp,struct request_sock * req)400 static inline void tcp_move_syn(struct tcp_sock *tp,
401 				struct request_sock *req)
402 {
403 	tp->saved_syn = req->saved_syn;
404 	req->saved_syn = NULL;
405 }
406 
tcp_saved_syn_free(struct tcp_sock * tp)407 static inline void tcp_saved_syn_free(struct tcp_sock *tp)
408 {
409 	kfree(tp->saved_syn);
410 	tp->saved_syn = NULL;
411 }
412 
413 #endif	/* _LINUX_TCP_H */
414