xref: /openbmc/linux/include/net/tcp.h (revision 95188aaf)
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/dmaengine.h>
31 #include <linux/crypto.h>
32 #include <linux/cryptohash.h>
33 #include <linux/kref.h>
34 
35 #include <net/inet_connection_sock.h>
36 #include <net/inet_timewait_sock.h>
37 #include <net/inet_hashtables.h>
38 #include <net/checksum.h>
39 #include <net/request_sock.h>
40 #include <net/sock.h>
41 #include <net/snmp.h>
42 #include <net/ip.h>
43 #include <net/tcp_states.h>
44 #include <net/inet_ecn.h>
45 #include <net/dst.h>
46 
47 #include <linux/seq_file.h>
48 #include <linux/memcontrol.h>
49 
50 extern struct inet_hashinfo tcp_hashinfo;
51 
52 extern struct percpu_counter tcp_orphan_count;
53 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
54 
55 #define MAX_TCP_HEADER	(128 + MAX_HEADER)
56 #define MAX_TCP_OPTION_SPACE 40
57 
58 /*
59  * Never offer a window over 32767 without using window scaling. Some
60  * poor stacks do signed 16bit maths!
61  */
62 #define MAX_TCP_WINDOW		32767U
63 
64 /* Offer an initial receive window of 10 mss. */
65 #define TCP_DEFAULT_INIT_RCVWND	10
66 
67 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
68 #define TCP_MIN_MSS		88U
69 
70 /* The least MTU to use for probing */
71 #define TCP_BASE_MSS		512
72 
73 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
74 #define TCP_FASTRETRANS_THRESH 3
75 
76 /* Maximal reordering. */
77 #define TCP_MAX_REORDERING	127
78 
79 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
80 #define TCP_MAX_QUICKACKS	16U
81 
82 /* urg_data states */
83 #define TCP_URG_VALID	0x0100
84 #define TCP_URG_NOTYET	0x0200
85 #define TCP_URG_READ	0x0400
86 
87 #define TCP_RETR1	3	/*
88 				 * This is how many retries it does before it
89 				 * tries to figure out if the gateway is
90 				 * down. Minimal RFC value is 3; it corresponds
91 				 * to ~3sec-8min depending on RTO.
92 				 */
93 
94 #define TCP_RETR2	15	/*
95 				 * This should take at least
96 				 * 90 minutes to time out.
97 				 * RFC1122 says that the limit is 100 sec.
98 				 * 15 is ~13-30min depending on RTO.
99 				 */
100 
101 #define TCP_SYN_RETRIES	 6	/* This is how many retries are done
102 				 * when active opening a connection.
103 				 * RFC1122 says the minimum retry MUST
104 				 * be at least 180secs.  Nevertheless
105 				 * this value is corresponding to
106 				 * 63secs of retransmission with the
107 				 * current initial RTO.
108 				 */
109 
110 #define TCP_SYNACK_RETRIES 5	/* This is how may retries are done
111 				 * when passive opening a connection.
112 				 * This is corresponding to 31secs of
113 				 * retransmission with the current
114 				 * initial RTO.
115 				 */
116 
117 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
118 				  * state, about 60 seconds	*/
119 #define TCP_FIN_TIMEOUT	TCP_TIMEWAIT_LEN
120                                  /* BSD style FIN_WAIT2 deadlock breaker.
121 				  * It used to be 3min, new value is 60sec,
122 				  * to combine FIN-WAIT-2 timeout with
123 				  * TIME-WAIT timer.
124 				  */
125 
126 #define TCP_DELACK_MAX	((unsigned)(HZ/5))	/* maximal time to delay before sending an ACK */
127 #if HZ >= 100
128 #define TCP_DELACK_MIN	((unsigned)(HZ/25))	/* minimal time to delay before sending an ACK */
129 #define TCP_ATO_MIN	((unsigned)(HZ/25))
130 #else
131 #define TCP_DELACK_MIN	4U
132 #define TCP_ATO_MIN	4U
133 #endif
134 #define TCP_RTO_MAX	((unsigned)(120*HZ))
135 #define TCP_RTO_MIN	((unsigned)(HZ/5))
136 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ))	/* RFC6298 2.1 initial RTO value	*/
137 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ))	/* RFC 1122 initial RTO value, now
138 						 * used as a fallback RTO for the
139 						 * initial data transmission if no
140 						 * valid RTT sample has been acquired,
141 						 * most likely due to retrans in 3WHS.
142 						 */
143 
144 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
145 					                 * for local resources.
146 					                 */
147 
148 #define TCP_KEEPALIVE_TIME	(120*60*HZ)	/* two hours */
149 #define TCP_KEEPALIVE_PROBES	9		/* Max of 9 keepalive probes	*/
150 #define TCP_KEEPALIVE_INTVL	(75*HZ)
151 
152 #define MAX_TCP_KEEPIDLE	32767
153 #define MAX_TCP_KEEPINTVL	32767
154 #define MAX_TCP_KEEPCNT		127
155 #define MAX_TCP_SYNCNT		127
156 
157 #define TCP_SYNQ_INTERVAL	(HZ/5)	/* Period of SYNACK timer */
158 
159 #define TCP_PAWS_24DAYS	(60 * 60 * 24 * 24)
160 #define TCP_PAWS_MSL	60		/* Per-host timestamps are invalidated
161 					 * after this time. It should be equal
162 					 * (or greater than) TCP_TIMEWAIT_LEN
163 					 * to provide reliability equal to one
164 					 * provided by timewait state.
165 					 */
166 #define TCP_PAWS_WINDOW	1		/* Replay window for per-host
167 					 * timestamps. It must be less than
168 					 * minimal timewait lifetime.
169 					 */
170 /*
171  *	TCP option
172  */
173 
174 #define TCPOPT_NOP		1	/* Padding */
175 #define TCPOPT_EOL		0	/* End of options */
176 #define TCPOPT_MSS		2	/* Segment size negotiating */
177 #define TCPOPT_WINDOW		3	/* Window scaling */
178 #define TCPOPT_SACK_PERM        4       /* SACK Permitted */
179 #define TCPOPT_SACK             5       /* SACK Block */
180 #define TCPOPT_TIMESTAMP	8	/* Better RTT estimations/PAWS */
181 #define TCPOPT_MD5SIG		19	/* MD5 Signature (RFC2385) */
182 #define TCPOPT_COOKIE		253	/* Cookie extension (experimental) */
183 #define TCPOPT_EXP		254	/* Experimental */
184 /* Magic number to be after the option value for sharing TCP
185  * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
186  */
187 #define TCPOPT_FASTOPEN_MAGIC	0xF989
188 
189 /*
190  *     TCP option lengths
191  */
192 
193 #define TCPOLEN_MSS            4
194 #define TCPOLEN_WINDOW         3
195 #define TCPOLEN_SACK_PERM      2
196 #define TCPOLEN_TIMESTAMP      10
197 #define TCPOLEN_MD5SIG         18
198 #define TCPOLEN_EXP_FASTOPEN_BASE  4
199 #define TCPOLEN_COOKIE_BASE    2	/* Cookie-less header extension */
200 #define TCPOLEN_COOKIE_PAIR    3	/* Cookie pair header extension */
201 #define TCPOLEN_COOKIE_MIN     (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN)
202 #define TCPOLEN_COOKIE_MAX     (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX)
203 
204 /* But this is what stacks really send out. */
205 #define TCPOLEN_TSTAMP_ALIGNED		12
206 #define TCPOLEN_WSCALE_ALIGNED		4
207 #define TCPOLEN_SACKPERM_ALIGNED	4
208 #define TCPOLEN_SACK_BASE		2
209 #define TCPOLEN_SACK_BASE_ALIGNED	4
210 #define TCPOLEN_SACK_PERBLOCK		8
211 #define TCPOLEN_MD5SIG_ALIGNED		20
212 #define TCPOLEN_MSS_ALIGNED		4
213 
214 /* Flags in tp->nonagle */
215 #define TCP_NAGLE_OFF		1	/* Nagle's algo is disabled */
216 #define TCP_NAGLE_CORK		2	/* Socket is corked	    */
217 #define TCP_NAGLE_PUSH		4	/* Cork is overridden for already queued data */
218 
219 /* TCP thin-stream limits */
220 #define TCP_THIN_LINEAR_RETRIES 6       /* After 6 linear retries, do exp. backoff */
221 
222 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
223 #define TCP_INIT_CWND		10
224 
225 /* Bit Flags for sysctl_tcp_fastopen */
226 #define	TFO_CLIENT_ENABLE	1
227 #define	TFO_SERVER_ENABLE	2
228 #define	TFO_CLIENT_NO_COOKIE	4	/* Data in SYN w/o cookie option */
229 
230 /* Process SYN data but skip cookie validation */
231 #define	TFO_SERVER_COOKIE_NOT_CHKED	0x100
232 /* Accept SYN data w/o any cookie option */
233 #define	TFO_SERVER_COOKIE_NOT_REQD	0x200
234 
235 /* Force enable TFO on all listeners, i.e., not requiring the
236  * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen.
237  */
238 #define	TFO_SERVER_WO_SOCKOPT1	0x400
239 #define	TFO_SERVER_WO_SOCKOPT2	0x800
240 /* Always create TFO child sockets on a TFO listener even when
241  * cookie/data not present. (For testing purpose!)
242  */
243 #define	TFO_SERVER_ALWAYS	0x1000
244 
245 extern struct inet_timewait_death_row tcp_death_row;
246 
247 /* sysctl variables for tcp */
248 extern int sysctl_tcp_timestamps;
249 extern int sysctl_tcp_window_scaling;
250 extern int sysctl_tcp_sack;
251 extern int sysctl_tcp_fin_timeout;
252 extern int sysctl_tcp_keepalive_time;
253 extern int sysctl_tcp_keepalive_probes;
254 extern int sysctl_tcp_keepalive_intvl;
255 extern int sysctl_tcp_syn_retries;
256 extern int sysctl_tcp_synack_retries;
257 extern int sysctl_tcp_retries1;
258 extern int sysctl_tcp_retries2;
259 extern int sysctl_tcp_orphan_retries;
260 extern int sysctl_tcp_syncookies;
261 extern int sysctl_tcp_fastopen;
262 extern int sysctl_tcp_retrans_collapse;
263 extern int sysctl_tcp_stdurg;
264 extern int sysctl_tcp_rfc1337;
265 extern int sysctl_tcp_abort_on_overflow;
266 extern int sysctl_tcp_max_orphans;
267 extern int sysctl_tcp_fack;
268 extern int sysctl_tcp_reordering;
269 extern int sysctl_tcp_dsack;
270 extern int sysctl_tcp_wmem[3];
271 extern int sysctl_tcp_rmem[3];
272 extern int sysctl_tcp_app_win;
273 extern int sysctl_tcp_adv_win_scale;
274 extern int sysctl_tcp_tw_reuse;
275 extern int sysctl_tcp_frto;
276 extern int sysctl_tcp_frto_response;
277 extern int sysctl_tcp_low_latency;
278 extern int sysctl_tcp_dma_copybreak;
279 extern int sysctl_tcp_nometrics_save;
280 extern int sysctl_tcp_moderate_rcvbuf;
281 extern int sysctl_tcp_tso_win_divisor;
282 extern int sysctl_tcp_mtu_probing;
283 extern int sysctl_tcp_base_mss;
284 extern int sysctl_tcp_workaround_signed_windows;
285 extern int sysctl_tcp_slow_start_after_idle;
286 extern int sysctl_tcp_max_ssthresh;
287 extern int sysctl_tcp_cookie_size;
288 extern int sysctl_tcp_thin_linear_timeouts;
289 extern int sysctl_tcp_thin_dupack;
290 extern int sysctl_tcp_early_retrans;
291 extern int sysctl_tcp_limit_output_bytes;
292 extern int sysctl_tcp_challenge_ack_limit;
293 
294 extern atomic_long_t tcp_memory_allocated;
295 extern struct percpu_counter tcp_sockets_allocated;
296 extern int tcp_memory_pressure;
297 
298 /*
299  * The next routines deal with comparing 32 bit unsigned ints
300  * and worry about wraparound (automatic with unsigned arithmetic).
301  */
302 
303 static inline bool before(__u32 seq1, __u32 seq2)
304 {
305         return (__s32)(seq1-seq2) < 0;
306 }
307 #define after(seq2, seq1) 	before(seq1, seq2)
308 
309 /* is s2<=s1<=s3 ? */
310 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
311 {
312 	return seq3 - seq2 >= seq1 - seq2;
313 }
314 
315 static inline bool tcp_out_of_memory(struct sock *sk)
316 {
317 	if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
318 	    sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
319 		return true;
320 	return false;
321 }
322 
323 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
324 {
325 	struct percpu_counter *ocp = sk->sk_prot->orphan_count;
326 	int orphans = percpu_counter_read_positive(ocp);
327 
328 	if (orphans << shift > sysctl_tcp_max_orphans) {
329 		orphans = percpu_counter_sum_positive(ocp);
330 		if (orphans << shift > sysctl_tcp_max_orphans)
331 			return true;
332 	}
333 	return false;
334 }
335 
336 extern bool tcp_check_oom(struct sock *sk, int shift);
337 
338 /* syncookies: remember time of last synqueue overflow */
339 static inline void tcp_synq_overflow(struct sock *sk)
340 {
341 	tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
342 }
343 
344 /* syncookies: no recent synqueue overflow on this listening socket? */
345 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
346 {
347 	unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
348 	return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
349 }
350 
351 extern struct proto tcp_prot;
352 
353 #define TCP_INC_STATS(net, field)	SNMP_INC_STATS((net)->mib.tcp_statistics, field)
354 #define TCP_INC_STATS_BH(net, field)	SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
355 #define TCP_DEC_STATS(net, field)	SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
356 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
357 #define TCP_ADD_STATS(net, field, val)	SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
358 
359 extern void tcp_init_mem(struct net *net);
360 
361 extern void tcp_tasklet_init(void);
362 
363 extern void tcp_v4_err(struct sk_buff *skb, u32);
364 
365 extern void tcp_shutdown (struct sock *sk, int how);
366 
367 extern void tcp_v4_early_demux(struct sk_buff *skb);
368 extern int tcp_v4_rcv(struct sk_buff *skb);
369 
370 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
371 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
372 		       size_t size);
373 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset,
374 			size_t size, int flags);
375 extern void tcp_release_cb(struct sock *sk);
376 extern void tcp_write_timer_handler(struct sock *sk);
377 extern void tcp_delack_timer_handler(struct sock *sk);
378 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
379 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
380 				 const struct tcphdr *th, unsigned int len);
381 extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
382 			       const struct tcphdr *th, unsigned int len);
383 extern void tcp_rcv_space_adjust(struct sock *sk);
384 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
385 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
386 extern void tcp_twsk_destructor(struct sock *sk);
387 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
388 			       struct pipe_inode_info *pipe, size_t len,
389 			       unsigned int flags);
390 
391 static inline void tcp_dec_quickack_mode(struct sock *sk,
392 					 const unsigned int pkts)
393 {
394 	struct inet_connection_sock *icsk = inet_csk(sk);
395 
396 	if (icsk->icsk_ack.quick) {
397 		if (pkts >= icsk->icsk_ack.quick) {
398 			icsk->icsk_ack.quick = 0;
399 			/* Leaving quickack mode we deflate ATO. */
400 			icsk->icsk_ack.ato   = TCP_ATO_MIN;
401 		} else
402 			icsk->icsk_ack.quick -= pkts;
403 	}
404 }
405 
406 #define	TCP_ECN_OK		1
407 #define	TCP_ECN_QUEUE_CWR	2
408 #define	TCP_ECN_DEMAND_CWR	4
409 #define	TCP_ECN_SEEN		8
410 
411 enum tcp_tw_status {
412 	TCP_TW_SUCCESS = 0,
413 	TCP_TW_RST = 1,
414 	TCP_TW_ACK = 2,
415 	TCP_TW_SYN = 3
416 };
417 
418 
419 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
420 						     struct sk_buff *skb,
421 						     const struct tcphdr *th);
422 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
423 				   struct request_sock *req,
424 				   struct request_sock **prev,
425 				   bool fastopen);
426 extern int tcp_child_process(struct sock *parent, struct sock *child,
427 			     struct sk_buff *skb);
428 extern bool tcp_use_frto(struct sock *sk);
429 extern void tcp_enter_frto(struct sock *sk);
430 extern void tcp_enter_loss(struct sock *sk, int how);
431 extern void tcp_clear_retrans(struct tcp_sock *tp);
432 extern void tcp_update_metrics(struct sock *sk);
433 extern void tcp_init_metrics(struct sock *sk);
434 extern void tcp_metrics_init(void);
435 extern bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, bool paws_check);
436 extern bool tcp_remember_stamp(struct sock *sk);
437 extern bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
438 extern void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
439 extern void tcp_disable_fack(struct tcp_sock *tp);
440 extern void tcp_close(struct sock *sk, long timeout);
441 extern void tcp_init_sock(struct sock *sk);
442 extern unsigned int tcp_poll(struct file * file, struct socket *sock,
443 			     struct poll_table_struct *wait);
444 extern int tcp_getsockopt(struct sock *sk, int level, int optname,
445 			  char __user *optval, int __user *optlen);
446 extern int tcp_setsockopt(struct sock *sk, int level, int optname,
447 			  char __user *optval, unsigned int optlen);
448 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
449 				 char __user *optval, int __user *optlen);
450 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
451 				 char __user *optval, unsigned int optlen);
452 extern void tcp_set_keepalive(struct sock *sk, int val);
453 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
454 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
455 		       size_t len, int nonblock, int flags, int *addr_len);
456 extern void tcp_parse_options(const struct sk_buff *skb,
457 			      struct tcp_options_received *opt_rx, const u8 **hvpp,
458 			      int estab, struct tcp_fastopen_cookie *foc);
459 extern const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
460 
461 /*
462  *	TCP v4 functions exported for the inet6 API
463  */
464 
465 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
466 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
467 extern struct sock * tcp_create_openreq_child(struct sock *sk,
468 					      struct request_sock *req,
469 					      struct sk_buff *skb);
470 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
471 					  struct request_sock *req,
472 					  struct dst_entry *dst);
473 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
474 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr,
475 			  int addr_len);
476 extern int tcp_connect(struct sock *sk);
477 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst,
478 					struct request_sock *req,
479 					struct request_values *rvp,
480 					struct tcp_fastopen_cookie *foc);
481 extern int tcp_disconnect(struct sock *sk, int flags);
482 
483 void tcp_connect_init(struct sock *sk);
484 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
485 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
486 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
487 
488 /* From syncookies.c */
489 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
490 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
491 				    struct ip_options *opt);
492 #ifdef CONFIG_SYN_COOKIES
493 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
494 				     __u16 *mss);
495 #else
496 static inline __u32 cookie_v4_init_sequence(struct sock *sk,
497 					    struct sk_buff *skb,
498 					    __u16 *mss)
499 {
500 	return 0;
501 }
502 #endif
503 
504 extern __u32 cookie_init_timestamp(struct request_sock *req);
505 extern bool cookie_check_timestamp(struct tcp_options_received *opt,
506 				struct net *net, bool *ecn_ok);
507 
508 /* From net/ipv6/syncookies.c */
509 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
510 #ifdef CONFIG_SYN_COOKIES
511 extern __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb,
512 				     __u16 *mss);
513 #else
514 static inline __u32 cookie_v6_init_sequence(struct sock *sk,
515 					    struct sk_buff *skb,
516 					    __u16 *mss)
517 {
518 	return 0;
519 }
520 #endif
521 /* tcp_output.c */
522 
523 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
524 				      int nonagle);
525 extern bool tcp_may_send_now(struct sock *sk);
526 extern int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
527 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
528 extern void tcp_retransmit_timer(struct sock *sk);
529 extern void tcp_xmit_retransmit_queue(struct sock *);
530 extern void tcp_simple_retransmit(struct sock *);
531 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
532 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
533 
534 extern void tcp_send_probe0(struct sock *);
535 extern void tcp_send_partial(struct sock *);
536 extern int tcp_write_wakeup(struct sock *);
537 extern void tcp_send_fin(struct sock *sk);
538 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
539 extern int tcp_send_synack(struct sock *);
540 extern bool tcp_syn_flood_action(struct sock *sk,
541 				 const struct sk_buff *skb,
542 				 const char *proto);
543 extern void tcp_push_one(struct sock *, unsigned int mss_now);
544 extern void tcp_send_ack(struct sock *sk);
545 extern void tcp_send_delayed_ack(struct sock *sk);
546 
547 /* tcp_input.c */
548 extern void tcp_cwnd_application_limited(struct sock *sk);
549 extern void tcp_resume_early_retransmit(struct sock *sk);
550 extern void tcp_rearm_rto(struct sock *sk);
551 extern void tcp_reset(struct sock *sk);
552 
553 /* tcp_timer.c */
554 extern void tcp_init_xmit_timers(struct sock *);
555 static inline void tcp_clear_xmit_timers(struct sock *sk)
556 {
557 	inet_csk_clear_xmit_timers(sk);
558 }
559 
560 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
561 extern unsigned int tcp_current_mss(struct sock *sk);
562 
563 /* Bound MSS / TSO packet size with the half of the window */
564 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
565 {
566 	int cutoff;
567 
568 	/* When peer uses tiny windows, there is no use in packetizing
569 	 * to sub-MSS pieces for the sake of SWS or making sure there
570 	 * are enough packets in the pipe for fast recovery.
571 	 *
572 	 * On the other hand, for extremely large MSS devices, handling
573 	 * smaller than MSS windows in this way does make sense.
574 	 */
575 	if (tp->max_window >= 512)
576 		cutoff = (tp->max_window >> 1);
577 	else
578 		cutoff = tp->max_window;
579 
580 	if (cutoff && pktsize > cutoff)
581 		return max_t(int, cutoff, 68U - tp->tcp_header_len);
582 	else
583 		return pktsize;
584 }
585 
586 /* tcp.c */
587 extern void tcp_get_info(const struct sock *, struct tcp_info *);
588 
589 /* Read 'sendfile()'-style from a TCP socket */
590 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
591 				unsigned int, size_t);
592 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
593 			 sk_read_actor_t recv_actor);
594 
595 extern void tcp_initialize_rcv_mss(struct sock *sk);
596 
597 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
598 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
599 extern void tcp_mtup_init(struct sock *sk);
600 extern void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt);
601 extern void tcp_init_buffer_space(struct sock *sk);
602 
603 static inline void tcp_bound_rto(const struct sock *sk)
604 {
605 	if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
606 		inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
607 }
608 
609 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
610 {
611 	return (tp->srtt >> 3) + tp->rttvar;
612 }
613 
614 extern void tcp_set_rto(struct sock *sk);
615 
616 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
617 {
618 	tp->pred_flags = htonl((tp->tcp_header_len << 26) |
619 			       ntohl(TCP_FLAG_ACK) |
620 			       snd_wnd);
621 }
622 
623 static inline void tcp_fast_path_on(struct tcp_sock *tp)
624 {
625 	__tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
626 }
627 
628 static inline void tcp_fast_path_check(struct sock *sk)
629 {
630 	struct tcp_sock *tp = tcp_sk(sk);
631 
632 	if (skb_queue_empty(&tp->out_of_order_queue) &&
633 	    tp->rcv_wnd &&
634 	    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
635 	    !tp->urg_data)
636 		tcp_fast_path_on(tp);
637 }
638 
639 /* Compute the actual rto_min value */
640 static inline u32 tcp_rto_min(struct sock *sk)
641 {
642 	const struct dst_entry *dst = __sk_dst_get(sk);
643 	u32 rto_min = TCP_RTO_MIN;
644 
645 	if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
646 		rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
647 	return rto_min;
648 }
649 
650 /* Compute the actual receive window we are currently advertising.
651  * Rcv_nxt can be after the window if our peer push more data
652  * than the offered window.
653  */
654 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
655 {
656 	s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
657 
658 	if (win < 0)
659 		win = 0;
660 	return (u32) win;
661 }
662 
663 /* Choose a new window, without checks for shrinking, and without
664  * scaling applied to the result.  The caller does these things
665  * if necessary.  This is a "raw" window selection.
666  */
667 extern u32 __tcp_select_window(struct sock *sk);
668 
669 void tcp_send_window_probe(struct sock *sk);
670 
671 /* TCP timestamps are only 32-bits, this causes a slight
672  * complication on 64-bit systems since we store a snapshot
673  * of jiffies in the buffer control blocks below.  We decided
674  * to use only the low 32-bits of jiffies and hide the ugly
675  * casts with the following macro.
676  */
677 #define tcp_time_stamp		((__u32)(jiffies))
678 
679 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
680 
681 #define TCPHDR_FIN 0x01
682 #define TCPHDR_SYN 0x02
683 #define TCPHDR_RST 0x04
684 #define TCPHDR_PSH 0x08
685 #define TCPHDR_ACK 0x10
686 #define TCPHDR_URG 0x20
687 #define TCPHDR_ECE 0x40
688 #define TCPHDR_CWR 0x80
689 
690 /* This is what the send packet queuing engine uses to pass
691  * TCP per-packet control information to the transmission code.
692  * We also store the host-order sequence numbers in here too.
693  * This is 44 bytes if IPV6 is enabled.
694  * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
695  */
696 struct tcp_skb_cb {
697 	union {
698 		struct inet_skb_parm	h4;
699 #if IS_ENABLED(CONFIG_IPV6)
700 		struct inet6_skb_parm	h6;
701 #endif
702 	} header;	/* For incoming frames		*/
703 	__u32		seq;		/* Starting sequence number	*/
704 	__u32		end_seq;	/* SEQ + FIN + SYN + datalen	*/
705 	__u32		when;		/* used to compute rtt's	*/
706 	__u8		tcp_flags;	/* TCP header flags. (tcp[13])	*/
707 
708 	__u8		sacked;		/* State flags for SACK/FACK.	*/
709 #define TCPCB_SACKED_ACKED	0x01	/* SKB ACK'd by a SACK block	*/
710 #define TCPCB_SACKED_RETRANS	0x02	/* SKB retransmitted		*/
711 #define TCPCB_LOST		0x04	/* SKB is lost			*/
712 #define TCPCB_TAGBITS		0x07	/* All tag bits			*/
713 #define TCPCB_EVER_RETRANS	0x80	/* Ever retransmitted frame	*/
714 #define TCPCB_RETRANS		(TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
715 
716 	__u8		ip_dsfield;	/* IPv4 tos or IPv6 dsfield	*/
717 	/* 1 byte hole */
718 	__u32		ack_seq;	/* Sequence number ACK'd	*/
719 };
720 
721 #define TCP_SKB_CB(__skb)	((struct tcp_skb_cb *)&((__skb)->cb[0]))
722 
723 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set
724  *
725  * If we receive a SYN packet with these bits set, it means a network is
726  * playing bad games with TOS bits. In order to avoid possible false congestion
727  * notifications, we disable TCP ECN negociation.
728  */
729 static inline void
730 TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb,
731 		struct net *net)
732 {
733 	const struct tcphdr *th = tcp_hdr(skb);
734 
735 	if (net->ipv4.sysctl_tcp_ecn && th->ece && th->cwr &&
736 	    INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield))
737 		inet_rsk(req)->ecn_ok = 1;
738 }
739 
740 /* Due to TSO, an SKB can be composed of multiple actual
741  * packets.  To keep these tracked properly, we use this.
742  */
743 static inline int tcp_skb_pcount(const struct sk_buff *skb)
744 {
745 	return skb_shinfo(skb)->gso_segs;
746 }
747 
748 /* This is valid iff tcp_skb_pcount() > 1. */
749 static inline int tcp_skb_mss(const struct sk_buff *skb)
750 {
751 	return skb_shinfo(skb)->gso_size;
752 }
753 
754 /* Events passed to congestion control interface */
755 enum tcp_ca_event {
756 	CA_EVENT_TX_START,	/* first transmit when no packets in flight */
757 	CA_EVENT_CWND_RESTART,	/* congestion window restart */
758 	CA_EVENT_COMPLETE_CWR,	/* end of congestion recovery */
759 	CA_EVENT_FRTO,		/* fast recovery timeout */
760 	CA_EVENT_LOSS,		/* loss timeout */
761 	CA_EVENT_FAST_ACK,	/* in sequence ack */
762 	CA_EVENT_SLOW_ACK,	/* other ack */
763 };
764 
765 /*
766  * Interface for adding new TCP congestion control handlers
767  */
768 #define TCP_CA_NAME_MAX	16
769 #define TCP_CA_MAX	128
770 #define TCP_CA_BUF_MAX	(TCP_CA_NAME_MAX*TCP_CA_MAX)
771 
772 #define TCP_CONG_NON_RESTRICTED 0x1
773 #define TCP_CONG_RTT_STAMP	0x2
774 
775 struct tcp_congestion_ops {
776 	struct list_head	list;
777 	unsigned long flags;
778 
779 	/* initialize private data (optional) */
780 	void (*init)(struct sock *sk);
781 	/* cleanup private data  (optional) */
782 	void (*release)(struct sock *sk);
783 
784 	/* return slow start threshold (required) */
785 	u32 (*ssthresh)(struct sock *sk);
786 	/* lower bound for congestion window (optional) */
787 	u32 (*min_cwnd)(const struct sock *sk);
788 	/* do new cwnd calculation (required) */
789 	void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
790 	/* call before changing ca_state (optional) */
791 	void (*set_state)(struct sock *sk, u8 new_state);
792 	/* call when cwnd event occurs (optional) */
793 	void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
794 	/* new value of cwnd after loss (optional) */
795 	u32  (*undo_cwnd)(struct sock *sk);
796 	/* hook for packet ack accounting (optional) */
797 	void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
798 	/* get info for inet_diag (optional) */
799 	void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
800 
801 	char 		name[TCP_CA_NAME_MAX];
802 	struct module 	*owner;
803 };
804 
805 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
806 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
807 
808 extern void tcp_init_congestion_control(struct sock *sk);
809 extern void tcp_cleanup_congestion_control(struct sock *sk);
810 extern int tcp_set_default_congestion_control(const char *name);
811 extern void tcp_get_default_congestion_control(char *name);
812 extern void tcp_get_available_congestion_control(char *buf, size_t len);
813 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
814 extern int tcp_set_allowed_congestion_control(char *allowed);
815 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
816 extern void tcp_slow_start(struct tcp_sock *tp);
817 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
818 
819 extern struct tcp_congestion_ops tcp_init_congestion_ops;
820 extern u32 tcp_reno_ssthresh(struct sock *sk);
821 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
822 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
823 extern struct tcp_congestion_ops tcp_reno;
824 
825 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
826 {
827 	struct inet_connection_sock *icsk = inet_csk(sk);
828 
829 	if (icsk->icsk_ca_ops->set_state)
830 		icsk->icsk_ca_ops->set_state(sk, ca_state);
831 	icsk->icsk_ca_state = ca_state;
832 }
833 
834 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
835 {
836 	const struct inet_connection_sock *icsk = inet_csk(sk);
837 
838 	if (icsk->icsk_ca_ops->cwnd_event)
839 		icsk->icsk_ca_ops->cwnd_event(sk, event);
840 }
841 
842 /* These functions determine how the current flow behaves in respect of SACK
843  * handling. SACK is negotiated with the peer, and therefore it can vary
844  * between different flows.
845  *
846  * tcp_is_sack - SACK enabled
847  * tcp_is_reno - No SACK
848  * tcp_is_fack - FACK enabled, implies SACK enabled
849  */
850 static inline int tcp_is_sack(const struct tcp_sock *tp)
851 {
852 	return tp->rx_opt.sack_ok;
853 }
854 
855 static inline bool tcp_is_reno(const struct tcp_sock *tp)
856 {
857 	return !tcp_is_sack(tp);
858 }
859 
860 static inline bool tcp_is_fack(const struct tcp_sock *tp)
861 {
862 	return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
863 }
864 
865 static inline void tcp_enable_fack(struct tcp_sock *tp)
866 {
867 	tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
868 }
869 
870 /* TCP early-retransmit (ER) is similar to but more conservative than
871  * the thin-dupack feature.  Enable ER only if thin-dupack is disabled.
872  */
873 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
874 {
875 	tp->do_early_retrans = sysctl_tcp_early_retrans &&
876 		!sysctl_tcp_thin_dupack && sysctl_tcp_reordering == 3;
877 	tp->early_retrans_delayed = 0;
878 }
879 
880 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
881 {
882 	tp->do_early_retrans = 0;
883 }
884 
885 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
886 {
887 	return tp->sacked_out + tp->lost_out;
888 }
889 
890 /* This determines how many packets are "in the network" to the best
891  * of our knowledge.  In many cases it is conservative, but where
892  * detailed information is available from the receiver (via SACK
893  * blocks etc.) we can make more aggressive calculations.
894  *
895  * Use this for decisions involving congestion control, use just
896  * tp->packets_out to determine if the send queue is empty or not.
897  *
898  * Read this equation as:
899  *
900  *	"Packets sent once on transmission queue" MINUS
901  *	"Packets left network, but not honestly ACKed yet" PLUS
902  *	"Packets fast retransmitted"
903  */
904 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
905 {
906 	return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
907 }
908 
909 #define TCP_INFINITE_SSTHRESH	0x7fffffff
910 
911 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
912 {
913 	return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
914 }
915 
916 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
917 {
918 	return (TCPF_CA_CWR | TCPF_CA_Recovery) &
919 	       (1 << inet_csk(sk)->icsk_ca_state);
920 }
921 
922 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
923  * The exception is cwnd reduction phase, when cwnd is decreasing towards
924  * ssthresh.
925  */
926 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
927 {
928 	const struct tcp_sock *tp = tcp_sk(sk);
929 
930 	if (tcp_in_cwnd_reduction(sk))
931 		return tp->snd_ssthresh;
932 	else
933 		return max(tp->snd_ssthresh,
934 			   ((tp->snd_cwnd >> 1) +
935 			    (tp->snd_cwnd >> 2)));
936 }
937 
938 /* Use define here intentionally to get WARN_ON location shown at the caller */
939 #define tcp_verify_left_out(tp)	WARN_ON(tcp_left_out(tp) > tp->packets_out)
940 
941 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
942 extern __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
943 
944 /* The maximum number of MSS of available cwnd for which TSO defers
945  * sending if not using sysctl_tcp_tso_win_divisor.
946  */
947 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
948 {
949 	return 3;
950 }
951 
952 /* Slow start with delack produces 3 packets of burst, so that
953  * it is safe "de facto".  This will be the default - same as
954  * the default reordering threshold - but if reordering increases,
955  * we must be able to allow cwnd to burst at least this much in order
956  * to not pull it back when holes are filled.
957  */
958 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
959 {
960 	return tp->reordering;
961 }
962 
963 /* Returns end sequence number of the receiver's advertised window */
964 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
965 {
966 	return tp->snd_una + tp->snd_wnd;
967 }
968 extern bool tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
969 
970 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
971 				       const struct sk_buff *skb)
972 {
973 	if (skb->len < mss)
974 		tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
975 }
976 
977 static inline void tcp_check_probe_timer(struct sock *sk)
978 {
979 	const struct tcp_sock *tp = tcp_sk(sk);
980 	const struct inet_connection_sock *icsk = inet_csk(sk);
981 
982 	if (!tp->packets_out && !icsk->icsk_pending)
983 		inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
984 					  icsk->icsk_rto, TCP_RTO_MAX);
985 }
986 
987 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
988 {
989 	tp->snd_wl1 = seq;
990 }
991 
992 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
993 {
994 	tp->snd_wl1 = seq;
995 }
996 
997 /*
998  * Calculate(/check) TCP checksum
999  */
1000 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1001 				   __be32 daddr, __wsum base)
1002 {
1003 	return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1004 }
1005 
1006 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1007 {
1008 	return __skb_checksum_complete(skb);
1009 }
1010 
1011 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1012 {
1013 	return !skb_csum_unnecessary(skb) &&
1014 		__tcp_checksum_complete(skb);
1015 }
1016 
1017 /* Prequeue for VJ style copy to user, combined with checksumming. */
1018 
1019 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1020 {
1021 	tp->ucopy.task = NULL;
1022 	tp->ucopy.len = 0;
1023 	tp->ucopy.memory = 0;
1024 	skb_queue_head_init(&tp->ucopy.prequeue);
1025 #ifdef CONFIG_NET_DMA
1026 	tp->ucopy.dma_chan = NULL;
1027 	tp->ucopy.wakeup = 0;
1028 	tp->ucopy.pinned_list = NULL;
1029 	tp->ucopy.dma_cookie = 0;
1030 #endif
1031 }
1032 
1033 /* Packet is added to VJ-style prequeue for processing in process
1034  * context, if a reader task is waiting. Apparently, this exciting
1035  * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1036  * failed somewhere. Latency? Burstiness? Well, at least now we will
1037  * see, why it failed. 8)8)				  --ANK
1038  *
1039  * NOTE: is this not too big to inline?
1040  */
1041 static inline bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1042 {
1043 	struct tcp_sock *tp = tcp_sk(sk);
1044 
1045 	if (sysctl_tcp_low_latency || !tp->ucopy.task)
1046 		return false;
1047 
1048 	if (skb->len <= tcp_hdrlen(skb) &&
1049 	    skb_queue_len(&tp->ucopy.prequeue) == 0)
1050 		return false;
1051 
1052 	__skb_queue_tail(&tp->ucopy.prequeue, skb);
1053 	tp->ucopy.memory += skb->truesize;
1054 	if (tp->ucopy.memory > sk->sk_rcvbuf) {
1055 		struct sk_buff *skb1;
1056 
1057 		BUG_ON(sock_owned_by_user(sk));
1058 
1059 		while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1060 			sk_backlog_rcv(sk, skb1);
1061 			NET_INC_STATS_BH(sock_net(sk),
1062 					 LINUX_MIB_TCPPREQUEUEDROPPED);
1063 		}
1064 
1065 		tp->ucopy.memory = 0;
1066 	} else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1067 		wake_up_interruptible_sync_poll(sk_sleep(sk),
1068 					   POLLIN | POLLRDNORM | POLLRDBAND);
1069 		if (!inet_csk_ack_scheduled(sk))
1070 			inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1071 						  (3 * tcp_rto_min(sk)) / 4,
1072 						  TCP_RTO_MAX);
1073 	}
1074 	return true;
1075 }
1076 
1077 
1078 #undef STATE_TRACE
1079 
1080 #ifdef STATE_TRACE
1081 static const char *statename[]={
1082 	"Unused","Established","Syn Sent","Syn Recv",
1083 	"Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1084 	"Close Wait","Last ACK","Listen","Closing"
1085 };
1086 #endif
1087 extern void tcp_set_state(struct sock *sk, int state);
1088 
1089 extern void tcp_done(struct sock *sk);
1090 
1091 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1092 {
1093 	rx_opt->dsack = 0;
1094 	rx_opt->num_sacks = 0;
1095 }
1096 
1097 /* Determine a window scaling and initial window to offer. */
1098 extern void tcp_select_initial_window(int __space, __u32 mss,
1099 				      __u32 *rcv_wnd, __u32 *window_clamp,
1100 				      int wscale_ok, __u8 *rcv_wscale,
1101 				      __u32 init_rcv_wnd);
1102 
1103 static inline int tcp_win_from_space(int space)
1104 {
1105 	return sysctl_tcp_adv_win_scale<=0 ?
1106 		(space>>(-sysctl_tcp_adv_win_scale)) :
1107 		space - (space>>sysctl_tcp_adv_win_scale);
1108 }
1109 
1110 /* Note: caller must be prepared to deal with negative returns */
1111 static inline int tcp_space(const struct sock *sk)
1112 {
1113 	return tcp_win_from_space(sk->sk_rcvbuf -
1114 				  atomic_read(&sk->sk_rmem_alloc));
1115 }
1116 
1117 static inline int tcp_full_space(const struct sock *sk)
1118 {
1119 	return tcp_win_from_space(sk->sk_rcvbuf);
1120 }
1121 
1122 static inline void tcp_openreq_init(struct request_sock *req,
1123 				    struct tcp_options_received *rx_opt,
1124 				    struct sk_buff *skb)
1125 {
1126 	struct inet_request_sock *ireq = inet_rsk(req);
1127 
1128 	req->rcv_wnd = 0;		/* So that tcp_send_synack() knows! */
1129 	req->cookie_ts = 0;
1130 	tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1131 	tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1;
1132 	tcp_rsk(req)->snt_synack = 0;
1133 	req->mss = rx_opt->mss_clamp;
1134 	req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1135 	ireq->tstamp_ok = rx_opt->tstamp_ok;
1136 	ireq->sack_ok = rx_opt->sack_ok;
1137 	ireq->snd_wscale = rx_opt->snd_wscale;
1138 	ireq->wscale_ok = rx_opt->wscale_ok;
1139 	ireq->acked = 0;
1140 	ireq->ecn_ok = 0;
1141 	ireq->rmt_port = tcp_hdr(skb)->source;
1142 	ireq->loc_port = tcp_hdr(skb)->dest;
1143 }
1144 
1145 /* Compute time elapsed between SYNACK and the ACK completing 3WHS */
1146 static inline void tcp_synack_rtt_meas(struct sock *sk,
1147 				       struct request_sock *req)
1148 {
1149 	if (tcp_rsk(req)->snt_synack)
1150 		tcp_valid_rtt_meas(sk,
1151 		    tcp_time_stamp - tcp_rsk(req)->snt_synack);
1152 }
1153 
1154 extern void tcp_enter_memory_pressure(struct sock *sk);
1155 
1156 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1157 {
1158 	return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1159 }
1160 
1161 static inline int keepalive_time_when(const struct tcp_sock *tp)
1162 {
1163 	return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1164 }
1165 
1166 static inline int keepalive_probes(const struct tcp_sock *tp)
1167 {
1168 	return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1169 }
1170 
1171 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1172 {
1173 	const struct inet_connection_sock *icsk = &tp->inet_conn;
1174 
1175 	return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1176 			  tcp_time_stamp - tp->rcv_tstamp);
1177 }
1178 
1179 static inline int tcp_fin_time(const struct sock *sk)
1180 {
1181 	int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1182 	const int rto = inet_csk(sk)->icsk_rto;
1183 
1184 	if (fin_timeout < (rto << 2) - (rto >> 1))
1185 		fin_timeout = (rto << 2) - (rto >> 1);
1186 
1187 	return fin_timeout;
1188 }
1189 
1190 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1191 				  int paws_win)
1192 {
1193 	if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1194 		return true;
1195 	if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1196 		return true;
1197 	/*
1198 	 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1199 	 * then following tcp messages have valid values. Ignore 0 value,
1200 	 * or else 'negative' tsval might forbid us to accept their packets.
1201 	 */
1202 	if (!rx_opt->ts_recent)
1203 		return true;
1204 	return false;
1205 }
1206 
1207 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1208 				   int rst)
1209 {
1210 	if (tcp_paws_check(rx_opt, 0))
1211 		return false;
1212 
1213 	/* RST segments are not recommended to carry timestamp,
1214 	   and, if they do, it is recommended to ignore PAWS because
1215 	   "their cleanup function should take precedence over timestamps."
1216 	   Certainly, it is mistake. It is necessary to understand the reasons
1217 	   of this constraint to relax it: if peer reboots, clock may go
1218 	   out-of-sync and half-open connections will not be reset.
1219 	   Actually, the problem would be not existing if all
1220 	   the implementations followed draft about maintaining clock
1221 	   via reboots. Linux-2.2 DOES NOT!
1222 
1223 	   However, we can relax time bounds for RST segments to MSL.
1224 	 */
1225 	if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1226 		return false;
1227 	return true;
1228 }
1229 
1230 static inline void tcp_mib_init(struct net *net)
1231 {
1232 	/* See RFC 2012 */
1233 	TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1234 	TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1235 	TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1236 	TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1237 }
1238 
1239 /* from STCP */
1240 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1241 {
1242 	tp->lost_skb_hint = NULL;
1243 	tp->scoreboard_skb_hint = NULL;
1244 }
1245 
1246 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1247 {
1248 	tcp_clear_retrans_hints_partial(tp);
1249 	tp->retransmit_skb_hint = NULL;
1250 }
1251 
1252 /* MD5 Signature */
1253 struct crypto_hash;
1254 
1255 union tcp_md5_addr {
1256 	struct in_addr  a4;
1257 #if IS_ENABLED(CONFIG_IPV6)
1258 	struct in6_addr	a6;
1259 #endif
1260 };
1261 
1262 /* - key database */
1263 struct tcp_md5sig_key {
1264 	struct hlist_node	node;
1265 	u8			keylen;
1266 	u8			family; /* AF_INET or AF_INET6 */
1267 	union tcp_md5_addr	addr;
1268 	u8			key[TCP_MD5SIG_MAXKEYLEN];
1269 	struct rcu_head		rcu;
1270 };
1271 
1272 /* - sock block */
1273 struct tcp_md5sig_info {
1274 	struct hlist_head	head;
1275 	struct rcu_head		rcu;
1276 };
1277 
1278 /* - pseudo header */
1279 struct tcp4_pseudohdr {
1280 	__be32		saddr;
1281 	__be32		daddr;
1282 	__u8		pad;
1283 	__u8		protocol;
1284 	__be16		len;
1285 };
1286 
1287 struct tcp6_pseudohdr {
1288 	struct in6_addr	saddr;
1289 	struct in6_addr daddr;
1290 	__be32		len;
1291 	__be32		protocol;	/* including padding */
1292 };
1293 
1294 union tcp_md5sum_block {
1295 	struct tcp4_pseudohdr ip4;
1296 #if IS_ENABLED(CONFIG_IPV6)
1297 	struct tcp6_pseudohdr ip6;
1298 #endif
1299 };
1300 
1301 /* - pool: digest algorithm, hash description and scratch buffer */
1302 struct tcp_md5sig_pool {
1303 	struct hash_desc	md5_desc;
1304 	union tcp_md5sum_block	md5_blk;
1305 };
1306 
1307 /* - functions */
1308 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1309 			       const struct sock *sk,
1310 			       const struct request_sock *req,
1311 			       const struct sk_buff *skb);
1312 extern int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1313 			  int family, const u8 *newkey,
1314 			  u8 newkeylen, gfp_t gfp);
1315 extern int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1316 			  int family);
1317 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1318 					 struct sock *addr_sk);
1319 
1320 #ifdef CONFIG_TCP_MD5SIG
1321 extern struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1322 			const union tcp_md5_addr *addr, int family);
1323 #define tcp_twsk_md5_key(twsk)	((twsk)->tw_md5_key)
1324 #else
1325 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1326 					 const union tcp_md5_addr *addr,
1327 					 int family)
1328 {
1329 	return NULL;
1330 }
1331 #define tcp_twsk_md5_key(twsk)	NULL
1332 #endif
1333 
1334 extern struct tcp_md5sig_pool __percpu *tcp_alloc_md5sig_pool(struct sock *);
1335 extern void tcp_free_md5sig_pool(void);
1336 
1337 extern struct tcp_md5sig_pool	*tcp_get_md5sig_pool(void);
1338 extern void tcp_put_md5sig_pool(void);
1339 
1340 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1341 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1342 				 unsigned int header_len);
1343 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1344 			    const struct tcp_md5sig_key *key);
1345 
1346 /* From tcp_fastopen.c */
1347 extern void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1348 				   struct tcp_fastopen_cookie *cookie,
1349 				   int *syn_loss, unsigned long *last_syn_loss);
1350 extern void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1351 				   struct tcp_fastopen_cookie *cookie,
1352 				   bool syn_lost);
1353 struct tcp_fastopen_request {
1354 	/* Fast Open cookie. Size 0 means a cookie request */
1355 	struct tcp_fastopen_cookie	cookie;
1356 	struct msghdr			*data;  /* data in MSG_FASTOPEN */
1357 	u16				copied;	/* queued in tcp_connect() */
1358 };
1359 void tcp_free_fastopen_req(struct tcp_sock *tp);
1360 
1361 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1362 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1363 void tcp_fastopen_cookie_gen(__be32 addr, struct tcp_fastopen_cookie *foc);
1364 
1365 #define TCP_FASTOPEN_KEY_LENGTH 16
1366 
1367 /* Fastopen key context */
1368 struct tcp_fastopen_context {
1369 	struct crypto_cipher __rcu	*tfm;
1370 	__u8				key[TCP_FASTOPEN_KEY_LENGTH];
1371 	struct rcu_head			rcu;
1372 };
1373 
1374 /* write queue abstraction */
1375 static inline void tcp_write_queue_purge(struct sock *sk)
1376 {
1377 	struct sk_buff *skb;
1378 
1379 	while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1380 		sk_wmem_free_skb(sk, skb);
1381 	sk_mem_reclaim(sk);
1382 	tcp_clear_all_retrans_hints(tcp_sk(sk));
1383 }
1384 
1385 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1386 {
1387 	return skb_peek(&sk->sk_write_queue);
1388 }
1389 
1390 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1391 {
1392 	return skb_peek_tail(&sk->sk_write_queue);
1393 }
1394 
1395 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1396 						   const struct sk_buff *skb)
1397 {
1398 	return skb_queue_next(&sk->sk_write_queue, skb);
1399 }
1400 
1401 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1402 						   const struct sk_buff *skb)
1403 {
1404 	return skb_queue_prev(&sk->sk_write_queue, skb);
1405 }
1406 
1407 #define tcp_for_write_queue(skb, sk)					\
1408 	skb_queue_walk(&(sk)->sk_write_queue, skb)
1409 
1410 #define tcp_for_write_queue_from(skb, sk)				\
1411 	skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1412 
1413 #define tcp_for_write_queue_from_safe(skb, tmp, sk)			\
1414 	skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1415 
1416 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1417 {
1418 	return sk->sk_send_head;
1419 }
1420 
1421 static inline bool tcp_skb_is_last(const struct sock *sk,
1422 				   const struct sk_buff *skb)
1423 {
1424 	return skb_queue_is_last(&sk->sk_write_queue, skb);
1425 }
1426 
1427 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1428 {
1429 	if (tcp_skb_is_last(sk, skb))
1430 		sk->sk_send_head = NULL;
1431 	else
1432 		sk->sk_send_head = tcp_write_queue_next(sk, skb);
1433 }
1434 
1435 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1436 {
1437 	if (sk->sk_send_head == skb_unlinked)
1438 		sk->sk_send_head = NULL;
1439 }
1440 
1441 static inline void tcp_init_send_head(struct sock *sk)
1442 {
1443 	sk->sk_send_head = NULL;
1444 }
1445 
1446 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1447 {
1448 	__skb_queue_tail(&sk->sk_write_queue, skb);
1449 }
1450 
1451 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1452 {
1453 	__tcp_add_write_queue_tail(sk, skb);
1454 
1455 	/* Queue it, remembering where we must start sending. */
1456 	if (sk->sk_send_head == NULL) {
1457 		sk->sk_send_head = skb;
1458 
1459 		if (tcp_sk(sk)->highest_sack == NULL)
1460 			tcp_sk(sk)->highest_sack = skb;
1461 	}
1462 }
1463 
1464 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1465 {
1466 	__skb_queue_head(&sk->sk_write_queue, skb);
1467 }
1468 
1469 /* Insert buff after skb on the write queue of sk.  */
1470 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1471 						struct sk_buff *buff,
1472 						struct sock *sk)
1473 {
1474 	__skb_queue_after(&sk->sk_write_queue, skb, buff);
1475 }
1476 
1477 /* Insert new before skb on the write queue of sk.  */
1478 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1479 						  struct sk_buff *skb,
1480 						  struct sock *sk)
1481 {
1482 	__skb_queue_before(&sk->sk_write_queue, skb, new);
1483 
1484 	if (sk->sk_send_head == skb)
1485 		sk->sk_send_head = new;
1486 }
1487 
1488 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1489 {
1490 	__skb_unlink(skb, &sk->sk_write_queue);
1491 }
1492 
1493 static inline bool tcp_write_queue_empty(struct sock *sk)
1494 {
1495 	return skb_queue_empty(&sk->sk_write_queue);
1496 }
1497 
1498 static inline void tcp_push_pending_frames(struct sock *sk)
1499 {
1500 	if (tcp_send_head(sk)) {
1501 		struct tcp_sock *tp = tcp_sk(sk);
1502 
1503 		__tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1504 	}
1505 }
1506 
1507 /* Start sequence of the skb just after the highest skb with SACKed
1508  * bit, valid only if sacked_out > 0 or when the caller has ensured
1509  * validity by itself.
1510  */
1511 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1512 {
1513 	if (!tp->sacked_out)
1514 		return tp->snd_una;
1515 
1516 	if (tp->highest_sack == NULL)
1517 		return tp->snd_nxt;
1518 
1519 	return TCP_SKB_CB(tp->highest_sack)->seq;
1520 }
1521 
1522 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1523 {
1524 	tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1525 						tcp_write_queue_next(sk, skb);
1526 }
1527 
1528 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1529 {
1530 	return tcp_sk(sk)->highest_sack;
1531 }
1532 
1533 static inline void tcp_highest_sack_reset(struct sock *sk)
1534 {
1535 	tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1536 }
1537 
1538 /* Called when old skb is about to be deleted (to be combined with new skb) */
1539 static inline void tcp_highest_sack_combine(struct sock *sk,
1540 					    struct sk_buff *old,
1541 					    struct sk_buff *new)
1542 {
1543 	if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1544 		tcp_sk(sk)->highest_sack = new;
1545 }
1546 
1547 /* Determines whether this is a thin stream (which may suffer from
1548  * increased latency). Used to trigger latency-reducing mechanisms.
1549  */
1550 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1551 {
1552 	return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1553 }
1554 
1555 /* /proc */
1556 enum tcp_seq_states {
1557 	TCP_SEQ_STATE_LISTENING,
1558 	TCP_SEQ_STATE_OPENREQ,
1559 	TCP_SEQ_STATE_ESTABLISHED,
1560 	TCP_SEQ_STATE_TIME_WAIT,
1561 };
1562 
1563 int tcp_seq_open(struct inode *inode, struct file *file);
1564 
1565 struct tcp_seq_afinfo {
1566 	char				*name;
1567 	sa_family_t			family;
1568 	const struct file_operations	*seq_fops;
1569 	struct seq_operations		seq_ops;
1570 };
1571 
1572 struct tcp_iter_state {
1573 	struct seq_net_private	p;
1574 	sa_family_t		family;
1575 	enum tcp_seq_states	state;
1576 	struct sock		*syn_wait_sk;
1577 	int			bucket, offset, sbucket, num;
1578 	kuid_t			uid;
1579 	loff_t			last_pos;
1580 };
1581 
1582 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1583 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1584 
1585 extern struct request_sock_ops tcp_request_sock_ops;
1586 extern struct request_sock_ops tcp6_request_sock_ops;
1587 
1588 extern void tcp_v4_destroy_sock(struct sock *sk);
1589 
1590 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1591 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb,
1592 				       netdev_features_t features);
1593 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1594 					struct sk_buff *skb);
1595 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1596 					 struct sk_buff *skb);
1597 extern int tcp_gro_complete(struct sk_buff *skb);
1598 extern int tcp4_gro_complete(struct sk_buff *skb);
1599 
1600 #ifdef CONFIG_PROC_FS
1601 extern int tcp4_proc_init(void);
1602 extern void tcp4_proc_exit(void);
1603 #endif
1604 
1605 /* TCP af-specific functions */
1606 struct tcp_sock_af_ops {
1607 #ifdef CONFIG_TCP_MD5SIG
1608 	struct tcp_md5sig_key	*(*md5_lookup) (struct sock *sk,
1609 						struct sock *addr_sk);
1610 	int			(*calc_md5_hash) (char *location,
1611 						  struct tcp_md5sig_key *md5,
1612 						  const struct sock *sk,
1613 						  const struct request_sock *req,
1614 						  const struct sk_buff *skb);
1615 	int			(*md5_parse) (struct sock *sk,
1616 					      char __user *optval,
1617 					      int optlen);
1618 #endif
1619 };
1620 
1621 struct tcp_request_sock_ops {
1622 #ifdef CONFIG_TCP_MD5SIG
1623 	struct tcp_md5sig_key	*(*md5_lookup) (struct sock *sk,
1624 						struct request_sock *req);
1625 	int			(*calc_md5_hash) (char *location,
1626 						  struct tcp_md5sig_key *md5,
1627 						  const struct sock *sk,
1628 						  const struct request_sock *req,
1629 						  const struct sk_buff *skb);
1630 #endif
1631 };
1632 
1633 /* Using SHA1 for now, define some constants.
1634  */
1635 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS)
1636 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4)
1637 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS)
1638 
1639 extern int tcp_cookie_generator(u32 *bakery);
1640 
1641 /**
1642  *	struct tcp_cookie_values - each socket needs extra space for the
1643  *	cookies, together with (optional) space for any SYN data.
1644  *
1645  *	A tcp_sock contains a pointer to the current value, and this is
1646  *	cloned to the tcp_timewait_sock.
1647  *
1648  * @cookie_pair:	variable data from the option exchange.
1649  *
1650  * @cookie_desired:	user specified tcpct_cookie_desired.  Zero
1651  *			indicates default (sysctl_tcp_cookie_size).
1652  *			After cookie sent, remembers size of cookie.
1653  *			Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX.
1654  *
1655  * @s_data_desired:	user specified tcpct_s_data_desired.  When the
1656  *			constant payload is specified (@s_data_constant),
1657  *			holds its length instead.
1658  *			Range 0 to TCP_MSS_DESIRED.
1659  *
1660  * @s_data_payload:	constant data that is to be included in the
1661  *			payload of SYN or SYNACK segments when the
1662  *			cookie option is present.
1663  */
1664 struct tcp_cookie_values {
1665 	struct kref	kref;
1666 	u8		cookie_pair[TCP_COOKIE_PAIR_SIZE];
1667 	u8		cookie_pair_size;
1668 	u8		cookie_desired;
1669 	u16		s_data_desired:11,
1670 			s_data_constant:1,
1671 			s_data_in:1,
1672 			s_data_out:1,
1673 			s_data_unused:2;
1674 	u8		s_data_payload[0];
1675 };
1676 
1677 static inline void tcp_cookie_values_release(struct kref *kref)
1678 {
1679 	kfree(container_of(kref, struct tcp_cookie_values, kref));
1680 }
1681 
1682 /* The length of constant payload data.  Note that s_data_desired is
1683  * overloaded, depending on s_data_constant: either the length of constant
1684  * data (returned here) or the limit on variable data.
1685  */
1686 static inline int tcp_s_data_size(const struct tcp_sock *tp)
1687 {
1688 	return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant)
1689 		? tp->cookie_values->s_data_desired
1690 		: 0;
1691 }
1692 
1693 /**
1694  *	struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace.
1695  *
1696  *	As tcp_request_sock has already been extended in other places, the
1697  *	only remaining method is to pass stack values along as function
1698  *	parameters.  These parameters are not needed after sending SYNACK.
1699  *
1700  * @cookie_bakery:	cryptographic secret and message workspace.
1701  *
1702  * @cookie_plus:	bytes in authenticator/cookie option, copied from
1703  *			struct tcp_options_received (above).
1704  */
1705 struct tcp_extend_values {
1706 	struct request_values		rv;
1707 	u32				cookie_bakery[COOKIE_WORKSPACE_WORDS];
1708 	u8				cookie_plus:6,
1709 					cookie_out_never:1,
1710 					cookie_in_always:1;
1711 };
1712 
1713 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp)
1714 {
1715 	return (struct tcp_extend_values *)rvp;
1716 }
1717 
1718 extern void tcp_v4_init(void);
1719 extern void tcp_init(void);
1720 
1721 #endif	/* _TCP_H */
1722