xref: /openbmc/linux/include/net/tcp.h (revision 861e10be)
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_ecn;
270 extern int sysctl_tcp_dsack;
271 extern int sysctl_tcp_wmem[3];
272 extern int sysctl_tcp_rmem[3];
273 extern int sysctl_tcp_app_win;
274 extern int sysctl_tcp_adv_win_scale;
275 extern int sysctl_tcp_tw_reuse;
276 extern int sysctl_tcp_frto;
277 extern int sysctl_tcp_frto_response;
278 extern int sysctl_tcp_low_latency;
279 extern int sysctl_tcp_dma_copybreak;
280 extern int sysctl_tcp_nometrics_save;
281 extern int sysctl_tcp_moderate_rcvbuf;
282 extern int sysctl_tcp_tso_win_divisor;
283 extern int sysctl_tcp_abc;
284 extern int sysctl_tcp_mtu_probing;
285 extern int sysctl_tcp_base_mss;
286 extern int sysctl_tcp_workaround_signed_windows;
287 extern int sysctl_tcp_slow_start_after_idle;
288 extern int sysctl_tcp_max_ssthresh;
289 extern int sysctl_tcp_cookie_size;
290 extern int sysctl_tcp_thin_linear_timeouts;
291 extern int sysctl_tcp_thin_dupack;
292 extern int sysctl_tcp_early_retrans;
293 extern int sysctl_tcp_limit_output_bytes;
294 extern int sysctl_tcp_challenge_ack_limit;
295 
296 extern atomic_long_t tcp_memory_allocated;
297 extern struct percpu_counter tcp_sockets_allocated;
298 extern int tcp_memory_pressure;
299 
300 /*
301  * The next routines deal with comparing 32 bit unsigned ints
302  * and worry about wraparound (automatic with unsigned arithmetic).
303  */
304 
305 static inline bool before(__u32 seq1, __u32 seq2)
306 {
307         return (__s32)(seq1-seq2) < 0;
308 }
309 #define after(seq2, seq1) 	before(seq1, seq2)
310 
311 /* is s2<=s1<=s3 ? */
312 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
313 {
314 	return seq3 - seq2 >= seq1 - seq2;
315 }
316 
317 static inline bool tcp_out_of_memory(struct sock *sk)
318 {
319 	if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
320 	    sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
321 		return true;
322 	return false;
323 }
324 
325 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
326 {
327 	struct percpu_counter *ocp = sk->sk_prot->orphan_count;
328 	int orphans = percpu_counter_read_positive(ocp);
329 
330 	if (orphans << shift > sysctl_tcp_max_orphans) {
331 		orphans = percpu_counter_sum_positive(ocp);
332 		if (orphans << shift > sysctl_tcp_max_orphans)
333 			return true;
334 	}
335 	return false;
336 }
337 
338 extern bool tcp_check_oom(struct sock *sk, int shift);
339 
340 /* syncookies: remember time of last synqueue overflow */
341 static inline void tcp_synq_overflow(struct sock *sk)
342 {
343 	tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
344 }
345 
346 /* syncookies: no recent synqueue overflow on this listening socket? */
347 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
348 {
349 	unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
350 	return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
351 }
352 
353 extern struct proto tcp_prot;
354 
355 #define TCP_INC_STATS(net, field)	SNMP_INC_STATS((net)->mib.tcp_statistics, field)
356 #define TCP_INC_STATS_BH(net, field)	SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
357 #define TCP_DEC_STATS(net, field)	SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
358 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
359 #define TCP_ADD_STATS(net, field, val)	SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
360 
361 extern void tcp_init_mem(struct net *net);
362 
363 extern void tcp_tasklet_init(void);
364 
365 extern void tcp_v4_err(struct sk_buff *skb, u32);
366 
367 extern void tcp_shutdown (struct sock *sk, int how);
368 
369 extern void tcp_v4_early_demux(struct sk_buff *skb);
370 extern int tcp_v4_rcv(struct sk_buff *skb);
371 
372 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
373 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
374 		       size_t size);
375 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset,
376 			size_t size, int flags);
377 extern void tcp_release_cb(struct sock *sk);
378 extern void tcp_write_timer_handler(struct sock *sk);
379 extern void tcp_delack_timer_handler(struct sock *sk);
380 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
381 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
382 				 const struct tcphdr *th, unsigned int len);
383 extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
384 			       const struct tcphdr *th, unsigned int len);
385 extern void tcp_rcv_space_adjust(struct sock *sk);
386 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
387 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
388 extern void tcp_twsk_destructor(struct sock *sk);
389 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
390 			       struct pipe_inode_info *pipe, size_t len,
391 			       unsigned int flags);
392 
393 static inline void tcp_dec_quickack_mode(struct sock *sk,
394 					 const unsigned int pkts)
395 {
396 	struct inet_connection_sock *icsk = inet_csk(sk);
397 
398 	if (icsk->icsk_ack.quick) {
399 		if (pkts >= icsk->icsk_ack.quick) {
400 			icsk->icsk_ack.quick = 0;
401 			/* Leaving quickack mode we deflate ATO. */
402 			icsk->icsk_ack.ato   = TCP_ATO_MIN;
403 		} else
404 			icsk->icsk_ack.quick -= pkts;
405 	}
406 }
407 
408 #define	TCP_ECN_OK		1
409 #define	TCP_ECN_QUEUE_CWR	2
410 #define	TCP_ECN_DEMAND_CWR	4
411 #define	TCP_ECN_SEEN		8
412 
413 enum tcp_tw_status {
414 	TCP_TW_SUCCESS = 0,
415 	TCP_TW_RST = 1,
416 	TCP_TW_ACK = 2,
417 	TCP_TW_SYN = 3
418 };
419 
420 
421 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
422 						     struct sk_buff *skb,
423 						     const struct tcphdr *th);
424 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
425 				   struct request_sock *req,
426 				   struct request_sock **prev,
427 				   bool fastopen);
428 extern int tcp_child_process(struct sock *parent, struct sock *child,
429 			     struct sk_buff *skb);
430 extern bool tcp_use_frto(struct sock *sk);
431 extern void tcp_enter_frto(struct sock *sk);
432 extern void tcp_enter_loss(struct sock *sk, int how);
433 extern void tcp_clear_retrans(struct tcp_sock *tp);
434 extern void tcp_update_metrics(struct sock *sk);
435 extern void tcp_init_metrics(struct sock *sk);
436 extern void tcp_metrics_init(void);
437 extern bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, bool paws_check);
438 extern bool tcp_remember_stamp(struct sock *sk);
439 extern bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
440 extern void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
441 extern void tcp_disable_fack(struct tcp_sock *tp);
442 extern void tcp_close(struct sock *sk, long timeout);
443 extern void tcp_init_sock(struct sock *sk);
444 extern unsigned int tcp_poll(struct file * file, struct socket *sock,
445 			     struct poll_table_struct *wait);
446 extern int tcp_getsockopt(struct sock *sk, int level, int optname,
447 			  char __user *optval, int __user *optlen);
448 extern int tcp_setsockopt(struct sock *sk, int level, int optname,
449 			  char __user *optval, unsigned int optlen);
450 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
451 				 char __user *optval, int __user *optlen);
452 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
453 				 char __user *optval, unsigned int optlen);
454 extern void tcp_set_keepalive(struct sock *sk, int val);
455 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
456 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
457 		       size_t len, int nonblock, int flags, int *addr_len);
458 extern void tcp_parse_options(const struct sk_buff *skb,
459 			      struct tcp_options_received *opt_rx, const u8 **hvpp,
460 			      int estab, struct tcp_fastopen_cookie *foc);
461 extern const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
462 
463 /*
464  *	TCP v4 functions exported for the inet6 API
465  */
466 
467 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
468 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
469 extern struct sock * tcp_create_openreq_child(struct sock *sk,
470 					      struct request_sock *req,
471 					      struct sk_buff *skb);
472 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
473 					  struct request_sock *req,
474 					  struct dst_entry *dst);
475 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
476 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr,
477 			  int addr_len);
478 extern int tcp_connect(struct sock *sk);
479 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst,
480 					struct request_sock *req,
481 					struct request_values *rvp,
482 					struct tcp_fastopen_cookie *foc);
483 extern int tcp_disconnect(struct sock *sk, int flags);
484 
485 void tcp_connect_init(struct sock *sk);
486 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
487 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
488 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
489 
490 /* From syncookies.c */
491 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
492 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
493 				    struct ip_options *opt);
494 #ifdef CONFIG_SYN_COOKIES
495 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
496 				     __u16 *mss);
497 #else
498 static inline __u32 cookie_v4_init_sequence(struct sock *sk,
499 					    struct sk_buff *skb,
500 					    __u16 *mss)
501 {
502 	return 0;
503 }
504 #endif
505 
506 extern __u32 cookie_init_timestamp(struct request_sock *req);
507 extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *);
508 
509 /* From net/ipv6/syncookies.c */
510 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
511 #ifdef CONFIG_SYN_COOKIES
512 extern __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb,
513 				     __u16 *mss);
514 #else
515 static inline __u32 cookie_v6_init_sequence(struct sock *sk,
516 					    struct sk_buff *skb,
517 					    __u16 *mss)
518 {
519 	return 0;
520 }
521 #endif
522 /* tcp_output.c */
523 
524 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
525 				      int nonagle);
526 extern bool tcp_may_send_now(struct sock *sk);
527 extern int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
528 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
529 extern void tcp_retransmit_timer(struct sock *sk);
530 extern void tcp_xmit_retransmit_queue(struct sock *);
531 extern void tcp_simple_retransmit(struct sock *);
532 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
533 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
534 
535 extern void tcp_send_probe0(struct sock *);
536 extern void tcp_send_partial(struct sock *);
537 extern int tcp_write_wakeup(struct sock *);
538 extern void tcp_send_fin(struct sock *sk);
539 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
540 extern int tcp_send_synack(struct sock *);
541 extern bool tcp_syn_flood_action(struct sock *sk,
542 				 const struct sk_buff *skb,
543 				 const char *proto);
544 extern void tcp_push_one(struct sock *, unsigned int mss_now);
545 extern void tcp_send_ack(struct sock *sk);
546 extern void tcp_send_delayed_ack(struct sock *sk);
547 
548 /* tcp_input.c */
549 extern void tcp_cwnd_application_limited(struct sock *sk);
550 extern void tcp_resume_early_retransmit(struct sock *sk);
551 extern void tcp_rearm_rto(struct sock *sk);
552 extern void tcp_reset(struct sock *sk);
553 
554 /* tcp_timer.c */
555 extern void tcp_init_xmit_timers(struct sock *);
556 static inline void tcp_clear_xmit_timers(struct sock *sk)
557 {
558 	inet_csk_clear_xmit_timers(sk);
559 }
560 
561 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
562 extern unsigned int tcp_current_mss(struct sock *sk);
563 
564 /* Bound MSS / TSO packet size with the half of the window */
565 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
566 {
567 	int cutoff;
568 
569 	/* When peer uses tiny windows, there is no use in packetizing
570 	 * to sub-MSS pieces for the sake of SWS or making sure there
571 	 * are enough packets in the pipe for fast recovery.
572 	 *
573 	 * On the other hand, for extremely large MSS devices, handling
574 	 * smaller than MSS windows in this way does make sense.
575 	 */
576 	if (tp->max_window >= 512)
577 		cutoff = (tp->max_window >> 1);
578 	else
579 		cutoff = tp->max_window;
580 
581 	if (cutoff && pktsize > cutoff)
582 		return max_t(int, cutoff, 68U - tp->tcp_header_len);
583 	else
584 		return pktsize;
585 }
586 
587 /* tcp.c */
588 extern void tcp_get_info(const struct sock *, struct tcp_info *);
589 
590 /* Read 'sendfile()'-style from a TCP socket */
591 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
592 				unsigned int, size_t);
593 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
594 			 sk_read_actor_t recv_actor);
595 
596 extern void tcp_initialize_rcv_mss(struct sock *sk);
597 
598 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
599 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
600 extern void tcp_mtup_init(struct sock *sk);
601 extern void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt);
602 extern void tcp_init_buffer_space(struct sock *sk);
603 
604 static inline void tcp_bound_rto(const struct sock *sk)
605 {
606 	if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
607 		inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
608 }
609 
610 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
611 {
612 	return (tp->srtt >> 3) + tp->rttvar;
613 }
614 
615 extern void tcp_set_rto(struct sock *sk);
616 
617 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
618 {
619 	tp->pred_flags = htonl((tp->tcp_header_len << 26) |
620 			       ntohl(TCP_FLAG_ACK) |
621 			       snd_wnd);
622 }
623 
624 static inline void tcp_fast_path_on(struct tcp_sock *tp)
625 {
626 	__tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
627 }
628 
629 static inline void tcp_fast_path_check(struct sock *sk)
630 {
631 	struct tcp_sock *tp = tcp_sk(sk);
632 
633 	if (skb_queue_empty(&tp->out_of_order_queue) &&
634 	    tp->rcv_wnd &&
635 	    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
636 	    !tp->urg_data)
637 		tcp_fast_path_on(tp);
638 }
639 
640 /* Compute the actual rto_min value */
641 static inline u32 tcp_rto_min(struct sock *sk)
642 {
643 	const struct dst_entry *dst = __sk_dst_get(sk);
644 	u32 rto_min = TCP_RTO_MIN;
645 
646 	if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
647 		rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
648 	return rto_min;
649 }
650 
651 /* Compute the actual receive window we are currently advertising.
652  * Rcv_nxt can be after the window if our peer push more data
653  * than the offered window.
654  */
655 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
656 {
657 	s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
658 
659 	if (win < 0)
660 		win = 0;
661 	return (u32) win;
662 }
663 
664 /* Choose a new window, without checks for shrinking, and without
665  * scaling applied to the result.  The caller does these things
666  * if necessary.  This is a "raw" window selection.
667  */
668 extern u32 __tcp_select_window(struct sock *sk);
669 
670 void tcp_send_window_probe(struct sock *sk);
671 
672 /* TCP timestamps are only 32-bits, this causes a slight
673  * complication on 64-bit systems since we store a snapshot
674  * of jiffies in the buffer control blocks below.  We decided
675  * to use only the low 32-bits of jiffies and hide the ugly
676  * casts with the following macro.
677  */
678 #define tcp_time_stamp		((__u32)(jiffies))
679 
680 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
681 
682 #define TCPHDR_FIN 0x01
683 #define TCPHDR_SYN 0x02
684 #define TCPHDR_RST 0x04
685 #define TCPHDR_PSH 0x08
686 #define TCPHDR_ACK 0x10
687 #define TCPHDR_URG 0x20
688 #define TCPHDR_ECE 0x40
689 #define TCPHDR_CWR 0x80
690 
691 /* This is what the send packet queuing engine uses to pass
692  * TCP per-packet control information to the transmission code.
693  * We also store the host-order sequence numbers in here too.
694  * This is 44 bytes if IPV6 is enabled.
695  * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
696  */
697 struct tcp_skb_cb {
698 	union {
699 		struct inet_skb_parm	h4;
700 #if IS_ENABLED(CONFIG_IPV6)
701 		struct inet6_skb_parm	h6;
702 #endif
703 	} header;	/* For incoming frames		*/
704 	__u32		seq;		/* Starting sequence number	*/
705 	__u32		end_seq;	/* SEQ + FIN + SYN + datalen	*/
706 	__u32		when;		/* used to compute rtt's	*/
707 	__u8		tcp_flags;	/* TCP header flags. (tcp[13])	*/
708 
709 	__u8		sacked;		/* State flags for SACK/FACK.	*/
710 #define TCPCB_SACKED_ACKED	0x01	/* SKB ACK'd by a SACK block	*/
711 #define TCPCB_SACKED_RETRANS	0x02	/* SKB retransmitted		*/
712 #define TCPCB_LOST		0x04	/* SKB is lost			*/
713 #define TCPCB_TAGBITS		0x07	/* All tag bits			*/
714 #define TCPCB_EVER_RETRANS	0x80	/* Ever retransmitted frame	*/
715 #define TCPCB_RETRANS		(TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
716 
717 	__u8		ip_dsfield;	/* IPv4 tos or IPv6 dsfield	*/
718 	/* 1 byte hole */
719 	__u32		ack_seq;	/* Sequence number ACK'd	*/
720 };
721 
722 #define TCP_SKB_CB(__skb)	((struct tcp_skb_cb *)&((__skb)->cb[0]))
723 
724 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set
725  *
726  * If we receive a SYN packet with these bits set, it means a network is
727  * playing bad games with TOS bits. In order to avoid possible false congestion
728  * notifications, we disable TCP ECN negociation.
729  */
730 static inline void
731 TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb)
732 {
733 	const struct tcphdr *th = tcp_hdr(skb);
734 
735 	if (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 	__skb_queue_tail(&tp->ucopy.prequeue, skb);
1049 	tp->ucopy.memory += skb->truesize;
1050 	if (tp->ucopy.memory > sk->sk_rcvbuf) {
1051 		struct sk_buff *skb1;
1052 
1053 		BUG_ON(sock_owned_by_user(sk));
1054 
1055 		while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1056 			sk_backlog_rcv(sk, skb1);
1057 			NET_INC_STATS_BH(sock_net(sk),
1058 					 LINUX_MIB_TCPPREQUEUEDROPPED);
1059 		}
1060 
1061 		tp->ucopy.memory = 0;
1062 	} else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1063 		wake_up_interruptible_sync_poll(sk_sleep(sk),
1064 					   POLLIN | POLLRDNORM | POLLRDBAND);
1065 		if (!inet_csk_ack_scheduled(sk))
1066 			inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1067 						  (3 * tcp_rto_min(sk)) / 4,
1068 						  TCP_RTO_MAX);
1069 	}
1070 	return true;
1071 }
1072 
1073 
1074 #undef STATE_TRACE
1075 
1076 #ifdef STATE_TRACE
1077 static const char *statename[]={
1078 	"Unused","Established","Syn Sent","Syn Recv",
1079 	"Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1080 	"Close Wait","Last ACK","Listen","Closing"
1081 };
1082 #endif
1083 extern void tcp_set_state(struct sock *sk, int state);
1084 
1085 extern void tcp_done(struct sock *sk);
1086 
1087 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1088 {
1089 	rx_opt->dsack = 0;
1090 	rx_opt->num_sacks = 0;
1091 }
1092 
1093 /* Determine a window scaling and initial window to offer. */
1094 extern void tcp_select_initial_window(int __space, __u32 mss,
1095 				      __u32 *rcv_wnd, __u32 *window_clamp,
1096 				      int wscale_ok, __u8 *rcv_wscale,
1097 				      __u32 init_rcv_wnd);
1098 
1099 static inline int tcp_win_from_space(int space)
1100 {
1101 	return sysctl_tcp_adv_win_scale<=0 ?
1102 		(space>>(-sysctl_tcp_adv_win_scale)) :
1103 		space - (space>>sysctl_tcp_adv_win_scale);
1104 }
1105 
1106 /* Note: caller must be prepared to deal with negative returns */
1107 static inline int tcp_space(const struct sock *sk)
1108 {
1109 	return tcp_win_from_space(sk->sk_rcvbuf -
1110 				  atomic_read(&sk->sk_rmem_alloc));
1111 }
1112 
1113 static inline int tcp_full_space(const struct sock *sk)
1114 {
1115 	return tcp_win_from_space(sk->sk_rcvbuf);
1116 }
1117 
1118 static inline void tcp_openreq_init(struct request_sock *req,
1119 				    struct tcp_options_received *rx_opt,
1120 				    struct sk_buff *skb)
1121 {
1122 	struct inet_request_sock *ireq = inet_rsk(req);
1123 
1124 	req->rcv_wnd = 0;		/* So that tcp_send_synack() knows! */
1125 	req->cookie_ts = 0;
1126 	tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1127 	tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1;
1128 	tcp_rsk(req)->snt_synack = 0;
1129 	req->mss = rx_opt->mss_clamp;
1130 	req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1131 	ireq->tstamp_ok = rx_opt->tstamp_ok;
1132 	ireq->sack_ok = rx_opt->sack_ok;
1133 	ireq->snd_wscale = rx_opt->snd_wscale;
1134 	ireq->wscale_ok = rx_opt->wscale_ok;
1135 	ireq->acked = 0;
1136 	ireq->ecn_ok = 0;
1137 	ireq->rmt_port = tcp_hdr(skb)->source;
1138 	ireq->loc_port = tcp_hdr(skb)->dest;
1139 }
1140 
1141 /* Compute time elapsed between SYNACK and the ACK completing 3WHS */
1142 static inline void tcp_synack_rtt_meas(struct sock *sk,
1143 				       struct request_sock *req)
1144 {
1145 	if (tcp_rsk(req)->snt_synack)
1146 		tcp_valid_rtt_meas(sk,
1147 		    tcp_time_stamp - tcp_rsk(req)->snt_synack);
1148 }
1149 
1150 extern void tcp_enter_memory_pressure(struct sock *sk);
1151 
1152 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1153 {
1154 	return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1155 }
1156 
1157 static inline int keepalive_time_when(const struct tcp_sock *tp)
1158 {
1159 	return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1160 }
1161 
1162 static inline int keepalive_probes(const struct tcp_sock *tp)
1163 {
1164 	return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1165 }
1166 
1167 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1168 {
1169 	const struct inet_connection_sock *icsk = &tp->inet_conn;
1170 
1171 	return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1172 			  tcp_time_stamp - tp->rcv_tstamp);
1173 }
1174 
1175 static inline int tcp_fin_time(const struct sock *sk)
1176 {
1177 	int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1178 	const int rto = inet_csk(sk)->icsk_rto;
1179 
1180 	if (fin_timeout < (rto << 2) - (rto >> 1))
1181 		fin_timeout = (rto << 2) - (rto >> 1);
1182 
1183 	return fin_timeout;
1184 }
1185 
1186 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1187 				  int paws_win)
1188 {
1189 	if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1190 		return true;
1191 	if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1192 		return true;
1193 	/*
1194 	 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1195 	 * then following tcp messages have valid values. Ignore 0 value,
1196 	 * or else 'negative' tsval might forbid us to accept their packets.
1197 	 */
1198 	if (!rx_opt->ts_recent)
1199 		return true;
1200 	return false;
1201 }
1202 
1203 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1204 				   int rst)
1205 {
1206 	if (tcp_paws_check(rx_opt, 0))
1207 		return false;
1208 
1209 	/* RST segments are not recommended to carry timestamp,
1210 	   and, if they do, it is recommended to ignore PAWS because
1211 	   "their cleanup function should take precedence over timestamps."
1212 	   Certainly, it is mistake. It is necessary to understand the reasons
1213 	   of this constraint to relax it: if peer reboots, clock may go
1214 	   out-of-sync and half-open connections will not be reset.
1215 	   Actually, the problem would be not existing if all
1216 	   the implementations followed draft about maintaining clock
1217 	   via reboots. Linux-2.2 DOES NOT!
1218 
1219 	   However, we can relax time bounds for RST segments to MSL.
1220 	 */
1221 	if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1222 		return false;
1223 	return true;
1224 }
1225 
1226 static inline void tcp_mib_init(struct net *net)
1227 {
1228 	/* See RFC 2012 */
1229 	TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1230 	TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1231 	TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1232 	TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1233 }
1234 
1235 /* from STCP */
1236 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1237 {
1238 	tp->lost_skb_hint = NULL;
1239 	tp->scoreboard_skb_hint = NULL;
1240 }
1241 
1242 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1243 {
1244 	tcp_clear_retrans_hints_partial(tp);
1245 	tp->retransmit_skb_hint = NULL;
1246 }
1247 
1248 /* MD5 Signature */
1249 struct crypto_hash;
1250 
1251 union tcp_md5_addr {
1252 	struct in_addr  a4;
1253 #if IS_ENABLED(CONFIG_IPV6)
1254 	struct in6_addr	a6;
1255 #endif
1256 };
1257 
1258 /* - key database */
1259 struct tcp_md5sig_key {
1260 	struct hlist_node	node;
1261 	u8			keylen;
1262 	u8			family; /* AF_INET or AF_INET6 */
1263 	union tcp_md5_addr	addr;
1264 	u8			key[TCP_MD5SIG_MAXKEYLEN];
1265 	struct rcu_head		rcu;
1266 };
1267 
1268 /* - sock block */
1269 struct tcp_md5sig_info {
1270 	struct hlist_head	head;
1271 	struct rcu_head		rcu;
1272 };
1273 
1274 /* - pseudo header */
1275 struct tcp4_pseudohdr {
1276 	__be32		saddr;
1277 	__be32		daddr;
1278 	__u8		pad;
1279 	__u8		protocol;
1280 	__be16		len;
1281 };
1282 
1283 struct tcp6_pseudohdr {
1284 	struct in6_addr	saddr;
1285 	struct in6_addr daddr;
1286 	__be32		len;
1287 	__be32		protocol;	/* including padding */
1288 };
1289 
1290 union tcp_md5sum_block {
1291 	struct tcp4_pseudohdr ip4;
1292 #if IS_ENABLED(CONFIG_IPV6)
1293 	struct tcp6_pseudohdr ip6;
1294 #endif
1295 };
1296 
1297 /* - pool: digest algorithm, hash description and scratch buffer */
1298 struct tcp_md5sig_pool {
1299 	struct hash_desc	md5_desc;
1300 	union tcp_md5sum_block	md5_blk;
1301 };
1302 
1303 /* - functions */
1304 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1305 			       const struct sock *sk,
1306 			       const struct request_sock *req,
1307 			       const struct sk_buff *skb);
1308 extern int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1309 			  int family, const u8 *newkey,
1310 			  u8 newkeylen, gfp_t gfp);
1311 extern int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1312 			  int family);
1313 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1314 					 struct sock *addr_sk);
1315 
1316 #ifdef CONFIG_TCP_MD5SIG
1317 extern struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1318 			const union tcp_md5_addr *addr, int family);
1319 #define tcp_twsk_md5_key(twsk)	((twsk)->tw_md5_key)
1320 #else
1321 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1322 					 const union tcp_md5_addr *addr,
1323 					 int family)
1324 {
1325 	return NULL;
1326 }
1327 #define tcp_twsk_md5_key(twsk)	NULL
1328 #endif
1329 
1330 extern struct tcp_md5sig_pool __percpu *tcp_alloc_md5sig_pool(struct sock *);
1331 extern void tcp_free_md5sig_pool(void);
1332 
1333 extern struct tcp_md5sig_pool	*tcp_get_md5sig_pool(void);
1334 extern void tcp_put_md5sig_pool(void);
1335 
1336 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1337 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1338 				 unsigned int header_len);
1339 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1340 			    const struct tcp_md5sig_key *key);
1341 
1342 /* From tcp_fastopen.c */
1343 extern void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1344 				   struct tcp_fastopen_cookie *cookie,
1345 				   int *syn_loss, unsigned long *last_syn_loss);
1346 extern void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1347 				   struct tcp_fastopen_cookie *cookie,
1348 				   bool syn_lost);
1349 struct tcp_fastopen_request {
1350 	/* Fast Open cookie. Size 0 means a cookie request */
1351 	struct tcp_fastopen_cookie	cookie;
1352 	struct msghdr			*data;  /* data in MSG_FASTOPEN */
1353 	u16				copied;	/* queued in tcp_connect() */
1354 };
1355 void tcp_free_fastopen_req(struct tcp_sock *tp);
1356 
1357 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1358 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1359 void tcp_fastopen_cookie_gen(__be32 addr, struct tcp_fastopen_cookie *foc);
1360 
1361 #define TCP_FASTOPEN_KEY_LENGTH 16
1362 
1363 /* Fastopen key context */
1364 struct tcp_fastopen_context {
1365 	struct crypto_cipher __rcu	*tfm;
1366 	__u8				key[TCP_FASTOPEN_KEY_LENGTH];
1367 	struct rcu_head			rcu;
1368 };
1369 
1370 /* write queue abstraction */
1371 static inline void tcp_write_queue_purge(struct sock *sk)
1372 {
1373 	struct sk_buff *skb;
1374 
1375 	while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1376 		sk_wmem_free_skb(sk, skb);
1377 	sk_mem_reclaim(sk);
1378 	tcp_clear_all_retrans_hints(tcp_sk(sk));
1379 }
1380 
1381 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1382 {
1383 	return skb_peek(&sk->sk_write_queue);
1384 }
1385 
1386 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1387 {
1388 	return skb_peek_tail(&sk->sk_write_queue);
1389 }
1390 
1391 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1392 						   const struct sk_buff *skb)
1393 {
1394 	return skb_queue_next(&sk->sk_write_queue, skb);
1395 }
1396 
1397 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1398 						   const struct sk_buff *skb)
1399 {
1400 	return skb_queue_prev(&sk->sk_write_queue, skb);
1401 }
1402 
1403 #define tcp_for_write_queue(skb, sk)					\
1404 	skb_queue_walk(&(sk)->sk_write_queue, skb)
1405 
1406 #define tcp_for_write_queue_from(skb, sk)				\
1407 	skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1408 
1409 #define tcp_for_write_queue_from_safe(skb, tmp, sk)			\
1410 	skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1411 
1412 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1413 {
1414 	return sk->sk_send_head;
1415 }
1416 
1417 static inline bool tcp_skb_is_last(const struct sock *sk,
1418 				   const struct sk_buff *skb)
1419 {
1420 	return skb_queue_is_last(&sk->sk_write_queue, skb);
1421 }
1422 
1423 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1424 {
1425 	if (tcp_skb_is_last(sk, skb))
1426 		sk->sk_send_head = NULL;
1427 	else
1428 		sk->sk_send_head = tcp_write_queue_next(sk, skb);
1429 }
1430 
1431 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1432 {
1433 	if (sk->sk_send_head == skb_unlinked)
1434 		sk->sk_send_head = NULL;
1435 }
1436 
1437 static inline void tcp_init_send_head(struct sock *sk)
1438 {
1439 	sk->sk_send_head = NULL;
1440 }
1441 
1442 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1443 {
1444 	__skb_queue_tail(&sk->sk_write_queue, skb);
1445 }
1446 
1447 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1448 {
1449 	__tcp_add_write_queue_tail(sk, skb);
1450 
1451 	/* Queue it, remembering where we must start sending. */
1452 	if (sk->sk_send_head == NULL) {
1453 		sk->sk_send_head = skb;
1454 
1455 		if (tcp_sk(sk)->highest_sack == NULL)
1456 			tcp_sk(sk)->highest_sack = skb;
1457 	}
1458 }
1459 
1460 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1461 {
1462 	__skb_queue_head(&sk->sk_write_queue, skb);
1463 }
1464 
1465 /* Insert buff after skb on the write queue of sk.  */
1466 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1467 						struct sk_buff *buff,
1468 						struct sock *sk)
1469 {
1470 	__skb_queue_after(&sk->sk_write_queue, skb, buff);
1471 }
1472 
1473 /* Insert new before skb on the write queue of sk.  */
1474 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1475 						  struct sk_buff *skb,
1476 						  struct sock *sk)
1477 {
1478 	__skb_queue_before(&sk->sk_write_queue, skb, new);
1479 
1480 	if (sk->sk_send_head == skb)
1481 		sk->sk_send_head = new;
1482 }
1483 
1484 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1485 {
1486 	__skb_unlink(skb, &sk->sk_write_queue);
1487 }
1488 
1489 static inline bool tcp_write_queue_empty(struct sock *sk)
1490 {
1491 	return skb_queue_empty(&sk->sk_write_queue);
1492 }
1493 
1494 static inline void tcp_push_pending_frames(struct sock *sk)
1495 {
1496 	if (tcp_send_head(sk)) {
1497 		struct tcp_sock *tp = tcp_sk(sk);
1498 
1499 		__tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1500 	}
1501 }
1502 
1503 /* Start sequence of the skb just after the highest skb with SACKed
1504  * bit, valid only if sacked_out > 0 or when the caller has ensured
1505  * validity by itself.
1506  */
1507 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1508 {
1509 	if (!tp->sacked_out)
1510 		return tp->snd_una;
1511 
1512 	if (tp->highest_sack == NULL)
1513 		return tp->snd_nxt;
1514 
1515 	return TCP_SKB_CB(tp->highest_sack)->seq;
1516 }
1517 
1518 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1519 {
1520 	tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1521 						tcp_write_queue_next(sk, skb);
1522 }
1523 
1524 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1525 {
1526 	return tcp_sk(sk)->highest_sack;
1527 }
1528 
1529 static inline void tcp_highest_sack_reset(struct sock *sk)
1530 {
1531 	tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1532 }
1533 
1534 /* Called when old skb is about to be deleted (to be combined with new skb) */
1535 static inline void tcp_highest_sack_combine(struct sock *sk,
1536 					    struct sk_buff *old,
1537 					    struct sk_buff *new)
1538 {
1539 	if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1540 		tcp_sk(sk)->highest_sack = new;
1541 }
1542 
1543 /* Determines whether this is a thin stream (which may suffer from
1544  * increased latency). Used to trigger latency-reducing mechanisms.
1545  */
1546 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1547 {
1548 	return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1549 }
1550 
1551 /* /proc */
1552 enum tcp_seq_states {
1553 	TCP_SEQ_STATE_LISTENING,
1554 	TCP_SEQ_STATE_OPENREQ,
1555 	TCP_SEQ_STATE_ESTABLISHED,
1556 	TCP_SEQ_STATE_TIME_WAIT,
1557 };
1558 
1559 int tcp_seq_open(struct inode *inode, struct file *file);
1560 
1561 struct tcp_seq_afinfo {
1562 	char				*name;
1563 	sa_family_t			family;
1564 	const struct file_operations	*seq_fops;
1565 	struct seq_operations		seq_ops;
1566 };
1567 
1568 struct tcp_iter_state {
1569 	struct seq_net_private	p;
1570 	sa_family_t		family;
1571 	enum tcp_seq_states	state;
1572 	struct sock		*syn_wait_sk;
1573 	int			bucket, offset, sbucket, num;
1574 	kuid_t			uid;
1575 	loff_t			last_pos;
1576 };
1577 
1578 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1579 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1580 
1581 extern struct request_sock_ops tcp_request_sock_ops;
1582 extern struct request_sock_ops tcp6_request_sock_ops;
1583 
1584 extern void tcp_v4_destroy_sock(struct sock *sk);
1585 
1586 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1587 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb,
1588 				       netdev_features_t features);
1589 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1590 					struct sk_buff *skb);
1591 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1592 					 struct sk_buff *skb);
1593 extern int tcp_gro_complete(struct sk_buff *skb);
1594 extern int tcp4_gro_complete(struct sk_buff *skb);
1595 
1596 #ifdef CONFIG_PROC_FS
1597 extern int tcp4_proc_init(void);
1598 extern void tcp4_proc_exit(void);
1599 #endif
1600 
1601 /* TCP af-specific functions */
1602 struct tcp_sock_af_ops {
1603 #ifdef CONFIG_TCP_MD5SIG
1604 	struct tcp_md5sig_key	*(*md5_lookup) (struct sock *sk,
1605 						struct sock *addr_sk);
1606 	int			(*calc_md5_hash) (char *location,
1607 						  struct tcp_md5sig_key *md5,
1608 						  const struct sock *sk,
1609 						  const struct request_sock *req,
1610 						  const struct sk_buff *skb);
1611 	int			(*md5_parse) (struct sock *sk,
1612 					      char __user *optval,
1613 					      int optlen);
1614 #endif
1615 };
1616 
1617 struct tcp_request_sock_ops {
1618 #ifdef CONFIG_TCP_MD5SIG
1619 	struct tcp_md5sig_key	*(*md5_lookup) (struct sock *sk,
1620 						struct request_sock *req);
1621 	int			(*calc_md5_hash) (char *location,
1622 						  struct tcp_md5sig_key *md5,
1623 						  const struct sock *sk,
1624 						  const struct request_sock *req,
1625 						  const struct sk_buff *skb);
1626 #endif
1627 };
1628 
1629 /* Using SHA1 for now, define some constants.
1630  */
1631 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS)
1632 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4)
1633 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS)
1634 
1635 extern int tcp_cookie_generator(u32 *bakery);
1636 
1637 /**
1638  *	struct tcp_cookie_values - each socket needs extra space for the
1639  *	cookies, together with (optional) space for any SYN data.
1640  *
1641  *	A tcp_sock contains a pointer to the current value, and this is
1642  *	cloned to the tcp_timewait_sock.
1643  *
1644  * @cookie_pair:	variable data from the option exchange.
1645  *
1646  * @cookie_desired:	user specified tcpct_cookie_desired.  Zero
1647  *			indicates default (sysctl_tcp_cookie_size).
1648  *			After cookie sent, remembers size of cookie.
1649  *			Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX.
1650  *
1651  * @s_data_desired:	user specified tcpct_s_data_desired.  When the
1652  *			constant payload is specified (@s_data_constant),
1653  *			holds its length instead.
1654  *			Range 0 to TCP_MSS_DESIRED.
1655  *
1656  * @s_data_payload:	constant data that is to be included in the
1657  *			payload of SYN or SYNACK segments when the
1658  *			cookie option is present.
1659  */
1660 struct tcp_cookie_values {
1661 	struct kref	kref;
1662 	u8		cookie_pair[TCP_COOKIE_PAIR_SIZE];
1663 	u8		cookie_pair_size;
1664 	u8		cookie_desired;
1665 	u16		s_data_desired:11,
1666 			s_data_constant:1,
1667 			s_data_in:1,
1668 			s_data_out:1,
1669 			s_data_unused:2;
1670 	u8		s_data_payload[0];
1671 };
1672 
1673 static inline void tcp_cookie_values_release(struct kref *kref)
1674 {
1675 	kfree(container_of(kref, struct tcp_cookie_values, kref));
1676 }
1677 
1678 /* The length of constant payload data.  Note that s_data_desired is
1679  * overloaded, depending on s_data_constant: either the length of constant
1680  * data (returned here) or the limit on variable data.
1681  */
1682 static inline int tcp_s_data_size(const struct tcp_sock *tp)
1683 {
1684 	return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant)
1685 		? tp->cookie_values->s_data_desired
1686 		: 0;
1687 }
1688 
1689 /**
1690  *	struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace.
1691  *
1692  *	As tcp_request_sock has already been extended in other places, the
1693  *	only remaining method is to pass stack values along as function
1694  *	parameters.  These parameters are not needed after sending SYNACK.
1695  *
1696  * @cookie_bakery:	cryptographic secret and message workspace.
1697  *
1698  * @cookie_plus:	bytes in authenticator/cookie option, copied from
1699  *			struct tcp_options_received (above).
1700  */
1701 struct tcp_extend_values {
1702 	struct request_values		rv;
1703 	u32				cookie_bakery[COOKIE_WORKSPACE_WORDS];
1704 	u8				cookie_plus:6,
1705 					cookie_out_never:1,
1706 					cookie_in_always:1;
1707 };
1708 
1709 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp)
1710 {
1711 	return (struct tcp_extend_values *)rvp;
1712 }
1713 
1714 extern void tcp_v4_init(void);
1715 extern void tcp_init(void);
1716 
1717 #endif	/* _TCP_H */
1718